2 * turbostat -- show CPU frequency and C-state residency
3 * on modern Intel turbo-capable processors.
5 * Copyright (c) 2013 Intel Corporation.
6 * Len Brown <len.brown@intel.com>
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms and conditions of the GNU General Public License,
10 * version 2, as published by the Free Software Foundation.
12 * This program is distributed in the hope it will be useful, but WITHOUT
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 * You should have received a copy of the GNU General Public License along with
18 * this program; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
24 #include INTEL_FAMILY_HEADER
29 #include <sys/types.h>
32 #include <sys/resource.h>
44 #include <linux/capability.h>
47 char *proc_stat = "/proc/stat";
50 struct timespec interval_ts = {5, 0};
53 unsigned int sums_need_wide_columns;
54 unsigned int rapl_joules;
55 unsigned int summary_only;
56 unsigned int list_header_only;
57 unsigned int dump_only;
58 unsigned int do_snb_cstates;
59 unsigned int do_knl_cstates;
60 unsigned int do_skl_residency;
61 unsigned int do_slm_cstates;
62 unsigned int use_c1_residency_msr;
63 unsigned int has_aperf;
65 unsigned int do_irtl_snb;
66 unsigned int do_irtl_hsw;
67 unsigned int units = 1000000; /* MHz etc */
68 unsigned int genuine_intel;
69 unsigned int has_invariant_tsc;
70 unsigned int do_nhm_platform_info;
71 unsigned int no_MSR_MISC_PWR_MGMT;
72 unsigned int aperf_mperf_multiplier = 1;
75 unsigned int has_base_hz;
76 double tsc_tweak = 1.0;
77 unsigned int show_pkg_only;
78 unsigned int show_core_only;
79 char *output_buffer, *outp;
83 unsigned long long gfx_cur_rc6_ms;
84 unsigned int gfx_cur_mhz;
85 unsigned int tcc_activation_temp;
86 unsigned int tcc_activation_temp_override;
87 double rapl_power_units, rapl_time_units;
88 double rapl_dram_energy_units, rapl_energy_units;
89 double rapl_joule_counter_range;
90 unsigned int do_core_perf_limit_reasons;
91 unsigned int do_gfx_perf_limit_reasons;
92 unsigned int do_ring_perf_limit_reasons;
93 unsigned int crystal_hz;
94 unsigned long long tsc_hz;
96 double discover_bclk(unsigned int family, unsigned int model);
97 unsigned int has_hwp; /* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
98 /* IA32_HWP_REQUEST, IA32_HWP_STATUS */
99 unsigned int has_hwp_notify; /* IA32_HWP_INTERRUPT */
100 unsigned int has_hwp_activity_window; /* IA32_HWP_REQUEST[bits 41:32] */
101 unsigned int has_hwp_epp; /* IA32_HWP_REQUEST[bits 31:24] */
102 unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */
103 unsigned int has_misc_feature_control;
105 #define RAPL_PKG (1 << 0)
106 /* 0x610 MSR_PKG_POWER_LIMIT */
107 /* 0x611 MSR_PKG_ENERGY_STATUS */
108 #define RAPL_PKG_PERF_STATUS (1 << 1)
109 /* 0x613 MSR_PKG_PERF_STATUS */
110 #define RAPL_PKG_POWER_INFO (1 << 2)
111 /* 0x614 MSR_PKG_POWER_INFO */
113 #define RAPL_DRAM (1 << 3)
114 /* 0x618 MSR_DRAM_POWER_LIMIT */
115 /* 0x619 MSR_DRAM_ENERGY_STATUS */
116 #define RAPL_DRAM_PERF_STATUS (1 << 4)
117 /* 0x61b MSR_DRAM_PERF_STATUS */
118 #define RAPL_DRAM_POWER_INFO (1 << 5)
119 /* 0x61c MSR_DRAM_POWER_INFO */
121 #define RAPL_CORES_POWER_LIMIT (1 << 6)
122 /* 0x638 MSR_PP0_POWER_LIMIT */
123 #define RAPL_CORE_POLICY (1 << 7)
124 /* 0x63a MSR_PP0_POLICY */
126 #define RAPL_GFX (1 << 8)
127 /* 0x640 MSR_PP1_POWER_LIMIT */
128 /* 0x641 MSR_PP1_ENERGY_STATUS */
129 /* 0x642 MSR_PP1_POLICY */
131 #define RAPL_CORES_ENERGY_STATUS (1 << 9)
132 /* 0x639 MSR_PP0_ENERGY_STATUS */
133 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
134 #define TJMAX_DEFAULT 100
136 #define MAX(a, b) ((a) > (b) ? (a) : (b))
139 * buffer size used by sscanf() for added column names
140 * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
142 #define NAME_BYTES 20
143 #define PATH_BYTES 128
148 #define CPU_SUBSET_MAXCPUS 1024 /* need to use before probe... */
149 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset;
150 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size;
151 #define MAX_ADDED_COUNTERS 16
154 unsigned long long tsc;
155 unsigned long long aperf;
156 unsigned long long mperf;
157 unsigned long long c1;
158 unsigned long long irq_count;
159 unsigned int smi_count;
162 #define CPU_IS_FIRST_THREAD_IN_CORE 0x2
163 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
164 unsigned long long counter[MAX_ADDED_COUNTERS];
165 } *thread_even, *thread_odd;
168 unsigned long long c3;
169 unsigned long long c6;
170 unsigned long long c7;
171 unsigned long long mc6_us; /* duplicate as per-core for now, even though per module */
172 unsigned int core_temp_c;
173 unsigned int core_id;
174 unsigned long long counter[MAX_ADDED_COUNTERS];
175 } *core_even, *core_odd;
178 unsigned long long pc2;
179 unsigned long long pc3;
180 unsigned long long pc6;
181 unsigned long long pc7;
182 unsigned long long pc8;
183 unsigned long long pc9;
184 unsigned long long pc10;
185 unsigned long long pkg_wtd_core_c0;
186 unsigned long long pkg_any_core_c0;
187 unsigned long long pkg_any_gfxe_c0;
188 unsigned long long pkg_both_core_gfxe_c0;
189 long long gfx_rc6_ms;
190 unsigned int gfx_mhz;
191 unsigned int package_id;
192 unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */
193 unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */
194 unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */
195 unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */
196 unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */
197 unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */
198 unsigned int pkg_temp_c;
199 unsigned long long counter[MAX_ADDED_COUNTERS];
200 } *package_even, *package_odd;
202 #define ODD_COUNTERS thread_odd, core_odd, package_odd
203 #define EVEN_COUNTERS thread_even, core_even, package_even
205 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
206 (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
207 topo.num_threads_per_core + \
208 (core_no) * topo.num_threads_per_core + (thread_no))
209 #define GET_CORE(core_base, core_no, pkg_no) \
210 (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
211 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
213 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
214 enum counter_type {COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC};
215 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
218 unsigned int msr_num;
219 char name[NAME_BYTES];
220 char path[PATH_BYTES];
222 enum counter_type type;
223 enum counter_format format;
224 struct msr_counter *next;
226 #define FLAGS_HIDE (1 << 0)
227 #define FLAGS_SHOW (1 << 1)
228 #define SYSFS_PERCPU (1 << 1)
231 struct sys_counters {
232 unsigned int added_thread_counters;
233 unsigned int added_core_counters;
234 unsigned int added_package_counters;
235 struct msr_counter *tp;
236 struct msr_counter *cp;
237 struct msr_counter *pp;
240 struct system_summary {
241 struct thread_data threads;
242 struct core_data cores;
243 struct pkg_data packages;
252 int num_cores_per_pkg;
253 int num_threads_per_core;
256 struct timeval tv_even, tv_odd, tv_delta;
258 int *irq_column_2_cpu; /* /proc/interrupts column numbers */
259 int *irqs_per_cpu; /* indexed by cpu_num */
261 void setup_all_buffers(void);
263 int cpu_is_not_present(int cpu)
265 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
268 * run func(thread, core, package) in topology order
269 * skip non-present cpus
272 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
273 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
275 int retval, pkg_no, core_no, thread_no;
277 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
278 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
279 for (thread_no = 0; thread_no <
280 topo.num_threads_per_core; ++thread_no) {
281 struct thread_data *t;
285 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
287 if (cpu_is_not_present(t->cpu_id))
290 c = GET_CORE(core_base, core_no, pkg_no);
291 p = GET_PKG(pkg_base, pkg_no);
293 retval = func(t, c, p);
302 int cpu_migrate(int cpu)
304 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
305 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
306 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
311 int get_msr_fd(int cpu)
321 sprintf(pathname, "/dev/cpu/%d/msr", cpu);
322 fd = open(pathname, O_RDONLY);
324 err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
331 int get_msr(int cpu, off_t offset, unsigned long long *msr)
335 retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
337 if (retval != sizeof *msr)
338 err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
344 * Each string in this array is compared in --show and --hide cmdline.
345 * Thus, strings that are proper sub-sets must follow their more specific peers.
347 struct msr_counter bic[] = {
353 { 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL},
389 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
390 #define BIC_Package (1ULL << 0)
391 #define BIC_Avg_MHz (1ULL << 1)
392 #define BIC_Bzy_MHz (1ULL << 2)
393 #define BIC_TSC_MHz (1ULL << 3)
394 #define BIC_IRQ (1ULL << 4)
395 #define BIC_SMI (1ULL << 5)
396 #define BIC_Busy (1ULL << 6)
397 #define BIC_CPU_c1 (1ULL << 7)
398 #define BIC_CPU_c3 (1ULL << 8)
399 #define BIC_CPU_c6 (1ULL << 9)
400 #define BIC_CPU_c7 (1ULL << 10)
401 #define BIC_ThreadC (1ULL << 11)
402 #define BIC_CoreTmp (1ULL << 12)
403 #define BIC_CoreCnt (1ULL << 13)
404 #define BIC_PkgTmp (1ULL << 14)
405 #define BIC_GFX_rc6 (1ULL << 15)
406 #define BIC_GFXMHz (1ULL << 16)
407 #define BIC_Pkgpc2 (1ULL << 17)
408 #define BIC_Pkgpc3 (1ULL << 18)
409 #define BIC_Pkgpc6 (1ULL << 19)
410 #define BIC_Pkgpc7 (1ULL << 20)
411 #define BIC_Pkgpc8 (1ULL << 21)
412 #define BIC_Pkgpc9 (1ULL << 22)
413 #define BIC_Pkgpc10 (1ULL << 23)
414 #define BIC_PkgWatt (1ULL << 24)
415 #define BIC_CorWatt (1ULL << 25)
416 #define BIC_GFXWatt (1ULL << 26)
417 #define BIC_PkgCnt (1ULL << 27)
418 #define BIC_RAMWatt (1ULL << 28)
419 #define BIC_PKG__ (1ULL << 29)
420 #define BIC_RAM__ (1ULL << 30)
421 #define BIC_Pkg_J (1ULL << 31)
422 #define BIC_Cor_J (1ULL << 32)
423 #define BIC_GFX_J (1ULL << 33)
424 #define BIC_RAM_J (1ULL << 34)
425 #define BIC_Core (1ULL << 35)
426 #define BIC_CPU (1ULL << 36)
427 #define BIC_Mod_c6 (1ULL << 37)
428 #define BIC_sysfs (1ULL << 38)
430 unsigned long long bic_enabled = 0xFFFFFFFFFFFFFFFFULL;
431 unsigned long long bic_present = BIC_sysfs;
433 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
434 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
435 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
437 #define MAX_DEFERRED 16
438 char *deferred_skip_names[MAX_DEFERRED];
439 int deferred_skip_index;
442 * HIDE_LIST - hide this list of counters, show the rest [default]
443 * SHOW_LIST - show this list of counters, hide the rest
445 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
450 "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
452 "Turbostat forks the specified COMMAND and prints statistics\n"
453 "when COMMAND completes.\n"
454 "If no COMMAND is specified, turbostat wakes every 5-seconds\n"
455 "to print statistics, until interrupted.\n"
456 "--add add a counter\n"
457 " eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
458 "--cpu cpu-set limit output to summary plus cpu-set:\n"
459 " {core | package | j,k,l..m,n-p }\n"
460 "--quiet skip decoding system configuration header\n"
461 "--interval sec Override default 5-second measurement interval\n"
462 "--help print this help message\n"
463 "--list list column headers only\n"
464 "--out file create or truncate \"file\" for all output\n"
465 "--version print version information\n"
467 "For more help, run \"man turbostat\"\n");
472 * for all the strings in comma separate name_list,
473 * set the approprate bit in return value.
