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 INTEL_FAMILY_HEADER
25 #include <stdarg.h>
26 #include <stdio.h>
27 #include <err.h>
28 #include <unistd.h>
29 #include <sys/types.h>
30 #include <sys/wait.h>
31 #include <sys/stat.h>
32 #include <sys/resource.h>
33 #include <fcntl.h>
34 #include <signal.h>
35 #include <sys/time.h>
36 #include <stdlib.h>
37 #include <getopt.h>
38 #include <dirent.h>
39 #include <string.h>
40 #include <ctype.h>
41 #include <sched.h>
42 #include <time.h>
43 #include <cpuid.h>
44 #include <linux/capability.h>
45 #include <errno.h>
46 
47 char *proc_stat = "/proc/stat";
48 FILE *outf;
49 int *fd_percpu;
50 struct timespec interval_ts = {5, 0};
51 unsigned int debug;
52 unsigned int quiet;
53 unsigned int rapl_joules;
54 unsigned int summary_only;
55 unsigned int dump_only;
56 unsigned int do_snb_cstates;
57 unsigned int do_knl_cstates;
58 unsigned int do_skl_residency;
59 unsigned int do_slm_cstates;
60 unsigned int use_c1_residency_msr;
61 unsigned int has_aperf;
62 unsigned int has_epb;
63 unsigned int do_irtl_snb;
64 unsigned int do_irtl_hsw;
65 unsigned int units = 1000000;	/* MHz etc */
66 unsigned int genuine_intel;
67 unsigned int has_invariant_tsc;
68 unsigned int do_nhm_platform_info;
69 unsigned int no_MSR_MISC_PWR_MGMT;
70 unsigned int aperf_mperf_multiplier = 1;
71 double bclk;
72 double base_hz;
73 unsigned int has_base_hz;
74 double tsc_tweak = 1.0;
75 unsigned int show_pkg_only;
76 unsigned int show_core_only;
77 char *output_buffer, *outp;
78 unsigned int do_rapl;
79 unsigned int do_dts;
80 unsigned int do_ptm;
81 unsigned long long  gfx_cur_rc6_ms;
82 unsigned int gfx_cur_mhz;
83 unsigned int tcc_activation_temp;
84 unsigned int tcc_activation_temp_override;
85 double rapl_power_units, rapl_time_units;
86 double rapl_dram_energy_units, rapl_energy_units;
87 double rapl_joule_counter_range;
88 unsigned int do_core_perf_limit_reasons;
89 unsigned int do_gfx_perf_limit_reasons;
90 unsigned int do_ring_perf_limit_reasons;
91 unsigned int crystal_hz;
92 unsigned long long tsc_hz;
93 int base_cpu;
94 double discover_bclk(unsigned int family, unsigned int model);
95 unsigned int has_hwp;	/* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
96 			/* IA32_HWP_REQUEST, IA32_HWP_STATUS */
97 unsigned int has_hwp_notify;		/* IA32_HWP_INTERRUPT */
98 unsigned int has_hwp_activity_window;	/* IA32_HWP_REQUEST[bits 41:32] */
99 unsigned int has_hwp_epp;		/* IA32_HWP_REQUEST[bits 31:24] */
100 unsigned int has_hwp_pkg;		/* IA32_HWP_REQUEST_PKG */
101 unsigned int has_misc_feature_control;
102 
103 #define RAPL_PKG		(1 << 0)
104 					/* 0x610 MSR_PKG_POWER_LIMIT */
105 					/* 0x611 MSR_PKG_ENERGY_STATUS */
106 #define RAPL_PKG_PERF_STATUS	(1 << 1)
107 					/* 0x613 MSR_PKG_PERF_STATUS */
108 #define RAPL_PKG_POWER_INFO	(1 << 2)
109 					/* 0x614 MSR_PKG_POWER_INFO */
110 
111 #define RAPL_DRAM		(1 << 3)
112 					/* 0x618 MSR_DRAM_POWER_LIMIT */
113 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
114 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
115 					/* 0x61b MSR_DRAM_PERF_STATUS */
116 #define RAPL_DRAM_POWER_INFO	(1 << 5)
117 					/* 0x61c MSR_DRAM_POWER_INFO */
118 
119 #define RAPL_CORES_POWER_LIMIT	(1 << 6)
120 					/* 0x638 MSR_PP0_POWER_LIMIT */
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 
129 #define RAPL_CORES_ENERGY_STATUS	(1 << 9)
130 					/* 0x639 MSR_PP0_ENERGY_STATUS */
131 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
132 #define	TJMAX_DEFAULT	100
133 
134 #define MAX(a, b) ((a) > (b) ? (a) : (b))
135 
136 /*
137  * buffer size used by sscanf() for added column names
138  * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
139  */
140 #define	NAME_BYTES 20
141 
142 int backwards_count;
143 char *progname;
144 
145 cpu_set_t *cpu_present_set, *cpu_affinity_set;
146 size_t cpu_present_setsize, cpu_affinity_setsize;
147 #define MAX_ADDED_COUNTERS 16
148 
149 struct thread_data {
150 	unsigned long long tsc;
151 	unsigned long long aperf;
152 	unsigned long long mperf;
153 	unsigned long long c1;
154 	unsigned int irq_count;
155 	unsigned int smi_count;
156 	unsigned int cpu_id;
157 	unsigned int flags;
158 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
159 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
160 	unsigned long long counter[MAX_ADDED_COUNTERS];
161 } *thread_even, *thread_odd;
162 
163 struct core_data {
164 	unsigned long long c3;
165 	unsigned long long c6;
166 	unsigned long long c7;
167 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
168 	unsigned int core_temp_c;
169 	unsigned int core_id;
170 	unsigned long long counter[MAX_ADDED_COUNTERS];
171 } *core_even, *core_odd;
172 
173 struct pkg_data {
174 	unsigned long long pc2;
175 	unsigned long long pc3;
176 	unsigned long long pc6;
177 	unsigned long long pc7;
178 	unsigned long long pc8;
179 	unsigned long long pc9;
180 	unsigned long long pc10;
181 	unsigned long long pkg_wtd_core_c0;
182 	unsigned long long pkg_any_core_c0;
183 	unsigned long long pkg_any_gfxe_c0;
184 	unsigned long long pkg_both_core_gfxe_c0;
185 	long long gfx_rc6_ms;
186 	unsigned int gfx_mhz;
187 	unsigned int package_id;
188 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
189 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
190 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
191 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
192 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
193 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
194 	unsigned int pkg_temp_c;
195 	unsigned long long counter[MAX_ADDED_COUNTERS];
196 } *package_even, *package_odd;
197 
198 #define ODD_COUNTERS thread_odd, core_odd, package_odd
199 #define EVEN_COUNTERS thread_even, core_even, package_even
200 
201 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
202 	(thread_base + (pkg_no) * topo.num_cores_per_pkg * \
203 		topo.num_threads_per_core + \
204 		(core_no) * topo.num_threads_per_core + (thread_no))
205 #define GET_CORE(core_base, core_no, pkg_no) \
206 	(core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
207 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
208 
209 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
210 enum counter_type {COUNTER_CYCLES, COUNTER_SECONDS};
211 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
212 
213 struct msr_counter {
214 	unsigned int msr_num;
215 	char name[NAME_BYTES];
216 	unsigned int width;
217 	enum counter_type type;
218 	enum counter_format format;
219 	struct msr_counter *next;
220 	unsigned int flags;
221 #define	FLAGS_HIDE	(1 << 0)
222 #define	FLAGS_SHOW	(1 << 1)
223 };
224 
225 struct sys_counters {
226 	unsigned int added_thread_counters;
227 	unsigned int added_core_counters;
228 	unsigned int added_package_counters;
229 	struct msr_counter *tp;
230 	struct msr_counter *cp;
231 	struct msr_counter *pp;
232 } sys;
233 
234 struct system_summary {
235 	struct thread_data threads;
236 	struct core_data cores;
237 	struct pkg_data packages;
238 } average;
239 
240 
241 struct topo_params {
242 	int num_packages;
243 	int num_cpus;
244 	int num_cores;
245 	int max_cpu_num;
246 	int num_cores_per_pkg;
247 	int num_threads_per_core;
248 } topo;
249 
250 struct timeval tv_even, tv_odd, tv_delta;
251 
252 int *irq_column_2_cpu;	/* /proc/interrupts column numbers */
253 int *irqs_per_cpu;		/* indexed by cpu_num */
254 
255 void setup_all_buffers(void);
256 
257 int cpu_is_not_present(int cpu)
258 {
259 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
260 }
261 /*
262  * run func(thread, core, package) in topology order
263  * skip non-present cpus
264  */
265 
266 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
267 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
268 {
269 	int retval, pkg_no, core_no, thread_no;
270 
271 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
272 		for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
273 			for (thread_no = 0; thread_no <
274 				topo.num_threads_per_core; ++thread_no) {
275 				struct thread_data *t;
276 				struct core_data *c;
277 				struct pkg_data *p;
278 
279 				t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
280 
281 				if (cpu_is_not_present(t->cpu_id))
282 					continue;
283 
284 				c = GET_CORE(core_base, core_no, pkg_no);
285 				p = GET_PKG(pkg_base, pkg_no);
286 
287 				retval = func(t, c, p);
288 				if (retval)
289 					return retval;
290 			}
291 		}
292 	}
293 	return 0;
294 }
295 
296 int cpu_migrate(int cpu)
297 {
298 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
299 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
300 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
301 		return -1;
302 	else
303 		return 0;
304 }
305 int get_msr_fd(int cpu)
306 {
307 	char pathname[32];
308 	int fd;
309 
310 	fd = fd_percpu[cpu];
311 
312 	if (fd)
313 		return fd;
314 
315 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
316 	fd = open(pathname, O_RDONLY);
317 	if (fd < 0)
318 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
319 
320 	fd_percpu[cpu] = fd;
321 
322 	return fd;
323 }
324 
325 int get_msr(int cpu, off_t offset, unsigned long long *msr)
326 {
327 	ssize_t retval;
328 
329 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
330 
331 	if (retval != sizeof *msr)
332 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
333 
334 	return 0;
335 }
336 
337 /*
338  * Each string in this array is compared in --show and --hide cmdline.
339  * Thus, strings that are proper sub-sets must follow their more specific peers.
