]>
Commit | Line | Data |
---|---|---|
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 <cpuid.h> | |
42 | #include <linux/capability.h> | |
43 | #include <errno.h> | |
44 | ||
45 | char *proc_stat = "/proc/stat"; | |
46 | unsigned int interval_sec = 5; | |
47 | unsigned int debug; | |
48 | unsigned int rapl_joules; | |
49 | unsigned int summary_only; | |
50 | unsigned int dump_only; | |
51 | unsigned int skip_c0; | |
52 | unsigned int skip_c1; | |
53 | unsigned int do_nhm_cstates; | |
54 | unsigned int do_snb_cstates; | |
55 | unsigned int do_knl_cstates; | |
56 | unsigned int do_pc2; | |
57 | unsigned int do_pc3; | |
58 | unsigned int do_pc6; | |
59 | unsigned int do_pc7; | |
60 | unsigned int do_c8_c9_c10; | |
61 | unsigned int do_skl_residency; | |
62 | unsigned int do_slm_cstates; | |
63 | unsigned int use_c1_residency_msr; | |
64 | unsigned int has_aperf; | |
65 | unsigned int has_epb; | |
66 | unsigned int units = 1000000; /* MHz etc */ | |
67 | unsigned int genuine_intel; | |
68 | unsigned int has_invariant_tsc; | |
69 | unsigned int do_nhm_platform_info; | |
70 | unsigned int extra_msr_offset32; | |
71 | unsigned int extra_msr_offset64; | |
72 | unsigned int extra_delta_offset32; | |
73 | unsigned int extra_delta_offset64; | |
74 | unsigned int aperf_mperf_multiplier = 1; | |
75 | int do_smi; | |
76 | double bclk; | |
77 | double base_hz; | |
78 | unsigned int has_base_hz; | |
79 | double tsc_tweak = 1.0; | |
80 | unsigned int show_pkg; | |
81 | unsigned int show_core; | |
82 | unsigned int show_cpu; | |
83 | unsigned int show_pkg_only; | |
84 | unsigned int show_core_only; | |
85 | char *output_buffer, *outp; | |
86 | unsigned int do_rapl; | |
87 | unsigned int do_dts; | |
88 | unsigned int do_ptm; | |
89 | unsigned int tcc_activation_temp; | |
90 | unsigned int tcc_activation_temp_override; | |
91 | double rapl_power_units, rapl_time_units; | |
92 | double rapl_dram_energy_units, rapl_energy_units; | |
93 | double rapl_joule_counter_range; | |
94 | unsigned int do_core_perf_limit_reasons; | |
95 | unsigned int do_gfx_perf_limit_reasons; | |
96 | unsigned int do_ring_perf_limit_reasons; | |
97 | unsigned int crystal_hz; | |
98 | unsigned long long tsc_hz; | |
99 | int base_cpu; | |
100 | double discover_bclk(unsigned int family, unsigned int model); | |
101 | ||
102 | #define RAPL_PKG (1 << 0) | |
103 | /* 0x610 MSR_PKG_POWER_LIMIT */ | |
104 | /* 0x611 MSR_PKG_ENERGY_STATUS */ | |
105 | #define RAPL_PKG_PERF_STATUS (1 << 1) | |
106 | /* 0x613 MSR_PKG_PERF_STATUS */ | |
107 | #define RAPL_PKG_POWER_INFO (1 << 2) | |
108 | /* 0x614 MSR_PKG_POWER_INFO */ | |
109 | ||
110 | #define RAPL_DRAM (1 << 3) | |
111 | /* 0x618 MSR_DRAM_POWER_LIMIT */ | |
112 | /* 0x619 MSR_DRAM_ENERGY_STATUS */ | |
113 | #define RAPL_DRAM_PERF_STATUS (1 << 4) | |
114 | /* 0x61b MSR_DRAM_PERF_STATUS */ | |
115 | #define RAPL_DRAM_POWER_INFO (1 << 5) | |
116 | /* 0x61c MSR_DRAM_POWER_INFO */ | |
117 | ||
118 | #define RAPL_CORES (1 << 6) | |
119 | /* 0x638 MSR_PP0_POWER_LIMIT */ | |
120 | /* 0x639 MSR_PP0_ENERGY_STATUS */ | |
121 | #define RAPL_CORE_POLICY (1 << 7) | |
122 | /* 0x63a MSR_PP0_POLICY */ | |
123 | ||
124 | #define RAPL_GFX (1 << 8) | |
125 | /* 0x640 MSR_PP1_POWER_LIMIT */ | |
126 | /* 0x641 MSR_PP1_ENERGY_STATUS */ | |
127 | /* 0x642 MSR_PP1_POLICY */ | |
128 | #define TJMAX_DEFAULT 100 | |
129 | ||
130 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
131 | ||
132 | int aperf_mperf_unstable; | |
133 | int backwards_count; | |
134 | char *progname; | |
135 | ||
136 | cpu_set_t *cpu_present_set, *cpu_affinity_set; | |
137 | size_t cpu_present_setsize, cpu_affinity_setsize; | |
138 | ||
139 | struct thread_data { | |
140 | unsigned long long tsc; | |
141 | unsigned long long aperf; | |
142 | unsigned long long mperf; | |
143 | unsigned long long c1; | |
144 | unsigned long long extra_msr64; | |
145 | unsigned long long extra_delta64; | |
146 | unsigned long long extra_msr32; | |
147 | unsigned long long extra_delta32; | |
148 | unsigned int smi_count; | |
149 | unsigned int cpu_id; | |
150 | unsigned int flags; | |
151 | #define CPU_IS_FIRST_THREAD_IN_CORE 0x2 | |
152 | #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 | |
153 | } *thread_even, *thread_odd; | |
154 | ||
155 | struct core_data { | |
156 | unsigned long long c3; | |
157 | unsigned long long c6; | |
158 | unsigned long long c7; | |
159 | unsigned int core_temp_c; | |
160 | unsigned int core_id; | |
161 | } *core_even, *core_odd; | |
162 | ||
163 | struct pkg_data { | |
164 | unsigned long long pc2; | |
165 | unsigned long long pc3; | |
166 | unsigned long long pc6; | |
167 | unsigned long long pc7; | |
168 | unsigned long long pc8; | |
169 | unsigned long long pc9; | |
170 | unsigned long long pc10; | |
171 | unsigned long long pkg_wtd_core_c0; | |
172 | unsigned long long pkg_any_core_c0; | |
173 | unsigned long long pkg_any_gfxe_c0; | |
174 | unsigned long long pkg_both_core_gfxe_c0; | |
175 | unsigned int package_id; | |
176 | unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */ | |
177 | unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */ | |
178 | unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */ | |
179 | unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */ | |
180 | unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ | |
181 | unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ | |
182 | unsigned int pkg_temp_c; | |
183 | ||
184 | } *package_even, *package_odd; | |
185 | ||
186 | #define ODD_COUNTERS thread_odd, core_odd, package_odd | |
187 | #define EVEN_COUNTERS thread_even, core_even, package_even | |
188 | ||
189 | #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \ | |
190 | (thread_base + (pkg_no) * topo.num_cores_per_pkg * \ | |
191 | topo.num_threads_per_core + \ | |
192 | (core_no) * topo.num_threads_per_core + (thread_no)) | |
193 | #define GET_CORE(core_base, core_no, pkg_no) \ | |
194 | (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no)) | |
195 | #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) | |
196 | ||
197 | struct system_summary { | |
198 | struct thread_data threads; | |
199 | struct core_data cores; | |
200 | struct pkg_data packages; | |
201 | } sum, average; | |
202 | ||
203 | ||
204 | struct topo_params { | |
205 | int num_packages; | |
206 | int num_cpus; | |
207 | int num_cores; | |
208 | int max_cpu_num; | |
209 | int num_cores_per_pkg; | |
210 | int num_threads_per_core; | |
211 | } topo; | |
212 | ||
213 | struct timeval tv_even, tv_odd, tv_delta; | |
214 | ||
215 | void setup_all_buffers(void); | |
216 | ||
217 | int cpu_is_not_present(int cpu) | |
218 | { | |
219 | return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); | |
220 | } | |
221 | /* | |
222 | * run func(thread, core, package) in topology order | |
223 | * skip non-present cpus | |
224 | */ | |
225 | ||
226 | int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *), | |
227 | struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base) | |
228 | { | |
229 | int retval, pkg_no, core_no, thread_no; | |
230 | ||
231 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { | |
232 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
233 | for (thread_no = 0; thread_no < | |
234 | topo.num_threads_per_core; ++thread_no) { | |
235 | struct thread_data *t; | |
236 | struct core_data *c; | |
237 | struct pkg_data *p; | |
238 | ||
239 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); | |
240 | ||
241 | if (cpu_is_not_present(t->cpu_id)) | |
242 | continue; | |
243 | ||
244 | c = GET_CORE(core_base, core_no, pkg_no); | |
245 | p = GET_PKG(pkg_base, pkg_no); | |
246 | ||
247 | retval = func(t, c, p); | |
248 | if (retval) | |
249 | return retval; | |
250 | } | |
251 | } | |
252 | } | |
253 | return 0; | |
254 | } | |
255 | ||
256 | int cpu_migrate(int cpu) | |
257 | { | |
258 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); | |
259 | CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set); | |
260 | if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) | |
261 | return -1; | |
262 | else | |
263 | return 0; | |
264 | } | |
265 | ||
266 | int get_msr(int cpu, off_t offset, unsigned long long *msr) | |
267 | { | |
268 | ssize_t retval; | |
269 | char pathname[32]; | |
270 | int fd; | |
271 | ||
272 | sprintf(pathname, "/dev/cpu/%d/msr", cpu); | |
273 | fd = open(pathname, O_RDONLY); | |
274 | if (fd < 0) | |
275 | err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname); | |
276 | ||
277 | retval = pread(fd, msr, sizeof *msr, offset); | |
278 | close(fd); | |
279 | ||
280 | if (retval != sizeof *msr) | |
281 | err(-1, "%s offset 0x%llx read failed", pathname, (unsigned long long)offset); | |
282 | ||
283 | return 0; | |
284 | } | |
285 | ||
286 | /* | |
287 | * Example Format w/ field column widths: | |
288 | * | |
289 | * Package Core CPU Avg_MHz Bzy_MHz TSC_MHz SMI %Busy CPU_%c1 CPU_%c3 CPU_%c6 CPU_%c7 CoreTmp PkgTmp Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 PkgWatt CorWatt GFXWatt | |
290 | * 123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678 | |
291 | */ | |
292 | ||
293 | void print_header(void) | |
294 | { | |
295 | if (show_pkg) | |
296 | outp += sprintf(outp, " Package"); | |
297 | if (show_core) | |
298 | outp += sprintf(outp, " Core"); | |
299 | if (show_cpu) | |
300 | outp += sprintf(outp, " CPU"); | |
301 | if (has_aperf) | |
302 | outp += sprintf(outp, " Avg_MHz"); | |
303 | if (has_aperf) | |
304 | outp += sprintf(outp, " %%Busy"); | |
305 | if (has_aperf) | |
306 | outp += sprintf(outp, " Bzy_MHz"); | |
307 | outp += sprintf(outp, " TSC_MHz"); | |
308 | ||
309 | if (extra_delta_offset32) | |
310 | outp += sprintf(outp, " count 0x%03X", extra_delta_offset32); | |
311 | if (extra_delta_offset64) | |
312 | outp += sprintf(outp, " COUNT 0x%03X", extra_delta_offset64); | |
313 | if (extra_msr_offset32) | |
314 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32); | |
315 | if (extra_msr_offset64) | |
316 | outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64); | |
317 | ||
318 | if (!debug) | |
319 | goto done; | |
320 | ||
321 | if (do_smi) | |
322 | outp += sprintf(outp, " SMI"); | |
323 | ||
324 | if (do_nhm_cstates) | |
325 | outp += sprintf(outp, " CPU%%c1"); | |
326 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) | |
327 | outp += sprintf(outp, " CPU%%c3"); | |
328 | if (do_nhm_cstates) | |
329 | outp += sprintf(outp, " CPU%%c6"); | |
330 | if (do_snb_cstates) | |
331 | outp += sprintf(outp, " CPU%%c7"); | |
332 | ||
333 | if (do_dts) | |
334 | outp += sprintf(outp, " CoreTmp"); | |
335 | if (do_ptm) | |
336 | outp += sprintf(outp, " PkgTmp"); | |
337 | ||
338 | if (do_skl_residency) { | |
339 | outp += sprintf(outp, " Totl%%C0"); | |
340 | outp += sprintf(outp, " Any%%C0"); | |
341 | outp += sprintf(outp, " GFX%%C0"); | |
342 | outp += sprintf(outp, " CPUGFX%%"); | |
343 | } | |
344 | ||
345 | if (do_pc2) | |
346 | outp += sprintf(outp, " Pkg%%pc2"); | |
347 | if (do_pc3) | |
348 | outp += sprintf(outp, " Pkg%%pc3"); | |
349 | if (do_pc6) | |
350 | outp += sprintf(outp, " Pkg%%pc6"); | |
351 | if (do_pc7) | |
352 | outp += sprintf(outp, " Pkg%%pc7"); | |
353 | if (do_c8_c9_c10) { | |
354 | outp += sprintf(outp, " Pkg%%pc8"); | |
355 | outp += sprintf(outp, " Pkg%%pc9"); | |
356 | outp += sprintf(outp, " Pk%%pc10"); | |
357 | } | |
358 | ||
359 | if (do_rapl && !rapl_joules) { | |
360 | if (do_rapl & RAPL_PKG) | |
361 | outp += sprintf(outp, " PkgWatt"); | |
362 | if (do_rapl & RAPL_CORES) | |
363 | outp += sprintf(outp, " CorWatt"); | |
364 | if (do_rapl & RAPL_GFX) | |
365 | outp += sprintf(outp, " GFXWatt"); | |
366 | if (do_rapl & RAPL_DRAM) | |
367 | outp += sprintf(outp, " RAMWatt"); | |
368 | if (do_rapl & RAPL_PKG_PERF_STATUS) | |
369 | outp += sprintf(outp, " PKG_%%"); | |
370 | if (do_rapl & RAPL_DRAM_PERF_STATUS) | |
371 | outp += sprintf(outp, " RAM_%%"); | |
372 | } else if (do_rapl && rapl_joules) { | |
373 | if (do_rapl & RAPL_PKG) | |
374 | outp += sprintf(outp, " Pkg_J"); | |
375 | if (do_rapl & RAPL_CORES) | |
376 | outp += sprintf(outp, " Cor_J"); | |
377 | if (do_rapl & RAPL_GFX) | |
378 | outp += sprintf(outp, " GFX_J"); | |
379 | if (do_rapl & RAPL_DRAM) | |
380 | outp += sprintf(outp, " RAM_J"); | |
381 | if (do_rapl & RAPL_PKG_PERF_STATUS) | |
382 | outp += sprintf(outp, " PKG_%%"); | |
383 | if (do_rapl & RAPL_DRAM_PERF_STATUS) | |
384 | outp += sprintf(outp, " RAM_%%"); | |
385 | outp += sprintf(outp, " time"); | |
386 | ||
387 | } | |
388 | done: | |
389 | outp += sprintf(outp, "\n"); | |
390 | } | |
391 | ||
392 | int dump_counters(struct thread_data *t, struct core_data *c, | |
393 | struct pkg_data *p) | |
394 | { | |
395 | outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p); | |
396 | ||
397 | if (t) { | |
398 | outp += sprintf(outp, "CPU: %d flags 0x%x\n", | |
399 | t->cpu_id, t->flags); | |
400 | outp += sprintf(outp, "TSC: %016llX\n", t->tsc); | |
401 | outp += sprintf(outp, "aperf: %016llX\n", t->aperf); | |
402 | outp += sprintf(outp, "mperf: %016llX\n", t->mperf); | |
403 | outp += sprintf(outp, "c1: %016llX\n", t->c1); | |
404 | outp += sprintf(outp, "msr0x%x: %08llX\n", | |
405 | extra_delta_offset32, t->extra_delta32); | |
406 | outp += sprintf(outp, "msr0x%x: %016llX\n", | |
407 | extra_delta_offset64, t->extra_delta64); | |
408 | outp += sprintf(outp, "msr0x%x: %08llX\n", | |
409 | extra_msr_offset32, t->extra_msr32); | |
410 | outp += sprintf(outp, "msr0x%x: %016llX\n", | |
411 | extra_msr_offset64, t->extra_msr64); | |
412 | if (do_smi) | |
413 | outp += sprintf(outp, "SMI: %08X\n", t->smi_count); | |
414 | } | |
415 | ||
416 | if (c) { | |
417 | outp += sprintf(outp, "core: %d\n", c->core_id); | |
