1 /*
2 * turbostat -- Log CPU frequency and C-state residency
3 * on modern Intel turbo-capable processors for collectd.
4 *
5 * Based on the 'turbostat' tool of the Linux kernel, found at
6 * linux/tools/power/x86/turbostat/turbostat.c:
7 * ----
8 * Copyright (c) 2013 Intel Corporation.
9 * Len Brown <len.brown@intel.com>
10 *
11 * This program is free software; you can redistribute it and/or modify it
12 * under the terms and conditions of the GNU General Public License,
13 * version 2, as published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope it will be useful, but WITHOUT
16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
17 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
18 * more details.
19 *
20 * You should have received a copy of the GNU General Public License along with
21 * this program; if not, write to the Free Software Foundation, Inc.,
22 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
23 * ----
24 * Ported to collectd by Vincent Brillault <git@lerya.net>
25 */
27 #define _GNU_SOURCE
28 #include <asm/msr-index.h>
29 #include <stdarg.h>
30 #include <stdio.h>
31 #include <err.h>
32 #include <unistd.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <sys/stat.h>
36 #include <sys/resource.h>
37 #include <fcntl.h>
38 #include <signal.h>
39 #include <sys/time.h>
40 #include <stdlib.h>
41 #include <dirent.h>
42 #include <string.h>
43 #include <ctype.h>
44 #include <sched.h>
45 #include <cpuid.h>
47 #include "collectd.h"
48 #include "common.h"
49 #include "plugin.h"
51 #define PLUGIN_NAME "turbostat"
53 static const char *proc_stat = "/proc/stat";
54 static unsigned int skip_c0;
55 static unsigned int skip_c1;
56 static unsigned int do_nhm_cstates;
57 static unsigned int do_snb_cstates;
58 static unsigned int do_c8_c9_c10;
59 static unsigned int do_slm_cstates;
60 static unsigned int has_aperf;
61 static unsigned int has_epb;
62 static unsigned int units = 1000000000; /* Ghz etc */
63 static unsigned int genuine_intel;
64 static unsigned int has_invariant_tsc;
65 static unsigned int do_nehalem_platform_info;
66 static int do_smi;
67 static unsigned int show_pkg;
68 static unsigned int show_core;
69 static unsigned int show_cpu;
70 static unsigned int do_rapl;
71 static unsigned int do_dts;
72 static unsigned int do_ptm;
73 static unsigned int tcc_activation_temp;
74 static unsigned int tcc_activation_temp_override;
75 static double rapl_power_units, rapl_energy_units, rapl_time_units;
76 static double rapl_joule_counter_range;
78 #define RAPL_PKG (1 << 0)
79 /* 0x610 MSR_PKG_POWER_LIMIT */
80 /* 0x611 MSR_PKG_ENERGY_STATUS */
81 #define RAPL_PKG_PERF_STATUS (1 << 1)
82 /* 0x613 MSR_PKG_PERF_STATUS */
83 #define RAPL_PKG_POWER_INFO (1 << 2)
84 /* 0x614 MSR_PKG_POWER_INFO */
86 #define RAPL_DRAM (1 << 3)
87 /* 0x618 MSR_DRAM_POWER_LIMIT */
88 /* 0x619 MSR_DRAM_ENERGY_STATUS */
89 /* 0x61c MSR_DRAM_POWER_INFO */
90 #define RAPL_DRAM_PERF_STATUS (1 << 4)
91 /* 0x61b MSR_DRAM_PERF_STATUS */
93 #define RAPL_CORES (1 << 5)
94 /* 0x638 MSR_PP0_POWER_LIMIT */
95 /* 0x639 MSR_PP0_ENERGY_STATUS */
96 #define RAPL_CORE_POLICY (1 << 6)
97 /* 0x63a MSR_PP0_POLICY */
100 #define RAPL_GFX (1 << 7)
101 /* 0x640 MSR_PP1_POWER_LIMIT */
102 /* 0x641 MSR_PP1_ENERGY_STATUS */
103 /* 0x642 MSR_PP1_POLICY */
104 #define TJMAX_DEFAULT 100
106 int aperf_mperf_unstable;
107 int backwards_count;
108 char *progname;
110 cpu_set_t *cpu_present_set, *cpu_affinity_set;
111 size_t cpu_present_setsize, cpu_affinity_setsize;
113 struct thread_data {
114 unsigned long long tsc;
115 unsigned long long aperf;
116 unsigned long long mperf;
117 unsigned long long c1;
118 unsigned int smi_count;
119 unsigned int cpu_id;
120 unsigned int flags;
121 #define CPU_IS_FIRST_THREAD_IN_CORE 0x2
122 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
123 } *thread_even, *thread_odd;
125 struct core_data {
126 unsigned long long c3;
127 unsigned long long c6;
128 unsigned long long c7;
129 unsigned int core_temp_c;
130 unsigned int core_id;
131 } *core_even, *core_odd;
133 struct pkg_data {
134 unsigned long long pc2;
135 unsigned long long pc3;
136 unsigned long long pc6;
137 unsigned long long pc7;
138 unsigned long long pc8;
139 unsigned long long pc9;
140 unsigned long long pc10;
141 unsigned int package_id;
142 unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */
143 unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */
144 unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */
145 unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */
