]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - tools/power/x86/turbostat/turbostat.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
turbostat: Check return value of fscanf
[mirror_ubuntu-artful-kernel.git] / tools / power / x86 / turbostat / turbostat.c
CommitLineData
103a8fea
LB		 1/*
2 * turbostat -- show CPU frequency and C-state residency
3 * on modern Intel turbo-capable processors.
4 *
144b44b1 5 * Copyright (c) 2013 Intel Corporation.
103a8fea
LB		 6 * Len Brown <len.brown@intel.com>
7 *
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.
11 *
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
15 * more details.
16 *
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.
20 */
21
88c3281f 22#define _GNU_SOURCE
b731f311 23#include MSRHEADER
103a8fea
LB		 24#include <stdio.h>
25#include <unistd.h>
26#include <sys/types.h>
27#include <sys/wait.h>
28#include <sys/stat.h>
29#include <sys/resource.h>
30#include <fcntl.h>
31#include <signal.h>
32#include <sys/time.h>
33#include <stdlib.h>
34#include <dirent.h>
35#include <string.h>
36#include <ctype.h>
88c3281f 37#include <sched.h>
2b92865e 38#include <cpuid.h>
103a8fea 39
103a8fea
LB		 40char *proc_stat = "/proc/stat";
41unsigned int interval_sec = 5; /* set with -i interval_sec */
42unsigned int verbose; /* set with -v */
889facbe
LB		 43unsigned int rapl_verbose; /* set with -R */
44unsigned int thermal_verbose; /* set with -T */
e23da037 45unsigned int summary_only; /* set with -s */
103a8fea
LB		 46unsigned int skip_c0;
47unsigned int skip_c1;
48unsigned int do_nhm_cstates;
49unsigned int do_snb_cstates;
ca58710f 50unsigned int do_c8_c9_c10;
144b44b1
LB		 51unsigned int do_slm_cstates;
52unsigned int use_c1_residency_msr;
103a8fea 53unsigned int has_aperf;
889facbe 54unsigned int has_epb;
103a8fea
LB		 55unsigned int units = 1000000000; /* Ghz etc */
56unsigned int genuine_intel;
57unsigned int has_invariant_tsc;
58unsigned int do_nehalem_platform_info;
59unsigned int do_nehalem_turbo_ratio_limit;
6574a5d5 60unsigned int do_ivt_turbo_ratio_limit;
2f32edf1
LB		 61unsigned int extra_msr_offset32;
62unsigned int extra_msr_offset64;
8e180f3c
LB		 63unsigned int extra_delta_offset32;
64unsigned int extra_delta_offset64;
1ed51011 65int do_smi;
103a8fea
LB		 66double bclk;
67unsigned int show_pkg;
68unsigned int show_core;
69unsigned int show_cpu;
c98d5d94
LB		 70unsigned int show_pkg_only;
71unsigned int show_core_only;
72char *output_buffer, *outp;
889facbe
LB		 73unsigned int do_rapl;
74unsigned int do_dts;
75unsigned int do_ptm;
76unsigned int tcc_activation_temp;
77unsigned int tcc_activation_temp_override;
78double rapl_power_units, rapl_energy_units, rapl_time_units;
79double rapl_joule_counter_range;
80
81#define RAPL_PKG (1 << 0)
82#define RAPL_CORES (1 << 1)
83#define RAPL_GFX (1 << 2)
84#define RAPL_DRAM (1 << 3)
85#define RAPL_PKG_PERF_STATUS (1 << 4)
86#define RAPL_DRAM_PERF_STATUS (1 << 5)
144b44b1
LB		 87#define RAPL_PKG_POWER_INFO (1 << 6)
88#define RAPL_CORE_POLICY (1 << 7)
889facbe
LB		 89#define TJMAX_DEFAULT 100
90
91#define MAX(a, b) ((a) > (b) ? (a) : (b))
103a8fea
LB		 92
93int aperf_mperf_unstable;
94int backwards_count;
95char *progname;
103a8fea 96
c98d5d94
LB		 97cpu_set_t *cpu_present_set, *cpu_affinity_set;
98size_t cpu_present_setsize, cpu_affinity_setsize;
99
100struct thread_data {
101 unsigned long long tsc;
102 unsigned long long aperf;
103 unsigned long long mperf;
144b44b1 104 unsigned long long c1;
2f32edf1 105 unsigned long long extra_msr64;
8e180f3c
LB		 106 unsigned long long extra_delta64;
107 unsigned long long extra_msr32;
108 unsigned long long extra_delta32;
1ed51011 109 unsigned int smi_count;
c98d5d94
LB		 110 unsigned int cpu_id;
111 unsigned int flags;
112#define CPU_IS_FIRST_THREAD_IN_CORE 0x2
113#define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
114} *thread_even, *thread_odd;
115
116struct core_data {
117 unsigned long long c3;
118 unsigned long long c6;
119 unsigned long long c7;
889facbe 120 unsigned int core_temp_c;
c98d5d94
LB		 121 unsigned int core_id;
122} *core_even, *core_odd;
123
124struct pkg_data {
125 unsigned long long pc2;
126 unsigned long long pc3;
127 unsigned long long pc6;
128 unsigned long long pc7;
ca58710f
KCA		 129 unsigned long long pc8;
130 unsigned long long pc9;
131 unsigned long long pc10;
c98d5d94 132 unsigned int package_id;
889facbe
LB		 133 unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */
134 unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */
135 unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */
136 unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */
137 unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */
138 unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */
139 unsigned int pkg_temp_c;
140
c98d5d94
LB		 141} *package_even, *package_odd;
142
143#define ODD_COUNTERS thread_odd, core_odd, package_odd
144#define EVEN_COUNTERS thread_even, core_even, package_even
145
146#define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
147 (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
148 topo.num_threads_per_core + \
149 (core_no) * topo.num_threads_per_core + (thread_no))
150#define GET_CORE(core_base, core_no, pkg_no) \
151 (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
152#define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
153
154struct system_summary {
155 struct thread_data threads;
156 struct core_data cores;
157 struct pkg_data packages;
158} sum, average;
159
160
161struct topo_params {
162 int num_packages;
163 int num_cpus;
164 int num_cores;
165 int max_cpu_num;
166 int num_cores_per_pkg;
167 int num_threads_per_core;
168} topo;
169
170struct timeval tv_even, tv_odd, tv_delta;
171
172void setup_all_buffers(void);
173
174int cpu_is_not_present(int cpu)
d15cf7c1 175{
c98d5d94 176 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
d15cf7c1 177}
88c3281f 178/*
c98d5d94
LB		 179 * run func(thread, core, package) in topology order
180 * skip non-present cpus
88c3281f 181 */
c98d5d94
LB		 182
183int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
184 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
88c3281f 185{
c98d5d94 186 int retval, pkg_no, core_no, thread_no;
d15cf7c1 187
c98d5d94
LB		 188 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
189 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
190 for (thread_no = 0; thread_no <
191 topo.num_threads_per_core; ++thread_no) {
192 struct thread_data *t;
193 struct core_data *c;
194 struct pkg_data *p;
88c3281f 195
c98d5d94
LB		 196 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
197
198 if (cpu_is_not_present(t->cpu_id))
199 continue;
200
201 c = GET_CORE(core_base, core_no, pkg_no);
202 p = GET_PKG(pkg_base, pkg_no);
203
204 retval = func(t, c, p);
205 if (retval)
206 return retval;
207 }
208 }
209 }
210 return 0;
88c3281f
LB		 211}
212
213int cpu_migrate(int cpu)
214{
c98d5d94
LB		 215 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
216 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
217 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
88c3281f
LB		 218 return -1;
219 else
220 return 0;
221}
222
15aaa346 223int get_msr(int cpu, off_t offset, unsigned long long *msr)
103a8fea
LB		 224{
225 ssize_t retval;
103a8fea
LB		 226 char pathname[32];
227 int fd;
228
229 sprintf(pathname, "/dev/cpu/%d/msr", cpu);
230 fd = open(pathname, O_RDONLY);
15aaa346
LB		 231 if (fd < 0)
232 return -1;
103a8fea 233
15aaa346 234 retval = pread(fd, msr, sizeof *msr, offset);
103a8fea 235 close(fd);
15aaa346 236
d91bb17c 237 if (retval != sizeof *msr) {
2e9c6bc7 238 fprintf(stderr, "%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
15aaa346 239 return -1;
d91bb17c 240 }
15aaa346
LB		 241
242 return 0;
103a8fea
LB		 243}
244
a829eb4d 245void print_header(void)
103a8fea
LB		 246{
247 if (show_pkg)
c98d5d94 248 outp += sprintf(outp, "pk");
e23da037 249 if (show_pkg)
c98d5d94 250 outp += sprintf(outp, " ");
103a8fea 251 if (show_core)
c98d5d94 252 outp += sprintf(outp, "cor");
103a8fea 253 if (show_cpu)
c98d5d94 254 outp += sprintf(outp, " CPU");
e23da037 255 if (show_pkg || show_core || show_cpu)
c98d5d94 256 outp += sprintf(outp, " ");
103a8fea 257 if (do_nhm_cstates)
c98d5d94 258 outp += sprintf(outp, " %%c0");
103a8fea 259 if (has_aperf)
c98d5d94
LB		 260 outp += sprintf(outp, " GHz");
261 outp += sprintf(outp, " TSC");
1ed51011
LB		 262 if (do_smi)
263 outp += sprintf(outp, " SMI");
8e180f3c 264 if (extra_delta_offset32)
f9240813 265 outp += sprintf(outp, " count 0x%03X", extra_delta_offset32);
8e180f3c 266 if (extra_delta_offset64)
f9240813 267 outp += sprintf(outp, " COUNT 0x%03X", extra_delta_offset64);
2f32edf1 268 if (extra_msr_offset32)
8e180f3c 269 outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32);
2f32edf1 270 if (extra_msr_offset64)
8e180f3c 271 outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64);
103a8fea 272 if (do_nhm_cstates)