340  */
341 struct msr_counter bic[] = {
342 	{ 0x0, "Package" },
343 	{ 0x0, "Avg_MHz" },
344 	{ 0x0, "Bzy_MHz" },
345 	{ 0x0, "TSC_MHz" },
346 	{ 0x0, "IRQ" },
347 	{ 0x0, "SMI", 32, 0, FORMAT_DELTA, NULL},
348 	{ 0x0, "Busy%" },
349 	{ 0x0, "CPU%c1" },
350 	{ 0x0, "CPU%c3" },
351 	{ 0x0, "CPU%c6" },
352 	{ 0x0, "CPU%c7" },
353 	{ 0x0, "ThreadC" },
354 	{ 0x0, "CoreTmp" },
355 	{ 0x0, "CoreCnt" },
356 	{ 0x0, "PkgTmp" },
357 	{ 0x0, "GFX%rc6" },
358 	{ 0x0, "GFXMHz" },
359 	{ 0x0, "Pkg%pc2" },
360 	{ 0x0, "Pkg%pc3" },
361 	{ 0x0, "Pkg%pc6" },
362 	{ 0x0, "Pkg%pc7" },
363 	{ 0x0, "Pkg%pc8" },
364 	{ 0x0, "Pkg%pc9" },
365 	{ 0x0, "Pkg%pc10" },
366 	{ 0x0, "PkgWatt" },
367 	{ 0x0, "CorWatt" },
368 	{ 0x0, "GFXWatt" },
369 	{ 0x0, "PkgCnt" },
370 	{ 0x0, "RAMWatt" },
371 	{ 0x0, "PKG_%" },
372 	{ 0x0, "RAM_%" },
373 	{ 0x0, "Pkg_J" },
374 	{ 0x0, "Cor_J" },
375 	{ 0x0, "GFX_J" },
376 	{ 0x0, "RAM_J" },
377 	{ 0x0, "Core" },
378 	{ 0x0, "CPU" },
379 	{ 0x0, "Mod%c6" },
380 };
381 
382 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
383 #define	BIC_Package	(1ULL << 0)
384 #define	BIC_Avg_MHz	(1ULL << 1)
385 #define	BIC_Bzy_MHz	(1ULL << 2)
386 #define	BIC_TSC_MHz	(1ULL << 3)
387 #define	BIC_IRQ		(1ULL << 4)
388 #define	BIC_SMI		(1ULL << 5)
389 #define	BIC_Busy	(1ULL << 6)
390 #define	BIC_CPU_c1	(1ULL << 7)
391 #define	BIC_CPU_c3	(1ULL << 8)
392 #define	BIC_CPU_c6	(1ULL << 9)
393 #define	BIC_CPU_c7	(1ULL << 10)
394 #define	BIC_ThreadC	(1ULL << 11)
395 #define	BIC_CoreTmp	(1ULL << 12)
396 #define	BIC_CoreCnt	(1ULL << 13)
397 #define	BIC_PkgTmp	(1ULL << 14)
398 #define	BIC_GFX_rc6	(1ULL << 15)
399 #define	BIC_GFXMHz	(1ULL << 16)
400 #define	BIC_Pkgpc2	(1ULL << 17)
401 #define	BIC_Pkgpc3	(1ULL << 18)
402 #define	BIC_Pkgpc6	(1ULL << 19)
403 #define	BIC_Pkgpc7	(1ULL << 20)
404 #define	BIC_Pkgpc8	(1ULL << 21)
405 #define	BIC_Pkgpc9	(1ULL << 22)
406 #define	BIC_Pkgpc10	(1ULL << 23)
407 #define	BIC_PkgWatt	(1ULL << 24)
408 #define	BIC_CorWatt	(1ULL << 25)
409 #define	BIC_GFXWatt	(1ULL << 26)
410 #define	BIC_PkgCnt	(1ULL << 27)
411 #define	BIC_RAMWatt	(1ULL << 28)
412 #define	BIC_PKG__	(1ULL << 29)
413 #define	BIC_RAM__	(1ULL << 30)
414 #define	BIC_Pkg_J	(1ULL << 31)
415 #define	BIC_Cor_J	(1ULL << 32)
416 #define	BIC_GFX_J	(1ULL << 33)
417 #define	BIC_RAM_J	(1ULL << 34)
418 #define	BIC_Core	(1ULL << 35)
419 #define	BIC_CPU		(1ULL << 36)
420 #define	BIC_Mod_c6	(1ULL << 37)
421 
422 unsigned long long bic_enabled = 0xFFFFFFFFFFFFFFFFULL;
423 unsigned long long bic_present;
424 
425 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
426 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
427 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
428 
429 /*
430  * bic_lookup
431  * for all the strings in comma separate name_list,
432  * set the approprate bit in return value.
433  */
434 unsigned long long bic_lookup(char *name_list)
435 {
436 	int i;
437 	unsigned long long retval = 0;
438 
439 	while (name_list) {
440 		char *comma;
441 
442 		comma = strchr(name_list, ',');
443 
444 		if (comma)
445 			*comma = '\0';
446 
447 		for (i = 0; i < MAX_BIC; ++i) {
448 			if (!strcmp(name_list, bic[i].name)) {
449 				retval |= (1ULL << i);
450 				break;
451 			}
452 		}
453 		if (i == MAX_BIC) {
454 			fprintf(stderr, "Invalid counter name: %s\n", name_list);
455 			exit(-1);
456 		}
457 
458 		name_list = comma;
459 		if (name_list)
460 			name_list++;
461 
462 	}
463 	return retval;
464 }
465 
466 void print_header(void)
467 {
468 	struct msr_counter *mp;
469 
470 	if (DO_BIC(BIC_Package))
471 		outp += sprintf(outp, "\tPackage");
472 	if (DO_BIC(BIC_Core))
473 		outp += sprintf(outp, "\tCore");
474 	if (DO_BIC(BIC_CPU))
475 		outp += sprintf(outp, "\tCPU");
476 	if (DO_BIC(BIC_Avg_MHz))
477 		outp += sprintf(outp, "\tAvg_MHz");
478 	if (DO_BIC(BIC_Busy))
479 		outp += sprintf(outp, "\tBusy%%");
480 	if (DO_BIC(BIC_Bzy_MHz))
481 		outp += sprintf(outp, "\tBzy_MHz");
482 	if (DO_BIC(BIC_TSC_MHz))
483 		outp += sprintf(outp, "\tTSC_MHz");
484 
485 	if (DO_BIC(BIC_IRQ))
486 		outp += sprintf(outp, "\tIRQ");
487 	if (DO_BIC(BIC_SMI))
488 		outp += sprintf(outp, "\tSMI");
489 
490 	if (DO_BIC(BIC_CPU_c1))
491 		outp += sprintf(outp, "\tCPU%%c1");
492 
493 	for (mp = sys.tp; mp; mp = mp->next) {
494 		if (mp->format == FORMAT_RAW) {
495 			if (mp->width == 64)
496 				outp += sprintf(outp, "\t%18.18s", mp->name);
497 			else
498 				outp += sprintf(outp, "\t%10.10s", mp->name);
499 		} else {
500 			outp += sprintf(outp, "\t%-7.7s", mp->name);
501 		}
502 	}
503 
504 	if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates)
505 		outp += sprintf(outp, "\tCPU%%c3");
506 	if (DO_BIC(BIC_CPU_c6))
507 		outp += sprintf(outp, "\tCPU%%c6");
508 	if (DO_BIC(BIC_CPU_c7))
509 		outp += sprintf(outp, "\tCPU%%c7");
510 
511 	if (DO_BIC(BIC_Mod_c6))
512 		outp += sprintf(outp, "\tMod%%c6");
513 
514 	if (DO_BIC(BIC_CoreTmp))
515 		outp += sprintf(outp, "\tCoreTmp");
516 
517 	for (mp = sys.cp; mp; mp = mp->next) {
518 		if (mp->format == FORMAT_RAW) {
519 			if (mp->width == 64)
520 				outp += sprintf(outp, "\t%18.18s", mp->name);
521 			else
522 				outp += sprintf(outp, "\t%10.10s", mp->name);
523 		} else {
524 			outp += sprintf(outp, "\t%-7.7s", mp->name);
525 		}
526 	}
527 
528 	if (DO_BIC(BIC_PkgTmp))
529 		outp += sprintf(outp, "\tPkgTmp");
530 
531 	if (DO_BIC(BIC_GFX_rc6))
532 		outp += sprintf(outp, "\tGFX%%rc6");
533 
534 	if (DO_BIC(BIC_GFXMHz))
535 		outp += sprintf(outp, "\tGFXMHz");
536 
537 	if (do_skl_residency) {
538 		outp += sprintf(outp, "\tTotl%%C0");
539 		outp += sprintf(outp, "\tAny%%C0");
540 		outp += sprintf(outp, "\tGFX%%C0");
541 		outp += sprintf(outp, "\tCPUGFX%%");
542 	}
543 
544 	if (DO_BIC(BIC_Pkgpc2))
545 		outp += sprintf(outp, "\tPkg%%pc2");
546 	if (DO_BIC(BIC_Pkgpc3))
547 		outp += sprintf(outp, "\tPkg%%pc3");
548 	if (DO_BIC(BIC_Pkgpc6))
549 		outp += sprintf(outp, "\tPkg%%pc6");
550 	if (DO_BIC(BIC_Pkgpc7))
551 		outp += sprintf(outp, "\tPkg%%pc7");
552 	if (DO_BIC(BIC_Pkgpc8))
553 		outp += sprintf(outp, "\tPkg%%pc8");
554 	if (DO_BIC(BIC_Pkgpc9))
555 		outp += sprintf(outp, "\tPkg%%pc9");
556 	if (DO_BIC(BIC_Pkgpc10))
557 		outp += sprintf(outp, "\tPk%%pc10");
558 
559 	if (do_rapl && !rapl_joules) {
560 		if (DO_BIC(BIC_PkgWatt))
561 			outp += sprintf(outp, "\tPkgWatt");
562 		if (DO_BIC(BIC_CorWatt))
563 			outp += sprintf(outp, "\tCorWatt");
564 		if (DO_BIC(BIC_GFXWatt))
565 			outp += sprintf(outp, "\tGFXWatt");
566 		if (DO_BIC(BIC_RAMWatt))
567 			outp += sprintf(outp, "\tRAMWatt");
568 		if (DO_BIC(BIC_PKG__))
569 			outp += sprintf(outp, "\tPKG_%%");
570 		if (DO_BIC(BIC_RAM__))
571 			outp += sprintf(outp, "\tRAM_%%");
572 	} else if (do_rapl && rapl_joules) {
573 		if (DO_BIC(BIC_Pkg_J))
574 			outp += sprintf(outp, "\tPkg_J");
575 		if (DO_BIC(BIC_Cor_J))
576 			outp += sprintf(outp, "\tCor_J");
577 		if (DO_BIC(BIC_GFX_J))
578 			outp += sprintf(outp, "\tGFX_J");
579 		if (DO_BIC(BIC_RAM_J))
580 			outp += sprintf(outp, "\tRAM_J");
581 		if (DO_BIC(BIC_PKG__))
582 			outp += sprintf(outp, "\tPKG_%%");
583 		if (DO_BIC(BIC_RAM__))
584 			outp += sprintf(outp, "\tRAM_%%");
585 	}
586 	for (mp = sys.pp; mp; mp = mp->next) {
587 		if (mp->format == FORMAT_RAW) {
588 			if (mp->width == 64)
589 				outp += sprintf(outp, "\t%18.18s", mp->name);
590 			else
591 				outp += sprintf(outp, "\t%10.10s", mp->name);
592 		} else {
593 			outp += sprintf(outp, "\t%-7.7s", mp->name);
594 		}
595 	}
596 
597 	outp += sprintf(outp, "\n");
598 }
599 
600 int dump_counters(struct thread_data *t, struct core_data *c,
601 	struct pkg_data *p)
602 {
603 	int i;
604 	struct msr_counter *mp;
605 
606 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
607 
608 	if (t) {
609 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
610 			t->cpu_id, t->flags);
611 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
612 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
613 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
614 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
615 
616 		if (DO_BIC(BIC_IRQ))
617 			outp += sprintf(outp, "IRQ: %08X\n", t->irq_count);
618 		if (DO_BIC(BIC_SMI))
619 			outp += sprintf(outp, "SMI: %08X\n", t->smi_count);
620 
621 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
622 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
623 				i, mp->msr_num, t->counter[i]);
624 		}
625 	}
626 
627 	if (c) {
628 		outp += sprintf(outp, "core: %d\n", c->core_id);
629 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
630 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
631 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
632 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
633 
634 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
635 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
636 				i, mp->msr_num, c->counter[i]);
637 		}
638 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
639 	}
640 
641 	if (p) {
642 		outp += sprintf(outp, "package: %d\n", p->package_id);
643 
644 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
645 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
646 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
647 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
648 
649 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
650 		if (DO_BIC(BIC_Pkgpc3))
651 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
652 		if (DO_BIC(BIC_Pkgpc6))
653 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
654 		if (DO_BIC(BIC_Pkgpc7))
655 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
656 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
657 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
658 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
659 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
660 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
661 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
662 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
663 		outp += sprintf(outp, "Throttle PKG: %0X\n",
664 			p->rapl_pkg_perf_status);
665 		outp += sprintf(outp, "Throttle RAM: %0X\n",
666 			p->rapl_dram_perf_status);
667 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
668 
669 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
670 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
671 				i, mp->msr_num, p->counter[i]);
672 		}
673 	}
674 
675 	outp += sprintf(outp, "\n");
676 
677 	return 0;
678 }
679 
680 /*
681  * column formatting convention & formats
682  */
683 int format_counters(struct thread_data *t, struct core_data *c,
684 	struct pkg_data *p)
685 {
686 	double interval_float, tsc;
687 	char *fmt8;
688 	int i;
689 	struct msr_counter *mp;
690 
691 	 /* if showing only 1st thread in core and this isn't one, bail out */
692 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
693 		return 0;
694 
695 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
696 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
697 		return 0;
698 
699 	interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
700 
