1 /*
2 * turbostat -- show CPU frequency and C-state residency
3 * on modern Intel turbo-capable processors.
4 *
5 * Copyright (c) 2013 Intel Corporation.
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
22 #define _GNU_SOURCE
23 #include MSRHEADER
24 #include <stdarg.h>
25 #include <stdio.h>
26 #include <err.h>
27 #include <unistd.h>
28 #include <sys/types.h>
29 #include <sys/wait.h>
30 #include <sys/stat.h>
31 #include <sys/resource.h>
32 #include <fcntl.h>
33 #include <signal.h>
34 #include <sys/time.h>
35 #include <stdlib.h>
36 #include <getopt.h>
37 #include <dirent.h>
38 #include <string.h>
39 #include <ctype.h>
40 #include <sched.h>
41 #include <time.h>
42 #include <cpuid.h>
43 #include <linux/capability.h>
44 #include <errno.h>
45
46 char *proc_stat = "/proc/stat";
47 FILE *outf;
48 int *fd_percpu;
49 struct timespec interval_ts = {5, 0};
50 unsigned int debug;
51 unsigned int rapl_joules;
52 unsigned int summary_only;
53 unsigned int dump_only;
54 unsigned int skip_c0;
55 unsigned int skip_c1;
56 unsigned int do_nhm_cstates;
57 unsigned int do_snb_cstates;
58 unsigned int do_knl_cstates;
59 unsigned int do_pc2;
60 unsigned int do_pc3;
61 unsigned int do_pc6;
62 unsigned int do_pc7;
63 unsigned int do_c8_c9_c10;
64 unsigned int do_skl_residency;
65 unsigned int do_slm_cstates;
66 unsigned int use_c1_residency_msr;
67 unsigned int has_aperf;
68 unsigned int has_epb;
69 unsigned int do_irtl_snb;
70 unsigned int do_irtl_hsw;
71 unsigned int units = 1000000; /* MHz etc */
72 unsigned int genuine_intel;
73 unsigned int has_invariant_tsc;
74 unsigned int do_nhm_platform_info;
75 unsigned int extra_msr_offset32;
76 unsigned int extra_msr_offset64;
77 unsigned int extra_delta_offset32;
78 unsigned int extra_delta_offset64;
79 unsigned int aperf_mperf_multiplier = 1;
80 int do_irq = 1;
81 int do_smi;
82 double bclk;
83 double base_hz;
84 unsigned int has_base_hz;
85 double tsc_tweak = 1.0;
86 unsigned int show_pkg;
87 unsigned int show_core;
88 unsigned int show_cpu;
89 unsigned int show_pkg_only;
90 unsigned int show_core_only;
91 char *output_buffer, *outp;
92 unsigned int do_rapl;
93 unsigned int do_dts;
94 unsigned int do_ptm;
95 unsigned int do_gfx_rc6_ms;
96 unsigned long long gfx_cur_rc6_ms;
97 unsigned int do_gfx_mhz;
98 unsigned int gfx_cur_mhz;
99 unsigned int tcc_activation_temp;
100 unsigned int tcc_activation_temp_override;
101 double rapl_power_units, rapl_time_units;
102 double rapl_dram_energy_units, rapl_energy_units;
103 double rapl_joule_counter_range;
104 unsigned int do_core_perf_limit_reasons;
105 unsigned int do_gfx_perf_limit_reasons;
106 unsigned int do_ring_perf_limit_reasons;
107 unsigned int crystal_hz;
108 unsigned long long tsc_hz;
109 int base_cpu;
110 double discover_bclk(unsigned int family, unsigned int model);
111 unsigned int has_hwp; /* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
112 /* IA32_HWP_REQUEST, IA32_HWP_STATUS */
113 unsigned int has_hwp_notify; /* IA32_HWP_INTERRUPT */
114 unsigned int has_hwp_activity_window; /* IA32_HWP_REQUEST[bits 41:32] */
115 unsigned int has_hwp_epp; /* IA32_HWP_REQUEST[bits 31:24] */
116 unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */
117
118 #define RAPL_PKG (1 << 0)
119 /* 0x610 MSR_PKG_POWER_LIMIT */
120 /* 0x611 MSR_PKG_ENERGY_STATUS */
121 #define RAPL_PKG_PERF_STATUS (1 << 1)
122 /* 0x613 MSR_PKG_PERF_STATUS */
123 #define RAPL_PKG_POWER_INFO (1 << 2)
124 /* 0x614 MSR_PKG_POWER_INFO */
125
126 #define RAPL_DRAM (1 << 3)
127 /* 0x618 MSR_DRAM_POWER_LIMIT */
128 /* 0x619 MSR_DRAM_ENERGY_STATUS */
129 #define RAPL_DRAM_PERF_STATUS (1 << 4)
130 /* 0x61b MSR_DRAM_PERF_STATUS */
131 #define RAPL_DRAM_POWER_INFO (1 << 5)
132 /* 0x61c MSR_DRAM_POWER_INFO */
133
134 #define RAPL_CORES (1 << 6)
135 /* 0x638 MSR_PP0_POWER_LIMIT */
136 /* 0x639 MSR_PP0_ENERGY_STATUS */
137 #define RAPL_CORE_POLICY (1 << 7)
138 /* 0x63a MSR_PP0_POLICY */
139
140 #define RAPL_GFX (1 << 8)
141 /* 0x640 MSR_PP1_POWER_LIMIT */
142 /* 0x641 MSR_PP1_ENERGY_STATUS */
143 /* 0x642 MSR_PP1_POLICY */
144 #define TJMAX_DEFAULT 100
145
146 #define MAX(a, b) ((a) > (b) ? (a) : (b))
147
148 int aperf_mperf_unstable;
149 int backwards_count;
150 char *progname;
151
152 cpu_set_t *cpu_present_set, *cpu_affinity_set;
153 size_t cpu_present_setsize, cpu_affinity_setsize;
154
155 struct thread_data {
156 unsigned long long tsc;
157 unsigned long long aperf;
158 unsigned long long mperf;
159 unsigned long long c1;
160 unsigned long long extra_msr64;
161 unsigned long long extra_delta64;
162 unsigned long long extra_msr32;
163 unsigned long long extra_delta32;
164 unsigned int irq_count;
165 unsigned int smi_count;
166 unsigned int cpu_id;
167 unsigned int flags;
168 #define CPU_IS_FIRST_THREAD_IN_CORE 0x2
169 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
170 } *thread_even, *thread_odd;
171
172 struct core_data {
173 unsigned long long c3;
174 unsigned long long c6;
175 unsigned long long c7;
176 unsigned int core_temp_c;
177 unsigned int core_id;
178 } *core_even, *core_odd;
179
180 struct pkg_data {
181 unsigned long long pc2;
182 unsigned long long pc3;
183 unsigned long long pc6;
184 unsigned long long pc7;
185 unsigned long long pc8;
186 unsigned long long pc9;
187 unsigned long long pc10;
188 unsigned long long pkg_wtd_core_c0;
189 unsigned long long pkg_any_core_c0;
190 unsigned long long pkg_any_gfxe_c0;
191 unsigned long long pkg_both_core_gfxe_c0;
192 long long gfx_rc6_ms;
193 unsigned int gfx_mhz;
194 unsigned int package_id;
195 unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */
196 unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */
197 unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */
198 unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */
199 unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */
200 unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */
201 unsigned int pkg_temp_c;
202
203 } *package_even, *package_odd;
204
205 #define ODD_COUNTERS thread_odd, core_odd, package_odd
206 #define EVEN_COUNTERS thread_even, core_even, package_even
207
208 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
209 (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
210 topo.num_threads_per_core + \
211 (core_no) * topo.num_threads_per_core + (thread_no))
212 #define GET_CORE(core_base, core_no, pkg_no) \
213 (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
214 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
215
216 struct system_summary {
217 struct thread_data threads;
218 struct core_data cores;
219 struct pkg_data packages;
220 } sum, average;
221
222
223 struct topo_params {
224 int num_packages;
225 int num_cpus;
226 int num_cores;
227 int max_cpu_num;
228 int num_cores_per_pkg;
229 int num_threads_per_core;
230 } topo;
231
232 struct timeval tv_even, tv_odd, tv_delta;
233
234 int *irq_column_2_cpu; /* /proc/interrupts column numbers */
235 int *irqs_per_cpu; /* indexed by cpu_num */
236
237 void setup_all_buffers(void);
238
cpu_is_not_present(int cpu)239 int cpu_is_not_present(int cpu)
240 {
241 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
242 }
243 /*
244 * run func(thread, core, package) in topology order
245 * skip non-present cpus
246 */
247
for_all_cpus(int (func)(struct thread_data *,struct core_data *,struct pkg_data *),struct thread_data * thread_base,struct core_data * core_base,struct pkg_data * pkg_base)248 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
249 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
250 {
251 int retval, pkg_no, core_no, thread_no;
252
253 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
254 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
255 for (thread_no = 0; thread_no <
256 topo.num_threads_per_core; ++thread_no) {
257 struct thread_data *t;
258 struct core_data *c;
259 struct pkg_data *p;
260
261 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
262
263 if (cpu_is_not_present(t->cpu_id))
264 continue;
265
266 c = GET_CORE(core_base, core_no, pkg_no);
267 p = GET_PKG(pkg_base, pkg_no);
268
269 retval = func(t, c, p);
270 if (retval)
271 return retval;
272 }
273 }
274 }
275 return 0;
276 }
277
cpu_migrate(int cpu)278 int cpu_migrate(int cpu)
279 {
280 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
281 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
282 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
283 return -1;
284 else
285 return 0;
286 }
get_msr_fd(int cpu)287 int get_msr_fd(int cpu)
288 {
289 char pathname[32];
290 int fd;
291
292 fd = fd_percpu[cpu];
293
294 if (fd)
295 return fd;
296
297 sprintf(pathname, "/dev/cpu/%d/msr", cpu);
298 fd = open(pathname, O_RDONLY);
299 if (fd < 0)
300 err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
301
302 fd_percpu[cpu] = fd;
303
304 return fd;
305 }
306
get_msr(int cpu,off_t offset,unsigned long long * msr)307 int get_msr(int cpu, off_t offset, unsigned long long *msr)
308 {
309 ssize_t retval;
310
311 retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
312
313 if (retval != sizeof *msr)
314 err(-1, "msr %d offset 0x%llx read failed", cpu, (unsigned long long)offset);
315
316 return 0;
317 }
318
319 /*
320 * Example Format w/ field column widths:
321 *
322 * Package Core CPU Avg_MHz Bzy_MHz TSC_MHz IRQ SMI Busy% CPU_%c1 CPU_%c3 CPU_%c6 CPU_%c7 CoreTmp PkgTmp GFXMHz Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 PkgWatt CorWatt GFXWatt
323 * 12345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678
324 */
325
print_header(void)326 void print_header(void)
327 {
328 if (show_pkg)
329 outp += sprintf(outp, " Package");
330 if (show_core)
331 outp += sprintf(outp, " Core");
332 if (show_cpu)
333 outp += sprintf(outp, " CPU");
334 if (has_aperf)
335 outp += sprintf(outp, " Avg_MHz");
336 if (has_aperf)
337 outp += sprintf(outp, " Busy%%");
338 if (has_aperf)
339 outp += sprintf(outp, " Bzy_MHz");
340 outp += sprintf(outp, " TSC_MHz");
341
342 if (extra_delta_offset32)
343 outp += sprintf(outp, " count 0x%03X", extra_delta_offset32);
344 if (extra_delta_offset64)
345 outp += sprintf(outp, " COUNT 0x%03X", extra_delta_offset64);
346 if (extra_msr_offset32)
347 outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32);
348 if (extra_msr_offset64)
349 outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64);
350
351 if (!debug)
352 goto done;
353
354 if (do_irq)
355 outp += sprintf(outp, " IRQ");
356 if (do_smi)
357 outp += sprintf(outp, " SMI");
358
359 if (do_nhm_cstates)
360 outp += sprintf(outp, " CPU%%c1");
361 if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates)
362 outp += sprintf(outp, " CPU%%c3");
363 if (do_nhm_cstates)
364 outp += sprintf(outp, " CPU%%c6");
365 if (do_snb_cstates)
366 outp += sprintf(outp, " CPU%%c7");
367
368 if (do_dts)
369 outp += sprintf(outp, " CoreTmp");
370 if (do_ptm)
371 outp += sprintf(outp, " PkgTmp");
372
373 if (do_gfx_rc6_ms)
374 outp += sprintf(outp, " GFX%%rc6");
375
376 if (do_gfx_mhz)
377 outp += sprintf(outp, " GFXMHz");
378
379 if (do_skl_residency) {
380 outp += sprintf(outp, " Totl%%C0");
381 outp += sprintf(outp, " Any%%C0");
382 outp += sprintf(outp, " GFX%%C0");
383 outp += sprintf(outp, " CPUGFX%%");
384 }
385
386 if (do_pc2)
387 outp += sprintf(outp, " Pkg%%pc2");
388 if (do_pc3)
389 outp += sprintf(outp, " Pkg%%pc3");
390 if (do_pc6)
391 outp += sprintf(outp, " Pkg%%pc6");
392 if (do_pc7)
393 outp += sprintf(outp, " Pkg%%pc7");
394 if (do_c8_c9_c10) {
395 outp += sprintf(outp, " Pkg%%pc8");
396 outp += sprintf(outp, " Pkg%%pc9");
397 outp += sprintf(outp, " Pk%%pc10");
398 }
399
400 if (do_rapl && !rapl_joules) {
401 if (do_rapl & RAPL_PKG)
402 outp += sprintf(outp, " PkgWatt");
403 if (do_rapl & RAPL_CORES)
404 outp += sprintf(outp, " CorWatt");
405 if (do_rapl & RAPL_GFX)
406 outp += sprintf(outp, " GFXWatt");
407 if (do_rapl & RAPL_DRAM)
408 outp += sprintf(outp, " RAMWatt");
409 if (do_rapl & RAPL_PKG_PERF_STATUS)
410 outp += sprintf(outp, " PKG_%%");
411 if (do_rapl & RAPL_DRAM_PERF_STATUS)
412 outp += sprintf(outp, " RAM_%%");
413 } else if (do_rapl && rapl_joules) {
414 if (do_rapl & RAPL_PKG)
415 outp += sprintf(outp, " Pkg_J");
416 if (do_rapl & RAPL_CORES)
417 outp += sprintf(outp, " Cor_J");
418 if (do_rapl & RAPL_GFX)
419 outp += sprintf(outp, " GFX_J");
420 if (do_rapl & RAPL_DRAM)
421 outp += sprintf(outp, " RAM_J");
422 if (do_rapl & RAPL_PKG_PERF_STATUS)
423 outp += sprintf(outp, " PKG_%%");
424 if (do_rapl & RAPL_DRAM_PERF_STATUS)
425 outp += sprintf(outp, " RAM_%%");
426 outp += sprintf(outp, " time");
427
428 }
429 done:
430 outp += sprintf(outp, "\n");
431 }
432
dump_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)433 int dump_counters(struct thread_data *t, struct core_data *c,
434 struct pkg_data *p)
435 {
436 outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
437
438 if (t) {
439 outp += sprintf(outp, "CPU: %d flags 0x%x\n",
440 t->cpu_id, t->flags);