c98d5d94 273 outp += sprintf(outp, " %%c1");
144b44b1 274 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 275 outp += sprintf(outp, " %%c3");
103a8fea 276 if (do_nhm_cstates)
c98d5d94 277 outp += sprintf(outp, " %%c6");
103a8fea 278 if (do_snb_cstates)
c98d5d94 279 outp += sprintf(outp, " %%c7");
889facbe
LB		 280
281 if (do_dts)
282 outp += sprintf(outp, " CTMP");
283 if (do_ptm)
284 outp += sprintf(outp, " PTMP");
285
103a8fea 286 if (do_snb_cstates)
c98d5d94 287 outp += sprintf(outp, " %%pc2");
144b44b1 288 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 289 outp += sprintf(outp, " %%pc3");
144b44b1 290 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 291 outp += sprintf(outp, " %%pc6");
103a8fea 292 if (do_snb_cstates)
c98d5d94 293 outp += sprintf(outp, " %%pc7");
ca58710f
KCA		 294 if (do_c8_c9_c10) {
295 outp += sprintf(outp, " %%pc8");
296 outp += sprintf(outp, " %%pc9");
297 outp += sprintf(outp, " %%pc10");
298 }
103a8fea 299
889facbe
LB		 300 if (do_rapl & RAPL_PKG)
301 outp += sprintf(outp, " Pkg_W");
302 if (do_rapl & RAPL_CORES)
303 outp += sprintf(outp, " Cor_W");
304 if (do_rapl & RAPL_GFX)
305 outp += sprintf(outp, " GFX_W");
306 if (do_rapl & RAPL_DRAM)
307 outp += sprintf(outp, " RAM_W");
308 if (do_rapl & RAPL_PKG_PERF_STATUS)
309 outp += sprintf(outp, " PKG_%%");
310 if (do_rapl & RAPL_DRAM_PERF_STATUS)
311 outp += sprintf(outp, " RAM_%%");
312
c98d5d94 313 outp += sprintf(outp, "\n");
103a8fea
LB		 314}
315
c98d5d94
LB		 316int dump_counters(struct thread_data *t, struct core_data *c,
317 struct pkg_data *p)
103a8fea 318{
c98d5d94
LB		 319 fprintf(stderr, "t %p, c %p, p %p\n", t, c, p);
320
321 if (t) {
322 fprintf(stderr, "CPU: %d flags 0x%x\n", t->cpu_id, t->flags);
323 fprintf(stderr, "TSC: %016llX\n", t->tsc);
324 fprintf(stderr, "aperf: %016llX\n", t->aperf);
325 fprintf(stderr, "mperf: %016llX\n", t->mperf);
326 fprintf(stderr, "c1: %016llX\n", t->c1);
8e180f3c
LB		 327 fprintf(stderr, "msr0x%x: %08llX\n",
328 extra_delta_offset32, t->extra_delta32);
329 fprintf(stderr, "msr0x%x: %016llX\n",
330 extra_delta_offset64, t->extra_delta64);
331 fprintf(stderr, "msr0x%x: %08llX\n",
2f32edf1 332 extra_msr_offset32, t->extra_msr32);
c98d5d94 333 fprintf(stderr, "msr0x%x: %016llX\n",
2f32edf1 334 extra_msr_offset64, t->extra_msr64);
1ed51011
LB		 335 if (do_smi)
336 fprintf(stderr, "SMI: %08X\n", t->smi_count);
c98d5d94 337 }
103a8fea 338
c98d5d94
LB		 339 if (c) {
340 fprintf(stderr, "core: %d\n", c->core_id);
341 fprintf(stderr, "c3: %016llX\n", c->c3);
342 fprintf(stderr, "c6: %016llX\n", c->c6);
343 fprintf(stderr, "c7: %016llX\n", c->c7);
889facbe 344 fprintf(stderr, "DTS: %dC\n", c->core_temp_c);
c98d5d94 345 }
103a8fea 346
c98d5d94
LB		 347 if (p) {
348 fprintf(stderr, "package: %d\n", p->package_id);
349 fprintf(stderr, "pc2: %016llX\n", p->pc2);
350 fprintf(stderr, "pc3: %016llX\n", p->pc3);
351 fprintf(stderr, "pc6: %016llX\n", p->pc6);
352 fprintf(stderr, "pc7: %016llX\n", p->pc7);
ca58710f
KCA		 353 fprintf(stderr, "pc8: %016llX\n", p->pc8);
354 fprintf(stderr, "pc9: %016llX\n", p->pc9);
355 fprintf(stderr, "pc10: %016llX\n", p->pc10);
889facbe
LB		 356 fprintf(stderr, "Joules PKG: %0X\n", p->energy_pkg);
357 fprintf(stderr, "Joules COR: %0X\n", p->energy_cores);
358 fprintf(stderr, "Joules GFX: %0X\n", p->energy_gfx);
359 fprintf(stderr, "Joules RAM: %0X\n", p->energy_dram);
360 fprintf(stderr, "Throttle PKG: %0X\n", p->rapl_pkg_perf_status);
361 fprintf(stderr, "Throttle RAM: %0X\n", p->rapl_dram_perf_status);
362 fprintf(stderr, "PTM: %dC\n", p->pkg_temp_c);
c98d5d94
LB		 363 }
364 return 0;
103a8fea
LB		 365}
366
e23da037
LB		 367/*
368 * column formatting convention & formats
369 * package: "pk" 2 columns %2d
370 * core: "cor" 3 columns %3d
371 * CPU: "CPU" 3 columns %3d
889facbe
LB		 372 * Pkg_W: %6.2
373 * Cor_W: %6.2
374 * GFX_W: %5.2
375 * RAM_W: %5.2
e23da037
LB		 376 * GHz: "GHz" 3 columns %3.2
377 * TSC: "TSC" 3 columns %3.2
1ed51011 378 * SMI: "SMI" 4 columns %4d
e23da037 379 * percentage " %pc3" %6.2
889facbe
LB		 380 * Perf Status percentage: %5.2
381 * "CTMP" 4 columns %4d
e23da037 382 */
c98d5d94
LB		 383int format_counters(struct thread_data *t, struct core_data *c,
384 struct pkg_data *p)
103a8fea
LB		 385{
386 double interval_float;
889facbe 387 char *fmt5, *fmt6;
103a8fea 388
c98d5d94
LB		 389 /* if showing only 1st thread in core and this isn't one, bail out */
390 if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
391 return 0;
392
393 /* if showing only 1st thread in pkg and this isn't one, bail out */
394 if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
395 return 0;
396
103a8fea
LB		 397 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
398
c98d5d94
LB		 399 /* topo columns, print blanks on 1st (average) line */
400 if (t == &average.threads) {
103a8fea 401 if (show_pkg)
c98d5d94 402 outp += sprintf(outp, " ");
e23da037 403 if (show_pkg && show_core)
c98d5d94 404 outp += sprintf(outp, " ");
103a8fea 405 if (show_core)
c98d5d94 406 outp += sprintf(outp, " ");
103a8fea 407 if (show_cpu)
c98d5d94 408 outp += sprintf(outp, " " " ");
103a8fea 409 } else {
c98d5d94
LB		 410 if (show_pkg) {
411 if (p)
412 outp += sprintf(outp, "%2d", p->package_id);
413 else
414 outp += sprintf(outp, " ");
415 }
e23da037 416 if (show_pkg && show_core)
c98d5d94
LB		 417 outp += sprintf(outp, " ");
418 if (show_core) {
419 if (c)
420 outp += sprintf(outp, "%3d", c->core_id);
421 else
422 outp += sprintf(outp, " ");
423 }
103a8fea 424 if (show_cpu)
c98d5d94 425 outp += sprintf(outp, " %3d", t->cpu_id);
103a8fea 426 }
103a8fea
LB		 427 /* %c0 */
428 if (do_nhm_cstates) {
e23da037 429 if (show_pkg || show_core || show_cpu)
c98d5d94 430 outp += sprintf(outp, " ");
103a8fea 431 if (!skip_c0)
c98d5d94 432 outp += sprintf(outp, "%6.2f", 100.0 * t->mperf/t->tsc);
103a8fea 433 else
c98d5d94 434 outp += sprintf(outp, " ****");
103a8fea
LB		 435 }
436
437 /* GHz */
438 if (has_aperf) {
439 if (!aperf_mperf_unstable) {
c98d5d94
LB		 440 outp += sprintf(outp, " %3.2f",
441 1.0 * t->tsc / units * t->aperf /
442 t->mperf / interval_float);
103a8fea 443 } else {
c98d5d94
LB		 444 if (t->aperf > t->tsc || t->mperf > t->tsc) {
445 outp += sprintf(outp, " ***");
103a8fea 446 } else {
c98d5d94
LB		 447 outp += sprintf(outp, "%3.1f*",
448 1.0 * t->tsc /
449 units * t->aperf /
450 t->mperf / interval_float);
103a8fea
LB		 451 }
452 }
453 }
454
455 /* TSC */
c98d5d94 456 outp += sprintf(outp, "%5.2f", 1.0 * t->tsc/units/interval_float);
103a8fea 457
1ed51011
LB		 458 /* SMI */
459 if (do_smi)
460 outp += sprintf(outp, "%4d", t->smi_count);
461
8e180f3c
LB		 462 /* delta */
463 if (extra_delta_offset32)
464 outp += sprintf(outp, " %11llu", t->extra_delta32);
465
466 /* DELTA */
467 if (extra_delta_offset64)
468 outp += sprintf(outp, " %11llu", t->extra_delta64);
2f32edf1
LB		 469 /* msr */
470 if (extra_msr_offset32)
8e180f3c 471 outp += sprintf(outp, " 0x%08llx", t->extra_msr32);
2f32edf1 472
130ff304 473 /* MSR */
2f32edf1
LB		 474 if (extra_msr_offset64)
475 outp += sprintf(outp, " 0x%016llx", t->extra_msr64);
130ff304 476
103a8fea
LB		 477 if (do_nhm_cstates) {
478 if (!skip_c1)
c98d5d94 479 outp += sprintf(outp, " %6.2f", 100.0 * t->c1/t->tsc);
103a8fea 480 else
c98d5d94 481 outp += sprintf(outp, " ****");
103a8fea 482 }
c98d5d94
LB		 483
484 /* print per-core data only for 1st thread in core */
485 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
486 goto done;
487
144b44b1 488 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 489 outp += sprintf(outp, " %6.2f", 100.0 * c->c3/t->tsc);
103a8fea 490 if (do_nhm_cstates)
c98d5d94 491 outp += sprintf(outp, " %6.2f", 100.0 * c->c6/t->tsc);
103a8fea 492 if (do_snb_cstates)
c98d5d94
LB		 493 outp += sprintf(outp, " %6.2f", 100.0 * c->c7/t->tsc);
494
889facbe
LB		 495 if (do_dts)
496 outp += sprintf(outp, " %4d", c->core_temp_c);
497
c98d5d94
LB		 498 /* print per-package data only for 1st core in package */
499 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
500 goto done;
501
889facbe
LB		 502 if (do_ptm)
503 outp += sprintf(outp, " %4d", p->pkg_temp_c);
504
103a8fea 505 if (do_snb_cstates)
c98d5d94 506 outp += sprintf(outp, " %6.2f", 100.0 * p->pc2/t->tsc);
144b44b1 507 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 508 outp += sprintf(outp, " %6.2f", 100.0 * p->pc3/t->tsc);
144b44b1 509 if (do_nhm_cstates && !do_slm_cstates)
c98d5d94 510 outp += sprintf(outp, " %6.2f", 100.0 * p->pc6/t->tsc);
103a8fea 511 if (do_snb_cstates)
c98d5d94 512 outp += sprintf(outp, " %6.2f", 100.0 * p->pc7/t->tsc);
ca58710f
KCA		 513 if (do_c8_c9_c10) {
514 outp += sprintf(outp, " %6.2f", 100.0 * p->pc8/t->tsc);
515 outp += sprintf(outp, " %6.2f", 100.0 * p->pc9/t->tsc);
516 outp += sprintf(outp, " %6.2f", 100.0 * p->pc10/t->tsc);
517 }
889facbe
LB		 518
519 /*
520 * If measurement interval exceeds minimum RAPL Joule Counter range,
521 * indicate that results are suspect by printing "**" in fraction place.