475 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
478 unsigned long long retval = 0;
483 comma = strchr(name_list, ',');
488 for (i = 0; i < MAX_BIC; ++i) {
489 if (!strcmp(name_list, bic[i].name)) {
490 retval |= (1ULL << i);
495 if (mode == SHOW_LIST) {
496 fprintf(stderr, "Invalid counter name: %s\n", name_list);
499 deferred_skip_names[deferred_skip_index++] = name_list;
501 fprintf(stderr, "deferred \"%s\"\n", name_list);
502 if (deferred_skip_index >= MAX_DEFERRED) {
503 fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
504 MAX_DEFERRED, name_list);
519 void print_header(char *delim)
521 struct msr_counter *mp;
524 if (DO_BIC(BIC_Package))
525 outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
526 if (DO_BIC(BIC_Core))
527 outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
529 outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
530 if (DO_BIC(BIC_Avg_MHz))
531 outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
532 if (DO_BIC(BIC_Busy))
533 outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
534 if (DO_BIC(BIC_Bzy_MHz))
535 outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
536 if (DO_BIC(BIC_TSC_MHz))
537 outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
539 if (DO_BIC(BIC_IRQ)) {
540 if (sums_need_wide_columns)
541 outp += sprintf(outp, "%s IRQ", (printed++ ? delim : ""));
543 outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
547 outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
549 for (mp = sys.tp; mp; mp = mp->next) {
551 if (mp->format == FORMAT_RAW) {
553 outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
555 outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
557 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
558 outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
560 outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
564 if (DO_BIC(BIC_CPU_c1))
565 outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
566 if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates)
567 outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
568 if (DO_BIC(BIC_CPU_c6))
569 outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
570 if (DO_BIC(BIC_CPU_c7))
571 outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
573 if (DO_BIC(BIC_Mod_c6))
574 outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
576 if (DO_BIC(BIC_CoreTmp))
577 outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
579 for (mp = sys.cp; mp; mp = mp->next) {
580 if (mp->format == FORMAT_RAW) {
582 outp += sprintf(outp, "%s%18.18s", delim, mp->name);
584 outp += sprintf(outp, "%s%10.10s", delim, mp->name);
586 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
587 outp += sprintf(outp, "%s%8s", delim, mp->name);
589 outp += sprintf(outp, "%s%s", delim, mp->name);
593 if (DO_BIC(BIC_PkgTmp))
594 outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
596 if (DO_BIC(BIC_GFX_rc6))
597 outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
599 if (DO_BIC(BIC_GFXMHz))
600 outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
602 if (do_skl_residency) {
603 outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
604 outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
605 outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
606 outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
609 if (DO_BIC(BIC_Pkgpc2))
610 outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
611 if (DO_BIC(BIC_Pkgpc3))
612 outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
613 if (DO_BIC(BIC_Pkgpc6))
614 outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
615 if (DO_BIC(BIC_Pkgpc7))
616 outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
617 if (DO_BIC(BIC_Pkgpc8))
618 outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
619 if (DO_BIC(BIC_Pkgpc9))
620 outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
621 if (DO_BIC(BIC_Pkgpc10))
622 outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
624 if (do_rapl && !rapl_joules) {
625 if (DO_BIC(BIC_PkgWatt))
626 outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
627 if (DO_BIC(BIC_CorWatt))
628 outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
629 if (DO_BIC(BIC_GFXWatt))
630 outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
631 if (DO_BIC(BIC_RAMWatt))
632 outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
633 if (DO_BIC(BIC_PKG__))
634 outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
635 if (DO_BIC(BIC_RAM__))
636 outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
637 } else if (do_rapl && rapl_joules) {
638 if (DO_BIC(BIC_Pkg_J))
639 outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
640 if (DO_BIC(BIC_Cor_J))
641 outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
642 if (DO_BIC(BIC_GFX_J))
643 outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
644 if (DO_BIC(BIC_RAM_J))
645 outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
646 if (DO_BIC(BIC_PKG__))
647 outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
648 if (DO_BIC(BIC_RAM__))
649 outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
651 for (mp = sys.pp; mp; mp = mp->next) {
652 if (mp->format == FORMAT_RAW) {
654 outp += sprintf(outp, "%s%18.18s", delim, mp->name);
656 outp += sprintf(outp, "%s%10.10s", delim, mp->name);
658 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
659 outp += sprintf(outp, "%s%8s", delim, mp->name);
661 outp += sprintf(outp, "%s%s", delim, mp->name);
665 outp += sprintf(outp, "\n");
668 int dump_counters(struct thread_data *t, struct core_data *c,
672 struct msr_counter *mp;
674 outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
677 outp += sprintf(outp, "CPU: %d flags 0x%x\n",
678 t->cpu_id, t->flags);
679 outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
680 outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
681 outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
682 outp += sprintf(outp, "c1: %016llX\n", t->c1);
685 outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
687 outp += sprintf(outp, "SMI: %d\n", t->smi_count);
689 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
690 outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
691 i, mp->msr_num, t->counter[i]);
696 outp += sprintf(outp, "core: %d\n", c->core_id);
697 outp += sprintf(outp, "c3: %016llX\n", c->c3);
698 outp += sprintf(outp, "c6: %016llX\n", c->c6);
699 outp += sprintf(outp, "c7: %016llX\n", c->c7);
700 outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
702 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
703 outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
704 i, mp->msr_num, c->counter[i]);
706 outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
710 outp += sprintf(outp, "package: %d\n", p->package_id);
712 outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
713 outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
714 outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
715 outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
717 outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
718 if (DO_BIC(BIC_Pkgpc3))
719 outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
720 if (DO_BIC(BIC_Pkgpc6))
721 outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
722 if (DO_BIC(BIC_Pkgpc7))
723 outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
724 outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
725 outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
726 outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
727 outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
728 outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
729 outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
730 outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
731 outp += sprintf(outp, "Throttle PKG: %0X\n",
732 p->rapl_pkg_perf_status);
733 outp += sprintf(outp, "Throttle RAM: %0X\n",
734 p->rapl_dram_perf_status);
735 outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
737 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
738 outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
739 i, mp->msr_num, p->counter[i]);
743 outp += sprintf(outp, "\n");
749 * column formatting convention & formats
751 int format_counters(struct thread_data *t, struct core_data *c,
754 double interval_float, tsc;
757 struct msr_counter *mp;
761 /* if showing only 1st thread in core and this isn't one, bail out */
762 if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
765 /* if showing only 1st thread in pkg and this isn't one, bail out */
766 if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
769 /*if not summary line and --cpu is used */
770 if ((t != &average.threads) &&
771 (cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
774 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
776 tsc = t->tsc * tsc_tweak;
778 /* topo columns, print blanks on 1st (average) line */
779 if (t == &average.threads) {
780 if (DO_BIC(BIC_Package))
781 outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
782 if (DO_BIC(BIC_Core))
783 outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
785 outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
787 if (DO_BIC(BIC_Package)) {
789 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
791 outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
793 if (DO_BIC(BIC_Core)) {
795 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
797 outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
800 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
803 if (DO_BIC(BIC_Avg_MHz))
804 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
805 1.0 / units * t->aperf / interval_float);
807 if (DO_BIC(BIC_Busy))
808 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf/tsc);
810 if (DO_BIC(BIC_Bzy_MHz)) {
812 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
814 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
815 tsc / units * t->aperf / t->mperf / interval_float);
818 if (DO_BIC(BIC_TSC_MHz))
819 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
822 if (DO_BIC(BIC_IRQ)) {
823 if (sums_need_wide_columns)
824 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
826 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
831 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
834 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
835 if (mp->format == FORMAT_RAW) {
837 outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) t->counter[i]);
839 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
840 } else if (mp->format == FORMAT_DELTA) {
841 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
842 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
844 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
845 } else if (mp->format == FORMAT_PERCENT) {
846 if (mp->type == COUNTER_USEC)
847 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), t->counter[i]/interval_float/10000);
849 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i]/tsc);
854 if (DO_BIC(BIC_CPU_c1))
855 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1/tsc);
858 /* print per-core data only for 1st thread in core */
859 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
862 if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates)
863 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3/tsc);
864 if (DO_BIC(BIC_CPU_c6))
865 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6/tsc);
866 if (DO_BIC(BIC_CPU_c7))
867 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7/tsc);
870 if (DO_BIC(BIC_Mod_c6))
871 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
873 if (DO_BIC(BIC_CoreTmp))
874 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
876 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
877 if (mp->format == FORMAT_RAW) {
879 outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) c->counter[i]);
881 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
882 } else if (mp->format == FORMAT_DELTA) {
883 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
884 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
886 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
887 } else if (mp->format == FORMAT_PERCENT) {
888 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i]/tsc);
892 /* print per-package data only for 1st core in package */
893 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
897 if (DO_BIC(BIC_PkgTmp))
898 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
901 if (DO_BIC(BIC_GFX_rc6)) {
902 if (p->gfx_rc6_ms == -1) { /* detect GFX counter reset */
903 outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
905 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
906 p->gfx_rc6_ms / 10.0 / interval_float);
911 if (DO_BIC(BIC_GFXMHz))
912 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
914 /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
915 if (do_skl_residency) {
916 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0/tsc);
917 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0/tsc);
918 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0/tsc);
919 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0/tsc);
922 if (DO_BIC(BIC_Pkgpc2))
923 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2/tsc);
924 if (DO_BIC(BIC_Pkgpc3))
925 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3/tsc);
926 if (DO_BIC(BIC_Pkgpc6))
927 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6/tsc);
928 if (DO_BIC(BIC_Pkgpc7))
929 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7/tsc);
930 if (DO_BIC(BIC_Pkgpc8))
931 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8/tsc);
932 if (DO_BIC(BIC_Pkgpc9))
933 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9/tsc);
934 if (DO_BIC(BIC_Pkgpc10))
935 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10/tsc);
938 * If measurement interval exceeds minimum RAPL Joule Counter range,
939 * indicate that results are suspect by printing "**" in fraction place.
941 if (interval_float < rapl_joule_counter_range)
946 if (DO_BIC(BIC_PkgWatt))
947 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
948 if (DO_BIC(BIC_CorWatt))
949 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
950 if (DO_BIC(BIC_GFXWatt))
951 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
952 if (DO_BIC(BIC_RAMWatt))
953 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units / interval_float);
954 if (DO_BIC(BIC_Pkg_J))
955 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
956 if (DO_BIC(BIC_Cor_J))
957 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
958 if (DO_BIC(BIC_GFX_J))
959 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
960 if (DO_BIC(BIC_RAM_J))
961 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
962 if (DO_BIC(BIC_PKG__))
963 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
964 if (DO_BIC(BIC_RAM__))
965 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
967 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
968 if (mp->format == FORMAT_RAW) {
970 outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) p->counter[i]);
972 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
973 } else if (mp->format == FORMAT_DELTA) {
974 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
975 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
977 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
978 } else if (mp->format == FORMAT_PERCENT) {
979 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i]/tsc);
984 outp += sprintf(outp, "\n");
989 void flush_output_stdout(void)
998 fputs(output_buffer, filep);
1001 outp = output_buffer;
1003 void flush_output_stderr(void)
1005 fputs(output_buffer, outf);
1007 outp = output_buffer;
1009 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1013 if (!printed || !summary_only)
1016 if (topo.num_cpus > 1)
1017 format_counters(&average.threads, &average.cores,
1025 for_all_cpus(format_counters, t, c, p);
1028 #define DELTA_WRAP32(new, old) \
1032 old = 0x100000000 + new - old; \
1036 delta_package(struct pkg_data *new, struct pkg_data *old)
1039 struct msr_counter *mp;
1041 if (do_skl_residency) {
1042 old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1043 old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1044 old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1045 old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1047 old->pc2 = new->pc2 - old->pc2;
1048 if (DO_BIC(BIC_Pkgpc3))
1049 old->pc3 = new->pc3 - old->pc3;
1050 if (DO_BIC(BIC_Pkgpc6))
1051 old->pc6 = new->pc6 - old->pc6;
1052 if (DO_BIC(BIC_Pkgpc7))
1053 old->pc7 = new->pc7 - old->pc7;
1054 old->pc8 = new->pc8 - old->pc8;
1055 old->pc9 = new->pc9 - old->pc9;
1056 old->pc10 = new->pc10 - old->pc10;
1057 old->pkg_temp_c = new->pkg_temp_c;
1059 /* flag an error when rc6 counter resets/wraps */
1060 if (old->gfx_rc6_ms > new->gfx_rc6_ms)
1061 old->gfx_rc6_ms = -1;
1063 old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1065 old->gfx_mhz = new->gfx_mhz;
1067 DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
1068 DELTA_WRAP32(new->energy_cores, old->energy_cores);
1069 DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
1070 DELTA_WRAP32(new->energy_dram, old->energy_dram);
1071 DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
1072 DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
1074 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1075 if (mp->format == FORMAT_RAW)
1076 old->counter[i] = new->counter[i];
1078 old->counter[i] = new->counter[i] - old->counter[i];
1085 delta_core(struct core_data *new, struct core_data *old)
1088 struct msr_counter *mp;
1090 old->c3 = new->c3 - old->c3;
1091 old->c6 = new->c6 - old->c6;
1092 old->c7 = new->c7 - old->c7;
1093 old->core_temp_c = new->core_temp_c;
1094 old->mc6_us = new->mc6_us - old->mc6_us;
1096 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1097 if (mp->format == FORMAT_RAW)
1098 old->counter[i] = new->counter[i];
1100 old->counter[i] = new->counter[i] - old->counter[i];
1108 delta_thread(struct thread_data *new, struct thread_data *old,
1109 struct core_data *core_delta)
1112 struct msr_counter *mp;
1114 old->tsc = new->tsc - old->tsc;
1116 /* check for TSC < 1 Mcycles over interval */
1117 if (old->tsc < (1000 * 1000))
1118 errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1119 "You can disable all c-states by booting with \"idle=poll\"\n"
1120 "or just the deep ones with \"processor.max_cstate=1\"");
1122 old->c1 = new->c1 - old->c1;
1124 if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1125 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1126 old->aperf = new->aperf - old->aperf;
1127 old->mperf = new->mperf - old->mperf;
1134 if (use_c1_residency_msr) {
1136 * Some models have a dedicated C1 residency MSR,
1137 * which should be more accurate than the derivation below.
1141 * As counter collection is not atomic,
1142 * it is possible for mperf's non-halted cycles + idle states
1143 * to exceed TSC's all cycles: show c1 = 0% in that case.