441 outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
442 outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
443 outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
444 outp += sprintf(outp, "c1: %016llX\n", t->c1);
445 outp += sprintf(outp, "msr0x%x: %08llX\n",
446 extra_delta_offset32, t->extra_delta32);
447 outp += sprintf(outp, "msr0x%x: %016llX\n",
448 extra_delta_offset64, t->extra_delta64);
449 outp += sprintf(outp, "msr0x%x: %08llX\n",
450 extra_msr_offset32, t->extra_msr32);
451 outp += sprintf(outp, "msr0x%x: %016llX\n",
452 extra_msr_offset64, t->extra_msr64);
453 if (do_irq)
454 outp += sprintf(outp, "IRQ: %08X\n", t->irq_count);
455 if (do_smi)
456 outp += sprintf(outp, "SMI: %08X\n", t->smi_count);
457 }
458
459 if (c) {
460 outp += sprintf(outp, "core: %d\n", c->core_id);
461 outp += sprintf(outp, "c3: %016llX\n", c->c3);
462 outp += sprintf(outp, "c6: %016llX\n", c->c6);
463 outp += sprintf(outp, "c7: %016llX\n", c->c7);
464 outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
465 }
466
467 if (p) {
468 outp += sprintf(outp, "package: %d\n", p->package_id);
469
470 outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
471 outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
472 outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
473 outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
474
475 outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
476 if (do_pc3)
477 outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
478 if (do_pc6)
479 outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
480 if (do_pc7)
481 outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
482 outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
483 outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
484 outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
485 outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
486 outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
487 outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
488 outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
489 outp += sprintf(outp, "Throttle PKG: %0X\n",
490 p->rapl_pkg_perf_status);
491 outp += sprintf(outp, "Throttle RAM: %0X\n",
492 p->rapl_dram_perf_status);
493 outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
494 }
495
496 outp += sprintf(outp, "\n");
497
498 return 0;
499 }
500
501 /*
502 * column formatting convention & formats
503 */
format_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)504 int format_counters(struct thread_data *t, struct core_data *c,
505 struct pkg_data *p)
506 {
507 double interval_float;
508 char *fmt8;
509
510 /* if showing only 1st thread in core and this isn't one, bail out */
511 if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
512 return 0;
513
514 /* if showing only 1st thread in pkg and this isn't one, bail out */
515 if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
516 return 0;
517
518 interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
519
520 /* topo columns, print blanks on 1st (average) line */
521 if (t == &average.threads) {
522 if (show_pkg)
523 outp += sprintf(outp, " -");
524 if (show_core)
525 outp += sprintf(outp, " -");
526 if (show_cpu)
527 outp += sprintf(outp, " -");
528 } else {
529 if (show_pkg) {
530 if (p)
531 outp += sprintf(outp, "%8d", p->package_id);
532 else
533 outp += sprintf(outp, " -");
534 }
535 if (show_core) {
536 if (c)
537 outp += sprintf(outp, "%8d", c->core_id);
538 else
539 outp += sprintf(outp, " -");
540 }
541 if (show_cpu)
542 outp += sprintf(outp, "%8d", t->cpu_id);
543 }
544
545 /* Avg_MHz */
546 if (has_aperf)
547 outp += sprintf(outp, "%8.0f",
548 1.0 / units * t->aperf / interval_float);
549
550 /* Busy% */
551 if (has_aperf) {
552 if (!skip_c0)
553 outp += sprintf(outp, "%8.2f", 100.0 * t->mperf/t->tsc/tsc_tweak);
554 else
555 outp += sprintf(outp, "********");
556 }
557
558 /* Bzy_MHz */
559 if (has_aperf) {
560 if (has_base_hz)
561 outp += sprintf(outp, "%8.0f", base_hz / units * t->aperf / t->mperf);
562 else
563 outp += sprintf(outp, "%8.0f",
564 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
565 }
566
567 /* TSC_MHz */
568 outp += sprintf(outp, "%8.0f", 1.0 * t->tsc/units/interval_float);
569
570 /* delta */
571 if (extra_delta_offset32)
572 outp += sprintf(outp, " %11llu", t->extra_delta32);
573
574 /* DELTA */
575 if (extra_delta_offset64)
576 outp += sprintf(outp, " %11llu", t->extra_delta64);
577 /* msr */
578 if (extra_msr_offset32)
579 outp += sprintf(outp, " 0x%08llx", t->extra_msr32);
580
581 /* MSR */
582 if (extra_msr_offset64)
583 outp += sprintf(outp, " 0x%016llx", t->extra_msr64);
584
585 if (!debug)
586 goto done;
587
588 /* IRQ */
589 if (do_irq)
590 outp += sprintf(outp, "%8d", t->irq_count);
591
592 /* SMI */
593 if (do_smi)
594 outp += sprintf(outp, "%8d", t->smi_count);
595
596 if (do_nhm_cstates) {
597 if (!skip_c1)
598 outp += sprintf(outp, "%8.2f", 100.0 * t->c1/t->tsc);
599 else
600 outp += sprintf(outp, "********");
601 }
602
603 /* print per-core data only for 1st thread in core */
604 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
605 goto done;
606
607 if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates)
608 outp += sprintf(outp, "%8.2f", 100.0 * c->c3/t->tsc);
609 if (do_nhm_cstates)
610 outp += sprintf(outp, "%8.2f", 100.0 * c->c6/t->tsc);
611 if (do_snb_cstates)
612 outp += sprintf(outp, "%8.2f", 100.0 * c->c7/t->tsc);
613
614 if (do_dts)
615 outp += sprintf(outp, "%8d", c->core_temp_c);
616
617 /* print per-package data only for 1st core in package */
618 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
619 goto done;
620
621 /* PkgTmp */
622 if (do_ptm)
623 outp += sprintf(outp, "%8d", p->pkg_temp_c);
624
625 /* GFXrc6 */
626 if (do_gfx_rc6_ms) {
627 if (p->gfx_rc6_ms == -1) { /* detect counter reset */
628 outp += sprintf(outp, " ***.**");
629 } else {
630 outp += sprintf(outp, "%8.2f",
631 p->gfx_rc6_ms / 10.0 / interval_float);
632 }
633 }
634
635 /* GFXMHz */
636 if (do_gfx_mhz)
637 outp += sprintf(outp, "%8d", p->gfx_mhz);
638
639 /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
640 if (do_skl_residency) {
641 outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_wtd_core_c0/t->tsc);
642 outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_core_c0/t->tsc);
643 outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_gfxe_c0/t->tsc);
644 outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_both_core_gfxe_c0/t->tsc);
645 }
646
647 if (do_pc2)
648 outp += sprintf(outp, "%8.2f", 100.0 * p->pc2/t->tsc);
649 if (do_pc3)
650 outp += sprintf(outp, "%8.2f", 100.0 * p->pc3/t->tsc);
651 if (do_pc6)
652 outp += sprintf(outp, "%8.2f", 100.0 * p->pc6/t->tsc);
653 if (do_pc7)
654 outp += sprintf(outp, "%8.2f", 100.0 * p->pc7/t->tsc);
655 if (do_c8_c9_c10) {
656 outp += sprintf(outp, "%8.2f", 100.0 * p->pc8/t->tsc);
657 outp += sprintf(outp, "%8.2f", 100.0 * p->pc9/t->tsc);
658 outp += sprintf(outp, "%8.2f", 100.0 * p->pc10/t->tsc);
659 }
660
661 /*
662 * If measurement interval exceeds minimum RAPL Joule Counter range,
663 * indicate that results are suspect by printing "**" in fraction place.
664 */
665 if (interval_float < rapl_joule_counter_range)
666 fmt8 = "%8.2f";
667 else
668 fmt8 = " %6.0f**";
669
670 if (do_rapl && !rapl_joules) {
671 if (do_rapl & RAPL_PKG)
672 outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float);
673 if (do_rapl & RAPL_CORES)
674 outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float);
675 if (do_rapl & RAPL_GFX)
676 outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float);
677 if (do_rapl & RAPL_DRAM)
678 outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units / interval_float);
679 if (do_rapl & RAPL_PKG_PERF_STATUS)
680 outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
681 if (do_rapl & RAPL_DRAM_PERF_STATUS)
682 outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
683 } else if (do_rapl && rapl_joules) {
684 if (do_rapl & RAPL_PKG)
685 outp += sprintf(outp, fmt8,
686 p->energy_pkg * rapl_energy_units);
687 if (do_rapl & RAPL_CORES)
688 outp += sprintf(outp, fmt8,
689 p->energy_cores * rapl_energy_units);
690 if (do_rapl & RAPL_GFX)
691 outp += sprintf(outp, fmt8,
692 p->energy_gfx * rapl_energy_units);
693 if (do_rapl & RAPL_DRAM)
694 outp += sprintf(outp, fmt8,
695 p->energy_dram * rapl_dram_energy_units);
696 if (do_rapl & RAPL_PKG_PERF_STATUS)
697 outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
698 if (do_rapl & RAPL_DRAM_PERF_STATUS)
699 outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
700
701 outp += sprintf(outp, fmt8, interval_float);
702 }
703 done:
704 outp += sprintf(outp, "\n");
705
706 return 0;
707 }
708
flush_output_stdout(void)709 void flush_output_stdout(void)
710 {
711 FILE *filep;
712
713 if (outf == stderr)
714 filep = stdout;
715 else
716 filep = outf;
717
718 fputs(output_buffer, filep);
719 fflush(filep);
720
721 outp = output_buffer;
722 }
flush_output_stderr(void)723 void flush_output_stderr(void)
724 {
725 fputs(output_buffer, outf);
726 fflush(outf);
727 outp = output_buffer;
728 }
format_all_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)729 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
730 {
731 static int printed;
732
733 if (!printed || !summary_only)
734 print_header();
735
736 if (topo.num_cpus > 1)
737 format_counters(&average.threads, &average.cores,
738 &average.packages);
739
740 printed = 1;
741
742 if (summary_only)
743 return;
744
745 for_all_cpus(format_counters, t, c, p);
746 }
747
748 #define DELTA_WRAP32(new, old) \
749 if (new > old) { \
750 old = new - old; \
751 } else { \
752 old = 0x100000000 + new - old; \
753 }
754
755 void
delta_package(struct pkg_data * new,struct pkg_data * old)756 delta_package(struct pkg_data *new, struct pkg_data *old)
757 {
758
759 if (do_skl_residency) {
760 old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
761 old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
762 old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
763 old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
764 }
765 old->pc2 = new->pc2 - old->pc2;
766 if (do_pc3)
767 old->pc3 = new->pc3 - old->pc3;
768 if (do_pc6)
769 old->pc6 = new->pc6 - old->pc6;
770 if (do_pc7)
771 old->pc7 = new->pc7 - old->pc7;
772 old->pc8 = new->pc8 - old->pc8;
773 old->pc9 = new->pc9 - old->pc9;
774 old->pc10 = new->pc10 - old->pc10;
775 old->pkg_temp_c = new->pkg_temp_c;
776
777 /* flag an error when rc6 counter resets/wraps */
778 if (old->gfx_rc6_ms > new->gfx_rc6_ms)
779 old->gfx_rc6_ms = -1;
780 else
781 old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
782
783 old->gfx_mhz = new->gfx_mhz;
784
785 DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
786 DELTA_WRAP32(new->energy_cores, old->energy_cores);
787 DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
788 DELTA_WRAP32(new->energy_dram, old->energy_dram);
789 DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
790 DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
791 }
792
793 void
delta_core(struct core_data * new,struct core_data * old)794 delta_core(struct core_data *new, struct core_data *old)
795 {
796 old->c3 = new->c3 - old->c3;
797 old->c6 = new->c6 - old->c6;
798 old->c7 = new->c7 - old->c7;
799 old->core_temp_c = new->core_temp_c;
800 }
801
802 /*
803 * old = new - old
804 */
805 void
delta_thread(struct thread_data * new,struct thread_data * old,struct core_data * core_delta)806 delta_thread(struct thread_data *new, struct thread_data *old,
807 struct core_data *core_delta)
808 {
809 old->tsc = new->tsc - old->tsc;
810
811 /* check for TSC < 1 Mcycles over interval */
812 if (old->tsc < (1000 * 1000))
813 errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
814 "You can disable all c-states by booting with \"idle=poll\"\n"
815 "or just the deep ones with \"processor.max_cstate=1\"");
816
817 old->c1 = new->c1 - old->c1;
818
819 if (has_aperf) {
820 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
821 old->aperf = new->aperf - old->aperf;
822 old->mperf = new->mperf - old->mperf;
823 } else {
824
825 if (!aperf_mperf_unstable) {
826 fprintf(outf, "%s: APERF or MPERF went backwards *\n", progname);
827 fprintf(outf, "* Frequency results do not cover entire interval *\n");
828 fprintf(outf, "* fix this by running Linux-2.6.30 or later *\n");
829
830 aperf_mperf_unstable = 1;
831 }
832 /*
833 * mperf delta is likely a huge "positive" number
834 * can not use it for calculating c0 time
835 */
836 skip_c0 = 1;
837 skip_c1 = 1;
838 }
839 }
840
841
842 if (use_c1_residency_msr) {
843 /*
844 * Some models have a dedicated C1 residency MSR,
845 * which should be more accurate than the derivation below.
846 */
847 } else {
848 /*
849 * As counter collection is not atomic,
850 * it is possible for mperf's non-halted cycles + idle states
851 * to exceed TSC's all cycles: show c1 = 0% in that case.