522 */
523 if (interval_float < rapl_joule_counter_range) {
524 fmt5 = " %5.2f";
525 fmt6 = " %6.2f";
526 } else {
527 fmt5 = " %3.0f**";
528 fmt6 = " %4.0f**";
529 }
530
531 if (do_rapl & RAPL_PKG)
532 outp += sprintf(outp, fmt6, p->energy_pkg * rapl_energy_units / interval_float);
533 if (do_rapl & RAPL_CORES)
534 outp += sprintf(outp, fmt6, p->energy_cores * rapl_energy_units / interval_float);
535 if (do_rapl & RAPL_GFX)
536 outp += sprintf(outp, fmt5, p->energy_gfx * rapl_energy_units / interval_float);
537 if (do_rapl & RAPL_DRAM)
538 outp += sprintf(outp, fmt5, p->energy_dram * rapl_energy_units / interval_float);
539 if (do_rapl & RAPL_PKG_PERF_STATUS )
540 outp += sprintf(outp, fmt5, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
541 if (do_rapl & RAPL_DRAM_PERF_STATUS )
542 outp += sprintf(outp, fmt5, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
543
c98d5d94 544done:
c98d5d94
LB		 545 outp += sprintf(outp, "\n");
546
547 return 0;
103a8fea
LB		 548}
549
c98d5d94
LB		 550void flush_stdout()
551{
552 fputs(output_buffer, stdout);
ddac0d68 553 fflush(stdout);
c98d5d94
LB		 554 outp = output_buffer;
555}
556void flush_stderr()
557{
558 fputs(output_buffer, stderr);
559 outp = output_buffer;
560}
561void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
103a8fea 562{
e23da037 563 static int printed;
103a8fea 564
e23da037
LB		 565 if (!printed || !summary_only)
566 print_header();
103a8fea 567
c98d5d94
LB		 568 if (topo.num_cpus > 1)
569 format_counters(&average.threads, &average.cores,
570 &average.packages);
103a8fea 571
e23da037
LB		 572 printed = 1;
573
574 if (summary_only)
575 return;
576
c98d5d94 577 for_all_cpus(format_counters, t, c, p);
103a8fea
LB		 578}
579
889facbe
LB		 580#define DELTA_WRAP32(new, old) \
581 if (new > old) { \
582 old = new - old; \
583 } else { \
584 old = 0x100000000 + new - old; \
585 }
586
c98d5d94
LB		 587void
588delta_package(struct pkg_data *new, struct pkg_data *old)
589{
590 old->pc2 = new->pc2 - old->pc2;
591 old->pc3 = new->pc3 - old->pc3;
592 old->pc6 = new->pc6 - old->pc6;
593 old->pc7 = new->pc7 - old->pc7;
ca58710f
KCA		 594 old->pc8 = new->pc8 - old->pc8;
595 old->pc9 = new->pc9 - old->pc9;
596 old->pc10 = new->pc10 - old->pc10;
889facbe
LB		 597 old->pkg_temp_c = new->pkg_temp_c;
598
599 DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
600 DELTA_WRAP32(new->energy_cores, old->energy_cores);
601 DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
602 DELTA_WRAP32(new->energy_dram, old->energy_dram);
603 DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
604 DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
c98d5d94 605}
103a8fea 606
c98d5d94
LB		 607void
608delta_core(struct core_data *new, struct core_data *old)
103a8fea 609{
c98d5d94
LB		 610 old->c3 = new->c3 - old->c3;
611 old->c6 = new->c6 - old->c6;
612 old->c7 = new->c7 - old->c7;
889facbe 613 old->core_temp_c = new->core_temp_c;
c98d5d94 614}
103a8fea 615
c3ae331d
LB		 616/*
617 * old = new - old
618 */
c98d5d94
LB		 619void
620delta_thread(struct thread_data *new, struct thread_data *old,
621 struct core_data *core_delta)
622{
623 old->tsc = new->tsc - old->tsc;
624
625 /* check for TSC < 1 Mcycles over interval */
626 if (old->tsc < (1000 * 1000)) {
627 fprintf(stderr, "Insanely slow TSC rate, TSC stops in idle?\n");
628 fprintf(stderr, "You can disable all c-states by booting with \"idle=poll\"\n");
629 fprintf(stderr, "or just the deep ones with \"processor.max_cstate=1\"\n");
630 exit(-3);
631 }
103a8fea 632
c98d5d94 633 old->c1 = new->c1 - old->c1;
103a8fea 634
c98d5d94
LB		 635 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
636 old->aperf = new->aperf - old->aperf;
637 old->mperf = new->mperf - old->mperf;
638 } else {
103a8fea 639
c98d5d94
LB		 640 if (!aperf_mperf_unstable) {
641 fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
642 fprintf(stderr, "* Frequency results do not cover entire interval *\n");
643 fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
103a8fea 644
c98d5d94 645 aperf_mperf_unstable = 1;
103a8fea 646 }
103a8fea 647 /*
c98d5d94
LB		 648 * mperf delta is likely a huge "positive" number
649 * can not use it for calculating c0 time
103a8fea 650 */
c98d5d94
LB		 651 skip_c0 = 1;
652 skip_c1 = 1;
653 }
103a8fea 654
103a8fea 655
144b44b1
LB		 656 if (use_c1_residency_msr) {
657 /*
658 * Some models have a dedicated C1 residency MSR,
659 * which should be more accurate than the derivation below.
660 */
661 } else {
662 /*
663 * As counter collection is not atomic,
664 * it is possible for mperf's non-halted cycles + idle states
665 * to exceed TSC's all cycles: show c1 = 0% in that case.
666 */
667 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
668 old->c1 = 0;
669 else {
670 /* normal case, derive c1 */
671 old->c1 = old->tsc - old->mperf - core_delta->c3
c98d5d94 672 - core_delta->c6 - core_delta->c7;
144b44b1 673 }
c98d5d94 674 }
c3ae331d 675
c98d5d94 676 if (old->mperf == 0) {
c3ae331d 677 if (verbose > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id);
c98d5d94 678 old->mperf = 1; /* divide by 0 protection */
103a8fea 679 }
c98d5d94 680
8e180f3c
LB		 681 old->extra_delta32 = new->extra_delta32 - old->extra_delta32;
682 old->extra_delta32 &= 0xFFFFFFFF;
683
684 old->extra_delta64 = new->extra_delta64 - old->extra_delta64;
685
c98d5d94 686 /*
8e180f3c 687 * Extra MSR is just a snapshot, simply copy latest w/o subtracting
c98d5d94 688 */
2f32edf1
LB		 689 old->extra_msr32 = new->extra_msr32;
690 old->extra_msr64 = new->extra_msr64;
1ed51011
LB		 691
692 if (do_smi)
693 old->smi_count = new->smi_count - old->smi_count;
c98d5d94
LB		 694}
695
696int delta_cpu(struct thread_data *t, struct core_data *c,
697 struct pkg_data *p, struct thread_data *t2,
698 struct core_data *c2, struct pkg_data *p2)
699{
700 /* calculate core delta only for 1st thread in core */
701 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
702 delta_core(c, c2);
703
704 /* always calculate thread delta */
705 delta_thread(t, t2, c2); /* c2 is core delta */
706
707 /* calculate package delta only for 1st core in package */
708 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
709 delta_package(p, p2);
710
103a8fea
LB		 711 return 0;
712}
713
c98d5d94
LB		 714void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
715{
716 t->tsc = 0;
717 t->aperf = 0;
718 t->mperf = 0;
719 t->c1 = 0;
720
1ed51011 721 t->smi_count = 0;
8e180f3c
LB		 722 t->extra_delta32 = 0;
723 t->extra_delta64 = 0;
724
c98d5d94
LB		 725 /* tells format_counters to dump all fields from this set */
726 t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
727
728 c->c3 = 0;
729 c->c6 = 0;
730 c->c7 = 0;
889facbe 731 c->core_temp_c = 0;
c98d5d94
LB		 732
733 p->pc2 = 0;
734 p->pc3 = 0;
735 p->pc6 = 0;
736 p->pc7 = 0;
ca58710f
KCA		 737 p->pc8 = 0;
738 p->pc9 = 0;
739 p->pc10 = 0;
889facbe
LB		 740
741 p->energy_pkg = 0;
742 p->energy_dram = 0;
743 p->energy_cores = 0;
744 p->energy_gfx = 0;
745 p->rapl_pkg_perf_status = 0;
746 p->rapl_dram_perf_status = 0;
747 p->pkg_temp_c = 0;
c98d5d94
LB		 748}
749int sum_counters(struct thread_data *t, struct core_data *c,
750 struct pkg_data *p)
103a8fea 751{
c98d5d94
LB		 752 average.threads.tsc += t->tsc;
753 average.threads.aperf += t->aperf;
754 average.threads.mperf += t->mperf;
755 average.threads.c1 += t->c1;
103a8fea 756
8e180f3c
LB		 757 average.threads.extra_delta32 += t->extra_delta32;
758 average.threads.extra_delta64 += t->extra_delta64;
759
c98d5d94
LB		 760 /* sum per-core values only for 1st thread in core */
761 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
762 return 0;
103a8fea 763
c98d5d94
LB		 764 average.cores.c3 += c->c3;
765 average.cores.c6 += c->c6;
766 average.cores.c7 += c->c7;
767
889facbe
LB		 768 average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
769
c98d5d94
LB		 770 /* sum per-pkg values only for 1st core in pkg */
771 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
772 return 0;
773
774 average.packages.pc2 += p->pc2;
775 average.packages.pc3 += p->pc3;
776 average.packages.pc6 += p->pc6;
777 average.packages.pc7 += p->pc7;
ca58710f
KCA		 778 average.packages.pc8 += p->pc8;
779 average.packages.pc9 += p->pc9;
780 average.packages.pc10 += p->pc10;
c98d5d94 781
889facbe
LB		 782 average.packages.energy_pkg += p->energy_pkg;
783 average.packages.energy_dram += p->energy_dram;
784 average.packages.energy_cores += p->energy_cores;
785 average.packages.energy_gfx += p->energy_gfx;
786
787 average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
788
789 average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
790 average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
c98d5d94
LB		 791 return 0;
792}
793/*
794 * sum the counters for all cpus in the system
795 * compute the weighted average
796 */
797void compute_average(struct thread_data *t, struct core_data *c,
798 struct pkg_data *p)
799{
800 clear_counters(&average.threads, &average.cores, &average.packages);
801
802 for_all_cpus(sum_counters, t, c, p);
803
804 average.threads.tsc /= topo.num_cpus;
805 average.threads.aperf /= topo.num_cpus;
806 average.threads.mperf /= topo.num_cpus;
807 average.threads.c1 /= topo.num_cpus;
808
8e180f3c
LB		 809 average.threads.extra_delta32 /= topo.num_cpus;
810 average.threads.extra_delta32 &= 0xFFFFFFFF;
811
812 average.threads.extra_delta64 /= topo.num_cpus;
813
c98d5d94
LB		 814 average.cores.c3 /= topo.num_cores;
815 average.cores.c6 /= topo.num_cores;
816 average.cores.c7 /= topo.num_cores;
817
818 average.packages.pc2 /= topo.num_packages;
819 average.packages.pc3 /= topo.num_packages;
820 average.packages.pc6 /= topo.num_packages;
821 average.packages.pc7 /= topo.num_packages;
ca58710f
KCA		 822
823 average.packages.pc8 /= topo.num_packages;
824 average.packages.pc9 /= topo.num_packages;
825 average.packages.pc10 /= topo.num_packages;
103a8fea
LB		 826}
827
c98d5d94 828static unsigned long long rdtsc(void)
103a8fea 829{
c98d5d94 830 unsigned int low, high;
15aaa346 831
c98d5d94 832 asm volatile("rdtsc" : "=a" (low), "=d" (high));
15aaa346 833
c98d5d94
LB		 834 return low | ((unsigned long long)high) << 32;
835}
15aaa346 836
15aaa346 837
c98d5d94
LB		 838/*
839 * get_counters(...)