1145 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1148 /* normal case, derive c1 */
1149 old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1150 - core_delta->c6 - core_delta->c7;
1154 if (old->mperf == 0) {
1156 fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1157 old->mperf = 1; /* divide by 0 protection */
1160 if (DO_BIC(BIC_IRQ))
1161 old->irq_count = new->irq_count - old->irq_count;
1163 if (DO_BIC(BIC_SMI))
1164 old->smi_count = new->smi_count - old->smi_count;
1166 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1167 if (mp->format == FORMAT_RAW)
1168 old->counter[i] = new->counter[i];
1170 old->counter[i] = new->counter[i] - old->counter[i];
1175 int delta_cpu(struct thread_data *t, struct core_data *c,
1176 struct pkg_data *p, struct thread_data *t2,
1177 struct core_data *c2, struct pkg_data *p2)
1181 /* calculate core delta only for 1st thread in core */
1182 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1185 /* always calculate thread delta */
1186 retval = delta_thread(t, t2, c2); /* c2 is core delta */
1190 /* calculate package delta only for 1st core in package */
1191 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1192 retval = delta_package(p, p2);
1197 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1200 struct msr_counter *mp;
1210 /* tells format_counters to dump all fields from this set */
1211 t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1219 p->pkg_wtd_core_c0 = 0;
1220 p->pkg_any_core_c0 = 0;
1221 p->pkg_any_gfxe_c0 = 0;
1222 p->pkg_both_core_gfxe_c0 = 0;
1225 if (DO_BIC(BIC_Pkgpc3))
1227 if (DO_BIC(BIC_Pkgpc6))
1229 if (DO_BIC(BIC_Pkgpc7))
1237 p->energy_cores = 0;
1239 p->rapl_pkg_perf_status = 0;
1240 p->rapl_dram_perf_status = 0;
1245 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1248 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1251 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1254 int sum_counters(struct thread_data *t, struct core_data *c,
1258 struct msr_counter *mp;
1260 average.threads.tsc += t->tsc;
1261 average.threads.aperf += t->aperf;
1262 average.threads.mperf += t->mperf;
1263 average.threads.c1 += t->c1;
1265 average.threads.irq_count += t->irq_count;
1266 average.threads.smi_count += t->smi_count;
1268 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1269 if (mp->format == FORMAT_RAW)
1271 average.threads.counter[i] += t->counter[i];
1274 /* sum per-core values only for 1st thread in core */
1275 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1278 average.cores.c3 += c->c3;
1279 average.cores.c6 += c->c6;
1280 average.cores.c7 += c->c7;
1281 average.cores.mc6_us += c->mc6_us;
1283 average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1285 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1286 if (mp->format == FORMAT_RAW)
1288 average.cores.counter[i] += c->counter[i];
1291 /* sum per-pkg values only for 1st core in pkg */
1292 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1295 if (do_skl_residency) {
1296 average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1297 average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1298 average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1299 average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1302 average.packages.pc2 += p->pc2;
1303 if (DO_BIC(BIC_Pkgpc3))
1304 average.packages.pc3 += p->pc3;
1305 if (DO_BIC(BIC_Pkgpc6))
1306 average.packages.pc6 += p->pc6;
1307 if (DO_BIC(BIC_Pkgpc7))
1308 average.packages.pc7 += p->pc7;
1309 average.packages.pc8 += p->pc8;
1310 average.packages.pc9 += p->pc9;
1311 average.packages.pc10 += p->pc10;
1313 average.packages.energy_pkg += p->energy_pkg;
1314 average.packages.energy_dram += p->energy_dram;
1315 average.packages.energy_cores += p->energy_cores;
1316 average.packages.energy_gfx += p->energy_gfx;
1318 average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1319 average.packages.gfx_mhz = p->gfx_mhz;
1321 average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1323 average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1324 average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1326 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1327 if (mp->format == FORMAT_RAW)
1329 average.packages.counter[i] += p->counter[i];
1334 * sum the counters for all cpus in the system
1335 * compute the weighted average
1337 void compute_average(struct thread_data *t, struct core_data *c,
1341 struct msr_counter *mp;
1343 clear_counters(&average.threads, &average.cores, &average.packages);
1345 for_all_cpus(sum_counters, t, c, p);
1347 average.threads.tsc /= topo.num_cpus;
1348 average.threads.aperf /= topo.num_cpus;
1349 average.threads.mperf /= topo.num_cpus;
1350 average.threads.c1 /= topo.num_cpus;
1352 if (average.threads.irq_count > 9999999)
1353 sums_need_wide_columns = 1;
1355 average.cores.c3 /= topo.num_cores;
1356 average.cores.c6 /= topo.num_cores;
1357 average.cores.c7 /= topo.num_cores;
1358 average.cores.mc6_us /= topo.num_cores;
1360 if (do_skl_residency) {
1361 average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1362 average.packages.pkg_any_core_c0 /= topo.num_packages;
1363 average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1364 average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1367 average.packages.pc2 /= topo.num_packages;
1368 if (DO_BIC(BIC_Pkgpc3))
1369 average.packages.pc3 /= topo.num_packages;
1370 if (DO_BIC(BIC_Pkgpc6))
1371 average.packages.pc6 /= topo.num_packages;
1372 if (DO_BIC(BIC_Pkgpc7))
1373 average.packages.pc7 /= topo.num_packages;
1375 average.packages.pc8 /= topo.num_packages;
1376 average.packages.pc9 /= topo.num_packages;
1377 average.packages.pc10 /= topo.num_packages;
1379 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1380 if (mp->format == FORMAT_RAW)
1382 if (mp->type == COUNTER_ITEMS) {
1383 if (average.threads.counter[i] > 9999999)
1384 sums_need_wide_columns = 1;
1387 average.threads.counter[i] /= topo.num_cpus;
1389 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1390 if (mp->format == FORMAT_RAW)
1392 if (mp->type == COUNTER_ITEMS) {
1393 if (average.cores.counter[i] > 9999999)
1394 sums_need_wide_columns = 1;
1396 average.cores.counter[i] /= topo.num_cores;
1398 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1399 if (mp->format == FORMAT_RAW)
1401 if (mp->type == COUNTER_ITEMS) {
1402 if (average.packages.counter[i] > 9999999)
1403 sums_need_wide_columns = 1;
1405 average.packages.counter[i] /= topo.num_packages;
1409 static unsigned long long rdtsc(void)
1411 unsigned int low, high;
1413 asm volatile("rdtsc" : "=a" (low), "=d" (high));
1415 return low | ((unsigned long long)high) << 32;
1419 * Open a file, and exit on failure
1421 FILE *fopen_or_die(const char *path, const char *mode)
1423 FILE *filep = fopen(path, mode);
1426 err(1, "%s: open failed", path);
1430 * snapshot_sysfs_counter()
1432 * return snapshot of given counter
1434 unsigned long long snapshot_sysfs_counter(char *path)
1438 unsigned long long counter;
1440 fp = fopen_or_die(path, "r");
1442 retval = fscanf(fp, "%lld", &counter);
1444 err(1, "snapshot_sysfs_counter(%s)", path);
1451 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1453 if (mp->msr_num != 0) {
1454 if (get_msr(cpu, mp->msr_num, counterp))
1459 if (mp->flags & SYSFS_PERCPU) {
1460 sprintf(path, "/sys/devices/system/cpu/cpu%d/%s",
1463 *counterp = snapshot_sysfs_counter(path);
1465 *counterp = snapshot_sysfs_counter(mp->path);
1475 * acquire and record local counters for that cpu
1477 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1479 int cpu = t->cpu_id;
1480 unsigned long long msr;
1481 int aperf_mperf_retry_count = 0;
1482 struct msr_counter *mp;
1485 if (cpu_migrate(cpu)) {
1486 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1491 t->tsc = rdtsc(); /* we are running on local CPU of interest */
1493 if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1494 unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1497 * The TSC, APERF and MPERF must be read together for
1498 * APERF/MPERF and MPERF/TSC to give accurate results.
1500 * Unfortunately, APERF and MPERF are read by
1501 * individual system call, so delays may occur
1502 * between them. If the time to read them
1503 * varies by a large amount, we re-read them.
1507 * This initial dummy APERF read has been seen to
1508 * reduce jitter in the subsequent reads.
1511 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1514 t->tsc = rdtsc(); /* re-read close to APERF */
1516 tsc_before = t->tsc;
1518 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1521 tsc_between = rdtsc();
1523 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1526 tsc_after = rdtsc();
1528 aperf_time = tsc_between - tsc_before;
1529 mperf_time = tsc_after - tsc_between;
1532 * If the system call latency to read APERF and MPERF
1533 * differ by more than 2x, then try again.
1535 if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1536 aperf_mperf_retry_count++;
1537 if (aperf_mperf_retry_count < 5)
1540 warnx("cpu%d jitter %lld %lld",
1541 cpu, aperf_time, mperf_time);
1543 aperf_mperf_retry_count = 0;
1545 t->aperf = t->aperf * aperf_mperf_multiplier;
1546 t->mperf = t->mperf * aperf_mperf_multiplier;
1549 if (DO_BIC(BIC_IRQ))
1550 t->irq_count = irqs_per_cpu[cpu];
1551 if (DO_BIC(BIC_SMI)) {
1552 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1554 t->smi_count = msr & 0xFFFFFFFF;
1556 if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1557 if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1561 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1562 if (get_mp(cpu, mp, &t->counter[i]))
1566 /* collect core counters only for 1st thread in core */
1567 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1570 if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates) {
1571 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1575 if (DO_BIC(BIC_CPU_c6) && !do_knl_cstates) {
1576 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1578 } else if (do_knl_cstates) {
1579 if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1583 if (DO_BIC(BIC_CPU_c7))
1584 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1587 if (DO_BIC(BIC_Mod_c6))
1588 if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1591 if (DO_BIC(BIC_CoreTmp)) {
1592 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1594 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1597 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1598 if (get_mp(cpu, mp, &c->counter[i]))
1602 /* collect package counters only for 1st core in package */
1603 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1606 if (do_skl_residency) {
1607 if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1609 if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1611 if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1613 if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1616 if (DO_BIC(BIC_Pkgpc3))
1617 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1619 if (DO_BIC(BIC_Pkgpc6)) {
1620 if (do_slm_cstates) {
1621 if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1624 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1629 if (DO_BIC(BIC_Pkgpc2))
1630 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1632 if (DO_BIC(BIC_Pkgpc7))
1633 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1635 if (DO_BIC(BIC_Pkgpc8))
1636 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1638 if (DO_BIC(BIC_Pkgpc9))
1639 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1641 if (DO_BIC(BIC_Pkgpc10))
1642 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1645 if (do_rapl & RAPL_PKG) {
1646 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1648 p->energy_pkg = msr & 0xFFFFFFFF;
1650 if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1651 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1653 p->energy_cores = msr & 0xFFFFFFFF;
1655 if (do_rapl & RAPL_DRAM) {
1656 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1658 p->energy_dram = msr & 0xFFFFFFFF;
1660 if (do_rapl & RAPL_GFX) {
1661 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1663 p->energy_gfx = msr & 0xFFFFFFFF;
1665 if (do_rapl & RAPL_PKG_PERF_STATUS) {
1666 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1668 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1670 if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1671 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1673 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1675 if (DO_BIC(BIC_PkgTmp)) {
1676 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1678 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1681 if (DO_BIC(BIC_GFX_rc6))
1682 p->gfx_rc6_ms = gfx_cur_rc6_ms;
1684 if (DO_BIC(BIC_GFXMHz))
1685 p->gfx_mhz = gfx_cur_mhz;
1687 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1688 if (get_mp(cpu, mp, &p->counter[i]))
1696 * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
1697 * If you change the values, note they are used both in comparisons
1698 * (>= PCL__7) and to index pkg_cstate_limit_strings[].
1701 #define PCLUKN 0 /* Unknown */
1702 #define PCLRSV 1 /* Reserved */
1703 #define PCL__0 2 /* PC0 */
1704 #define PCL__1 3 /* PC1 */
1705 #define PCL__2 4 /* PC2 */
1706 #define PCL__3 5 /* PC3 */
1707 #define PCL__4 6 /* PC4 */
1708 #define PCL__6 7 /* PC6 */
1709 #define PCL_6N 8 /* PC6 No Retention */
1710 #define PCL_6R 9 /* PC6 Retention */
1711 #define PCL__7 10 /* PC7 */
1712 #define PCL_7S 11 /* PC7 Shrink */
1713 #define PCL__8 12 /* PC8 */
1714 #define PCL__9 13 /* PC9 */
1715 #define PCLUNL 14 /* Unlimited */
1717 int pkg_cstate_limit = PCLUKN;
1718 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
1719 "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"};
1721 int nhm_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1722 int snb_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1723 int hsw_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1724 int slv_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7};
1725 int amt_pkg_cstate_limits[16] = {PCLUNL, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1726 int phi_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1727 int bxt_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1728 int skx_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1732 calculate_tsc_tweak()
1734 tsc_tweak = base_hz / tsc_hz;
1738 dump_nhm_platform_info(void)
1740 unsigned long long msr;
1743 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
1745 fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
1747 ratio = (msr >> 40) & 0xFF;
1748 fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
1749 ratio, bclk, ratio * bclk);
1751 ratio = (msr >> 8) & 0xFF;
1752 fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
1753 ratio, bclk, ratio * bclk);
1755 get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
1756 fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
1757 base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
1763 dump_hsw_turbo_ratio_limits(void)
1765 unsigned long long msr;
1768 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
1770 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
1772 ratio = (msr >> 8) & 0xFF;
1774 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
1775 ratio, bclk, ratio * bclk);
1777 ratio = (msr >> 0) & 0xFF;
1779 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
1780 ratio, bclk, ratio * bclk);
1785 dump_ivt_turbo_ratio_limits(void)
1787 unsigned long long msr;
1790 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
1792 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
1794 ratio = (msr >> 56) & 0xFF;
1796 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
1797 ratio, bclk, ratio * bclk);
1799 ratio = (msr >> 48) & 0xFF;
1801 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
1802 ratio, bclk, ratio * bclk);
1804 ratio = (msr >> 40) & 0xFF;
1806 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
1807 ratio, bclk, ratio * bclk);
1809 ratio = (msr >> 32) & 0xFF;
1811 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
1812 ratio, bclk, ratio * bclk);
1814 ratio = (msr >> 24) & 0xFF;
1816 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
1817 ratio, bclk, ratio * bclk);
1819 ratio = (msr >> 16) & 0xFF;
1821 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
1822 ratio, bclk, ratio * bclk);
1824 ratio = (msr >> 8) & 0xFF;
1826 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
1827 ratio, bclk, ratio * bclk);
1829 ratio = (msr >> 0) & 0xFF;
1831 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
1832 ratio, bclk, ratio * bclk);
1835 int has_turbo_ratio_group_limits(int family, int model)
1842 case INTEL_FAM6_ATOM_GOLDMONT:
1843 case INTEL_FAM6_SKYLAKE_X:
1844 case INTEL_FAM6_ATOM_DENVERTON:
1851 dump_turbo_ratio_limits(int family, int model)
1853 unsigned long long msr, core_counts;
1854 unsigned int ratio, group_size;
1856 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1857 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
1859 if (has_turbo_ratio_group_limits(family, model)) {
1860 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
1861 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
1863 core_counts = 0x0807060504030201;
1866 ratio = (msr >> 56) & 0xFF;
1867 group_size = (core_counts >> 56) & 0xFF;
1869 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1870 ratio, bclk, ratio * bclk, group_size);
1872 ratio = (msr >> 48) & 0xFF;
1873 group_size = (core_counts >> 48) & 0xFF;
1875 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1876 ratio, bclk, ratio * bclk, group_size);
1878 ratio = (msr >> 40) & 0xFF;
1879 group_size = (core_counts >> 40) & 0xFF;
1881 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1882 ratio, bclk, ratio * bclk, group_size);
1884 ratio = (msr >> 32) & 0xFF;
1885 group_size = (core_counts >> 32) & 0xFF;
1887 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1888 ratio, bclk, ratio * bclk, group_size);
1890 ratio = (msr >> 24) & 0xFF;
1891 group_size = (core_counts >> 24) & 0xFF;
1893 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1894 ratio, bclk, ratio * bclk, group_size);
1896 ratio = (msr >> 16) & 0xFF;
1897 group_size = (core_counts >> 16) & 0xFF;
1899 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1900 ratio, bclk, ratio * bclk, group_size);
1902 ratio = (msr >> 8) & 0xFF;
1903 group_size = (core_counts >> 8) & 0xFF;
1905 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1906 ratio, bclk, ratio * bclk, group_size);
1908 ratio = (msr >> 0) & 0xFF;
1909 group_size = (core_counts >> 0) & 0xFF;
1911 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1912 ratio, bclk, ratio * bclk, group_size);
1917 dump_atom_turbo_ratio_limits(void)
1919 unsigned long long msr;
1922 get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
1923 fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
1925 ratio = (msr >> 0) & 0x3F;
1927 fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
1928 ratio, bclk, ratio * bclk);
1930 ratio = (msr >> 8) & 0x3F;
1932 fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
1933 ratio, bclk, ratio * bclk);
1935 ratio = (msr >> 16) & 0x3F;
1937 fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
1938 ratio, bclk, ratio * bclk);
1940 get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
1941 fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
1943 ratio = (msr >> 24) & 0x3F;
1945 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
1946 ratio, bclk, ratio * bclk);
1948 ratio = (msr >> 16) & 0x3F;
1950 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
1951 ratio, bclk, ratio * bclk);
1953 ratio = (msr >> 8) & 0x3F;
1955 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
1956 ratio, bclk, ratio * bclk);
1958 ratio = (msr >> 0) & 0x3F;
1960 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
1961 ratio, bclk, ratio * bclk);
1965 dump_knl_turbo_ratio_limits(void)
1967 const unsigned int buckets_no = 7;
1969 unsigned long long msr;
1970 int delta_cores, delta_ratio;
1972 unsigned int cores[buckets_no];
1973 unsigned int ratio[buckets_no];
1975 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1977 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
1981 * Turbo encoding in KNL is as follows:
1983 * [7:1] -- Base value of number of active cores of bucket 1.
