xref: /linux/tools/power/x86/turbostat/turbostat.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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_pc2;
56 unsigned int do_pc3;
57 unsigned int do_pc6;
58 unsigned int do_pc7;
59 unsigned int do_c8_c9_c10;
60 unsigned int do_slm_cstates;
61 unsigned int use_c1_residency_msr;
62 unsigned int has_aperf;
63 unsigned int has_epb;
64 unsigned int units = 1000000;	/* MHz etc */
65 unsigned int genuine_intel;
66 unsigned int has_invariant_tsc;
67 unsigned int do_nhm_platform_info;
68 unsigned int do_nhm_turbo_ratio_limit;
69 unsigned int do_ivt_turbo_ratio_limit;
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 int do_smi;
75 double bclk;
76 unsigned int show_pkg;
77 unsigned int show_core;
78 unsigned int show_cpu;
79 unsigned int show_pkg_only;
80 unsigned int show_core_only;
81 char *output_buffer, *outp;
82 unsigned int do_rapl;
83 unsigned int do_dts;
84 unsigned int do_ptm;
85 unsigned int tcc_activation_temp;
86 unsigned int tcc_activation_temp_override;
87 double rapl_power_units, rapl_energy_units, rapl_time_units;
88 double rapl_joule_counter_range;
89 unsigned int do_core_perf_limit_reasons;
90 unsigned int do_gfx_perf_limit_reasons;
91 unsigned int do_ring_perf_limit_reasons;
92 
93 #define RAPL_PKG		(1 << 0)
94 					/* 0x610 MSR_PKG_POWER_LIMIT */
95 					/* 0x611 MSR_PKG_ENERGY_STATUS */
96 #define RAPL_PKG_PERF_STATUS	(1 << 1)
97 					/* 0x613 MSR_PKG_PERF_STATUS */
98 #define RAPL_PKG_POWER_INFO	(1 << 2)
99 					/* 0x614 MSR_PKG_POWER_INFO */
100 
101 #define RAPL_DRAM		(1 << 3)
102 					/* 0x618 MSR_DRAM_POWER_LIMIT */
103 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
104 					/* 0x61c MSR_DRAM_POWER_INFO */
105 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
106 					/* 0x61b MSR_DRAM_PERF_STATUS */
107 
108 #define RAPL_CORES		(1 << 5)
109 					/* 0x638 MSR_PP0_POWER_LIMIT */
110 					/* 0x639 MSR_PP0_ENERGY_STATUS */
111 #define RAPL_CORE_POLICY	(1 << 6)
112 					/* 0x63a MSR_PP0_POLICY */
113 
114 
115 #define RAPL_GFX		(1 << 7)
116 					/* 0x640 MSR_PP1_POWER_LIMIT */
117 					/* 0x641 MSR_PP1_ENERGY_STATUS */
118 					/* 0x642 MSR_PP1_POLICY */
119 #define	TJMAX_DEFAULT	100
120 
121 #define MAX(a, b) ((a) > (b) ? (a) : (b))
122 
123 int aperf_mperf_unstable;
124 int backwards_count;
125 char *progname;
126 
127 cpu_set_t *cpu_present_set, *cpu_affinity_set;
128 size_t cpu_present_setsize, cpu_affinity_setsize;
129 
130 struct thread_data {
131 	unsigned long long tsc;
132 	unsigned long long aperf;
133 	unsigned long long mperf;
134 	unsigned long long c1;
135 	unsigned long long extra_msr64;
136 	unsigned long long extra_delta64;
137 	unsigned long long extra_msr32;
138 	unsigned long long extra_delta32;
139 	unsigned int smi_count;
140 	unsigned int cpu_id;
141 	unsigned int flags;
142 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
143 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
144 } *thread_even, *thread_odd;
145 
146 struct core_data {
147 	unsigned long long c3;
148 	unsigned long long c6;
149 	unsigned long long c7;
150 	unsigned int core_temp_c;
151 	unsigned int core_id;
152 } *core_even, *core_odd;
153 
154 struct pkg_data {
155 	unsigned long long pc2;
156 	unsigned long long pc3;
157 	unsigned long long pc6;
158 	unsigned long long pc7;
159 	unsigned long long pc8;
160 	unsigned long long pc9;
161 	unsigned long long pc10;
162 	unsigned int package_id;
163 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
164 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
165 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
166 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
167 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
168 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
169 	unsigned int pkg_temp_c;
170 
171 } *package_even, *package_odd;
172 
173 #define ODD_COUNTERS thread_odd, core_odd, package_odd
174 #define EVEN_COUNTERS thread_even, core_even, package_even
175 
176 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
177 	(thread_base + (pkg_no) * topo.num_cores_per_pkg * \
178 		topo.num_threads_per_core + \
179 		(core_no) * topo.num_threads_per_core + (thread_no))
180 #define GET_CORE(core_base, core_no, pkg_no) \
181 	(core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
182 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
183 
184 struct system_summary {
185 	struct thread_data threads;
186 	struct core_data cores;
187 	struct pkg_data packages;
188 } sum, average;
189 
190 
191 struct topo_params {
192 	int num_packages;
193 	int num_cpus;
194 	int num_cores;
195 	int max_cpu_num;
196 	int num_cores_per_pkg;
197 	int num_threads_per_core;
198 } topo;
199 
200 struct timeval tv_even, tv_odd, tv_delta;
201 
202 void setup_all_buffers(void);
203 
204 int cpu_is_not_present(int cpu)
205 {
206 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
207 }
208 /*
209  * run func(thread, core, package) in topology order
210  * skip non-present cpus
211  */
212 
213 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
214 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
215 {
216 	int retval, pkg_no, core_no, thread_no;
217 
218 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
219 		for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
220 			for (thread_no = 0; thread_no <
221 				topo.num_threads_per_core; ++thread_no) {
222 				struct thread_data *t;
223 				struct core_data *c;
224 				struct pkg_data *p;
225 
226 				t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
227 
228 				if (cpu_is_not_present(t->cpu_id))
229 					continue;
230 
231 				c = GET_CORE(core_base, core_no, pkg_no);
232 				p = GET_PKG(pkg_base, pkg_no);
233 
234 				retval = func(t, c, p);
235 				if (retval)
236 					return retval;
237 			}
238 		}
239 	}
240 	return 0;
241 }
242 
243 int cpu_migrate(int cpu)
244 {
245 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
246 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
247 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
248 		return -1;
249 	else
250 		return 0;
251 }
252 
253 int get_msr(int cpu, off_t offset, unsigned long long *msr)
254 {
255 	ssize_t retval;
256 	char pathname[32];
257 	int fd;
258 
259 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
260 	fd = open(pathname, O_RDONLY);
261 	if (fd < 0)
262 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
263 
264 	retval = pread(fd, msr, sizeof *msr, offset);
265 	close(fd);
266 
267 	if (retval != sizeof *msr)
268 		err(-1, "%s offset 0x%llx read failed", pathname, (unsigned long long)offset);
269 
270 	return 0;
271 }
272 
273 /*
274  * Example Format w/ field column widths:
275  *
276  *  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
277  * 123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678123456781234567812345678
278  */
279 
280 void print_header(void)
281 {
282 	if (show_pkg)
283 		outp += sprintf(outp, " Package");
284 	if (show_core)
285 		outp += sprintf(outp, "    Core");
286 	if (show_cpu)
287 		outp += sprintf(outp, "     CPU");
288 	if (has_aperf)
289 		outp += sprintf(outp, " Avg_MHz");
290 	if (has_aperf)
291 		outp += sprintf(outp, "   %%Busy");
292 	if (has_aperf)
293 		outp += sprintf(outp, " Bzy_MHz");
294 	outp += sprintf(outp, " TSC_MHz");
295 	if (do_smi)
296 		outp += sprintf(outp, "     SMI");
297 	if (extra_delta_offset32)
298 		outp += sprintf(outp, "  count 0x%03X", extra_delta_offset32);
299 	if (extra_delta_offset64)
300 		outp += sprintf(outp, "  COUNT 0x%03X", extra_delta_offset64);
301 	if (extra_msr_offset32)
302 		outp += sprintf(outp, "   MSR 0x%03X", extra_msr_offset32);
303 	if (extra_msr_offset64)
304 		outp += sprintf(outp, "           MSR 0x%03X", extra_msr_offset64);
305 	if (do_nhm_cstates)
306 		outp += sprintf(outp, "  CPU%%c1");
307 	if (do_nhm_cstates && !do_slm_cstates)
308 		outp += sprintf(outp, "  CPU%%c3");
309 	if (do_nhm_cstates)
310 		outp += sprintf(outp, "  CPU%%c6");
311 	if (do_snb_cstates)
312 		outp += sprintf(outp, "  CPU%%c7");
313 
314 	if (do_dts)
315 		outp += sprintf(outp, " CoreTmp");
316 	if (do_ptm)
317 		outp += sprintf(outp, "  PkgTmp");
318 
319 	if (do_pc2)
320 		outp += sprintf(outp, " Pkg%%pc2");
321 	if (do_pc3)
322 		outp += sprintf(outp, " Pkg%%pc3");
323 	if (do_pc6)
324 		outp += sprintf(outp, " Pkg%%pc6");
325 	if (do_pc7)
326 		outp += sprintf(outp, " Pkg%%pc7");
327 	if (do_c8_c9_c10) {
328 		outp += sprintf(outp, " Pkg%%pc8");
329 		outp += sprintf(outp, " Pkg%%pc9");
330 		outp += sprintf(outp, " Pk%%pc10");
331 	}
332 
333 	if (do_rapl && !rapl_joules) {
334 		if (do_rapl & RAPL_PKG)
335 			outp += sprintf(outp, " PkgWatt");
336 		if (do_rapl & RAPL_CORES)
337 			outp += sprintf(outp, " CorWatt");
338 		if (do_rapl & RAPL_GFX)
339 			outp += sprintf(outp, " GFXWatt");
340 		if (do_rapl & RAPL_DRAM)
341 			outp += sprintf(outp, " RAMWatt");
342 		if (do_rapl & RAPL_PKG_PERF_STATUS)
343 			outp += sprintf(outp, "   PKG_%%");
344 		if (do_rapl & RAPL_DRAM_PERF_STATUS)
345 			outp += sprintf(outp, "   RAM_%%");
346 	} else if (do_rapl && rapl_joules) {
347 		if (do_rapl & RAPL_PKG)
348 			outp += sprintf(outp, "   Pkg_J");
349 		if (do_rapl & RAPL_CORES)
350 			outp += sprintf(outp, "   Cor_J");
351 		if (do_rapl & RAPL_GFX)
352 			outp += sprintf(outp, "   GFX_J");
353 		if (do_rapl & RAPL_DRAM)
354 			outp += sprintf(outp, "   RAM_W");
355 		if (do_rapl & RAPL_PKG_PERF_STATUS)
356 			outp += sprintf(outp, "   PKG_%%");
357 		if (do_rapl & RAPL_DRAM_PERF_STATUS)
358 			outp += sprintf(outp, "   RAM_%%");
359 		outp += sprintf(outp, "   time");
360 
361 	}
362 	outp += sprintf(outp, "\n");
363 }
364 
365 int dump_counters(struct thread_data *t, struct core_data *c,
366 	struct pkg_data *p)
367 {
368 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
369 
370 	if (t) {
371 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
372 			t->cpu_id, t->flags);
373 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
374 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
375 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
376 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
377 		outp += sprintf(outp, "msr0x%x: %08llX\n",
378 			extra_delta_offset32, t->extra_delta32);
379 		outp += sprintf(outp, "msr0x%x: %016llX\n",
380 			extra_delta_offset64, t->extra_delta64);
381 		outp += sprintf(outp, "msr0x%x: %08llX\n",