840 * migrate to cpu
841 * acquire and record local counters for that cpu
842 */
843int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
844{
845 int cpu = t->cpu_id;
889facbe 846 unsigned long long msr;
88c3281f 847
e52966c0
LB		 848 if (cpu_migrate(cpu)) {
849 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
c98d5d94 850 return -1;
e52966c0 851 }
15aaa346 852
c98d5d94
LB		 853 t->tsc = rdtsc(); /* we are running on local CPU of interest */
854
855 if (has_aperf) {
9c63a650 856 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
c98d5d94 857 return -3;
9c63a650 858 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
c98d5d94
LB		 859 return -4;
860 }
861
1ed51011
LB		 862 if (do_smi) {
863 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
864 return -5;
865 t->smi_count = msr & 0xFFFFFFFF;
866 }
8e180f3c 867 if (extra_delta_offset32) {
889facbe 868 if (get_msr(cpu, extra_delta_offset32, &msr))
8e180f3c 869 return -5;
889facbe 870 t->extra_delta32 = msr & 0xFFFFFFFF;
8e180f3c
LB		 871 }
872
873 if (extra_delta_offset64)
874 if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64))
2f32edf1
LB		 875 return -5;
876
8e180f3c 877 if (extra_msr_offset32) {
889facbe 878 if (get_msr(cpu, extra_msr_offset32, &msr))
8e180f3c 879 return -5;
889facbe 880 t->extra_msr32 = msr & 0xFFFFFFFF;
8e180f3c
LB		 881 }
882
2f32edf1
LB		 883 if (extra_msr_offset64)
884 if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64))
c98d5d94
LB		 885 return -5;
886
144b44b1
LB		 887 if (use_c1_residency_msr) {
888 if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
889 return -6;
890 }
891
c98d5d94
LB		 892 /* collect core counters only for 1st thread in core */
893 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
894 return 0;
895
144b44b1 896 if (do_nhm_cstates && !do_slm_cstates) {
c98d5d94
LB		 897 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
898 return -6;
144b44b1
LB		 899 }
900
901 if (do_nhm_cstates) {
c98d5d94
LB		 902 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
903 return -7;
904 }
905
906 if (do_snb_cstates)
907 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
908 return -8;
909
889facbe
LB		 910 if (do_dts) {
911 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
912 return -9;
913 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
914 }
915
916
c98d5d94
LB		 917 /* collect package counters only for 1st core in package */
918 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
919 return 0;
920
144b44b1 921 if (do_nhm_cstates && !do_slm_cstates) {
c98d5d94
LB		 922 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
923 return -9;
924 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
925 return -10;
926 }
927 if (do_snb_cstates) {
928 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
929 return -11;
930 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
931 return -12;
103a8fea 932 }
ca58710f
KCA		 933 if (do_c8_c9_c10) {
934 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
935 return -13;
936 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
937 return -13;
938 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
939 return -13;
940 }
889facbe
LB		 941 if (do_rapl & RAPL_PKG) {
942 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
943 return -13;
944 p->energy_pkg = msr & 0xFFFFFFFF;
945 }
946 if (do_rapl & RAPL_CORES) {
947 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
948 return -14;
949 p->energy_cores = msr & 0xFFFFFFFF;
950 }
951 if (do_rapl & RAPL_DRAM) {
952 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
953 return -15;
954 p->energy_dram = msr & 0xFFFFFFFF;
955 }
956 if (do_rapl & RAPL_GFX) {
957 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
958 return -16;
959 p->energy_gfx = msr & 0xFFFFFFFF;
960 }
961 if (do_rapl & RAPL_PKG_PERF_STATUS) {
962 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
963 return -16;
964 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
965 }
966 if (do_rapl & RAPL_DRAM_PERF_STATUS) {
967 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
968 return -16;
969 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
970 }
971 if (do_ptm) {
972 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
973 return -17;
974 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
975 }
15aaa346 976 return 0;
103a8fea
LB		 977}
978
c98d5d94 979void print_verbose_header(void)
103a8fea
LB		 980{
981 unsigned long long msr;
982 unsigned int ratio;
983
984 if (!do_nehalem_platform_info)
985 return;
986
9c63a650 987 get_msr(0, MSR_NHM_PLATFORM_INFO, &msr);
103a8fea 988
67920418 989 fprintf(stderr, "cpu0: MSR_NHM_PLATFORM_INFO: 0x%08llx\n", msr);
6574a5d5 990
103a8fea
LB		 991 ratio = (msr >> 40) & 0xFF;
992 fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
993 ratio, bclk, ratio * bclk);
994
995 ratio = (msr >> 8) & 0xFF;
996 fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
997 ratio, bclk, ratio * bclk);
998
67920418 999 get_msr(0, MSR_IA32_POWER_CTL, &msr);
144b44b1 1000 fprintf(stderr, "cpu0: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
67920418
LB		 1001 msr, msr & 0x2 ? "EN" : "DIS");
1002
6574a5d5
LB		 1003 if (!do_ivt_turbo_ratio_limit)
1004 goto print_nhm_turbo_ratio_limits;
1005
1006 get_msr(0, MSR_IVT_TURBO_RATIO_LIMIT, &msr);
1007
67920418 1008 fprintf(stderr, "cpu0: MSR_IVT_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
6574a5d5
LB		 1009
1010 ratio = (msr >> 56) & 0xFF;
1011 if (ratio)
1012 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n",
1013 ratio, bclk, ratio * bclk);
1014
1015 ratio = (msr >> 48) & 0xFF;
1016 if (ratio)
1017 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n",
1018 ratio, bclk, ratio * bclk);
1019
1020 ratio = (msr >> 40) & 0xFF;
1021 if (ratio)
1022 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n",
1023 ratio, bclk, ratio * bclk);
1024
1025 ratio = (msr >> 32) & 0xFF;
1026 if (ratio)
1027 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n",
1028 ratio, bclk, ratio * bclk);
1029
1030 ratio = (msr >> 24) & 0xFF;
1031 if (ratio)
1032 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n",
1033 ratio, bclk, ratio * bclk);
1034
1035 ratio = (msr >> 16) & 0xFF;
1036 if (ratio)
1037 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n",
1038 ratio, bclk, ratio * bclk);
1039
1040 ratio = (msr >> 8) & 0xFF;
1041 if (ratio)
1042 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n",
1043 ratio, bclk, ratio * bclk);
1044
1045 ratio = (msr >> 0) & 0xFF;
1046 if (ratio)
1047 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n",
1048 ratio, bclk, ratio * bclk);
1049
1050print_nhm_turbo_ratio_limits:
889facbe
LB		 1051 get_msr(0, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
1052
1053#define SNB_C1_AUTO_UNDEMOTE (1UL << 27)
1054#define SNB_C3_AUTO_UNDEMOTE (1UL << 28)
1055
1056 fprintf(stderr, "cpu0: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", msr);
1057
1058 fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: ",
1059 (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
1060 (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
1061 (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
1062 (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
1063 (msr & (1 << 15)) ? "" : "UN",
1064 (unsigned int)msr & 7);
1065
1066
1067 switch(msr & 0x7) {
1068 case 0:
144b44b1 1069 fprintf(stderr, do_slm_cstates ? "no pkg states" : "pc0");
889facbe
LB		 1070 break;
1071 case 1:
144b44b1 1072 fprintf(stderr, do_slm_cstates ? "no pkg states" : do_snb_cstates ? "pc2" : "pc0");
889facbe
LB		 1073 break;
1074 case 2:
144b44b1 1075 fprintf(stderr, do_slm_cstates ? "invalid" : do_snb_cstates ? "pc6-noret" : "pc3");
889facbe
LB		 1076 break;
1077 case 3:
144b44b1 1078 fprintf(stderr, do_slm_cstates ? "invalid" : "pc6");
889facbe
LB		 1079 break;
1080 case 4:
144b44b1 1081 fprintf(stderr, do_slm_cstates ? "pc4" : "pc7");
889facbe
LB		 1082 break;
1083 case 5:
144b44b1
LB		 1084 fprintf(stderr, do_slm_cstates ? "invalid" : do_snb_cstates ? "pc7s" : "invalid");
1085 break;
1086 case 6:
1087 fprintf(stderr, do_slm_cstates ? "pc6" : "invalid");
889facbe
LB		 1088 break;
1089 case 7:
144b44b1 1090 fprintf(stderr, do_slm_cstates ? "pc7" : "unlimited");
889facbe
LB		 1091 break;
1092 default:
1093 fprintf(stderr, "invalid");
1094 }
1095 fprintf(stderr, ")\n");
103a8fea
LB		 1096
1097 if (!do_nehalem_turbo_ratio_limit)
1098 return;
1099
9c63a650 1100 get_msr(0, MSR_NHM_TURBO_RATIO_LIMIT, &msr);
103a8fea 1101
67920418 1102 fprintf(stderr, "cpu0: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
6574a5d5
LB		 1103
1104 ratio = (msr >> 56) & 0xFF;
1105 if (ratio)
1106 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n",
1107 ratio, bclk, ratio * bclk);
1108
1109 ratio = (msr >> 48) & 0xFF;
1110 if (ratio)
1111 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n",
1112 ratio, bclk, ratio * bclk);
1113
1114 ratio = (msr >> 40) & 0xFF;