1984 * [15:8] -- Base value of freq ratio of bucket 1.
1985 * [20:16] -- +ve delta of number of active cores of bucket 2.
1986 * i.e. active cores of bucket 2 =
1987 * active cores of bucket 1 + delta
1988 * [23:21] -- Negative delta of freq ratio of bucket 2.
1989 * i.e. freq ratio of bucket 2 =
1990 * freq ratio of bucket 1 - delta
1991 * [28:24]-- +ve delta of number of active cores of bucket 3.
1992 * [31:29]-- -ve delta of freq ratio of bucket 3.
1993 * [36:32]-- +ve delta of number of active cores of bucket 4.
1994 * [39:37]-- -ve delta of freq ratio of bucket 4.
1995 * [44:40]-- +ve delta of number of active cores of bucket 5.
1996 * [47:45]-- -ve delta of freq ratio of bucket 5.
1997 * [52:48]-- +ve delta of number of active cores of bucket 6.
1998 * [55:53]-- -ve delta of freq ratio of bucket 6.
1999 * [60:56]-- +ve delta of number of active cores of bucket 7.
2000 * [63:61]-- -ve delta of freq ratio of bucket 7.
2004 cores[b_nr] = (msr & 0xFF) >> 1;
2005 ratio[b_nr] = (msr >> 8) & 0xFF;
2007 for (i = 16; i < 64; i += 8) {
2008 delta_cores = (msr >> i) & 0x1F;
2009 delta_ratio = (msr >> (i + 5)) & 0x7;
2011 cores[b_nr + 1] = cores[b_nr] + delta_cores;
2012 ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2016 for (i = buckets_no - 1; i >= 0; i--)
2017 if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2019 "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2020 ratio[i], bclk, ratio[i] * bclk, cores[i]);
2024 dump_nhm_cst_cfg(void)
2026 unsigned long long msr;
2028 get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2030 #define SNB_C1_AUTO_UNDEMOTE (1UL << 27)
2031 #define SNB_C3_AUTO_UNDEMOTE (1UL << 28)
2033 fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2035 fprintf(outf, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n",
2036 (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2037 (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2038 (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2039 (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2040 (msr & (1 << 15)) ? "" : "UN",
2041 (unsigned int)msr & 0xF,
2042 pkg_cstate_limit_strings[pkg_cstate_limit]);
2047 dump_config_tdp(void)
2049 unsigned long long msr;
2051 get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2052 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2053 fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2055 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2056 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2058 fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2059 fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2060 fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2061 fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2063 fprintf(outf, ")\n");
2065 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2066 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2068 fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2069 fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2070 fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2071 fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2073 fprintf(outf, ")\n");
2075 get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2076 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2078 fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2079 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2080 fprintf(outf, ")\n");
2082 get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2083 fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2084 fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2085 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2086 fprintf(outf, ")\n");
2089 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2091 void print_irtl(void)
2093 unsigned long long msr;
2095 get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2096 fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2097 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2098 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2100 get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2101 fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2102 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2103 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2105 get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2106 fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2107 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2108 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2113 get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2114 fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2115 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2116 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2118 get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2119 fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2120 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2121 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2123 get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2124 fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2125 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2126 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2129 void free_fd_percpu(void)
2133 for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2134 if (fd_percpu[i] != 0)
2135 close(fd_percpu[i]);
2141 void free_all_buffers(void)
2143 CPU_FREE(cpu_present_set);
2144 cpu_present_set = NULL;
2145 cpu_present_setsize = 0;
2147 CPU_FREE(cpu_affinity_set);
2148 cpu_affinity_set = NULL;
2149 cpu_affinity_setsize = 0;
2157 package_even = NULL;
2167 free(output_buffer);
2168 output_buffer = NULL;
2173 free(irq_column_2_cpu);
2179 * Parse a file containing a single int.
2181 int parse_int_file(const char *fmt, ...)
2184 char path[PATH_MAX];
2188 va_start(args, fmt);
2189 vsnprintf(path, sizeof(path), fmt, args);
2191 filep = fopen_or_die(path, "r");
2192 if (fscanf(filep, "%d", &value) != 1)
2193 err(1, "%s: failed to parse number from file", path);
2199 * get_cpu_position_in_core(cpu)
2200 * return the position of the CPU among its HT siblings in the core
2201 * return -1 if the sibling is not in list
2203 int get_cpu_position_in_core(int cpu)
2212 "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list",
2214 filep = fopen(path, "r");
2215 if (filep == NULL) {
2220 for (i = 0; i < topo.num_threads_per_core; i++) {
2221 fscanf(filep, "%d", &this_cpu);
2222 if (this_cpu == cpu) {
2227 /* Account for no separator after last thread*/
2228 if (i != (topo.num_threads_per_core - 1))
2229 fscanf(filep, "%c", &character);
2237 * cpu_is_first_core_in_package(cpu)
2238 * return 1 if given CPU is 1st core in package
2240 int cpu_is_first_core_in_package(int cpu)
2242 return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2245 int get_physical_package_id(int cpu)
2247 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2250 int get_core_id(int cpu)
2252 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2255 int get_num_ht_siblings(int cpu)
2265 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
2266 filep = fopen_or_die(path, "r");
2270 * A ',' separated or '-' separated set of numbers
2271 * (eg 1-2 or 1,3,4,5)
2273 fscanf(filep, "%d%c\n", &sib1, &character);
2274 fseek(filep, 0, SEEK_SET);
2275 fgets(str, 100, filep);
2276 ch = strchr(str, character);
2277 while (ch != NULL) {
2279 ch = strchr(ch+1, character);
2287 * run func(thread, core, package) in topology order
2288 * skip non-present cpus
2291 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2292 struct pkg_data *, struct thread_data *, struct core_data *,
2293 struct pkg_data *), struct thread_data *thread_base,
2294 struct core_data *core_base, struct pkg_data *pkg_base,
2295 struct thread_data *thread_base2, struct core_data *core_base2,
2296 struct pkg_data *pkg_base2)
2298 int retval, pkg_no, core_no, thread_no;
2300 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2301 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
2302 for (thread_no = 0; thread_no <
2303 topo.num_threads_per_core; ++thread_no) {
2304 struct thread_data *t, *t2;
2305 struct core_data *c, *c2;
2306 struct pkg_data *p, *p2;
2308 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
2310 if (cpu_is_not_present(t->cpu_id))
2313 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
2315 c = GET_CORE(core_base, core_no, pkg_no);
2316 c2 = GET_CORE(core_base2, core_no, pkg_no);
2318 p = GET_PKG(pkg_base, pkg_no);
2319 p2 = GET_PKG(pkg_base2, pkg_no);
2321 retval = func(t, c, p, t2, c2, p2);
2331 * run func(cpu) on every cpu in /proc/stat
2332 * return max_cpu number
2334 int for_all_proc_cpus(int (func)(int))
2340 fp = fopen_or_die(proc_stat, "r");
2342 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2344 err(1, "%s: failed to parse format", proc_stat);
2347 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2351 retval = func(cpu_num);
2361 void re_initialize(void)
2364 setup_all_buffers();
2365 printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2371 * remember the last one seen, it will be the max
2373 int count_cpus(int cpu)
2375 if (topo.max_cpu_num < cpu)
2376 topo.max_cpu_num = cpu;
2381 int mark_cpu_present(int cpu)
2383 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2388 * snapshot_proc_interrupts()
2390 * read and record summary of /proc/interrupts
2392 * return 1 if config change requires a restart, else return 0
2394 int snapshot_proc_interrupts(void)
2400 fp = fopen_or_die("/proc/interrupts", "r");
2404 /* read 1st line of /proc/interrupts to get cpu* name for each column */
2405 for (column = 0; column < topo.num_cpus; ++column) {
2408 retval = fscanf(fp, " CPU%d", &cpu_number);
2412 if (cpu_number > topo.max_cpu_num) {
2413 warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2417 irq_column_2_cpu[column] = cpu_number;
2418 irqs_per_cpu[cpu_number] = 0;
2421 /* read /proc/interrupt count lines and sum up irqs per cpu */
2426 retval = fscanf(fp, " %s:", buf); /* flush irq# "N:" */
2430 /* read the count per cpu */
2431 for (column = 0; column < topo.num_cpus; ++column) {
2433 int cpu_number, irq_count;
2435 retval = fscanf(fp, " %d", &irq_count);
2439 cpu_number = irq_column_2_cpu[column];
2440 irqs_per_cpu[cpu_number] += irq_count;
2444 while (getc(fp) != '\n')
2445 ; /* flush interrupt description */
2451 * snapshot_gfx_rc6_ms()
2453 * record snapshot of
2454 * /sys/class/drm/card0/power/rc6_residency_ms
2456 * return 1 if config change requires a restart, else return 0
2458 int snapshot_gfx_rc6_ms(void)
2463 fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2465 retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2474 * snapshot_gfx_mhz()
2476 * record snapshot of
2477 * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2479 * return 1 if config change requires a restart, else return 0
2481 int snapshot_gfx_mhz(void)
2487 fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2493 retval = fscanf(fp, "%d", &gfx_cur_mhz);
2501 * snapshot /proc and /sys files
2503 * return 1 if configuration restart needed, else return 0
2505 int snapshot_proc_sysfs_files(void)
2507 if (DO_BIC(BIC_IRQ))
2508 if (snapshot_proc_interrupts())
2511 if (DO_BIC(BIC_GFX_rc6))
2512 snapshot_gfx_rc6_ms();
2514 if (DO_BIC(BIC_GFXMHz))
2520 void turbostat_loop()
2528 snapshot_proc_sysfs_files();
2529 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2532 } else if (retval == -1) {
2533 if (restarted > 1) {
2540 gettimeofday(&tv_even, (struct timezone *)NULL);
2543 if (for_all_proc_cpus(cpu_is_not_present)) {
2547 nanosleep(&interval_ts, NULL);
2548 if (snapshot_proc_sysfs_files())
2550 retval = for_all_cpus(get_counters, ODD_COUNTERS);
2553 } else if (retval == -1) {
2557 gettimeofday(&tv_odd, (struct timezone *)NULL);
2558 timersub(&tv_odd, &tv_even, &tv_delta);
2559 if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
2563 compute_average(EVEN_COUNTERS);
2564 format_all_counters(EVEN_COUNTERS);
2565 flush_output_stdout();
2566 nanosleep(&interval_ts, NULL);
2567 if (snapshot_proc_sysfs_files())
2569 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2572 } else if (retval == -1) {
2576 gettimeofday(&tv_even, (struct timezone *)NULL);
2577 timersub(&tv_even, &tv_odd, &tv_delta);
2578 if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
2582 compute_average(ODD_COUNTERS);
2583 format_all_counters(ODD_COUNTERS);
2584 flush_output_stdout();
2588 void check_dev_msr()
2593 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2594 if (stat(pathname, &sb))
2595 if (system("/sbin/modprobe msr > /dev/null 2>&1"))
2596 err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
2599 void check_permissions()
2601 struct __user_cap_header_struct cap_header_data;
2602 cap_user_header_t cap_header = &cap_header_data;
2603 struct __user_cap_data_struct cap_data_data;
2604 cap_user_data_t cap_data = &cap_data_data;
2605 extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
2609 /* check for CAP_SYS_RAWIO */
2610 cap_header->pid = getpid();
2611 cap_header->version = _LINUX_CAPABILITY_VERSION;
2612 if (capget(cap_header, cap_data) < 0)
2613 err(-6, "capget(2) failed");
2615 if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
2617 warnx("capget(CAP_SYS_RAWIO) failed,"
2618 " try \"# setcap cap_sys_rawio=ep %s\"", progname);
2621 /* test file permissions */
2622 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2623 if (euidaccess(pathname, R_OK)) {
2625 warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
2628 /* if all else fails, thell them to be root */
2631 warnx("... or simply run as root");
2638 * NHM adds support for additional MSRs:
2640 * MSR_SMI_COUNT 0x00000034
2642 * MSR_PLATFORM_INFO 0x000000ce