382 			extra_msr_offset32, t->extra_msr32);
383 		outp += sprintf(outp, "msr0x%x: %016llX\n",
384 			extra_msr_offset64, t->extra_msr64);
385 		if (do_smi)
386 			outp += sprintf(outp, "SMI: %08X\n", t->smi_count);
387 	}
388 
389 	if (c) {
390 		outp += sprintf(outp, "core: %d\n", c->core_id);
391 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
392 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
393 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
394 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
395 	}
396 
397 	if (p) {
398 		outp += sprintf(outp, "package: %d\n", p->package_id);
399 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
400 		if (do_pc3)
401 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
402 		if (do_pc6)
403 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
404 		if (do_pc7)
405 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
406 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
407 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
408 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
409 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
410 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
411 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
412 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
413 		outp += sprintf(outp, "Throttle PKG: %0X\n",
414 			p->rapl_pkg_perf_status);
415 		outp += sprintf(outp, "Throttle RAM: %0X\n",
416 			p->rapl_dram_perf_status);
417 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
418 	}
419 
420 	outp += sprintf(outp, "\n");
421 
422 	return 0;
423 }
424 
425 /*
426  * column formatting convention & formats
427  */
428 int format_counters(struct thread_data *t, struct core_data *c,
429 	struct pkg_data *p)
430 {
431 	double interval_float;
432 	char *fmt8;
433 
434 	 /* if showing only 1st thread in core and this isn't one, bail out */
435 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
436 		return 0;
437 
438 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
439 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
440 		return 0;
441 
442 	interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
443 
444 	/* topo columns, print blanks on 1st (average) line */
445 	if (t == &average.threads) {
446 		if (show_pkg)
447 			outp += sprintf(outp, "       -");
448 		if (show_core)
449 			outp += sprintf(outp, "       -");
450 		if (show_cpu)
451 			outp += sprintf(outp, "       -");
452 	} else {
453 		if (show_pkg) {
454 			if (p)
455 				outp += sprintf(outp, "%8d", p->package_id);
456 			else
457 				outp += sprintf(outp, "       -");
458 		}
459 		if (show_core) {
460 			if (c)
461 				outp += sprintf(outp, "%8d", c->core_id);
462 			else
463 				outp += sprintf(outp, "       -");
464 		}
465 		if (show_cpu)
466 			outp += sprintf(outp, "%8d", t->cpu_id);
467 	}
468 
469 	/* Avg_MHz */
470 	if (has_aperf)
471 		outp += sprintf(outp, "%8.0f",
472 			1.0 / units * t->aperf / interval_float);
473 
474 	/* %Busy */
475 	if (has_aperf) {
476 		if (!skip_c0)
477 			outp += sprintf(outp, "%8.2f", 100.0 * t->mperf/t->tsc);
478 		else
479 			outp += sprintf(outp, "********");
480 	}
481 
482 	/* Bzy_MHz */
483 	if (has_aperf)
484 		outp += sprintf(outp, "%8.0f",
485 			1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
486 
487 	/* TSC_MHz */
488 	outp += sprintf(outp, "%8.0f", 1.0 * t->tsc/units/interval_float);
489 
490 	/* SMI */
491 	if (do_smi)
492 		outp += sprintf(outp, "%8d", t->smi_count);
493 
494 	/* delta */
495 	if (extra_delta_offset32)
496 		outp += sprintf(outp, "  %11llu", t->extra_delta32);
497 
498 	/* DELTA */
499 	if (extra_delta_offset64)
500 		outp += sprintf(outp, "  %11llu", t->extra_delta64);
501 	/* msr */
502 	if (extra_msr_offset32)
503 		outp += sprintf(outp, "  0x%08llx", t->extra_msr32);
504 
505 	/* MSR */
506 	if (extra_msr_offset64)
507 		outp += sprintf(outp, "  0x%016llx", t->extra_msr64);
508 
509 	if (do_nhm_cstates) {
510 		if (!skip_c1)
511 			outp += sprintf(outp, "%8.2f", 100.0 * t->c1/t->tsc);
512 		else
513 			outp += sprintf(outp, "********");
514 	}
515 
516 	/* print per-core data only for 1st thread in core */
517 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
518 		goto done;
519 
520 	if (do_nhm_cstates && !do_slm_cstates)
521 		outp += sprintf(outp, "%8.2f", 100.0 * c->c3/t->tsc);
522 	if (do_nhm_cstates)
523 		outp += sprintf(outp, "%8.2f", 100.0 * c->c6/t->tsc);
524 	if (do_snb_cstates)
525 		outp += sprintf(outp, "%8.2f", 100.0 * c->c7/t->tsc);
526 
527 	if (do_dts)
528 		outp += sprintf(outp, "%8d", c->core_temp_c);
529 
530 	/* print per-package data only for 1st core in package */
531 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
532 		goto done;
533 
534 	if (do_ptm)
535 		outp += sprintf(outp, "%8d", p->pkg_temp_c);
536 
537 	if (do_pc2)
538 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc2/t->tsc);
539 	if (do_pc3)
540 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc3/t->tsc);
541 	if (do_pc6)
542 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc6/t->tsc);
543 	if (do_pc7)
544 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc7/t->tsc);
545 	if (do_c8_c9_c10) {
546 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc8/t->tsc);
547 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc9/t->tsc);
548 		outp += sprintf(outp, "%8.2f", 100.0 * p->pc10/t->tsc);
549 	}
550 
551 	/*
552  	 * If measurement interval exceeds minimum RAPL Joule Counter range,
553  	 * indicate that results are suspect by printing "**" in fraction place.
554  	 */
555 	if (interval_float < rapl_joule_counter_range)
556 		fmt8 = "%8.2f";
557 	else
558 		fmt8 = " %6.0f**";
559 
560 	if (do_rapl && !rapl_joules) {
561 		if (do_rapl & RAPL_PKG)
562 			outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float);
563 		if (do_rapl & RAPL_CORES)
564 			outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float);
565 		if (do_rapl & RAPL_GFX)
566 			outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float);
567 		if (do_rapl & RAPL_DRAM)
568 			outp += sprintf(outp, fmt8, p->energy_dram * rapl_energy_units / interval_float);
569 		if (do_rapl & RAPL_PKG_PERF_STATUS)
570 			outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
571 		if (do_rapl & RAPL_DRAM_PERF_STATUS)
572 			outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
573 	} else if (do_rapl && rapl_joules) {
574 		if (do_rapl & RAPL_PKG)
575 			outp += sprintf(outp, fmt8,
576 					p->energy_pkg * rapl_energy_units);
577 		if (do_rapl & RAPL_CORES)
578 			outp += sprintf(outp, fmt8,
579 					p->energy_cores * rapl_energy_units);
580 		if (do_rapl & RAPL_GFX)
581 			outp += sprintf(outp, fmt8,
582 					p->energy_gfx * rapl_energy_units);
583 		if (do_rapl & RAPL_DRAM)
584 			outp += sprintf(outp, fmt8,
585 					p->energy_dram * rapl_energy_units);
586 		if (do_rapl & RAPL_PKG_PERF_STATUS)
587 			outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
588 		if (do_rapl & RAPL_DRAM_PERF_STATUS)
589 			outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
590 
591 		outp += sprintf(outp, fmt8, interval_float);
592 	}
593 done:
594 	outp += sprintf(outp, "\n");
595 
596 	return 0;
597 }
598 
599 void flush_stdout()
600 {
601 	fputs(output_buffer, stdout);
602 	fflush(stdout);
603 	outp = output_buffer;
604 }
605 void flush_stderr()
606 {
607 	fputs(output_buffer, stderr);
608 	outp = output_buffer;
609 }
610 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
611 {
612 	static int printed;
613 
614 	if (!printed || !summary_only)
615 		print_header();
616 
617 	if (topo.num_cpus > 1)
618 		format_counters(&average.threads, &average.cores,
619 			&average.packages);
620 
621 	printed = 1;
622 
623 	if (summary_only)
624 		return;
625 
626 	for_all_cpus(format_counters, t, c, p);
627 }
628 
629 #define DELTA_WRAP32(new, old)			\
630 	if (new > old) {			\
631 		old = new - old;		\
632 	} else {				\
633 		old = 0x100000000 + new - old;	\
634 	}
635 
636 void
637 delta_package(struct pkg_data *new, struct pkg_data *old)
638 {
639 	old->pc2 = new->pc2 - old->pc2;
640 	if (do_pc3)
641 		old->pc3 = new->pc3 - old->pc3;
642 	if (do_pc6)
643 		old->pc6 = new->pc6 - old->pc6;
644 	if (do_pc7)
645 		old->pc7 = new->pc7 - old->pc7;
646 	old->pc8 = new->pc8 - old->pc8;
647 	old->pc9 = new->pc9 - old->pc9;
648 	old->pc10 = new->pc10 - old->pc10;
649 	old->pkg_temp_c = new->pkg_temp_c;
650 
651 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
652 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
653 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
654 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
655 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
656 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
657 }
658 
659 void
660 delta_core(struct core_data *new, struct core_data *old)
661 {
662 	old->c3 = new->c3 - old->c3;
663 	old->c6 = new->c6 - old->c6;
664 	old->c7 = new->c7 - old->c7;
665 	old->core_temp_c = new->core_temp_c;
666 }
667 
668 /*
669  * old = new - old
670  */
671 void
672 delta_thread(struct thread_data *new, struct thread_data *old,
673 	struct core_data *core_delta)
674 {
675 	old->tsc = new->tsc - old->tsc;
676 
677 	/* check for TSC < 1 Mcycles over interval */
678 	if (old->tsc < (1000 * 1000))
679 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
680 		     "You can disable all c-states by booting with \"idle=poll\"\n"
681 		     "or just the deep ones with \"processor.max_cstate=1\"");
682 
683 	old->c1 = new->c1 - old->c1;
684 
685 	if (has_aperf) {
686 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
687 			old->aperf = new->aperf - old->aperf;
688 			old->mperf = new->mperf - old->mperf;
689 		} else {
690 
691 			if (!aperf_mperf_unstable) {
692 				fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
693 				fprintf(stderr, "* Frequency results do not cover entire interval *\n");
694 				fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
695 
696 				aperf_mperf_unstable = 1;
697 			}
698 			/*
699 			 * mperf delta is likely a huge "positive" number
700 			 * can not use it for calculating c0 time
701 			 */
702 			skip_c0 = 1;
703 			skip_c1 = 1;
704 		}
705 	}
706 
707 
708 	if (use_c1_residency_msr) {
709 		/*
710 		 * Some models have a dedicated C1 residency MSR,
711 		 * which should be more accurate than the derivation below.