146 unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */
147 unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */
148 unsigned int pkg_temp_c;
150 } *package_even, *package_odd;
152 #define ODD_COUNTERS thread_odd, core_odd, package_odd
153 #define EVEN_COUNTERS thread_even, core_even, package_even
154 static _Bool is_even = 1;
156 static _Bool allocated = 0;
157 static _Bool initialized = 0;
159 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
160 (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
161 topo.num_threads_per_core + \
162 (core_no) * topo.num_threads_per_core + (thread_no))
163 #define GET_CORE(core_base, core_no, pkg_no) \
164 (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
165 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
167 struct topo_params {
168 int num_packages;
169 int num_cpus;
170 int num_cores;
171 int max_cpu_num;
172 int num_cores_per_pkg;
173 int num_threads_per_core;
174 } topo;
176 struct timeval tv_even, tv_odd, tv_delta;
178 enum return_values {
179 OK = 0,
180 ERR_CPU_MIGRATE,
181 ERR_MSR_IA32_APERF,
182 ERR_MSR_IA32_MPERF,
183 ERR_MSR_SMI_COUNT,
184 ERR_MSR_CORE_C3_RESIDENCY,
185 ERR_MSR_CORE_C6_RESIDENCY,
186 ERR_MSR_CORE_C7_RESIDENCY,
187 ERR_MSR_IA32_THERM_STATUS,
188 ERR_MSR_PKG_C3_RESIDENCY,
189 ERR_MSR_PKG_C6_RESIDENCY,
190 ERR_MSR_PKG_C2_RESIDENCY,
191 ERR_MSR_PKG_C7_RESIDENCY,
192 ERR_MSR_PKG_C8_RESIDENCY,
193 ERR_MSR_PKG_C9_RESIDENCY,
194 ERR_MSR_PKG_C10_RESIDENCY,
195 ERR_MSR_PKG_ENERGY_STATUS,
196 ERR_MSR_PP0_ENERGY_STATUS,
197 ERR_MSR_DRAM_ENERGY_STATUS,
198 ERR_MSR_PP1_ENERGY_STATUS,
199 ERR_MSR_PKG_PERF_STATUS,
200 ERR_MSR_DRAM_PERF_STATUS,
201 ERR_MSR_IA32_PACKAGE_THERM_STATUS,
202 ERR_CPU_NOT_PRESENT,
203 ERR_NO_MSR,
204 ERR_CANT_OPEN_FILE,
205 ERR_CANT_READ_NUMBER,
206 ERR_CANT_READ_PROC_STAT,
207 ERR_NO_INVARIANT_TSC,
208 ERR_NO_APERF,
209 ERR_CALLOC,
210 ERR_CPU_ALLOC,
211 ERR_NOT_ROOT,
212 };
214 static int setup_all_buffers(void);
216 static int
217 cpu_is_not_present(int cpu)
218 {
219 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
220 }
221 /*
222 * run func(thread, core, package) in topology order
223 * skip non-present cpus
224 */
226 static int __attribute__((warn_unused_result))
227 for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
228 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
229 {
230 int retval, pkg_no, core_no, thread_no;
232 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
233 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
234 for (thread_no = 0; thread_no <
235 topo.num_threads_per_core; ++thread_no) {
236 struct thread_data *t;
237 struct core_data *c;
238 struct pkg_data *p;
240 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
242 if (cpu_is_not_present(t->cpu_id))
243 continue;
245 c = GET_CORE(core_base, core_no, pkg_no);
246 p = GET_PKG(pkg_base, pkg_no);
248 retval = func(t, c, p);
249 if (retval)
250 return retval;
251 }
252 }
253 }
254 return 0;
255 }
257 static int __attribute__((warn_unused_result))
258 cpu_migrate(int cpu)
259 {
260 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
261 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
262 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
263 return -ERR_CPU_MIGRATE;
264 else
265 return 0;
266 }
268 static int __attribute__((warn_unused_result))
269 get_msr(int cpu, off_t offset, unsigned long long *msr)
270 {
271 ssize_t retval;
272 char pathname[32];
273 int fd;
275 ssnprintf(pathname, 32, "/dev/cpu/%d/msr", cpu);
276 fd = open(pathname, O_RDONLY);
277 if (fd < 0)
278 return -1;
280 retval = pread(fd, msr, sizeof *msr, offset);
281 close(fd);
283 if (retval != sizeof *msr) {
284 ERROR ("%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
285 return -1;
286 }
288 return 0;
289 }
291 #define DELTA_WRAP32(new, old) \
292 if (new > old) { \
293 old = new - old; \
294 } else { \
295 old = 0x100000000 + new - old; \
296 }
298 static void
299 delta_package(struct pkg_data *new, struct pkg_data *old)
300 {
301 old->pc2 = new->pc2 - old->pc2;
302 old->pc3 = new->pc3 - old->pc3;
303 old->pc6 = new->pc6 - old->pc6;
304 old->pc7 = new->pc7 - old->pc7;
305 old->pc8 = new->pc8 - old->pc8;
306 old->pc9 = new->pc9 - old->pc9;
307 old->pc10 = new->pc10 - old->pc10;
308 old->pkg_temp_c = new->pkg_temp_c;
310 DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
311 DELTA_WRAP32(new->energy_cores, old->energy_cores);
312 DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
313 DELTA_WRAP32(new->energy_dram, old->energy_dram);