2643 * MSR_PKG_CST_CONFIG_CONTROL 0x000000e2
2645 * MSR_MISC_PWR_MGMT 0x000001aa
2647 * MSR_PKG_C3_RESIDENCY 0x000003f8
2648 * MSR_PKG_C6_RESIDENCY 0x000003f9
2649 * MSR_CORE_C3_RESIDENCY 0x000003fc
2650 * MSR_CORE_C6_RESIDENCY 0x000003fd
2653 * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
2654 * sets has_misc_feature_control
2656 int probe_nhm_msrs(unsigned int family, unsigned int model)
2658 unsigned long long msr;
2659 unsigned int base_ratio;
2660 int *pkg_cstate_limits;
2668 bclk = discover_bclk(family, model);
2671 case INTEL_FAM6_NEHALEM_EP: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
2672 case INTEL_FAM6_NEHALEM: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
2673 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
2674 case INTEL_FAM6_WESTMERE: /* Westmere Client - Clarkdale, Arrandale */
2675 case INTEL_FAM6_WESTMERE_EP: /* Westmere EP - Gulftown */
2676 case INTEL_FAM6_NEHALEM_EX: /* Nehalem-EX Xeon - Beckton */
2677 case INTEL_FAM6_WESTMERE_EX: /* Westmere-EX Xeon - Eagleton */
2678 pkg_cstate_limits = nhm_pkg_cstate_limits;
2680 case INTEL_FAM6_SANDYBRIDGE: /* SNB */
2681 case INTEL_FAM6_SANDYBRIDGE_X: /* SNB Xeon */
2682 case INTEL_FAM6_IVYBRIDGE: /* IVB */
2683 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */
2684 pkg_cstate_limits = snb_pkg_cstate_limits;
2685 has_misc_feature_control = 1;
2687 case INTEL_FAM6_HASWELL_CORE: /* HSW */
2688 case INTEL_FAM6_HASWELL_X: /* HSX */
2689 case INTEL_FAM6_HASWELL_ULT: /* HSW */
2690 case INTEL_FAM6_HASWELL_GT3E: /* HSW */
2691 case INTEL_FAM6_BROADWELL_CORE: /* BDW */
2692 case INTEL_FAM6_BROADWELL_GT3E: /* BDW */
2693 case INTEL_FAM6_BROADWELL_X: /* BDX */
2694 case INTEL_FAM6_BROADWELL_XEON_D: /* BDX-DE */
2695 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
2696 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
2697 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
2698 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
2699 pkg_cstate_limits = hsw_pkg_cstate_limits;
2700 has_misc_feature_control = 1;
2702 case INTEL_FAM6_SKYLAKE_X: /* SKX */
2703 pkg_cstate_limits = skx_pkg_cstate_limits;
2704 has_misc_feature_control = 1;
2706 case INTEL_FAM6_ATOM_SILVERMONT1: /* BYT */
2707 no_MSR_MISC_PWR_MGMT = 1;
2708 case INTEL_FAM6_ATOM_SILVERMONT2: /* AVN */
2709 pkg_cstate_limits = slv_pkg_cstate_limits;
2711 case INTEL_FAM6_ATOM_AIRMONT: /* AMT */
2712 pkg_cstate_limits = amt_pkg_cstate_limits;
2713 no_MSR_MISC_PWR_MGMT = 1;
2715 case INTEL_FAM6_XEON_PHI_KNL: /* PHI */
2716 case INTEL_FAM6_XEON_PHI_KNM:
2717 pkg_cstate_limits = phi_pkg_cstate_limits;
2719 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */
2720 case INTEL_FAM6_ATOM_GEMINI_LAKE:
2721 case INTEL_FAM6_ATOM_DENVERTON: /* DNV */
2722 pkg_cstate_limits = bxt_pkg_cstate_limits;
2727 get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2728 pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
2730 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2731 base_ratio = (msr >> 8) & 0xFF;
2733 base_hz = base_ratio * bclk * 1000000;
2738 * SLV client has support for unique MSRs:
2740 * MSR_CC6_DEMOTION_POLICY_CONFIG
2741 * MSR_MC6_DEMOTION_POLICY_CONFIG
2744 int has_slv_msrs(unsigned int family, unsigned int model)
2750 case INTEL_FAM6_ATOM_SILVERMONT1:
2751 case INTEL_FAM6_ATOM_MERRIFIELD:
2752 case INTEL_FAM6_ATOM_MOOREFIELD:
2757 int is_dnv(unsigned int family, unsigned int model)
2764 case INTEL_FAM6_ATOM_DENVERTON:
2769 int is_bdx(unsigned int family, unsigned int model)
2776 case INTEL_FAM6_BROADWELL_X:
2777 case INTEL_FAM6_BROADWELL_XEON_D:
2782 int is_skx(unsigned int family, unsigned int model)
2789 case INTEL_FAM6_SKYLAKE_X:
2795 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
2797 if (has_slv_msrs(family, model))
2801 /* Nehalem compatible, but do not include turbo-ratio limit support */
2802 case INTEL_FAM6_NEHALEM_EX: /* Nehalem-EX Xeon - Beckton */
2803 case INTEL_FAM6_WESTMERE_EX: /* Westmere-EX Xeon - Eagleton */
2804 case INTEL_FAM6_XEON_PHI_KNL: /* PHI - Knights Landing (different MSR definition) */
2805 case INTEL_FAM6_XEON_PHI_KNM:
2811 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
2813 if (has_slv_msrs(family, model))
2818 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
2827 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */
2828 case INTEL_FAM6_HASWELL_X: /* HSW Xeon */
2834 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
2843 case INTEL_FAM6_HASWELL_X: /* HSW Xeon */
2850 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
2859 case INTEL_FAM6_XEON_PHI_KNL: /* Knights Landing */
2860 case INTEL_FAM6_XEON_PHI_KNM:
2866 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
2875 case INTEL_FAM6_ATOM_GOLDMONT:
2876 case INTEL_FAM6_SKYLAKE_X:
2882 int has_config_tdp(unsigned int family, unsigned int model)
2891 case INTEL_FAM6_IVYBRIDGE: /* IVB */
2892 case INTEL_FAM6_HASWELL_CORE: /* HSW */
2893 case INTEL_FAM6_HASWELL_X: /* HSX */
2894 case INTEL_FAM6_HASWELL_ULT: /* HSW */
2895 case INTEL_FAM6_HASWELL_GT3E: /* HSW */
2896 case INTEL_FAM6_BROADWELL_CORE: /* BDW */
2897 case INTEL_FAM6_BROADWELL_GT3E: /* BDW */
2898 case INTEL_FAM6_BROADWELL_X: /* BDX */
2899 case INTEL_FAM6_BROADWELL_XEON_D: /* BDX-DE */
2900 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
2901 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
2902 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
2903 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
2904 case INTEL_FAM6_SKYLAKE_X: /* SKX */
2906 case INTEL_FAM6_XEON_PHI_KNL: /* Knights Landing */
2907 case INTEL_FAM6_XEON_PHI_KNM:
2915 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
2917 if (!do_nhm_platform_info)
2920 dump_nhm_platform_info();
2922 if (has_hsw_turbo_ratio_limit(family, model))
2923 dump_hsw_turbo_ratio_limits();
2925 if (has_ivt_turbo_ratio_limit(family, model))
2926 dump_ivt_turbo_ratio_limits();
2928 if (has_turbo_ratio_limit(family, model))
2929 dump_turbo_ratio_limits(family, model);
2931 if (has_atom_turbo_ratio_limit(family, model))
2932 dump_atom_turbo_ratio_limits();
2934 if (has_knl_turbo_ratio_limit(family, model))
2935 dump_knl_turbo_ratio_limits();
2937 if (has_config_tdp(family, model))
2944 dump_sysfs_cstate_config(void)
2953 if (!DO_BIC(BIC_sysfs))
2956 for (state = 0; state < 10; ++state) {
2958 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
2960 input = fopen(path, "r");
2963 fgets(name_buf, sizeof(name_buf), input);
2965 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
2966 sp = strchr(name_buf, '-');
2968 sp = strchrnul(name_buf, '\n');
2973 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc",
2975 input = fopen(path, "r");
2978 fgets(desc, sizeof(desc), input);
2980 fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
2985 dump_sysfs_pstate_config(void)
2988 char driver_buf[64];
2989 char governor_buf[64];
2993 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver",
2995 input = fopen(path, "r");
2996 if (input == NULL) {
2997 fprintf(stderr, "NSFOD %s\n", path);
3000 fgets(driver_buf, sizeof(driver_buf), input);
3003 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor",
3005 input = fopen(path, "r");
3006 if (input == NULL) {
3007 fprintf(stderr, "NSFOD %s\n", path);
3010 fgets(governor_buf, sizeof(governor_buf), input);
3013 fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
3014 fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
3016 sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
3017 input = fopen(path, "r");
3018 if (input != NULL) {
3019 fscanf(input, "%d", &turbo);
3020 fprintf(outf, "cpufreq boost: %d\n", turbo);
3024 sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
3025 input = fopen(path, "r");
3026 if (input != NULL) {
3027 fscanf(input, "%d", &turbo);
3028 fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
3036 * Decode the ENERGY_PERF_BIAS MSR
3038 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3040 unsigned long long msr;
3049 /* EPB is per-package */
3050 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3053 if (cpu_migrate(cpu)) {
3054 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3058 if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
3061 switch (msr & 0xF) {
3062 case ENERGY_PERF_BIAS_PERFORMANCE:
3063 epb_string = "performance";
3065 case ENERGY_PERF_BIAS_NORMAL:
3066 epb_string = "balanced";
3068 case ENERGY_PERF_BIAS_POWERSAVE:
3069 epb_string = "powersave";
3072 epb_string = "custom";
3075 fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
3081 * Decode the MSR_HWP_CAPABILITIES
3083 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3085 unsigned long long msr;
3093 /* MSR_HWP_CAPABILITIES is per-package */
3094 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3097 if (cpu_migrate(cpu)) {
3098 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3102 if (get_msr(cpu, MSR_PM_ENABLE, &msr))
3105 fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
3106 cpu, msr, (msr & (1 << 0)) ? "" : "No-");
3108 /* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
3109 if ((msr & (1 << 0)) == 0)
3112 if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
3115 fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
3116 "(high %d guar %d eff %d low %d)\n",
3118 (unsigned int)HWP_HIGHEST_PERF(msr),
3119 (unsigned int)HWP_GUARANTEED_PERF(msr),
3120 (unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
3121 (unsigned int)HWP_LOWEST_PERF(msr));
3123 if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
3126 fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
3127 "(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
3129 (unsigned int)(((msr) >> 0) & 0xff),
3130 (unsigned int)(((msr) >> 8) & 0xff),
3131 (unsigned int)(((msr) >> 16) & 0xff),
3132 (unsigned int)(((msr) >> 24) & 0xff),
3133 (unsigned int)(((msr) >> 32) & 0xff3),
3134 (unsigned int)(((msr) >> 42) & 0x1));
3137 if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
3140 fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
3141 "(min %d max %d des %d epp 0x%x window 0x%x)\n",
3143 (unsigned int)(((msr) >> 0) & 0xff),
3144 (unsigned int)(((msr) >> 8) & 0xff),
3145 (unsigned int)(((msr) >> 16) & 0xff),
3146 (unsigned int)(((msr) >> 24) & 0xff),
3147 (unsigned int)(((msr) >> 32) & 0xff3));
3149 if (has_hwp_notify) {
3150 if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
3153 fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
3154 "(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
3156 ((msr) & 0x1) ? "EN" : "Dis",
3157 ((msr) & 0x2) ? "EN" : "Dis");
3159 if (get_msr(cpu, MSR_HWP_STATUS, &msr))
3162 fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
3163 "(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
3165 ((msr) & 0x1) ? "" : "No-",
3166 ((msr) & 0x2) ? "" : "No-");
3172 * print_perf_limit()
3174 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3176 unsigned long long msr;
3182 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3185 if (cpu_migrate(cpu)) {
3186 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3190 if (do_core_perf_limit_reasons) {
3191 get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
3192 fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3193 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
3194 (msr & 1 << 15) ? "bit15, " : "",
3195 (msr & 1 << 14) ? "bit14, " : "",
3196 (msr & 1 << 13) ? "Transitions, " : "",
3197 (msr & 1 << 12) ? "MultiCoreTurbo, " : "",
3198 (msr & 1 << 11) ? "PkgPwrL2, " : "",
3199 (msr & 1 << 10) ? "PkgPwrL1, " : "",
3200 (msr & 1 << 9) ? "CorePwr, " : "",
3201 (msr & 1 << 8) ? "Amps, " : "",
3202 (msr & 1 << 6) ? "VR-Therm, " : "",
3203 (msr & 1 << 5) ? "Auto-HWP, " : "",
3204 (msr & 1 << 4) ? "Graphics, " : "",
3205 (msr & 1 << 2) ? "bit2, " : "",
3206 (msr & 1 << 1) ? "ThermStatus, " : "",
3207 (msr & 1 << 0) ? "PROCHOT, " : "");
3208 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
3209 (msr & 1 << 31) ? "bit31, " : "",
3210 (msr & 1 << 30) ? "bit30, " : "",
3211 (msr & 1 << 29) ? "Transitions, " : "",
3212 (msr & 1 << 28) ? "MultiCoreTurbo, " : "",
3213 (msr & 1 << 27) ? "PkgPwrL2, " : "",
3214 (msr & 1 << 26) ? "PkgPwrL1, " : "",
3215 (msr & 1 << 25) ? "CorePwr, " : "",
3216 (msr & 1 << 24) ? "Amps, " : "",
3217 (msr & 1 << 22) ? "VR-Therm, " : "",
3218 (msr & 1 << 21) ? "Auto-HWP, " : "",
3219 (msr & 1 << 20) ? "Graphics, " : "",
3220 (msr & 1 << 18) ? "bit18, " : "",
3221 (msr & 1 << 17) ? "ThermStatus, " : "",
3222 (msr & 1 << 16) ? "PROCHOT, " : "");
3225 if (do_gfx_perf_limit_reasons) {
3226 get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
3227 fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3228 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
3229 (msr & 1 << 0) ? "PROCHOT, " : "",
3230 (msr & 1 << 1) ? "ThermStatus, " : "",
3231 (msr & 1 << 4) ? "Graphics, " : "",
3232 (msr & 1 << 6) ? "VR-Therm, " : "",
3233 (msr & 1 << 8) ? "Amps, " : "",
3234 (msr & 1 << 9) ? "GFXPwr, " : "",
3235 (msr & 1 << 10) ? "PkgPwrL1, " : "",
3236 (msr & 1 << 11) ? "PkgPwrL2, " : "");
3237 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
3238 (msr & 1 << 16) ? "PROCHOT, " : "",
3239 (msr & 1 << 17) ? "ThermStatus, " : "",
3240 (msr & 1 << 20) ? "Graphics, " : "",
3241 (msr & 1 << 22) ? "VR-Therm, " : "",
3242 (msr & 1 << 24) ? "Amps, " : "",
3243 (msr & 1 << 25) ? "GFXPwr, " : "",
3244 (msr & 1 << 26) ? "PkgPwrL1, " : "",
3245 (msr & 1 << 27) ? "PkgPwrL2, " : "");
3247 if (do_ring_perf_limit_reasons) {
3248 get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
3249 fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3250 fprintf(outf, " (Active: %s%s%s%s%s%s)",
3251 (msr & 1 << 0) ? "PROCHOT, " : "",