1115 if (ratio)
1116 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n",
1117 ratio, bclk, ratio * bclk);
1118
1119 ratio = (msr >> 32) & 0xFF;
1120 if (ratio)
1121 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n",
1122 ratio, bclk, ratio * bclk);
1123
103a8fea
LB		 1124 ratio = (msr >> 24) & 0xFF;
1125 if (ratio)
1126 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
1127 ratio, bclk, ratio * bclk);
1128
1129 ratio = (msr >> 16) & 0xFF;
1130 if (ratio)
1131 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
1132 ratio, bclk, ratio * bclk);
1133
1134 ratio = (msr >> 8) & 0xFF;
1135 if (ratio)
1136 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
1137 ratio, bclk, ratio * bclk);
1138
1139 ratio = (msr >> 0) & 0xFF;
1140 if (ratio)
1141 fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
1142 ratio, bclk, ratio * bclk);
103a8fea
LB		 1143}
1144
c98d5d94 1145void free_all_buffers(void)
103a8fea 1146{
c98d5d94
LB		 1147 CPU_FREE(cpu_present_set);
1148 cpu_present_set = NULL;
1149 cpu_present_set = 0;
103a8fea 1150
c98d5d94
LB		 1151 CPU_FREE(cpu_affinity_set);
1152 cpu_affinity_set = NULL;
1153 cpu_affinity_setsize = 0;
103a8fea 1154
c98d5d94
LB		 1155 free(thread_even);
1156 free(core_even);
1157 free(package_even);
103a8fea 1158
c98d5d94
LB		 1159 thread_even = NULL;
1160 core_even = NULL;
1161 package_even = NULL;
103a8fea 1162
c98d5d94
LB		 1163 free(thread_odd);
1164 free(core_odd);
1165 free(package_odd);
103a8fea 1166
c98d5d94
LB		 1167 thread_odd = NULL;
1168 core_odd = NULL;
1169 package_odd = NULL;
103a8fea 1170
c98d5d94
LB		 1171 free(output_buffer);
1172 output_buffer = NULL;
1173 outp = NULL;
103a8fea
LB		 1174}
1175
c98d5d94
LB		 1176/*
1177 * cpu_is_first_sibling_in_core(cpu)
1178 * return 1 if given CPU is 1st HT sibling in the core
1179 */
1180int cpu_is_first_sibling_in_core(int cpu)
103a8fea 1181{
c98d5d94
LB		 1182 char path[64];
1183 FILE *filep;
1184 int first_cpu;
103a8fea 1185
c98d5d94
LB		 1186 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1187 filep = fopen(path, "r");
1188 if (filep == NULL) {
1189 perror(path);
1190 exit(1);
103a8fea 1191 }
74823419
JT		 1192 if (fscanf(filep, "%d", &first_cpu) != 1) {
1193 perror(path);
1194 exit(1);
1195 }
c98d5d94
LB		 1196 fclose(filep);
1197 return (cpu == first_cpu);
103a8fea
LB		 1198}
1199
c98d5d94
LB		 1200/*
1201 * cpu_is_first_core_in_package(cpu)
1202 * return 1 if given CPU is 1st core in package
1203 */
1204int cpu_is_first_core_in_package(int cpu)
103a8fea 1205{
c98d5d94
LB		 1206 char path[64];
1207 FILE *filep;
1208 int first_cpu;
103a8fea 1209
c98d5d94
LB		 1210 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
1211 filep = fopen(path, "r");
1212 if (filep == NULL) {
1213 perror(path);
103a8fea
LB		 1214 exit(1);
1215 }
74823419
JT		 1216 if (fscanf(filep, "%d", &first_cpu) != 1) {
1217 perror(path);
1218 exit(1);
1219 }
c98d5d94
LB		 1220 fclose(filep);
1221 return (cpu == first_cpu);
103a8fea
LB		 1222}
1223
1224int get_physical_package_id(int cpu)
1225{
c98d5d94 1226 char path[80];
103a8fea
LB		 1227 FILE *filep;
1228 int pkg;
1229
1230 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
1231 filep = fopen(path, "r");
1232 if (filep == NULL) {
1233 perror(path);
1234 exit(1);
1235 }
74823419
JT		 1236 if (fscanf(filep, "%d", &pkg) != 1) {
1237 perror(path);
1238 exit(1);
1239 }
103a8fea
LB		 1240 fclose(filep);
1241 return pkg;
1242}
1243
1244int get_core_id(int cpu)
1245{
c98d5d94 1246 char path[80];
103a8fea
LB		 1247 FILE *filep;
1248 int core;
1249
1250 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
1251 filep = fopen(path, "r");
1252 if (filep == NULL) {
1253 perror(path);
1254 exit(1);
1255 }
74823419
JT		 1256 if (fscanf(filep, "%d", &core) != 1) {
1257 perror(path);
1258 exit(1);
1259 }
103a8fea
LB		 1260 fclose(filep);
1261 return core;
1262}
1263
c98d5d94
LB		 1264int get_num_ht_siblings(int cpu)
1265{
1266 char path[80];
1267 FILE *filep;
1268 int sib1, sib2;
1269 int matches;
1270 char character;
1271
1272 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1273 filep = fopen(path, "r");
1274 if (filep == NULL) {
1275 perror(path);
1276 exit(1);
1277 }
1278 /*
1279 * file format:
1280 * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
1281 * otherwinse 1 sibling (self).
1282 */
1283 matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
1284
1285 fclose(filep);
1286
1287 if (matches == 3)
1288 return 2;
1289 else
1290 return 1;
1291}
1292
103a8fea 1293/*
c98d5d94
LB		 1294 * run func(thread, core, package) in topology order
1295 * skip non-present cpus
103a8fea
LB		 1296 */
1297
c98d5d94
LB		 1298int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
1299 struct pkg_data *, struct thread_data *, struct core_data *,
1300 struct pkg_data *), struct thread_data *thread_base,
1301 struct core_data *core_base, struct pkg_data *pkg_base,
1302 struct thread_data *thread_base2, struct core_data *core_base2,
1303 struct pkg_data *pkg_base2)
1304{
1305 int retval, pkg_no, core_no, thread_no;
1306
1307 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
1308 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
1309 for (thread_no = 0; thread_no <
1310 topo.num_threads_per_core; ++thread_no) {
1311 struct thread_data *t, *t2;
1312 struct core_data *c, *c2;
1313 struct pkg_data *p, *p2;
1314
1315 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
1316
1317 if (cpu_is_not_present(t->cpu_id))
1318 continue;
1319
1320 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
1321
1322 c = GET_CORE(core_base, core_no, pkg_no);
1323 c2 = GET_CORE(core_base2, core_no, pkg_no);
1324
1325 p = GET_PKG(pkg_base, pkg_no);
1326 p2 = GET_PKG(pkg_base2, pkg_no);
1327
1328 retval = func(t, c, p, t2, c2, p2);
1329 if (retval)
1330 return retval;
1331 }
1332 }
1333 }
1334 return 0;
1335}
1336
1337/*
1338 * run func(cpu) on every cpu in /proc/stat
1339 * return max_cpu number
1340 */
1341int for_all_proc_cpus(int (func)(int))
103a8fea
LB		 1342{
1343 FILE *fp;
c98d5d94 1344 int cpu_num;
103a8fea
LB		 1345 int retval;
1346
1347 fp = fopen(proc_stat, "r");
1348 if (fp == NULL) {
1349 perror(proc_stat);
1350 exit(1);
1351 }
1352
1353 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
1354 if (retval != 0) {
1355 perror("/proc/stat format");
1356 exit(1);
1357 }
1358
c98d5d94
LB		 1359 while (1) {
1360 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
103a8fea
LB		 1361 if (retval != 1)
1362 break;
1363
c98d5d94
LB		 1364 retval = func(cpu_num);
1365 if (retval) {
1366 fclose(fp);
1367 return(retval);
1368 }
103a8fea
LB		 1369 }
1370 fclose(fp);
c98d5d94 1371 return 0;
103a8fea
LB		 1372}
1373
1374void re_initialize(void)
1375{
c98d5d94
LB		 1376 free_all_buffers();
1377 setup_all_buffers();
1378 printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
103a8fea
LB		 1379}
1380
c98d5d94 1381
103a8fea 1382/*
c98d5d94
LB		 1383 * count_cpus()
1384 * remember the last one seen, it will be the max
103a8fea 1385 */
c98d5d94 1386int count_cpus(int cpu)
103a8fea 1387{
c98d5d94
LB		 1388 if (topo.max_cpu_num < cpu)
1389 topo.max_cpu_num = cpu;
103a8fea 1390
c98d5d94
LB		 1391 topo.num_cpus += 1;
1392 return 0;
1393}
1394int mark_cpu_present(int cpu)
1395{
1396 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
15aaa346 1397 return 0;
103a8fea
LB		 1398}
1399
1400void turbostat_loop()
1401{
c98d5d94 1402 int retval;
e52966c0 1403 int restarted = 0;
c98d5d94 1404
103a8fea 1405restart:
e52966c0
LB		 1406 restarted++;
1407
c98d5d94 1408 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
d91bb17c
LB		 1409 if (retval < -1) {
1410 exit(retval);
1411 } else if (retval == -1) {
e52966c0
LB		 1412 if (restarted > 1) {
1413 exit(retval);
1414 }
c98d5d94
LB		 1415 re_initialize();
1416 goto restart;
1417 }
e52966c0 1418 restarted = 0;
103a8fea
LB		 1419 gettimeofday(&tv_even, (struct timezone *)NULL);
1420
1421 while (1) {
c98d5d94 1422 if (for_all_proc_cpus(cpu_is_not_present)) {
103a8fea
LB		 1423 re_initialize();
1424 goto restart;
1425 }
1426 sleep(interval_sec);
c98d5d94 1427 retval = for_all_cpus(get_counters, ODD_COUNTERS);
d91bb17c
LB		 1428 if (retval < -1) {
1429 exit(retval);
1430 } else if (retval == -1) {
15aaa346
LB		 1431 re_initialize();
1432 goto restart;
1433 }
103a8fea 1434 gettimeofday(&tv_odd, (struct timezone *)NULL);
103a8fea 1435 timersub(&tv_odd, &tv_even, &tv_delta);
c98d5d94
LB		 1436 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
1437 compute_average(EVEN_COUNTERS);
1438 format_all_counters(EVEN_COUNTERS);
1439 flush_stdout();
15aaa346 1440 sleep(interval_sec);
c98d5d94 1441 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
d91bb17c
LB		 1442 if (retval < -1) {
1443 exit(retval);
1444 } else if (retval == -1) {
103a8fea
LB		 1445 re_initialize();
1446 goto restart;
1447 }
103a8fea 1448 gettimeofday(&tv_even, (struct timezone *)NULL);