712 		 */
713 	} else {
714 		/*
715 		 * As counter collection is not atomic,
716 		 * it is possible for mperf's non-halted cycles + idle states
717 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
718 		 */
719 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
720 			old->c1 = 0;
721 		else {
722 			/* normal case, derive c1 */
723 			old->c1 = old->tsc - old->mperf - core_delta->c3
724 				- core_delta->c6 - core_delta->c7;
725 		}
726 	}
727 
728 	if (old->mperf == 0) {
729 		if (debug > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id);
730 		old->mperf = 1;	/* divide by 0 protection */
731 	}
732 
733 	old->extra_delta32 = new->extra_delta32 - old->extra_delta32;
734 	old->extra_delta32 &= 0xFFFFFFFF;
735 
736 	old->extra_delta64 = new->extra_delta64 - old->extra_delta64;
737 
738 	/*
739 	 * Extra MSR is just a snapshot, simply copy latest w/o subtracting
740 	 */
741 	old->extra_msr32 = new->extra_msr32;
742 	old->extra_msr64 = new->extra_msr64;
743 
744 	if (do_smi)
745 		old->smi_count = new->smi_count - old->smi_count;
746 }
747 
748 int delta_cpu(struct thread_data *t, struct core_data *c,
749 	struct pkg_data *p, struct thread_data *t2,
750 	struct core_data *c2, struct pkg_data *p2)
751 {
752 	/* calculate core delta only for 1st thread in core */
753 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
754 		delta_core(c, c2);
755 
756 	/* always calculate thread delta */
757 	delta_thread(t, t2, c2);	/* c2 is core delta */
758 
759 	/* calculate package delta only for 1st core in package */
760 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
761 		delta_package(p, p2);
762 
763 	return 0;
764 }
765 
766 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
767 {
768 	t->tsc = 0;
769 	t->aperf = 0;
770 	t->mperf = 0;
771 	t->c1 = 0;
772 
773 	t->smi_count = 0;
774 	t->extra_delta32 = 0;
775 	t->extra_delta64 = 0;
776 
777 	/* tells format_counters to dump all fields from this set */
778 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
779 
780 	c->c3 = 0;
781 	c->c6 = 0;
782 	c->c7 = 0;
783 	c->core_temp_c = 0;
784 
785 	p->pc2 = 0;
786 	if (do_pc3)
787 		p->pc3 = 0;
788 	if (do_pc6)
789 		p->pc6 = 0;
790 	if (do_pc7)
791 		p->pc7 = 0;
792 	p->pc8 = 0;
793 	p->pc9 = 0;
794 	p->pc10 = 0;
795 
796 	p->energy_pkg = 0;
797 	p->energy_dram = 0;
798 	p->energy_cores = 0;
799 	p->energy_gfx = 0;
800 	p->rapl_pkg_perf_status = 0;
801 	p->rapl_dram_perf_status = 0;
802 	p->pkg_temp_c = 0;
803 }
804 int sum_counters(struct thread_data *t, struct core_data *c,
805 	struct pkg_data *p)
806 {
807 	average.threads.tsc += t->tsc;
808 	average.threads.aperf += t->aperf;
809 	average.threads.mperf += t->mperf;
810 	average.threads.c1 += t->c1;
811 
812 	average.threads.extra_delta32 += t->extra_delta32;
813 	average.threads.extra_delta64 += t->extra_delta64;
814 
815 	/* sum per-core values only for 1st thread in core */
816 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
817 		return 0;
818 
819 	average.cores.c3 += c->c3;
820 	average.cores.c6 += c->c6;
821 	average.cores.c7 += c->c7;
822 
823 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
824 
825 	/* sum per-pkg values only for 1st core in pkg */
826 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
827 		return 0;
828 
829 	average.packages.pc2 += p->pc2;
830 	if (do_pc3)
831 		average.packages.pc3 += p->pc3;
832 	if (do_pc6)
833 		average.packages.pc6 += p->pc6;
834 	if (do_pc7)
835 		average.packages.pc7 += p->pc7;
836 	average.packages.pc8 += p->pc8;
837 	average.packages.pc9 += p->pc9;
838 	average.packages.pc10 += p->pc10;
839 
840 	average.packages.energy_pkg += p->energy_pkg;
841 	average.packages.energy_dram += p->energy_dram;
842 	average.packages.energy_cores += p->energy_cores;
843 	average.packages.energy_gfx += p->energy_gfx;
844 
845 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
846 
847 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
848 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
849 	return 0;
850 }
851 /*
852  * sum the counters for all cpus in the system
853  * compute the weighted average
854  */
855 void compute_average(struct thread_data *t, struct core_data *c,
856 	struct pkg_data *p)
857 {
858 	clear_counters(&average.threads, &average.cores, &average.packages);
859 
860 	for_all_cpus(sum_counters, t, c, p);
861 
862 	average.threads.tsc /= topo.num_cpus;
863 	average.threads.aperf /= topo.num_cpus;
864 	average.threads.mperf /= topo.num_cpus;
865 	average.threads.c1 /= topo.num_cpus;
866 
867 	average.threads.extra_delta32 /= topo.num_cpus;
868 	average.threads.extra_delta32 &= 0xFFFFFFFF;
869 
870 	average.threads.extra_delta64 /= topo.num_cpus;
871 
872 	average.cores.c3 /= topo.num_cores;
873 	average.cores.c6 /= topo.num_cores;
874 	average.cores.c7 /= topo.num_cores;
875 
876 	average.packages.pc2 /= topo.num_packages;
877 	if (do_pc3)
878 		average.packages.pc3 /= topo.num_packages;
879 	if (do_pc6)
880 		average.packages.pc6 /= topo.num_packages;
881 	if (do_pc7)
882 		average.packages.pc7 /= topo.num_packages;
883 
884 	average.packages.pc8 /= topo.num_packages;
885 	average.packages.pc9 /= topo.num_packages;
886 	average.packages.pc10 /= topo.num_packages;
887 }
888 
889 static unsigned long long rdtsc(void)
890 {
891 	unsigned int low, high;
892 
893 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
894 
895 	return low | ((unsigned long long)high) << 32;
896 }
897 
898 
899 /*
900  * get_counters(...)
901  * migrate to cpu
902  * acquire and record local counters for that cpu
903  */
904 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
905 {
906 	int cpu = t->cpu_id;
907 	unsigned long long msr;
908 
909 	if (cpu_migrate(cpu)) {
910 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
911 		return -1;
912 	}
913 
914 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
915 
916 	if (has_aperf) {
917 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
918 			return -3;
919 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
920 			return -4;
921 	}
922 
923 	if (do_smi) {
924 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
925 			return -5;
926 		t->smi_count = msr & 0xFFFFFFFF;
927 	}
928 	if (extra_delta_offset32) {
929 		if (get_msr(cpu, extra_delta_offset32, &msr))
930 			return -5;
931 		t->extra_delta32 = msr & 0xFFFFFFFF;
932 	}
933 
934 	if (extra_delta_offset64)
935 		if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64))
936 			return -5;
937 
938 	if (extra_msr_offset32) {
939 		if (get_msr(cpu, extra_msr_offset32, &msr))
940 			return -5;
941 		t->extra_msr32 = msr & 0xFFFFFFFF;
942 	}
943 
944 	if (extra_msr_offset64)
945 		if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64))
946 			return -5;
947 
948 	if (use_c1_residency_msr) {
949 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
950 			return -6;
951 	}
952 
953 	/* collect core counters only for 1st thread in core */
954 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
955 		return 0;
956 
957 	if (do_nhm_cstates && !do_slm_cstates) {
958 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
959 			return -6;
960 	}
961 
962 	if (do_nhm_cstates) {
963 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
964 			return -7;
965 	}
966 
967 	if (do_snb_cstates)
968 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
969 			return -8;
970 
971 	if (do_dts) {
972 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
973 			return -9;
974 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
975 	}
976 
977 
978 	/* collect package counters only for 1st core in package */
979 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
980 		return 0;
981 
982 	if (do_pc3)
983 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
984 			return -9;
985 	if (do_pc6)
986 		if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
987 			return -10;
988 	if (do_pc2)
989 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
990 			return -11;
991 	if (do_pc7)
992 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
993 			return -12;
994 	if (do_c8_c9_c10) {
995 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
996 			return -13;
997 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
998 			return -13;
999 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1000 			return -13;
1001 	}
1002 	if (do_rapl & RAPL_PKG) {
1003 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1004 			return -13;
1005 		p->energy_pkg = msr & 0xFFFFFFFF;
1006 	}
1007 	if (do_rapl & RAPL_CORES) {
1008 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1009 			return -14;
1010 		p->energy_cores = msr & 0xFFFFFFFF;
1011 	}
1012 	if (do_rapl & RAPL_DRAM) {
1013 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1014 			return -15;
1015 		p->energy_dram = msr & 0xFFFFFFFF;
1016 	}
1017 	if (do_rapl & RAPL_GFX) {
1018 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1019 			return -16;
1020 		p->energy_gfx = msr & 0xFFFFFFFF;
1021 	}