3252 (msr & 1 << 1) ? "ThermStatus, " : "",
3253 (msr & 1 << 6) ? "VR-Therm, " : "",
3254 (msr & 1 << 8) ? "Amps, " : "",
3255 (msr & 1 << 10) ? "PkgPwrL1, " : "",
3256 (msr & 1 << 11) ? "PkgPwrL2, " : "");
3257 fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3258 (msr & 1 << 16) ? "PROCHOT, " : "",
3259 (msr & 1 << 17) ? "ThermStatus, " : "",
3260 (msr & 1 << 22) ? "VR-Therm, " : "",
3261 (msr & 1 << 24) ? "Amps, " : "",
3262 (msr & 1 << 26) ? "PkgPwrL1, " : "",
3263 (msr & 1 << 27) ? "PkgPwrL2, " : "");
3268 #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */
3269 #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */
3271 double get_tdp(unsigned int model)
3273 unsigned long long msr;
3275 if (do_rapl & RAPL_PKG_POWER_INFO)
3276 if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3277 return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3280 case INTEL_FAM6_ATOM_SILVERMONT1:
3281 case INTEL_FAM6_ATOM_SILVERMONT2:
3289 * rapl_dram_energy_units_probe()
3290 * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3293 rapl_dram_energy_units_probe(int model, double rapl_energy_units)
3295 /* only called for genuine_intel, family 6 */
3298 case INTEL_FAM6_HASWELL_X: /* HSX */
3299 case INTEL_FAM6_BROADWELL_X: /* BDX */
3300 case INTEL_FAM6_BROADWELL_XEON_D: /* BDX-DE */
3301 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */
3302 case INTEL_FAM6_XEON_PHI_KNM:
3303 return (rapl_dram_energy_units = 15.3 / 1000000);
3305 return (rapl_energy_units);
3313 * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
3315 void rapl_probe(unsigned int family, unsigned int model)
3317 unsigned long long msr;
3318 unsigned int time_unit;
3328 case INTEL_FAM6_SANDYBRIDGE:
3329 case INTEL_FAM6_IVYBRIDGE:
3330 case INTEL_FAM6_HASWELL_CORE: /* HSW */
3331 case INTEL_FAM6_HASWELL_ULT: /* HSW */
3332 case INTEL_FAM6_HASWELL_GT3E: /* HSW */
3333 case INTEL_FAM6_BROADWELL_CORE: /* BDW */
3334 case INTEL_FAM6_BROADWELL_GT3E: /* BDW */
3335 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3337 BIC_PRESENT(BIC_Pkg_J);
3338 BIC_PRESENT(BIC_Cor_J);
3339 BIC_PRESENT(BIC_GFX_J);
3341 BIC_PRESENT(BIC_PkgWatt);
3342 BIC_PRESENT(BIC_CorWatt);
3343 BIC_PRESENT(BIC_GFXWatt);
3346 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */
3347 case INTEL_FAM6_ATOM_GEMINI_LAKE:
3348 do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3350 BIC_PRESENT(BIC_Pkg_J);
3352 BIC_PRESENT(BIC_PkgWatt);
3354 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
3355 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
3356 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
3357 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
3358 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_GFX | RAPL_PKG_POWER_INFO;
3359 BIC_PRESENT(BIC_PKG__);
3360 BIC_PRESENT(BIC_RAM__);
3362 BIC_PRESENT(BIC_Pkg_J);
3363 BIC_PRESENT(BIC_Cor_J);
3364 BIC_PRESENT(BIC_RAM_J);
3365 BIC_PRESENT(BIC_GFX_J);
3367 BIC_PRESENT(BIC_PkgWatt);
3368 BIC_PRESENT(BIC_CorWatt);
3369 BIC_PRESENT(BIC_RAMWatt);
3370 BIC_PRESENT(BIC_GFXWatt);
3373 case INTEL_FAM6_HASWELL_X: /* HSX */
3374 case INTEL_FAM6_BROADWELL_X: /* BDX */
3375 case INTEL_FAM6_BROADWELL_XEON_D: /* BDX-DE */
3376 case INTEL_FAM6_SKYLAKE_X: /* SKX */
3377 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */
3378 case INTEL_FAM6_XEON_PHI_KNM:
3379 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3380 BIC_PRESENT(BIC_PKG__);
3381 BIC_PRESENT(BIC_RAM__);
3383 BIC_PRESENT(BIC_Pkg_J);
3384 BIC_PRESENT(BIC_RAM_J);
3386 BIC_PRESENT(BIC_PkgWatt);
3387 BIC_PRESENT(BIC_RAMWatt);
3390 case INTEL_FAM6_SANDYBRIDGE_X:
3391 case INTEL_FAM6_IVYBRIDGE_X:
3392 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
3393 BIC_PRESENT(BIC_PKG__);
3394 BIC_PRESENT(BIC_RAM__);
3396 BIC_PRESENT(BIC_Pkg_J);
3397 BIC_PRESENT(BIC_Cor_J);
3398 BIC_PRESENT(BIC_RAM_J);
3400 BIC_PRESENT(BIC_PkgWatt);
3401 BIC_PRESENT(BIC_CorWatt);
3402 BIC_PRESENT(BIC_RAMWatt);
3405 case INTEL_FAM6_ATOM_SILVERMONT1: /* BYT */
3406 case INTEL_FAM6_ATOM_SILVERMONT2: /* AVN */
3407 do_rapl = RAPL_PKG | RAPL_CORES;
3409 BIC_PRESENT(BIC_Pkg_J);
3410 BIC_PRESENT(BIC_Cor_J);
3412 BIC_PRESENT(BIC_PkgWatt);
3413 BIC_PRESENT(BIC_CorWatt);
3416 case INTEL_FAM6_ATOM_DENVERTON: /* DNV */
3417 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO | RAPL_CORES_ENERGY_STATUS;
3418 BIC_PRESENT(BIC_PKG__);
3419 BIC_PRESENT(BIC_RAM__);
3421 BIC_PRESENT(BIC_Pkg_J);
3422 BIC_PRESENT(BIC_Cor_J);
3423 BIC_PRESENT(BIC_RAM_J);
3425 BIC_PRESENT(BIC_PkgWatt);
3426 BIC_PRESENT(BIC_CorWatt);
3427 BIC_PRESENT(BIC_RAMWatt);
3434 /* units on package 0, verify later other packages match */
3435 if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3438 rapl_power_units = 1.0 / (1 << (msr & 0xF));
3439 if (model == INTEL_FAM6_ATOM_SILVERMONT1)
3440 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3442 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3444 rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3446 time_unit = msr >> 16 & 0xF;
3450 rapl_time_units = 1.0 / (1 << (time_unit));
3452 tdp = get_tdp(model);
3454 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3456 fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3461 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
3470 case INTEL_FAM6_HASWELL_CORE: /* HSW */
3471 case INTEL_FAM6_HASWELL_ULT: /* HSW */
3472 case INTEL_FAM6_HASWELL_GT3E: /* HSW */
3473 do_gfx_perf_limit_reasons = 1;
3474 case INTEL_FAM6_HASWELL_X: /* HSX */
3475 do_core_perf_limit_reasons = 1;
3476 do_ring_perf_limit_reasons = 1;
3482 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3484 unsigned long long msr;
3485 unsigned int dts, dts2;
3488 if (!(do_dts || do_ptm))
3493 /* DTS is per-core, no need to print for each thread */
3494 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
3497 if (cpu_migrate(cpu)) {
3498 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3502 if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
3503 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
3506 dts = (msr >> 16) & 0x7F;
3507 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
3508 cpu, msr, tcc_activation_temp - dts);
3510 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
3513 dts = (msr >> 16) & 0x7F;
3514 dts2 = (msr >> 8) & 0x7F;
3515 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
3516 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
3520 if (do_dts && debug) {
3521 unsigned int resolution;
3523 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
3526 dts = (msr >> 16) & 0x7F;
3527 resolution = (msr >> 27) & 0xF;
3528 fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
3529 cpu, msr, tcc_activation_temp - dts, resolution);
3531 if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
3534 dts = (msr >> 16) & 0x7F;
3535 dts2 = (msr >> 8) & 0x7F;
3536 fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
3537 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
3543 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
3545 fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
3547 ((msr >> 15) & 1) ? "EN" : "DIS",
3548 ((msr >> 0) & 0x7FFF) * rapl_power_units,
3549 (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
3550 (((msr >> 16) & 1) ? "EN" : "DIS"));
3555 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3557 unsigned long long msr;
3563 /* RAPL counters are per package, so print only for 1st thread/package */
3564 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3568 if (cpu_migrate(cpu)) {
3569 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3573 if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
3576 fprintf(outf, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr,
3577 rapl_power_units, rapl_energy_units, rapl_time_units);
3579 if (do_rapl & RAPL_PKG_POWER_INFO) {
3581 if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
3585 fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
3587 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3588 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3589 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3590 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
3593 if (do_rapl & RAPL_PKG) {
3595 if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
3598 fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
3599 cpu, msr, (msr >> 63) & 1 ? "" : "UN");
3601 print_power_limit_msr(cpu, msr, "PKG Limit #1");
3602 fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
3604 ((msr >> 47) & 1) ? "EN" : "DIS",
3605 ((msr >> 32) & 0x7FFF) * rapl_power_units,
3606 (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
3607 ((msr >> 48) & 1) ? "EN" : "DIS");
3610 if (do_rapl & RAPL_DRAM_POWER_INFO) {
3611 if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
3614 fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
3616 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3617 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3618 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3619 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
3621 if (do_rapl & RAPL_DRAM) {
3622 if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
3624 fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
3625 cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3627 print_power_limit_msr(cpu, msr, "DRAM Limit");
3629 if (do_rapl & RAPL_CORE_POLICY) {
3630 if (get_msr(cpu, MSR_PP0_POLICY, &msr))
3633 fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
3635 if (do_rapl & RAPL_CORES_POWER_LIMIT) {
3636 if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
3638 fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
3639 cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3640 print_power_limit_msr(cpu, msr, "Cores Limit");
3642 if (do_rapl & RAPL_GFX) {
3643 if (get_msr(cpu, MSR_PP1_POLICY, &msr))
3646 fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
3648 if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
3650 fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
3651 cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3652 print_power_limit_msr(cpu, msr, "GFX Limit");
3658 * SNB adds support for additional MSRs:
3660 * MSR_PKG_C7_RESIDENCY 0x000003fa
3661 * MSR_CORE_C7_RESIDENCY 0x000003fe
3662 * MSR_PKG_C2_RESIDENCY 0x0000060d
3665 int has_snb_msrs(unsigned int family, unsigned int model)
3671 case INTEL_FAM6_SANDYBRIDGE:
3672 case INTEL_FAM6_SANDYBRIDGE_X:
3673 case INTEL_FAM6_IVYBRIDGE: /* IVB */
3674 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */
3675 case INTEL_FAM6_HASWELL_CORE: /* HSW */
3676 case INTEL_FAM6_HASWELL_X: /* HSW */
3677 case INTEL_FAM6_HASWELL_ULT: /* HSW */
3678 case INTEL_FAM6_HASWELL_GT3E: /* HSW */
3679 case INTEL_FAM6_BROADWELL_CORE: /* BDW */
3680 case INTEL_FAM6_BROADWELL_GT3E: /* BDW */
3681 case INTEL_FAM6_BROADWELL_X: /* BDX */
3682 case INTEL_FAM6_BROADWELL_XEON_D: /* BDX-DE */
3683 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
3684 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
3685 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
3686 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
3687 case INTEL_FAM6_SKYLAKE_X: /* SKX */
3688 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */
3689 case INTEL_FAM6_ATOM_GEMINI_LAKE:
3690 case INTEL_FAM6_ATOM_DENVERTON: /* DNV */
3697 * HSW adds support for additional MSRs:
3699 * MSR_PKG_C8_RESIDENCY 0x00000630
3700 * MSR_PKG_C9_RESIDENCY 0x00000631
3701 * MSR_PKG_C10_RESIDENCY 0x00000632
3703 * MSR_PKGC8_IRTL 0x00000633
3704 * MSR_PKGC9_IRTL 0x00000634
3705 * MSR_PKGC10_IRTL 0x00000635
3708 int has_hsw_msrs(unsigned int family, unsigned int model)
3714 case INTEL_FAM6_HASWELL_ULT: /* HSW */
3715 case INTEL_FAM6_BROADWELL_CORE: /* BDW */
3716 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
3717 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
3718 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
3719 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
3720 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */
3721 case INTEL_FAM6_ATOM_GEMINI_LAKE:
3728 * SKL adds support for additional MSRS:
3730 * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658
3731 * MSR_PKG_ANY_CORE_C0_RES 0x00000659
3732 * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A
3733 * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B
3735 int has_skl_msrs(unsigned int family, unsigned int model)
3741 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
3742 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
3743 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
3744 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
3750 int is_slm(unsigned int family, unsigned int model)
3755 case INTEL_FAM6_ATOM_SILVERMONT1: /* BYT */
3756 case INTEL_FAM6_ATOM_SILVERMONT2: /* AVN */
3762 int is_knl(unsigned int family, unsigned int model)
3767 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */
3768 case INTEL_FAM6_XEON_PHI_KNM:
3774 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
3776 if (is_knl(family, model))
3781 #define SLM_BCLK_FREQS 5
3782 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
3784 double slm_bclk(void)
3786 unsigned long long msr = 3;
3790 if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
3791 fprintf(outf, "SLM BCLK: unknown\n");
3794 if (i >= SLM_BCLK_FREQS) {
3795 fprintf(outf, "SLM BCLK[%d] invalid\n", i);
3798 freq = slm_freq_table[i];
3801 fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
3806 double discover_bclk(unsigned int family, unsigned int model)
3808 if (has_snb_msrs(family, model) || is_knl(family, model))
3810 else if (is_slm(family, model))
3817 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
3818 * the Thermal Control Circuit (TCC) activates.