314 DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
315 DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
316 }
318 static void
319 delta_core(struct core_data *new, struct core_data *old)
320 {
321 old->c3 = new->c3 - old->c3;
322 old->c6 = new->c6 - old->c6;
323 old->c7 = new->c7 - old->c7;
324 old->core_temp_c = new->core_temp_c;
325 }
327 /*
328 * old = new - old
329 */
330 static int __attribute__((warn_unused_result))
331 delta_thread(struct thread_data *new, struct thread_data *old,
332 struct core_data *core_delta)
333 {
334 old->tsc = new->tsc - old->tsc;
336 /* check for TSC < 1 Mcycles over interval */
337 if (old->tsc < (1000 * 1000)) {
338 WARNING("Insanely slow TSC rate, TSC stops in idle?\n"
339 "You can disable all c-states by booting with \"idle=poll\"\n"
340 "or just the deep ones with \"processor.max_cstate=1\"");
341 return -1;
342 }
344 old->c1 = new->c1 - old->c1;
346 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
347 old->aperf = new->aperf - old->aperf;
348 old->mperf = new->mperf - old->mperf;
349 } else {
351 if (!aperf_mperf_unstable) {
352 WARNING("%s: APERF or MPERF went backwards *\n", progname);
353 WARNING("* Frequency results do not cover entire interval *\n");
354 WARNING("* fix this by running Linux-2.6.30 or later *\n");
356 aperf_mperf_unstable = 1;
357 }
358 /*
359 * mperf delta is likely a huge "positive" number
360 * can not use it for calculating c0 time
361 */
362 skip_c0 = 1;
363 skip_c1 = 1;
364 }
367 /*
368 * As counter collection is not atomic,
369 * it is possible for mperf's non-halted cycles + idle states
370 * to exceed TSC's all cycles: show c1 = 0% in that case.
371 */
372 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
373 old->c1 = 0;
374 else {
375 /* normal case, derive c1 */
376 old->c1 = old->tsc - old->mperf - core_delta->c3
377 - core_delta->c6 - core_delta->c7;
378 }
380 if (old->mperf == 0) {
381 WARNING("cpu%d MPERF 0!\n", old->cpu_id);
382 old->mperf = 1; /* divide by 0 protection */
383 }
385 if (do_smi)
386 old->smi_count = new->smi_count - old->smi_count;
388 return 0;
389 }
391 static int __attribute__((warn_unused_result))
392 delta_cpu(struct thread_data *t, struct core_data *c,
393 struct pkg_data *p, struct thread_data *t2,
394 struct core_data *c2, struct pkg_data *p2)
395 {
396 int ret;
398 /* calculate core delta only for 1st thread in core */
399 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
400 delta_core(c, c2);
402 /* always calculate thread delta */
403 ret = delta_thread(t, t2, c2); /* c2 is core delta */
404 if (ret != 0)
405 return ret;
407 /* calculate package delta only for 1st core in package */
408 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
409 delta_package(p, p2);
411 return 0;
412 }
414 static unsigned long long
415 rdtsc(void)
416 {
417 unsigned int low, high;
419 asm volatile("rdtsc" : "=a" (low), "=d" (high));
421 return low | ((unsigned long long)high) << 32;
422 }
425 /*
426 * get_counters(...)
427 * migrate to cpu
428 * acquire and record local counters for that cpu
429 */
430 static int __attribute__((warn_unused_result))
431 get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
432 {
433 int cpu = t->cpu_id;
434 unsigned long long msr;
436 if (cpu_migrate(cpu)) {
437 WARNING("Could not migrate to CPU %d\n", cpu);
438 return -ERR_CPU_MIGRATE;
439 }
441 t->tsc = rdtsc(); /* we are running on local CPU of interest */
443 if (has_aperf) {
444 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
445 return -ERR_MSR_IA32_APERF;
446 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
447 return -ERR_MSR_IA32_MPERF;
448 }
450 if (do_smi) {
451 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
452 return -ERR_MSR_SMI_COUNT;
453 t->smi_count = msr & 0xFFFFFFFF;
454 }
456 /* collect core counters only for 1st thread in core */
457 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
458 return 0;
460 if (do_nhm_cstates && !do_slm_cstates) {
461 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
462 return -ERR_MSR_CORE_C3_RESIDENCY;
463 }
465 if (do_nhm_cstates) {
466 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
467 return -ERR_MSR_CORE_C6_RESIDENCY;
468 }
470 if (do_snb_cstates)
471 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
472 return -ERR_MSR_CORE_C7_RESIDENCY;
474 if (do_dts) {
475 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
476 return -ERR_MSR_IA32_THERM_STATUS;
477 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
478 }
481 /* collect package counters only for 1st core in package */