418 | outp += sprintf(outp, "c3: %016llX\n", c->c3); | |
419 | outp += sprintf(outp, "c6: %016llX\n", c->c6); | |
420 | outp += sprintf(outp, "c7: %016llX\n", c->c7); | |
421 | outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c); | |
422 | } | |
423 | ||
424 | if (p) { | |
425 | outp += sprintf(outp, "package: %d\n", p->package_id); | |
426 | ||
427 | outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0); | |
428 | outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0); | |
429 | outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0); | |
430 | outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0); | |
431 | ||
432 | outp += sprintf(outp, "pc2: %016llX\n", p->pc2); | |
433 | if (do_pc3) | |
434 | outp += sprintf(outp, "pc3: %016llX\n", p->pc3); | |
435 | if (do_pc6) | |
436 | outp += sprintf(outp, "pc6: %016llX\n", p->pc6); | |
437 | if (do_pc7) | |
438 | outp += sprintf(outp, "pc7: %016llX\n", p->pc7); | |
439 | outp += sprintf(outp, "pc8: %016llX\n", p->pc8); | |
440 | outp += sprintf(outp, "pc9: %016llX\n", p->pc9); | |
441 | outp += sprintf(outp, "pc10: %016llX\n", p->pc10); | |
442 | outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg); | |
443 | outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores); | |
444 | outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx); | |
445 | outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram); | |
446 | outp += sprintf(outp, "Throttle PKG: %0X\n", | |
447 | p->rapl_pkg_perf_status); | |
448 | outp += sprintf(outp, "Throttle RAM: %0X\n", | |
449 | p->rapl_dram_perf_status); | |
450 | outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); | |
451 | } | |
452 | ||
453 | outp += sprintf(outp, "\n"); | |
454 | ||
455 | return 0; | |
456 | } | |
457 | ||
458 | /* | |
459 | * column formatting convention & formats | |
460 | */ | |
461 | int format_counters(struct thread_data *t, struct core_data *c, | |
462 | struct pkg_data *p) | |
463 | { | |
464 | double interval_float; | |
465 | char *fmt8; | |
466 | ||
467 | /* if showing only 1st thread in core and this isn't one, bail out */ | |
468 | if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
469 | return 0; | |
470 | ||
471 | /* if showing only 1st thread in pkg and this isn't one, bail out */ | |
472 | if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
473 | return 0; | |
474 | ||
475 | interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; | |
476 | ||
477 | /* topo columns, print blanks on 1st (average) line */ | |
478 | if (t == &average.threads) { | |
479 | if (show_pkg) | |
480 | outp += sprintf(outp, " -"); | |
481 | if (show_core) | |
482 | outp += sprintf(outp, " -"); | |
483 | if (show_cpu) | |
484 | outp += sprintf(outp, " -"); | |
485 | } else { | |
486 | if (show_pkg) { | |
487 | if (p) | |
488 | outp += sprintf(outp, "%8d", p->package_id); | |
489 | else | |
490 | outp += sprintf(outp, " -"); | |
491 | } | |
492 | if (show_core) { | |
493 | if (c) | |
494 | outp += sprintf(outp, "%8d", c->core_id); | |
495 | else | |
496 | outp += sprintf(outp, " -"); | |
497 | } | |
498 | if (show_cpu) | |
499 | outp += sprintf(outp, "%8d", t->cpu_id); | |
500 | } | |
501 | ||
502 | /* Avg_MHz */ | |
503 | if (has_aperf) | |
504 | outp += sprintf(outp, "%8.0f", | |
505 | 1.0 / units * t->aperf / interval_float); | |
506 | ||
507 | /* %Busy */ | |
508 | if (has_aperf) { | |
509 | if (!skip_c0) | |
510 | outp += sprintf(outp, "%8.2f", 100.0 * t->mperf/t->tsc/tsc_tweak); | |
511 | else | |
512 | outp += sprintf(outp, "********"); | |
513 | } | |
514 | ||
515 | /* Bzy_MHz */ | |
516 | if (has_aperf) { | |
517 | if (has_base_hz) | |
518 | outp += sprintf(outp, "%8.0f", base_hz / units * t->aperf / t->mperf); | |
519 | else | |
520 | outp += sprintf(outp, "%8.0f", | |
521 | 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float); | |
522 | } | |
523 | ||
524 | /* TSC_MHz */ | |
525 | outp += sprintf(outp, "%8.0f", 1.0 * t->tsc/units/interval_float); | |
526 | ||
527 | /* delta */ | |
528 | if (extra_delta_offset32) | |
529 | outp += sprintf(outp, " %11llu", t->extra_delta32); | |
530 | ||
531 | /* DELTA */ | |
532 | if (extra_delta_offset64) | |
533 | outp += sprintf(outp, " %11llu", t->extra_delta64); | |
534 | /* msr */ | |
535 | if (extra_msr_offset32) | |
536 | outp += sprintf(outp, " 0x%08llx", t->extra_msr32); | |
537 | ||
538 | /* MSR */ | |
539 | if (extra_msr_offset64) | |
540 | outp += sprintf(outp, " 0x%016llx", t->extra_msr64); | |
541 | ||
542 | if (!debug) | |
543 | goto done; | |
544 | ||
545 | /* SMI */ | |
546 | if (do_smi) | |
547 | outp += sprintf(outp, "%8d", t->smi_count); | |
548 | ||
549 | if (do_nhm_cstates) { | |
550 | if (!skip_c1) | |
551 | outp += sprintf(outp, "%8.2f", 100.0 * t->c1/t->tsc); | |
552 | else | |
553 | outp += sprintf(outp, "********"); | |
554 | } | |
555 | ||
556 | /* print per-core data only for 1st thread in core */ | |
557 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
558 | goto done; | |
559 | ||
560 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) | |
561 | outp += sprintf(outp, "%8.2f", 100.0 * c->c3/t->tsc); | |
562 | if (do_nhm_cstates) | |
563 | outp += sprintf(outp, "%8.2f", 100.0 * c->c6/t->tsc); | |
564 | if (do_snb_cstates) | |
565 | outp += sprintf(outp, "%8.2f", 100.0 * c->c7/t->tsc); | |
566 | ||
567 | if (do_dts) | |
568 | outp += sprintf(outp, "%8d", c->core_temp_c); | |
569 | ||
570 | /* print per-package data only for 1st core in package */ | |
571 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
572 | goto done; | |
573 | ||
574 | /* PkgTmp */ | |
575 | if (do_ptm) | |
576 | outp += sprintf(outp, "%8d", p->pkg_temp_c); | |
577 | ||
578 | /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */ | |
579 | if (do_skl_residency) { | |
580 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_wtd_core_c0/t->tsc); | |
581 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_core_c0/t->tsc); | |
582 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_any_gfxe_c0/t->tsc); | |
583 | outp += sprintf(outp, "%8.2f", 100.0 * p->pkg_both_core_gfxe_c0/t->tsc); | |
584 | } | |
585 | ||
586 | if (do_pc2) | |
587 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc2/t->tsc); | |
588 | if (do_pc3) | |
589 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc3/t->tsc); | |
590 | if (do_pc6) | |
591 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc6/t->tsc); | |
592 | if (do_pc7) | |
593 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc7/t->tsc); | |
594 | if (do_c8_c9_c10) { | |
595 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc8/t->tsc); | |
596 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc9/t->tsc); | |
597 | outp += sprintf(outp, "%8.2f", 100.0 * p->pc10/t->tsc); | |
598 | } | |
599 | ||
600 | /* | |
601 | * If measurement interval exceeds minimum RAPL Joule Counter range, | |
602 | * indicate that results are suspect by printing "**" in fraction place. | |
603 | */ | |
604 | if (interval_float < rapl_joule_counter_range) | |
605 | fmt8 = "%8.2f"; | |
606 | else | |
607 | fmt8 = " %6.0f**"; | |
608 | ||
609 | if (do_rapl && !rapl_joules) { | |
610 | if (do_rapl & RAPL_PKG) | |
611 | outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float); | |
612 | if (do_rapl & RAPL_CORES) | |
613 | outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float); | |
614 | if (do_rapl & RAPL_GFX) | |
615 | outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float); | |
616 | if (do_rapl & RAPL_DRAM) | |
617 | outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units / interval_float); | |
618 | if (do_rapl & RAPL_PKG_PERF_STATUS) | |
619 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); | |
620 | if (do_rapl & RAPL_DRAM_PERF_STATUS) | |
621 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); | |
622 | } else if (do_rapl && rapl_joules) { | |
623 | if (do_rapl & RAPL_PKG) | |
624 | outp += sprintf(outp, fmt8, | |
625 | p->energy_pkg * rapl_energy_units); | |
626 | if (do_rapl & RAPL_CORES) | |
627 | outp += sprintf(outp, fmt8, | |
628 | p->energy_cores * rapl_energy_units); | |
629 | if (do_rapl & RAPL_GFX) | |
630 | outp += sprintf(outp, fmt8, | |
631 | p->energy_gfx * rapl_energy_units); | |
632 | if (do_rapl & RAPL_DRAM) | |
633 | outp += sprintf(outp, fmt8, | |
634 | p->energy_dram * rapl_dram_energy_units); | |
635 | if (do_rapl & RAPL_PKG_PERF_STATUS) | |
636 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); | |
637 | if (do_rapl & RAPL_DRAM_PERF_STATUS) | |
638 | outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); | |
639 | ||
640 | outp += sprintf(outp, fmt8, interval_float); | |
641 | } | |
642 | done: | |
643 | outp += sprintf(outp, "\n"); | |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
648 | void flush_stdout() | |
649 | { | |
650 | fputs(output_buffer, stdout); | |
651 | fflush(stdout); | |
652 | outp = output_buffer; | |
653 | } | |
654 | void flush_stderr() | |
655 | { | |
656 | fputs(output_buffer, stderr); | |
657 | outp = output_buffer; | |
658 | } | |
659 | void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
660 | { | |
661 | static int printed; | |
662 | ||
663 | if (!printed || !summary_only) | |
664 | print_header(); | |
665 | ||
666 | if (topo.num_cpus > 1) | |
667 | format_counters(&average.threads, &average.cores, | |
668 | &average.packages); | |
669 | ||
670 | printed = 1; | |
671 | ||
672 | if (summary_only) | |
673 | return; | |
674 | ||
675 | for_all_cpus(format_counters, t, c, p); | |
676 | } | |
677 | ||
678 | #define DELTA_WRAP32(new, old) \ | |
679 | if (new > old) { \ | |
680 | old = new - old; \ | |
681 | } else { \ | |
682 | old = 0x100000000 + new - old; \ | |
683 | } | |
684 | ||
685 | void | |
686 | delta_package(struct pkg_data *new, struct pkg_data *old) | |
687 | { | |
688 | ||
689 | if (do_skl_residency) { | |
690 | old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0; | |
691 | old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0; | |
692 | old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0; | |
693 | old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0; | |
694 | } | |
695 | old->pc2 = new->pc2 - old->pc2; | |
696 | if (do_pc3) | |
697 | old->pc3 = new->pc3 - old->pc3; | |
698 | if (do_pc6) | |
699 | old->pc6 = new->pc6 - old->pc6; | |
700 | if (do_pc7) | |
701 | old->pc7 = new->pc7 - old->pc7; | |
702 | old->pc8 = new->pc8 - old->pc8; | |
703 | old->pc9 = new->pc9 - old->pc9; | |
704 | old->pc10 = new->pc10 - old->pc10; | |
705 | old->pkg_temp_c = new->pkg_temp_c; | |
706 | ||
707 | DELTA_WRAP32(new->energy_pkg, old->energy_pkg); | |
708 | DELTA_WRAP32(new->energy_cores, old->energy_cores); | |
709 | DELTA_WRAP32(new->energy_gfx, old->energy_gfx); | |
710 | DELTA_WRAP32(new->energy_dram, old->energy_dram); | |
711 | DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status); | |
712 | DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status); | |
713 | } | |
714 | ||
715 | void | |
716 | delta_core(struct core_data *new, struct core_data *old) | |
717 | { | |
718 | old->c3 = new->c3 - old->c3; | |
719 | old->c6 = new->c6 - old->c6; | |
720 | old->c7 = new->c7 - old->c7; | |
721 | old->core_temp_c = new->core_temp_c; | |
722 | } | |
723 | ||
724 | /* | |
725 | * old = new - old | |
726 | */ | |
727 | void | |
728 | delta_thread(struct thread_data *new, struct thread_data *old, | |
729 | struct core_data *core_delta) | |
730 | { | |
731 | old->tsc = new->tsc - old->tsc; | |
732 | ||
733 | /* check for TSC < 1 Mcycles over interval */ | |
734 | if (old->tsc < (1000 * 1000)) | |
735 | errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n" | |
736 | "You can disable all c-states by booting with \"idle=poll\"\n" | |
737 | "or just the deep ones with \"processor.max_cstate=1\""); | |
738 | ||
739 | old->c1 = new->c1 - old->c1; | |
740 | ||
741 | if (has_aperf) { | |
742 | if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) { | |
743 | old->aperf = new->aperf - old->aperf; | |
744 | old->mperf = new->mperf - old->mperf; | |
745 | } else { | |
746 | ||
747 | if (!aperf_mperf_unstable) { | |
748 | fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname); | |
749 | fprintf(stderr, "* Frequency results do not cover entire interval *\n"); | |
750 | fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n"); | |
751 | ||
752 | aperf_mperf_unstable = 1; | |
753 | } | |
754 | /* | |
755 | * mperf delta is likely a huge "positive" number | |
756 | * can not use it for calculating c0 time | |
757 | */ | |
758 | skip_c0 = 1; | |
759 | skip_c1 = 1; | |
760 | } | |
761 | } | |
762 | ||
763 | ||
764 | if (use_c1_residency_msr) { | |
765 | /* | |
766 | * Some models have a dedicated C1 residency MSR, | |
767 | * which should be more accurate than the derivation below. | |
768 | */ | |
769 | } else { | |
770 | /* | |
771 | * As counter collection is not atomic, | |
772 | * it is possible for mperf's non-halted cycles + idle states | |
773 | * to exceed TSC's all cycles: show c1 = 0% in that case. | |
774 | */ | |
775 | if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc) | |
776 | old->c1 = 0; | |
777 | else { | |
778 | /* normal case, derive c1 */ | |
779 | old->c1 = old->tsc - old->mperf - core_delta->c3 | |
780 | - core_delta->c6 - core_delta->c7; | |
781 | } | |
782 | } | |
783 | ||
784 | if (old->mperf == 0) { | |
785 | if (debug > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id); | |
786 | old->mperf = 1; /* divide by 0 protection */ | |
787 | } | |
788 | ||
789 | old->extra_delta32 = new->extra_delta32 - old->extra_delta32; | |
790 | old->extra_delta32 &= 0xFFFFFFFF; | |
791 | ||
792 | old->extra_delta64 = new->extra_delta64 - old->extra_delta64; | |