1022 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1023 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1024 			return -16;
1025 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1026 	}
1027 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1028 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1029 			return -16;
1030 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1031 	}
1032 	if (do_ptm) {
1033 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1034 			return -17;
1035 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1036 	}
1037 	return 0;
1038 }
1039 
1040 /*
1041  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
1042  * If you change the values, note they are used both in comparisons
1043  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
1044  */
1045 
1046 #define PCLUKN 0 /* Unknown */
1047 #define PCLRSV 1 /* Reserved */
1048 #define PCL__0 2 /* PC0 */
1049 #define PCL__1 3 /* PC1 */
1050 #define PCL__2 4 /* PC2 */
1051 #define PCL__3 5 /* PC3 */
1052 #define PCL__4 6 /* PC4 */
1053 #define PCL__6 7 /* PC6 */
1054 #define PCL_6N 8 /* PC6 No Retention */
1055 #define PCL_6R 9 /* PC6 Retention */
1056 #define PCL__7 10 /* PC7 */
1057 #define PCL_7S 11 /* PC7 Shrink */
1058 #define PCLUNL 12 /* Unlimited */
1059 
1060 int pkg_cstate_limit = PCLUKN;
1061 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
1062 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "unlimited"};
1063 
1064 int nhm_pkg_cstate_limits[8] = {PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL};
1065 int snb_pkg_cstate_limits[8] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL};
1066 int hsw_pkg_cstate_limits[8] = {PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCLRSV, PCLUNL};
1067 int slv_pkg_cstate_limits[8] = {PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7};
1068 int amt_pkg_cstate_limits[8] = {PCL__0, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7};
1069 int phi_pkg_cstate_limits[8] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL};
1070 
1071 void print_verbose_header(void)
1072 {
1073 	unsigned long long msr;
1074 	unsigned int ratio;
1075 
1076 	if (!do_nhm_platform_info)
1077 		return;
1078 
1079 	get_msr(0, MSR_NHM_PLATFORM_INFO, &msr);
1080 
1081 	fprintf(stderr, "cpu0: MSR_NHM_PLATFORM_INFO: 0x%08llx\n", msr);
1082 
1083 	ratio = (msr >> 40) & 0xFF;
1084 	fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
1085 		ratio, bclk, ratio * bclk);
1086 
1087 	ratio = (msr >> 8) & 0xFF;
1088 	fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
1089 		ratio, bclk, ratio * bclk);
1090 
1091 	get_msr(0, MSR_IA32_POWER_CTL, &msr);
1092 	fprintf(stderr, "cpu0: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
1093 		msr, msr & 0x2 ? "EN" : "DIS");
1094 
1095 	if (!do_ivt_turbo_ratio_limit)
1096 		goto print_nhm_turbo_ratio_limits;
1097 
1098 	get_msr(0, MSR_IVT_TURBO_RATIO_LIMIT, &msr);
1099 
1100 	fprintf(stderr, "cpu0: MSR_IVT_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
1101 
1102 	ratio = (msr >> 56) & 0xFF;
1103 	if (ratio)
1104 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n",
1105 			ratio, bclk, ratio * bclk);
1106 
1107 	ratio = (msr >> 48) & 0xFF;
1108 	if (ratio)
1109 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n",
1110 			ratio, bclk, ratio * bclk);
1111 
1112 	ratio = (msr >> 40) & 0xFF;
1113 	if (ratio)
1114 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n",
1115 			ratio, bclk, ratio * bclk);
1116 
1117 	ratio = (msr >> 32) & 0xFF;
1118 	if (ratio)
1119 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n",
1120 			ratio, bclk, ratio * bclk);
1121 
1122 	ratio = (msr >> 24) & 0xFF;
1123 	if (ratio)
1124 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n",
1125 			ratio, bclk, ratio * bclk);
1126 
1127 	ratio = (msr >> 16) & 0xFF;
1128 	if (ratio)
1129 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n",
1130 			ratio, bclk, ratio * bclk);
1131 
1132 	ratio = (msr >> 8) & 0xFF;
1133 	if (ratio)
1134 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n",
1135 			ratio, bclk, ratio * bclk);
1136 
1137 	ratio = (msr >> 0) & 0xFF;
1138 	if (ratio)
1139 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n",
1140 			ratio, bclk, ratio * bclk);
1141 
1142 print_nhm_turbo_ratio_limits:
1143 	get_msr(0, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
1144 
1145 #define SNB_C1_AUTO_UNDEMOTE              (1UL << 27)
1146 #define SNB_C3_AUTO_UNDEMOTE              (1UL << 28)
1147 
1148 	fprintf(stderr, "cpu0: MSR_NHM_SNB_PKG_CST_CFG_CTL: 0x%08llx", msr);
1149 
1150 	fprintf(stderr, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n",
1151 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
1152 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
1153 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
1154 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
1155 		(msr & (1 << 15)) ? "" : "UN",
1156 		(unsigned int)msr & 7,
1157 		pkg_cstate_limit_strings[pkg_cstate_limit]);
1158 
1159 	if (!do_nhm_turbo_ratio_limit)
1160 		return;
1161 
1162 	get_msr(0, MSR_NHM_TURBO_RATIO_LIMIT, &msr);
1163 
1164 	fprintf(stderr, "cpu0: MSR_NHM_TURBO_RATIO_LIMIT: 0x%08llx\n", msr);
1165 
1166 	ratio = (msr >> 56) & 0xFF;
1167 	if (ratio)
1168 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n",
1169 			ratio, bclk, ratio * bclk);
1170 
1171 	ratio = (msr >> 48) & 0xFF;
1172 	if (ratio)
1173 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n",
1174 			ratio, bclk, ratio * bclk);
1175 
1176 	ratio = (msr >> 40) & 0xFF;
1177 	if (ratio)
1178 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n",
1179 			ratio, bclk, ratio * bclk);
1180 
1181 	ratio = (msr >> 32) & 0xFF;
1182 	if (ratio)
1183 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n",
1184 			ratio, bclk, ratio * bclk);
1185 
1186 	ratio = (msr >> 24) & 0xFF;
1187 	if (ratio)
1188 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
1189 			ratio, bclk, ratio * bclk);
1190 
1191 	ratio = (msr >> 16) & 0xFF;
1192 	if (ratio)
1193 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
1194 			ratio, bclk, ratio * bclk);
1195 
1196 	ratio = (msr >> 8) & 0xFF;
1197 	if (ratio)
1198 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
1199 			ratio, bclk, ratio * bclk);
1200 
1201 	ratio = (msr >> 0) & 0xFF;
1202 	if (ratio)
1203 		fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
1204 			ratio, bclk, ratio * bclk);
1205 
1206 }
1207 
1208 void free_all_buffers(void)
1209 {
1210 	CPU_FREE(cpu_present_set);
1211 	cpu_present_set = NULL;
1212 	cpu_present_set = 0;
1213 
1214 	CPU_FREE(cpu_affinity_set);
1215 	cpu_affinity_set = NULL;
1216 	cpu_affinity_setsize = 0;
1217 
1218 	free(thread_even);
1219 	free(core_even);
1220 	free(package_even);
1221 
1222 	thread_even = NULL;
1223 	core_even = NULL;
1224 	package_even = NULL;
1225 
1226 	free(thread_odd);
1227 	free(core_odd);
1228 	free(package_odd);
1229 
1230 	thread_odd = NULL;
1231 	core_odd = NULL;
1232 	package_odd = NULL;
1233 
1234 	free(output_buffer);
1235 	output_buffer = NULL;
1236 	outp = NULL;
1237 }
1238 
1239 /*
1240  * Open a file, and exit on failure
1241  */
1242 FILE *fopen_or_die(const char *path, const char *mode)
1243 {
1244 	FILE *filep = fopen(path, "r");
1245 	if (!filep)
1246 		err(1, "%s: open failed", path);
1247 	return filep;
1248 }
1249 
1250 /*
1251  * Parse a file containing a single int.
1252  */
1253 int parse_int_file(const char *fmt, ...)
1254 {
1255 	va_list args;
1256 	char path[PATH_MAX];
1257 	FILE *filep;
1258 	int value;
1259 
1260 	va_start(args, fmt);
1261 	vsnprintf(path, sizeof(path), fmt, args);
1262 	va_end(args);
1263 	filep = fopen_or_die(path, "r");
1264 	if (fscanf(filep, "%d", &value) != 1)
1265 		err(1, "%s: failed to parse number from file", path);
1266 	fclose(filep);
1267 	return value;
1268 }
1269 
1270 /*
1271  * cpu_is_first_sibling_in_core(cpu)
1272  * return 1 if given CPU is 1st HT sibling in the core
1273  */
1274 int cpu_is_first_sibling_in_core(int cpu)
1275 {
1276 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1277 }
1278 
1279 /*
1280  * cpu_is_first_core_in_package(cpu)
1281  * return 1 if given CPU is 1st core in package
1282  */
1283 int cpu_is_first_core_in_package(int cpu)
1284 {
1285 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
1286 }
1287 
1288 int get_physical_package_id(int cpu)
1289 {
1290 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
1291 }
1292 
1293 int get_core_id(int cpu)
1294 {
1295 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
1296 }
1297 
1298 int get_num_ht_siblings(int cpu)
1299 {
1300 	char path[80];
1301 	FILE *filep;
1302 	int sib1, sib2;
1303 	int matches;
1304 	char character;
1305 
1306 	sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
1307 	filep = fopen_or_die(path, "r");
1308 	/*
1309 	 * file format:
1310 	 * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
1311 	 * otherwinse 1 sibling (self).