701 	tsc = t->tsc * tsc_tweak;
702 
703 	/* topo columns, print blanks on 1st (average) line */
704 	if (t == &average.threads) {
705 		if (DO_BIC(BIC_Package))
706 			outp += sprintf(outp, "\t-");
707 		if (DO_BIC(BIC_Core))
708 			outp += sprintf(outp, "\t-");
709 		if (DO_BIC(BIC_CPU))
710 			outp += sprintf(outp, "\t-");
711 	} else {
712 		if (DO_BIC(BIC_Package)) {
713 			if (p)
714 				outp += sprintf(outp, "\t%d", p->package_id);
715 			else
716 				outp += sprintf(outp, "\t-");
717 		}
718 		if (DO_BIC(BIC_Core)) {
719 			if (c)
720 				outp += sprintf(outp, "\t%d", c->core_id);
721 			else
722 				outp += sprintf(outp, "\t-");
723 		}
724 		if (DO_BIC(BIC_CPU))
725 			outp += sprintf(outp, "\t%d", t->cpu_id);
726 	}
727 
728 	if (DO_BIC(BIC_Avg_MHz))
729 		outp += sprintf(outp, "\t%.0f",
730 			1.0 / units * t->aperf / interval_float);
731 
732 	if (DO_BIC(BIC_Busy))
733 		outp += sprintf(outp, "\t%.2f", 100.0 * t->mperf/tsc);
734 
735 	if (DO_BIC(BIC_Bzy_MHz)) {
736 		if (has_base_hz)
737 			outp += sprintf(outp, "\t%.0f", base_hz / units * t->aperf / t->mperf);
738 		else
739 			outp += sprintf(outp, "\t%.0f",
740 				tsc / units * t->aperf / t->mperf / interval_float);
741 	}
742 
743 	if (DO_BIC(BIC_TSC_MHz))
744 		outp += sprintf(outp, "\t%.0f", 1.0 * t->tsc/units/interval_float);
745 
746 	/* IRQ */
747 	if (DO_BIC(BIC_IRQ))
748 		outp += sprintf(outp, "\t%d", t->irq_count);
749 
750 	/* SMI */
751 	if (DO_BIC(BIC_SMI))
752 		outp += sprintf(outp, "\t%d", t->smi_count);
753 
754 	/* C1 */
755 	if (DO_BIC(BIC_CPU_c1))
756 		outp += sprintf(outp, "\t%.2f", 100.0 * t->c1/tsc);
757 
758 	/* Added counters */
759 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
760 		if (mp->format == FORMAT_RAW) {
761 			if (mp->width == 32)
762 				outp += sprintf(outp, "\t0x%08lx", (unsigned long) t->counter[i]);
763 			else
764 				outp += sprintf(outp, "\t0x%016llx", t->counter[i]);
765 		} else if (mp->format == FORMAT_DELTA) {
766 			outp += sprintf(outp, "\t%lld", t->counter[i]);
767 		} else if (mp->format == FORMAT_PERCENT) {
768 			outp += sprintf(outp, "\t%.2f", 100.0 * t->counter[i]/tsc);
769 		}
770 	}
771 
772 	/* print per-core data only for 1st thread in core */
773 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
774 		goto done;
775 
776 	if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates)
777 		outp += sprintf(outp, "\t%.2f", 100.0 * c->c3/tsc);
778 	if (DO_BIC(BIC_CPU_c6))
779 		outp += sprintf(outp, "\t%.2f", 100.0 * c->c6/tsc);
780 	if (DO_BIC(BIC_CPU_c7))
781 		outp += sprintf(outp, "\t%.2f", 100.0 * c->c7/tsc);
782 
783 	/* Mod%c6 */
784 	if (DO_BIC(BIC_Mod_c6))
785 		outp += sprintf(outp, "\t%.2f", 100.0 * c->mc6_us / tsc);
786 
787 	if (DO_BIC(BIC_CoreTmp))
788 		outp += sprintf(outp, "\t%d", c->core_temp_c);
789 
790 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
791 		if (mp->format == FORMAT_RAW) {
792 			if (mp->width == 32)
793 				outp += sprintf(outp, "\t0x%08lx", (unsigned long) c->counter[i]);
794 			else
795 				outp += sprintf(outp, "\t0x%016llx", c->counter[i]);
796 		} else if (mp->format == FORMAT_DELTA) {
797 			outp += sprintf(outp, "\t%lld", c->counter[i]);
798 		} else if (mp->format == FORMAT_PERCENT) {
799 			outp += sprintf(outp, "\t%.2f", 100.0 * c->counter[i]/tsc);
800 		}
801 	}
802 
803 	/* print per-package data only for 1st core in package */
804 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
805 		goto done;
806 
807 	/* PkgTmp */
808 	if (DO_BIC(BIC_PkgTmp))
809 		outp += sprintf(outp, "\t%d", p->pkg_temp_c);
810 
811 	/* GFXrc6 */
812 	if (DO_BIC(BIC_GFX_rc6)) {
813 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
814 			outp += sprintf(outp, "\t**.**");
815 		} else {
816 			outp += sprintf(outp, "\t%.2f",
817 				p->gfx_rc6_ms / 10.0 / interval_float);
818 		}
819 	}
820 
821 	/* GFXMHz */
822 	if (DO_BIC(BIC_GFXMHz))
823 		outp += sprintf(outp, "\t%d", p->gfx_mhz);
824 
825 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
826 	if (do_skl_residency) {
827 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pkg_wtd_core_c0/tsc);
828 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pkg_any_core_c0/tsc);
829 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pkg_any_gfxe_c0/tsc);
830 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pkg_both_core_gfxe_c0/tsc);
831 	}
832 
833 	if (DO_BIC(BIC_Pkgpc2))
834 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc2/tsc);
835 	if (DO_BIC(BIC_Pkgpc3))
836 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc3/tsc);
837 	if (DO_BIC(BIC_Pkgpc6))
838 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc6/tsc);
839 	if (DO_BIC(BIC_Pkgpc7))
840 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc7/tsc);
841 	if (DO_BIC(BIC_Pkgpc8))
842 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc8/tsc);
843 	if (DO_BIC(BIC_Pkgpc9))
844 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc9/tsc);
845 	if (DO_BIC(BIC_Pkgpc10))
846 		outp += sprintf(outp, "\t%.2f", 100.0 * p->pc10/tsc);
847 
848 	/*
849  	 * If measurement interval exceeds minimum RAPL Joule Counter range,
850  	 * indicate that results are suspect by printing "**" in fraction place.
851  	 */
852 	if (interval_float < rapl_joule_counter_range)
853 		fmt8 = "\t%.2f";
854 	else
855 		fmt8 = "%6.0f**";
856 
857 	if (DO_BIC(BIC_PkgWatt))
858 		outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units / interval_float);
859 	if (DO_BIC(BIC_CorWatt))
860 		outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units / interval_float);
861 	if (DO_BIC(BIC_GFXWatt))
862 		outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units / interval_float);
863 	if (DO_BIC(BIC_RAMWatt))
864 		outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units / interval_float);
865 	if (DO_BIC(BIC_Pkg_J))
866 		outp += sprintf(outp, fmt8, p->energy_pkg * rapl_energy_units);
867 	if (DO_BIC(BIC_Cor_J))
868 		outp += sprintf(outp, fmt8, p->energy_cores * rapl_energy_units);
869 	if (DO_BIC(BIC_GFX_J))
870 		outp += sprintf(outp, fmt8, p->energy_gfx * rapl_energy_units);
871 	if (DO_BIC(BIC_RAM_J))
872 		outp += sprintf(outp, fmt8, p->energy_dram * rapl_dram_energy_units);
873 	if (DO_BIC(BIC_PKG__))
874 		outp += sprintf(outp, fmt8, 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
875 	if (DO_BIC(BIC_RAM__))
876 		outp += sprintf(outp, fmt8, 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
877 
878 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
879 		if (mp->format == FORMAT_RAW) {
880 			if (mp->width == 32)
881 				outp += sprintf(outp, "\t0x%08lx", (unsigned long) p->counter[i]);
882 			else
883 				outp += sprintf(outp, "\t0x%016llx", p->counter[i]);
884 		} else if (mp->format == FORMAT_DELTA) {
885 			outp += sprintf(outp, "\t%lld", p->counter[i]);
886 		} else if (mp->format == FORMAT_PERCENT) {
887 			outp += sprintf(outp, "\t%.2f", 100.0 * p->counter[i]/tsc);
888 		}
889 	}
890 
891 done:
892 	outp += sprintf(outp, "\n");
893 
894 	return 0;
895 }
896 
897 void flush_output_stdout(void)
898 {
899 	FILE *filep;
900 
901 	if (outf == stderr)
902 		filep = stdout;
903 	else
904 		filep = outf;
905 
906 	fputs(output_buffer, filep);
907 	fflush(filep);
908 
909 	outp = output_buffer;
910 }
911 void flush_output_stderr(void)
912 {
913 	fputs(output_buffer, outf);
914 	fflush(outf);
915 	outp = output_buffer;
916 }
917 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
918 {
919 	static int printed;
920 
921 	if (!printed || !summary_only)
922 		print_header();
923 
924 	if (topo.num_cpus > 1)
925 		format_counters(&average.threads, &average.cores,
926 			&average.packages);
927 
928 	printed = 1;
929 
930 	if (summary_only)
931 		return;
932 
933 	for_all_cpus(format_counters, t, c, p);
934 }
935 
936 #define DELTA_WRAP32(new, old)			\
937 	if (new > old) {			\
938 		old = new - old;		\
939 	} else {				\
940 		old = 0x100000000 + new - old;	\
941 	}
942 
943 int
944 delta_package(struct pkg_data *new, struct pkg_data *old)
945 {
946 	int i;
947 	struct msr_counter *mp;
948 
949 	if (do_skl_residency) {
950 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
951 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
952 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
953 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
954 	}
955 	old->pc2 = new->pc2 - old->pc2;
956 	if (DO_BIC(BIC_Pkgpc3))
957 		old->pc3 = new->pc3 - old->pc3;
958 	if (DO_BIC(BIC_Pkgpc6))
959 		old->pc6 = new->pc6 - old->pc6;
960 	if (DO_BIC(BIC_Pkgpc7))
961 		old->pc7 = new->pc7 - old->pc7;
962 	old->pc8 = new->pc8 - old->pc8;
963 	old->pc9 = new->pc9 - old->pc9;
964 	old->pc10 = new->pc10 - old->pc10;
965 	old->pkg_temp_c = new->pkg_temp_c;
966 
967 	/* flag an error when rc6 counter resets/wraps */
968 	if (old->gfx_rc6_ms >  new->gfx_rc6_ms)
969 		old->gfx_rc6_ms = -1;
970 	else
971 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
972 
973 	old->gfx_mhz = new->gfx_mhz;
974 
975 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
976 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
977 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
978 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
979 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
980 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
981 
982 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
983 		if (mp->format == FORMAT_RAW)
984 			old->counter[i] = new->counter[i];
985 		else
986 			old->counter[i] = new->counter[i] - old->counter[i];
987 	}
988 
989 	return 0;
990 }
991 
992 void
993 delta_core(struct core_data *new, struct core_data *old)
994 {
995 	int i;
996 	struct msr_counter *mp;
997 
998 	old->c3 = new->c3 - old->c3;
999 	old->c6 = new->c6 - old->c6;
1000 	old->c7 = new->c7 - old->c7;
1001 	old->core_temp_c = new->core_temp_c;
1002 	old->mc6_us = new->mc6_us - old->mc6_us;
1003 
1004 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1005 		if (mp->format == FORMAT_RAW)
1006 			old->counter[i] = new->counter[i];
1007 		else
1008 			old->counter[i] = new->counter[i] - old->counter[i];
1009 	}
1010 }
1011 
1012 /*
1013  * old = new - old
1014  */
1015 int
1016 delta_thread(struct thread_data *new, struct thread_data *old,
1017 	struct core_data *core_delta)
1018 {
1019 	int i;
1020 	struct msr_counter *mp;
1021 
1022 	old->tsc = new->tsc - old->tsc;
1023 
1024 	/* check for TSC < 1 Mcycles over interval */
1025 	if (old->tsc < (1000 * 1000))
1026 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1027 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1028 		     "or just the deep ones with \"processor.max_cstate=1\"");
1029 
1030 	old->c1 = new->c1 - old->c1;
1031 
1032 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1033 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1034 			old->aperf = new->aperf - old->aperf;
1035 			old->mperf = new->mperf - old->mperf;
1036 		} else {
1037 			return -1;
1038 		}
1039 	}
1040 
1041 
1042 	if (use_c1_residency_msr) {
1043 		/*
1044 		 * Some models have a dedicated C1 residency MSR,
1045 		 * which should be more accurate than the derivation below.