3819 * This is usually equal to tjMax.
3821 * Older processors do not have this MSR, so there we guess,
3822 * but also allow cmdline over-ride with -T.
3824 * Several MSR temperature values are in units of degrees-C
3825 * below this value, including the Digital Thermal Sensor (DTS),
3826 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
3828 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3830 unsigned long long msr;
3831 unsigned int target_c_local;
3834 /* tcc_activation_temp is used only for dts or ptm */
3835 if (!(do_dts || do_ptm))
3838 /* this is a per-package concept */
3839 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3843 if (cpu_migrate(cpu)) {
3844 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3848 if (tcc_activation_temp_override != 0) {
3849 tcc_activation_temp = tcc_activation_temp_override;
3850 fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
3851 cpu, tcc_activation_temp);
3855 /* Temperature Target MSR is Nehalem and newer only */
3856 if (!do_nhm_platform_info)
3859 if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
3862 target_c_local = (msr >> 16) & 0xFF;
3865 fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
3866 cpu, msr, target_c_local);
3868 if (!target_c_local)
3871 tcc_activation_temp = target_c_local;
3876 tcc_activation_temp = TJMAX_DEFAULT;
3877 fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
3878 cpu, tcc_activation_temp);
3883 void decode_feature_control_msr(void)
3885 unsigned long long msr;
3887 if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
3888 fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
3890 msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
3891 msr & (1 << 18) ? "SGX" : "");
3894 void decode_misc_enable_msr(void)
3896 unsigned long long msr;
3898 if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
3899 fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
3901 msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
3902 msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
3903 msr & MSR_IA32_MISC_ENABLE_MWAIT ? "No-" : "",
3904 msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
3905 msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
3908 void decode_misc_feature_control(void)
3910 unsigned long long msr;
3912 if (!has_misc_feature_control)
3915 if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
3916 fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
3918 msr & (0 << 0) ? "No-" : "",
3919 msr & (1 << 0) ? "No-" : "",
3920 msr & (2 << 0) ? "No-" : "",
3921 msr & (3 << 0) ? "No-" : "");
3924 * Decode MSR_MISC_PWR_MGMT
3926 * Decode the bits according to the Nehalem documentation
3927 * bit[0] seems to continue to have same meaning going forward
3930 void decode_misc_pwr_mgmt_msr(void)
3932 unsigned long long msr;
3934 if (!do_nhm_platform_info)
3937 if (no_MSR_MISC_PWR_MGMT)
3940 if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
3941 fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
3943 msr & (1 << 0) ? "DIS" : "EN",
3944 msr & (1 << 1) ? "EN" : "DIS",
3945 msr & (1 << 8) ? "EN" : "DIS");
3948 * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
3950 * This MSRs are present on Silvermont processors,
3951 * Intel Atom processor E3000 series (Baytrail), and friends.
3953 void decode_c6_demotion_policy_msr(void)
3955 unsigned long long msr;
3957 if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
3958 fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
3959 base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
3961 if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
3962 fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
3963 base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
3966 void process_cpuid()
3968 unsigned int eax, ebx, ecx, edx, max_level, max_extended_level;
3969 unsigned int fms, family, model, stepping;
3970 unsigned int has_turbo;
3972 eax = ebx = ecx = edx = 0;
3974 __cpuid(0, max_level, ebx, ecx, edx);
3976 if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
3980 fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
3981 (char *)&ebx, (char *)&edx, (char *)&ecx);
3983 __cpuid(1, fms, ebx, ecx, edx);
3984 family = (fms >> 8) & 0xf;
3985 model = (fms >> 4) & 0xf;
3986 stepping = fms & 0xf;
3987 if (family == 6 || family == 0xf)
3988 model += ((fms >> 16) & 0xf) << 4;
3991 fprintf(outf, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
3992 max_level, family, model, stepping, family, model, stepping);
3993 fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s\n",
3994 ecx & (1 << 0) ? "SSE3" : "-",
3995 ecx & (1 << 3) ? "MONITOR" : "-",
3996 ecx & (1 << 6) ? "SMX" : "-",
3997 ecx & (1 << 7) ? "EIST" : "-",
3998 ecx & (1 << 8) ? "TM2" : "-",
3999 edx & (1 << 4) ? "TSC" : "-",
4000 edx & (1 << 5) ? "MSR" : "-",
4001 edx & (1 << 22) ? "ACPI-TM" : "-",
4002 edx & (1 << 29) ? "TM" : "-");
4005 if (!(edx & (1 << 5)))
4006 errx(1, "CPUID: no MSR");
4009 * check max extended function levels of CPUID.
4010 * This is needed to check for invariant TSC.
4011 * This check is valid for both Intel and AMD.
4013 ebx = ecx = edx = 0;
4014 __cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
4016 if (max_extended_level >= 0x80000007) {
4019 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
4020 * this check is valid for both Intel and AMD
4022 __cpuid(0x80000007, eax, ebx, ecx, edx);
4023 has_invariant_tsc = edx & (1 << 8);
4027 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
4028 * this check is valid for both Intel and AMD
4031 __cpuid(0x6, eax, ebx, ecx, edx);
4032 has_aperf = ecx & (1 << 0);
4034 BIC_PRESENT(BIC_Avg_MHz);
4035 BIC_PRESENT(BIC_Busy);
4036 BIC_PRESENT(BIC_Bzy_MHz);
4038 do_dts = eax & (1 << 0);
4040 BIC_PRESENT(BIC_CoreTmp);
4041 has_turbo = eax & (1 << 1);
4042 do_ptm = eax & (1 << 6);
4044 BIC_PRESENT(BIC_PkgTmp);
4045 has_hwp = eax & (1 << 7);
4046 has_hwp_notify = eax & (1 << 8);
4047 has_hwp_activity_window = eax & (1 << 9);
4048 has_hwp_epp = eax & (1 << 10);
4049 has_hwp_pkg = eax & (1 << 11);
4050 has_epb = ecx & (1 << 3);
4053 fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
4054 "%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
4055 has_aperf ? "" : "No-",
4056 has_turbo ? "" : "No-",
4057 do_dts ? "" : "No-",
4058 do_ptm ? "" : "No-",
4059 has_hwp ? "" : "No-",
4060 has_hwp_notify ? "" : "No-",
4061 has_hwp_activity_window ? "" : "No-",
4062 has_hwp_epp ? "" : "No-",
4063 has_hwp_pkg ? "" : "No-",
4064 has_epb ? "" : "No-");
4067 decode_misc_enable_msr();
4070 if (max_level >= 0x7 && !quiet) {
4075 __cpuid_count(0x7, 0, eax, ebx, ecx, edx);
4077 has_sgx = ebx & (1 << 2);
4078 fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
4081 decode_feature_control_msr();
4084 if (max_level >= 0x15) {
4085 unsigned int eax_crystal;
4086 unsigned int ebx_tsc;
4089 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
4091 eax_crystal = ebx_tsc = crystal_hz = edx = 0;
4092 __cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
4096 if (!quiet && (ebx != 0))
4097 fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
4098 eax_crystal, ebx_tsc, crystal_hz);
4100 if (crystal_hz == 0)
4102 case INTEL_FAM6_SKYLAKE_MOBILE: /* SKL */
4103 case INTEL_FAM6_SKYLAKE_DESKTOP: /* SKL */
4104 case INTEL_FAM6_KABYLAKE_MOBILE: /* KBL */
4105 case INTEL_FAM6_KABYLAKE_DESKTOP: /* KBL */
4106 crystal_hz = 24000000; /* 24.0 MHz */
4108 case INTEL_FAM6_SKYLAKE_X: /* SKX */
4109 case INTEL_FAM6_ATOM_DENVERTON: /* DNV */
4110 crystal_hz = 25000000; /* 25.0 MHz */
4112 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */
4113 case INTEL_FAM6_ATOM_GEMINI_LAKE:
4114 crystal_hz = 19200000; /* 19.2 MHz */
4121 tsc_hz = (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
4123 fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
4124 tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal);
4128 if (max_level >= 0x16) {
4129 unsigned int base_mhz, max_mhz, bus_mhz, edx;
4132 * CPUID 16H Base MHz, Max MHz, Bus MHz
4134 base_mhz = max_mhz = bus_mhz = edx = 0;
4136 __cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
4138 fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
4139 base_mhz, max_mhz, bus_mhz);
4143 aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
4145 BIC_PRESENT(BIC_IRQ);
4146 BIC_PRESENT(BIC_TSC_MHz);
4148 if (probe_nhm_msrs(family, model)) {
4149 do_nhm_platform_info = 1;
4150 BIC_PRESENT(BIC_CPU_c1);
4151 BIC_PRESENT(BIC_CPU_c3);
4152 BIC_PRESENT(BIC_CPU_c6);
4153 BIC_PRESENT(BIC_SMI);
4155 do_snb_cstates = has_snb_msrs(family, model);
4158 BIC_PRESENT(BIC_CPU_c7);
4160 do_irtl_snb = has_snb_msrs(family, model);
4161 if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
4162 BIC_PRESENT(BIC_Pkgpc2);
4163 if (pkg_cstate_limit >= PCL__3)
4164 BIC_PRESENT(BIC_Pkgpc3);
4165 if (pkg_cstate_limit >= PCL__6)
4166 BIC_PRESENT(BIC_Pkgpc6);
4167 if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
4168 BIC_PRESENT(BIC_Pkgpc7);
4169 if (has_slv_msrs(family, model)) {
4170 BIC_NOT_PRESENT(BIC_Pkgpc2);
4171 BIC_NOT_PRESENT(BIC_Pkgpc3);
4172 BIC_PRESENT(BIC_Pkgpc6);
4173 BIC_NOT_PRESENT(BIC_Pkgpc7);
4174 BIC_PRESENT(BIC_Mod_c6);
4175 use_c1_residency_msr = 1;
4177 if (is_dnv(family, model)) {
4178 BIC_PRESENT(BIC_CPU_c1);
4179 BIC_NOT_PRESENT(BIC_CPU_c3);
4180 BIC_NOT_PRESENT(BIC_Pkgpc3);
4181 BIC_NOT_PRESENT(BIC_CPU_c7);
4182 BIC_NOT_PRESENT(BIC_Pkgpc7);
4183 use_c1_residency_msr = 1;
4185 if (is_skx(family, model)) {
4186 BIC_NOT_PRESENT(BIC_CPU_c3);
4187 BIC_NOT_PRESENT(BIC_Pkgpc3);
4188 BIC_NOT_PRESENT(BIC_CPU_c7);
4189 BIC_NOT_PRESENT(BIC_Pkgpc7);
4191 if (is_bdx(family, model)) {
4192 BIC_NOT_PRESENT(BIC_CPU_c7);
4193 BIC_NOT_PRESENT(BIC_Pkgpc7);
4195 if (has_hsw_msrs(family, model)) {
4196 BIC_PRESENT(BIC_Pkgpc8);
4197 BIC_PRESENT(BIC_Pkgpc9);
4198 BIC_PRESENT(BIC_Pkgpc10);
4200 do_irtl_hsw = has_hsw_msrs(family, model);
4201 do_skl_residency = has_skl_msrs(family, model);
4202 do_slm_cstates = is_slm(family, model);
4203 do_knl_cstates = is_knl(family, model);
4206 decode_misc_pwr_mgmt_msr();
4208 if (!quiet && has_slv_msrs(family, model))
4209 decode_c6_demotion_policy_msr();
4211 rapl_probe(family, model);
4212 perf_limit_reasons_probe(family, model);
4215 dump_cstate_pstate_config_info(family, model);
4218 dump_sysfs_cstate_config();
4220 dump_sysfs_pstate_config();
4222 if (has_skl_msrs(family, model))
4223 calculate_tsc_tweak();
4225 if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
4226 BIC_PRESENT(BIC_GFX_rc6);
4228 if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
4229 BIC_PRESENT(BIC_GFXMHz);
4232 decode_misc_feature_control();
4239 * in /dev/cpu/ return success for names that are numbers
4240 * ie. filter out ".", "..", "microcode".