852 */
853 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
854 old->c1 = 0;
855 else {
856 /* normal case, derive c1 */
857 old->c1 = old->tsc - old->mperf - core_delta->c3
858 - core_delta->c6 - core_delta->c7;
859 }
860 }
861
862 if (old->mperf == 0) {
863 if (debug > 1)
864 fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
865 old->mperf = 1; /* divide by 0 protection */
866 }
867
868 old->extra_delta32 = new->extra_delta32 - old->extra_delta32;
869 old->extra_delta32 &= 0xFFFFFFFF;
870
871 old->extra_delta64 = new->extra_delta64 - old->extra_delta64;
872
873 /*
874 * Extra MSR is just a snapshot, simply copy latest w/o subtracting
875 */
876 old->extra_msr32 = new->extra_msr32;
877 old->extra_msr64 = new->extra_msr64;
878
879 if (do_irq)
880 old->irq_count = new->irq_count - old->irq_count;
881
882 if (do_smi)
883 old->smi_count = new->smi_count - old->smi_count;
884 }
885
delta_cpu(struct thread_data * t,struct core_data * c,struct pkg_data * p,struct thread_data * t2,struct core_data * c2,struct pkg_data * p2)886 int delta_cpu(struct thread_data *t, struct core_data *c,
887 struct pkg_data *p, struct thread_data *t2,
888 struct core_data *c2, struct pkg_data *p2)
889 {
890 /* calculate core delta only for 1st thread in core */
891 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
892 delta_core(c, c2);
893
894 /* always calculate thread delta */
895 delta_thread(t, t2, c2); /* c2 is core delta */
896
897 /* calculate package delta only for 1st core in package */
898 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
899 delta_package(p, p2);
900
901 return 0;
902 }
903
clear_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)904 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
905 {
906 t->tsc = 0;
907 t->aperf = 0;
908 t->mperf = 0;
909 t->c1 = 0;
910
911 t->extra_delta32 = 0;
912 t->extra_delta64 = 0;
913
914 t->irq_count = 0;
915 t->smi_count = 0;
916
917 /* tells format_counters to dump all fields from this set */
918 t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
919
920 c->c3 = 0;
921 c->c6 = 0;
922 c->c7 = 0;
923 c->core_temp_c = 0;
924
925 p->pkg_wtd_core_c0 = 0;
926 p->pkg_any_core_c0 = 0;
927 p->pkg_any_gfxe_c0 = 0;
928 p->pkg_both_core_gfxe_c0 = 0;
929
930 p->pc2 = 0;
931 if (do_pc3)
932 p->pc3 = 0;
933 if (do_pc6)
934 p->pc6 = 0;
935 if (do_pc7)
936 p->pc7 = 0;
937 p->pc8 = 0;
938 p->pc9 = 0;
939 p->pc10 = 0;
940
941 p->energy_pkg = 0;
942 p->energy_dram = 0;
943 p->energy_cores = 0;
944 p->energy_gfx = 0;
945 p->rapl_pkg_perf_status = 0;
946 p->rapl_dram_perf_status = 0;
947 p->pkg_temp_c = 0;
948
949 p->gfx_rc6_ms = 0;
950 p->gfx_mhz = 0;
951 }
sum_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)952 int sum_counters(struct thread_data *t, struct core_data *c,
953 struct pkg_data *p)
954 {
955 average.threads.tsc += t->tsc;
956 average.threads.aperf += t->aperf;
957 average.threads.mperf += t->mperf;
958 average.threads.c1 += t->c1;
959
960 average.threads.extra_delta32 += t->extra_delta32;
961 average.threads.extra_delta64 += t->extra_delta64;
962
963 average.threads.irq_count += t->irq_count;
964 average.threads.smi_count += t->smi_count;
965
966 /* sum per-core values only for 1st thread in core */
967 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
968 return 0;
969
970 average.cores.c3 += c->c3;
971 average.cores.c6 += c->c6;
972 average.cores.c7 += c->c7;
973
974 average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
975
976 /* sum per-pkg values only for 1st core in pkg */
977 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
978 return 0;
979
980 if (do_skl_residency) {
981 average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
982 average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
983 average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
984 average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
985 }
986
987 average.packages.pc2 += p->pc2;
988 if (do_pc3)
989 average.packages.pc3 += p->pc3;
990 if (do_pc6)
991 average.packages.pc6 += p->pc6;
992 if (do_pc7)
993 average.packages.pc7 += p->pc7;
994 average.packages.pc8 += p->pc8;
995 average.packages.pc9 += p->pc9;
996 average.packages.pc10 += p->pc10;
997
998 average.packages.energy_pkg += p->energy_pkg;
999 average.packages.energy_dram += p->energy_dram;
1000 average.packages.energy_cores += p->energy_cores;
1001 average.packages.energy_gfx += p->energy_gfx;
1002
1003 average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1004 average.packages.gfx_mhz = p->gfx_mhz;
1005
1006 average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1007
1008 average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1009 average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1010 return 0;
1011 }
1012 /*
1013 * sum the counters for all cpus in the system
1014 * compute the weighted average
1015 */
compute_average(struct thread_data * t,struct core_data * c,struct pkg_data * p)1016 void compute_average(struct thread_data *t, struct core_data *c,
1017 struct pkg_data *p)
1018 {
1019 clear_counters(&average.threads, &average.cores, &average.packages);
1020
1021 for_all_cpus(sum_counters, t, c, p);
1022
1023 average.threads.tsc /= topo.num_cpus;
1024 average.threads.aperf /= topo.num_cpus;
1025 average.threads.mperf /= topo.num_cpus;
1026 average.threads.c1 /= topo.num_cpus;
1027
1028 average.threads.extra_delta32 /= topo.num_cpus;
1029 average.threads.extra_delta32 &= 0xFFFFFFFF;
1030
1031 average.threads.extra_delta64 /= topo.num_cpus;
1032
1033 average.cores.c3 /= topo.num_cores;
1034 average.cores.c6 /= topo.num_cores;
1035 average.cores.c7 /= topo.num_cores;
1036
1037 if (do_skl_residency) {
1038 average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1039 average.packages.pkg_any_core_c0 /= topo.num_packages;
1040 average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1041 average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1042 }
1043
1044 average.packages.pc2 /= topo.num_packages;
1045 if (do_pc3)
1046 average.packages.pc3 /= topo.num_packages;
1047 if (do_pc6)
1048 average.packages.pc6 /= topo.num_packages;
1049 if (do_pc7)
1050 average.packages.pc7 /= topo.num_packages;
1051
1052 average.packages.pc8 /= topo.num_packages;
1053 average.packages.pc9 /= topo.num_packages;
1054 average.packages.pc10 /= topo.num_packages;
1055 }
1056
rdtsc(void)1057 static unsigned long long rdtsc(void)
1058 {
1059 unsigned int low, high;
1060
1061 asm volatile("rdtsc" : "=a" (low), "=d" (high));
1062
1063 return low | ((unsigned long long)high) << 32;
1064 }
1065
1066 /*
1067 * get_counters(...)
1068 * migrate to cpu
1069 * acquire and record local counters for that cpu
1070 */
get_counters(struct thread_data * t,struct core_data * c,struct pkg_data * p)1071 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1072 {
1073 int cpu = t->cpu_id;
1074 unsigned long long msr;
1075 int aperf_mperf_retry_count = 0;
1076
1077 if (cpu_migrate(cpu)) {
1078 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1079 return -1;
1080 }
1081
1082 retry:
1083 t->tsc = rdtsc(); /* we are running on local CPU of interest */
1084
1085 if (has_aperf) {
1086 unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1087
1088 /*
1089 * The TSC, APERF and MPERF must be read together for
1090 * APERF/MPERF and MPERF/TSC to give accurate results.
1091 *
1092 * Unfortunately, APERF and MPERF are read by
1093 * individual system call, so delays may occur
1094 * between them. If the time to read them
1095 * varies by a large amount, we re-read them.
1096 */
1097
1098 /*
1099 * This initial dummy APERF read has been seen to
1100 * reduce jitter in the subsequent reads.
1101 */
1102
1103 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1104 return -3;
1105
1106 t->tsc = rdtsc(); /* re-read close to APERF */
1107
1108 tsc_before = t->tsc;
1109
1110 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1111 return -3;
1112
1113 tsc_between = rdtsc();
1114
1115 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1116 return -4;
1117
1118 tsc_after = rdtsc();
1119
1120 aperf_time = tsc_between - tsc_before;
1121 mperf_time = tsc_after - tsc_between;
1122
1123 /*
1124 * If the system call latency to read APERF and MPERF
1125 * differ by more than 2x, then try again.
1126 */
1127 if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1128 aperf_mperf_retry_count++;
1129 if (aperf_mperf_retry_count < 5)
1130 goto retry;
1131 else
1132 warnx("cpu%d jitter %lld %lld",
1133 cpu, aperf_time, mperf_time);
1134 }
1135 aperf_mperf_retry_count = 0;
1136
1137 t->aperf = t->aperf * aperf_mperf_multiplier;
1138 t->mperf = t->mperf * aperf_mperf_multiplier;
1139 }
1140
1141 if (do_irq)
1142 t->irq_count = irqs_per_cpu[cpu];
1143 if (do_smi) {
1144 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1145 return -5;
1146 t->smi_count = msr & 0xFFFFFFFF;
1147 }
1148 if (extra_delta_offset32) {
1149 if (get_msr(cpu, extra_delta_offset32, &msr))
1150 return -5;
1151 t->extra_delta32 = msr & 0xFFFFFFFF;
1152 }
1153
1154 if (extra_delta_offset64)
1155 if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64))
1156 return -5;
1157
1158 if (extra_msr_offset32) {
1159 if (get_msr(cpu, extra_msr_offset32, &msr))
1160 return -5;
1161 t->extra_msr32 = msr & 0xFFFFFFFF;
1162 }
1163
1164 if (extra_msr_offset64)
1165 if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64))
1166 return -5;
1167
1168 if (use_c1_residency_msr) {
1169 if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1170 return -6;
1171 }
1172
1173 /* collect core counters only for 1st thread in core */
1174 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1175 return 0;
1176
1177 if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) {
1178 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1179 return -6;
1180 }
1181
1182 if (do_nhm_cstates && !do_knl_cstates) {
1183 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1184 return -7;
1185 } else if (do_knl_cstates) {
1186 if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1187 return -7;
1188 }
1189
1190 if (do_snb_cstates)
1191 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1192 return -8;
1193
1194 if (do_dts) {
1195 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1196 return -9;
1197 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1198 }
1199
1200
1201 /* collect package counters only for 1st core in package */
1202 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1203 return 0;
1204
1205 if (do_skl_residency) {
1206 if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1207 return -10;
1208 if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1209 return -11;
1210 if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1211 return -12;
1212 if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1213 return -13;
1214 }
1215 if (do_pc3)
1216 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1217 return -9;
1218 if (do_pc6)
1219 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1220 return -10;
1221 if (do_pc2)
1222 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1223 return -11;
1224 if (do_pc7)
1225 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1226 return -12;
1227 if (do_c8_c9_c10) {
1228 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1229 return -13;
1230 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1231 return -13;
1232 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1233 return -13;
1234 }
1235 if (do_rapl & RAPL_PKG) {
1236 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1237 return -13;
1238 p->energy_pkg = msr & 0xFFFFFFFF;
1239 }
1240 if (do_rapl & RAPL_CORES) {
1241 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1242 return -14;
1243 p->energy_cores = msr & 0xFFFFFFFF;
1244 }
1245 if (do_rapl & RAPL_DRAM) {
1246 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1247 return -15;
1248 p->energy_dram = msr & 0xFFFFFFFF;
1249 }
1250 if (do_rapl & RAPL_GFX) {
1251 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1252 return -16;
1253 p->energy_gfx = msr & 0xFFFFFFFF;
1254 }
1255 if (do_rapl & RAPL_PKG_PERF_STATUS) {
1256 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1257 return -16;
1258 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1259 }
1260 if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1261 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1262 return -16;
1263 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1264 }
1265 if (do_ptm) {
1266 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1267 return -17;
1268 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1269 }
1270
1271 if (do_gfx_rc6_ms)
1272 p->gfx_rc6_ms = gfx_cur_rc6_ms;
1273
1274 if (do_gfx_mhz)
1275 p->gfx_mhz = gfx_cur_mhz;
1276
1277 return 0;
1278 }
1279
1280 /*
1281 * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
1282 * If you change the values, note they are used both in comparisons
1283 * (>= PCL__7) and to index pkg_cstate_limit_strings[].
1284 */
1285
1286 #define PCLUKN 0 /* Unknown */
1287 #define PCLRSV 1 /* Reserved */
1288 #define PCL__0 2 /* PC0 */
1289 #define PCL__1 3 /* PC1 */
1290 #define PCL__2 4 /* PC2 */
1291 #define PCL__3 5 /* PC3 */
1292 #define PCL__4 6 /* PC4 */
1293 #define PCL__6 7 /* PC6 */
1294 #define PCL_6N 8 /* PC6 No Retention */
1295 #define PCL_6R 9 /* PC6 Retention */
1296 #define PCL__7 10 /* PC7 */
1297 #define PCL_7S 11 /* PC7 Shrink */
1298 #define PCL__8 12 /* PC8 */
1299 #define PCL__9 13 /* PC9 */
1300 #define PCLUNL 14 /* Unlimited */
1301
1302 int pkg_cstate_limit = PCLUKN;
1303 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
1304 "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"};
1305
1306 int nhm_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1307 int snb_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1308 int hsw_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1309 int slv_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1310 int amt_pkg_cstate_limits[16] = {PCL__0, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1311 int phi_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1312 int bxt_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1313
1314
1315 static void
calculate_tsc_tweak()1316 calculate_tsc_tweak()
1317 {
1318 tsc_tweak = base_hz / tsc_hz;
1319 }
1320
1321 static void
dump_nhm_platform_info(void)1322 dump_nhm_platform_info(void)
1323 {
1324 unsigned long long msr;
1325 unsigned int ratio;
1326
1327 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
1328
1329 fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
1330
1331 ratio = (msr >> 40) & 0xFF;
1332 fprintf(outf, "%d * %.0f = %.0f MHz max efficiency frequency\n",
1333 ratio, bclk, ratio * bclk);
1334
1335 ratio = (msr >> 8) & 0xFF;
1336 fprintf(outf, "%d * %.0f = %.0f MHz base frequency\n",
1337 ratio, bclk, ratio * bclk);
1338
1339 get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
1340 fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
1341 base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
1342
1343 return;
1344 }
1345
1346 static void
dump_hsw_turbo_ratio_limits(void)1347 dump_hsw_turbo_ratio_limits(void)
1348 {
1349 unsigned long long msr;
1350 unsigned int ratio;
1351
1352 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
1353
1354 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
1355
1356 ratio = (msr >> 8) & 0xFF;
1357 if (ratio)
1358 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 18 active cores\n",
1359 ratio, bclk, ratio * bclk);
1360
1361 ratio = (msr >> 0) & 0xFF;
1362 if (ratio)
1363 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 17 active cores\n",
1364 ratio, bclk, ratio * bclk);
1365 return;
1366 }
1367
1368 static void
dump_ivt_turbo_ratio_limits(void)1369 dump_ivt_turbo_ratio_limits(void)
1370 {
1371 unsigned long long msr;
1372 unsigned int ratio;
1373
1374 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
1375
1376 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
1377
1378 ratio = (msr >> 56) & 0xFF;
1379 if (ratio)
1380 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 16 active cores\n",
1381 ratio, bclk, ratio * bclk);
1382
1383 ratio = (msr >> 48) & 0xFF;
1384 if (ratio)
1385 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 15 active cores\n",
1386 ratio, bclk, ratio * bclk);
1387
1388 ratio = (msr >> 40) & 0xFF;
1389 if (ratio)
1390 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 14 active cores\n",
1391 ratio, bclk, ratio * bclk);
1392
1393 ratio = (msr >> 32) & 0xFF;
1394 if (ratio)
1395 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 13 active cores\n",
1396 ratio, bclk, ratio * bclk);
1397
1398 ratio = (msr >> 24) & 0xFF;
1399 if (ratio)
1400 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 12 active cores\n",
1401 ratio, bclk, ratio * bclk);
1402
1403 ratio = (msr >> 16) & 0xFF;
1404 if (ratio)
1405 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 11 active cores\n",
1406 ratio, bclk, ratio * bclk);
1407
1408 ratio = (msr >> 8) & 0xFF;
1409 if (ratio)
1410 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 10 active cores\n",
1411 ratio, bclk, ratio * bclk);
1412
1413 ratio = (msr >> 0) & 0xFF;
1414 if (ratio)
1415 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 9 active cores\n",
1416 ratio, bclk, ratio * bclk);
1417 return;
1418 }
1419
1420 static void
dump_nhm_turbo_ratio_limits(void)1421 dump_nhm_turbo_ratio_limits(void)
1422 {
1423 unsigned long long msr;
1424 unsigned int ratio;
1425
1426 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1427
1428 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
1429
1430 ratio = (msr >> 56) & 0xFF;
1431 if (ratio)
1432 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 8 active cores\n",
1433 ratio, bclk, ratio * bclk);
1434
1435 ratio = (msr >> 48) & 0xFF;
1436 if (ratio)
1437 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 7 active cores\n",
1438 ratio, bclk, ratio * bclk);
1439
1440 ratio = (msr >> 40) & 0xFF;
1441 if (ratio)
1442 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 6 active cores\n",
1443 ratio, bclk, ratio * bclk);
1444
1445 ratio = (msr >> 32) & 0xFF;
1446 if (ratio)
1447 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 5 active cores\n",
1448 ratio, bclk, ratio * bclk);
1449
1450 ratio = (msr >> 24) & 0xFF;
1451 if (ratio)
1452 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
1453 ratio, bclk, ratio * bclk);
1454
1455 ratio = (msr >> 16) & 0xFF;
1456 if (ratio)
1457 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
1458 ratio, bclk, ratio * bclk);
1459
1460 ratio = (msr >> 8) & 0xFF;
1461 if (ratio)
1462 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
1463 ratio, bclk, ratio * bclk);
1464
1465 ratio = (msr >> 0) & 0xFF;
1466 if (ratio)
1467 fprintf(outf, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
1468 ratio, bclk, ratio * bclk);
1469 return;
1470 }
1471
1472 static void
dump_knl_turbo_ratio_limits(void)1473 dump_knl_turbo_ratio_limits(void)
1474 {
1475 const unsigned int buckets_no = 7;
1476
1477 unsigned long long msr;
1478 int delta_cores, delta_ratio;
1479 int i, b_nr;
1480 unsigned int cores[buckets_no];
1481 unsigned int ratio[buckets_no];
1482
1483 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1484
1485 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
1486 base_cpu, msr);
1487
1488 /**
1489 * Turbo encoding in KNL is as follows:
1490 * [0] -- Reserved
1491 * [7:1] -- Base value of number of active cores of bucket 1.