1046 		 */
1047 	} else {
1048 		/*
1049 		 * As counter collection is not atomic,
1050 		 * it is possible for mperf's non-halted cycles + idle states
1051 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1052 		 */
1053 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
1054 			old->c1 = 0;
1055 		else {
1056 			/* normal case, derive c1 */
1057 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1058 				- core_delta->c6 - core_delta->c7;
1059 		}
1060 	}
1061 
1062 	if (old->mperf == 0) {
1063 		if (debug > 1)
1064 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1065 		old->mperf = 1;	/* divide by 0 protection */
1066 	}
1067 
1068 	if (DO_BIC(BIC_IRQ))
1069 		old->irq_count = new->irq_count - old->irq_count;
1070 
1071 	if (DO_BIC(BIC_SMI))
1072 		old->smi_count = new->smi_count - old->smi_count;
1073 
1074 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1075 		if (mp->format == FORMAT_RAW)
1076 			old->counter[i] = new->counter[i];
1077 		else
1078 			old->counter[i] = new->counter[i] - old->counter[i];
1079 	}
1080 	return 0;
1081 }
1082 
1083 int delta_cpu(struct thread_data *t, struct core_data *c,
1084 	struct pkg_data *p, struct thread_data *t2,
1085 	struct core_data *c2, struct pkg_data *p2)
1086 {
1087 	int retval = 0;
1088 
1089 	/* calculate core delta only for 1st thread in core */
1090 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1091 		delta_core(c, c2);
1092 
1093 	/* always calculate thread delta */
1094 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1095 	if (retval)
1096 		return retval;
1097 
1098 	/* calculate package delta only for 1st core in package */
1099 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1100 		retval = delta_package(p, p2);
1101 
1102 	return retval;
1103 }
1104 
1105 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1106 {
1107 	int i;
1108 	struct msr_counter  *mp;
1109 
1110 	t->tsc = 0;
1111 	t->aperf = 0;
1112 	t->mperf = 0;
1113 	t->c1 = 0;
1114 
1115 	t->irq_count = 0;
1116 	t->smi_count = 0;
1117 
1118 	/* tells format_counters to dump all fields from this set */
1119 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1120 
1121 	c->c3 = 0;
1122 	c->c6 = 0;
1123 	c->c7 = 0;
1124 	c->mc6_us = 0;
1125 	c->core_temp_c = 0;
1126 
1127 	p->pkg_wtd_core_c0 = 0;
1128 	p->pkg_any_core_c0 = 0;
1129 	p->pkg_any_gfxe_c0 = 0;
1130 	p->pkg_both_core_gfxe_c0 = 0;
1131 
1132 	p->pc2 = 0;
1133 	if (DO_BIC(BIC_Pkgpc3))
1134 		p->pc3 = 0;
1135 	if (DO_BIC(BIC_Pkgpc6))
1136 		p->pc6 = 0;
1137 	if (DO_BIC(BIC_Pkgpc7))
1138 		p->pc7 = 0;
1139 	p->pc8 = 0;
1140 	p->pc9 = 0;
1141 	p->pc10 = 0;
1142 
1143 	p->energy_pkg = 0;
1144 	p->energy_dram = 0;
1145 	p->energy_cores = 0;
1146 	p->energy_gfx = 0;
1147 	p->rapl_pkg_perf_status = 0;
1148 	p->rapl_dram_perf_status = 0;
1149 	p->pkg_temp_c = 0;
1150 
1151 	p->gfx_rc6_ms = 0;
1152 	p->gfx_mhz = 0;
1153 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1154 		t->counter[i] = 0;
1155 
1156 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1157 		c->counter[i] = 0;
1158 
1159 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1160 		p->counter[i] = 0;
1161 }
1162 int sum_counters(struct thread_data *t, struct core_data *c,
1163 	struct pkg_data *p)
1164 {
1165 	int i;
1166 	struct msr_counter *mp;
1167 
1168 	average.threads.tsc += t->tsc;
1169 	average.threads.aperf += t->aperf;
1170 	average.threads.mperf += t->mperf;
1171 	average.threads.c1 += t->c1;
1172 
1173 	average.threads.irq_count += t->irq_count;
1174 	average.threads.smi_count += t->smi_count;
1175 
1176 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1177 		if (mp->format == FORMAT_RAW)
1178 			continue;
1179 		average.threads.counter[i] += t->counter[i];
1180 	}
1181 
1182 	/* sum per-core values only for 1st thread in core */
1183 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1184 		return 0;
1185 
1186 	average.cores.c3 += c->c3;
1187 	average.cores.c6 += c->c6;
1188 	average.cores.c7 += c->c7;
1189 	average.cores.mc6_us += c->mc6_us;
1190 
1191 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1192 
1193 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1194 		if (mp->format == FORMAT_RAW)
1195 			continue;
1196 		average.cores.counter[i] += c->counter[i];
1197 	}
1198 
1199 	/* sum per-pkg values only for 1st core in pkg */
1200 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1201 		return 0;
1202 
1203 	if (do_skl_residency) {
1204 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1205 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1206 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1207 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1208 	}
1209 
1210 	average.packages.pc2 += p->pc2;
1211 	if (DO_BIC(BIC_Pkgpc3))
1212 		average.packages.pc3 += p->pc3;
1213 	if (DO_BIC(BIC_Pkgpc6))
1214 		average.packages.pc6 += p->pc6;
1215 	if (DO_BIC(BIC_Pkgpc7))
1216 		average.packages.pc7 += p->pc7;
1217 	average.packages.pc8 += p->pc8;
1218 	average.packages.pc9 += p->pc9;
1219 	average.packages.pc10 += p->pc10;
1220 
1221 	average.packages.energy_pkg += p->energy_pkg;
1222 	average.packages.energy_dram += p->energy_dram;
1223 	average.packages.energy_cores += p->energy_cores;
1224 	average.packages.energy_gfx += p->energy_gfx;
1225 
1226 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1227 	average.packages.gfx_mhz = p->gfx_mhz;
1228 
1229 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1230 
1231 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1232 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1233 
1234 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1235 		if (mp->format == FORMAT_RAW)
1236 			continue;
1237 		average.packages.counter[i] += p->counter[i];
1238 	}
1239 	return 0;
1240 }
1241 /*
1242  * sum the counters for all cpus in the system
1243  * compute the weighted average
1244  */
1245 void compute_average(struct thread_data *t, struct core_data *c,
1246 	struct pkg_data *p)
1247 {
1248 	int i;
1249 	struct msr_counter *mp;
1250 
1251 	clear_counters(&average.threads, &average.cores, &average.packages);
1252 
1253 	for_all_cpus(sum_counters, t, c, p);
1254 
1255 	average.threads.tsc /= topo.num_cpus;
1256 	average.threads.aperf /= topo.num_cpus;
1257 	average.threads.mperf /= topo.num_cpus;
1258 	average.threads.c1 /= topo.num_cpus;
1259 
1260 	average.cores.c3 /= topo.num_cores;
1261 	average.cores.c6 /= topo.num_cores;
1262 	average.cores.c7 /= topo.num_cores;
1263 	average.cores.mc6_us /= topo.num_cores;
1264 
1265 	if (do_skl_residency) {
1266 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1267 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1268 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1269 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1270 	}
1271 
1272 	average.packages.pc2 /= topo.num_packages;
1273 	if (DO_BIC(BIC_Pkgpc3))
1274 		average.packages.pc3 /= topo.num_packages;
1275 	if (DO_BIC(BIC_Pkgpc6))
1276 		average.packages.pc6 /= topo.num_packages;
1277 	if (DO_BIC(BIC_Pkgpc7))
1278 		average.packages.pc7 /= topo.num_packages;
1279 
1280 	average.packages.pc8 /= topo.num_packages;
1281 	average.packages.pc9 /= topo.num_packages;
1282 	average.packages.pc10 /= topo.num_packages;
1283 
1284 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1285 		if (mp->format == FORMAT_RAW)
1286 			continue;
1287 		average.threads.counter[i] /= topo.num_cpus;
1288 	}
1289 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1290 		if (mp->format == FORMAT_RAW)
1291 			continue;
1292 		average.cores.counter[i] /= topo.num_cores;
1293 	}
1294 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1295 		if (mp->format == FORMAT_RAW)
1296 			continue;
1297 		average.packages.counter[i] /= topo.num_packages;
1298 	}
1299 }
1300 
1301 static unsigned long long rdtsc(void)
1302 {
1303 	unsigned int low, high;
1304 
1305 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
1306 
1307 	return low | ((unsigned long long)high) << 32;
1308 }
1309 
1310 /*
1311  * get_counters(...)
1312  * migrate to cpu
1313  * acquire and record local counters for that cpu
1314  */
1315 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1316 {
1317 	int cpu = t->cpu_id;
1318 	unsigned long long msr;
1319 	int aperf_mperf_retry_count = 0;
1320 	struct msr_counter *mp;
1321 	int i;
1322 
1323 	if (cpu_migrate(cpu)) {
1324 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1325 		return -1;
1326 	}
1327 
1328 retry:
1329 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
1330 
1331 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1332 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1333 
1334 		/*
1335 		 * The TSC, APERF and MPERF must be read together for
1336 		 * APERF/MPERF and MPERF/TSC to give accurate results.
1337 		 *
1338 		 * Unfortunately, APERF and MPERF are read by
1339 		 * individual system call, so delays may occur
1340 		 * between them.  If the time to read them
1341 		 * varies by a large amount, we re-read them.
1342 		 */
1343 
1344 		/*
1345 		 * This initial dummy APERF read has been seen to
1346 		 * reduce jitter in the subsequent reads.
1347 		 */
1348 
1349 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1350 			return -3;
1351 
1352 		t->tsc = rdtsc();	/* re-read close to APERF */
1353 
1354 		tsc_before = t->tsc;
1355 
1356 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1357 			return -3;
1358 
1359 		tsc_between = rdtsc();
1360 
1361 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1362 			return -4;
1363 
1364 		tsc_after = rdtsc();
1365 
1366 		aperf_time = tsc_between - tsc_before;
1367 		mperf_time = tsc_after - tsc_between;
1368 
1369 		/*
1370 		 * If the system call latency to read APERF and MPERF
1371 		 * differ by more than 2x, then try again.