793 | ||
794 | /* | |
795 | * Extra MSR is just a snapshot, simply copy latest w/o subtracting | |
796 | */ | |
797 | old->extra_msr32 = new->extra_msr32; | |
798 | old->extra_msr64 = new->extra_msr64; | |
799 | ||
800 | if (do_smi) | |
801 | old->smi_count = new->smi_count - old->smi_count; | |
802 | } | |
803 | ||
804 | int delta_cpu(struct thread_data *t, struct core_data *c, | |
805 | struct pkg_data *p, struct thread_data *t2, | |
806 | struct core_data *c2, struct pkg_data *p2) | |
807 | { | |
808 | /* calculate core delta only for 1st thread in core */ | |
809 | if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE) | |
810 | delta_core(c, c2); | |
811 | ||
812 | /* always calculate thread delta */ | |
813 | delta_thread(t, t2, c2); /* c2 is core delta */ | |
814 | ||
815 | /* calculate package delta only for 1st core in package */ | |
816 | if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) | |
817 | delta_package(p, p2); | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
822 | void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
823 | { | |
824 | t->tsc = 0; | |
825 | t->aperf = 0; | |
826 | t->mperf = 0; | |
827 | t->c1 = 0; | |
828 | ||
829 | t->smi_count = 0; | |
830 | t->extra_delta32 = 0; | |
831 | t->extra_delta64 = 0; | |
832 | ||
833 | /* tells format_counters to dump all fields from this set */ | |
834 | t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE; | |
835 | ||
836 | c->c3 = 0; | |
837 | c->c6 = 0; | |
838 | c->c7 = 0; | |
839 | c->core_temp_c = 0; | |
840 | ||
841 | p->pkg_wtd_core_c0 = 0; | |
842 | p->pkg_any_core_c0 = 0; | |
843 | p->pkg_any_gfxe_c0 = 0; | |
844 | p->pkg_both_core_gfxe_c0 = 0; | |
845 | ||
846 | p->pc2 = 0; | |
847 | if (do_pc3) | |
848 | p->pc3 = 0; | |
849 | if (do_pc6) | |
850 | p->pc6 = 0; | |
851 | if (do_pc7) | |
852 | p->pc7 = 0; | |
853 | p->pc8 = 0; | |
854 | p->pc9 = 0; | |
855 | p->pc10 = 0; | |
856 | ||
857 | p->energy_pkg = 0; | |
858 | p->energy_dram = 0; | |
859 | p->energy_cores = 0; | |
860 | p->energy_gfx = 0; | |
861 | p->rapl_pkg_perf_status = 0; | |
862 | p->rapl_dram_perf_status = 0; | |
863 | p->pkg_temp_c = 0; | |
864 | } | |
865 | int sum_counters(struct thread_data *t, struct core_data *c, | |
866 | struct pkg_data *p) | |
867 | { | |
868 | average.threads.tsc += t->tsc; | |
869 | average.threads.aperf += t->aperf; | |
870 | average.threads.mperf += t->mperf; | |
871 | average.threads.c1 += t->c1; | |
872 | ||
873 | average.threads.extra_delta32 += t->extra_delta32; | |
874 | average.threads.extra_delta64 += t->extra_delta64; | |
875 | ||
876 | /* sum per-core values only for 1st thread in core */ | |
877 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
878 | return 0; | |
879 | ||
880 | average.cores.c3 += c->c3; | |
881 | average.cores.c6 += c->c6; | |
882 | average.cores.c7 += c->c7; | |
883 | ||
884 | average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c); | |
885 | ||
886 | /* sum per-pkg values only for 1st core in pkg */ | |
887 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
888 | return 0; | |
889 | ||
890 | if (do_skl_residency) { | |
891 | average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0; | |
892 | average.packages.pkg_any_core_c0 += p->pkg_any_core_c0; | |
893 | average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0; | |
894 | average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0; | |
895 | } | |
896 | ||
897 | average.packages.pc2 += p->pc2; | |
898 | if (do_pc3) | |
899 | average.packages.pc3 += p->pc3; | |
900 | if (do_pc6) | |
901 | average.packages.pc6 += p->pc6; | |
902 | if (do_pc7) | |
903 | average.packages.pc7 += p->pc7; | |
904 | average.packages.pc8 += p->pc8; | |
905 | average.packages.pc9 += p->pc9; | |
906 | average.packages.pc10 += p->pc10; | |
907 | ||
908 | average.packages.energy_pkg += p->energy_pkg; | |
909 | average.packages.energy_dram += p->energy_dram; | |
910 | average.packages.energy_cores += p->energy_cores; | |
911 | average.packages.energy_gfx += p->energy_gfx; | |
912 | ||
913 | average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); | |
914 | ||
915 | average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status; | |
916 | average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; | |
917 | return 0; | |
918 | } | |
919 | /* | |
920 | * sum the counters for all cpus in the system | |
921 | * compute the weighted average | |
922 | */ | |
923 | void compute_average(struct thread_data *t, struct core_data *c, | |
924 | struct pkg_data *p) | |
925 | { | |
926 | clear_counters(&average.threads, &average.cores, &average.packages); | |
927 | ||
928 | for_all_cpus(sum_counters, t, c, p); | |
929 | ||
930 | average.threads.tsc /= topo.num_cpus; | |
931 | average.threads.aperf /= topo.num_cpus; | |
932 | average.threads.mperf /= topo.num_cpus; | |
933 | average.threads.c1 /= topo.num_cpus; | |
934 | ||
935 | average.threads.extra_delta32 /= topo.num_cpus; | |
936 | average.threads.extra_delta32 &= 0xFFFFFFFF; | |
937 | ||
938 | average.threads.extra_delta64 /= topo.num_cpus; | |
939 | ||
940 | average.cores.c3 /= topo.num_cores; | |
941 | average.cores.c6 /= topo.num_cores; | |
942 | average.cores.c7 /= topo.num_cores; | |
943 | ||
944 | if (do_skl_residency) { | |
945 | average.packages.pkg_wtd_core_c0 /= topo.num_packages; | |
946 | average.packages.pkg_any_core_c0 /= topo.num_packages; | |
947 | average.packages.pkg_any_gfxe_c0 /= topo.num_packages; | |
948 | average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages; | |
949 | } | |
950 | ||
951 | average.packages.pc2 /= topo.num_packages; | |
952 | if (do_pc3) | |
953 | average.packages.pc3 /= topo.num_packages; | |
954 | if (do_pc6) | |
955 | average.packages.pc6 /= topo.num_packages; | |
956 | if (do_pc7) | |
957 | average.packages.pc7 /= topo.num_packages; | |
958 | ||
959 | average.packages.pc8 /= topo.num_packages; | |
960 | average.packages.pc9 /= topo.num_packages; | |
961 | average.packages.pc10 /= topo.num_packages; | |
962 | } | |
963 | ||
964 | static unsigned long long rdtsc(void) | |
965 | { | |
966 | unsigned int low, high; | |
967 | ||
968 | asm volatile("rdtsc" : "=a" (low), "=d" (high)); | |
969 | ||
970 | return low | ((unsigned long long)high) << 32; | |
971 | } | |
972 | ||
973 | ||
974 | /* | |
975 | * get_counters(...) | |
976 | * migrate to cpu | |
977 | * acquire and record local counters for that cpu | |
978 | */ | |
979 | int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
980 | { | |
981 | int cpu = t->cpu_id; | |
982 | unsigned long long msr; | |
983 | ||
984 | if (cpu_migrate(cpu)) { | |
985 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
986 | return -1; | |
987 | } | |
988 | ||
989 | t->tsc = rdtsc(); /* we are running on local CPU of interest */ | |
990 | ||
991 | if (has_aperf) { | |
992 | if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) | |
993 | return -3; | |
994 | if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) | |
995 | return -4; | |
996 | t->aperf = t->aperf * aperf_mperf_multiplier; | |
997 | t->mperf = t->mperf * aperf_mperf_multiplier; | |
998 | } | |
999 | ||
1000 | if (do_smi) { | |
1001 | if (get_msr(cpu, MSR_SMI_COUNT, &msr)) | |
1002 | return -5; | |
1003 | t->smi_count = msr & 0xFFFFFFFF; | |
1004 | } | |
1005 | if (extra_delta_offset32) { | |
1006 | if (get_msr(cpu, extra_delta_offset32, &msr)) | |
1007 | return -5; | |
1008 | t->extra_delta32 = msr & 0xFFFFFFFF; | |
1009 | } | |
1010 | ||
1011 | if (extra_delta_offset64) | |
1012 | if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64)) | |
1013 | return -5; | |
1014 | ||
1015 | if (extra_msr_offset32) { | |
1016 | if (get_msr(cpu, extra_msr_offset32, &msr)) | |
1017 | return -5; | |
1018 | t->extra_msr32 = msr & 0xFFFFFFFF; | |
1019 | } | |
1020 | ||
1021 | if (extra_msr_offset64) | |
1022 | if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64)) | |
1023 | return -5; | |
1024 | ||
1025 | if (use_c1_residency_msr) { | |
1026 | if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1)) | |
1027 | return -6; | |
1028 | } | |
1029 | ||
1030 | /* collect core counters only for 1st thread in core */ | |
1031 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
1032 | return 0; | |
1033 | ||
1034 | if (do_nhm_cstates && !do_slm_cstates && !do_knl_cstates) { | |
1035 | if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) | |
1036 | return -6; | |
1037 | } | |
1038 | ||
1039 | if (do_nhm_cstates && !do_knl_cstates) { | |
1040 | if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) | |
1041 | return -7; | |
1042 | } else if (do_knl_cstates) { | |
1043 | if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6)) | |
1044 | return -7; | |
1045 | } | |
1046 | ||
1047 | if (do_snb_cstates) | |
1048 | if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) | |
1049 | return -8; | |
1050 | ||
1051 | if (do_dts) { | |
1052 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
1053 | return -9; | |
1054 | c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1055 | } | |
1056 | ||
1057 | ||
1058 | /* collect package counters only for 1st core in package */ | |
1059 | if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
1060 | return 0; | |
1061 | ||
1062 | if (do_skl_residency) { | |
1063 | if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0)) | |
1064 | return -10; | |
1065 | if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0)) | |
1066 | return -11; | |
1067 | if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0)) | |
1068 | return -12; | |
1069 | if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0)) | |
1070 | return -13; | |
1071 | } | |
1072 | if (do_pc3) | |
1073 | if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) | |
1074 | return -9; | |
1075 | if (do_pc6) | |
1076 | if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) | |
1077 | return -10; | |
1078 | if (do_pc2) | |
1079 | if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) | |
1080 | return -11; | |
1081 | if (do_pc7) | |
1082 | if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) | |
1083 | return -12; | |
1084 | if (do_c8_c9_c10) { | |
1085 | if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8)) | |
1086 | return -13; | |
1087 | if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9)) | |
1088 | return -13; | |
1089 | if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10)) | |
1090 | return -13; | |
1091 | } | |
1092 | if (do_rapl & RAPL_PKG) { | |
1093 | if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr)) | |
1094 | return -13; | |
1095 | p->energy_pkg = msr & 0xFFFFFFFF; | |
1096 | } | |
1097 | if (do_rapl & RAPL_CORES) { | |
1098 | if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr)) | |
1099 | return -14; | |
1100 | p->energy_cores = msr & 0xFFFFFFFF; | |
1101 | } | |
1102 | if (do_rapl & RAPL_DRAM) { | |
1103 | if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) | |
1104 | return -15; | |
1105 | p->energy_dram = msr & 0xFFFFFFFF; | |
1106 | } | |
1107 | if (do_rapl & RAPL_GFX) { | |
1108 | if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr)) | |
1109 | return -16; | |
1110 | p->energy_gfx = msr & 0xFFFFFFFF; | |
1111 | } | |
1112 | if (do_rapl & RAPL_PKG_PERF_STATUS) { | |
1113 | if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr)) | |
1114 | return -16; | |
1115 | p->rapl_pkg_perf_status = msr & 0xFFFFFFFF; | |
1116 | } | |
1117 | if (do_rapl & RAPL_DRAM_PERF_STATUS) { | |
1118 | if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr)) | |
1119 | return -16; | |
1120 | p->rapl_dram_perf_status = msr & 0xFFFFFFFF; | |
1121 | } | |
1122 | if (do_ptm) { | |
1123 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
1124 | return -17; | |
1125 | p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F); | |
1126 | } | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: | |
1132 | * If you change the values, note they are used both in comparisons | |
1133 | * (>= PCL__7) and to index pkg_cstate_limit_strings[]. | |
1134 | */ | |
1135 | ||
1136 | #define PCLUKN 0 /* Unknown */ | |
1137 | #define PCLRSV 1 /* Reserved */ | |
1138 | #define PCL__0 2 /* PC0 */ | |
1139 | #define PCL__1 3 /* PC1 */ | |
1140 | #define PCL__2 4 /* PC2 */ | |
1141 | #define PCL__3 5 /* PC3 */ | |
1142 | #define PCL__4 6 /* PC4 */ | |
1143 | #define PCL__6 7 /* PC6 */ | |
1144 | #define PCL_6N 8 /* PC6 No Retention */ | |
1145 | #define PCL_6R 9 /* PC6 Retention */ | |
1146 | #define PCL__7 10 /* PC7 */ | |
1147 | #define PCL_7S 11 /* PC7 Shrink */ | |
1148 | #define PCL__8 12 /* PC8 */ | |
1149 | #define PCL__9 13 /* PC9 */ | |
1150 | #define PCLUNL 14 /* Unlimited */ | |
1151 | ||
1152 | int pkg_cstate_limit = PCLUKN; | |
1153 | char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2", | |
1154 | "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"}; | |
1155 | ||
1156 | 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}; | |
1157 | 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}; | |
1158 | 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}; | |
1159 | 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}; | |
1160 | 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}; | |
1161 | 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}; | |
1162 | ||
1163 | ||
1164 | static void | |
1165 | calculate_tsc_tweak() | |
1166 | { | |
1167 | tsc_tweak = base_hz / tsc_hz; | |
1168 | } | |
1169 | ||
1170 | static void | |
1171 | dump_nhm_platform_info(void) | |
1172 | { | |
1173 | unsigned long long msr; | |
1174 | unsigned int ratio; | |
1175 | ||
1176 | get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); | |
1177 | ||
1178 | fprintf(stderr, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr); | |
1179 | ||