1492 * [15:8] -- Base value of freq ratio of bucket 1.
1493 * [20:16] -- +ve delta of number of active cores of bucket 2.
1494 * i.e. active cores of bucket 2 =
1495 * active cores of bucket 1 + delta
1496 * [23:21] -- Negative delta of freq ratio of bucket 2.
1497 * i.e. freq ratio of bucket 2 =
1498 * freq ratio of bucket 1 - delta
1499 * [28:24]-- +ve delta of number of active cores of bucket 3.
1500 * [31:29]-- -ve delta of freq ratio of bucket 3.
1501 * [36:32]-- +ve delta of number of active cores of bucket 4.
1502 * [39:37]-- -ve delta of freq ratio of bucket 4.
1503 * [44:40]-- +ve delta of number of active cores of bucket 5.
1504 * [47:45]-- -ve delta of freq ratio of bucket 5.
1505 * [52:48]-- +ve delta of number of active cores of bucket 6.
1506 * [55:53]-- -ve delta of freq ratio of bucket 6.
1507 * [60:56]-- +ve delta of number of active cores of bucket 7.
1508 * [63:61]-- -ve delta of freq ratio of bucket 7.
1509 */
1510
1511 b_nr = 0;
1512 cores[b_nr] = (msr & 0xFF) >> 1;
1513 ratio[b_nr] = (msr >> 8) & 0xFF;
1514
1515 for (i = 16; i < 64; i += 8) {
1516 delta_cores = (msr >> i) & 0x1F;
1517 delta_ratio = (msr >> (i + 5)) & 0x7;
1518
1519 cores[b_nr + 1] = cores[b_nr] + delta_cores;
1520 ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
1521 b_nr++;
1522 }
1523
1524 for (i = buckets_no - 1; i >= 0; i--)
1525 if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
1526 fprintf(outf,
1527 "%d * %.0f = %.0f MHz max turbo %d active cores\n",
1528 ratio[i], bclk, ratio[i] * bclk, cores[i]);
1529 }
1530
1531 static void
dump_nhm_cst_cfg(void)1532 dump_nhm_cst_cfg(void)
1533 {
1534 unsigned long long msr;
1535
1536 get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
1537
1538 #define SNB_C1_AUTO_UNDEMOTE (1UL << 27)
1539 #define SNB_C3_AUTO_UNDEMOTE (1UL << 28)
1540
1541 fprintf(outf, "cpu%d: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", base_cpu, msr);
1542
1543 fprintf(outf, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n",
1544 (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
1545 (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
1546 (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
1547 (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
1548 (msr & (1 << 15)) ? "" : "UN",
1549 (unsigned int)msr & 0xF,
1550 pkg_cstate_limit_strings[pkg_cstate_limit]);
1551 return;
1552 }
1553
1554 static void
dump_config_tdp(void)1555 dump_config_tdp(void)
1556 {
1557 unsigned long long msr;
1558
1559 get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
1560 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
1561 fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
1562
1563 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
1564 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
1565 if (msr) {
1566 fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
1567 fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
1568 fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
1569 fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
1570 }
1571 fprintf(outf, ")\n");
1572
1573 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
1574 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
1575 if (msr) {
1576 fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
1577 fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
1578 fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
1579 fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
1580 }
1581 fprintf(outf, ")\n");
1582
1583 get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
1584 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
1585 if ((msr) & 0x3)
1586 fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
1587 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
1588 fprintf(outf, ")\n");
1589
1590 get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
1591 fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
1592 fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
1593 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
1594 fprintf(outf, ")\n");
1595 }
1596
1597 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
1598
print_irtl(void)1599 void print_irtl(void)
1600 {
1601 unsigned long long msr;
1602
1603 get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
1604 fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
1605 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1606 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1607
1608 get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
1609 fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
1610 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1611 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1612
1613 get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
1614 fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
1615 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1616 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1617
1618 if (!do_irtl_hsw)
1619 return;
1620
1621 get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
1622 fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
1623 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1624 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1625
1626 get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
1627 fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
1628 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1629 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1630
1631 get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
1632 fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
1633 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1634 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1635
1636 }
free_fd_percpu(void)1637 void free_fd_percpu(void)
1638 {
1639 int i;
1640
1641 for (i = 0; i < topo.max_cpu_num; ++i) {
1642 if (fd_percpu[i] != 0)
1643 close(fd_percpu[i]);
1644 }
1645
1646 free(fd_percpu);
1647 }
1648
free_all_buffers(void)1649 void free_all_buffers(void)
1650 {
1651 CPU_FREE(cpu_present_set);
1652 cpu_present_set = NULL;
1653 cpu_present_setsize = 0;
1654
1655 CPU_FREE(cpu_affinity_set);
1656 cpu_affinity_set = NULL;
1657 cpu_affinity_setsize = 0;
1658
1659 free(thread_even);
1660 free(core_even);
1661 free(package_even);
1662
1663 thread_even = NULL;
1664 core_even = NULL;
1665 package_even = NULL;
1666
1667 free(thread_odd);
1668 free(core_odd);
1669 free(package_odd);
1670
1671 thread_odd = NULL;
1672 core_odd = NULL;
1673 package_odd = NULL;
1674
1675 free(output_buffer);
1676 output_buffer = NULL;
1677 outp = NULL;
1678
1679 free_fd_percpu();
1680
1681 free(irq_column_2_cpu);
1682 free(irqs_per_cpu);
1683 }
1684
1685 /*
1686 * Open a file, and exit on failure
1687 */
fopen_or_die(const char * path,const char * mode)1688 FILE *fopen_or_die(const char *path, const char *mode)
1689 {
1690 FILE *filep = fopen(path, mode);
1691 if (!filep)
1692 err(1, "%s: open failed", path);
1693 return filep;
1694 }
1695
1696 /*
1697 * Parse a file containing a single int.
1698 */
parse_int_file(const char * fmt,...)1699 int parse_int_file(const char *fmt, ...)
1700 {
1701 va_list args;
1702 char path[PATH_MAX];
1703 FILE *filep;
1704 int value;
1705
1706 va_start(args, fmt);
1707 vsnprintf(path, sizeof(path), fmt, args);
1708 va_end(args);
1709 filep = fopen_or_die(path, "r");
1710 if (fscanf(filep, "%d", &value) != 1)
1711 err(1, "%s: failed to parse number from file", path);
1712 fclose(filep);
1713 return value;
1714 }
1715
1716 /*
1717 * get_cpu_position_in_core(cpu)
1718 * return the position of the CPU among its HT siblings in the core
1719 * return -1 if the sibling is not in list
1720 */
get_cpu_position_in_core(int cpu)1721 int get_cpu_position_in_core(int cpu)
1722 {
1723 char path[64];
1724 FILE *filep;
1725 int this_cpu;
1726 char character;
1727 int i;
1728
1729 sprintf(path,
1730 "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list",
1731 cpu);
1732 filep = fopen(path, "r");
1733 if (filep == NULL) {
1734 perror(path);
1735 exit(1);
1736 }
1737
1738 for (i = 0; i < topo.num_threads_per_core; i++) {
1739 fscanf(filep, "%d", &this_cpu);
1740 if (this_cpu == cpu) {
1741 fclose(filep);
1742 return i;
1743 }
1744
1745 /* Account for no separator after last thread*/
1746 if (i != (topo.num_threads_per_core - 1))
1747 fscanf(filep, "%c", &character);
1748 }
1749
1750 fclose(filep);
1751 return -1;
1752 }
1753
1754 /*
1755 * cpu_is_first_core_in_package(cpu)
1756 * return 1 if given CPU is 1st core in package
1757 */
cpu_is_first_core_in_package(int cpu)1758 int cpu_is_first_core_in_package(int cpu)
1759 {
1760 return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
1761 }
1762
get_physical_package_id(int cpu)1763 int get_physical_package_id(int cpu)
1764 {
1765 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
1766 }
1767
get_core_id(int cpu)1768 int get_core_id(int cpu)
1769 {
1770 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
1771 }
1772
get_num_ht_siblings(int cpu)1773 int get_num_ht_siblings(int cpu)
1774 {
1775 char path[80];
1776 FILE *filep;
1777 int sib1;
1778 int matches = 0;
1779 char character;
1780 char str[100];
1781 char *ch;
1782
1783 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1784 filep = fopen_or_die(path, "r");
1785
1786 /*
1787 * file format:
1788 * A ',' separated or '-' separated set of numbers
1789 * (eg 1-2 or 1,3,4,5)
1790 */
1791 fscanf(filep, "%d%c\n", &sib1, &character);
1792 fseek(filep, 0, SEEK_SET);
1793 fgets(str, 100, filep);
1794 ch = strchr(str, character);
1795 while (ch != NULL) {
1796 matches++;
1797 ch = strchr(ch+1, character);
1798 }
1799
1800 fclose(filep);
1801 return matches+1;
1802 }
1803
1804 /*
1805 * run func(thread, core, package) in topology order
1806 * skip non-present cpus
1807 */
1808
for_all_cpus_2(int (func)(struct thread_data *,struct core_data *,struct pkg_data *,struct thread_data *,struct core_data *,struct pkg_data *),struct thread_data * thread_base,struct core_data * core_base,struct pkg_data * pkg_base,struct thread_data * thread_base2,struct core_data * core_base2,struct pkg_data * pkg_base2)1809 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
1810 struct pkg_data *, struct thread_data *, struct core_data *,
1811 struct pkg_data *), struct thread_data *thread_base,
1812 struct core_data *core_base, struct pkg_data *pkg_base,
1813 struct thread_data *thread_base2, struct core_data *core_base2,
1814 struct pkg_data *pkg_base2)
1815 {
1816 int retval, pkg_no, core_no, thread_no;
1817
1818 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
1819 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
1820 for (thread_no = 0; thread_no <
1821 topo.num_threads_per_core; ++thread_no) {
1822 struct thread_data *t, *t2;
1823 struct core_data *c, *c2;
1824 struct pkg_data *p, *p2;
1825
1826 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
1827
1828 if (cpu_is_not_present(t->cpu_id))
1829 continue;
1830
1831 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
1832
1833 c = GET_CORE(core_base, core_no, pkg_no);
1834 c2 = GET_CORE(core_base2, core_no, pkg_no);
1835
1836 p = GET_PKG(pkg_base, pkg_no);
1837 p2 = GET_PKG(pkg_base2, pkg_no);
1838
1839 retval = func(t, c, p, t2, c2, p2);
1840 if (retval)
1841 return retval;
1842 }
1843 }
1844 }
1845 return 0;
1846 }
1847
1848 /*
1849 * run func(cpu) on every cpu in /proc/stat
1850 * return max_cpu number
1851 */
for_all_proc_cpus(int (func)(int))1852 int for_all_proc_cpus(int (func)(int))
1853 {
1854 FILE *fp;
1855 int cpu_num;
1856 int retval;
1857
1858 fp = fopen_or_die(proc_stat, "r");
1859
1860 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
1861 if (retval != 0)
1862 err(1, "%s: failed to parse format", proc_stat);
1863
1864 while (1) {
1865 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
1866 if (retval != 1)
1867 break;
1868
1869 retval = func(cpu_num);
1870 if (retval) {
1871 fclose(fp);
1872 return(retval);
1873 }
1874 }
1875 fclose(fp);
1876 return 0;
1877 }
1878
re_initialize(void)1879 void re_initialize(void)
1880 {
1881 free_all_buffers();
1882 setup_all_buffers();
1883 printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
1884 }
1885
1886
1887 /*
1888 * count_cpus()
1889 * remember the last one seen, it will be the max
1890 */
count_cpus(int cpu)1891 int count_cpus(int cpu)
1892 {
1893 if (topo.max_cpu_num < cpu)
1894 topo.max_cpu_num = cpu;
1895
1896 topo.num_cpus += 1;
1897 return 0;
1898 }
mark_cpu_present(int cpu)1899 int mark_cpu_present(int cpu)
1900 {
1901 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
1902 return 0;
1903 }
1904
1905 /*
1906 * snapshot_proc_interrupts()
1907 *
1908 * read and record summary of /proc/interrupts
1909 *
1910 * return 1 if config change requires a restart, else return 0
1911 */
snapshot_proc_interrupts(void)1912 int snapshot_proc_interrupts(void)
1913 {
1914 static FILE *fp;
1915 int column, retval;
1916
1917 if (fp == NULL)
1918 fp = fopen_or_die("/proc/interrupts", "r");
1919 else
1920 rewind(fp);
1921
1922 /* read 1st line of /proc/interrupts to get cpu* name for each column */
1923 for (column = 0; column < topo.num_cpus; ++column) {
1924 int cpu_number;
1925
1926 retval = fscanf(fp, " CPU%d", &cpu_number);
1927 if (retval != 1)
1928 break;
1929
1930 if (cpu_number > topo.max_cpu_num) {
1931 warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
1932 return 1;
1933 }
1934
1935 irq_column_2_cpu[column] = cpu_number;
1936 irqs_per_cpu[cpu_number] = 0;
1937 }
1938
1939 /* read /proc/interrupt count lines and sum up irqs per cpu */
1940 while (1) {
1941 int column;
1942 char buf[64];
1943
1944 retval = fscanf(fp, " %s:", buf); /* flush irq# "N:" */
1945 if (retval != 1)
1946 break;
1947
1948 /* read the count per cpu */
1949 for (column = 0; column < topo.num_cpus; ++column) {
1950
1951 int cpu_number, irq_count;
1952