1372 		 */
1373 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1374 			aperf_mperf_retry_count++;
1375 			if (aperf_mperf_retry_count < 5)
1376 				goto retry;
1377 			else
1378 				warnx("cpu%d jitter %lld %lld",
1379 					cpu, aperf_time, mperf_time);
1380 		}
1381 		aperf_mperf_retry_count = 0;
1382 
1383 		t->aperf = t->aperf * aperf_mperf_multiplier;
1384 		t->mperf = t->mperf * aperf_mperf_multiplier;
1385 	}
1386 
1387 	if (DO_BIC(BIC_IRQ))
1388 		t->irq_count = irqs_per_cpu[cpu];
1389 	if (DO_BIC(BIC_SMI)) {
1390 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1391 			return -5;
1392 		t->smi_count = msr & 0xFFFFFFFF;
1393 	}
1394 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1395 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1396 			return -6;
1397 	}
1398 
1399 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1400 		if (get_msr(cpu, mp->msr_num, &t->counter[i]))
1401 			return -10;
1402 	}
1403 
1404 
1405 	/* collect core counters only for 1st thread in core */
1406 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1407 		return 0;
1408 
1409 	if (DO_BIC(BIC_CPU_c3) && !do_slm_cstates && !do_knl_cstates) {
1410 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1411 			return -6;
1412 	}
1413 
1414 	if (DO_BIC(BIC_CPU_c6) && !do_knl_cstates) {
1415 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1416 			return -7;
1417 	} else if (do_knl_cstates) {
1418 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1419 			return -7;
1420 	}
1421 
1422 	if (DO_BIC(BIC_CPU_c7))
1423 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1424 			return -8;
1425 
1426 	if (DO_BIC(BIC_Mod_c6))
1427 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1428 			return -8;
1429 
1430 	if (DO_BIC(BIC_CoreTmp)) {
1431 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1432 			return -9;
1433 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1434 	}
1435 
1436 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1437 		if (get_msr(cpu, mp->msr_num, &c->counter[i]))
1438 			return -10;
1439 	}
1440 
1441 	/* collect package counters only for 1st core in package */
1442 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1443 		return 0;
1444 
1445 	if (do_skl_residency) {
1446 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1447 			return -10;
1448 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1449 			return -11;
1450 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1451 			return -12;
1452 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1453 			return -13;
1454 	}
1455 	if (DO_BIC(BIC_Pkgpc3))
1456 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1457 			return -9;
1458 	if (DO_BIC(BIC_Pkgpc6)) {
1459 		if (do_slm_cstates) {
1460 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1461 				return -10;
1462 		} else {
1463 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1464 				return -10;
1465 		}
1466 	}
1467 
1468 	if (DO_BIC(BIC_Pkgpc2))
1469 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1470 			return -11;
1471 	if (DO_BIC(BIC_Pkgpc7))
1472 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1473 			return -12;
1474 	if (DO_BIC(BIC_Pkgpc8))
1475 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1476 			return -13;
1477 	if (DO_BIC(BIC_Pkgpc9))
1478 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1479 			return -13;
1480 	if (DO_BIC(BIC_Pkgpc10))
1481 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1482 			return -13;
1483 
1484 	if (do_rapl & RAPL_PKG) {
1485 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1486 			return -13;
1487 		p->energy_pkg = msr & 0xFFFFFFFF;
1488 	}
1489 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1490 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1491 			return -14;
1492 		p->energy_cores = msr & 0xFFFFFFFF;
1493 	}
1494 	if (do_rapl & RAPL_DRAM) {
1495 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1496 			return -15;
1497 		p->energy_dram = msr & 0xFFFFFFFF;
1498 	}
1499 	if (do_rapl & RAPL_GFX) {
1500 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1501 			return -16;
1502 		p->energy_gfx = msr & 0xFFFFFFFF;
1503 	}
1504 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1505 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1506 			return -16;
1507 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1508 	}
1509 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1510 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1511 			return -16;
1512 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1513 	}
1514 	if (DO_BIC(BIC_PkgTmp)) {
1515 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1516 			return -17;
1517 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1518 	}
1519 
1520 	if (DO_BIC(BIC_GFX_rc6))
1521 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
1522 
1523 	if (DO_BIC(BIC_GFXMHz))
1524 		p->gfx_mhz = gfx_cur_mhz;
1525 
1526 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1527 		if (get_msr(cpu, mp->msr_num, &p->counter[i]))
1528 			return -10;
1529 	}
1530 
1531 	return 0;
1532 }
1533 
1534 /*
1535  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
1536  * If you change the values, note they are used both in comparisons
1537  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
1538  */
1539 
1540 #define PCLUKN 0 /* Unknown */
1541 #define PCLRSV 1 /* Reserved */
1542 #define PCL__0 2 /* PC0 */
1543 #define PCL__1 3 /* PC1 */
1544 #define PCL__2 4 /* PC2 */
1545 #define PCL__3 5 /* PC3 */
1546 #define PCL__4 6 /* PC4 */
1547 #define PCL__6 7 /* PC6 */
1548 #define PCL_6N 8 /* PC6 No Retention */
1549 #define PCL_6R 9 /* PC6 Retention */
1550 #define PCL__7 10 /* PC7 */
1551 #define PCL_7S 11 /* PC7 Shrink */
1552 #define PCL__8 12 /* PC8 */
1553 #define PCL__9 13 /* PC9 */
1554 #define PCLUNL 14 /* Unlimited */
1555 
1556 int pkg_cstate_limit = PCLUKN;
1557 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
1558 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "unlimited"};
1559 
1560 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};
1561 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};
1562 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};
1563 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, PCL__6, PCL__7};
1564 int amt_pkg_cstate_limits[16] = {PCLUNL, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1565 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};
1566 int bxt_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1567 int skx_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
1568 
1569 
1570 static void
1571 calculate_tsc_tweak()
1572 {
1573 	tsc_tweak = base_hz / tsc_hz;
1574 }
1575 
1576 static void
1577 dump_nhm_platform_info(void)
1578 {
1579 	unsigned long long msr;
1580 	unsigned int ratio;
1581 
1582 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
1583 
1584 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
1585 
1586 	ratio = (msr >> 40) & 0xFF;
1587 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
1588 		ratio, bclk, ratio * bclk);
1589 
1590 	ratio = (msr >> 8) & 0xFF;
1591 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
1592 		ratio, bclk, ratio * bclk);
1593 
1594 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
1595 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
1596 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
1597 
1598 	return;
1599 }
1600 
1601 static void
1602 dump_hsw_turbo_ratio_limits(void)
1603 {
1604 	unsigned long long msr;
1605 	unsigned int ratio;
1606 
1607 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
1608 
1609 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
1610 
1611 	ratio = (msr >> 8) & 0xFF;
1612 	if (ratio)
1613 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
1614 			ratio, bclk, ratio * bclk);
1615 
1616 	ratio = (msr >> 0) & 0xFF;
1617 	if (ratio)
1618 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
1619 			ratio, bclk, ratio * bclk);
1620 	return;
1621 }
1622 
1623 static void
1624 dump_ivt_turbo_ratio_limits(void)
1625 {
1626 	unsigned long long msr;
1627 	unsigned int ratio;
1628 
1629 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
1630 
1631 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
1632 
1633 	ratio = (msr >> 56) & 0xFF;
1634 	if (ratio)
1635 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
1636 			ratio, bclk, ratio * bclk);
1637 
1638 	ratio = (msr >> 48) & 0xFF;
1639 	if (ratio)
1640 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
1641 			ratio, bclk, ratio * bclk);
1642 
1643 	ratio = (msr >> 40) & 0xFF;
1644 	if (ratio)
1645 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
1646 			ratio, bclk, ratio * bclk);
1647 
1648 	ratio = (msr >> 32) & 0xFF;
1649 	if (ratio)
1650 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
1651 			ratio, bclk, ratio * bclk);
1652 
1653 	ratio = (msr >> 24) & 0xFF;
1654 	if (ratio)
1655 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
1656 			ratio, bclk, ratio * bclk);
1657 
1658 	ratio = (msr >> 16) & 0xFF;
1659 	if (ratio)
1660 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
1661 			ratio, bclk, ratio * bclk);
1662 
1663 	ratio = (msr >> 8) & 0xFF;
1664 	if (ratio)
1665 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
1666 			ratio, bclk, ratio * bclk);
1667 
1668 	ratio = (msr >> 0) & 0xFF;
1669 	if (ratio)
1670 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
1671 			ratio, bclk, ratio * bclk);
1672 	return;
1673 }
1674 int has_turbo_ratio_group_limits(int family, int model)
1675 {
1676 
1677 	if (!genuine_intel)
1678 		return 0;
1679 
1680 	switch (model) {
1681 	case INTEL_FAM6_ATOM_GOLDMONT:
1682 	case INTEL_FAM6_SKYLAKE_X:
1683 	case INTEL_FAM6_ATOM_DENVERTON:
1684 		return 1;
1685 	}
1686 	return 0;
1687 }
1688 
1689 static void
1690 dump_turbo_ratio_limits(int family, int model)
1691 {
1692 	unsigned long long msr, core_counts;
1693 	unsigned int ratio, group_size;
1694 
1695 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1696 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
1697 
1698 	if (has_turbo_ratio_group_limits(family, model)) {
1699 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
1700 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
1701 	} else {
1702 		core_counts = 0x0807060504030201;
1703 	}
1704 
1705 	ratio = (msr >> 56) & 0xFF;
1706 	group_size = (core_counts >> 56) & 0xFF;
1707 	if (ratio)
1708 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1709 			ratio, bclk, ratio * bclk, group_size);
1710 
1711 	ratio = (msr >> 48) & 0xFF;
1712 	group_size = (core_counts >> 48) & 0xFF;
1713 	if (ratio)
1714 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1715 			ratio, bclk, ratio * bclk, group_size);
1716 
1717 	ratio = (msr >> 40) & 0xFF;
1718 	group_size = (core_counts >> 40) & 0xFF;
1719 	if (ratio)
1720 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1721 			ratio, bclk, ratio * bclk, group_size);
1722 
1723 	ratio = (msr >> 32) & 0xFF;
1724 	group_size = (core_counts >> 32) & 0xFF;
1725 	if (ratio)
1726 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1727 			ratio, bclk, ratio * bclk, group_size);
1728 
1729 	ratio = (msr >> 24) & 0xFF;
1730 	group_size = (core_counts >> 24) & 0xFF;
1731 	if (ratio)
1732 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1733 			ratio, bclk, ratio * bclk, group_size);
1734 
1735 	ratio = (msr >> 16) & 0xFF;
1736 	group_size = (core_counts >> 16) & 0xFF;
1737 	if (ratio)
1738 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1739 			ratio, bclk, ratio * bclk, group_size);
1740 
1741 	ratio = (msr >> 8) & 0xFF;
1742 	group_size = (core_counts >> 8) & 0xFF;
1743 	if (ratio)
1744 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1745 			ratio, bclk, ratio * bclk, group_size);
1746 
1747 	ratio = (msr >> 0) & 0xFF;
1748 	group_size = (core_counts >> 0) & 0xFF;
1749 	if (ratio)
1750 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
1751 			ratio, bclk, ratio * bclk, group_size);
1752 	return;
1753 }
1754 
1755 static void
1756 dump_atom_turbo_ratio_limits(void)
1757 {
1758 	unsigned long long msr;
1759 	unsigned int ratio;
1760 
1761 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
1762 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
1763 
1764 	ratio = (msr >> 0) & 0x3F;
1765 	if (ratio)
1766 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
1767 			ratio, bclk, ratio * bclk);
1768 
1769 	ratio = (msr >> 8) & 0x3F;
1770 	if (ratio)
1771 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
1772 			ratio, bclk, ratio * bclk);
1773 
1774 	ratio = (msr >> 16) & 0x3F;
1775 	if (ratio)
1776 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
1777 			ratio, bclk, ratio * bclk);
1778 
1779 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
1780 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
1781 
1782 	ratio = (msr >> 24) & 0x3F;
1783 	if (ratio)
1784 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
1785 			ratio, bclk, ratio * bclk);
1786 
1787 	ratio = (msr >> 16) & 0x3F;
1788 	if (ratio)
1789 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
1790 			ratio, bclk, ratio * bclk);
1791 
1792 	ratio = (msr >> 8) & 0x3F;
1793 	if (ratio)
1794 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
1795 			ratio, bclk, ratio * bclk);
1796 
1797 	ratio = (msr >> 0) & 0x3F;
1798 	if (ratio)
1799 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
1800 			ratio, bclk, ratio * bclk);
1801 }
1802 
1803 static void
1804 dump_knl_turbo_ratio_limits(void)
1805 {
1806 	const unsigned int buckets_no = 7;
1807 
1808 	unsigned long long msr;
1809 	int delta_cores, delta_ratio;
1810 	int i, b_nr;
1811 	unsigned int cores[buckets_no];
1812 	unsigned int ratio[buckets_no];
1813 
1814 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
1815 
1816 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
1817 		base_cpu, msr);
1818 
1819 	/**
1820 	 * Turbo encoding in KNL is as follows:
1821 	 * [0] -- Reserved
1822 	 * [7:1] -- Base value of number of active cores of bucket 1.