1180 | ratio = (msr >> 40) & 0xFF; | |
1181 | fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency frequency\n", | |
1182 | ratio, bclk, ratio * bclk); | |
1183 | ||
1184 | ratio = (msr >> 8) & 0xFF; | |
1185 | fprintf(stderr, "%d * %.0f = %.0f MHz base frequency\n", | |
1186 | ratio, bclk, ratio * bclk); | |
1187 | ||
1188 | get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr); | |
1189 | fprintf(stderr, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n", | |
1190 | base_cpu, msr, msr & 0x2 ? "EN" : "DIS"); | |
1191 | ||
1192 | return; | |
1193 | } | |
1194 | ||
1195 | static void | |
1196 | dump_hsw_turbo_ratio_limits(void) | |
1197 | { | |
1198 | unsigned long long msr; | |
1199 | unsigned int ratio; | |
1200 | ||
1201 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr); | |
1202 | ||
1203 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr); | |
1204 | ||
1205 | ratio = (msr >> 8) & 0xFF; | |
1206 | if (ratio) | |
1207 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 18 active cores\n", | |
1208 | ratio, bclk, ratio * bclk); | |
1209 | ||
1210 | ratio = (msr >> 0) & 0xFF; | |
1211 | if (ratio) | |
1212 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 17 active cores\n", | |
1213 | ratio, bclk, ratio * bclk); | |
1214 | return; | |
1215 | } | |
1216 | ||
1217 | static void | |
1218 | dump_ivt_turbo_ratio_limits(void) | |
1219 | { | |
1220 | unsigned long long msr; | |
1221 | unsigned int ratio; | |
1222 | ||
1223 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr); | |
1224 | ||
1225 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr); | |
1226 | ||
1227 | ratio = (msr >> 56) & 0xFF; | |
1228 | if (ratio) | |
1229 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n", | |
1230 | ratio, bclk, ratio * bclk); | |
1231 | ||
1232 | ratio = (msr >> 48) & 0xFF; | |
1233 | if (ratio) | |
1234 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n", | |
1235 | ratio, bclk, ratio * bclk); | |
1236 | ||
1237 | ratio = (msr >> 40) & 0xFF; | |
1238 | if (ratio) | |
1239 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n", | |
1240 | ratio, bclk, ratio * bclk); | |
1241 | ||
1242 | ratio = (msr >> 32) & 0xFF; | |
1243 | if (ratio) | |
1244 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n", | |
1245 | ratio, bclk, ratio * bclk); | |
1246 | ||
1247 | ratio = (msr >> 24) & 0xFF; | |
1248 | if (ratio) | |
1249 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n", | |
1250 | ratio, bclk, ratio * bclk); | |
1251 | ||
1252 | ratio = (msr >> 16) & 0xFF; | |
1253 | if (ratio) | |
1254 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n", | |
1255 | ratio, bclk, ratio * bclk); | |
1256 | ||
1257 | ratio = (msr >> 8) & 0xFF; | |
1258 | if (ratio) | |
1259 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n", | |
1260 | ratio, bclk, ratio * bclk); | |
1261 | ||
1262 | ratio = (msr >> 0) & 0xFF; | |
1263 | if (ratio) | |
1264 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n", | |
1265 | ratio, bclk, ratio * bclk); | |
1266 | return; | |
1267 | } | |
1268 | ||
1269 | static void | |
1270 | dump_nhm_turbo_ratio_limits(void) | |
1271 | { | |
1272 | unsigned long long msr; | |
1273 | unsigned int ratio; | |
1274 | ||
1275 | get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr); | |
1276 | ||
1277 | fprintf(stderr, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr); | |
1278 | ||
1279 | ratio = (msr >> 56) & 0xFF; | |
1280 | if (ratio) | |
1281 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n", | |
1282 | ratio, bclk, ratio * bclk); | |
1283 | ||
1284 | ratio = (msr >> 48) & 0xFF; | |
1285 | if (ratio) | |
1286 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n", | |
1287 | ratio, bclk, ratio * bclk); | |
1288 | ||
1289 | ratio = (msr >> 40) & 0xFF; | |
1290 | if (ratio) | |
1291 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n", | |
1292 | ratio, bclk, ratio * bclk); | |
1293 | ||
1294 | ratio = (msr >> 32) & 0xFF; | |
1295 | if (ratio) | |
1296 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n", | |
1297 | ratio, bclk, ratio * bclk); | |
1298 | ||
1299 | ratio = (msr >> 24) & 0xFF; | |
1300 | if (ratio) | |
1301 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n", | |
1302 | ratio, bclk, ratio * bclk); | |
1303 | ||
1304 | ratio = (msr >> 16) & 0xFF; | |
1305 | if (ratio) | |
1306 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n", | |
1307 | ratio, bclk, ratio * bclk); | |
1308 | ||
1309 | ratio = (msr >> 8) & 0xFF; | |
1310 | if (ratio) | |
1311 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n", | |
1312 | ratio, bclk, ratio * bclk); | |
1313 | ||
1314 | ratio = (msr >> 0) & 0xFF; | |
1315 | if (ratio) | |
1316 | fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n", | |
1317 | ratio, bclk, ratio * bclk); | |
1318 | return; | |
1319 | } | |
1320 | ||
1321 | static void | |
1322 | dump_knl_turbo_ratio_limits(void) | |
1323 | { | |
1324 | int cores; | |
1325 | unsigned int ratio; | |
1326 | unsigned long long msr; | |
1327 | int delta_cores; | |
1328 | int delta_ratio; | |
1329 | int i; | |
1330 | ||
1331 | get_msr(base_cpu, MSR_NHM_TURBO_RATIO_LIMIT, &msr); | |
1332 | ||
1333 | fprintf(stderr, "cpu%d: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", | |
1334 | base_cpu, msr); | |
1335 | ||
1336 | /** | |
1337 | * Turbo encoding in KNL is as follows: | |
1338 | * [7:0] -- Base value of number of active cores of bucket 1. | |
1339 | * [15:8] -- Base value of freq ratio of bucket 1. | |
1340 | * [20:16] -- +ve delta of number of active cores of bucket 2. | |
1341 | * i.e. active cores of bucket 2 = | |
1342 | * active cores of bucket 1 + delta | |
1343 | * [23:21] -- Negative delta of freq ratio of bucket 2. | |
1344 | * i.e. freq ratio of bucket 2 = | |
1345 | * freq ratio of bucket 1 - delta | |
1346 | * [28:24]-- +ve delta of number of active cores of bucket 3. | |
1347 | * [31:29]-- -ve delta of freq ratio of bucket 3. | |
1348 | * [36:32]-- +ve delta of number of active cores of bucket 4. | |
1349 | * [39:37]-- -ve delta of freq ratio of bucket 4. | |
1350 | * [44:40]-- +ve delta of number of active cores of bucket 5. | |
1351 | * [47:45]-- -ve delta of freq ratio of bucket 5. | |
1352 | * [52:48]-- +ve delta of number of active cores of bucket 6. | |
1353 | * [55:53]-- -ve delta of freq ratio of bucket 6. | |
1354 | * [60:56]-- +ve delta of number of active cores of bucket 7. | |
1355 | * [63:61]-- -ve delta of freq ratio of bucket 7. | |
1356 | */ | |
1357 | cores = msr & 0xFF; | |
1358 | ratio = (msr >> 8) && 0xFF; | |
1359 | if (ratio > 0) | |
1360 | fprintf(stderr, | |
1361 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1362 | ratio, bclk, ratio * bclk, cores); | |
1363 | ||
1364 | for (i = 16; i < 64; i = i + 8) { | |
1365 | delta_cores = (msr >> i) & 0x1F; | |
1366 | delta_ratio = (msr >> (i + 5)) && 0x7; | |
1367 | if (!delta_cores || !delta_ratio) | |
1368 | return; | |
1369 | cores = cores + delta_cores; | |
1370 | ratio = ratio - delta_ratio; | |
1371 | ||
1372 | /** -ve ratios will make successive ratio calculations | |
1373 | * negative. Hence return instead of carrying on. | |
1374 | */ | |
1375 | if (ratio > 0) | |
1376 | fprintf(stderr, | |
1377 | "%d * %.0f = %.0f MHz max turbo %d active cores\n", | |
1378 | ratio, bclk, ratio * bclk, cores); | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | static void | |
1383 | dump_nhm_cst_cfg(void) | |
1384 | { | |
1385 | unsigned long long msr; | |
1386 | ||
1387 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); | |
1388 | ||
1389 | #define SNB_C1_AUTO_UNDEMOTE (1UL << 27) | |
1390 | #define SNB_C3_AUTO_UNDEMOTE (1UL << 28) | |
1391 | ||
1392 | fprintf(stderr, "cpu%d: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", base_cpu, msr); | |
1393 | ||
1394 | fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n", | |
1395 | (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "", | |
1396 | (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "", | |
1397 | (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "", | |
1398 | (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "", | |
1399 | (msr & (1 << 15)) ? "" : "UN", | |
1400 | (unsigned int)msr & 7, | |
1401 | pkg_cstate_limit_strings[pkg_cstate_limit]); | |
1402 | return; | |
1403 | } | |
1404 | ||
1405 | static void | |
1406 | dump_config_tdp(void) | |
1407 | { | |
1408 | unsigned long long msr; | |
1409 | ||
1410 | get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr); | |
1411 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr); | |
1412 | fprintf(stderr, " (base_ratio=%d)\n", (unsigned int)msr & 0xEF); | |
1413 | ||
1414 | get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr); | |
1415 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr); | |
1416 | if (msr) { | |
1417 | fprintf(stderr, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0xEFFF); | |
1418 | fprintf(stderr, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0xEFFF); | |
1419 | fprintf(stderr, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xEF); | |
1420 | fprintf(stderr, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0xEFFF); | |
1421 | } | |
1422 | fprintf(stderr, ")\n"); | |
1423 | ||
1424 | get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr); | |
1425 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr); | |
1426 | if (msr) { | |
1427 | fprintf(stderr, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0xEFFF); | |
1428 | fprintf(stderr, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0xEFFF); | |
1429 | fprintf(stderr, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xEF); | |
1430 | fprintf(stderr, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0xEFFF); | |
1431 | } | |
1432 | fprintf(stderr, ")\n"); | |
1433 | ||
1434 | get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr); | |
1435 | fprintf(stderr, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr); | |
1436 | if ((msr) & 0x3) | |
1437 | fprintf(stderr, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3); | |
1438 | fprintf(stderr, " lock=%d", (unsigned int)(msr >> 31) & 1); | |
1439 | fprintf(stderr, ")\n"); | |
1440 | ||
1441 | get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr); | |
1442 | fprintf(stderr, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr); | |
1443 | fprintf(stderr, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0x7F); | |
1444 | fprintf(stderr, " lock=%d", (unsigned int)(msr >> 31) & 1); | |
1445 | fprintf(stderr, ")\n"); | |
1446 | } | |
1447 | ||
1448 | void free_all_buffers(void) | |
1449 | { | |
1450 | CPU_FREE(cpu_present_set); | |
1451 | cpu_present_set = NULL; | |
1452 | cpu_present_set = 0; | |
1453 | ||
1454 | CPU_FREE(cpu_affinity_set); | |
1455 | cpu_affinity_set = NULL; | |
1456 | cpu_affinity_setsize = 0; | |
1457 | ||
1458 | free(thread_even); | |
1459 | free(core_even); | |
1460 | free(package_even); | |
1461 | ||
1462 | thread_even = NULL; | |
1463 | core_even = NULL; | |
1464 | package_even = NULL; | |
1465 | ||
1466 | free(thread_odd); | |
1467 | free(core_odd); | |
1468 | free(package_odd); | |
1469 | ||
1470 | thread_odd = NULL; | |
1471 | core_odd = NULL; | |
1472 | package_odd = NULL; | |
1473 | ||
1474 | free(output_buffer); | |
1475 | output_buffer = NULL; | |
1476 | outp = NULL; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Open a file, and exit on failure | |
1481 | */ | |
1482 | FILE *fopen_or_die(const char *path, const char *mode) | |
1483 | { | |
1484 | FILE *filep = fopen(path, "r"); | |
1485 | if (!filep) | |
1486 | err(1, "%s: open failed", path); | |
1487 | return filep; | |
1488 | } | |
1489 | ||
1490 | /* | |
1491 | * Parse a file containing a single int. | |
1492 | */ | |
1493 | int parse_int_file(const char *fmt, ...) | |
1494 | { | |
1495 | va_list args; | |
1496 | char path[PATH_MAX]; | |
1497 | FILE *filep; | |
1498 | int value; | |
1499 | ||
1500 | va_start(args, fmt); | |
1501 | vsnprintf(path, sizeof(path), fmt, args); | |
1502 | va_end(args); | |
1503 | filep = fopen_or_die(path, "r"); | |
1504 | if (fscanf(filep, "%d", &value) != 1) | |
1505 | err(1, "%s: failed to parse number from file", path); | |
1506 | fclose(filep); | |
1507 | return value; | |
1508 | } | |
1509 | ||
1510 | /* | |
1511 | * get_cpu_position_in_core(cpu) | |
1512 | * return the position of the CPU among its HT siblings in the core | |
1513 | * return -1 if the sibling is not in list | |
1514 | */ | |
1515 | int get_cpu_position_in_core(int cpu) | |
1516 | { | |
1517 | char path[64]; | |
1518 | FILE *filep; | |
1519 | int this_cpu; | |
1520 | char character; | |
1521 | int i; | |
1522 | ||
1523 | sprintf(path, | |
1524 | "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", | |
1525 | cpu); | |
1526 | filep = fopen(path, "r"); | |
1527 | if (filep == NULL) { | |
1528 | perror(path); | |
1529 | exit(1); | |
1530 | } | |
1531 | ||
1532 | for (i = 0; i < topo.num_threads_per_core; i++) { | |
1533 | fscanf(filep, "%d", &this_cpu); | |
1534 | if (this_cpu == cpu) { | |
1535 | fclose(filep); | |
1536 | return i; | |
1537 | } | |
1538 | ||
1539 | /* Account for no separator after last thread*/ | |
1540 | if (i != (topo.num_threads_per_core - 1)) | |
1541 | fscanf(filep, "%c", &character); | |
1542 | } | |
1543 | ||
1544 | fclose(filep); | |
1545 | return -1; | |
1546 | } | |
1547 | ||
1548 | /* | |
1549 | * cpu_is_first_core_in_package(cpu) | |
1550 | * return 1 if given CPU is 1st core in package | |
1551 | */ | |
1552 | int cpu_is_first_core_in_package(int cpu) | |
1553 | { | |
1554 | return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu); | |
1555 | } | |
1556 | ||
1557 | int get_physical_package_id(int cpu) | |
1558 | { | |