482 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
483 return 0;
485 if (do_nhm_cstates && !do_slm_cstates) {
486 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
487 return -ERR_MSR_PKG_C3_RESIDENCY;
488 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
489 return -ERR_MSR_PKG_C6_RESIDENCY;
490 }
491 if (do_snb_cstates) {
492 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
493 return -ERR_MSR_PKG_C2_RESIDENCY;
494 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
495 return -ERR_MSR_PKG_C7_RESIDENCY;
496 }
497 if (do_c8_c9_c10) {
498 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
499 return -ERR_MSR_PKG_C8_RESIDENCY;
500 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
501 return -ERR_MSR_PKG_C9_RESIDENCY;
502 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
503 return -ERR_MSR_PKG_C10_RESIDENCY;
504 }
505 if (do_rapl & RAPL_PKG) {
506 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
507 return -ERR_MSR_PKG_ENERGY_STATUS;
508 p->energy_pkg = msr & 0xFFFFFFFF;
509 }
510 if (do_rapl & RAPL_CORES) {
511 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
512 return MSR_PP0_ENERGY_STATUS;
513 p->energy_cores = msr & 0xFFFFFFFF;
514 }
515 if (do_rapl & RAPL_DRAM) {
516 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
517 return -ERR_MSR_DRAM_ENERGY_STATUS;
518 p->energy_dram = msr & 0xFFFFFFFF;
519 }
520 if (do_rapl & RAPL_GFX) {
521 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
522 return -ERR_MSR_PP1_ENERGY_STATUS;
523 p->energy_gfx = msr & 0xFFFFFFFF;
524 }
525 if (do_rapl & RAPL_PKG_PERF_STATUS) {
526 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
527 return -ERR_MSR_PKG_PERF_STATUS;
528 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
529 }
530 if (do_rapl & RAPL_DRAM_PERF_STATUS) {
531 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
532 return -ERR_MSR_DRAM_PERF_STATUS;
533 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
534 }
535 if (do_ptm) {
536 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
537 return -ERR_MSR_IA32_PACKAGE_THERM_STATUS;
538 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
539 }
540 return 0;
541 }
543 static void
544 free_all_buffers(void)
545 {
546 allocated = 0;
547 initialized = 0;
549 CPU_FREE(cpu_present_set);
550 cpu_present_set = NULL;
551 cpu_present_set = 0;
553 CPU_FREE(cpu_affinity_set);
554 cpu_affinity_set = NULL;
555 cpu_affinity_setsize = 0;
557 free(thread_even);
558 free(core_even);
559 free(package_even);
561 thread_even = NULL;
562 core_even = NULL;
563 package_even = NULL;
565 free(thread_odd);
566 free(core_odd);
567 free(package_odd);
569 thread_odd = NULL;
570 core_odd = NULL;
571 package_odd = NULL;
572 }
574 /*
575 * Parse a file containing a single int.
576 */
577 static int __attribute__ ((format(printf,1,2)))
578 parse_int_file(const char *fmt, ...)
579 {
580 va_list args;
581 char path[PATH_MAX];
582 FILE *filep;
583 int value;
585 va_start(args, fmt);
586 vsnprintf(path, sizeof(path), fmt, args);
587 va_end(args);
588 filep = fopen(path, "r");
589 if (!filep) {
590 ERROR("%s: open failed", path);
591 return -ERR_CANT_OPEN_FILE;
592 }
593 if (fscanf(filep, "%d", &value) != 1) {
594 ERROR("%s: failed to parse number from file", path);
595 return -ERR_CANT_READ_NUMBER;
596 }
597 fclose(filep);
598 return value;
599 }
601 /*
602 * cpu_is_first_sibling_in_core(cpu)
603 * return 1 if given CPU is 1st HT sibling in the core
604 */
605 static int
606 cpu_is_first_sibling_in_core(int cpu)
607 {
608 return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
609 }
611 /*
612 * cpu_is_first_core_in_package(cpu)
613 * return 1 if given CPU is 1st core in package
614 */
615 static int
616 cpu_is_first_core_in_package(int cpu)
617 {
618 return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
619 }
621 static int
622 get_physical_package_id(int cpu)
623 {
624 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
625 }
627 static int
628 get_core_id(int cpu)
629 {
630 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
631 }
633 static int
634 get_num_ht_siblings(int cpu)
635 {
636 char path[80];
637 FILE *filep;
638 int sib1, sib2;
639 int matches;
640 char character;
642 ssnprintf(path, 80, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
643 filep = fopen(path, "r");
644 if (!filep) {
645 ERROR("%s: open failed", path);
646 return -ERR_CANT_OPEN_FILE;
647 }
648 /*
649 * file format:
650 * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
651 * otherwinse 1 sibling (self).