103a8fea 1449 timersub(&tv_even, &tv_odd, &tv_delta);
c98d5d94
LB		 1450 for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS);
1451 compute_average(ODD_COUNTERS);
1452 format_all_counters(ODD_COUNTERS);
1453 flush_stdout();
103a8fea
LB		 1454 }
1455}
1456
1457void check_dev_msr()
1458{
1459 struct stat sb;
1460
1461 if (stat("/dev/cpu/0/msr", &sb)) {
1462 fprintf(stderr, "no /dev/cpu/0/msr\n");
1463 fprintf(stderr, "Try \"# modprobe msr\"\n");
1464 exit(-5);
1465 }
1466}
1467
1468void check_super_user()
1469{
1470 if (getuid() != 0) {
1471 fprintf(stderr, "must be root\n");
1472 exit(-6);
1473 }
1474}
1475
1476int has_nehalem_turbo_ratio_limit(unsigned int family, unsigned int model)
1477{
1478 if (!genuine_intel)
1479 return 0;
1480
1481 if (family != 6)
1482 return 0;
1483
1484 switch (model) {
1485 case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
1486 case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
1487 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
1488 case 0x25: /* Westmere Client - Clarkdale, Arrandale */
1489 case 0x2C: /* Westmere EP - Gulftown */
1490 case 0x2A: /* SNB */
1491 case 0x2D: /* SNB Xeon */
553575f1 1492 case 0x3A: /* IVB */
1300651b 1493 case 0x3E: /* IVB Xeon */
70b43400
LB		 1494 case 0x3C: /* HSW */
1495 case 0x3F: /* HSW */
1496 case 0x45: /* HSW */
149c2319 1497 case 0x46: /* HSW */
144b44b1
LB		 1498 case 0x37: /* BYT */
1499 case 0x4D: /* AVN */
103a8fea
LB		 1500 return 1;
1501 case 0x2E: /* Nehalem-EX Xeon - Beckton */
1502 case 0x2F: /* Westmere-EX Xeon - Eagleton */
1503 default:
1504 return 0;
1505 }
1506}
6574a5d5
LB		 1507int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
1508{
1509 if (!genuine_intel)
1510 return 0;
1511
1512 if (family != 6)
1513 return 0;
1514
1515 switch (model) {
1516 case 0x3E: /* IVB Xeon */
1517 return 1;
1518 default:
1519 return 0;
1520 }
1521}
1522
889facbe
LB		 1523/*
1524 * print_epb()
1525 * Decode the ENERGY_PERF_BIAS MSR
1526 */
1527int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1528{
1529 unsigned long long msr;
1530 char *epb_string;
1531 int cpu;
1532
1533 if (!has_epb)
1534 return 0;
1535
1536 cpu = t->cpu_id;
1537
1538 /* EPB is per-package */
1539 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1540 return 0;
1541
1542 if (cpu_migrate(cpu)) {
1543 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1544 return -1;
1545 }
1546
1547 if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
1548 return 0;
1549
1550 switch (msr & 0x7) {
1551 case ENERGY_PERF_BIAS_PERFORMANCE:
1552 epb_string = "performance";
1553 break;
1554 case ENERGY_PERF_BIAS_NORMAL:
1555 epb_string = "balanced";
1556 break;
1557 case ENERGY_PERF_BIAS_POWERSAVE:
1558 epb_string = "powersave";
1559 break;
1560 default:
1561 epb_string = "custom";
1562 break;
1563 }
1564 fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
1565
1566 return 0;
1567}
1568
1569#define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */
1570#define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */
1571
144b44b1
LB		 1572double get_tdp(model)
1573{
1574 unsigned long long msr;
1575
1576 if (do_rapl & RAPL_PKG_POWER_INFO)
1577 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
1578 return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
1579
1580 switch (model) {
1581 case 0x37:
1582 case 0x4D:
1583 return 30.0;
1584 default:
1585 return 135.0;
1586 }
1587}
1588
1589
889facbe
LB		 1590/*
1591 * rapl_probe()
1592 *
144b44b1 1593 * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
889facbe
LB		 1594 */
1595void rapl_probe(unsigned int family, unsigned int model)
1596{
1597 unsigned long long msr;
144b44b1 1598 unsigned int time_unit;
889facbe
LB		 1599 double tdp;
1600
1601 if (!genuine_intel)
1602 return;
1603
1604 if (family != 6)
1605 return;
1606
1607 switch (model) {
1608 case 0x2A:
1609 case 0x3A:
70b43400
LB		 1610 case 0x3C: /* HSW */
1611 case 0x3F: /* HSW */
1612 case 0x45: /* HSW */
149c2319 1613 case 0x46: /* HSW */
144b44b1 1614 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
889facbe
LB		 1615 break;
1616 case 0x2D:
1617 case 0x3E:
144b44b1
LB		 1618 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1619 break;
1620 case 0x37: /* BYT */
1621 case 0x4D: /* AVN */
1622 do_rapl = RAPL_PKG | RAPL_CORES ;
889facbe
LB		 1623 break;
1624 default:
1625 return;
1626 }
1627
1628 /* units on package 0, verify later other packages match */
1629 if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1630 return;
1631
1632 rapl_power_units = 1.0 / (1 << (msr & 0xF));
144b44b1
LB		 1633 if (model == 0x37)
1634 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1635 else
1636 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
889facbe 1637
144b44b1
LB		 1638 time_unit = msr >> 16 & 0xF;
1639 if (time_unit == 0)
1640 time_unit = 0xA;
889facbe 1641
144b44b1 1642 rapl_time_units = 1.0 / (1 << (time_unit));
889facbe 1643
144b44b1 1644 tdp = get_tdp(model);
889facbe 1645
144b44b1 1646 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
889facbe 1647 if (verbose)
144b44b1 1648 fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
889facbe
LB		 1649
1650 return;
1651}
1652
1653int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1654{
1655 unsigned long long msr;
1656 unsigned int dts;
1657 int cpu;
1658
1659 if (!(do_dts || do_ptm))
1660 return 0;
1661
1662 cpu = t->cpu_id;
1663
1664 /* DTS is per-core, no need to print for each thread */
1665 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1666 return 0;
1667
1668 if (cpu_migrate(cpu)) {
1669 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1670 return -1;
1671 }
1672
1673 if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
1674 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1675 return 0;
1676
1677 dts = (msr >> 16) & 0x7F;
1678 fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
1679 cpu, msr, tcc_activation_temp - dts);
1680
1681#ifdef THERM_DEBUG
1682 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
1683 return 0;
1684
1685 dts = (msr >> 16) & 0x7F;
1686 dts2 = (msr >> 8) & 0x7F;
1687 fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1688 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1689#endif
1690 }
1691
1692
1693 if (do_dts) {
1694 unsigned int resolution;
1695
1696 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1697 return 0;
1698
1699 dts = (msr >> 16) & 0x7F;
1700 resolution = (msr >> 27) & 0xF;
1701 fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
1702 cpu, msr, tcc_activation_temp - dts, resolution);
1703
1704#ifdef THERM_DEBUG
1705 if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
1706 return 0;
1707
1708 dts = (msr >> 16) & 0x7F;
1709 dts2 = (msr >> 8) & 0x7F;
1710 fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1711 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1712#endif
1713 }
1714
1715 return 0;
1716}
1717
1718void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
1719{
1720 fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
1721 cpu, label,
1722 ((msr >> 15) & 1) ? "EN" : "DIS",
1723 ((msr >> 0) & 0x7FFF) * rapl_power_units,
1724 (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
1725 (((msr >> 16) & 1) ? "EN" : "DIS"));
1726
1727 return;
1728}
1729
1730int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1731{
1732 unsigned long long msr;
1733 int cpu;
889facbe
LB		 1734
1735 if (!do_rapl)
1736 return 0;
1737
1738 /* RAPL counters are per package, so print only for 1st thread/package */
1739 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1740 return 0;
1741
1742 cpu = t->cpu_id;
1743 if (cpu_migrate(cpu)) {
1744 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1745 return -1;
1746 }
1747
1748 if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
1749 return -1;
1750
889facbe
LB		 1751 if (verbose) {
1752 fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx "
1753 "(%f Watts, %f Joules, %f sec.)\n", cpu, msr,
144b44b1 1754 rapl_power_units, rapl_energy_units, rapl_time_units);
889facbe 1755 }
144b44b1
LB		 1756 if (do_rapl & RAPL_PKG_POWER_INFO) {
1757
889facbe
LB		 1758 if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
1759 return -5;
1760
1761
1762 fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
1763 cpu, msr,
1764 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1765 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1766 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1767 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
1768
144b44b1
LB		 1769 }
1770 if (do_rapl & RAPL_PKG) {
1771
889facbe
LB		 1772 if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
1773 return -9;
1774
1775 fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
1776 cpu, msr, (msr >> 63) & 1 ? "": "UN");
1777
1778 print_power_limit_msr(cpu, msr, "PKG Limit #1");
1779 fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
1780 cpu,
1781 ((msr >> 47) & 1) ? "EN" : "DIS",