4242 int dir_filter(const struct dirent *dirp)
4244 if (isdigit(dirp->d_name[0]))
4250 int open_dev_cpu_msr(int dummy1)
4255 void topology_probe()
4258 int max_core_id = 0;
4259 int max_package_id = 0;
4260 int max_siblings = 0;
4261 struct cpu_topology {
4263 int physical_package_id;
4266 /* Initialize num_cpus, max_cpu_num */
4268 topo.max_cpu_num = 0;
4269 for_all_proc_cpus(count_cpus);
4270 if (!summary_only && topo.num_cpus > 1)
4271 BIC_PRESENT(BIC_CPU);
4274 fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4276 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
4278 err(1, "calloc cpus");
4281 * Allocate and initialize cpu_present_set
4283 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4284 if (cpu_present_set == NULL)
4285 err(3, "CPU_ALLOC");
4286 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4287 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4288 for_all_proc_cpus(mark_cpu_present);
4291 * Validate that all cpus in cpu_subset are also in cpu_present_set
4293 for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
4294 if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
4295 if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
4296 err(1, "cpu%d not present", i);
4300 * Allocate and initialize cpu_affinity_set
4302 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4303 if (cpu_affinity_set == NULL)
4304 err(3, "CPU_ALLOC");
4305 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4306 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4311 * find max_core_id, max_package_id
4313 for (i = 0; i <= topo.max_cpu_num; ++i) {
4316 if (cpu_is_not_present(i)) {
4318 fprintf(outf, "cpu%d NOT PRESENT\n", i);
4321 cpus[i].core_id = get_core_id(i);
4322 if (cpus[i].core_id > max_core_id)
4323 max_core_id = cpus[i].core_id;
4325 cpus[i].physical_package_id = get_physical_package_id(i);
4326 if (cpus[i].physical_package_id > max_package_id)
4327 max_package_id = cpus[i].physical_package_id;
4329 siblings = get_num_ht_siblings(i);
4330 if (siblings > max_siblings)
4331 max_siblings = siblings;
4333 fprintf(outf, "cpu %d pkg %d core %d\n",
4334 i, cpus[i].physical_package_id, cpus[i].core_id);
4336 topo.num_cores_per_pkg = max_core_id + 1;
4338 fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
4339 max_core_id, topo.num_cores_per_pkg);
4340 if (!summary_only && topo.num_cores_per_pkg > 1)
4341 BIC_PRESENT(BIC_Core);
4343 topo.num_packages = max_package_id + 1;
4345 fprintf(outf, "max_package_id %d, sizing for %d packages\n",
4346 max_package_id, topo.num_packages);
4347 if (!summary_only && topo.num_packages > 1)
4348 BIC_PRESENT(BIC_Package);
4350 topo.num_threads_per_core = max_siblings;
4352 fprintf(outf, "max_siblings %d\n", max_siblings);
4358 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
4362 *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
4363 topo.num_packages, sizeof(struct thread_data));
4367 for (i = 0; i < topo.num_threads_per_core *
4368 topo.num_cores_per_pkg * topo.num_packages; i++)
4369 (*t)[i].cpu_id = -1;
4371 *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
4372 sizeof(struct core_data));
4376 for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
4377 (*c)[i].core_id = -1;
4379 *p = calloc(topo.num_packages, sizeof(struct pkg_data));
4383 for (i = 0; i < topo.num_packages; i++)
4384 (*p)[i].package_id = i;
4388 err(1, "calloc counters");
4393 * set cpu_id, core_num, pkg_num
4394 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
4396 * increment topo.num_cores when 1st core in pkg seen
4398 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
4399 struct pkg_data *pkg_base, int thread_num, int core_num,
4400 int pkg_num, int cpu_id)
4402 struct thread_data *t;
4403 struct core_data *c;
4406 t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
4407 c = GET_CORE(core_base, core_num, pkg_num);
4408 p = GET_PKG(pkg_base, pkg_num);
4411 if (thread_num == 0) {
4412 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
4413 if (cpu_is_first_core_in_package(cpu_id))
4414 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
4417 c->core_id = core_num;
4418 p->package_id = pkg_num;
4422 int initialize_counters(int cpu_id)
4424 int my_thread_id, my_core_id, my_package_id;
4426 my_package_id = get_physical_package_id(cpu_id);
4427 my_core_id = get_core_id(cpu_id);
4428 my_thread_id = get_cpu_position_in_core(cpu_id);
4432 init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
4433 init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
4437 void allocate_output_buffer()
4439 output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
4440 outp = output_buffer;
4442 err(-1, "calloc output buffer");
4444 void allocate_fd_percpu(void)
4446 fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
4447 if (fd_percpu == NULL)
4448 err(-1, "calloc fd_percpu");
4450 void allocate_irq_buffers(void)
4452 irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
4453 if (irq_column_2_cpu == NULL)
4454 err(-1, "calloc %d", topo.num_cpus);
4456 irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
4457 if (irqs_per_cpu == NULL)
4458 err(-1, "calloc %d", topo.max_cpu_num + 1);
4460 void setup_all_buffers(void)
4463 allocate_irq_buffers();
4464 allocate_fd_percpu();
4465 allocate_counters(&thread_even, &core_even, &package_even);
4466 allocate_counters(&thread_odd, &core_odd, &package_odd);
4467 allocate_output_buffer();
4468 for_all_proc_cpus(initialize_counters);
4471 void set_base_cpu(void)
4473 base_cpu = sched_getcpu();
4475 err(-ENODEV, "No valid cpus found");
4478 fprintf(outf, "base_cpu = %d\n", base_cpu);
4481 void turbostat_init()
4483 setup_all_buffers();
4486 check_permissions();
4491 for_all_cpus(print_hwp, ODD_COUNTERS);
4494 for_all_cpus(print_epb, ODD_COUNTERS);
4497 for_all_cpus(print_perf_limit, ODD_COUNTERS);
4500 for_all_cpus(print_rapl, ODD_COUNTERS);
4502 for_all_cpus(set_temperature_target, ODD_COUNTERS);
4505 for_all_cpus(print_thermal, ODD_COUNTERS);
4507 if (!quiet && do_irtl_snb)
4511 int fork_it(char **argv)
4516 snapshot_proc_sysfs_files();
4517 status = for_all_cpus(get_counters, EVEN_COUNTERS);
4520 /* clear affinity side-effect of get_counters() */
4521 sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
4522 gettimeofday(&tv_even, (struct timezone *)NULL);
4527 execvp(argv[0], argv);
4528 err(errno, "exec %s", argv[0]);
4532 if (child_pid == -1)
4535 signal(SIGINT, SIG_IGN);
4536 signal(SIGQUIT, SIG_IGN);
4537 if (waitpid(child_pid, &status, 0) == -1)
4538 err(status, "waitpid");
4541 * n.b. fork_it() does not check for errors from for_all_cpus()
4542 * because re-starting is problematic when forking
4544 snapshot_proc_sysfs_files();
4545 for_all_cpus(get_counters, ODD_COUNTERS);
4546 gettimeofday(&tv_odd, (struct timezone *)NULL);
4547 timersub(&tv_odd, &tv_even, &tv_delta);
4548 if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
4549 fprintf(outf, "%s: Counter reset detected\n", progname);
4551 compute_average(EVEN_COUNTERS);
4552 format_all_counters(EVEN_COUNTERS);
4555 fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
4557 flush_output_stderr();
4562 int get_and_dump_counters(void)
4566 snapshot_proc_sysfs_files();
4567 status = for_all_cpus(get_counters, ODD_COUNTERS);
4571 status = for_all_cpus(dump_counters, ODD_COUNTERS);
4575 flush_output_stdout();
4580 void print_version() {
4581 fprintf(outf, "turbostat version 17.04.12"
4582 " - Len Brown <lenb@kernel.org>\n");
4585 int add_counter(unsigned int msr_num, char *path, char *name,
4586 unsigned int width, enum counter_scope scope,
4587 enum counter_type type, enum counter_format format, int flags)
4589 struct msr_counter *msrp;
4591 msrp = calloc(1, sizeof(struct msr_counter));
4597 msrp->msr_num = msr_num;
4598 strncpy(msrp->name, name, NAME_BYTES);
4600 strncpy(msrp->path, path, PATH_BYTES);
4601 msrp->width = width;
4603 msrp->format = format;
4604 msrp->flags = flags;
4609 msrp->next = sys.tp;
4611 sys.added_thread_counters++;
4612 if (sys.added_thread_counters > MAX_ADDED_COUNTERS) {
4613 fprintf(stderr, "exceeded max %d added thread counters\n",
4614 MAX_ADDED_COUNTERS);
4620 msrp->next = sys.cp;
4622 sys.added_core_counters++;
4623 if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
4624 fprintf(stderr, "exceeded max %d added core counters\n",
4625 MAX_ADDED_COUNTERS);
4631 msrp->next = sys.pp;
4633 sys.added_package_counters++;
4634 if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
4635 fprintf(stderr, "exceeded max %d added package counters\n",
4636 MAX_ADDED_COUNTERS);
4645 void parse_add_command(char *add_command)
4649 char name_buffer[NAME_BYTES] = "";
4652 enum counter_scope scope = SCOPE_CPU;
4653 enum counter_type type = COUNTER_CYCLES;
4654 enum counter_format format = FORMAT_DELTA;
4656 while (add_command) {
4658 if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
4661 if (sscanf(add_command, "msr%d", &msr_num) == 1)
4664 if (*add_command == '/') {
4669 if (sscanf(add_command, "u%d", &width) == 1) {
4670 if ((width == 32) || (width == 64))
4674 if (!strncmp(add_command, "cpu", strlen("cpu"))) {
4678 if (!strncmp(add_command, "core", strlen("core"))) {
4682 if (!strncmp(add_command, "package", strlen("package"))) {
4683 scope = SCOPE_PACKAGE;
4686 if (!strncmp(add_command, "cycles", strlen("cycles"))) {
4687 type = COUNTER_CYCLES;
4690 if (!strncmp(add_command, "seconds", strlen("seconds"))) {
4691 type = COUNTER_SECONDS;
4694 if (!strncmp(add_command, "usec", strlen("usec"))) {
4695 type = COUNTER_USEC;
4698 if (!strncmp(add_command, "raw", strlen("raw"))) {
4699 format = FORMAT_RAW;
4702 if (!strncmp(add_command, "delta", strlen("delta"))) {
4703 format = FORMAT_DELTA;
4706 if (!strncmp(add_command, "percent", strlen("percent"))) {
4707 format = FORMAT_PERCENT;
4711 if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) { /* 18 < NAME_BYTES */
4714 eos = strchr(name_buffer, ',');
4721 add_command = strchr(add_command, ',');
4723 *add_command = '\0';
4728 if ((msr_num == 0) && (path == NULL)) {
4729 fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
4733 /* generate default column header */
4734 if (*name_buffer == '\0') {
4736 sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
4738 sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
4741 if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
4750 int is_deferred_skip(char *name)
4754 for (i = 0; i < deferred_skip_index; ++i)
4755 if (!strcmp(name, deferred_skip_names[i]))
4760 void probe_sysfs(void)
4768 if (!DO_BIC(BIC_sysfs))
4771 for (state = 10; state > 0; --state) {
4773 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
4775 input = fopen(path, "r");
4778 fgets(name_buf, sizeof(name_buf), input);
4780 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
4781 sp = strchr(name_buf, '-');
4783 sp = strchrnul(name_buf, '\n');
4789 sprintf(path, "cpuidle/state%d/time", state);
4791 if (is_deferred_skip(name_buf))
4794 add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC,
4795 FORMAT_PERCENT, SYSFS_PERCPU);
4798 for (state = 10; state > 0; --state) {
4800 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
4802 input = fopen(path, "r");
4805 fgets(name_buf, sizeof(name_buf), input);
4806 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
4807 sp = strchr(name_buf, '-');
4809 sp = strchrnul(name_buf, '\n');
4813 sprintf(path, "cpuidle/state%d/usage", state);
4815 if (is_deferred_skip(name_buf))
4818 add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS,
4819 FORMAT_DELTA, SYSFS_PERCPU);
4826 * parse cpuset with following syntax
4827 * 1,2,4..6,8-10 and set bits in cpu_subset
4829 void parse_cpu_command(char *optarg)
4831 unsigned int start, end;
4834 if (!strcmp(optarg, "core")) {
4840 if (!strcmp(optarg, "package")) {
4846 if (show_core_only || show_pkg_only)
4849 cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
4850 if (cpu_subset == NULL)
4851 err(3, "CPU_ALLOC");
4852 cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
4854 CPU_ZERO_S(cpu_subset_size, cpu_subset);
4858 while (next && *next) {
4860 if (*next == '-') /* no negative cpu numbers */
4863 start = strtoul(next, &next, 10);
4865 if (start >= CPU_SUBSET_MAXCPUS)
4867 CPU_SET_S(start, cpu_subset_size, cpu_subset);
4878 next += 1; /* start range */
4879 } else if (*next == '.') {
4882 next += 1; /* start range */
4887 end = strtoul(next, &next, 10);
4891 while (++start <= end) {
4892 if (start >= CPU_SUBSET_MAXCPUS)
4894 CPU_SET_S(start, cpu_subset_size, cpu_subset);
4899 else if (*next != '\0')
4906 fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
4913 * parse_show_hide() - process cmdline to set default counter action
4915 void parse_show_hide(char *optarg, enum show_hide_mode new_mode)
4918 * --show: show only those specified
4919 * The 1st invocation will clear and replace the enabled mask
4920 * subsequent invocations can add to it.
4922 if (new_mode == SHOW_LIST) {
4924 bic_enabled = bic_lookup(optarg, new_mode);
4926 bic_enabled |= bic_lookup(optarg, new_mode);
4933 * --hide: do not show those specified
4934 * multiple invocations simply clear more bits in enabled mask
4936 bic_enabled &= ~bic_lookup(optarg, new_mode);
4940 void cmdline(int argc, char **argv)
4943 int option_index = 0;
4944 static struct option long_options[] = {
4945 {"add", required_argument, 0, 'a'},
4946 {"cpu", required_argument, 0, 'c'},
4947 {"Dump", no_argument, 0, 'D'},
4948 {"debug", no_argument, 0, 'd'}, /* internal, not documented */
4949 {"interval", required_argument, 0, 'i'},
4950 {"help", no_argument, 0, 'h'},
4951 {"hide", required_argument, 0, 'H'}, // meh, -h taken by --help
4952 {"Joules", no_argument, 0, 'J'},
4953 {"list", no_argument, 0, 'l'},
4954 {"out", required_argument, 0, 'o'},
4955 {"quiet", no_argument, 0, 'q'},
4956 {"show", required_argument, 0, 's'},
4957 {"Summary", no_argument, 0, 'S'},
4958 {"TCC", required_argument, 0, 'T'},
4959 {"version", no_argument, 0, 'v' },
4965 while ((opt = getopt_long_only(argc, argv, "+C:c:Ddhi:JM:m:o:qST:v",
4966 long_options, &option_index)) != -1) {
4969 parse_add_command(optarg);
4972 parse_cpu_command(optarg);
4981 parse_show_hide(optarg, HIDE_LIST);
4989 double interval = strtod(optarg, NULL);
4991 if (interval < 0.001) {
4992 fprintf(outf, "interval %f seconds is too small\n",
4997 interval_ts.tv_sec = interval;
4998 interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
5009 outf = fopen_or_die(optarg, "w");
5015 parse_show_hide(optarg, SHOW_LIST);
5021 tcc_activation_temp_override = atoi(optarg);
5031 int main(int argc, char **argv)
5035 cmdline(argc, argv);
5044 /* dump counters and exit */
5046 return get_and_dump_counters();
5048 /* list header and exit */
5049 if (list_header_only) {
5051 flush_output_stdout();
5056 * if any params left, it must be a command to fork
5059 return fork_it(argv + optind);