1312 	 */
1313 	matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
1314 
1315 	fclose(filep);
1316 
1317 	if (matches == 3)
1318 		return 2;
1319 	else
1320 		return 1;
1321 }
1322 
1323 /*
1324  * run func(thread, core, package) in topology order
1325  * skip non-present cpus
1326  */
1327 
1328 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
1329 	struct pkg_data *, struct thread_data *, struct core_data *,
1330 	struct pkg_data *), struct thread_data *thread_base,
1331 	struct core_data *core_base, struct pkg_data *pkg_base,
1332 	struct thread_data *thread_base2, struct core_data *core_base2,
1333 	struct pkg_data *pkg_base2)
1334 {
1335 	int retval, pkg_no, core_no, thread_no;
1336 
1337 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
1338 		for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
1339 			for (thread_no = 0; thread_no <
1340 				topo.num_threads_per_core; ++thread_no) {
1341 				struct thread_data *t, *t2;
1342 				struct core_data *c, *c2;
1343 				struct pkg_data *p, *p2;
1344 
1345 				t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
1346 
1347 				if (cpu_is_not_present(t->cpu_id))
1348 					continue;
1349 
1350 				t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
1351 
1352 				c = GET_CORE(core_base, core_no, pkg_no);
1353 				c2 = GET_CORE(core_base2, core_no, pkg_no);
1354 
1355 				p = GET_PKG(pkg_base, pkg_no);
1356 				p2 = GET_PKG(pkg_base2, pkg_no);
1357 
1358 				retval = func(t, c, p, t2, c2, p2);
1359 				if (retval)
1360 					return retval;
1361 			}
1362 		}
1363 	}
1364 	return 0;
1365 }
1366 
1367 /*
1368  * run func(cpu) on every cpu in /proc/stat
1369  * return max_cpu number
1370  */
1371 int for_all_proc_cpus(int (func)(int))
1372 {
1373 	FILE *fp;
1374 	int cpu_num;
1375 	int retval;
1376 
1377 	fp = fopen_or_die(proc_stat, "r");
1378 
1379 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
1380 	if (retval != 0)
1381 		err(1, "%s: failed to parse format", proc_stat);
1382 
1383 	while (1) {
1384 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
1385 		if (retval != 1)
1386 			break;
1387 
1388 		retval = func(cpu_num);
1389 		if (retval) {
1390 			fclose(fp);
1391 			return(retval);
1392 		}
1393 	}
1394 	fclose(fp);
1395 	return 0;
1396 }
1397 
1398 void re_initialize(void)
1399 {
1400 	free_all_buffers();
1401 	setup_all_buffers();
1402 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
1403 }
1404 
1405 
1406 /*
1407  * count_cpus()
1408  * remember the last one seen, it will be the max
1409  */
1410 int count_cpus(int cpu)
1411 {
1412 	if (topo.max_cpu_num < cpu)
1413 		topo.max_cpu_num = cpu;
1414 
1415 	topo.num_cpus += 1;
1416 	return 0;
1417 }
1418 int mark_cpu_present(int cpu)
1419 {
1420 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
1421 	return 0;
1422 }
1423 
1424 void turbostat_loop()
1425 {
1426 	int retval;
1427 	int restarted = 0;
1428 
1429 restart:
1430 	restarted++;
1431 
1432 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
1433 	if (retval < -1) {
1434 		exit(retval);
1435 	} else if (retval == -1) {
1436 		if (restarted > 1) {
1437 			exit(retval);
1438 		}
1439 		re_initialize();
1440 		goto restart;
1441 	}
1442 	restarted = 0;
1443 	gettimeofday(&tv_even, (struct timezone *)NULL);
1444 
1445 	while (1) {
1446 		if (for_all_proc_cpus(cpu_is_not_present)) {
1447 			re_initialize();
1448 			goto restart;
1449 		}
1450 		sleep(interval_sec);
1451 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
1452 		if (retval < -1) {
1453 			exit(retval);
1454 		} else if (retval == -1) {
1455 			re_initialize();
1456 			goto restart;
1457 		}
1458 		gettimeofday(&tv_odd, (struct timezone *)NULL);
1459 		timersub(&tv_odd, &tv_even, &tv_delta);
1460 		for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
1461 		compute_average(EVEN_COUNTERS);
1462 		format_all_counters(EVEN_COUNTERS);
1463 		flush_stdout();
1464 		sleep(interval_sec);
1465 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
1466 		if (retval < -1) {
1467 			exit(retval);
1468 		} else if (retval == -1) {
1469 			re_initialize();
1470 			goto restart;
1471 		}
1472 		gettimeofday(&tv_even, (struct timezone *)NULL);
1473 		timersub(&tv_even, &tv_odd, &tv_delta);
1474 		for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS);
1475 		compute_average(ODD_COUNTERS);
1476 		format_all_counters(ODD_COUNTERS);
1477 		flush_stdout();
1478 	}
1479 }
1480 
1481 void check_dev_msr()
1482 {
1483 	struct stat sb;
1484 
1485 	if (stat("/dev/cpu/0/msr", &sb))
1486 		err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
1487 }
1488 
1489 void check_permissions()
1490 {
1491 	struct __user_cap_header_struct cap_header_data;
1492 	cap_user_header_t cap_header = &cap_header_data;
1493 	struct __user_cap_data_struct cap_data_data;
1494 	cap_user_data_t cap_data = &cap_data_data;
1495 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
1496 	int do_exit = 0;
1497 
1498 	/* check for CAP_SYS_RAWIO */
1499 	cap_header->pid = getpid();
1500 	cap_header->version = _LINUX_CAPABILITY_VERSION;
1501 	if (capget(cap_header, cap_data) < 0)
1502 		err(-6, "capget(2) failed");
1503 
1504 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
1505 		do_exit++;
1506 		warnx("capget(CAP_SYS_RAWIO) failed,"
1507 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
1508 	}
1509 
1510 	/* test file permissions */
1511 	if (euidaccess("/dev/cpu/0/msr", R_OK)) {
1512 		do_exit++;
1513 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
1514 	}
1515 
1516 	/* if all else fails, thell them to be root */
1517 	if (do_exit)
1518 		if (getuid() != 0)
1519 			warnx("... or simply run as root");
1520 
1521 	if (do_exit)
1522 		exit(-6);
1523 }
1524 
1525 /*
1526  * NHM adds support for additional MSRs:
1527  *
1528  * MSR_SMI_COUNT                   0x00000034
1529  *
1530  * MSR_NHM_PLATFORM_INFO           0x000000ce
1531  * MSR_NHM_SNB_PKG_CST_CFG_CTL     0x000000e2
1532  *
1533  * MSR_PKG_C3_RESIDENCY            0x000003f8
1534  * MSR_PKG_C6_RESIDENCY            0x000003f9
1535  * MSR_CORE_C3_RESIDENCY           0x000003fc
1536  * MSR_CORE_C6_RESIDENCY           0x000003fd
1537  *
1538  * Side effect:
1539  * sets global pkg_cstate_limit to decode MSR_NHM_SNB_PKG_CST_CFG_CTL
1540  */
1541 int probe_nhm_msrs(unsigned int family, unsigned int model)
1542 {
1543 	unsigned long long msr;
1544 	int *pkg_cstate_limits;
1545 
1546 	if (!genuine_intel)
1547 		return 0;
1548 
1549 	if (family != 6)
1550 		return 0;
1551 
1552 	switch (model) {
1553 	case 0x1A:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
1554 	case 0x1E:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
1555 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
1556 	case 0x25:	/* Westmere Client - Clarkdale, Arrandale */
1557 	case 0x2C:	/* Westmere EP - Gulftown */
1558 	case 0x2E:	/* Nehalem-EX Xeon - Beckton */
1559 	case 0x2F:	/* Westmere-EX Xeon - Eagleton */
1560 		pkg_cstate_limits = nhm_pkg_cstate_limits;
1561 		break;
1562 	case 0x2A:	/* SNB */
1563 	case 0x2D:	/* SNB Xeon */
1564 	case 0x3A:	/* IVB */
1565 	case 0x3E:	/* IVB Xeon */
1566 		pkg_cstate_limits = snb_pkg_cstate_limits;
1567 		break;
1568 	case 0x3C:	/* HSW */
1569 	case 0x3F:	/* HSX */
1570 	case 0x45:	/* HSW */
1571 	case 0x46:	/* HSW */
1572 	case 0x3D:	/* BDW */
1573 	case 0x47:	/* BDW */
1574 	case 0x4F:	/* BDX */
1575 	case 0x56:	/* BDX-DE */
1576 		pkg_cstate_limits = hsw_pkg_cstate_limits;
1577 		break;
1578 	case 0x37:	/* BYT */
1579 	case 0x4D:	/* AVN */
1580 		pkg_cstate_limits = slv_pkg_cstate_limits;
1581 		break;
1582 	case 0x4C:	/* AMT */
1583 		pkg_cstate_limits = amt_pkg_cstate_limits;
1584 		break;
1585 	case 0x57:	/* PHI */
1586 		pkg_cstate_limits = phi_pkg_cstate_limits;
1587 		break;
1588 	default:
1589 		return 0;
1590 	}
1591 	get_msr(0, MSR_NHM_SNB_PKG_CST_CFG_CTL, &msr);
1592 
1593 	pkg_cstate_limit = pkg_cstate_limits[msr & 0x7];
1594 
1595 	return 1;
1596 }
1597 int has_nhm_turbo_ratio_limit(unsigned int family, unsigned int model)
1598 {
1599 	switch (model) {
1600 	/* Nehalem compatible, but do not include turbo-ratio limit support */
1601 	case 0x2E:	/* Nehalem-EX Xeon - Beckton */
1602 	case 0x2F:	/* Westmere-EX Xeon - Eagleton */
1603 		return 0;
1604 	default:
1605 		return 1;
1606 	}
1607 }
1608 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
1609 {
1610 	if (!genuine_intel)
1611 		return 0;
1612 
1613 	if (family != 6)
1614 		return 0;
1615 
1616 	switch (model) {
1617 	case 0x3E:	/* IVB Xeon */
1618 		return 1;
1619 	default:
1620 		return 0;
1621 	}
1622 }
1623 
1624 /*
1625  * print_epb()
1626  * Decode the ENERGY_PERF_BIAS MSR
1627  */
1628 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1629 {
1630 	unsigned long long msr;
1631 	char *epb_string;
1632 	int cpu;
1633 
1634 	if (!has_epb)
1635 		return 0;
1636 
1637 	cpu = t->cpu_id;
1638 
1639 	/* EPB is per-package */
1640 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1641 		return 0;
1642 
1643 	if (cpu_migrate(cpu)) {
1644 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1645 		return -1;
1646 	}
1647 
1648 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
1649 		return 0;
1650 
1651 	switch (msr & 0x7) {
1652 	case ENERGY_PERF_BIAS_PERFORMANCE:
1653 		epb_string = "performance";
1654 		break;
1655 	case ENERGY_PERF_BIAS_NORMAL:
1656 		epb_string = "balanced";
1657 		break;
1658 	case ENERGY_PERF_BIAS_POWERSAVE:
1659 		epb_string = "powersave";
1660 		break;
1661 	default:
1662 		epb_string = "custom";
1663 		break;
1664 	}
1665 	fprintf(stderr, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
1666 
1667 	return 0;
1668 }
1669 
1670 /*
1671  * print_perf_limit()
1672  */
1673 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1674 {
1675 	unsigned long long msr;
1676 	int cpu;
1677 
1678 	cpu = t->cpu_id;
1679 
1680 	/* per-package */
1681 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1682 		return 0;
1683 
1684 	if (cpu_migrate(cpu)) {
1685 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1686 		return -1;
1687 	}
1688 
1689 	if (do_core_perf_limit_reasons) {
1690 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
1691 		fprintf(stderr, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
1692 		fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
1693 			(msr & 1 << 0) ? "PROCHOT, " : "",
1694 			(msr & 1 << 1) ? "ThermStatus, " : "",
1695 			(msr & 1 << 2) ? "bit2, " : "",
1696 			(msr & 1 << 4) ? "Graphics, " : "",
1697 			(msr & 1 << 5) ? "Auto-HWP, " : "",
1698 			(msr & 1 << 6) ? "VR-Therm, " : "",
1699 			(msr & 1 << 8) ? "Amps, " : "",
1700 			(msr & 1 << 9) ? "CorePwr, " : "",
1701 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
1702 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
1703 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
1704 			(msr & 1 << 13) ? "Transitions, " : "",
1705 			(msr & 1 << 14) ? "bit14, " : "",
1706 			(msr & 1 << 15) ? "bit15, " : "");
1707 		fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
1708 			(msr & 1 << 16) ? "PROCHOT, " : "",
1709 			(msr & 1 << 17) ? "ThermStatus, " : "",
1710 			(msr & 1 << 18) ? "bit18, " : "",
1711 			(msr & 1 << 20) ? "Graphics, " : "",
1712 			(msr & 1 << 21) ? "Auto-HWP, " : "",
1713 			(msr & 1 << 22) ? "VR-Therm, " : "",
1714 			(msr & 1 << 24) ? "Amps, " : "",
1715 			(msr & 1 << 25) ? "CorePwr, " : "",
1716 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
1717 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
1718 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
1719 			(msr & 1 << 29) ? "Transitions, " : "",
1720 			(msr & 1 << 30) ? "bit30, " : "",
1721 			(msr & 1 << 31) ? "bit31, " : "");
1722 
1723 	}
1724 	if (do_gfx_perf_limit_reasons) {
1725 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
1726 		fprintf(stderr, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
1727 		fprintf(stderr, " (Active: %s%s%s%s%s%s%s%s)",
1728 			(msr & 1 << 0) ? "PROCHOT, " : "",
1729 			(msr & 1 << 1) ? "ThermStatus, " : "",
1730 			(msr & 1 << 4) ? "Graphics, " : "",
1731 			(msr & 1 << 6) ? "VR-Therm, " : "",
1732 			(msr & 1 << 8) ? "Amps, " : "",
1733 			(msr & 1 << 9) ? "GFXPwr, " : "",
1734 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
1735 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
1736 		fprintf(stderr, " (Logged: %s%s%s%s%s%s%s%s)\n",