1823 	 * [15:8] -- Base value of freq ratio of bucket 1.
1824 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
1825 	 * i.e. active cores of bucket 2 =
1826 	 * active cores of bucket 1 + delta
1827 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
1828 	 * i.e. freq ratio of bucket 2 =
1829 	 * freq ratio of bucket 1 - delta
1830 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
1831 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
1832 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
1833 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
1834 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
1835 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
1836 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
1837 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
1838 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
1839 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
1840 	 */
1841 
1842 	b_nr = 0;
1843 	cores[b_nr] = (msr & 0xFF) >> 1;
1844 	ratio[b_nr] = (msr >> 8) & 0xFF;
1845 
1846 	for (i = 16; i < 64; i += 8) {
1847 		delta_cores = (msr >> i) & 0x1F;
1848 		delta_ratio = (msr >> (i + 5)) & 0x7;
1849 
1850 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
1851 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
1852 		b_nr++;
1853 	}
1854 
1855 	for (i = buckets_no - 1; i >= 0; i--)
1856 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
1857 			fprintf(outf,
1858 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
1859 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
1860 }
1861 
1862 static void
1863 dump_nhm_cst_cfg(void)
1864 {
1865 	unsigned long long msr;
1866 
1867 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
1868 
1869 #define SNB_C1_AUTO_UNDEMOTE              (1UL << 27)
1870 #define SNB_C3_AUTO_UNDEMOTE              (1UL << 28)
1871 
1872 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
1873 
1874 	fprintf(outf, " (%s%s%s%s%slocked: pkg-cstate-limit=%d: %s)\n",
1875 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
1876 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
1877 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
1878 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
1879 		(msr & (1 << 15)) ? "" : "UN",
1880 		(unsigned int)msr & 0xF,
1881 		pkg_cstate_limit_strings[pkg_cstate_limit]);
1882 	return;
1883 }
1884 
1885 static void
1886 dump_config_tdp(void)
1887 {
1888 	unsigned long long msr;
1889 
1890 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
1891 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
1892 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
1893 
1894 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
1895 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
1896 	if (msr) {
1897 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
1898 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
1899 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
1900 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
1901 	}
1902 	fprintf(outf, ")\n");
1903 
1904 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
1905 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
1906 	if (msr) {
1907 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
1908 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
1909 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
1910 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
1911 	}
1912 	fprintf(outf, ")\n");
1913 
1914 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
1915 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
1916 	if ((msr) & 0x3)
1917 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
1918 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
1919 	fprintf(outf, ")\n");
1920 
1921 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
1922 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
1923 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
1924 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
1925 	fprintf(outf, ")\n");
1926 }
1927 
1928 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
1929 
1930 void print_irtl(void)
1931 {
1932 	unsigned long long msr;
1933 
1934 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
1935 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
1936 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1937 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1938 
1939 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
1940 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
1941 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1942 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1943 
1944 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
1945 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
1946 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1947 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1948 
1949 	if (!do_irtl_hsw)
1950 		return;
1951 
1952 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
1953 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
1954 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1955 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1956 
1957 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
1958 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
1959 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1960 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1961 
1962 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
1963 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
1964 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
1965 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
1966 
1967 }
1968 void free_fd_percpu(void)
1969 {
1970 	int i;
1971 
1972 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
1973 		if (fd_percpu[i] != 0)
1974 			close(fd_percpu[i]);
1975 	}
1976 
1977 	free(fd_percpu);
1978 }
1979 
1980 void free_all_buffers(void)
1981 {
1982 	CPU_FREE(cpu_present_set);
1983 	cpu_present_set = NULL;
1984 	cpu_present_setsize = 0;
1985 
1986 	CPU_FREE(cpu_affinity_set);
1987 	cpu_affinity_set = NULL;
1988 	cpu_affinity_setsize = 0;
1989 
1990 	free(thread_even);
1991 	free(core_even);
1992 	free(package_even);
1993 
1994 	thread_even = NULL;
1995 	core_even = NULL;
1996 	package_even = NULL;
1997 
1998 	free(thread_odd);
1999 	free(core_odd);
2000 	free(package_odd);
2001 
2002 	thread_odd = NULL;
2003 	core_odd = NULL;
2004 	package_odd = NULL;
2005 
2006 	free(output_buffer);
2007 	output_buffer = NULL;
2008 	outp = NULL;
2009 
2010 	free_fd_percpu();
2011 
2012 	free(irq_column_2_cpu);
2013 	free(irqs_per_cpu);
2014 }
2015 
2016 /*
2017  * Open a file, and exit on failure
2018  */
2019 FILE *fopen_or_die(const char *path, const char *mode)
2020 {
2021 	FILE *filep = fopen(path, mode);
2022 	if (!filep)
2023 		err(1, "%s: open failed", path);
2024 	return filep;
2025 }
2026 
2027 /*
2028  * Parse a file containing a single int.
2029  */
2030 int parse_int_file(const char *fmt, ...)
2031 {
2032 	va_list args;
2033 	char path[PATH_MAX];
2034 	FILE *filep;
2035 	int value;
2036 
2037 	va_start(args, fmt);
2038 	vsnprintf(path, sizeof(path), fmt, args);
2039 	va_end(args);
2040 	filep = fopen_or_die(path, "r");
2041 	if (fscanf(filep, "%d", &value) != 1)
2042 		err(1, "%s: failed to parse number from file", path);
2043 	fclose(filep);
2044 	return value;
2045 }
2046 
2047 /*
2048  * get_cpu_position_in_core(cpu)
2049  * return the position of the CPU among its HT siblings in the core
2050  * return -1 if the sibling is not in list
2051  */
2052 int get_cpu_position_in_core(int cpu)
2053 {
2054 	char path[64];
2055 	FILE *filep;
2056 	int this_cpu;
2057 	char character;
2058 	int i;
2059 
2060 	sprintf(path,
2061 		"/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list",
2062 		cpu);
2063 	filep = fopen(path, "r");
2064 	if (filep == NULL) {
2065 		perror(path);
2066 		exit(1);
2067 	}
2068 
2069 	for (i = 0; i < topo.num_threads_per_core; i++) {
2070 		fscanf(filep, "%d", &this_cpu);
2071 		if (this_cpu == cpu) {
2072 			fclose(filep);
2073 			return i;
2074 		}
2075 
2076 		/* Account for no separator after last thread*/
2077 		if (i != (topo.num_threads_per_core - 1))
2078 			fscanf(filep, "%c", &character);
2079 	}
2080 
2081 	fclose(filep);
2082 	return -1;
2083 }
2084 
2085 /*
2086  * cpu_is_first_core_in_package(cpu)
2087  * return 1 if given CPU is 1st core in package
2088  */
2089 int cpu_is_first_core_in_package(int cpu)
2090 {
2091 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2092 }
2093 
2094 int get_physical_package_id(int cpu)
2095 {
2096 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2097 }
2098 
2099 int get_core_id(int cpu)
2100 {
2101 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2102 }
2103 
2104 int get_num_ht_siblings(int cpu)
2105 {
2106 	char path[80];
2107 	FILE *filep;
2108 	int sib1;
2109 	int matches = 0;
2110 	char character;
2111 	char str[100];
2112 	char *ch;
2113 
2114 	sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
2115 	filep = fopen_or_die(path, "r");
2116 
2117 	/*
2118 	 * file format:
2119 	 * A ',' separated or '-' separated set of numbers
2120 	 * (eg 1-2 or 1,3,4,5)
2121 	 */
2122 	fscanf(filep, "%d%c\n", &sib1, &character);
2123 	fseek(filep, 0, SEEK_SET);
2124 	fgets(str, 100, filep);
2125 	ch = strchr(str, character);
2126 	while (ch != NULL) {
2127 		matches++;
2128 		ch = strchr(ch+1, character);
2129 	}
2130 
2131 	fclose(filep);
2132 	return matches+1;
2133 }
2134 
2135 /*
2136  * run func(thread, core, package) in topology order
2137  * skip non-present cpus
2138  */
2139 
2140 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2141 	struct pkg_data *, struct thread_data *, struct core_data *,
2142 	struct pkg_data *), struct thread_data *thread_base,
2143 	struct core_data *core_base, struct pkg_data *pkg_base,
2144 	struct thread_data *thread_base2, struct core_data *core_base2,
2145 	struct pkg_data *pkg_base2)
2146 {
2147 	int retval, pkg_no, core_no, thread_no;
2148 
2149 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2150 		for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
2151 			for (thread_no = 0; thread_no <
2152 				topo.num_threads_per_core; ++thread_no) {
2153 				struct thread_data *t, *t2;
2154 				struct core_data *c, *c2;
2155 				struct pkg_data *p, *p2;
2156 
2157 				t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
2158 
2159 				if (cpu_is_not_present(t->cpu_id))
2160 					continue;
2161 
2162 				t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
2163 
2164 				c = GET_CORE(core_base, core_no, pkg_no);
2165 				c2 = GET_CORE(core_base2, core_no, pkg_no);
2166 
2167 				p = GET_PKG(pkg_base, pkg_no);
2168 				p2 = GET_PKG(pkg_base2, pkg_no);
2169 
2170 				retval = func(t, c, p, t2, c2, p2);
2171 				if (retval)
2172 					return retval;
2173 			}
2174 		}
2175 	}
2176 	return 0;
2177 }
2178 
2179 /*
2180  * run func(cpu) on every cpu in /proc/stat
2181  * return max_cpu number
2182  */
2183 int for_all_proc_cpus(int (func)(int))
2184 {
2185 	FILE *fp;
2186 	int cpu_num;
2187 	int retval;
2188 
2189 	fp = fopen_or_die(proc_stat, "r");
2190 
2191 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2192 	if (retval != 0)
2193 		err(1, "%s: failed to parse format", proc_stat);
2194 
2195 	while (1) {
2196 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2197 		if (retval != 1)
2198 			break;
2199 
2200 		retval = func(cpu_num);
2201 		if (retval) {
2202 			fclose(fp);
2203 			return(retval);
2204 		}
2205 	}
2206 	fclose(fp);
2207 	return 0;
2208 }
2209 
2210 void re_initialize(void)
2211 {
2212 	free_all_buffers();
2213 	setup_all_buffers();
2214 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2215 }
2216 
2217 
2218 /*
2219  * count_cpus()