1782 ((msr >> 32) & 0x7FFF) * rapl_power_units,
1783 (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
1784 ((msr >> 48) & 1) ? "EN" : "DIS");
1785 }
1786
1787 if (do_rapl & RAPL_DRAM) {
1788 if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
1789 return -6;
1790
1791
1792 fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
1793 cpu, msr,
1794 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1795 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1796 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1797 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
1798
1799
1800 if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
1801 return -9;
1802 fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
1803 cpu, msr, (msr >> 31) & 1 ? "": "UN");
1804
1805 print_power_limit_msr(cpu, msr, "DRAM Limit");
1806 }
144b44b1 1807 if (do_rapl & RAPL_CORE_POLICY) {
889facbe
LB		 1808 if (verbose) {
1809 if (get_msr(cpu, MSR_PP0_POLICY, &msr))
1810 return -7;
1811
1812 fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
144b44b1
LB		 1813 }
1814 }
1815 if (do_rapl & RAPL_CORES) {
1816 if (verbose) {
889facbe
LB		 1817
1818 if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
1819 return -9;
1820 fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
1821 cpu, msr, (msr >> 31) & 1 ? "": "UN");
1822 print_power_limit_msr(cpu, msr, "Cores Limit");
1823 }
1824 }
1825 if (do_rapl & RAPL_GFX) {
1826 if (verbose) {
1827 if (get_msr(cpu, MSR_PP1_POLICY, &msr))
1828 return -8;
1829
1830 fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
1831
1832 if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
1833 return -9;
1834 fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
1835 cpu, msr, (msr >> 31) & 1 ? "": "UN");
1836 print_power_limit_msr(cpu, msr, "GFX Limit");
1837 }
1838 }
1839 return 0;
1840}
1841
103a8fea
LB		 1842
1843int is_snb(unsigned int family, unsigned int model)
1844{
1845 if (!genuine_intel)
1846 return 0;
1847
1848 switch (model) {
1849 case 0x2A:
1850 case 0x2D:
650a37f3 1851 case 0x3A: /* IVB */
1300651b 1852 case 0x3E: /* IVB Xeon */
70b43400
LB		 1853 case 0x3C: /* HSW */
1854 case 0x3F: /* HSW */
1855 case 0x45: /* HSW */
149c2319 1856 case 0x46: /* HSW */
103a8fea
LB		 1857 return 1;
1858 }
1859 return 0;
1860}
1861
ca58710f
KCA		 1862int has_c8_c9_c10(unsigned int family, unsigned int model)
1863{
1864 if (!genuine_intel)
1865 return 0;
1866
1867 switch (model) {
1868 case 0x45:
1869 return 1;
1870 }
1871 return 0;
1872}
1873
1874
144b44b1
LB		 1875int is_slm(unsigned int family, unsigned int model)
1876{
1877 if (!genuine_intel)
1878 return 0;
1879 switch (model) {
1880 case 0x37: /* BYT */
1881 case 0x4D: /* AVN */
1882 return 1;
1883 }
1884 return 0;
1885}
1886
1887#define SLM_BCLK_FREQS 5
1888double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
1889
1890double slm_bclk(void)
1891{
1892 unsigned long long msr = 3;
1893 unsigned int i;
1894 double freq;
1895
1896 if (get_msr(0, MSR_FSB_FREQ, &msr))
1897 fprintf(stderr, "SLM BCLK: unknown\n");
1898
1899 i = msr & 0xf;
1900 if (i >= SLM_BCLK_FREQS) {
1901 fprintf(stderr, "SLM BCLK[%d] invalid\n", i);
1902 msr = 3;
1903 }
1904 freq = slm_freq_table[i];
1905
1906 fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq);
1907
1908 return freq;
1909}
1910
103a8fea
LB		 1911double discover_bclk(unsigned int family, unsigned int model)
1912{
1913 if (is_snb(family, model))
1914 return 100.00;
144b44b1
LB		 1915 else if (is_slm(family, model))
1916 return slm_bclk();
103a8fea
LB		 1917 else
1918 return 133.33;
1919}
1920
889facbe
LB		 1921/*
1922 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1923 * the Thermal Control Circuit (TCC) activates.
1924 * This is usually equal to tjMax.
1925 *
1926 * Older processors do not have this MSR, so there we guess,
1927 * but also allow cmdline over-ride with -T.
1928 *
1929 * Several MSR temperature values are in units of degrees-C
1930 * below this value, including the Digital Thermal Sensor (DTS),
1931 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1932 */
1933int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1934{
1935 unsigned long long msr;
1936 unsigned int target_c_local;
1937 int cpu;
1938
1939 /* tcc_activation_temp is used only for dts or ptm */
1940 if (!(do_dts || do_ptm))
1941 return 0;
1942
1943 /* this is a per-package concept */
1944 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1945 return 0;
1946
1947 cpu = t->cpu_id;
1948 if (cpu_migrate(cpu)) {
1949 fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1950 return -1;
1951 }
1952
1953 if (tcc_activation_temp_override != 0) {
1954 tcc_activation_temp = tcc_activation_temp_override;
1955 fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n",
1956 cpu, tcc_activation_temp);
1957 return 0;
1958 }
1959
1960 /* Temperature Target MSR is Nehalem and newer only */
1961 if (!do_nehalem_platform_info)
1962 goto guess;
1963
1964 if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1965 goto guess;
1966
1967 target_c_local = (msr >> 16) & 0x7F;
1968
1969 if (verbose)
1970 fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
1971 cpu, msr, target_c_local);
1972
144b44b1 1973 if (target_c_local < 85 || target_c_local > 127)
889facbe
LB		 1974 goto guess;
1975
1976 tcc_activation_temp = target_c_local;
1977
1978 return 0;
1979
1980guess:
1981 tcc_activation_temp = TJMAX_DEFAULT;
1982 fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
1983 cpu, tcc_activation_temp);
1984
1985 return 0;
1986}
103a8fea
LB		 1987void check_cpuid()
1988{
1989 unsigned int eax, ebx, ecx, edx, max_level;
1990 unsigned int fms, family, model, stepping;
1991
1992 eax = ebx = ecx = edx = 0;
1993
2b92865e 1994 __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
103a8fea
LB		 1995
1996 if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
1997 genuine_intel = 1;
1998
1999 if (verbose)
889facbe 2000 fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ",
103a8fea
LB		 2001 (char *)&ebx, (char *)&edx, (char *)&ecx);
2002
2b92865e 2003 __get_cpuid(1, &fms, &ebx, &ecx, &edx);
103a8fea
LB		 2004 family = (fms >> 8) & 0xf;
2005 model = (fms >> 4) & 0xf;
2006 stepping = fms & 0xf;
2007 if (family == 6 || family == 0xf)
2008 model += ((fms >> 16) & 0xf) << 4;
2009
2010 if (verbose)
2011 fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
2012 max_level, family, model, stepping, family, model, stepping);
2013
2014 if (!(edx & (1 << 5))) {
2015 fprintf(stderr, "CPUID: no MSR\n");
2016 exit(1);
2017 }
2018
2019 /*
2020 * check max extended function levels of CPUID.
2021 * This is needed to check for invariant TSC.
2022 * This check is valid for both Intel and AMD.
2023 */
2024 ebx = ecx = edx = 0;
2b92865e 2025 __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
103a8fea
LB		 2026
2027 if (max_level < 0x80000007) {
2028 fprintf(stderr, "CPUID: no invariant TSC (max_level 0x%x)\n", max_level);
2029 exit(1);
2030 }
2031
2032 /*
2033 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
2034 * this check is valid for both Intel and AMD
2035 */
2b92865e 2036 __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
8209e054 2037 has_invariant_tsc = edx & (1 << 8);
103a8fea
LB		 2038
2039 if (!has_invariant_tsc) {
2040 fprintf(stderr, "No invariant TSC\n");
2041 exit(1);
2042 }
2043
2044 /*
2045 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
2046 * this check is valid for both Intel and AMD
2047 */
2048
2b92865e 2049 __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
8209e054 2050 has_aperf = ecx & (1 << 0);
889facbe
LB		 2051 do_dts = eax & (1 << 0);
2052 do_ptm = eax & (1 << 6);
2053 has_epb = ecx & (1 << 3);
2054
2055 if (verbose)
2056 fprintf(stderr, "CPUID(6): %s%s%s%s\n",
2057 has_aperf ? "APERF" : "No APERF!",
2058 do_dts ? ", DTS" : "",
2059 do_ptm ? ", PTM": "",
2060 has_epb ? ", EPB": "");
2061
2062 if (!has_aperf)
2063 exit(-1);
103a8fea
LB		 2064
2065 do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
2066 do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
1ed51011 2067 do_smi = do_nhm_cstates;
103a8fea 2068 do_snb_cstates = is_snb(family, model);
ca58710f 2069 do_c8_c9_c10 = has_c8_c9_c10(family, model);
144b44b1 2070 do_slm_cstates = is_slm(family, model);
103a8fea
LB		 2071 bclk = discover_bclk(family, model);
2072
2073 do_nehalem_turbo_ratio_limit = has_nehalem_turbo_ratio_limit(family, model);
6574a5d5 2074 do_ivt_turbo_ratio_limit = has_ivt_turbo_ratio_limit(family, model);
889facbe
LB		 2075 rapl_probe(family, model);
2076
2077 return;
103a8fea
LB		 2078}
2079
2080
2081void usage()
2082{
889facbe 2083 fprintf(stderr, "%s: [-v][-R][-T][-p|-P|-S][-c MSR# | -s]][-C MSR#][-m MSR#][-M MSR#][-i interval_sec | command ...]\n",
103a8fea
LB		 2084 progname);
2085 exit(1);
2086}
2087
2088
2089/*
2090 * in /dev/cpu/ return success for names that are numbers
2091 * ie. filter out ".", "..", "microcode".