1737 			(msr & 1 << 16) ? "PROCHOT, " : "",
1738 			(msr & 1 << 17) ? "ThermStatus, " : "",
1739 			(msr & 1 << 20) ? "Graphics, " : "",
1740 			(msr & 1 << 22) ? "VR-Therm, " : "",
1741 			(msr & 1 << 24) ? "Amps, " : "",
1742 			(msr & 1 << 25) ? "GFXPwr, " : "",
1743 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
1744 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
1745 	}
1746 	if (do_ring_perf_limit_reasons) {
1747 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
1748 		fprintf(stderr, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
1749 		fprintf(stderr, " (Active: %s%s%s%s%s%s)",
1750 			(msr & 1 << 0) ? "PROCHOT, " : "",
1751 			(msr & 1 << 1) ? "ThermStatus, " : "",
1752 			(msr & 1 << 6) ? "VR-Therm, " : "",
1753 			(msr & 1 << 8) ? "Amps, " : "",
1754 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
1755 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
1756 		fprintf(stderr, " (Logged: %s%s%s%s%s%s)\n",
1757 			(msr & 1 << 16) ? "PROCHOT, " : "",
1758 			(msr & 1 << 17) ? "ThermStatus, " : "",
1759 			(msr & 1 << 22) ? "VR-Therm, " : "",
1760 			(msr & 1 << 24) ? "Amps, " : "",
1761 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
1762 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
1763 	}
1764 	return 0;
1765 }
1766 
1767 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
1768 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
1769 
1770 double get_tdp(model)
1771 {
1772 	unsigned long long msr;
1773 
1774 	if (do_rapl & RAPL_PKG_POWER_INFO)
1775 		if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
1776 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
1777 
1778 	switch (model) {
1779 	case 0x37:
1780 	case 0x4D:
1781 		return 30.0;
1782 	default:
1783 		return 135.0;
1784 	}
1785 }
1786 
1787 
1788 /*
1789  * rapl_probe()
1790  *
1791  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
1792  */
1793 void rapl_probe(unsigned int family, unsigned int model)
1794 {
1795 	unsigned long long msr;
1796 	unsigned int time_unit;
1797 	double tdp;
1798 
1799 	if (!genuine_intel)
1800 		return;
1801 
1802 	if (family != 6)
1803 		return;
1804 
1805 	switch (model) {
1806 	case 0x2A:
1807 	case 0x3A:
1808 	case 0x3C:	/* HSW */
1809 	case 0x45:	/* HSW */
1810 	case 0x46:	/* HSW */
1811 	case 0x3D:	/* BDW */
1812 	case 0x47:	/* BDW */
1813 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1814 		break;
1815 	case 0x3F:	/* HSX */
1816 	case 0x4F:	/* BDX */
1817 	case 0x56:	/* BDX-DE */
1818 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1819 		break;
1820 	case 0x2D:
1821 	case 0x3E:
1822 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1823 		break;
1824 	case 0x37:	/* BYT */
1825 	case 0x4D:	/* AVN */
1826 		do_rapl = RAPL_PKG | RAPL_CORES ;
1827 		break;
1828 	default:
1829 		return;
1830 	}
1831 
1832 	/* units on package 0, verify later other packages match */
1833 	if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1834 		return;
1835 
1836 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
1837 	if (model == 0x37)
1838 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1839 	else
1840 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1841 
1842 	time_unit = msr >> 16 & 0xF;
1843 	if (time_unit == 0)
1844 		time_unit = 0xA;
1845 
1846 	rapl_time_units = 1.0 / (1 << (time_unit));
1847 
1848 	tdp = get_tdp(model);
1849 
1850 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1851 	if (debug)
1852 		fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1853 
1854 	return;
1855 }
1856 
1857 void perf_limit_reasons_probe(family, model)
1858 {
1859 	if (!genuine_intel)
1860 		return;
1861 
1862 	if (family != 6)
1863 		return;
1864 
1865 	switch (model) {
1866 	case 0x3C:	/* HSW */
1867 	case 0x45:	/* HSW */
1868 	case 0x46:	/* HSW */
1869 		do_gfx_perf_limit_reasons = 1;
1870 	case 0x3F:	/* HSX */
1871 		do_core_perf_limit_reasons = 1;
1872 		do_ring_perf_limit_reasons = 1;
1873 	default:
1874 		return;
1875 	}
1876 }
1877 
1878 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1879 {
1880 	unsigned long long msr;
1881 	unsigned int dts;
1882 	int cpu;
1883 
1884 	if (!(do_dts || do_ptm))
1885 		return 0;
1886 
1887 	cpu = t->cpu_id;
1888 
1889 	/* DTS is per-core, no need to print for each thread */
1890 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1891 		return 0;
1892 
1893 	if (cpu_migrate(cpu)) {
1894 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1895 		return -1;
1896 	}
1897 
1898 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
1899 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1900 			return 0;
1901 
1902 		dts = (msr >> 16) & 0x7F;
1903 		fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
1904 			cpu, msr, tcc_activation_temp - dts);
1905 
1906 #ifdef	THERM_DEBUG
1907 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
1908 			return 0;
1909 
1910 		dts = (msr >> 16) & 0x7F;
1911 		dts2 = (msr >> 8) & 0x7F;
1912 		fprintf(stderr, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1913 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1914 #endif
1915 	}
1916 
1917 
1918 	if (do_dts) {
1919 		unsigned int resolution;
1920 
1921 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1922 			return 0;
1923 
1924 		dts = (msr >> 16) & 0x7F;
1925 		resolution = (msr >> 27) & 0xF;
1926 		fprintf(stderr, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
1927 			cpu, msr, tcc_activation_temp - dts, resolution);
1928 
1929 #ifdef THERM_DEBUG
1930 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
1931 			return 0;
1932 
1933 		dts = (msr >> 16) & 0x7F;
1934 		dts2 = (msr >> 8) & 0x7F;
1935 		fprintf(stderr, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
1936 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
1937 #endif
1938 	}
1939 
1940 	return 0;
1941 }
1942 
1943 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
1944 {
1945 	fprintf(stderr, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
1946 		cpu, label,
1947 		((msr >> 15) & 1) ? "EN" : "DIS",
1948 		((msr >> 0) & 0x7FFF) * rapl_power_units,
1949 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
1950 		(((msr >> 16) & 1) ? "EN" : "DIS"));
1951 
1952 	return;
1953 }
1954 
1955 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1956 {
1957 	unsigned long long msr;
1958 	int cpu;
1959 
1960 	if (!do_rapl)
1961 		return 0;
1962 
1963 	/* RAPL counters are per package, so print only for 1st thread/package */
1964 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1965 		return 0;
1966 
1967 	cpu = t->cpu_id;
1968 	if (cpu_migrate(cpu)) {
1969 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
1970 		return -1;
1971 	}
1972 
1973 	if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
1974 		return -1;
1975 
1976 	if (debug) {
1977 		fprintf(stderr, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx "
1978 			"(%f Watts, %f Joules, %f sec.)\n", cpu, msr,
1979 			rapl_power_units, rapl_energy_units, rapl_time_units);
1980 	}
1981 	if (do_rapl & RAPL_PKG_POWER_INFO) {
1982 
1983 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
1984                 	return -5;
1985 
1986 
1987 		fprintf(stderr, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
1988 			cpu, msr,
1989 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1990 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1991 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
1992 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
1993 
1994 	}
1995 	if (do_rapl & RAPL_PKG) {
1996 
1997 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
1998 			return -9;
1999 
2000 		fprintf(stderr, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
2001 			cpu, msr, (msr >> 63) & 1 ? "": "UN");
2002 
2003 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
2004 		fprintf(stderr, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
2005 			cpu,
2006 			((msr >> 47) & 1) ? "EN" : "DIS",
2007 			((msr >> 32) & 0x7FFF) * rapl_power_units,
2008 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
2009 			((msr >> 48) & 1) ? "EN" : "DIS");
2010 	}
2011 
2012 	if (do_rapl & RAPL_DRAM) {
2013 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
2014                 	return -6;
2015 
2016 
2017 		fprintf(stderr, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
2018 			cpu, msr,
2019 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2020 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2021 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
2022 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
2023 
2024 
2025 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
2026 			return -9;
2027 		fprintf(stderr, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
2028 				cpu, msr, (msr >> 31) & 1 ? "": "UN");
2029 
2030 		print_power_limit_msr(cpu, msr, "DRAM Limit");
2031 	}
2032 	if (do_rapl & RAPL_CORE_POLICY) {
2033 		if (debug) {
2034 			if (get_msr(cpu, MSR_PP0_POLICY, &msr))
2035 				return -7;
2036 
2037 			fprintf(stderr, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
2038 		}
2039 	}
2040 	if (do_rapl & RAPL_CORES) {
2041 		if (debug) {
2042 
2043 			if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
2044 				return -9;
2045 			fprintf(stderr, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
2046 					cpu, msr, (msr >> 31) & 1 ? "": "UN");
2047 			print_power_limit_msr(cpu, msr, "Cores Limit");
2048 		}
2049 	}
2050 	if (do_rapl & RAPL_GFX) {
2051 		if (debug) {
2052 			if (get_msr(cpu, MSR_PP1_POLICY, &msr))
2053 				return -8;
2054 
2055 			fprintf(stderr, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
2056 
2057 			if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
2058 				return -9;
2059 			fprintf(stderr, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
2060 					cpu, msr, (msr >> 31) & 1 ? "": "UN");
2061 			print_power_limit_msr(cpu, msr, "GFX Limit");
2062 		}
2063 	}
2064 	return 0;
2065 }
2066 
2067 /*
2068  * SNB adds support for additional MSRs:
2069  *
2070  * MSR_PKG_C7_RESIDENCY            0x000003fa
2071  * MSR_CORE_C7_RESIDENCY           0x000003fe
2072  * MSR_PKG_C2_RESIDENCY            0x0000060d
2073  */
2074 
2075 int has_snb_msrs(unsigned int family, unsigned int model)
2076 {
2077 	if (!genuine_intel)
2078 		return 0;
2079 
2080 	switch (model) {
2081 	case 0x2A:
2082 	case 0x2D:
2083 	case 0x3A:	/* IVB */
2084 	case 0x3E:	/* IVB Xeon */
2085 	case 0x3C:	/* HSW */
2086 	case 0x3F:	/* HSW */
2087 	case 0x45:	/* HSW */
2088 	case 0x46:	/* HSW */
2089 	case 0x3D:	/* BDW */
2090 	case 0x47:	/* BDW */
2091 	case 0x4F:	/* BDX */
2092 	case 0x56:	/* BDX-DE */
2093 		return 1;
2094 	}
2095 	return 0;
2096 }
2097 
2098 /*
2099  * HSW adds support for additional MSRs:
2100  *
2101  * MSR_PKG_C8_RESIDENCY            0x00000630
2102  * MSR_PKG_C9_RESIDENCY            0x00000631
2103  * MSR_PKG_C10_RESIDENCY           0x00000632
2104  */
2105 int has_hsw_msrs(unsigned int family, unsigned int model)
2106 {
2107 	if (!genuine_intel)
2108 		return 0;
2109 
2110 	switch (model) {
2111 	case 0x45:	/* HSW */
2112 	case 0x3D:	/* BDW */
2113 		return 1;
2114 	}
2115 	return 0;
2116 }
2117 
2118 
2119 int is_slm(unsigned int family, unsigned int model)
2120 {
2121 	if (!genuine_intel)
2122 		return 0;
2123 	switch (model) {
2124 	case 0x37:	/* BYT */
2125 	case 0x4D:	/* AVN */
2126 		return 1;
2127 	}
2128 	return 0;
2129 }
2130 
2131 #define SLM_BCLK_FREQS 5
2132 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
2133 
2134 double slm_bclk(void)
2135 {
2136 	unsigned long long msr = 3;
2137 	unsigned int i;
2138 	double freq;
2139 
2140 	if (get_msr(0, MSR_FSB_FREQ, &msr))
2141 		fprintf(stderr, "SLM BCLK: unknown\n");
2142 
2143 	i = msr & 0xf;
2144 	if (i >= SLM_BCLK_FREQS) {
2145 		fprintf(stderr, "SLM BCLK[%d] invalid\n", i);
2146 		msr = 3;
2147 	}
2148 	freq = slm_freq_table[i];
2149 
2150 	fprintf(stderr, "SLM BCLK: %.1f Mhz\n", freq);
2151 
2152 	return freq;
2153 }
2154 
2155 double discover_bclk(unsigned int family, unsigned int model)
2156 {
2157 	if (has_snb_msrs(family, model))
2158 		return 100.00;
2159 	else if (is_slm(family, model))
2160 		return slm_bclk();
2161 	else
2162 		return 133.33;
2163 }
2164 
2165 /*
2166  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
2167  * the Thermal Control Circuit (TCC) activates.