1953 retval = fscanf(fp, " %d", &irq_count);
1954 if (retval != 1)
1955 break;
1956
1957 cpu_number = irq_column_2_cpu[column];
1958 irqs_per_cpu[cpu_number] += irq_count;
1959
1960 }
1961
1962 while (getc(fp) != '\n')
1963 ; /* flush interrupt description */
1964
1965 }
1966 return 0;
1967 }
1968 /*
1969 * snapshot_gfx_rc6_ms()
1970 *
1971 * record snapshot of
1972 * /sys/class/drm/card0/power/rc6_residency_ms
1973 *
1974 * return 1 if config change requires a restart, else return 0
1975 */
snapshot_gfx_rc6_ms(void)1976 int snapshot_gfx_rc6_ms(void)
1977 {
1978 FILE *fp;
1979 int retval;
1980
1981 fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
1982
1983 retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
1984 if (retval != 1)
1985 err(1, "GFX rc6");
1986
1987 fclose(fp);
1988
1989 return 0;
1990 }
1991 /*
1992 * snapshot_gfx_mhz()
1993 *
1994 * record snapshot of
1995 * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
1996 *
1997 * return 1 if config change requires a restart, else return 0
1998 */
snapshot_gfx_mhz(void)1999 int snapshot_gfx_mhz(void)
2000 {
2001 static FILE *fp;
2002 int retval;
2003
2004 if (fp == NULL)
2005 fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2006 else {
2007 rewind(fp);
2008 fflush(fp);
2009 }
2010
2011 retval = fscanf(fp, "%d", &gfx_cur_mhz);
2012 if (retval != 1)
2013 err(1, "GFX MHz");
2014
2015 return 0;
2016 }
2017
2018 /*
2019 * snapshot /proc and /sys files
2020 *
2021 * return 1 if configuration restart needed, else return 0
2022 */
snapshot_proc_sysfs_files(void)2023 int snapshot_proc_sysfs_files(void)
2024 {
2025 if (snapshot_proc_interrupts())
2026 return 1;
2027
2028 if (do_gfx_rc6_ms)
2029 snapshot_gfx_rc6_ms();
2030
2031 if (do_gfx_mhz)
2032 snapshot_gfx_mhz();
2033
2034 return 0;
2035 }
2036
turbostat_loop()2037 void turbostat_loop()
2038 {
2039 int retval;
2040 int restarted = 0;
2041
2042 restart:
2043 restarted++;
2044
2045 snapshot_proc_sysfs_files();
2046 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2047 if (retval < -1) {
2048 exit(retval);
2049 } else if (retval == -1) {
2050 if (restarted > 1) {
2051 exit(retval);
2052 }
2053 re_initialize();
2054 goto restart;
2055 }
2056 restarted = 0;
2057 gettimeofday(&tv_even, (struct timezone *)NULL);
2058
2059 while (1) {
2060 if (for_all_proc_cpus(cpu_is_not_present)) {
2061 re_initialize();
2062 goto restart;
2063 }
2064 nanosleep(&interval_ts, NULL);
2065 if (snapshot_proc_sysfs_files())
2066 goto restart;
2067 retval = for_all_cpus(get_counters, ODD_COUNTERS);
2068 if (retval < -1) {
2069 exit(retval);
2070 } else if (retval == -1) {
2071 re_initialize();
2072 goto restart;
2073 }
2074 gettimeofday(&tv_odd, (struct timezone *)NULL);
2075 timersub(&tv_odd, &tv_even, &tv_delta);
2076 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
2077 compute_average(EVEN_COUNTERS);
2078 format_all_counters(EVEN_COUNTERS);
2079 flush_output_stdout();
2080 nanosleep(&interval_ts, NULL);
2081 if (snapshot_proc_sysfs_files())
2082 goto restart;
2083 retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2084 if (retval < -1) {
2085 exit(retval);
2086 } else if (retval == -1) {
2087 re_initialize();
2088 goto restart;
2089 }
2090 gettimeofday(&tv_even, (struct timezone *)NULL);
2091 timersub(&tv_even, &tv_odd, &tv_delta);
2092 for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS);
2093 compute_average(ODD_COUNTERS);
2094 format_all_counters(ODD_COUNTERS);
2095 flush_output_stdout();
2096 }
2097 }
2098
check_dev_msr()2099 void check_dev_msr()
2100 {
2101 struct stat sb;
2102 char pathname[32];
2103
2104 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2105 if (stat(pathname, &sb))
2106 if (system("/sbin/modprobe msr > /dev/null 2>&1"))
2107 err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
2108 }
2109
check_permissions()2110 void check_permissions()
2111 {
2112 struct __user_cap_header_struct cap_header_data;
2113 cap_user_header_t cap_header = &cap_header_data;
2114 struct __user_cap_data_struct cap_data_data;
2115 cap_user_data_t cap_data = &cap_data_data;
2116 extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
2117 int do_exit = 0;
2118 char pathname[32];
2119
2120 /* check for CAP_SYS_RAWIO */
2121 cap_header->pid = getpid();
2122 cap_header->version = _LINUX_CAPABILITY_VERSION;
2123 if (capget(cap_header, cap_data) < 0)
2124 err(-6, "capget(2) failed");
2125
2126 if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
2127 do_exit++;
2128 warnx("capget(CAP_SYS_RAWIO) failed,"
2129 " try \"# setcap cap_sys_rawio=ep %s\"", progname);
2130 }
2131
2132 /* test file permissions */
2133 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2134 if (euidaccess(pathname, R_OK)) {
2135 do_exit++;
2136 warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
2137 }
2138
2139 /* if all else fails, thell them to be root */
2140 if (do_exit)
2141 if (getuid() != 0)
2142 warnx("... or simply run as root");
2143
2144 if (do_exit)
2145 exit(-6);
2146 }
2147
2148 /*
2149 * NHM adds support for additional MSRs:
2150 *
2151 * MSR_SMI_COUNT 0x00000034
2152 *
2153 * MSR_PLATFORM_INFO 0x000000ce
2154 * MSR_NHM_SNB_PKG_CST_CFG_CTL 0x000000e2
2155 *
2156 * MSR_PKG_C3_RESIDENCY 0x000003f8
2157 * MSR_PKG_C6_RESIDENCY 0x000003f9
2158 * MSR_CORE_C3_RESIDENCY 0x000003fc
2159 * MSR_CORE_C6_RESIDENCY 0x000003fd
2160 *
2161 * Side effect:
2162 * sets global pkg_cstate_limit to decode MSR_NHM_SNB_PKG_CST_CFG_CTL
2163 */
probe_nhm_msrs(unsigned int family,unsigned int model)2164 int probe_nhm_msrs(unsigned int family, unsigned int model)
2165 {
2166 unsigned long long msr;
2167 unsigned int base_ratio;
2168 int *pkg_cstate_limits;
2169
2170 if (!genuine_intel)
2171 return 0;
2172
2173 if (family != 6)
2174 return 0;
2175
2176 bclk = discover_bclk(family, model);
2177
2178 switch (model) {
2179 case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
2180 case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
2181 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
2182 case 0x25: /* Westmere Client - Clarkdale, Arrandale */
2183 case 0x2C: /* Westmere EP - Gulftown */
2184 case 0x2E: /* Nehalem-EX Xeon - Beckton */
2185 case 0x2F: /* Westmere-EX Xeon - Eagleton */
2186 pkg_cstate_limits = nhm_pkg_cstate_limits;
2187 break;
2188 case 0x2A: /* SNB */
2189 case 0x2D: /* SNB Xeon */
2190 case 0x3A: /* IVB */
2191 case 0x3E: /* IVB Xeon */
2192 pkg_cstate_limits = snb_pkg_cstate_limits;
2193 break;
2194 case 0x3C: /* HSW */
2195 case 0x3F: /* HSX */
2196 case 0x45: /* HSW */
2197 case 0x46: /* HSW */
2198 case 0x3D: /* BDW */
2199 case 0x47: /* BDW */
2200 case 0x4F: /* BDX */
2201 case 0x56: /* BDX-DE */
2202 case 0x4E: /* SKL */
2203 case 0x5E: /* SKL */
2204 case 0x8E: /* KBL */
2205 case 0x9E: /* KBL */
2206 case 0x55: /* SKX */
2207 pkg_cstate_limits = hsw_pkg_cstate_limits;
2208 break;
2209 case 0x37: /* BYT */
2210 case 0x4D: /* AVN */
2211 pkg_cstate_limits = slv_pkg_cstate_limits;
2212 break;
2213 case 0x4C: /* AMT */
2214 pkg_cstate_limits = amt_pkg_cstate_limits;
2215 break;
2216 case 0x57: /* PHI */
2217 pkg_cstate_limits = phi_pkg_cstate_limits;
2218 break;
2219 case 0x5C: /* BXT */
2220 pkg_cstate_limits = bxt_pkg_cstate_limits;
2221 break;
2222 default:
2223 return 0;
2224 }
2225 get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
2226 pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
2227
2228 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2229 base_ratio = (msr >> 8) & 0xFF;
2230
2231 base_hz = base_ratio * bclk * 1000000;
2232 has_base_hz = 1;
2233 return 1;
2234 }
has_nhm_turbo_ratio_limit(unsigned int family,unsigned int model)2235 int has_nhm_turbo_ratio_limit(unsigned int family, unsigned int model)
2236 {
2237 switch (model) {
2238 /* Nehalem compatible, but do not include turbo-ratio limit support */
2239 case 0x2E: /* Nehalem-EX Xeon - Beckton */
2240 case 0x2F: /* Westmere-EX Xeon - Eagleton */
2241 case 0x57: /* PHI - Knights Landing (different MSR definition) */
2242 return 0;
2243 default:
2244 return 1;
2245 }
2246 }
has_ivt_turbo_ratio_limit(unsigned int family,unsigned int model)2247 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
2248 {
2249 if (!genuine_intel)
2250 return 0;
2251
2252 if (family != 6)
2253 return 0;
2254
2255 switch (model) {
2256 case 0x3E: /* IVB Xeon */
2257 case 0x3F: /* HSW Xeon */
2258 return 1;
2259 default:
2260 return 0;
2261 }
2262 }
has_hsw_turbo_ratio_limit(unsigned int family,unsigned int model)2263 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
2264 {
2265 if (!genuine_intel)
2266 return 0;
2267
2268 if (family != 6)
2269 return 0;
2270
2271 switch (model) {
2272 case 0x3F: /* HSW Xeon */
2273 return 1;
2274 default:
2275 return 0;
2276 }
2277 }
2278
has_knl_turbo_ratio_limit(unsigned int family,unsigned int model)2279 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
2280 {
2281 if (!genuine_intel)
2282 return 0;
2283
2284 if (family != 6)
2285 return 0;
2286
2287 switch (model) {
2288 case 0x57: /* Knights Landing */
2289 return 1;
2290 default:
2291 return 0;
2292 }
2293 }
has_config_tdp(unsigned int family,unsigned int model)2294 int has_config_tdp(unsigned int family, unsigned int model)
2295 {
2296 if (!genuine_intel)
2297 return 0;
2298
2299 if (family != 6)
2300 return 0;
2301
2302 switch (model) {
2303 case 0x3A: /* IVB */
2304 case 0x3C: /* HSW */
2305 case 0x3F: /* HSX */
2306 case 0x45: /* HSW */
2307 case 0x46: /* HSW */
2308 case 0x3D: /* BDW */
2309 case 0x47: /* BDW */
2310 case 0x4F: /* BDX */
2311 case 0x56: /* BDX-DE */
2312 case 0x4E: /* SKL */
2313 case 0x5E: /* SKL */
2314 case 0x8E: /* KBL */
2315 case 0x9E: /* KBL */
2316 case 0x55: /* SKX */
2317
2318 case 0x57: /* Knights Landing */
2319 return 1;
2320 default:
2321 return 0;
2322 }
2323 }
2324
2325 static void
dump_cstate_pstate_config_info(unsigned int family,unsigned int model)2326 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
2327 {
2328 if (!do_nhm_platform_info)
2329 return;
2330
2331 dump_nhm_platform_info();
2332
2333 if (has_hsw_turbo_ratio_limit(family, model))
2334 dump_hsw_turbo_ratio_limits();
2335
2336 if (has_ivt_turbo_ratio_limit(family, model))
2337 dump_ivt_turbo_ratio_limits();
2338
2339 if (has_nhm_turbo_ratio_limit(family, model))
2340 dump_nhm_turbo_ratio_limits();
2341
2342 if (has_knl_turbo_ratio_limit(family, model))
2343 dump_knl_turbo_ratio_limits();
2344
2345 if (has_config_tdp(family, model))
2346 dump_config_tdp();
2347
2348 dump_nhm_cst_cfg();
2349 }
2350
2351
2352 /*
2353 * print_epb()
2354 * Decode the ENERGY_PERF_BIAS MSR
2355 */
print_epb(struct thread_data * t,struct core_data * c,struct pkg_data * p)2356 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2357 {
2358 unsigned long long msr;
2359 char *epb_string;
2360 int cpu;
2361
2362 if (!has_epb)
2363 return 0;
2364
2365 cpu = t->cpu_id;
2366
2367 /* EPB is per-package */
2368 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2369 return 0;
2370
2371 if (cpu_migrate(cpu)) {
2372 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2373 return -1;
2374 }
2375
2376 if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
2377 return 0;
2378
2379 switch (msr & 0xF) {
2380 case ENERGY_PERF_BIAS_PERFORMANCE:
2381 epb_string = "performance";
2382 break;
2383 case ENERGY_PERF_BIAS_NORMAL:
2384 epb_string = "balanced";
2385 break;
2386 case ENERGY_PERF_BIAS_POWERSAVE:
2387 epb_string = "powersave";
2388 break;
2389 default:
2390 epb_string = "custom";
2391 break;
2392 }
2393 fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
2394
2395 return 0;
2396 }
2397 /*
2398 * print_hwp()
2399 * Decode the MSR_HWP_CAPABILITIES
2400 */
print_hwp(struct thread_data * t,struct core_data * c,struct pkg_data * p)2401 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2402 {
2403 unsigned long long msr;
2404 int cpu;
2405
2406 if (!has_hwp)
2407 return 0;
2408
2409 cpu = t->cpu_id;
2410
2411 /* MSR_HWP_CAPABILITIES is per-package */
2412 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2413 return 0;
2414
2415 if (cpu_migrate(cpu)) {
2416 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2417 return -1;
2418 }
2419
2420 if (get_msr(cpu, MSR_PM_ENABLE, &msr))
2421 return 0;
2422
2423 fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
2424 cpu, msr, (msr & (1 << 0)) ? "" : "No-");
2425
2426 /* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
2427 if ((msr & (1 << 0)) == 0)
2428 return 0;
2429
2430 if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
2431 return 0;
2432
2433 fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
2434 "(high 0x%x guar 0x%x eff 0x%x low 0x%x)\n",
2435 cpu, msr,
2436 (unsigned int)HWP_HIGHEST_PERF(msr),
2437 (unsigned int)HWP_GUARANTEED_PERF(msr),
2438 (unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
2439 (unsigned int)HWP_LOWEST_PERF(msr));
2440
2441 if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
2442 return 0;
2443
2444 fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
2445 "(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x pkg 0x%x)\n",
2446 cpu, msr,
2447 (unsigned int)(((msr) >> 0) & 0xff),
2448 (unsigned int)(((msr) >> 8) & 0xff),
2449 (unsigned int)(((msr) >> 16) & 0xff),
2450 (unsigned int)(((msr) >> 24) & 0xff),
2451 (unsigned int)(((msr) >> 32) & 0xff3),
2452 (unsigned int)(((msr) >> 42) & 0x1));
2453
2454 if (has_hwp_pkg) {
2455 if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
2456 return 0;
2457
2458 fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
2459 "(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x)\n",
2460 cpu, msr,
2461 (unsigned int)(((msr) >> 0) & 0xff),
2462 (unsigned int)(((msr) >> 8) & 0xff),
2463 (unsigned int)(((msr) >> 16) & 0xff),
2464 (unsigned int)(((msr) >> 24) & 0xff),
2465 (unsigned int)(((msr) >> 32) & 0xff3));
2466 }
2467 if (has_hwp_notify) {
2468 if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