1559 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu); | |
1560 | } | |
1561 | ||
1562 | int get_core_id(int cpu) | |
1563 | { | |
1564 | return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu); | |
1565 | } | |
1566 | ||
1567 | int get_num_ht_siblings(int cpu) | |
1568 | { | |
1569 | char path[80]; | |
1570 | FILE *filep; | |
1571 | int sib1; | |
1572 | int matches = 0; | |
1573 | char character; | |
1574 | char str[100]; | |
1575 | char *ch; | |
1576 | ||
1577 | sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); | |
1578 | filep = fopen_or_die(path, "r"); | |
1579 | ||
1580 | /* | |
1581 | * file format: | |
1582 | * A ',' separated or '-' separated set of numbers | |
1583 | * (eg 1-2 or 1,3,4,5) | |
1584 | */ | |
1585 | fscanf(filep, "%d%c\n", &sib1, &character); | |
1586 | fseek(filep, 0, SEEK_SET); | |
1587 | fgets(str, 100, filep); | |
1588 | ch = strchr(str, character); | |
1589 | while (ch != NULL) { | |
1590 | matches++; | |
1591 | ch = strchr(ch+1, character); | |
1592 | } | |
1593 | ||
1594 | fclose(filep); | |
1595 | return matches+1; | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * run func(thread, core, package) in topology order | |
1600 | * skip non-present cpus | |
1601 | */ | |
1602 | ||
1603 | int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *, | |
1604 | struct pkg_data *, struct thread_data *, struct core_data *, | |
1605 | struct pkg_data *), struct thread_data *thread_base, | |
1606 | struct core_data *core_base, struct pkg_data *pkg_base, | |
1607 | struct thread_data *thread_base2, struct core_data *core_base2, | |
1608 | struct pkg_data *pkg_base2) | |
1609 | { | |
1610 | int retval, pkg_no, core_no, thread_no; | |
1611 | ||
1612 | for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { | |
1613 | for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { | |
1614 | for (thread_no = 0; thread_no < | |
1615 | topo.num_threads_per_core; ++thread_no) { | |
1616 | struct thread_data *t, *t2; | |
1617 | struct core_data *c, *c2; | |
1618 | struct pkg_data *p, *p2; | |
1619 | ||
1620 | t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); | |
1621 | ||
1622 | if (cpu_is_not_present(t->cpu_id)) | |
1623 | continue; | |
1624 | ||
1625 | t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no); | |
1626 | ||
1627 | c = GET_CORE(core_base, core_no, pkg_no); | |
1628 | c2 = GET_CORE(core_base2, core_no, pkg_no); | |
1629 | ||
1630 | p = GET_PKG(pkg_base, pkg_no); | |
1631 | p2 = GET_PKG(pkg_base2, pkg_no); | |
1632 | ||
1633 | retval = func(t, c, p, t2, c2, p2); | |
1634 | if (retval) | |
1635 | return retval; | |
1636 | } | |
1637 | } | |
1638 | } | |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * run func(cpu) on every cpu in /proc/stat | |
1644 | * return max_cpu number | |
1645 | */ | |
1646 | int for_all_proc_cpus(int (func)(int)) | |
1647 | { | |
1648 | FILE *fp; | |
1649 | int cpu_num; | |
1650 | int retval; | |
1651 | ||
1652 | fp = fopen_or_die(proc_stat, "r"); | |
1653 | ||
1654 | retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); | |
1655 | if (retval != 0) | |
1656 | err(1, "%s: failed to parse format", proc_stat); | |
1657 | ||
1658 | while (1) { | |
1659 | retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num); | |
1660 | if (retval != 1) | |
1661 | break; | |
1662 | ||
1663 | retval = func(cpu_num); | |
1664 | if (retval) { | |
1665 | fclose(fp); | |
1666 | return(retval); | |
1667 | } | |
1668 | } | |
1669 | fclose(fp); | |
1670 | return 0; | |
1671 | } | |
1672 | ||
1673 | void re_initialize(void) | |
1674 | { | |
1675 | free_all_buffers(); | |
1676 | setup_all_buffers(); | |
1677 | printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus); | |
1678 | } | |
1679 | ||
1680 | ||
1681 | /* | |
1682 | * count_cpus() | |
1683 | * remember the last one seen, it will be the max | |
1684 | */ | |
1685 | int count_cpus(int cpu) | |
1686 | { | |
1687 | if (topo.max_cpu_num < cpu) | |
1688 | topo.max_cpu_num = cpu; | |
1689 | ||
1690 | topo.num_cpus += 1; | |
1691 | return 0; | |
1692 | } | |
1693 | int mark_cpu_present(int cpu) | |
1694 | { | |
1695 | CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); | |
1696 | return 0; | |
1697 | } | |
1698 | ||
1699 | void turbostat_loop() | |
1700 | { | |
1701 | int retval; | |
1702 | int restarted = 0; | |
1703 | ||
1704 | restart: | |
1705 | restarted++; | |
1706 | ||
1707 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); | |
1708 | if (retval < -1) { | |
1709 | exit(retval); | |
1710 | } else if (retval == -1) { | |
1711 | if (restarted > 1) { | |
1712 | exit(retval); | |
1713 | } | |
1714 | re_initialize(); | |
1715 | goto restart; | |
1716 | } | |
1717 | restarted = 0; | |
1718 | gettimeofday(&tv_even, (struct timezone *)NULL); | |
1719 | ||
1720 | while (1) { | |
1721 | if (for_all_proc_cpus(cpu_is_not_present)) { | |
1722 | re_initialize(); | |
1723 | goto restart; | |
1724 | } | |
1725 | sleep(interval_sec); | |
1726 | retval = for_all_cpus(get_counters, ODD_COUNTERS); | |
1727 | if (retval < -1) { | |
1728 | exit(retval); | |
1729 | } else if (retval == -1) { | |
1730 | re_initialize(); | |
1731 | goto restart; | |
1732 | } | |
1733 | gettimeofday(&tv_odd, (struct timezone *)NULL); | |
1734 | timersub(&tv_odd, &tv_even, &tv_delta); | |
1735 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); | |
1736 | compute_average(EVEN_COUNTERS); | |
1737 | format_all_counters(EVEN_COUNTERS); | |
1738 | flush_stdout(); | |
1739 | sleep(interval_sec); | |
1740 | retval = for_all_cpus(get_counters, EVEN_COUNTERS); | |
1741 | if (retval < -1) { | |
1742 | exit(retval); | |
1743 | } else if (retval == -1) { | |
1744 | re_initialize(); | |
1745 | goto restart; | |
1746 | } | |
1747 | gettimeofday(&tv_even, (struct timezone *)NULL); | |
1748 | timersub(&tv_even, &tv_odd, &tv_delta); | |
1749 | for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS); | |
1750 | compute_average(ODD_COUNTERS); | |
1751 | format_all_counters(ODD_COUNTERS); | |
1752 | flush_stdout(); | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | void check_dev_msr() | |
1757 | { | |
1758 | struct stat sb; | |
1759 | char pathname[32]; | |
1760 | ||
1761 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); | |
1762 | if (stat(pathname, &sb)) | |
1763 | if (system("/sbin/modprobe msr > /dev/null 2>&1")) | |
1764 | err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" "); | |
1765 | } | |
1766 | ||
1767 | void check_permissions() | |
1768 | { | |
1769 | struct __user_cap_header_struct cap_header_data; | |
1770 | cap_user_header_t cap_header = &cap_header_data; | |
1771 | struct __user_cap_data_struct cap_data_data; | |
1772 | cap_user_data_t cap_data = &cap_data_data; | |
1773 | extern int capget(cap_user_header_t hdrp, cap_user_data_t datap); | |
1774 | int do_exit = 0; | |
1775 | char pathname[32]; | |
1776 | ||
1777 | /* check for CAP_SYS_RAWIO */ | |
1778 | cap_header->pid = getpid(); | |
1779 | cap_header->version = _LINUX_CAPABILITY_VERSION; | |
1780 | if (capget(cap_header, cap_data) < 0) | |
1781 | err(-6, "capget(2) failed"); | |
1782 | ||
1783 | if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) { | |
1784 | do_exit++; | |
1785 | warnx("capget(CAP_SYS_RAWIO) failed," | |
1786 | " try \"# setcap cap_sys_rawio=ep %s\"", progname); | |
1787 | } | |
1788 | ||
1789 | /* test file permissions */ | |
1790 | sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); | |
1791 | if (euidaccess(pathname, R_OK)) { | |
1792 | do_exit++; | |
1793 | warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr"); | |
1794 | } | |
1795 | ||
1796 | /* if all else fails, thell them to be root */ | |
1797 | if (do_exit) | |
1798 | if (getuid() != 0) | |
1799 | warnx("... or simply run as root"); | |
1800 | ||
1801 | if (do_exit) | |
1802 | exit(-6); | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * NHM adds support for additional MSRs: | |
1807 | * | |
1808 | * MSR_SMI_COUNT 0x00000034 | |
1809 | * | |
1810 | * MSR_PLATFORM_INFO 0x000000ce | |
1811 | * MSR_NHM_SNB_PKG_CST_CFG_CTL 0x000000e2 | |
1812 | * | |
1813 | * MSR_PKG_C3_RESIDENCY 0x000003f8 | |
1814 | * MSR_PKG_C6_RESIDENCY 0x000003f9 | |
1815 | * MSR_CORE_C3_RESIDENCY 0x000003fc | |
1816 | * MSR_CORE_C6_RESIDENCY 0x000003fd | |
1817 | * | |
1818 | * Side effect: | |
1819 | * sets global pkg_cstate_limit to decode MSR_NHM_SNB_PKG_CST_CFG_CTL | |
1820 | */ | |
1821 | int probe_nhm_msrs(unsigned int family, unsigned int model) | |
1822 | { | |
1823 | unsigned long long msr; | |
1824 | unsigned int base_ratio; | |
1825 | int *pkg_cstate_limits; | |
1826 | ||
1827 | if (!genuine_intel) | |
1828 | return 0; | |
1829 | ||
1830 | if (family != 6) | |
1831 | return 0; | |
1832 | ||
1833 | bclk = discover_bclk(family, model); | |
1834 | ||
1835 | switch (model) { | |
1836 | case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */ | |
1837 | case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */ | |
1838 | case 0x1F: /* Core i7 and i5 Processor - Nehalem */ | |
1839 | case 0x25: /* Westmere Client - Clarkdale, Arrandale */ | |
1840 | case 0x2C: /* Westmere EP - Gulftown */ | |
1841 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ | |
1842 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1843 | pkg_cstate_limits = nhm_pkg_cstate_limits; | |
1844 | break; | |
1845 | case 0x2A: /* SNB */ | |
1846 | case 0x2D: /* SNB Xeon */ | |
1847 | case 0x3A: /* IVB */ | |
1848 | case 0x3E: /* IVB Xeon */ | |
1849 | pkg_cstate_limits = snb_pkg_cstate_limits; | |
1850 | break; | |
1851 | case 0x3C: /* HSW */ | |
1852 | case 0x3F: /* HSX */ | |
1853 | case 0x45: /* HSW */ | |
1854 | case 0x46: /* HSW */ | |
1855 | case 0x3D: /* BDW */ | |
1856 | case 0x47: /* BDW */ | |
1857 | case 0x4F: /* BDX */ | |
1858 | case 0x56: /* BDX-DE */ | |
1859 | case 0x4E: /* SKL */ | |
1860 | case 0x5E: /* SKL */ | |
1861 | pkg_cstate_limits = hsw_pkg_cstate_limits; | |
1862 | break; | |
1863 | case 0x37: /* BYT */ | |
1864 | case 0x4D: /* AVN */ | |
1865 | pkg_cstate_limits = slv_pkg_cstate_limits; | |
1866 | break; | |
1867 | case 0x4C: /* AMT */ | |
1868 | pkg_cstate_limits = amt_pkg_cstate_limits; | |
1869 | break; | |
1870 | case 0x57: /* PHI */ | |
1871 | pkg_cstate_limits = phi_pkg_cstate_limits; | |
1872 | break; | |
1873 | default: | |
1874 | return 0; | |
1875 | } | |
1876 | get_msr(base_cpu, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr); | |
1877 | pkg_cstate_limit = pkg_cstate_limits[msr & 0xF]; | |
1878 | ||
1879 | get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); | |
1880 | base_ratio = (msr >> 8) & 0xFF; | |
1881 | ||
1882 | base_hz = base_ratio * bclk * 1000000; | |
1883 | has_base_hz = 1; | |
1884 | return 1; | |
1885 | } | |
1886 | int has_nhm_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1887 | { | |
1888 | switch (model) { | |
1889 | /* Nehalem compatible, but do not include turbo-ratio limit support */ | |
1890 | case 0x2E: /* Nehalem-EX Xeon - Beckton */ | |
1891 | case 0x2F: /* Westmere-EX Xeon - Eagleton */ | |
1892 | return 0; | |
1893 | default: | |
1894 | return 1; | |
1895 | } | |
1896 | } | |
1897 | int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1898 | { | |
1899 | if (!genuine_intel) | |
1900 | return 0; | |
1901 | ||
1902 | if (family != 6) | |
1903 | return 0; | |
1904 | ||
1905 | switch (model) { | |
1906 | case 0x3E: /* IVB Xeon */ | |
1907 | case 0x3F: /* HSW Xeon */ | |
1908 | return 1; | |
1909 | default: | |
1910 | return 0; | |
1911 | } | |
1912 | } | |
1913 | int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1914 | { | |
1915 | if (!genuine_intel) | |
1916 | return 0; | |
1917 | ||
1918 | if (family != 6) | |
1919 | return 0; | |
1920 | ||
1921 | switch (model) { | |
1922 | case 0x3F: /* HSW Xeon */ | |
1923 | return 1; | |
1924 | default: | |
1925 | return 0; | |
1926 | } | |
1927 | } | |
1928 | ||
1929 | int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model) | |
1930 | { | |
1931 | if (!genuine_intel) | |
1932 | return 0; | |
1933 | ||
1934 | if (family != 6) | |
1935 | return 0; | |
1936 | ||
1937 | switch (model) { | |
1938 | case 0x57: /* Knights Landing */ | |
1939 | return 1; | |
1940 | default: | |
1941 | return 0; | |
1942 | } | |
1943 | } | |
1944 | int has_config_tdp(unsigned int family, unsigned int model) | |
1945 | { | |
1946 | if (!genuine_intel) | |
1947 | return 0; | |
1948 | ||
1949 | if (family != 6) | |
1950 | return 0; | |
1951 | ||
1952 | switch (model) { | |
1953 | case 0x3A: /* IVB */ | |
1954 | case 0x3C: /* HSW */ | |
1955 | case 0x3F: /* HSX */ | |
1956 | case 0x45: /* HSW */ | |
1957 | case 0x46: /* HSW */ | |
1958 | case 0x3D: /* BDW */ | |
1959 | case 0x47: /* BDW */ | |
1960 | case 0x4F: /* BDX */ | |
1961 | case 0x56: /* BDX-DE */ | |
1962 | case 0x4E: /* SKL */ | |
1963 | case 0x5E: /* SKL */ | |
1964 | ||
1965 | case 0x57: /* Knights Landing */ | |
1966 | return 1; | |
1967 | default: | |
1968 | return 0; | |
1969 | } | |
1970 | } | |
1971 | ||
1972 | static void | |
1973 | dump_cstate_pstate_config_info(family, model) | |
1974 | { | |
1975 | if (!do_nhm_platform_info) | |
1976 | return; | |
1977 | ||
1978 | dump_nhm_platform_info(); | |
1979 | ||
1980 | if (has_hsw_turbo_ratio_limit(family, model)) | |
1981 | dump_hsw_turbo_ratio_limits(); | |
1982 | ||
1983 | if (has_ivt_turbo_ratio_limit(family, model)) | |
1984 | dump_ivt_turbo_ratio_limits(); | |
1985 | ||
1986 | if (has_nhm_turbo_ratio_limit(family, model)) | |
1987 | dump_nhm_turbo_ratio_limits(); | |
1988 | ||
1989 | if (has_knl_turbo_ratio_limit(family, model)) | |
1990 | dump_knl_turbo_ratio_limits(); | |
1991 | ||
1992 | if (has_config_tdp(family, model)) | |
1993 | dump_config_tdp(); | |
1994 | ||
1995 | dump_nhm_cst_cfg(); | |