2168  * This is usually equal to tjMax.
2169  *
2170  * Older processors do not have this MSR, so there we guess,
2171  * but also allow cmdline over-ride with -T.
2172  *
2173  * Several MSR temperature values are in units of degrees-C
2174  * below this value, including the Digital Thermal Sensor (DTS),
2175  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
2176  */
2177 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2178 {
2179 	unsigned long long msr;
2180 	unsigned int target_c_local;
2181 	int cpu;
2182 
2183 	/* tcc_activation_temp is used only for dts or ptm */
2184 	if (!(do_dts || do_ptm))
2185 		return 0;
2186 
2187 	/* this is a per-package concept */
2188 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2189 		return 0;
2190 
2191 	cpu = t->cpu_id;
2192 	if (cpu_migrate(cpu)) {
2193 		fprintf(stderr, "Could not migrate to CPU %d\n", cpu);
2194 		return -1;
2195 	}
2196 
2197 	if (tcc_activation_temp_override != 0) {
2198 		tcc_activation_temp = tcc_activation_temp_override;
2199 		fprintf(stderr, "cpu%d: Using cmdline TCC Target (%d C)\n",
2200 			cpu, tcc_activation_temp);
2201 		return 0;
2202 	}
2203 
2204 	/* Temperature Target MSR is Nehalem and newer only */
2205 	if (!do_nhm_platform_info)
2206 		goto guess;
2207 
2208 	if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
2209 		goto guess;
2210 
2211 	target_c_local = (msr >> 16) & 0xFF;
2212 
2213 	if (debug)
2214 		fprintf(stderr, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
2215 			cpu, msr, target_c_local);
2216 
2217 	if (!target_c_local)
2218 		goto guess;
2219 
2220 	tcc_activation_temp = target_c_local;
2221 
2222 	return 0;
2223 
2224 guess:
2225 	tcc_activation_temp = TJMAX_DEFAULT;
2226 	fprintf(stderr, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
2227 		cpu, tcc_activation_temp);
2228 
2229 	return 0;
2230 }
2231 void check_cpuid()
2232 {
2233 	unsigned int eax, ebx, ecx, edx, max_level;
2234 	unsigned int fms, family, model, stepping;
2235 
2236 	eax = ebx = ecx = edx = 0;
2237 
2238 	__get_cpuid(0, &max_level, &ebx, &ecx, &edx);
2239 
2240 	if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
2241 		genuine_intel = 1;
2242 
2243 	if (debug)
2244 		fprintf(stderr, "CPUID(0): %.4s%.4s%.4s ",
2245 			(char *)&ebx, (char *)&edx, (char *)&ecx);
2246 
2247 	__get_cpuid(1, &fms, &ebx, &ecx, &edx);
2248 	family = (fms >> 8) & 0xf;
2249 	model = (fms >> 4) & 0xf;
2250 	stepping = fms & 0xf;
2251 	if (family == 6 || family == 0xf)
2252 		model += ((fms >> 16) & 0xf) << 4;
2253 
2254 	if (debug)
2255 		fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
2256 			max_level, family, model, stepping, family, model, stepping);
2257 
2258 	if (!(edx & (1 << 5)))
2259 		errx(1, "CPUID: no MSR");
2260 
2261 	/*
2262 	 * check max extended function levels of CPUID.
2263 	 * This is needed to check for invariant TSC.
2264 	 * This check is valid for both Intel and AMD.
2265 	 */
2266 	ebx = ecx = edx = 0;
2267 	__get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
2268 
2269 	if (max_level >= 0x80000007) {
2270 
2271 		/*
2272 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
2273 		 * this check is valid for both Intel and AMD
2274 		 */
2275 		__get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
2276 		has_invariant_tsc = edx & (1 << 8);
2277 	}
2278 
2279 	/*
2280 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
2281 	 * this check is valid for both Intel and AMD
2282 	 */
2283 
2284 	__get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
2285 	has_aperf = ecx & (1 << 0);
2286 	do_dts = eax & (1 << 0);
2287 	do_ptm = eax & (1 << 6);
2288 	has_epb = ecx & (1 << 3);
2289 
2290 	if (debug)
2291 		fprintf(stderr, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sEPB\n",
2292 			has_aperf ? "" : "No ",
2293 			do_dts ? "" : "No ",
2294 			do_ptm ? "" : "No ",
2295 			has_epb ? "" : "No ");
2296 
2297 	do_nhm_platform_info = do_nhm_cstates = do_smi = probe_nhm_msrs(family, model);
2298 	do_snb_cstates = has_snb_msrs(family, model);
2299 	do_pc2 = do_snb_cstates && (pkg_cstate_limit >= PCL__2);
2300 	do_pc3 = (pkg_cstate_limit >= PCL__3);
2301 	do_pc6 = (pkg_cstate_limit >= PCL__6);
2302 	do_pc7 = do_snb_cstates && (pkg_cstate_limit >= PCL__7);
2303 	do_c8_c9_c10 = has_hsw_msrs(family, model);
2304 	do_slm_cstates = is_slm(family, model);
2305 	bclk = discover_bclk(family, model);
2306 
2307 	do_nhm_turbo_ratio_limit = do_nhm_platform_info && has_nhm_turbo_ratio_limit(family, model);
2308 	do_ivt_turbo_ratio_limit = has_ivt_turbo_ratio_limit(family, model);
2309 	rapl_probe(family, model);
2310 	perf_limit_reasons_probe(family, model);
2311 
2312 	return;
2313 }
2314 
2315 
2316 void help()
2317 {
2318 	fprintf(stderr,
2319 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
2320 	"\n"
2321 	"Turbostat forks the specified COMMAND and prints statistics\n"
2322 	"when COMMAND completes.\n"
2323 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
2324 	"to print statistics, until interrupted.\n"
2325 	"--debug	run in \"debug\" mode\n"
2326 	"--interval sec	Override default 5-second measurement interval\n"
2327 	"--help		print this help message\n"
2328 	"--counter msr	print 32-bit counter at address \"msr\"\n"
2329 	"--Counter msr	print 64-bit Counter at address \"msr\"\n"
2330 	"--msr msr	print 32-bit value at address \"msr\"\n"
2331 	"--MSR msr	print 64-bit Value at address \"msr\"\n"
2332 	"--version	print version information\n"
2333 	"\n"
2334 	"For more help, run \"man turbostat\"\n");
2335 }
2336 
2337 
2338 /*
2339  * in /dev/cpu/ return success for names that are numbers
2340  * ie. filter out ".", "..", "microcode".