652 */
653 matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
655 fclose(filep);
657 if (matches == 3)
658 return 2;
659 else
660 return 1;
661 }
663 /*
664 * run func(thread, core, package) in topology order
665 * skip non-present cpus
666 */
669 static int __attribute__((warn_unused_result))
670 for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
671 struct pkg_data *, struct thread_data *, struct core_data *,
672 struct pkg_data *), struct thread_data *thread_base,
673 struct core_data *core_base, struct pkg_data *pkg_base,
674 struct thread_data *thread_base2, struct core_data *core_base2,
675 struct pkg_data *pkg_base2)
676 {
677 int retval, pkg_no, core_no, thread_no;
679 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
680 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
681 for (thread_no = 0; thread_no <
682 topo.num_threads_per_core; ++thread_no) {
683 struct thread_data *t, *t2;
684 struct core_data *c, *c2;
685 struct pkg_data *p, *p2;
687 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
689 if (cpu_is_not_present(t->cpu_id))
690 continue;
692 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
694 c = GET_CORE(core_base, core_no, pkg_no);
695 c2 = GET_CORE(core_base2, core_no, pkg_no);
697 p = GET_PKG(pkg_base, pkg_no);
698 p2 = GET_PKG(pkg_base2, pkg_no);
700 retval = func(t, c, p, t2, c2, p2);
701 if (retval)
702 return retval;
703 }
704 }
705 }
706 return 0;
707 }
709 /*
710 * run func(cpu) on every cpu in /proc/stat
711 * return max_cpu number
712 */
713 static int __attribute__((warn_unused_result))
714 for_all_proc_cpus(int (func)(int))
715 {
716 FILE *fp;
717 int cpu_num;
718 int retval;
720 fp = fopen(proc_stat, "r");
721 if (!fp) {
722 ERROR("%s: open failed", proc_stat);
723 return -ERR_CANT_OPEN_FILE;
724 }
726 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
727 if (retval != 0) {
728 ERROR("%s: failed to parse format", proc_stat);
729 return -ERR_CANT_READ_PROC_STAT;
730 }
732 while (1) {
733 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
734 if (retval != 1)
735 break;
737 retval = func(cpu_num);
738 if (retval) {
739 fclose(fp);
740 return(retval);
741 }
742 }
743 fclose(fp);
744 return 0;
745 }
747 /*
748 * count_cpus()
749 * remember the last one seen, it will be the max
750 */
751 static int
752 count_cpus(int cpu)
753 {
754 if (topo.max_cpu_num < cpu)
755 topo.max_cpu_num = cpu;
757 topo.num_cpus += 1;
758 return 0;
759 }
760 static int
761 mark_cpu_present(int cpu)
762 {
763 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
764 return 0;
765 }
768 static void
769 turbostat_submit (const char *plugin_instance,
770 const char *type, const char *type_instance,
771 gauge_t value)
772 {
773 value_list_t vl = VALUE_LIST_INIT;
774 value_t v;
776 v.gauge = value;
777 vl.values = &v;
778 vl.values_len = 1;
779 sstrncpy (vl.host, hostname_g, sizeof (vl.host));
780 sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
781 if (plugin_instance != NULL)
782 sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
783 sstrncpy (vl.type, type, sizeof (vl.type));
784 if (type_instance != NULL)
785 sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
787 plugin_dispatch_values (&vl);
788 }
790 /*
791 * column formatting convention & formats
792 * package: "pk" 2 columns %2d
793 * core: "cor" 3 columns %3d
794 * CPU: "CPU" 3 columns %3d
795 * Pkg_W: %6.2
796 * Cor_W: %6.2
797 * GFX_W: %5.2
798 * RAM_W: %5.2
799 * GHz: "GHz" 3 columns %3.2
800 * TSC: "TSC" 3 columns %3.2
801 * SMI: "SMI" 4 columns %4d
802 * percentage " %pc3" %6.2
803 * Perf Status percentage: %5.2
804 * "CTMP" 4 columns %4d
805 */
806 #define NAME_LEN 12
807 static int
808 submit_counters(struct thread_data *t, struct core_data *c,
809 struct pkg_data *p)
810 {
811 char name[NAME_LEN];
812 double interval_float;
814 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
816 snprintf(name, NAME_LEN, "cpu%02d", t->cpu_id);
818 if (do_nhm_cstates) {
819 if (!skip_c0)
820 turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
821 if (!skip_c1)
822 turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
823 }
825 /* GHz */
826 if (has_aperf && ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc))))
827 turbostat_submit(NULL, "frequency", name, 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
829 /* SMI */
830 if (do_smi)
831 turbostat_submit(NULL, "current", name, t->smi_count);
833 /* print per-core data only for 1st thread in core */
834 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
835 goto done;
837 snprintf(name, NAME_LEN, "core%02d", c->core_id);
839 if (do_nhm_cstates && !do_slm_cstates)
840 turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
841 if (do_nhm_cstates)
842 turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
843 if (do_snb_cstates)
844 turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
846 if (do_dts)
847 turbostat_submit(NULL, "temperature", name, c->core_temp_c);
849 /* print per-package data only for 1st core in package */
850 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
851 goto done;
853 snprintf(name, NAME_LEN, "pkg%02d", p->package_id);
855 if (do_ptm)
856 turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
858 if (do_snb_cstates)
859 turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
860 if (do_nhm_cstates && !do_slm_cstates)
861 turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
862 if (do_nhm_cstates && !do_slm_cstates)
863 turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
864 if (do_snb_cstates)
865 turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
866 if (do_c8_c9_c10) {