1996 | } | |
1997 | ||
1998 | ||
1999 | /* | |
2000 | * print_epb() | |
2001 | * Decode the ENERGY_PERF_BIAS MSR | |
2002 | */ | |
2003 | int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2004 | { | |
2005 | unsigned long long msr; | |
2006 | char *epb_string; | |
2007 | int cpu; | |
2008 | ||
2009 | if (!has_epb) | |
2010 | return 0; | |
2011 | ||
2012 | cpu = t->cpu_id; | |
2013 | ||
2014 | /* EPB is per-package */ | |
2015 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2016 | return 0; | |
2017 | ||
2018 | if (cpu_migrate(cpu)) { | |
2019 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2020 | return -1; | |
2021 | } | |
2022 | ||
2023 | if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr)) | |
2024 | return 0; | |
2025 | ||
2026 | switch (msr & 0xF) { | |
2027 | case ENERGY_PERF_BIAS_PERFORMANCE: | |
2028 | epb_string = "performance"; | |
2029 | break; | |
2030 | case ENERGY_PERF_BIAS_NORMAL: | |
2031 | epb_string = "balanced"; | |
2032 | break; | |
2033 | case ENERGY_PERF_BIAS_POWERSAVE: | |
2034 | epb_string = "powersave"; | |
2035 | break; | |
2036 | default: | |
2037 | epb_string = "custom"; | |
2038 | break; | |
2039 | } | |
2040 | fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string); | |
2041 | ||
2042 | return 0; | |
2043 | } | |
2044 | ||
2045 | /* | |
2046 | * print_perf_limit() | |
2047 | */ | |
2048 | int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2049 | { | |
2050 | unsigned long long msr; | |
2051 | int cpu; | |
2052 | ||
2053 | cpu = t->cpu_id; | |
2054 | ||
2055 | /* per-package */ | |
2056 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2057 | return 0; | |
2058 | ||
2059 | if (cpu_migrate(cpu)) { | |
2060 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2061 | return -1; | |
2062 | } | |
2063 | ||
2064 | if (do_core_perf_limit_reasons) { | |
2065 | get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr); | |
2066 | fprintf(stderr, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2067 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)", | |
2068 | (msr & 1 << 15) ? "bit15, " : "", | |
2069 | (msr & 1 << 14) ? "bit14, " : "", | |
2070 | (msr & 1 << 13) ? "Transitions, " : "", | |
2071 | (msr & 1 << 12) ? "MultiCoreTurbo, " : "", | |
2072 | (msr & 1 << 11) ? "PkgPwrL2, " : "", | |
2073 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2074 | (msr & 1 << 9) ? "CorePwr, " : "", | |
2075 | (msr & 1 << 8) ? "Amps, " : "", | |
2076 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2077 | (msr & 1 << 5) ? "Auto-HWP, " : "", | |
2078 | (msr & 1 << 4) ? "Graphics, " : "", | |
2079 | (msr & 1 << 2) ? "bit2, " : "", | |
2080 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2081 | (msr & 1 << 0) ? "PROCHOT, " : ""); | |
2082 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n", | |
2083 | (msr & 1 << 31) ? "bit31, " : "", | |
2084 | (msr & 1 << 30) ? "bit30, " : "", | |
2085 | (msr & 1 << 29) ? "Transitions, " : "", | |
2086 | (msr & 1 << 28) ? "MultiCoreTurbo, " : "", | |
2087 | (msr & 1 << 27) ? "PkgPwrL2, " : "", | |
2088 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2089 | (msr & 1 << 25) ? "CorePwr, " : "", | |
2090 | (msr & 1 << 24) ? "Amps, " : "", | |
2091 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2092 | (msr & 1 << 21) ? "Auto-HWP, " : "", | |
2093 | (msr & 1 << 20) ? "Graphics, " : "", | |
2094 | (msr & 1 << 18) ? "bit18, " : "", | |
2095 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2096 | (msr & 1 << 16) ? "PROCHOT, " : ""); | |
2097 | ||
2098 | } | |
2099 | if (do_gfx_perf_limit_reasons) { | |
2100 | get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr); | |
2101 | fprintf(stderr, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2102 | fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s)", | |
2103 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2104 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2105 | (msr & 1 << 4) ? "Graphics, " : "", | |
2106 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2107 | (msr & 1 << 8) ? "Amps, " : "", | |
2108 | (msr & 1 << 9) ? "GFXPwr, " : "", | |
2109 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2110 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2111 | fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s)\n", | |
2112 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2113 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2114 | (msr & 1 << 20) ? "Graphics, " : "", | |
2115 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2116 | (msr & 1 << 24) ? "Amps, " : "", | |
2117 | (msr & 1 << 25) ? "GFXPwr, " : "", | |
2118 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2119 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2120 | } | |
2121 | if (do_ring_perf_limit_reasons) { | |
2122 | get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr); | |
2123 | fprintf(stderr, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); | |
2124 | fprintf(stderr, " (Active: %s%s%s%s%s%s)", | |
2125 | (msr & 1 << 0) ? "PROCHOT, " : "", | |
2126 | (msr & 1 << 1) ? "ThermStatus, " : "", | |
2127 | (msr & 1 << 6) ? "VR-Therm, " : "", | |
2128 | (msr & 1 << 8) ? "Amps, " : "", | |
2129 | (msr & 1 << 10) ? "PkgPwrL1, " : "", | |
2130 | (msr & 1 << 11) ? "PkgPwrL2, " : ""); | |
2131 | fprintf(stderr, " (Logged: %s%s%s%s%s%s)\n", | |
2132 | (msr & 1 << 16) ? "PROCHOT, " : "", | |
2133 | (msr & 1 << 17) ? "ThermStatus, " : "", | |
2134 | (msr & 1 << 22) ? "VR-Therm, " : "", | |
2135 | (msr & 1 << 24) ? "Amps, " : "", | |
2136 | (msr & 1 << 26) ? "PkgPwrL1, " : "", | |
2137 | (msr & 1 << 27) ? "PkgPwrL2, " : ""); | |
2138 | } | |
2139 | return 0; | |
2140 | } | |
2141 | ||
2142 | #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */ | |
2143 | #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */ | |
2144 | ||
2145 | double get_tdp(model) | |
2146 | { | |
2147 | unsigned long long msr; | |
2148 | ||
2149 | if (do_rapl & RAPL_PKG_POWER_INFO) | |
2150 | if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr)) | |
2151 | return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units; | |
2152 | ||
2153 | switch (model) { | |
2154 | case 0x37: | |
2155 | case 0x4D: | |
2156 | return 30.0; | |
2157 | default: | |
2158 | return 135.0; | |
2159 | } | |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * rapl_dram_energy_units_probe() | |
2164 | * Energy units are either hard-coded, or come from RAPL Energy Unit MSR. | |
2165 | */ | |
2166 | static double | |
2167 | rapl_dram_energy_units_probe(int model, double rapl_energy_units) | |
2168 | { | |
2169 | /* only called for genuine_intel, family 6 */ | |
2170 | ||
2171 | switch (model) { | |
2172 | case 0x3F: /* HSX */ | |
2173 | case 0x4F: /* BDX */ | |
2174 | case 0x56: /* BDX-DE */ | |
2175 | case 0x57: /* KNL */ | |
2176 | return (rapl_dram_energy_units = 15.3 / 1000000); | |
2177 | default: | |
2178 | return (rapl_energy_units); | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | ||
2183 | /* | |
2184 | * rapl_probe() | |
2185 | * | |
2186 | * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units | |
2187 | */ | |
2188 | void rapl_probe(unsigned int family, unsigned int model) | |
2189 | { | |
2190 | unsigned long long msr; | |
2191 | unsigned int time_unit; | |
2192 | double tdp; | |
2193 | ||
2194 | if (!genuine_intel) | |
2195 | return; | |
2196 | ||
2197 | if (family != 6) | |
2198 | return; | |
2199 | ||
2200 | switch (model) { | |
2201 | case 0x2A: | |
2202 | case 0x3A: | |
2203 | case 0x3C: /* HSW */ | |
2204 | case 0x45: /* HSW */ | |
2205 | case 0x46: /* HSW */ | |
2206 | case 0x3D: /* BDW */ | |
2207 | case 0x47: /* BDW */ | |
2208 | do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO; | |
2209 | break; | |
2210 | case 0x4E: /* SKL */ | |
2211 | case 0x5E: /* SKL */ | |
2212 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; | |
2213 | break; | |
2214 | case 0x3F: /* HSX */ | |
2215 | case 0x4F: /* BDX */ | |
2216 | case 0x56: /* BDX-DE */ | |
2217 | case 0x57: /* KNL */ | |
2218 | do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; | |
2219 | break; | |
2220 | case 0x2D: | |
2221 | case 0x3E: | |
2222 | 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; | |
2223 | break; | |
2224 | case 0x37: /* BYT */ | |
2225 | case 0x4D: /* AVN */ | |
2226 | do_rapl = RAPL_PKG | RAPL_CORES ; | |
2227 | break; | |
2228 | default: | |
2229 | return; | |
2230 | } | |
2231 | ||
2232 | /* units on package 0, verify later other packages match */ | |
2233 | if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr)) | |
2234 | return; | |
2235 | ||
2236 | rapl_power_units = 1.0 / (1 << (msr & 0xF)); | |
2237 | if (model == 0x37) | |
2238 | rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000; | |
2239 | else | |
2240 | rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F)); | |
2241 | ||
2242 | rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units); | |
2243 | ||
2244 | time_unit = msr >> 16 & 0xF; | |
2245 | if (time_unit == 0) | |
2246 | time_unit = 0xA; | |
2247 | ||
2248 | rapl_time_units = 1.0 / (1 << (time_unit)); | |
2249 | ||
2250 | tdp = get_tdp(model); | |
2251 | ||
2252 | rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp; | |
2253 | if (debug) | |
2254 | fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp); | |
2255 | ||
2256 | return; | |
2257 | } | |
2258 | ||
2259 | void perf_limit_reasons_probe(family, model) | |
2260 | { | |
2261 | if (!genuine_intel) | |
2262 | return; | |
2263 | ||
2264 | if (family != 6) | |
2265 | return; | |
2266 | ||
2267 | switch (model) { | |
2268 | case 0x3C: /* HSW */ | |
2269 | case 0x45: /* HSW */ | |
2270 | case 0x46: /* HSW */ | |
2271 | do_gfx_perf_limit_reasons = 1; | |
2272 | case 0x3F: /* HSX */ | |
2273 | do_core_perf_limit_reasons = 1; | |
2274 | do_ring_perf_limit_reasons = 1; | |
2275 | default: | |
2276 | return; | |
2277 | } | |
2278 | } | |
2279 | ||
2280 | int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2281 | { | |
2282 | unsigned long long msr; | |
2283 | unsigned int dts; | |
2284 | int cpu; | |
2285 | ||
2286 | if (!(do_dts || do_ptm)) | |
2287 | return 0; | |
2288 | ||
2289 | cpu = t->cpu_id; | |
2290 | ||
2291 | /* DTS is per-core, no need to print for each thread */ | |
2292 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) | |
2293 | return 0; | |
2294 | ||
2295 | if (cpu_migrate(cpu)) { | |
2296 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2297 | return -1; | |
2298 | } | |
2299 | ||
2300 | if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) { | |
2301 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) | |
2302 | return 0; | |
2303 | ||
2304 | dts = (msr >> 16) & 0x7F; | |
2305 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n", | |
2306 | cpu, msr, tcc_activation_temp - dts); | |
2307 | ||
2308 | #ifdef THERM_DEBUG | |
2309 | if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr)) | |
2310 | return 0; | |
2311 | ||
2312 | dts = (msr >> 16) & 0x7F; | |
2313 | dts2 = (msr >> 8) & 0x7F; | |
2314 | fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2315 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2316 | #endif | |
2317 | } | |
2318 | ||
2319 | ||
2320 | if (do_dts) { | |
2321 | unsigned int resolution; | |
2322 | ||
2323 | if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) | |
2324 | return 0; | |
2325 | ||
2326 | dts = (msr >> 16) & 0x7F; | |
2327 | resolution = (msr >> 27) & 0xF; | |
2328 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n", | |
2329 | cpu, msr, tcc_activation_temp - dts, resolution); | |
2330 | ||
2331 | #ifdef THERM_DEBUG | |
2332 | if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr)) | |
2333 | return 0; | |
2334 | ||
2335 | dts = (msr >> 16) & 0x7F; | |
2336 | dts2 = (msr >> 8) & 0x7F; | |
2337 | fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", | |
2338 | cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2); | |
2339 | #endif | |
2340 | } | |
2341 | ||
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | void print_power_limit_msr(int cpu, unsigned long long msr, char *label) | |
2346 | { | |
2347 | fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n", | |
2348 | cpu, label, | |
2349 | ((msr >> 15) & 1) ? "EN" : "DIS", | |
2350 | ((msr >> 0) & 0x7FFF) * rapl_power_units, | |
2351 | (1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units, | |
2352 | (((msr >> 16) & 1) ? "EN" : "DIS")); | |
2353 | ||
2354 | return; | |
2355 | } | |
2356 | ||
2357 | int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2358 | { | |
2359 | unsigned long long msr; | |
2360 | int cpu; | |
2361 | ||
2362 | if (!do_rapl) | |
2363 | return 0; | |
2364 | ||
2365 | /* RAPL counters are per package, so print only for 1st thread/package */ | |
2366 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2367 | return 0; | |
2368 | ||
2369 | cpu = t->cpu_id; | |
2370 | if (cpu_migrate(cpu)) { | |
2371 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2372 | return -1; | |
2373 | } | |
2374 | ||
2375 | if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr)) | |
2376 | return -1; | |
2377 | ||
2378 | if (debug) { | |
2379 | fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx " | |
2380 | "(%f Watts, %f Joules, %f sec.)\n", cpu, msr, | |
2381 | rapl_power_units, rapl_energy_units, rapl_time_units); | |
2382 | } | |
2383 | if (do_rapl & RAPL_PKG_POWER_INFO) { | |
2384 | ||
2385 | if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr)) | |
2386 | return -5; | |
2387 | ||
2388 | ||
2389 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", | |
2390 | cpu, msr, | |
2391 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2392 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2393 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2394 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