2092 */
2093int dir_filter(const struct dirent *dirp)
2094{
2095 if (isdigit(dirp->d_name[0]))
2096 return 1;
2097 else
2098 return 0;
2099}
2100
2101int open_dev_cpu_msr(int dummy1)
2102{
2103 return 0;
2104}
2105
c98d5d94
LB		 2106void topology_probe()
2107{
2108 int i;
2109 int max_core_id = 0;
2110 int max_package_id = 0;
2111 int max_siblings = 0;
2112 struct cpu_topology {
2113 int core_id;
2114 int physical_package_id;
2115 } *cpus;
2116
2117 /* Initialize num_cpus, max_cpu_num */
2118 topo.num_cpus = 0;
2119 topo.max_cpu_num = 0;
2120 for_all_proc_cpus(count_cpus);
2121 if (!summary_only && topo.num_cpus > 1)
2122 show_cpu = 1;
2123
2124 if (verbose > 1)
2125 fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
2126
2127 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
2128 if (cpus == NULL) {
2129 perror("calloc cpus");
2130 exit(1);
2131 }
2132
2133 /*
2134 * Allocate and initialize cpu_present_set
2135 */
2136 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
2137 if (cpu_present_set == NULL) {
2138 perror("CPU_ALLOC");
2139 exit(3);
2140 }
2141 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2142 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
2143 for_all_proc_cpus(mark_cpu_present);
2144
2145 /*
2146 * Allocate and initialize cpu_affinity_set
2147 */
2148 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
2149 if (cpu_affinity_set == NULL) {
2150 perror("CPU_ALLOC");
2151 exit(3);
2152 }
2153 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2154 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
2155
2156
2157 /*
2158 * For online cpus
2159 * find max_core_id, max_package_id
2160 */
2161 for (i = 0; i <= topo.max_cpu_num; ++i) {
2162 int siblings;
2163
2164 if (cpu_is_not_present(i)) {
2165 if (verbose > 1)
2166 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
2167 continue;
2168 }
2169 cpus[i].core_id = get_core_id(i);
2170 if (cpus[i].core_id > max_core_id)
2171 max_core_id = cpus[i].core_id;
2172
2173 cpus[i].physical_package_id = get_physical_package_id(i);
2174 if (cpus[i].physical_package_id > max_package_id)
2175 max_package_id = cpus[i].physical_package_id;
2176
2177 siblings = get_num_ht_siblings(i);
2178 if (siblings > max_siblings)
2179 max_siblings = siblings;
2180 if (verbose > 1)
2181 fprintf(stderr, "cpu %d pkg %d core %d\n",
2182 i, cpus[i].physical_package_id, cpus[i].core_id);
2183 }
2184 topo.num_cores_per_pkg = max_core_id + 1;
2185 if (verbose > 1)
2186 fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n",
2187 max_core_id, topo.num_cores_per_pkg);
2188 if (!summary_only && topo.num_cores_per_pkg > 1)
2189 show_core = 1;
2190
2191 topo.num_packages = max_package_id + 1;
2192 if (verbose > 1)
2193 fprintf(stderr, "max_package_id %d, sizing for %d packages\n",
2194 max_package_id, topo.num_packages);
2195 if (!summary_only && topo.num_packages > 1)
2196 show_pkg = 1;
2197
2198 topo.num_threads_per_core = max_siblings;
2199 if (verbose > 1)
2200 fprintf(stderr, "max_siblings %d\n", max_siblings);
2201
2202 free(cpus);
2203}
2204
2205void
2206allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
2207{
2208 int i;
2209
2210 *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
2211 topo.num_packages, sizeof(struct thread_data));
2212 if (*t == NULL)
2213 goto error;
2214
2215 for (i = 0; i < topo.num_threads_per_core *
2216 topo.num_cores_per_pkg * topo.num_packages; i++)
2217 (*t)[i].cpu_id = -1;
2218
2219 *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
2220 sizeof(struct core_data));
2221 if (*c == NULL)
2222 goto error;
2223
2224 for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
2225 (*c)[i].core_id = -1;
2226
2227 *p = calloc(topo.num_packages, sizeof(struct pkg_data));
2228 if (*p == NULL)
2229 goto error;
2230
2231 for (i = 0; i < topo.num_packages; i++)
2232 (*p)[i].package_id = i;
2233
2234 return;
2235error:
2236 perror("calloc counters");
2237 exit(1);
2238}
2239/*
2240 * init_counter()
2241 *
2242 * set cpu_id, core_num, pkg_num
2243 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
2244 *
2245 * increment topo.num_cores when 1st core in pkg seen
2246 */
2247void init_counter(struct thread_data *thread_base, struct core_data *core_base,
2248 struct pkg_data *pkg_base, int thread_num, int core_num,
2249 int pkg_num, int cpu_id)
2250{
2251 struct thread_data *t;
2252 struct core_data *c;
2253 struct pkg_data *p;
2254
2255 t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
2256 c = GET_CORE(core_base, core_num, pkg_num);
2257 p = GET_PKG(pkg_base, pkg_num);
2258
2259 t->cpu_id = cpu_id;
2260 if (thread_num == 0) {
2261 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
2262 if (cpu_is_first_core_in_package(cpu_id))
2263 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
2264 }
2265
2266 c->core_id = core_num;
2267 p->package_id = pkg_num;
2268}
2269
2270
2271int initialize_counters(int cpu_id)
2272{
2273 int my_thread_id, my_core_id, my_package_id;
2274
2275 my_package_id = get_physical_package_id(cpu_id);
2276 my_core_id = get_core_id(cpu_id);
2277
2278 if (cpu_is_first_sibling_in_core(cpu_id)) {
2279 my_thread_id = 0;
2280 topo.num_cores++;
2281 } else {
2282 my_thread_id = 1;
2283 }
2284
2285 init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2286 init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2287 return 0;
2288}
2289
2290void allocate_output_buffer()
2291{
b844db31 2292 output_buffer = calloc(1, (1 + topo.num_cpus) * 256);
c98d5d94
LB		 2293 outp = output_buffer;
2294 if (outp == NULL) {
2295 perror("calloc");
2296 exit(-1);
2297 }
2298}
2299
2300void setup_all_buffers(void)
2301{
2302 topology_probe();
2303 allocate_counters(&thread_even, &core_even, &package_even);
2304 allocate_counters(&thread_odd, &core_odd, &package_odd);
2305 allocate_output_buffer();
2306 for_all_proc_cpus(initialize_counters);
2307}
103a8fea
LB		 2308void turbostat_init()
2309{
2310 check_cpuid();
2311
2312 check_dev_msr();
2313 check_super_user();
2314
c98d5d94 2315 setup_all_buffers();
103a8fea
LB		 2316
2317 if (verbose)
c98d5d94 2318 print_verbose_header();
889facbe
LB		 2319
2320 if (verbose)
2321 for_all_cpus(print_epb, ODD_COUNTERS);
2322
2323 if (verbose)
2324 for_all_cpus(print_rapl, ODD_COUNTERS);
2325
2326 for_all_cpus(set_temperature_target, ODD_COUNTERS);
2327
2328 if (verbose)
2329 for_all_cpus(print_thermal, ODD_COUNTERS);
103a8fea
LB		 2330}
2331
2332int fork_it(char **argv)
2333{
103a8fea 2334 pid_t child_pid;
d91bb17c 2335 int status;
d15cf7c1 2336
d91bb17c
LB		 2337 status = for_all_cpus(get_counters, EVEN_COUNTERS);
2338 if (status)
2339 exit(status);
c98d5d94
LB		 2340 /* clear affinity side-effect of get_counters() */
2341 sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
103a8fea
LB		 2342 gettimeofday(&tv_even, (struct timezone *)NULL);
2343
2344 child_pid = fork();
2345 if (!child_pid) {
2346 /* child */
2347 execvp(argv[0], argv);
2348 } else {
103a8fea
LB		 2349
2350 /* parent */
2351 if (child_pid == -1) {
2352 perror("fork");
2353 exit(1);
2354 }
2355
2356 signal(SIGINT, SIG_IGN);
2357 signal(SIGQUIT, SIG_IGN);
2358 if (waitpid(child_pid, &status, 0) == -1) {
2359 perror("wait");
d91bb17c 2360 exit(status);
103a8fea
LB		 2361 }
2362 }
c98d5d94
LB		 2363 /*
2364 * n.b. fork_it() does not check for errors from for_all_cpus()
2365 * because re-starting is problematic when forking
2366 */
2367 for_all_cpus(get_counters, ODD_COUNTERS);
103a8fea 2368 gettimeofday(&tv_odd, (struct timezone *)NULL);
103a8fea 2369 timersub(&tv_odd, &tv_even, &tv_delta);
c98d5d94
LB		 2370 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
2371 compute_average(EVEN_COUNTERS);
2372 format_all_counters(EVEN_COUNTERS);
2373 flush_stderr();
103a8fea 2374
6eab04a8 2375 fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
103a8fea 2376
d91bb17c 2377 return status;
103a8fea
LB		 2378}
2379
2380void cmdline(int argc, char **argv)
2381{
2382 int opt;
2383
2384 progname = argv[0];
2385
889facbe 2386 while ((opt = getopt(argc, argv, "+pPSvi:sc:sC:m:M:RT:")) != -1) {
103a8fea 2387 switch (opt) {
f9240813 2388 case 'p':
c98d5d94
LB		 2389 show_core_only++;
2390 break;
f9240813 2391 case 'P':
c98d5d94
LB		 2392 show_pkg_only++;
2393 break;
f9240813 2394 case 'S':
e23da037
LB		 2395 summary_only++;
2396 break;
103a8fea
LB		 2397 case 'v':
2398 verbose++;
2399 break;
2400 case 'i':
2401 interval_sec = atoi(optarg);
2402 break;
f9240813 2403 case 'c':
8e180f3c
LB		 2404 sscanf(optarg, "%x", &extra_delta_offset32);
2405 break;
f9240813 2406 case 'C':
8e180f3c
LB		 2407 sscanf(optarg, "%x", &extra_delta_offset64);
2408 break;
2f32edf1
LB		 2409 case 'm':
2410 sscanf(optarg, "%x", &extra_msr_offset32);
2f32edf1 2411 break;
103a8fea 2412 case 'M':
2f32edf1 2413 sscanf(optarg, "%x", &extra_msr_offset64);
103a8fea 2414 break;
889facbe
LB		 2415 case 'R':
2416 rapl_verbose++;
2417 break;
2418 case 'T':
2419 tcc_activation_temp_override = atoi(optarg);
2420 break;
103a8fea
LB		 2421 default:
2422 usage();
2423 }
2424 }
2425}
2426
2427int main(int argc, char **argv)
2428{
2429 cmdline(argc, argv);
2430
889facbe 2431 if (verbose)
144b44b1 2432 fprintf(stderr, "turbostat v3.5 April 26, 2013"
103a8fea 2433 " - Len Brown <lenb@kernel.org>\n");
103a8fea
LB		 2434
2435 turbostat_init();
2436
2437 /*
2438 * if any params left, it must be a command to fork
2439 */
2440 if (argc - optind)
2441 return fork_it(argv + optind);
2442 else
2443 turbostat_loop();
2444
2445 return 0;
2446}
Ubuntu Artful kernel mirror