867 turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
868 turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
869 turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
870 }
872 if (do_rapl) {
873 if (do_rapl & RAPL_PKG)
874 turbostat_submit(name, "power", "Pkg_W", p->energy_pkg * rapl_energy_units / interval_float);
875 if (do_rapl & RAPL_CORES)
876 turbostat_submit(name, "power", "Cor_W", p->energy_cores * rapl_energy_units / interval_float);
877 if (do_rapl & RAPL_GFX)
878 turbostat_submit(name, "power", "GFX_W", p->energy_gfx * rapl_energy_units / interval_float);
879 if (do_rapl & RAPL_DRAM)
880 turbostat_submit(name, "power", "RAM_W", p->energy_dram * rapl_energy_units / interval_float);
881 }
882 done:
883 return 0;
884 }
886 static int
887 turbostat_read(user_data_t * not_used)
888 {
889 int ret;
891 if (!allocated) {
892 if ((ret = setup_all_buffers()) < 0)
893 return ret;
894 }
896 if (for_all_proc_cpus(cpu_is_not_present)) {
897 free_all_buffers();
898 if ((ret = setup_all_buffers()) < 0)
899 return ret;
900 if (for_all_proc_cpus(cpu_is_not_present))
901 return -ERR_CPU_NOT_PRESENT;
902 }
904 if (!initialized) {
905 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
906 return ret;
907 gettimeofday(&tv_even, (struct timezone *)NULL);
908 is_even = 1;
909 initialized = 1;
910 return 0;
911 }
913 if (is_even) {
914 if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
915 return ret;
916 gettimeofday(&tv_odd, (struct timezone *)NULL);
917 is_even = 0;
918 timersub(&tv_odd, &tv_even, &tv_delta);
919 if ((ret = for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) < 0)
920 return ret;
921 if ((ret = for_all_cpus(submit_counters, EVEN_COUNTERS)) < 0)
922 return ret;
923 } else {
924 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
925 return ret;
926 gettimeofday(&tv_even, (struct timezone *)NULL);
927 is_even = 1;
928 timersub(&tv_even, &tv_odd, &tv_delta);
929 if ((ret = for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) < 0)
930 return ret;
931 if ((ret = for_all_cpus(submit_counters, ODD_COUNTERS)) < 0)
932 return ret;
933 }
934 return 0;
935 }
937 static int __attribute__((warn_unused_result))
938 check_dev_msr()
939 {
940 struct stat sb;
942 if (stat("/dev/cpu/0/msr", &sb)) {
943 ERROR("no /dev/cpu/0/msr\n"
944 "Try \"# modprobe msr\"");
945 return -ERR_NO_MSR;
946 }
947 return 0;
948 }
950 static int __attribute__((warn_unused_result))
951 check_super_user()
952 {
953 if (getuid() != 0) {
954 ERROR("must be root");
955 return -ERR_NOT_ROOT;
956 }
957 return 0;
958 }
961 #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */
962 #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */
964 static double
965 get_tdp(unsigned int model)
966 {
967 unsigned long long msr;
969 if (do_rapl & RAPL_PKG_POWER_INFO)
970 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
971 return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
973 switch (model) {
974 case 0x37:
975 case 0x4D:
976 return 30.0;
977 default:
978 return 135.0;
979 }
980 }
983 /*
984 * rapl_probe()
985 *
986 * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
987 */
988 static void
989 rapl_probe(unsigned int family, unsigned int model)
990 {
991 unsigned long long msr;
992 unsigned int time_unit;
993 double tdp;
995 if (!genuine_intel)
996 return;
998 if (family != 6)
999 return;
1001 switch (model) {
1002 case 0x2A:
1003 case 0x3A:
1004 case 0x3C: /* HSW */
1005 case 0x45: /* HSW */
1006 case 0x46: /* HSW */
1007 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1008 break;
1009 case 0x3F: /* HSX */
1010 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1011 break;
1012 case 0x2D:
1013 case 0x3E:
1014 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1015 break;
1016 case 0x37: /* BYT */
1017 case 0x4D: /* AVN */
1018 do_rapl = RAPL_PKG | RAPL_CORES ;
1019 break;
1020 default:
1021 return;
1022 }
1024 /* units on package 0, verify later other packages match */
1025 if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1026 return;
1028 rapl_power_units = 1.0 / (1 << (msr & 0xF));
1029 if (model == 0x37)
1030 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1031 else
1032 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1034 time_unit = msr >> 16 & 0xF;
1035 if (time_unit == 0)
1036 time_unit = 0xA;
1038 rapl_time_units = 1.0 / (1 << (time_unit));
1040 tdp = get_tdp(model);
1042 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1043 // if (verbose)
1044 // fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1046 return;
1047 }
1049 static int
1050 is_snb(unsigned int family, unsigned int model)
1051 {
1052 if (!genuine_intel)
1053 return 0;
1055 switch (model) {
1056 case 0x2A:
1057 case 0x2D:
1058 case 0x3A: /* IVB */
1059 case 0x3E: /* IVB Xeon */
1060 case 0x3C: /* HSW */
1061 case 0x3F: /* HSW */
1062 case 0x45: /* HSW */
1063 case 0x46: /* HSW */
1064 return 1;
1065 }
1066 return 0;
1067 }
1069 static int
1070 has_c8_c9_c10(unsigned int family, unsigned int model)
1071 {
1072 if (!genuine_intel)
1073 return 0;
1075 switch (model) {
1076 case 0x45:
1077 return 1;
1078 }
1079 return 0;
1080 }
1083 static int
1084 is_slm(unsigned int family, unsigned int model)
1085 {
1086 if (!genuine_intel)
1087 return 0;
1088 switch (model) {
1089 case 0x37: /* BYT */
1090 case 0x4D: /* AVN */
1091 return 1;
1092 }
1093 return 0;
1094 }
1096 /*
1097 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1098 * the Thermal Control Circuit (TCC) activates.
1099 * This is usually equal to tjMax.
1100 *
1101 * Older processors do not have this MSR, so there we guess,
1102 * but also allow cmdline over-ride with -T.