2395 | ||
2396 | } | |
2397 | if (do_rapl & RAPL_PKG) { | |
2398 | ||
2399 | if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr)) | |
2400 | return -9; | |
2401 | ||
2402 | fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2403 | cpu, msr, (msr >> 63) & 1 ? "": "UN"); | |
2404 | ||
2405 | print_power_limit_msr(cpu, msr, "PKG Limit #1"); | |
2406 | fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n", | |
2407 | cpu, | |
2408 | ((msr >> 47) & 1) ? "EN" : "DIS", | |
2409 | ((msr >> 32) & 0x7FFF) * rapl_power_units, | |
2410 | (1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units, | |
2411 | ((msr >> 48) & 1) ? "EN" : "DIS"); | |
2412 | } | |
2413 | ||
2414 | if (do_rapl & RAPL_DRAM_POWER_INFO) { | |
2415 | if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr)) | |
2416 | return -6; | |
2417 | ||
2418 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", | |
2419 | cpu, msr, | |
2420 | ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2421 | ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2422 | ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, | |
2423 | ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); | |
2424 | } | |
2425 | if (do_rapl & RAPL_DRAM) { | |
2426 | if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr)) | |
2427 | return -9; | |
2428 | fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2429 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2430 | ||
2431 | print_power_limit_msr(cpu, msr, "DRAM Limit"); | |
2432 | } | |
2433 | if (do_rapl & RAPL_CORE_POLICY) { | |
2434 | if (debug) { | |
2435 | if (get_msr(cpu, MSR_PP0_POLICY, &msr)) | |
2436 | return -7; | |
2437 | ||
2438 | fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF); | |
2439 | } | |
2440 | } | |
2441 | if (do_rapl & RAPL_CORES) { | |
2442 | if (debug) { | |
2443 | ||
2444 | if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr)) | |
2445 | return -9; | |
2446 | fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2447 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2448 | print_power_limit_msr(cpu, msr, "Cores Limit"); | |
2449 | } | |
2450 | } | |
2451 | if (do_rapl & RAPL_GFX) { | |
2452 | if (debug) { | |
2453 | if (get_msr(cpu, MSR_PP1_POLICY, &msr)) | |
2454 | return -8; | |
2455 | ||
2456 | fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF); | |
2457 | ||
2458 | if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr)) | |
2459 | return -9; | |
2460 | fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n", | |
2461 | cpu, msr, (msr >> 31) & 1 ? "": "UN"); | |
2462 | print_power_limit_msr(cpu, msr, "GFX Limit"); | |
2463 | } | |
2464 | } | |
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | /* | |
2469 | * SNB adds support for additional MSRs: | |
2470 | * | |
2471 | * MSR_PKG_C7_RESIDENCY 0x000003fa | |
2472 | * MSR_CORE_C7_RESIDENCY 0x000003fe | |
2473 | * MSR_PKG_C2_RESIDENCY 0x0000060d | |
2474 | */ | |
2475 | ||
2476 | int has_snb_msrs(unsigned int family, unsigned int model) | |
2477 | { | |
2478 | if (!genuine_intel) | |
2479 | return 0; | |
2480 | ||
2481 | switch (model) { | |
2482 | case 0x2A: | |
2483 | case 0x2D: | |
2484 | case 0x3A: /* IVB */ | |
2485 | case 0x3E: /* IVB Xeon */ | |
2486 | case 0x3C: /* HSW */ | |
2487 | case 0x3F: /* HSW */ | |
2488 | case 0x45: /* HSW */ | |
2489 | case 0x46: /* HSW */ | |
2490 | case 0x3D: /* BDW */ | |
2491 | case 0x47: /* BDW */ | |
2492 | case 0x4F: /* BDX */ | |
2493 | case 0x56: /* BDX-DE */ | |
2494 | case 0x4E: /* SKL */ | |
2495 | case 0x5E: /* SKL */ | |
2496 | return 1; | |
2497 | } | |
2498 | return 0; | |
2499 | } | |
2500 | ||
2501 | /* | |
2502 | * HSW adds support for additional MSRs: | |
2503 | * | |
2504 | * MSR_PKG_C8_RESIDENCY 0x00000630 | |
2505 | * MSR_PKG_C9_RESIDENCY 0x00000631 | |
2506 | * MSR_PKG_C10_RESIDENCY 0x00000632 | |
2507 | */ | |
2508 | int has_hsw_msrs(unsigned int family, unsigned int model) | |
2509 | { | |
2510 | if (!genuine_intel) | |
2511 | return 0; | |
2512 | ||
2513 | switch (model) { | |
2514 | case 0x45: /* HSW */ | |
2515 | case 0x3D: /* BDW */ | |
2516 | case 0x4E: /* SKL */ | |
2517 | case 0x5E: /* SKL */ | |
2518 | return 1; | |
2519 | } | |
2520 | return 0; | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * SKL adds support for additional MSRS: | |
2525 | * | |
2526 | * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658 | |
2527 | * MSR_PKG_ANY_CORE_C0_RES 0x00000659 | |
2528 | * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A | |
2529 | * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B | |
2530 | */ | |
2531 | int has_skl_msrs(unsigned int family, unsigned int model) | |
2532 | { | |
2533 | if (!genuine_intel) | |
2534 | return 0; | |
2535 | ||
2536 | switch (model) { | |
2537 | case 0x4E: /* SKL */ | |
2538 | case 0x5E: /* SKL */ | |
2539 | return 1; | |
2540 | } | |
2541 | return 0; | |
2542 | } | |
2543 | ||
2544 | ||
2545 | ||
2546 | int is_slm(unsigned int family, unsigned int model) | |
2547 | { | |
2548 | if (!genuine_intel) | |
2549 | return 0; | |
2550 | switch (model) { | |
2551 | case 0x37: /* BYT */ | |
2552 | case 0x4D: /* AVN */ | |
2553 | return 1; | |
2554 | } | |
2555 | return 0; | |
2556 | } | |
2557 | ||
2558 | int is_knl(unsigned int family, unsigned int model) | |
2559 | { | |
2560 | if (!genuine_intel) | |
2561 | return 0; | |
2562 | switch (model) { | |
2563 | case 0x57: /* KNL */ | |
2564 | return 1; | |
2565 | } | |
2566 | return 0; | |
2567 | } | |
2568 | ||
2569 | unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model) | |
2570 | { | |
2571 | if (is_knl(family, model)) | |
2572 | return 1024; | |
2573 | return 1; | |
2574 | } | |
2575 | ||
2576 | #define SLM_BCLK_FREQS 5 | |
2577 | double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0}; | |
2578 | ||
2579 | double slm_bclk(void) | |
2580 | { | |
2581 | unsigned long long msr = 3; | |
2582 | unsigned int i; | |
2583 | double freq; | |
2584 | ||
2585 | if (get_msr(base_cpu, MSR_FSB_FREQ, &msr)) | |
2586 | fprintf(stderr, "SLM BCLK: unknown\n"); | |
2587 | ||
2588 | i = msr & 0xf; | |
2589 | if (i >= SLM_BCLK_FREQS) { | |
2590 | fprintf(stderr, "SLM BCLK[%d] invalid\n", i); | |
2591 | msr = 3; | |
2592 | } | |
2593 | freq = slm_freq_table[i]; | |
2594 | ||
2595 | fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq); | |
2596 | ||
2597 | return freq; | |
2598 | } | |
2599 | ||
2600 | double discover_bclk(unsigned int family, unsigned int model) | |
2601 | { | |
2602 | if (has_snb_msrs(family, model)) | |
2603 | return 100.00; | |
2604 | else if (is_slm(family, model)) | |
2605 | return slm_bclk(); | |
2606 | else | |
2607 | return 133.33; | |
2608 | } | |
2609 | ||
2610 | /* | |
2611 | * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where | |
2612 | * the Thermal Control Circuit (TCC) activates. | |
2613 | * This is usually equal to tjMax. | |
2614 | * | |
2615 | * Older processors do not have this MSR, so there we guess, | |
2616 | * but also allow cmdline over-ride with -T. | |
2617 | * | |
2618 | * Several MSR temperature values are in units of degrees-C | |
2619 | * below this value, including the Digital Thermal Sensor (DTS), | |
2620 | * Package Thermal Management Sensor (PTM), and thermal event thresholds. | |
2621 | */ | |
2622 | int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p) | |
2623 | { | |
2624 | unsigned long long msr; | |
2625 | unsigned int target_c_local; | |
2626 | int cpu; | |
2627 | ||
2628 | /* tcc_activation_temp is used only for dts or ptm */ | |
2629 | if (!(do_dts || do_ptm)) | |
2630 | return 0; | |
2631 | ||
2632 | /* this is a per-package concept */ | |
2633 | if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) | |
2634 | return 0; | |
2635 | ||
2636 | cpu = t->cpu_id; | |
2637 | if (cpu_migrate(cpu)) { | |
2638 | fprintf(stderr, "Could not migrate to CPU %d\n", cpu); | |
2639 | return -1; | |
2640 | } | |
2641 | ||
2642 | if (tcc_activation_temp_override != 0) { | |
2643 | tcc_activation_temp = tcc_activation_temp_override; | |
2644 | fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n", | |
2645 | cpu, tcc_activation_temp); | |
2646 | return 0; | |
2647 | } | |
2648 | ||
2649 | /* Temperature Target MSR is Nehalem and newer only */ | |
2650 | if (!do_nhm_platform_info) | |
2651 | goto guess; | |
2652 | ||
2653 | if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr)) | |
2654 | goto guess; | |
2655 | ||
2656 | target_c_local = (msr >> 16) & 0xFF; | |
2657 | ||
2658 | if (debug) | |
2659 | fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n", | |
2660 | cpu, msr, target_c_local); | |
2661 | ||
2662 | if (!target_c_local) | |
2663 | goto guess; | |
2664 | ||
2665 | tcc_activation_temp = target_c_local; | |
2666 | ||
2667 | return 0; | |
2668 | ||
2669 | guess: | |
2670 | tcc_activation_temp = TJMAX_DEFAULT; | |
2671 | fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n", | |
2672 | cpu, tcc_activation_temp); | |
2673 | ||
2674 | return 0; | |
2675 | } | |
2676 | void process_cpuid() | |
2677 | { | |
2678 | unsigned int eax, ebx, ecx, edx, max_level; | |
2679 | unsigned int fms, family, model, stepping; | |
2680 | ||
2681 | eax = ebx = ecx = edx = 0; | |
2682 | ||
2683 | __get_cpuid(0, &max_level, &ebx, &ecx, &edx); | |
2684 | ||
2685 | if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e) | |
2686 | genuine_intel = 1; | |
2687 | ||
2688 | if (debug) | |
2689 | fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ", | |
2690 | (char *)&ebx, (char *)&edx, (char *)&ecx); | |
2691 | ||
2692 | __get_cpuid(1, &fms, &ebx, &ecx, &edx); | |
2693 | family = (fms >> 8) & 0xf; | |
2694 | model = (fms >> 4) & 0xf; | |
2695 | stepping = fms & 0xf; | |
2696 | if (family == 6 || family == 0xf) | |
2697 | model += ((fms >> 16) & 0xf) << 4; | |
2698 | ||
2699 | if (debug) | |
2700 | fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n", | |
2701 | max_level, family, model, stepping, family, model, stepping); | |
2702 | ||
2703 | if (!(edx & (1 << 5))) | |
2704 | errx(1, "CPUID: no MSR"); | |
2705 | ||
2706 | /* | |
2707 | * check max extended function levels of CPUID. | |
2708 | * This is needed to check for invariant TSC. | |
2709 | * This check is valid for both Intel and AMD. | |
2710 | */ | |
2711 | ebx = ecx = edx = 0; | |
2712 | __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx); | |
2713 | ||
2714 | if (max_level >= 0x80000007) { | |
2715 | ||
2716 | /* | |
2717 | * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8 | |
2718 | * this check is valid for both Intel and AMD | |
2719 | */ | |
2720 | __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx); | |
2721 | has_invariant_tsc = edx & (1 << 8); | |
2722 | } | |
2723 | ||
2724 | /* | |
2725 | * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0 | |
2726 | * this check is valid for both Intel and AMD | |
2727 | */ | |
2728 | ||
2729 | __get_cpuid(0x6, &eax, &ebx, &ecx, &edx); | |
2730 | has_aperf = ecx & (1 << 0); | |
2731 | do_dts = eax & (1 << 0); | |
2732 | do_ptm = eax & (1 << 6); | |
2733 | has_epb = ecx & (1 << 3); | |
2734 | ||
2735 | if (debug) | |
2736 | fprintf(stderr, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sEPB\n", | |
2737 | has_aperf ? "" : "No ", | |
2738 | do_dts ? "" : "No ", | |
2739 | do_ptm ? "" : "No ", | |
2740 | has_epb ? "" : "No "); | |
2741 | ||
2742 | if (max_level > 0x15) { | |
2743 | unsigned int eax_crystal; | |
2744 | unsigned int ebx_tsc; | |
2745 | ||
2746 | /* | |
2747 | * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz | |
2748 | */ | |
2749 | eax_crystal = ebx_tsc = crystal_hz = edx = 0; | |
2750 | __get_cpuid(0x15, &eax_crystal, &ebx_tsc, &crystal_hz, &edx); | |
2751 | ||
2752 | if (ebx_tsc != 0) { | |
2753 | ||
2754 | if (debug && (ebx != 0)) | |
2755 | fprintf(stderr, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n", | |
2756 | eax_crystal, ebx_tsc, crystal_hz); | |
2757 | ||
2758 | if (crystal_hz == 0) | |
2759 | switch(model) { | |
2760 | case 0x4E: /* SKL */ | |
2761 | case 0x5E: /* SKL */ | |
2762 | crystal_hz = 24000000; /* 24 MHz */ | |
2763 | break; | |
2764 | default: | |
2765 | crystal_hz = 0; | |
2766 | } | |
2767 | ||
2768 | if (crystal_hz) { | |
2769 | tsc_hz = (unsigned long long) crystal_hz * ebx_tsc / eax_crystal; | |
2770 | if (debug) | |
2771 | fprintf(stderr, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n", | |
2772 | tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal); | |
2773 | } | |
2774 | } | |
2775 | } | |
2776 | ||
2777 | if (has_aperf) | |
2778 | aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model); | |
2779 | ||
2780 | do_nhm_platform_info = do_nhm_cstates = do_smi = probe_nhm_msrs(family, model); | |
2781 | do_snb_cstates = has_snb_msrs(family, model); | |
2782 | do_pc2 = do_snb_cstates && (pkg_cstate_limit >= PCL__2); | |
2783 | do_pc3 = (pkg_cstate_limit >= PCL__3); | |
2784 | do_pc6 = (pkg_cstate_limit >= PCL__6); | |
2785 | do_pc7 = do_snb_cstates && (pkg_cstate_limit >= PCL__7); | |
2786 | do_c8_c9_c10 = has_hsw_msrs(family, model); | |
2787 | do_skl_residency = has_skl_msrs(family, model); | |
2788 | do_slm_cstates = is_slm(family, model); | |
2789 | do_knl_cstates = is_knl(family, model); | |
2790 | ||
2791 | rapl_probe(family, model); | |
2792 | perf_limit_reasons_probe(family, model); | |
2793 | ||
2794 | if (debug) | |
2795 | dump_cstate_pstate_config_info(); | |
2796 | ||
2797 | if (has_skl_msrs(family, model)) | |
2798 | calculate_tsc_tweak(); | |
2799 | ||
2800 | return; | |
2801 | } | |
2802 | ||
2803 | void help() | |
2804 | { | |
2805 | fprintf(stderr, | |
2806 | "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n" | |