2220  * remember the last one seen, it will be the max
2221  */
2222 int count_cpus(int cpu)
2223 {
2224 	if (topo.max_cpu_num < cpu)
2225 		topo.max_cpu_num = cpu;
2226 
2227 	topo.num_cpus += 1;
2228 	return 0;
2229 }
2230 int mark_cpu_present(int cpu)
2231 {
2232 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2233 	return 0;
2234 }
2235 
2236 /*
2237  * snapshot_proc_interrupts()
2238  *
2239  * read and record summary of /proc/interrupts
2240  *
2241  * return 1 if config change requires a restart, else return 0
2242  */
2243 int snapshot_proc_interrupts(void)
2244 {
2245 	static FILE *fp;
2246 	int column, retval;
2247 
2248 	if (fp == NULL)
2249 		fp = fopen_or_die("/proc/interrupts", "r");
2250 	else
2251 		rewind(fp);
2252 
2253 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
2254 	for (column = 0; column < topo.num_cpus; ++column) {
2255 		int cpu_number;
2256 
2257 		retval = fscanf(fp, " CPU%d", &cpu_number);
2258 		if (retval != 1)
2259 			break;
2260 
2261 		if (cpu_number > topo.max_cpu_num) {
2262 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2263 			return 1;
2264 		}
2265 
2266 		irq_column_2_cpu[column] = cpu_number;
2267 		irqs_per_cpu[cpu_number] = 0;
2268 	}
2269 
2270 	/* read /proc/interrupt count lines and sum up irqs per cpu */
2271 	while (1) {
2272 		int column;
2273 		char buf[64];
2274 
2275 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
2276 		if (retval != 1)
2277 			break;
2278 
2279 		/* read the count per cpu */
2280 		for (column = 0; column < topo.num_cpus; ++column) {
2281 
2282 			int cpu_number, irq_count;
2283 
2284 			retval = fscanf(fp, " %d", &irq_count);
2285 			if (retval != 1)
2286 				break;
2287 
2288 			cpu_number = irq_column_2_cpu[column];
2289 			irqs_per_cpu[cpu_number] += irq_count;
2290 
2291 		}
2292 
2293 		while (getc(fp) != '\n')
2294 			;	/* flush interrupt description */
2295 
2296 	}
2297 	return 0;
2298 }
2299 /*
2300  * snapshot_gfx_rc6_ms()
2301  *
2302  * record snapshot of
2303  * /sys/class/drm/card0/power/rc6_residency_ms
2304  *
2305  * return 1 if config change requires a restart, else return 0
2306  */
2307 int snapshot_gfx_rc6_ms(void)
2308 {
2309 	FILE *fp;
2310 	int retval;
2311 
2312 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2313 
2314 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2315 	if (retval != 1)
2316 		err(1, "GFX rc6");
2317 
2318 	fclose(fp);
2319 
2320 	return 0;
2321 }
2322 /*
2323  * snapshot_gfx_mhz()
2324  *
2325  * record snapshot of
2326  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2327  *
2328  * return 1 if config change requires a restart, else return 0
2329  */
2330 int snapshot_gfx_mhz(void)
2331 {
2332 	static FILE *fp;
2333 	int retval;
2334 
2335 	if (fp == NULL)
2336 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2337 	else
2338 		rewind(fp);
2339 
2340 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
2341 	if (retval != 1)
2342 		err(1, "GFX MHz");
2343 
2344 	return 0;
2345 }
2346 
2347 /*
2348  * snapshot /proc and /sys files
2349  *
2350  * return 1 if configuration restart needed, else return 0
2351  */
2352 int snapshot_proc_sysfs_files(void)
2353 {
2354 	if (snapshot_proc_interrupts())
2355 		return 1;
2356 
2357 	if (DO_BIC(BIC_GFX_rc6))
2358 		snapshot_gfx_rc6_ms();
2359 
2360 	if (DO_BIC(BIC_GFXMHz))
2361 		snapshot_gfx_mhz();
2362 
2363 	return 0;
2364 }
2365 
2366 void turbostat_loop()
2367 {
2368 	int retval;
2369 	int restarted = 0;
2370 
2371 restart:
2372 	restarted++;
2373 
2374 	snapshot_proc_sysfs_files();
2375 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2376 	if (retval < -1) {
2377 		exit(retval);
2378 	} else if (retval == -1) {
2379 		if (restarted > 1) {
2380 			exit(retval);
2381 		}
2382 		re_initialize();
2383 		goto restart;
2384 	}
2385 	restarted = 0;
2386 	gettimeofday(&tv_even, (struct timezone *)NULL);
2387 
2388 	while (1) {
2389 		if (for_all_proc_cpus(cpu_is_not_present)) {
2390 			re_initialize();
2391 			goto restart;
2392 		}
2393 		nanosleep(&interval_ts, NULL);
2394 		if (snapshot_proc_sysfs_files())
2395 			goto restart;
2396 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
2397 		if (retval < -1) {
2398 			exit(retval);
2399 		} else if (retval == -1) {
2400 			re_initialize();
2401 			goto restart;
2402 		}
2403 		gettimeofday(&tv_odd, (struct timezone *)NULL);
2404 		timersub(&tv_odd, &tv_even, &tv_delta);
2405 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
2406 			re_initialize();
2407 			goto restart;
2408 		}
2409 		compute_average(EVEN_COUNTERS);
2410 		format_all_counters(EVEN_COUNTERS);
2411 		flush_output_stdout();
2412 		nanosleep(&interval_ts, NULL);
2413 		if (snapshot_proc_sysfs_files())
2414 			goto restart;
2415 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
2416 		if (retval < -1) {
2417 			exit(retval);
2418 		} else if (retval == -1) {
2419 			re_initialize();
2420 			goto restart;
2421 		}
2422 		gettimeofday(&tv_even, (struct timezone *)NULL);
2423 		timersub(&tv_even, &tv_odd, &tv_delta);
2424 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
2425 			re_initialize();
2426 			goto restart;
2427 		}
2428 		compute_average(ODD_COUNTERS);
2429 		format_all_counters(ODD_COUNTERS);
2430 		flush_output_stdout();
2431 	}
2432 }
2433 
2434 void check_dev_msr()
2435 {
2436 	struct stat sb;
2437 	char pathname[32];
2438 
2439 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2440 	if (stat(pathname, &sb))
2441  		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
2442 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
2443 }
2444 
2445 void check_permissions()
2446 {
2447 	struct __user_cap_header_struct cap_header_data;
2448 	cap_user_header_t cap_header = &cap_header_data;
2449 	struct __user_cap_data_struct cap_data_data;
2450 	cap_user_data_t cap_data = &cap_data_data;
2451 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
2452 	int do_exit = 0;
2453 	char pathname[32];
2454 
2455 	/* check for CAP_SYS_RAWIO */
2456 	cap_header->pid = getpid();
2457 	cap_header->version = _LINUX_CAPABILITY_VERSION;
2458 	if (capget(cap_header, cap_data) < 0)
2459 		err(-6, "capget(2) failed");
2460 
2461 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
2462 		do_exit++;
2463 		warnx("capget(CAP_SYS_RAWIO) failed,"
2464 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
2465 	}
2466 
2467 	/* test file permissions */
2468 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
2469 	if (euidaccess(pathname, R_OK)) {
2470 		do_exit++;
2471 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
2472 	}
2473 
2474 	/* if all else fails, thell them to be root */
2475 	if (do_exit)
2476 		if (getuid() != 0)
2477 			warnx("... or simply run as root");
2478 
2479 	if (do_exit)
2480 		exit(-6);
2481 }
2482 
2483 /*
2484  * NHM adds support for additional MSRs:
2485  *
2486  * MSR_SMI_COUNT                   0x00000034
2487  *
2488  * MSR_PLATFORM_INFO               0x000000ce
2489  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
2490  *
2491  * MSR_MISC_PWR_MGMT               0x000001aa
2492  *
2493  * MSR_PKG_C3_RESIDENCY            0x000003f8
2494  * MSR_PKG_C6_RESIDENCY            0x000003f9
2495  * MSR_CORE_C3_RESIDENCY           0x000003fc
2496  * MSR_CORE_C6_RESIDENCY           0x000003fd
2497  *
2498  * Side effect:
2499  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
2500  * sets has_misc_feature_control
2501  */
2502 int probe_nhm_msrs(unsigned int family, unsigned int model)
2503 {
2504 	unsigned long long msr;
2505 	unsigned int base_ratio;
2506 	int *pkg_cstate_limits;
2507 
2508 	if (!genuine_intel)
2509 		return 0;
2510 
2511 	if (family != 6)
2512 		return 0;
2513 
2514 	bclk = discover_bclk(family, model);
2515 
2516 	switch (model) {
2517 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
2518 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
2519 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
2520 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
2521 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
2522 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
2523 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
2524 		pkg_cstate_limits = nhm_pkg_cstate_limits;
2525 		break;
2526 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
2527 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
2528 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
2529 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
2530 		pkg_cstate_limits = snb_pkg_cstate_limits;
2531 		has_misc_feature_control = 1;
2532 		break;
2533 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
2534 	case INTEL_FAM6_HASWELL_X:	/* HSX */
2535 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
2536 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
2537 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
2538 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
2539 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
2540 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
2541 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
2542 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
2543 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
2544 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
2545 		pkg_cstate_limits = hsw_pkg_cstate_limits;
2546 		has_misc_feature_control = 1;
2547 		break;
2548 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
2549 		pkg_cstate_limits = skx_pkg_cstate_limits;
2550 		has_misc_feature_control = 1;
2551 		break;
2552 	case INTEL_FAM6_ATOM_SILVERMONT1:	/* BYT */
2553 		no_MSR_MISC_PWR_MGMT = 1;
2554 	case INTEL_FAM6_ATOM_SILVERMONT2:	/* AVN */
2555 		pkg_cstate_limits = slv_pkg_cstate_limits;
2556 		break;
2557 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
2558 		pkg_cstate_limits = amt_pkg_cstate_limits;
2559 		no_MSR_MISC_PWR_MGMT = 1;
2560 		break;
2561 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
2562 	case INTEL_FAM6_XEON_PHI_KNM:
2563 		pkg_cstate_limits = phi_pkg_cstate_limits;
2564 		break;
2565 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
2566 	case INTEL_FAM6_ATOM_GEMINI_LAKE:
2567 	case INTEL_FAM6_ATOM_DENVERTON:	/* DNV */
2568 		pkg_cstate_limits = bxt_pkg_cstate_limits;
2569 		break;
2570 	default:
2571 		return 0;
2572 	}
2573 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2574 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
2575 
2576 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2577 	base_ratio = (msr >> 8) & 0xFF;
2578 
2579 	base_hz = base_ratio * bclk * 1000000;
2580 	has_base_hz = 1;
2581 	return 1;
2582 }
2583 /*
2584  * SLV client has supporet for unique MSRs:
2585  *
2586  * MSR_CC6_DEMOTION_POLICY_CONFIG
2587  * MSR_MC6_DEMOTION_POLICY_CONFIG
2588  */
2589 
2590 int has_slv_msrs(unsigned int family, unsigned int model)
2591 {
2592 	if (!genuine_intel)
2593 		return 0;
2594 
2595 	switch (model) {
2596 	case INTEL_FAM6_ATOM_SILVERMONT1:
2597 	case INTEL_FAM6_ATOM_MERRIFIELD:
2598 	case INTEL_FAM6_ATOM_MOOREFIELD:
2599 		return 1;
2600 	}
2601 	return 0;
2602 }
2603 int is_dnv(unsigned int family, unsigned int model)
2604 {
2605 
2606 	if (!genuine_intel)
2607 		return 0;
2608 
2609 	switch (model) {
2610 	case INTEL_FAM6_ATOM_DENVERTON:
2611 		return 1;
2612 	}
2613 	return 0;
2614 }
2615 int is_skx(unsigned int family, unsigned int model)
2616 {
2617 
2618 	if (!genuine_intel)
2619 		return 0;
2620 
2621 	switch (model) {
2622 	case INTEL_FAM6_SKYLAKE_X:
2623 		return 1;
2624 	}
2625 	return 0;
2626 }
2627 
2628 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
2629 {
2630 	if (has_slv_msrs(family, model))
2631 		return 0;
2632 
2633 	switch (model) {
2634 	/* Nehalem compatible, but do not include turbo-ratio limit support */
2635 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