2469 return 0;
2470
2471 fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
2472 "(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
2473 cpu, msr,
2474 ((msr) & 0x1) ? "EN" : "Dis",
2475 ((msr) & 0x2) ? "EN" : "Dis");
2476 }
2477 if (get_msr(cpu, MSR_HWP_STATUS, &msr))
2478 return 0;
2479
2480 fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
2481 "(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
2482 cpu, msr,
2483 ((msr) & 0x1) ? "" : "No-",
2484 ((msr) & 0x2) ? "" : "No-");
2485
2486 return 0;
2487 }
2488
2489 /*
2490 * print_perf_limit()
2491 */
print_perf_limit(struct thread_data * t,struct core_data * c,struct pkg_data * p)2492 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2493 {
2494 unsigned long long msr;
2495 int cpu;
2496
2497 cpu = t->cpu_id;
2498
2499 /* per-package */
2500 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2501 return 0;
2502
2503 if (cpu_migrate(cpu)) {
2504 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2505 return -1;
2506 }
2507
2508 if (do_core_perf_limit_reasons) {
2509 get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
2510 fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2511 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
2512 (msr & 1 << 15) ? "bit15, " : "",
2513 (msr & 1 << 14) ? "bit14, " : "",
2514 (msr & 1 << 13) ? "Transitions, " : "",
2515 (msr & 1 << 12) ? "MultiCoreTurbo, " : "",
2516 (msr & 1 << 11) ? "PkgPwrL2, " : "",
2517 (msr & 1 << 10) ? "PkgPwrL1, " : "",
2518 (msr & 1 << 9) ? "CorePwr, " : "",
2519 (msr & 1 << 8) ? "Amps, " : "",
2520 (msr & 1 << 6) ? "VR-Therm, " : "",
2521 (msr & 1 << 5) ? "Auto-HWP, " : "",
2522 (msr & 1 << 4) ? "Graphics, " : "",
2523 (msr & 1 << 2) ? "bit2, " : "",
2524 (msr & 1 << 1) ? "ThermStatus, " : "",
2525 (msr & 1 << 0) ? "PROCHOT, " : "");
2526 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
2527 (msr & 1 << 31) ? "bit31, " : "",
2528 (msr & 1 << 30) ? "bit30, " : "",
2529 (msr & 1 << 29) ? "Transitions, " : "",
2530 (msr & 1 << 28) ? "MultiCoreTurbo, " : "",
2531 (msr & 1 << 27) ? "PkgPwrL2, " : "",
2532 (msr & 1 << 26) ? "PkgPwrL1, " : "",
2533 (msr & 1 << 25) ? "CorePwr, " : "",
2534 (msr & 1 << 24) ? "Amps, " : "",
2535 (msr & 1 << 22) ? "VR-Therm, " : "",
2536 (msr & 1 << 21) ? "Auto-HWP, " : "",
2537 (msr & 1 << 20) ? "Graphics, " : "",
2538 (msr & 1 << 18) ? "bit18, " : "",
2539 (msr & 1 << 17) ? "ThermStatus, " : "",
2540 (msr & 1 << 16) ? "PROCHOT, " : "");
2541
2542 }
2543 if (do_gfx_perf_limit_reasons) {
2544 get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
2545 fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2546 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
2547 (msr & 1 << 0) ? "PROCHOT, " : "",
2548 (msr & 1 << 1) ? "ThermStatus, " : "",
2549 (msr & 1 << 4) ? "Graphics, " : "",
2550 (msr & 1 << 6) ? "VR-Therm, " : "",
2551 (msr & 1 << 8) ? "Amps, " : "",
2552 (msr & 1 << 9) ? "GFXPwr, " : "",
2553 (msr & 1 << 10) ? "PkgPwrL1, " : "",
2554 (msr & 1 << 11) ? "PkgPwrL2, " : "");
2555 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
2556 (msr & 1 << 16) ? "PROCHOT, " : "",
2557 (msr & 1 << 17) ? "ThermStatus, " : "",
2558 (msr & 1 << 20) ? "Graphics, " : "",
2559 (msr & 1 << 22) ? "VR-Therm, " : "",
2560 (msr & 1 << 24) ? "Amps, " : "",
2561 (msr & 1 << 25) ? "GFXPwr, " : "",
2562 (msr & 1 << 26) ? "PkgPwrL1, " : "",
2563 (msr & 1 << 27) ? "PkgPwrL2, " : "");
2564 }
2565 if (do_ring_perf_limit_reasons) {
2566 get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
2567 fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2568 fprintf(outf, " (Active: %s%s%s%s%s%s)",
2569 (msr & 1 << 0) ? "PROCHOT, " : "",
2570 (msr & 1 << 1) ? "ThermStatus, " : "",
2571 (msr & 1 << 6) ? "VR-Therm, " : "",
2572 (msr & 1 << 8) ? "Amps, " : "",
2573 (msr & 1 << 10) ? "PkgPwrL1, " : "",
2574 (msr & 1 << 11) ? "PkgPwrL2, " : "");
2575 fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
2576 (msr & 1 << 16) ? "PROCHOT, " : "",
2577 (msr & 1 << 17) ? "ThermStatus, " : "",
2578 (msr & 1 << 22) ? "VR-Therm, " : "",
2579 (msr & 1 << 24) ? "Amps, " : "",
2580 (msr & 1 << 26) ? "PkgPwrL1, " : "",
2581 (msr & 1 << 27) ? "PkgPwrL2, " : "");
2582 }
2583 return 0;
2584 }
2585
2586 #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */
2587 #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */
2588
get_tdp(unsigned int model)2589 double get_tdp(unsigned int model)
2590 {
2591 unsigned long long msr;
2592
2593 if (do_rapl & RAPL_PKG_POWER_INFO)
2594 if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
2595 return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
2596
2597 switch (model) {
2598 case 0x37:
2599 case 0x4D:
2600 return 30.0;
2601 default:
2602 return 135.0;
2603 }
2604 }
2605
2606 /*
2607 * rapl_dram_energy_units_probe()
2608 * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
2609 */
2610 static double
rapl_dram_energy_units_probe(int model,double rapl_energy_units)2611 rapl_dram_energy_units_probe(int model, double rapl_energy_units)
2612 {
2613 /* only called for genuine_intel, family 6 */
2614
2615 switch (model) {
2616 case 0x3F: /* HSX */
2617 case 0x4F: /* BDX */
2618 case 0x56: /* BDX-DE */
2619 case 0x57: /* KNL */
2620 return (rapl_dram_energy_units = 15.3 / 1000000);
2621 default:
2622 return (rapl_energy_units);
2623 }
2624 }
2625
2626
2627 /*
2628 * rapl_probe()
2629 *
2630 * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
2631 */
rapl_probe(unsigned int family,unsigned int model)2632 void rapl_probe(unsigned int family, unsigned int model)
2633 {
2634 unsigned long long msr;
2635 unsigned int time_unit;
2636 double tdp;
2637
2638 if (!genuine_intel)
2639 return;
2640
2641 if (family != 6)
2642 return;
2643
2644 switch (model) {
2645 case 0x2A:
2646 case 0x3A:
2647 case 0x3C: /* HSW */
2648 case 0x45: /* HSW */
2649 case 0x46: /* HSW */
2650 case 0x3D: /* BDW */
2651 case 0x47: /* BDW */
2652 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
2653 break;
2654 case 0x5C: /* BXT */
2655 do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
2656 break;
2657 case 0x4E: /* SKL */
2658 case 0x5E: /* SKL */
2659 case 0x8E: /* KBL */
2660 case 0x9E: /* KBL */
2661 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
2662 break;
2663 case 0x3F: /* HSX */
2664 case 0x4F: /* BDX */
2665 case 0x56: /* BDX-DE */
2666 case 0x55: /* SKX */
2667 case 0x57: /* KNL */
2668 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
2669 break;
2670 case 0x2D:
2671 case 0x3E:
2672 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
2673 break;
2674 case 0x37: /* BYT */
2675 case 0x4D: /* AVN */
2676 do_rapl = RAPL_PKG | RAPL_CORES ;
2677 break;
2678 default:
2679 return;
2680 }
2681
2682 /* units on package 0, verify later other packages match */
2683 if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
2684 return;
2685
2686 rapl_power_units = 1.0 / (1 << (msr & 0xF));
2687 if (model == 0x37)
2688 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
2689 else
2690 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
2691
2692 rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
2693
2694 time_unit = msr >> 16 & 0xF;
2695 if (time_unit == 0)
2696 time_unit = 0xA;
2697
2698 rapl_time_units = 1.0 / (1 << (time_unit));
2699
2700 tdp = get_tdp(model);
2701
2702 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
2703 if (debug)
2704 fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
2705
2706 return;
2707 }
2708
perf_limit_reasons_probe(unsigned int family,unsigned int model)2709 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
2710 {
2711 if (!genuine_intel)
2712 return;
2713
2714 if (family != 6)
2715 return;
2716
2717 switch (model) {
2718 case 0x3C: /* HSW */
2719 case 0x45: /* HSW */
2720 case 0x46: /* HSW */
2721 do_gfx_perf_limit_reasons = 1;
2722 case 0x3F: /* HSX */
2723 do_core_perf_limit_reasons = 1;
2724 do_ring_perf_limit_reasons = 1;
2725 default:
2726 return;
2727 }
2728 }
2729
print_thermal(struct thread_data * t,struct core_data * c,struct pkg_data * p)2730 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2731 {
2732 unsigned long long msr;
2733 unsigned int dts;
2734 int cpu;
2735
2736 if (!(do_dts || do_ptm))
2737 return 0;
2738
2739 cpu = t->cpu_id;
2740
2741 /* DTS is per-core, no need to print for each thread */
2742 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
2743 return 0;
2744
2745 if (cpu_migrate(cpu)) {
2746 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2747 return -1;
2748 }
2749
2750 if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
2751 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
2752 return 0;
2753
2754 dts = (msr >> 16) & 0x7F;
2755 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
2756 cpu, msr, tcc_activation_temp - dts);
2757
2758 #ifdef THERM_DEBUG
2759 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
2760 return 0;
2761
2762 dts = (msr >> 16) & 0x7F;
2763 dts2 = (msr >> 8) & 0x7F;
2764 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
2765 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
2766 #endif
2767 }
2768
2769
2770 if (do_dts) {
2771 unsigned int resolution;
2772
2773 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
2774 return 0;
2775
2776 dts = (msr >> 16) & 0x7F;
2777 resolution = (msr >> 27) & 0xF;
2778 fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
2779 cpu, msr, tcc_activation_temp - dts, resolution);
2780
2781 #ifdef THERM_DEBUG
2782 if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
2783 return 0;
2784
2785 dts = (msr >> 16) & 0x7F;
2786 dts2 = (msr >> 8) & 0x7F;
2787 fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
2788 cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
2789 #endif
2790 }
2791
2792 return 0;
2793 }
2794
print_power_limit_msr(int cpu,unsigned long long msr,char * label)2795 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
2796 {
2797 fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
2798 cpu, label,
2799 ((msr >> 15) & 1) ? "EN" : "DIS",
2800 ((msr >> 0) & 0x7FFF) * rapl_power_units,
2801 (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
2802 (((msr >> 16) & 1) ? "EN" : "DIS"));
2803
2804 return;
2805 }
2806
print_rapl(struct thread_data * t,struct core_data * c,struct pkg_data * p)2807 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2808 {
2809 unsigned long long msr;
2810 int cpu;
2811
2812 if (!do_rapl)
2813 return 0;
2814
2815 /* RAPL counters are per package, so print only for 1st thread/package */
2816 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2817 return 0;
2818
2819 cpu = t->cpu_id;
2820 if (cpu_migrate(cpu)) {
2821 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2822 return -1;
2823 }
2824
2825 if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
2826 return -1;
2827
2828 if (debug) {
2829 fprintf(outf, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx "
2830 "(%f Watts, %f Joules, %f sec.)\n", cpu, msr,
2831 rapl_power_units, rapl_energy_units, rapl_time_units);
2832 }
2833 if (do_rapl & RAPL_PKG_POWER_INFO) {
2834
2835 if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
2836 return -5;
2837
2838
2839 fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
2840 cpu, msr,
2841 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2842 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2843 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2844 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
2845
2846 }
2847 if (do_rapl & RAPL_PKG) {
2848
2849 if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
2850 return -9;
2851
2852 fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
2853 cpu, msr, (msr >> 63) & 1 ? "": "UN");
2854
2855 print_power_limit_msr(cpu, msr, "PKG Limit #1");
2856 fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
2857 cpu,
2858 ((msr >> 47) & 1) ? "EN" : "DIS",
2859 ((msr >> 32) & 0x7FFF) * rapl_power_units,
2860 (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
2861 ((msr >> 48) & 1) ? "EN" : "DIS");
2862 }
2863
2864 if (do_rapl & RAPL_DRAM_POWER_INFO) {
2865 if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
2866 return -6;
2867
2868 fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
2869 cpu, msr,
2870 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2871 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2872 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2873 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
2874 }
2875 if (do_rapl & RAPL_DRAM) {
2876 if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
2877 return -9;
2878 fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
2879 cpu, msr, (msr >> 31) & 1 ? "": "UN");
2880
2881 print_power_limit_msr(cpu, msr, "DRAM Limit");
2882 }
2883 if (do_rapl & RAPL_CORE_POLICY) {
2884 if (debug) {
2885 if (get_msr(cpu, MSR_PP0_POLICY, &msr))
2886 return -7;
2887
2888 fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
2889 }
2890 }
2891 if (do_rapl & RAPL_CORES) {
2892 if (debug) {
2893
2894 if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
2895 return -9;
2896 fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
2897 cpu, msr, (msr >> 31) & 1 ? "": "UN");
2898 print_power_limit_msr(cpu, msr, "Cores Limit");
2899 }
2900 }
2901 if (do_rapl & RAPL_GFX) {
2902 if (debug) {
2903 if (get_msr(cpu, MSR_PP1_POLICY, &msr))
2904 return -8;
2905
2906 fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
2907
2908 if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
2909 return -9;
2910 fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
2911 cpu, msr, (msr >> 31) & 1 ? "": "UN");
2912 print_power_limit_msr(cpu, msr, "GFX Limit");
2913 }
2914 }
2915 return 0;
2916 }
2917
2918 /*
2919 * SNB adds support for additional MSRs:
2920 *
2921 * MSR_PKG_C7_RESIDENCY 0x000003fa
2922 * MSR_CORE_C7_RESIDENCY 0x000003fe
2923 * MSR_PKG_C2_RESIDENCY 0x0000060d
2924 */
2925