2341  */
2342 int dir_filter(const struct dirent *dirp)
2343 {
2344 	if (isdigit(dirp->d_name[0]))
2345 		return 1;
2346 	else
2347 		return 0;
2348 }
2349 
2350 int open_dev_cpu_msr(int dummy1)
2351 {
2352 	return 0;
2353 }
2354 
2355 void topology_probe()
2356 {
2357 	int i;
2358 	int max_core_id = 0;
2359 	int max_package_id = 0;
2360 	int max_siblings = 0;
2361 	struct cpu_topology {
2362 		int core_id;
2363 		int physical_package_id;
2364 	} *cpus;
2365 
2366 	/* Initialize num_cpus, max_cpu_num */
2367 	topo.num_cpus = 0;
2368 	topo.max_cpu_num = 0;
2369 	for_all_proc_cpus(count_cpus);
2370 	if (!summary_only && topo.num_cpus > 1)
2371 		show_cpu = 1;
2372 
2373 	if (debug > 1)
2374 		fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
2375 
2376 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
2377 	if (cpus == NULL)
2378 		err(1, "calloc cpus");
2379 
2380 	/*
2381 	 * Allocate and initialize cpu_present_set
2382 	 */
2383 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
2384 	if (cpu_present_set == NULL)
2385 		err(3, "CPU_ALLOC");
2386 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2387 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
2388 	for_all_proc_cpus(mark_cpu_present);
2389 
2390 	/*
2391 	 * Allocate and initialize cpu_affinity_set
2392 	 */
2393 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
2394 	if (cpu_affinity_set == NULL)
2395 		err(3, "CPU_ALLOC");
2396 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2397 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
2398 
2399 
2400 	/*
2401 	 * For online cpus
2402 	 * find max_core_id, max_package_id
2403 	 */
2404 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2405 		int siblings;
2406 
2407 		if (cpu_is_not_present(i)) {
2408 			if (debug > 1)
2409 				fprintf(stderr, "cpu%d NOT PRESENT\n", i);
2410 			continue;
2411 		}
2412 		cpus[i].core_id = get_core_id(i);
2413 		if (cpus[i].core_id > max_core_id)
2414 			max_core_id = cpus[i].core_id;
2415 
2416 		cpus[i].physical_package_id = get_physical_package_id(i);
2417 		if (cpus[i].physical_package_id > max_package_id)
2418 			max_package_id = cpus[i].physical_package_id;
2419 
2420 		siblings = get_num_ht_siblings(i);
2421 		if (siblings > max_siblings)
2422 			max_siblings = siblings;
2423 		if (debug > 1)
2424 			fprintf(stderr, "cpu %d pkg %d core %d\n",
2425 				i, cpus[i].physical_package_id, cpus[i].core_id);
2426 	}
2427 	topo.num_cores_per_pkg = max_core_id + 1;
2428 	if (debug > 1)
2429 		fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n",
2430 			max_core_id, topo.num_cores_per_pkg);
2431 	if (!summary_only && topo.num_cores_per_pkg > 1)
2432 		show_core = 1;
2433 
2434 	topo.num_packages = max_package_id + 1;
2435 	if (debug > 1)
2436 		fprintf(stderr, "max_package_id %d, sizing for %d packages\n",
2437 			max_package_id, topo.num_packages);
2438 	if (!summary_only && topo.num_packages > 1)
2439 		show_pkg = 1;
2440 
2441 	topo.num_threads_per_core = max_siblings;
2442 	if (debug > 1)
2443 		fprintf(stderr, "max_siblings %d\n", max_siblings);
2444 
2445 	free(cpus);
2446 }
2447 
2448 void
2449 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
2450 {
2451 	int i;
2452 
2453 	*t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
2454 		topo.num_packages, sizeof(struct thread_data));
2455 	if (*t == NULL)
2456 		goto error;
2457 
2458 	for (i = 0; i < topo.num_threads_per_core *
2459 		topo.num_cores_per_pkg * topo.num_packages; i++)
2460 		(*t)[i].cpu_id = -1;
2461 
2462 	*c = calloc(topo.num_cores_per_pkg * topo.num_packages,
2463 		sizeof(struct core_data));
2464 	if (*c == NULL)
2465 		goto error;
2466 
2467 	for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
2468 		(*c)[i].core_id = -1;
2469 
2470 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
2471 	if (*p == NULL)
2472 		goto error;
2473 
2474 	for (i = 0; i < topo.num_packages; i++)
2475 		(*p)[i].package_id = i;
2476 
2477 	return;
2478 error:
2479 	err(1, "calloc counters");
2480 }
2481 /*
2482  * init_counter()
2483  *
2484  * set cpu_id, core_num, pkg_num
2485  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
2486  *
2487  * increment topo.num_cores when 1st core in pkg seen
2488  */
2489 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
2490 	struct pkg_data *pkg_base, int thread_num, int core_num,
2491 	int pkg_num, int cpu_id)
2492 {
2493 	struct thread_data *t;
2494 	struct core_data *c;
2495 	struct pkg_data *p;
2496 
2497 	t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
2498 	c = GET_CORE(core_base, core_num, pkg_num);
2499 	p = GET_PKG(pkg_base, pkg_num);
2500 
2501 	t->cpu_id = cpu_id;
2502 	if (thread_num == 0) {
2503 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
2504 		if (cpu_is_first_core_in_package(cpu_id))
2505 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
2506 	}
2507 
2508 	c->core_id = core_num;
2509 	p->package_id = pkg_num;
2510 }
2511 
2512 
2513 int initialize_counters(int cpu_id)
2514 {
2515 	int my_thread_id, my_core_id, my_package_id;
2516 
2517 	my_package_id = get_physical_package_id(cpu_id);
2518 	my_core_id = get_core_id(cpu_id);
2519 
2520 	if (cpu_is_first_sibling_in_core(cpu_id)) {
2521 		my_thread_id = 0;
2522 		topo.num_cores++;
2523 	} else {
2524 		my_thread_id = 1;
2525 	}
2526 
2527 	init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2528 	init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
2529 	return 0;
2530 }
2531 
2532 void allocate_output_buffer()
2533 {
2534 	output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
2535 	outp = output_buffer;
2536 	if (outp == NULL)
2537 		err(-1, "calloc output buffer");
2538 }
2539 
2540 void setup_all_buffers(void)
2541 {
2542 	topology_probe();
2543 	allocate_counters(&thread_even, &core_even, &package_even);
2544 	allocate_counters(&thread_odd, &core_odd, &package_odd);
2545 	allocate_output_buffer();
2546 	for_all_proc_cpus(initialize_counters);
2547 }
2548 
2549 void turbostat_init()
2550 {
2551 	check_dev_msr();
2552 	check_permissions();
2553 	check_cpuid();
2554 
2555 	setup_all_buffers();
2556 
2557 	if (debug)
2558 		print_verbose_header();
2559 
2560 	if (debug)
2561 		for_all_cpus(print_epb, ODD_COUNTERS);
2562 
2563 	if (debug)
2564 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
2565 
2566 	if (debug)
2567 		for_all_cpus(print_rapl, ODD_COUNTERS);
2568 
2569 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
2570 
2571 	if (debug)
2572 		for_all_cpus(print_thermal, ODD_COUNTERS);
2573 }
2574 
2575 int fork_it(char **argv)
2576 {
2577 	pid_t child_pid;
2578 	int status;
2579 
2580 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
2581 	if (status)
2582 		exit(status);
2583 	/* clear affinity side-effect of get_counters() */
2584 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
2585 	gettimeofday(&tv_even, (struct timezone *)NULL);
2586 
2587 	child_pid = fork();
2588 	if (!child_pid) {
2589 		/* child */
2590 		execvp(argv[0], argv);
2591 	} else {
2592 
2593 		/* parent */
2594 		if (child_pid == -1)
2595 			err(1, "fork");
2596 
2597 		signal(SIGINT, SIG_IGN);
2598 		signal(SIGQUIT, SIG_IGN);
2599 		if (waitpid(child_pid, &status, 0) == -1)
2600 			err(status, "waitpid");
2601 	}
2602 	/*
2603 	 * n.b. fork_it() does not check for errors from for_all_cpus()
2604 	 * because re-starting is problematic when forking
2605 	 */
2606 	for_all_cpus(get_counters, ODD_COUNTERS);
2607 	gettimeofday(&tv_odd, (struct timezone *)NULL);
2608 	timersub(&tv_odd, &tv_even, &tv_delta);
2609 	for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS);
2610 	compute_average(EVEN_COUNTERS);
2611 	format_all_counters(EVEN_COUNTERS);
2612 	flush_stderr();
2613 
2614 	fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
2615 
2616 	return status;
2617 }
2618 
2619 int get_and_dump_counters(void)
2620 {
2621 	int status;
2622 
2623 	status = for_all_cpus(get_counters, ODD_COUNTERS);
2624 	if (status)
2625 		return status;
2626 
2627 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
2628 	if (status)
2629 		return status;
2630 
2631 	flush_stdout();
2632 
2633 	return status;
2634 }
2635 
2636 void print_version() {
2637 	fprintf(stderr, "turbostat version 4.1 10-Feb, 2015"
2638 		" - Len Brown <lenb@kernel.org>\n");
2639 }
2640 
2641 void cmdline(int argc, char **argv)
2642 {
2643 	int opt;
2644 	int option_index = 0;
2645 	static struct option long_options[] = {
2646 		{"Counter",	required_argument,	0, 'C'},
2647 		{"counter",	required_argument,	0, 'c'},
2648 		{"Dump",	no_argument,		0, 'D'},
2649 		{"debug",	no_argument,		0, 'd'},
2650 		{"interval",	required_argument,	0, 'i'},
2651 		{"help",	no_argument,		0, 'h'},
2652 		{"Joules",	no_argument,		0, 'J'},
2653 		{"MSR",		required_argument,	0, 'M'},
2654 		{"msr",		required_argument,	0, 'm'},
2655 		{"Package",	no_argument,		0, 'p'},
2656 		{"processor",	no_argument,		0, 'p'},
2657 		{"Summary",	no_argument,		0, 'S'},
2658 		{"TCC",		required_argument,	0, 'T'},
2659 		{"version",	no_argument,		0, 'v' },
2660 		{0,		0,			0,  0 }
2661 	};
2662 
2663 	progname = argv[0];
2664 
2665 	while ((opt = getopt_long_only(argc, argv, "C:c:Ddhi:JM:m:PpST:v",
2666 				long_options, &option_index)) != -1) {
2667 		switch (opt) {
2668 		case 'C':
2669 			sscanf(optarg, "%x", &extra_delta_offset64);
2670 			break;
2671 		case 'c':
2672 			sscanf(optarg, "%x", &extra_delta_offset32);
2673 			break;
2674 		case 'D':
2675 			dump_only++;
2676 			break;
2677 		case 'd':
2678 			debug++;
2679 			break;
2680 		case 'h':
2681 		default:
2682 			help();
2683 			exit(1);
2684 		case 'i':
2685 			interval_sec = atoi(optarg);
2686 			break;
2687 		case 'J':
2688 			rapl_joules++;
2689 			break;
2690 		case 'M':
2691 			sscanf(optarg, "%x", &extra_msr_offset64);
2692 			break;
2693 		case 'm':
2694 			sscanf(optarg, "%x", &extra_msr_offset32);
2695 			break;
2696 		case 'P':
2697 			show_pkg_only++;
2698 			break;
2699 		case 'p':
2700 			show_core_only++;
2701 			break;
2702 		case 'S':
2703 			summary_only++;
2704 			break;
2705 		case 'T':
2706 			tcc_activation_temp_override = atoi(optarg);
2707 			break;
2708 		case 'v':
2709 			print_version();
2710 			exit(0);
2711 			break;
2712 		}
2713 	}
2714 }
2715 
2716 int main(int argc, char **argv)
2717 {
2718 	cmdline(argc, argv);
2719 
2720 	if (debug)
2721 		print_version();
2722 
2723 	turbostat_init();
2724 
2725 	/* dump counters and exit */
2726 	if (dump_only)
2727 		return get_and_dump_counters();
2728 
2729 	/*
2730 	 * if any params left, it must be a command to fork
2731 	 */
2732 	if (argc - optind)
2733 		return fork_it(argv + optind);
2734 	else
2735 		turbostat_loop();
2736 
2737 	return 0;
2738 }
2739