2807 | "\n" | |
2808 | "Turbostat forks the specified COMMAND and prints statistics\n" | |
2809 | "when COMMAND completes.\n" | |
2810 | "If no COMMAND is specified, turbostat wakes every 5-seconds\n" | |
2811 | "to print statistics, until interrupted.\n" | |
2812 | "--debug run in \"debug\" mode\n" | |
2813 | "--interval sec Override default 5-second measurement interval\n" | |
2814 | "--help print this help message\n" | |
2815 | "--counter msr print 32-bit counter at address \"msr\"\n" | |
2816 | "--Counter msr print 64-bit Counter at address \"msr\"\n" | |
2817 | "--msr msr print 32-bit value at address \"msr\"\n" | |
2818 | "--MSR msr print 64-bit Value at address \"msr\"\n" | |
2819 | "--version print version information\n" | |
2820 | "\n" | |
2821 | "For more help, run \"man turbostat\"\n"); | |
2822 | } | |
2823 | ||
2824 | ||
2825 | /* | |
2826 | * in /dev/cpu/ return success for names that are numbers | |
2827 | * ie. filter out ".", "..", "microcode". | |
2828 | */ | |
2829 | int dir_filter(const struct dirent *dirp) | |
2830 | { | |
2831 | if (isdigit(dirp->d_name[0])) | |
2832 | return 1; | |
2833 | else | |
2834 | return 0; | |
2835 | } | |
2836 | ||
2837 | int open_dev_cpu_msr(int dummy1) | |
2838 | { | |
2839 | return 0; | |
2840 | } | |
2841 | ||
2842 | void topology_probe() | |
2843 | { | |
2844 | int i; | |
2845 | int max_core_id = 0; | |
2846 | int max_package_id = 0; | |
2847 | int max_siblings = 0; | |
2848 | struct cpu_topology { | |
2849 | int core_id; | |
2850 | int physical_package_id; | |
2851 | } *cpus; | |
2852 | ||
2853 | /* Initialize num_cpus, max_cpu_num */ | |
2854 | topo.num_cpus = 0; | |
2855 | topo.max_cpu_num = 0; | |
2856 | for_all_proc_cpus(count_cpus); | |
2857 | if (!summary_only && topo.num_cpus > 1) | |
2858 | show_cpu = 1; | |
2859 | ||
2860 | if (debug > 1) | |
2861 | fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num); | |
2862 | ||
2863 | cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology)); | |
2864 | if (cpus == NULL) | |
2865 | err(1, "calloc cpus"); | |
2866 | ||
2867 | /* | |
2868 | * Allocate and initialize cpu_present_set | |
2869 | */ | |
2870 | cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
2871 | if (cpu_present_set == NULL) | |
2872 | err(3, "CPU_ALLOC"); | |
2873 | cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); | |
2874 | CPU_ZERO_S(cpu_present_setsize, cpu_present_set); | |
2875 | for_all_proc_cpus(mark_cpu_present); | |
2876 | ||
2877 | /* | |
2878 | * Allocate and initialize cpu_affinity_set | |
2879 | */ | |
2880 | cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1)); | |
2881 | if (cpu_affinity_set == NULL) | |
2882 | err(3, "CPU_ALLOC"); | |
2883 | cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); | |
2884 | CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); | |
2885 | ||
2886 | ||
2887 | /* | |
2888 | * For online cpus | |
2889 | * find max_core_id, max_package_id | |
2890 | */ | |
2891 | for (i = 0; i <= topo.max_cpu_num; ++i) { | |
2892 | int siblings; | |
2893 | ||
2894 | if (cpu_is_not_present(i)) { | |
2895 | if (debug > 1) | |
2896 | fprintf(stderr, "cpu%d NOT PRESENT\n", i); | |
2897 | continue; | |
2898 | } | |
2899 | cpus[i].core_id = get_core_id(i); | |
2900 | if (cpus[i].core_id > max_core_id) | |
2901 | max_core_id = cpus[i].core_id; | |
2902 | ||
2903 | cpus[i].physical_package_id = get_physical_package_id(i); | |
2904 | if (cpus[i].physical_package_id > max_package_id) | |
2905 | max_package_id = cpus[i].physical_package_id; | |
2906 | ||
2907 | siblings = get_num_ht_siblings(i); | |
2908 | if (siblings > max_siblings) | |
2909 | max_siblings = siblings; | |
2910 | if (debug > 1) | |
2911 | fprintf(stderr, "cpu %d pkg %d core %d\n", | |
2912 | i, cpus[i].physical_package_id, cpus[i].core_id); | |
2913 | } | |
2914 | topo.num_cores_per_pkg = max_core_id + 1; | |
2915 | if (debug > 1) | |
2916 | fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n", | |
2917 | max_core_id, topo.num_cores_per_pkg); | |
2918 | if (debug && !summary_only && topo.num_cores_per_pkg > 1) | |
2919 | show_core = 1; | |
2920 | ||
2921 | topo.num_packages = max_package_id + 1; | |
2922 | if (debug > 1) | |
2923 | fprintf(stderr, "max_package_id %d, sizing for %d packages\n", | |
2924 | max_package_id, topo.num_packages); | |
2925 | if (debug && !summary_only && topo.num_packages > 1) | |
2926 | show_pkg = 1; | |
2927 | ||
2928 | topo.num_threads_per_core = max_siblings; | |
2929 | if (debug > 1) | |
2930 | fprintf(stderr, "max_siblings %d\n", max_siblings); | |
2931 | ||
2932 | free(cpus); | |
2933 | } | |
2934 | ||
2935 | void | |
2936 | allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p) | |
2937 | { | |
2938 | int i; | |
2939 | ||
2940 | *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg * | |
2941 | topo.num_packages, sizeof(struct thread_data)); | |
2942 | if (*t == NULL) | |
2943 | goto error; | |
2944 | ||
2945 | for (i = 0; i < topo.num_threads_per_core * | |
2946 | topo.num_cores_per_pkg * topo.num_packages; i++) | |
2947 | (*t)[i].cpu_id = -1; | |
2948 | ||
2949 | *c = calloc(topo.num_cores_per_pkg * topo.num_packages, | |
2950 | sizeof(struct core_data)); | |
2951 | if (*c == NULL) | |
2952 | goto error; | |
2953 | ||
2954 | for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++) | |
2955 | (*c)[i].core_id = -1; | |
2956 | ||
2957 | *p = calloc(topo.num_packages, sizeof(struct pkg_data)); | |
2958 | if (*p == NULL) | |
2959 | goto error; | |
2960 | ||
2961 | for (i = 0; i < topo.num_packages; i++) | |
2962 | (*p)[i].package_id = i; | |
2963 | ||
2964 | return; | |
2965 | error: | |
2966 | err(1, "calloc counters"); | |
2967 | } | |
2968 | /* | |
2969 | * init_counter() | |
2970 | * | |
2971 | * set cpu_id, core_num, pkg_num | |
2972 | * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE | |
2973 | * | |
2974 | * increment topo.num_cores when 1st core in pkg seen | |
2975 | */ | |
2976 | void init_counter(struct thread_data *thread_base, struct core_data *core_base, | |
2977 | struct pkg_data *pkg_base, int thread_num, int core_num, | |
2978 | int pkg_num, int cpu_id) | |
2979 | { | |
2980 | struct thread_data *t; | |
2981 | struct core_data *c; | |
2982 | struct pkg_data *p; | |
2983 | ||
2984 | t = GET_THREAD(thread_base, thread_num, core_num, pkg_num); | |
2985 | c = GET_CORE(core_base, core_num, pkg_num); | |
2986 | p = GET_PKG(pkg_base, pkg_num); | |
2987 | ||
2988 | t->cpu_id = cpu_id; | |
2989 | if (thread_num == 0) { | |
2990 | t->flags |= CPU_IS_FIRST_THREAD_IN_CORE; | |
2991 | if (cpu_is_first_core_in_package(cpu_id)) | |
2992 | t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE; | |
2993 | } | |
2994 | ||
2995 | c->core_id = core_num; | |
2996 | p->package_id = pkg_num; | |
2997 | } | |
2998 | ||
2999 | ||
3000 | int initialize_counters(int cpu_id) | |
3001 | { | |
3002 | int my_thread_id, my_core_id, my_package_id; | |
3003 | ||
3004 | my_package_id = get_physical_package_id(cpu_id); | |
3005 | my_core_id = get_core_id(cpu_id); | |
3006 | my_thread_id = get_cpu_position_in_core(cpu_id); | |
3007 | if (!my_thread_id) | |
3008 | topo.num_cores++; | |
3009 | ||
3010 | init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
3011 | init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); | |
3012 | return 0; | |
3013 | } | |
3014 | ||
3015 | void allocate_output_buffer() | |
3016 | { | |
3017 | output_buffer = calloc(1, (1 + topo.num_cpus) * 1024); | |
3018 | outp = output_buffer; | |
3019 | if (outp == NULL) | |
3020 | err(-1, "calloc output buffer"); | |
3021 | } | |
3022 | ||
3023 | void setup_all_buffers(void) | |
3024 | { | |
3025 | topology_probe(); | |
3026 | allocate_counters(&thread_even, &core_even, &package_even); | |
3027 | allocate_counters(&thread_odd, &core_odd, &package_odd); | |
3028 | allocate_output_buffer(); | |
3029 | for_all_proc_cpus(initialize_counters); | |
3030 | } | |
3031 | ||
3032 | void set_base_cpu(void) | |
3033 | { | |
3034 | base_cpu = sched_getcpu(); | |
3035 | if (base_cpu < 0) | |
3036 | err(-ENODEV, "No valid cpus found"); | |
3037 | ||
3038 | if (debug > 1) | |
3039 | fprintf(stderr, "base_cpu = %d\n", base_cpu); | |
3040 | } | |
3041 | ||
3042 | void turbostat_init() | |
3043 | { | |
3044 | setup_all_buffers(); | |
3045 | set_base_cpu(); | |
3046 | check_dev_msr(); | |
3047 | check_permissions(); | |
3048 | process_cpuid(); | |
3049 | ||
3050 | ||
3051 | if (debug) | |
3052 | for_all_cpus(print_epb, ODD_COUNTERS); | |
3053 | ||
3054 | if (debug) | |
3055 | for_all_cpus(print_perf_limit, ODD_COUNTERS); | |
3056 | ||
3057 | if (debug) | |
3058 | for_all_cpus(print_rapl, ODD_COUNTERS); | |
3059 | ||
3060 | for_all_cpus(set_temperature_target, ODD_COUNTERS); | |
3061 | ||
3062 | if (debug) | |
3063 | for_all_cpus(print_thermal, ODD_COUNTERS); | |
3064 | } | |
3065 | ||
3066 | int fork_it(char **argv) | |
3067 | { | |
3068 | pid_t child_pid; | |
3069 | int status; | |
3070 | ||
3071 | status = for_all_cpus(get_counters, EVEN_COUNTERS); | |
3072 | if (status) | |
3073 | exit(status); | |
3074 | /* clear affinity side-effect of get_counters() */ | |
3075 | sched_setaffinity(0, cpu_present_setsize, cpu_present_set); | |
3076 | gettimeofday(&tv_even, (struct timezone *)NULL); | |
3077 | ||
3078 | child_pid = fork(); | |
3079 | if (!child_pid) { | |
3080 | /* child */ | |
3081 | execvp(argv[0], argv); | |
3082 | } else { | |
3083 | ||
3084 | /* parent */ | |
3085 | if (child_pid == -1) | |
3086 | err(1, "fork"); | |
3087 | ||
3088 | signal(SIGINT, SIG_IGN); | |
3089 | signal(SIGQUIT, SIG_IGN); | |
3090 | if (waitpid(child_pid, &status, 0) == -1) | |
3091 | err(status, "waitpid"); | |
3092 | } | |
3093 | /* | |
3094 | * n.b. fork_it() does not check for errors from for_all_cpus() | |
3095 | * because re-starting is problematic when forking | |
3096 | */ | |
3097 | for_all_cpus(get_counters, ODD_COUNTERS); | |
3098 | gettimeofday(&tv_odd, (struct timezone *)NULL); | |
3099 | timersub(&tv_odd, &tv_even, &tv_delta); | |
3100 | for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); | |
3101 | compute_average(EVEN_COUNTERS); | |
3102 | format_all_counters(EVEN_COUNTERS); | |
3103 | flush_stderr(); | |
3104 | ||
3105 | fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0); | |
3106 | ||
3107 | return status; | |
3108 | } | |
3109 | ||
3110 | int get_and_dump_counters(void) | |
3111 | { | |
3112 | int status; | |
3113 | ||
3114 | status = for_all_cpus(get_counters, ODD_COUNTERS); | |
3115 | if (status) | |
3116 | return status; | |
3117 | ||
3118 | status = for_all_cpus(dump_counters, ODD_COUNTERS); | |
3119 | if (status) | |
3120 | return status; | |
3121 | ||
3122 | flush_stdout(); | |
3123 | ||
3124 | return status; | |
3125 | } | |
3126 | ||
3127 | void print_version() { | |
3128 | fprintf(stderr, "turbostat version 4.8 26-Sep, 2015" | |
3129 | " - Len Brown <lenb@kernel.org>\n"); | |
3130 | } | |
3131 | ||
3132 | void cmdline(int argc, char **argv) | |
3133 | { | |
3134 | int opt; | |
3135 | int option_index = 0; | |
3136 | static struct option long_options[] = { | |
3137 | {"Counter", required_argument, 0, 'C'}, | |
3138 | {"counter", required_argument, 0, 'c'}, | |
3139 | {"Dump", no_argument, 0, 'D'}, | |
3140 | {"debug", no_argument, 0, 'd'}, | |
3141 | {"interval", required_argument, 0, 'i'}, | |
3142 | {"help", no_argument, 0, 'h'}, | |
3143 | {"Joules", no_argument, 0, 'J'}, | |
3144 | {"MSR", required_argument, 0, 'M'}, | |
3145 | {"msr", required_argument, 0, 'm'}, | |
3146 | {"Package", no_argument, 0, 'p'}, | |
3147 | {"processor", no_argument, 0, 'p'}, | |
3148 | {"Summary", no_argument, 0, 'S'}, | |
3149 | {"TCC", required_argument, 0, 'T'}, | |
3150 | {"version", no_argument, 0, 'v' }, | |
3151 | {0, 0, 0, 0 } | |
3152 | }; | |
3153 | ||
3154 | progname = argv[0]; | |
3155 | ||
3156 | while ((opt = getopt_long_only(argc, argv, "+C:c:Ddhi:JM:m:PpST:v", | |
3157 | long_options, &option_index)) != -1) { | |
3158 | switch (opt) { | |
3159 | case 'C': | |
3160 | sscanf(optarg, "%x", &extra_delta_offset64); | |
3161 | break; | |
3162 | case 'c': | |
3163 | sscanf(optarg, "%x", &extra_delta_offset32); | |
3164 | break; | |
3165 | case 'D': | |
3166 | dump_only++; | |
3167 | break; | |
3168 | case 'd': | |
3169 | debug++; | |
3170 | break; | |
3171 | case 'h': | |
3172 | default: | |
3173 | help(); | |
3174 | exit(1); | |
3175 | case 'i': | |
3176 | interval_sec = atoi(optarg); | |
3177 | break; | |
3178 | case 'J': | |
3179 | rapl_joules++; | |
3180 | break; | |
3181 | case 'M': | |
3182 | sscanf(optarg, "%x", &extra_msr_offset64); | |
3183 | break; | |
3184 | case 'm': | |
3185 | sscanf(optarg, "%x", &extra_msr_offset32); | |
3186 | break; | |
3187 | case 'P': | |
3188 | show_pkg_only++; | |
3189 | break; | |
3190 | case 'p': | |
3191 | show_core_only++; | |
3192 | break; | |
3193 | case 'S': | |
3194 | summary_only++; | |
3195 | break; | |
3196 | case 'T': | |
3197 | tcc_activation_temp_override = atoi(optarg); | |
3198 | break; | |
3199 | case 'v': | |
3200 | print_version(); | |
3201 | exit(0); | |
3202 | break; | |
3203 | } | |
3204 | } | |
3205 | } | |
3206 | ||
3207 | int main(int argc, char **argv) | |
3208 | { | |
3209 | cmdline(argc, argv); | |
3210 | ||
3211 | if (debug) | |
3212 | print_version(); | |
3213 | ||
3214 | turbostat_init(); | |
3215 | ||
3216 | /* dump counters and exit */ | |
3217 | if (dump_only) | |
3218 | return get_and_dump_counters(); | |
3219 | ||
3220 | /* | |
3221 | * if any params left, it must be a command to fork | |
3222 | */ | |
3223 | if (argc - optind) | |
3224 | return fork_it(argv + optind); | |
3225 | else | |
3226 | turbostat_loop(); | |
3227 | ||
3228 | return 0; | |
3229 | } |