has_snb_msrs(unsigned int family,unsigned int model)2926 int has_snb_msrs(unsigned int family, unsigned int model)
2927 {
2928 if (!genuine_intel)
2929 return 0;
2930
2931 switch (model) {
2932 case 0x2A:
2933 case 0x2D:
2934 case 0x3A: /* IVB */
2935 case 0x3E: /* IVB Xeon */
2936 case 0x3C: /* HSW */
2937 case 0x3F: /* HSW */
2938 case 0x45: /* HSW */
2939 case 0x46: /* HSW */
2940 case 0x3D: /* BDW */
2941 case 0x47: /* BDW */
2942 case 0x4F: /* BDX */
2943 case 0x56: /* BDX-DE */
2944 case 0x4E: /* SKL */
2945 case 0x5E: /* SKL */
2946 case 0x8E: /* KBL */
2947 case 0x9E: /* KBL */
2948 case 0x55: /* SKX */
2949 case 0x5C: /* BXT */
2950 return 1;
2951 }
2952 return 0;
2953 }
2954
2955 /*
2956 * HSW adds support for additional MSRs:
2957 *
2958 * MSR_PKG_C8_RESIDENCY 0x00000630
2959 * MSR_PKG_C9_RESIDENCY 0x00000631
2960 * MSR_PKG_C10_RESIDENCY 0x00000632
2961 *
2962 * MSR_PKGC8_IRTL 0x00000633
2963 * MSR_PKGC9_IRTL 0x00000634
2964 * MSR_PKGC10_IRTL 0x00000635
2965 *
2966 */
has_hsw_msrs(unsigned int family,unsigned int model)2967 int has_hsw_msrs(unsigned int family, unsigned int model)
2968 {
2969 if (!genuine_intel)
2970 return 0;
2971
2972 switch (model) {
2973 case 0x45: /* HSW */
2974 case 0x3D: /* BDW */
2975 case 0x4E: /* SKL */
2976 case 0x5E: /* SKL */
2977 case 0x8E: /* KBL */
2978 case 0x9E: /* KBL */
2979 case 0x5C: /* BXT */
2980 return 1;
2981 }
2982 return 0;
2983 }
2984
2985 /*
2986 * SKL adds support for additional MSRS:
2987 *
2988 * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658
2989 * MSR_PKG_ANY_CORE_C0_RES 0x00000659
2990 * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A
2991 * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B
2992 */
has_skl_msrs(unsigned int family,unsigned int model)2993 int has_skl_msrs(unsigned int family, unsigned int model)
2994 {
2995 if (!genuine_intel)
2996 return 0;
2997
2998 switch (model) {
2999 case 0x4E: /* SKL */
3000 case 0x5E: /* SKL */
3001 case 0x8E: /* KBL */
3002 case 0x9E: /* KBL */
3003 return 1;
3004 }
3005 return 0;
3006 }
3007
3008
3009
is_slm(unsigned int family,unsigned int model)3010 int is_slm(unsigned int family, unsigned int model)
3011 {
3012 if (!genuine_intel)
3013 return 0;
3014 switch (model) {
3015 case 0x37: /* BYT */
3016 case 0x4D: /* AVN */
3017 return 1;
3018 }
3019 return 0;
3020 }
3021
is_knl(unsigned int family,unsigned int model)3022 int is_knl(unsigned int family, unsigned int model)
3023 {
3024 if (!genuine_intel)
3025 return 0;
3026 switch (model) {
3027 case 0x57: /* KNL */
3028 return 1;
3029 }
3030 return 0;
3031 }
3032
get_aperf_mperf_multiplier(unsigned int family,unsigned int model)3033 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
3034 {
3035 if (is_knl(family, model))
3036 return 1024;
3037 return 1;
3038 }
3039
3040 #define SLM_BCLK_FREQS 5
3041 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
3042
slm_bclk(void)3043 double slm_bclk(void)
3044 {
3045 unsigned long long msr = 3;
3046 unsigned int i;
3047 double freq;
3048
3049 if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
3050 fprintf(outf, "SLM BCLK: unknown\n");
3051
3052 i = msr & 0xf;
3053 if (i >= SLM_BCLK_FREQS) {
3054 fprintf(outf, "SLM BCLK[%d] invalid\n", i);
3055 msr = 3;
3056 }
3057 freq = slm_freq_table[i];
3058
3059 fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
3060
3061 return freq;
3062 }
3063
discover_bclk(unsigned int family,unsigned int model)3064 double discover_bclk(unsigned int family, unsigned int model)
3065 {
3066 if (has_snb_msrs(family, model) || is_knl(family, model))
3067 return 100.00;
3068 else if (is_slm(family, model))
3069 return slm_bclk();
3070 else
3071 return 133.33;
3072 }
3073
3074 /*
3075 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
3076 * the Thermal Control Circuit (TCC) activates.
3077 * This is usually equal to tjMax.
3078 *
3079 * Older processors do not have this MSR, so there we guess,
3080 * but also allow cmdline over-ride with -T.
3081 *
3082 * Several MSR temperature values are in units of degrees-C
3083 * below this value, including the Digital Thermal Sensor (DTS),
3084 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
3085 */
set_temperature_target(struct thread_data * t,struct core_data * c,struct pkg_data * p)3086 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3087 {
3088 unsigned long long msr;
3089 unsigned int target_c_local;
3090 int cpu;
3091
3092 /* tcc_activation_temp is used only for dts or ptm */
3093 if (!(do_dts || do_ptm))
3094 return 0;
3095
3096 /* this is a per-package concept */
3097 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3098 return 0;
3099
3100 cpu = t->cpu_id;
3101 if (cpu_migrate(cpu)) {
3102 fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3103 return -1;
3104 }
3105
3106 if (tcc_activation_temp_override != 0) {
3107 tcc_activation_temp = tcc_activation_temp_override;
3108 fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
3109 cpu, tcc_activation_temp);
3110 return 0;
3111 }
3112
3113 /* Temperature Target MSR is Nehalem and newer only */
3114 if (!do_nhm_platform_info)
3115 goto guess;
3116
3117 if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
3118 goto guess;
3119
3120 target_c_local = (msr >> 16) & 0xFF;
3121
3122 if (debug)
3123 fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
3124 cpu, msr, target_c_local);
3125
3126 if (!target_c_local)
3127 goto guess;
3128
3129 tcc_activation_temp = target_c_local;
3130
3131 return 0;
3132
3133 guess:
3134 tcc_activation_temp = TJMAX_DEFAULT;
3135 fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
3136 cpu, tcc_activation_temp);
3137
3138 return 0;
3139 }
3140
decode_feature_control_msr(void)3141 void decode_feature_control_msr(void)
3142 {
3143 unsigned long long msr;
3144
3145 if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
3146 fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
3147 base_cpu, msr,
3148 msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
3149 msr & (1 << 18) ? "SGX" : "");
3150 }
3151
decode_misc_enable_msr(void)3152 void decode_misc_enable_msr(void)
3153 {
3154 unsigned long long msr;
3155
3156 if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
3157 fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%s %s %s)\n",
3158 base_cpu, msr,
3159 msr & (1 << 3) ? "TCC" : "",
3160 msr & (1 << 16) ? "EIST" : "",
3161 msr & (1 << 18) ? "MONITOR" : "");
3162 }
3163
3164 /*
3165 * Decode MSR_MISC_PWR_MGMT
3166 *
3167 * Decode the bits according to the Nehalem documentation
3168 * bit[0] seems to continue to have same meaning going forward
3169 * bit[1] less so...
3170 */
decode_misc_pwr_mgmt_msr(void)3171 void decode_misc_pwr_mgmt_msr(void)
3172 {
3173 unsigned long long msr;
3174
3175 if (!do_nhm_platform_info)
3176 return;
3177
3178 if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
3179 fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB)\n",
3180 base_cpu, msr,
3181 msr & (1 << 0) ? "DIS" : "EN",
3182 msr & (1 << 1) ? "EN" : "DIS");
3183 }
3184
process_cpuid()3185 void process_cpuid()
3186 {
3187 unsigned int eax, ebx, ecx, edx, max_level, max_extended_level;
3188 unsigned int fms, family, model, stepping;
3189
3190 eax = ebx = ecx = edx = 0;
3191
3192 __cpuid(0, max_level, ebx, ecx, edx);
3193
3194 if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
3195 genuine_intel = 1;
3196
3197 if (debug)
3198 fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
3199 (char *)&ebx, (char *)&edx, (char *)&ecx);
3200
3201 __cpuid(1, fms, ebx, ecx, edx);
3202 family = (fms >> 8) & 0xf;
3203 model = (fms >> 4) & 0xf;
3204 stepping = fms & 0xf;
3205 if (family == 6 || family == 0xf)
3206 model += ((fms >> 16) & 0xf) << 4;
3207
3208 if (debug) {
3209 fprintf(outf, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
3210 max_level, family, model, stepping, family, model, stepping);
3211 fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s\n",
3212 ecx & (1 << 0) ? "SSE3" : "-",
3213 ecx & (1 << 3) ? "MONITOR" : "-",
3214 ecx & (1 << 6) ? "SMX" : "-",
3215 ecx & (1 << 7) ? "EIST" : "-",
3216 ecx & (1 << 8) ? "TM2" : "-",
3217 edx & (1 << 4) ? "TSC" : "-",
3218 edx & (1 << 5) ? "MSR" : "-",
3219 edx & (1 << 22) ? "ACPI-TM" : "-",
3220 edx & (1 << 29) ? "TM" : "-");
3221 }
3222
3223 if (!(edx & (1 << 5)))
3224 errx(1, "CPUID: no MSR");
3225
3226 /*
3227 * check max extended function levels of CPUID.
3228 * This is needed to check for invariant TSC.
3229 * This check is valid for both Intel and AMD.
3230 */
3231 ebx = ecx = edx = 0;
3232 __cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
3233
3234 if (max_extended_level >= 0x80000007) {
3235
3236 /*
3237 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
3238 * this check is valid for both Intel and AMD
3239 */
3240 __cpuid(0x80000007, eax, ebx, ecx, edx);
3241 has_invariant_tsc = edx & (1 << 8);
3242 }
3243
3244 /*
3245 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
3246 * this check is valid for both Intel and AMD
3247 */
3248
3249 __cpuid(0x6, eax, ebx, ecx, edx);
3250 has_aperf = ecx & (1 << 0);
3251 do_dts = eax & (1 << 0);
3252 do_ptm = eax & (1 << 6);
3253 has_hwp = eax & (1 << 7);
3254 has_hwp_notify = eax & (1 << 8);
3255 has_hwp_activity_window = eax & (1 << 9);
3256 has_hwp_epp = eax & (1 << 10);
3257 has_hwp_pkg = eax & (1 << 11);
3258 has_epb = ecx & (1 << 3);
3259
3260 if (debug)
3261 fprintf(outf, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sHWP, "
3262 "%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
3263 has_aperf ? "" : "No-",
3264 do_dts ? "" : "No-",
3265 do_ptm ? "" : "No-",
3266 has_hwp ? "" : "No-",
3267 has_hwp_notify ? "" : "No-",
3268 has_hwp_activity_window ? "" : "No-",
3269 has_hwp_epp ? "" : "No-",
3270 has_hwp_pkg ? "" : "No-",
3271 has_epb ? "" : "No-");
3272
3273 if (debug)
3274 decode_misc_enable_msr();
3275
3276 if (max_level >= 0x7 && debug) {
3277 int has_sgx;
3278
3279 ecx = 0;
3280
3281 __cpuid_count(0x7, 0, eax, ebx, ecx, edx);
3282
3283 has_sgx = ebx & (1 << 2);
3284 fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
3285
3286 if (has_sgx)
3287 decode_feature_control_msr();
3288 }
3289
3290 if (max_level >= 0x15) {
3291 unsigned int eax_crystal;
3292 unsigned int ebx_tsc;
3293
3294 /*
3295 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
3296 */
3297 eax_crystal = ebx_tsc = crystal_hz = edx = 0;
3298 __cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
3299
3300 if (ebx_tsc != 0) {
3301
3302 if (debug && (ebx != 0))
3303 fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
3304 eax_crystal, ebx_tsc, crystal_hz);
3305
3306 if (crystal_hz == 0)
3307 switch(model) {
3308 case 0x4E: /* SKL */
3309 case 0x5E: /* SKL */
3310 case 0x8E: /* KBL */
3311 case 0x9E: /* KBL */
3312 crystal_hz = 24000000; /* 24.0 MHz */
3313 break;
3314 case 0x55: /* SKX */
3315 crystal_hz = 25000000; /* 25.0 MHz */
3316 break;
3317 case 0x5C: /* BXT */
3318 crystal_hz = 19200000; /* 19.2 MHz */
3319 break;
3320 default:
3321 crystal_hz = 0;
3322 }
3323
3324 if (crystal_hz) {
3325 tsc_hz = (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
3326 if (debug)
3327 fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
3328 tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal);
3329 }
3330 }
3331 }
3332 if (max_level >= 0x16) {
3333 unsigned int base_mhz, max_mhz, bus_mhz, edx;
3334
3335 /*
3336 * CPUID 16H Base MHz, Max MHz, Bus MHz
3337 */
3338 base_mhz = max_mhz = bus_mhz = edx = 0;
3339
3340 __cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
3341 if (debug)
3342 fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
3343 base_mhz, max_mhz, bus_mhz);
3344 }
3345
3346 if (has_aperf)
3347 aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
3348
3349 do_nhm_platform_info = do_nhm_cstates = do_smi = probe_nhm_msrs(family, model);
3350 do_snb_cstates = has_snb_msrs(family, model);
3351 do_irtl_snb = has_snb_msrs(family, model);
3352 do_pc2 = do_snb_cstates && (pkg_cstate_limit >= PCL__2);
3353 do_pc3 = (pkg_cstate_limit >= PCL__3);
3354 do_pc6 = (pkg_cstate_limit >= PCL__6);
3355 do_pc7 = do_snb_cstates && (pkg_cstate_limit >= PCL__7);
3356 do_c8_c9_c10 = has_hsw_msrs(family, model);
3357 do_irtl_hsw = has_hsw_msrs(family, model);
3358 do_skl_residency = has_skl_msrs(family, model);
3359 do_slm_cstates = is_slm(family, model);
3360 do_knl_cstates = is_knl(family, model);
3361
3362 if (debug)
3363 decode_misc_pwr_mgmt_msr();
3364
3365 rapl_probe(family, model);
3366 perf_limit_reasons_probe(family, model);
3367
3368 if (debug)
3369 dump_cstate_pstate_config_info(family, model);
3370
3371 if (has_skl_msrs(family, model))
3372 calculate_tsc_tweak();
3373
3374 do_gfx_rc6_ms = !access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK);
3375
3376 do_gfx_mhz = !access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK);
3377
3378 return;
3379 }
3380
help()3381 void help()
3382 {
3383 fprintf(outf,
3384 "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
3385 "\n"
3386 "Turbostat forks the specified COMMAND and prints statistics\n"
3387 "when COMMAND completes.\n"
3388 "If no COMMAND is specified, turbostat wakes every 5-seconds\n"
3389 "to print statistics, until interrupted.\n"
3390 "--debug run in \"debug\" mode\n"
3391 "--interval sec Override default 5-second measurement interval\n"
3392 "--help print this help message\n"
3393 "--counter msr print 32-bit counter at address \"msr\"\n"
3394 "--Counter msr print 64-bit Counter at address \"msr\"\n"
3395 "--out file create or truncate \"file\" for all output\n"
3396 "--msr msr print 32-bit value at address \"msr\"\n"
3397 "--MSR msr print 64-bit Value at address \"msr\"\n"
3398 "--version print version information\n"
3399 "\n"
3400 "For more help, run \"man turbostat\"\n");
3401 }
3402
3403
3404 /*
3405 * in /dev/cpu/ return success for names that are numbers
3406 * ie. filter out ".", "..", "microcode".