1103 *
1104 * Several MSR temperature values are in units of degrees-C
1105 * below this value, including the Digital Thermal Sensor (DTS),
1106 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1107 */
1108 static int __attribute__((warn_unused_result))
1109 set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1110 {
1111 unsigned long long msr;
1112 unsigned int target_c_local;
1113 int cpu;
1115 /* tcc_activation_temp is used only for dts or ptm */
1116 if (!(do_dts || do_ptm))
1117 return 0;
1119 /* this is a per-package concept */
1120 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1121 return 0;
1123 cpu = t->cpu_id;
1124 if (cpu_migrate(cpu)) {
1125 ERROR("Could not migrate to CPU %d\n", cpu);
1126 return -ERR_CPU_MIGRATE;
1127 }
1129 if (tcc_activation_temp_override != 0) {
1130 tcc_activation_temp = tcc_activation_temp_override;
1131 ERROR("cpu%d: Using cmdline TCC Target (%d C)\n",
1132 cpu, tcc_activation_temp);
1133 return 0;
1134 }
1136 /* Temperature Target MSR is Nehalem and newer only */
1137 if (!do_nehalem_platform_info)
1138 goto guess;
1140 if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1141 goto guess;
1143 target_c_local = (msr >> 16) & 0x7F;
1145 if (target_c_local < 85 || target_c_local > 127)
1146 goto guess;
1148 tcc_activation_temp = target_c_local;
1150 return 0;
1152 guess:
1153 tcc_activation_temp = TJMAX_DEFAULT;
1154 WARNING("cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
1155 cpu, tcc_activation_temp);
1157 return 0;
1158 }
1160 static int __attribute__((warn_unused_result))
1161 check_cpuid()
1162 {
1163 unsigned int eax, ebx, ecx, edx, max_level;
1164 unsigned int fms, family, model;
1166 eax = ebx = ecx = edx = 0;
1168 __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
1170 if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
1171 genuine_intel = 1;
1173 fms = 0;
1174 __get_cpuid(1, &fms, &ebx, &ecx, &edx);
1175 family = (fms >> 8) & 0xf;
1176 model = (fms >> 4) & 0xf;
1177 if (family == 6 || family == 0xf)
1178 model += ((fms >> 16) & 0xf) << 4;
1180 if (!(edx & (1 << 5))) {
1181 ERROR("CPUID: no MSR");
1182 return -ERR_NO_MSR;
1183 }
1185 /*
1186 * check max extended function levels of CPUID.
1187 * This is needed to check for invariant TSC.
1188 * This check is valid for both Intel and AMD.
1189 */
1190 ebx = ecx = edx = 0;
1191 __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
1193 if (max_level < 0x80000007) {
1194 ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level);
1195 return -ERR_NO_INVARIANT_TSC;
1196 }
1198 /*
1199 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
1200 * this check is valid for both Intel and AMD
1201 */
1202 __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1203 has_invariant_tsc = edx & (1 << 8);
1205 if (!has_invariant_tsc) {
1206 ERROR("No invariant TSC");
1207 return -ERR_NO_INVARIANT_TSC;
1208 }
1210 /*
1211 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
1212 * this check is valid for both Intel and AMD
1213 */
1215 __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
1216 has_aperf = ecx & (1 << 0);
1217 do_dts = eax & (1 << 0);
1218 do_ptm = eax & (1 << 6);
1219 has_epb = ecx & (1 << 3);
1221 if (!has_aperf) {
1222 ERROR("No APERF");
1223 return -ERR_NO_APERF;
1224 }
1226 do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
1227 do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
1228 do_smi = do_nhm_cstates;
1229 do_snb_cstates = is_snb(family, model);
1230 do_c8_c9_c10 = has_c8_c9_c10(family, model);
1231 do_slm_cstates = is_slm(family, model);
1233 rapl_probe(family, model);
1235 return 0;
1236 }
1240 static int __attribute__((warn_unused_result))
1241 topology_probe()
1242 {
1243 int i;
1244 int ret;
1245 int max_core_id = 0;
1246 int max_package_id = 0;
1247 int max_siblings = 0;
1248 struct cpu_topology {
1249 int core_id;
1250 int physical_package_id;
1251 } *cpus;
1253 /* Initialize num_cpus, max_cpu_num */
1254 topo.num_cpus = 0;
1255 topo.max_cpu_num = 0;
1256 ret = for_all_proc_cpus(count_cpus);
1257 if (ret < 0)
1258 return ret;
1259 if (topo.num_cpus > 1)
1260 show_cpu = 1;
1262 DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
1264 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
1265 if (cpus == NULL) {
1266 ERROR("calloc cpus");
1267 return -ERR_CALLOC;
1268 }
1270 /*
1271 * Allocate and initialize cpu_present_set
1272 */
1273 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
1274 if (cpu_present_set == NULL) {
1275 free(cpus);
1276 ERROR("CPU_ALLOC");
1277 return -ERR_CPU_ALLOC;
1278 }
1279 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1280 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
1281 ret = for_all_proc_cpus(mark_cpu_present);
1282 if (ret < 0) {
1283 free(cpus);
1284 return ret;
1285 }
1287 /*
1288 * Allocate and initialize cpu_affinity_set
1289 */
1290 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
1291 if (cpu_affinity_set == NULL) {
1292 free(cpus);
1293 ERROR("CPU_ALLOC");
1294 return -ERR_CPU_ALLOC;
1295 }
1296 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1297 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
1300 /*