3407 */
dir_filter(const struct dirent * dirp)3408 int dir_filter(const struct dirent *dirp)
3409 {
3410 if (isdigit(dirp->d_name[0]))
3411 return 1;
3412 else
3413 return 0;
3414 }
3415
open_dev_cpu_msr(int dummy1)3416 int open_dev_cpu_msr(int dummy1)
3417 {
3418 return 0;
3419 }
3420
topology_probe()3421 void topology_probe()
3422 {
3423 int i;
3424 int max_core_id = 0;
3425 int max_package_id = 0;
3426 int max_siblings = 0;
3427 struct cpu_topology {
3428 int core_id;
3429 int physical_package_id;
3430 } *cpus;
3431
3432 /* Initialize num_cpus, max_cpu_num */
3433 topo.num_cpus = 0;
3434 topo.max_cpu_num = 0;
3435 for_all_proc_cpus(count_cpus);
3436 if (!summary_only && topo.num_cpus > 1)
3437 show_cpu = 1;
3438
3439 if (debug > 1)
3440 fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
3441
3442 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
3443 if (cpus == NULL)
3444 err(1, "calloc cpus");
3445
3446 /*
3447 * Allocate and initialize cpu_present_set
3448 */
3449 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
3450 if (cpu_present_set == NULL)
3451 err(3, "CPU_ALLOC");
3452 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
3453 CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
3454 for_all_proc_cpus(mark_cpu_present);
3455
3456 /*
3457 * Allocate and initialize cpu_affinity_set
3458 */
3459 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
3460 if (cpu_affinity_set == NULL)
3461 err(3, "CPU_ALLOC");
3462 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
3463 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
3464
3465
3466 /*
3467 * For online cpus
3468 * find max_core_id, max_package_id
3469 */
3470 for (i = 0; i <= topo.max_cpu_num; ++i) {
3471 int siblings;
3472
3473 if (cpu_is_not_present(i)) {
3474 if (debug > 1)
3475 fprintf(outf, "cpu%d NOT PRESENT\n", i);
3476 continue;
3477 }
3478 cpus[i].core_id = get_core_id(i);
3479 if (cpus[i].core_id > max_core_id)
3480 max_core_id = cpus[i].core_id;
3481
3482 cpus[i].physical_package_id = get_physical_package_id(i);
3483 if (cpus[i].physical_package_id > max_package_id)
3484 max_package_id = cpus[i].physical_package_id;
3485
3486 siblings = get_num_ht_siblings(i);
3487 if (siblings > max_siblings)
3488 max_siblings = siblings;
3489 if (debug > 1)
3490 fprintf(outf, "cpu %d pkg %d core %d\n",
3491 i, cpus[i].physical_package_id, cpus[i].core_id);
3492 }
3493 topo.num_cores_per_pkg = max_core_id + 1;
3494 if (debug > 1)
3495 fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
3496 max_core_id, topo.num_cores_per_pkg);
3497 if (debug && !summary_only && topo.num_cores_per_pkg > 1)
3498 show_core = 1;
3499
3500 topo.num_packages = max_package_id + 1;
3501 if (debug > 1)
3502 fprintf(outf, "max_package_id %d, sizing for %d packages\n",
3503 max_package_id, topo.num_packages);
3504 if (debug && !summary_only && topo.num_packages > 1)
3505 show_pkg = 1;
3506
3507 topo.num_threads_per_core = max_siblings;
3508 if (debug > 1)
3509 fprintf(outf, "max_siblings %d\n", max_siblings);
3510
3511 free(cpus);
3512 }
3513
3514 void
allocate_counters(struct thread_data ** t,struct core_data ** c,struct pkg_data ** p)3515 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
3516 {
3517 int i;
3518
3519 *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
3520 topo.num_packages, sizeof(struct thread_data));
3521 if (*t == NULL)
3522 goto error;
3523
3524 for (i = 0; i < topo.num_threads_per_core *
3525 topo.num_cores_per_pkg * topo.num_packages; i++)
3526 (*t)[i].cpu_id = -1;
3527
3528 *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
3529 sizeof(struct core_data));
3530 if (*c == NULL)
3531 goto error;
3532
3533 for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
3534 (*c)[i].core_id = -1;
3535
3536 *p = calloc(topo.num_packages, sizeof(struct pkg_data));
3537 if (*p == NULL)
3538 goto error;
3539
3540 for (i = 0; i < topo.num_packages; i++)
3541 (*p)[i].package_id = i;
3542
3543 return;
3544 error:
3545 err(1, "calloc counters");
3546 }
3547 /*
3548 * init_counter()
3549 *
3550 * set cpu_id, core_num, pkg_num
3551 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
3552 *
3553 * increment topo.num_cores when 1st core in pkg seen
3554 */
init_counter(struct thread_data * thread_base,struct core_data * core_base,struct pkg_data * pkg_base,int thread_num,int core_num,int pkg_num,int cpu_id)3555 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
3556 struct pkg_data *pkg_base, int thread_num, int core_num,
3557 int pkg_num, int cpu_id)
3558 {
3559 struct thread_data *t;
3560 struct core_data *c;
3561 struct pkg_data *p;
3562
3563 t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
3564 c = GET_CORE(core_base, core_num, pkg_num);
3565 p = GET_PKG(pkg_base, pkg_num);
3566
3567 t->cpu_id = cpu_id;
3568 if (thread_num == 0) {
3569 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
3570 if (cpu_is_first_core_in_package(cpu_id))
3571 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
3572 }
3573
3574 c->core_id = core_num;
3575 p->package_id = pkg_num;
3576 }
3577
3578
initialize_counters(int cpu_id)3579 int initialize_counters(int cpu_id)
3580 {
3581 int my_thread_id, my_core_id, my_package_id;
3582
3583 my_package_id = get_physical_package_id(cpu_id);
3584 my_core_id = get_core_id(cpu_id);
3585 my_thread_id = get_cpu_position_in_core(cpu_id);
3586 if (!my_thread_id)
3587 topo.num_cores++;
3588
3589 init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
3590 init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
3591 return 0;
3592 }
3593
allocate_output_buffer()3594 void allocate_output_buffer()
3595 {
3596 output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
3597 outp = output_buffer;
3598 if (outp == NULL)
3599 err(-1, "calloc output buffer");
3600 }
allocate_fd_percpu(void)3601 void allocate_fd_percpu(void)
3602 {
3603 fd_percpu = calloc(topo.max_cpu_num, sizeof(int));
3604 if (fd_percpu == NULL)
3605 err(-1, "calloc fd_percpu");
3606 }
allocate_irq_buffers(void)3607 void allocate_irq_buffers(void)
3608 {
3609 irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
3610 if (irq_column_2_cpu == NULL)
3611 err(-1, "calloc %d", topo.num_cpus);
3612
3613 irqs_per_cpu = calloc(topo.max_cpu_num, sizeof(int));
3614 if (irqs_per_cpu == NULL)
3615 err(-1, "calloc %d", topo.max_cpu_num);
3616 }
setup_all_buffers(void)3617 void setup_all_buffers(void)
3618 {
3619 topology_probe();
3620 allocate_irq_buffers();
3621 allocate_fd_percpu();
3622 allocate_counters(&thread_even, &core_even, &package_even);
3623 allocate_counters(&thread_odd, &core_odd, &package_odd);
3624 allocate_output_buffer();
3625 for_all_proc_cpus(initialize_counters);
3626 }
3627
set_base_cpu(void)3628 void set_base_cpu(void)
3629 {
3630 base_cpu = sched_getcpu();
3631 if (base_cpu < 0)
3632 err(-ENODEV, "No valid cpus found");
3633
3634 if (debug > 1)
3635 fprintf(outf, "base_cpu = %d\n", base_cpu);
3636 }
3637
turbostat_init()3638 void turbostat_init()
3639 {
3640 setup_all_buffers();
3641 set_base_cpu();
3642 check_dev_msr();
3643 check_permissions();
3644 process_cpuid();
3645
3646
3647 if (debug)
3648 for_all_cpus(print_hwp, ODD_COUNTERS);
3649
3650 if (debug)
3651 for_all_cpus(print_epb, ODD_COUNTERS);
3652
3653 if (debug)
3654 for_all_cpus(print_perf_limit, ODD_COUNTERS);
3655
3656 if (debug)
3657 for_all_cpus(print_rapl, ODD_COUNTERS);
3658
3659 for_all_cpus(set_temperature_target, ODD_COUNTERS);
3660
3661 if (debug)
3662 for_all_cpus(print_thermal, ODD_COUNTERS);
3663
3664 if (debug && do_irtl_snb)
3665 print_irtl();
3666 }
3667
fork_it(char ** argv)3668 int fork_it(char **argv)
3669 {
3670 pid_t child_pid;
3671 int status;
3672
3673 status = for_all_cpus(get_counters, EVEN_COUNTERS);
3674 if (status)
3675 exit(status);
3676 /* clear affinity side-effect of get_counters() */
3677 sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
3678 gettimeofday(&tv_even, (struct timezone *)NULL);
3679
3680 child_pid = fork();
3681 if (!child_pid) {
3682 /* child */
3683 execvp(argv[0], argv);
3684 } else {
3685
3686 /* parent */
3687 if (child_pid == -1)
3688 err(1, "fork");
3689
3690 signal(SIGINT, SIG_IGN);
3691 signal(SIGQUIT, SIG_IGN);
3692 if (waitpid(child_pid, &status, 0) == -1)
3693 err(status, "waitpid");
3694 }
3695 /*
3696 * n.b. fork_it() does not check for errors from for_all_cpus()
3697 * because re-starting is problematic when forking
3698 */
3699 for_all_cpus(get_counters, ODD_COUNTERS);
3700 gettimeofday(&tv_odd, (struct timezone *)NULL);
3701 timersub(&tv_odd, &tv_even, &tv_delta);
3702 for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
3703 compute_average(EVEN_COUNTERS);
3704 format_all_counters(EVEN_COUNTERS);
3705
3706 fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
3707
3708 flush_output_stderr();
3709
3710 return status;
3711 }
3712
get_and_dump_counters(void)3713 int get_and_dump_counters(void)
3714 {
3715 int status;
3716
3717 status = for_all_cpus(get_counters, ODD_COUNTERS);
3718 if (status)
3719 return status;
3720
3721 status = for_all_cpus(dump_counters, ODD_COUNTERS);
3722 if (status)
3723 return status;
3724
3725 flush_output_stdout();
3726
3727 return status;
3728 }
3729
print_version()3730 void print_version() {
3731 fprintf(outf, "turbostat version 4.12 5 Apr 2016"
3732 " - Len Brown <lenb@kernel.org>\n");
3733 }
3734
cmdline(int argc,char ** argv)3735 void cmdline(int argc, char **argv)
3736 {
3737 int opt;
3738 int option_index = 0;
3739 static struct option long_options[] = {
3740 {"Counter", required_argument, 0, 'C'},
3741 {"counter", required_argument, 0, 'c'},
3742 {"Dump", no_argument, 0, 'D'},
3743 {"debug", no_argument, 0, 'd'},
3744 {"interval", required_argument, 0, 'i'},
3745 {"help", no_argument, 0, 'h'},
3746 {"Joules", no_argument, 0, 'J'},
3747 {"MSR", required_argument, 0, 'M'},
3748 {"msr", required_argument, 0, 'm'},
3749 {"out", required_argument, 0, 'o'},
3750 {"Package", no_argument, 0, 'p'},
3751 {"processor", no_argument, 0, 'p'},
3752 {"Summary", no_argument, 0, 'S'},
3753 {"TCC", required_argument, 0, 'T'},
3754 {"version", no_argument, 0, 'v' },
3755 {0, 0, 0, 0 }
3756 };
3757
3758 progname = argv[0];
3759
3760 while ((opt = getopt_long_only(argc, argv, "+C:c:Ddhi:JM:m:o:PpST:v",
3761 long_options, &option_index)) != -1) {
3762 switch (opt) {
3763 case 'C':
3764 sscanf(optarg, "%x", &extra_delta_offset64);
3765 break;
3766 case 'c':
3767 sscanf(optarg, "%x", &extra_delta_offset32);
3768 break;
3769 case 'D':
3770 dump_only++;
3771 break;
3772 case 'd':
3773 debug++;
3774 break;
3775 case 'h':
3776 default:
3777 help();
3778 exit(1);
3779 case 'i':
3780 {
3781 double interval = strtod(optarg, NULL);
3782
3783 if (interval < 0.001) {
3784 fprintf(outf, "interval %f seconds is too small\n",
3785 interval);
3786 exit(2);
3787 }
3788
3789 interval_ts.tv_sec = interval;
3790 interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
3791 }
3792 break;
3793 case 'J':
3794 rapl_joules++;
3795 break;
3796 case 'M':
3797 sscanf(optarg, "%x", &extra_msr_offset64);
3798 break;
3799 case 'm':
3800 sscanf(optarg, "%x", &extra_msr_offset32);
3801 break;
3802 case 'o':
3803 outf = fopen_or_die(optarg, "w");
3804 break;
3805 case 'P':
3806 show_pkg_only++;
3807 break;
3808 case 'p':
3809 show_core_only++;
3810 break;
3811 case 'S':
3812 summary_only++;
3813 break;
3814 case 'T':
3815 tcc_activation_temp_override = atoi(optarg);
3816 break;
3817 case 'v':
3818 print_version();
3819 exit(0);
3820 break;
3821 }
3822 }
3823 }
3824
main(int argc,char ** argv)3825 int main(int argc, char **argv)
3826 {
3827 outf = stderr;
3828
3829 cmdline(argc, argv);
3830
3831 if (debug)
3832 print_version();
3833
3834 turbostat_init();
3835
3836 /* dump counters and exit */
3837 if (dump_only)
3838 return get_and_dump_counters();
3839
3840 /*
3841 * if any params left, it must be a command to fork
3842 */
3843 if (argc - optind)
3844 return fork_it(argv + optind);
3845 else
3846 turbostat_loop();
3847
3848 return 0;
3849 }
3850