1301 * For online cpus
1302 * find max_core_id, max_package_id
1303 */
1304 for (i = 0; i <= topo.max_cpu_num; ++i) {
1305 int siblings;
1307 if (cpu_is_not_present(i)) {
1308 //if (verbose > 1)
1309 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
1310 continue;
1311 }
1312 cpus[i].core_id = get_core_id(i);
1313 if (cpus[i].core_id < 0)
1314 return cpus[i].core_id;
1315 if (cpus[i].core_id > max_core_id)
1316 max_core_id = cpus[i].core_id;
1318 cpus[i].physical_package_id = get_physical_package_id(i);
1319 if (cpus[i].physical_package_id < 0)
1320 return cpus[i].physical_package_id;
1321 if (cpus[i].physical_package_id > max_package_id)
1322 max_package_id = cpus[i].physical_package_id;
1324 siblings = get_num_ht_siblings(i);
1325 if (siblings < 0)
1326 return siblings;
1327 if (siblings > max_siblings)
1328 max_siblings = siblings;
1329 DEBUG("cpu %d pkg %d core %d\n",
1330 i, cpus[i].physical_package_id, cpus[i].core_id);
1331 }
1332 topo.num_cores_per_pkg = max_core_id + 1;
1333 DEBUG("max_core_id %d, sizing for %d cores per package\n",
1334 max_core_id, topo.num_cores_per_pkg);
1335 if (topo.num_cores_per_pkg > 1)
1336 show_core = 1;
1338 topo.num_packages = max_package_id + 1;
1339 DEBUG("max_package_id %d, sizing for %d packages\n",
1340 max_package_id, topo.num_packages);
1341 if (topo.num_packages > 1)
1342 show_pkg = 1;
1344 topo.num_threads_per_core = max_siblings;
1345 DEBUG("max_siblings %d\n", max_siblings);
1347 free(cpus);
1348 return 0;
1349 }
1351 static int
1352 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
1353 {
1354 int i;
1356 *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
1357 topo.num_packages, sizeof(struct thread_data));
1358 if (*t == NULL)
1359 goto error;
1361 for (i = 0; i < topo.num_threads_per_core *
1362 topo.num_cores_per_pkg * topo.num_packages; i++)
1363 (*t)[i].cpu_id = -1;
1365 *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
1366 sizeof(struct core_data));
1367 if (*c == NULL)
1368 goto error;
1370 for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
1371 (*c)[i].core_id = -1;
1373 *p = calloc(topo.num_packages, sizeof(struct pkg_data));
1374 if (*p == NULL)
1375 goto error;
1377 for (i = 0; i < topo.num_packages; i++)
1378 (*p)[i].package_id = i;
1380 return 0;
1381 error:
1382 ERROR("calloc counters");
1383 return -ERR_CALLOC;
1384 }
1385 /*
1386 * init_counter()
1387 *
1388 * set cpu_id, core_num, pkg_num
1389 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
1390 *
1391 * increment topo.num_cores when 1st core in pkg seen
1392 */
1393 static int
1394 init_counter(struct thread_data *thread_base, struct core_data *core_base,
1395 struct pkg_data *pkg_base, int thread_num, int core_num,
1396 int pkg_num, int cpu_id)
1397 {
1398 int ret;
1399 struct thread_data *t;
1400 struct core_data *c;
1401 struct pkg_data *p;
1403 t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
1404 c = GET_CORE(core_base, core_num, pkg_num);
1405 p = GET_PKG(pkg_base, pkg_num);
1407 t->cpu_id = cpu_id;
1408 if (thread_num == 0) {
1409 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1410 if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
1411 return ret;
1412 } else if (ret != 0) {
1413 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1414 }
1415 }
1417 c->core_id = core_num;
1418 p->package_id = pkg_num;
1420 return 0;
1421 }
1424 static int
1425 initialize_counters(int cpu_id)
1426 {
1427 int my_thread_id, my_core_id, my_package_id;
1428 int ret;
1430 my_package_id = get_physical_package_id(cpu_id);
1431 if (my_package_id < 0)
1432 return my_package_id;
1433 my_core_id = get_core_id(cpu_id);
1434 if (my_core_id < 0)
1435 return my_core_id;
1437 if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
1438 return ret;
1439 } else if (ret != 0) {
1440 my_thread_id = 0;
1441 topo.num_cores++;
1442 } else {
1443 my_thread_id = 1;
1444 }
1446 ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1447 if (ret < 0)
1448 return ret;
1449 ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1450 if (ret < 0)
1451 return ret;
1452 return 0;
1453 }
1455 #define DO_OR_GOTO_ERR(something) \
1456 do { \
1457 ret = (something); \
1458 if (ret < 0) \
1459 goto err; \
1460 } while (0)
1462 static int setup_all_buffers(void)
1463 {
1464 int ret;
1466 DO_OR_GOTO_ERR(topology_probe());
1467 DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
1468 DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
1469 DO_OR_GOTO_ERR(for_all_proc_cpus(initialize_counters));
1471 allocated = 1;
1472 return 0;
1473 err:
1474 free_all_buffers();
1475 return ret;
1476 }
1478 static int
1479 turbostat_init(void)
1480 {
1481 int ret;
1483 DO_OR_GOTO_ERR(check_cpuid());
1484 DO_OR_GOTO_ERR(check_dev_msr());
1485 DO_OR_GOTO_ERR(check_super_user());
1486 DO_OR_GOTO_ERR(setup_all_buffers());
1487 DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
1489 plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, NULL, NULL);
1491 return 0;
1492 err:
1493 free_all_buffers();
1494 return ret;
1495 }
1497 void module_register(void);
1498 void module_register(void)
1499 {
1500 plugin_register_init(PLUGIN_NAME, turbostat_init);
1501 }