2636 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
2637 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
2638 	case INTEL_FAM6_XEON_PHI_KNM:
2639 		return 0;
2640 	default:
2641 		return 1;
2642 	}
2643 }
2644 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
2645 {
2646 	if (has_slv_msrs(family, model))
2647 		return 1;
2648 
2649 	return 0;
2650 }
2651 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
2652 {
2653 	if (!genuine_intel)
2654 		return 0;
2655 
2656 	if (family != 6)
2657 		return 0;
2658 
2659 	switch (model) {
2660 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
2661 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
2662 		return 1;
2663 	default:
2664 		return 0;
2665 	}
2666 }
2667 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
2668 {
2669 	if (!genuine_intel)
2670 		return 0;
2671 
2672 	if (family != 6)
2673 		return 0;
2674 
2675 	switch (model) {
2676 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
2677 		return 1;
2678 	default:
2679 		return 0;
2680 	}
2681 }
2682 
2683 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
2684 {
2685 	if (!genuine_intel)
2686 		return 0;
2687 
2688 	if (family != 6)
2689 		return 0;
2690 
2691 	switch (model) {
2692 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
2693 	case INTEL_FAM6_XEON_PHI_KNM:
2694 		return 1;
2695 	default:
2696 		return 0;
2697 	}
2698 }
2699 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
2700 {
2701 	if (!genuine_intel)
2702 		return 0;
2703 
2704 	if (family != 6)
2705 		return 0;
2706 
2707 	switch (model) {
2708 	case INTEL_FAM6_ATOM_GOLDMONT:
2709 	case INTEL_FAM6_SKYLAKE_X:
2710 		return 1;
2711 	default:
2712 		return 0;
2713 	}
2714 }
2715 int has_config_tdp(unsigned int family, unsigned int model)
2716 {
2717 	if (!genuine_intel)
2718 		return 0;
2719 
2720 	if (family != 6)
2721 		return 0;
2722 
2723 	switch (model) {
2724 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
2725 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
2726 	case INTEL_FAM6_HASWELL_X:	/* HSX */
2727 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
2728 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
2729 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
2730 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
2731 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
2732 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
2733 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
2734 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
2735 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
2736 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
2737 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
2738 
2739 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
2740 	case INTEL_FAM6_XEON_PHI_KNM:
2741 		return 1;
2742 	default:
2743 		return 0;
2744 	}
2745 }
2746 
2747 static void
2748 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
2749 {
2750 	if (!do_nhm_platform_info)
2751 		return;
2752 
2753 	dump_nhm_platform_info();
2754 
2755 	if (has_hsw_turbo_ratio_limit(family, model))
2756 		dump_hsw_turbo_ratio_limits();
2757 
2758 	if (has_ivt_turbo_ratio_limit(family, model))
2759 		dump_ivt_turbo_ratio_limits();
2760 
2761 	if (has_turbo_ratio_limit(family, model))
2762 		dump_turbo_ratio_limits(family, model);
2763 
2764 	if (has_atom_turbo_ratio_limit(family, model))
2765 		dump_atom_turbo_ratio_limits();
2766 
2767 	if (has_knl_turbo_ratio_limit(family, model))
2768 		dump_knl_turbo_ratio_limits();
2769 
2770 	if (has_config_tdp(family, model))
2771 		dump_config_tdp();
2772 
2773 	dump_nhm_cst_cfg();
2774 }
2775 
2776 
2777 /*
2778  * print_epb()
2779  * Decode the ENERGY_PERF_BIAS MSR
2780  */
2781 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2782 {
2783 	unsigned long long msr;
2784 	char *epb_string;
2785 	int cpu;
2786 
2787 	if (!has_epb)
2788 		return 0;
2789 
2790 	cpu = t->cpu_id;
2791 
2792 	/* EPB is per-package */
2793 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2794 		return 0;
2795 
2796 	if (cpu_migrate(cpu)) {
2797 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2798 		return -1;
2799 	}
2800 
2801 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
2802 		return 0;
2803 
2804 	switch (msr & 0xF) {
2805 	case ENERGY_PERF_BIAS_PERFORMANCE:
2806 		epb_string = "performance";
2807 		break;
2808 	case ENERGY_PERF_BIAS_NORMAL:
2809 		epb_string = "balanced";
2810 		break;
2811 	case ENERGY_PERF_BIAS_POWERSAVE:
2812 		epb_string = "powersave";
2813 		break;
2814 	default:
2815 		epb_string = "custom";
2816 		break;
2817 	}
2818 	fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
2819 
2820 	return 0;
2821 }
2822 /*
2823  * print_hwp()
2824  * Decode the MSR_HWP_CAPABILITIES
2825  */
2826 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2827 {
2828 	unsigned long long msr;
2829 	int cpu;
2830 
2831 	if (!has_hwp)
2832 		return 0;
2833 
2834 	cpu = t->cpu_id;
2835 
2836 	/* MSR_HWP_CAPABILITIES is per-package */
2837 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2838 		return 0;
2839 
2840 	if (cpu_migrate(cpu)) {
2841 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2842 		return -1;
2843 	}
2844 
2845 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
2846 		return 0;
2847 
2848 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
2849 		cpu, msr, (msr & (1 << 0)) ? "" : "No-");
2850 
2851 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
2852 	if ((msr & (1 << 0)) == 0)
2853 		return 0;
2854 
2855 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
2856 		return 0;
2857 
2858 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
2859 			"(high 0x%x guar 0x%x eff 0x%x low 0x%x)\n",
2860 			cpu, msr,
2861 			(unsigned int)HWP_HIGHEST_PERF(msr),
2862 			(unsigned int)HWP_GUARANTEED_PERF(msr),
2863 			(unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
2864 			(unsigned int)HWP_LOWEST_PERF(msr));
2865 
2866 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
2867 		return 0;
2868 
2869 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
2870 			"(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x pkg 0x%x)\n",
2871 			cpu, msr,
2872 			(unsigned int)(((msr) >> 0) & 0xff),
2873 			(unsigned int)(((msr) >> 8) & 0xff),
2874 			(unsigned int)(((msr) >> 16) & 0xff),
2875 			(unsigned int)(((msr) >> 24) & 0xff),
2876 			(unsigned int)(((msr) >> 32) & 0xff3),
2877 			(unsigned int)(((msr) >> 42) & 0x1));
2878 
2879 	if (has_hwp_pkg) {
2880 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
2881 			return 0;
2882 
2883 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
2884 			"(min 0x%x max 0x%x des 0x%x epp 0x%x window 0x%x)\n",
2885 			cpu, msr,
2886 			(unsigned int)(((msr) >> 0) & 0xff),
2887 			(unsigned int)(((msr) >> 8) & 0xff),
2888 			(unsigned int)(((msr) >> 16) & 0xff),
2889 			(unsigned int)(((msr) >> 24) & 0xff),
2890 			(unsigned int)(((msr) >> 32) & 0xff3));
2891 	}
2892 	if (has_hwp_notify) {
2893 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
2894 			return 0;
2895 
2896 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
2897 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
2898 			cpu, msr,
2899 			((msr) & 0x1) ? "EN" : "Dis",
2900 			((msr) & 0x2) ? "EN" : "Dis");
2901 	}
2902 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
2903 		return 0;
2904 
2905 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
2906 			"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
2907 			cpu, msr,
2908 			((msr) & 0x1) ? "" : "No-",
2909 			((msr) & 0x2) ? "" : "No-");
2910 
2911 	return 0;
2912 }
2913 
2914 /*
2915  * print_perf_limit()
2916  */
2917 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2918 {
2919 	unsigned long long msr;
2920 	int cpu;
2921 
2922 	cpu = t->cpu_id;
2923 
2924 	/* per-package */
2925 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2926 		return 0;
2927 
2928 	if (cpu_migrate(cpu)) {
2929 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
2930 		return -1;
2931 	}
2932 
2933 	if (do_core_perf_limit_reasons) {
2934 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
2935 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2936 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
2937 			(msr & 1 << 15) ? "bit15, " : "",
2938 			(msr & 1 << 14) ? "bit14, " : "",
2939 			(msr & 1 << 13) ? "Transitions, " : "",
2940 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
2941 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
2942 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
2943 			(msr & 1 << 9) ? "CorePwr, " : "",
2944 			(msr & 1 << 8) ? "Amps, " : "",
2945 			(msr & 1 << 6) ? "VR-Therm, " : "",
2946 			(msr & 1 << 5) ? "Auto-HWP, " : "",
2947 			(msr & 1 << 4) ? "Graphics, " : "",
2948 			(msr & 1 << 2) ? "bit2, " : "",
2949 			(msr & 1 << 1) ? "ThermStatus, " : "",
2950 			(msr & 1 << 0) ? "PROCHOT, " : "");
2951 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
2952 			(msr & 1 << 31) ? "bit31, " : "",
2953 			(msr & 1 << 30) ? "bit30, " : "",
2954 			(msr & 1 << 29) ? "Transitions, " : "",
2955 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
2956 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
2957 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
2958 			(msr & 1 << 25) ? "CorePwr, " : "",
2959 			(msr & 1 << 24) ? "Amps, " : "",
2960 			(msr & 1 << 22) ? "VR-Therm, " : "",
2961 			(msr & 1 << 21) ? "Auto-HWP, " : "",
2962 			(msr & 1 << 20) ? "Graphics, " : "",
2963 			(msr & 1 << 18) ? "bit18, " : "",
2964 			(msr & 1 << 17) ? "ThermStatus, " : "",
2965 			(msr & 1 << 16) ? "PROCHOT, " : "");
2966 
2967 	}
2968 	if (do_gfx_perf_limit_reasons) {
2969 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
2970 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2971 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
2972 			(msr & 1 << 0) ? "PROCHOT, " : "",
2973 			(msr & 1 << 1) ? "ThermStatus, " : "",
2974 			(msr & 1 << 4) ? "Graphics, " : "",
2975 			(msr & 1 << 6) ? "VR-Therm, " : "",
2976 			(msr & 1 << 8) ? "Amps, " : "",
2977 			(msr & 1 << 9) ? "GFXPwr, " : "",
2978 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
2979 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
2980 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
2981 			(msr & 1 << 16) ? "PROCHOT, " : "",
2982 			(msr & 1 << 17) ? "ThermStatus, " : "",
2983 			(msr & 1 << 20) ? "Graphics, " : "",
2984 			(msr & 1 << 22) ? "VR-Therm, " : "",
2985 			(msr & 1 << 24) ? "Amps, " : "",
2986 			(msr & 1 << 25) ? "GFXPwr, " : "",
2987 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
2988 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
2989 	}
2990 	if (do_ring_perf_limit_reasons) {
2991 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
2992 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
2993 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
2994 			(msr & 1 << 0) ? "PROCHOT, " : "",
2995 			(msr & 1 << 1) ? "ThermStatus, " : "",
2996 			(msr & 1 << 6) ? "VR-Therm, " : "",
2997 			(msr & 1 << 8) ? "Amps, " : "",
2998 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
2999 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3000 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3001 			(msr & 1 << 16) ? "PROCHOT, " : "",
3002 			(msr & 1 << 17) ? "ThermStatus, " : "",
3003 			(msr & 1 << 22) ? "VR-Therm, " : "",
3004 			(msr & 1 << 24) ? "Amps, " : "",
3005 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3006 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3007 	}
3008 	return 0;
3009 }
3010 
3011 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
3012 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
3013 
3014 double get_tdp(unsigned int model)
3015 {
3016 	unsigned long long msr;
3017 
3018 	if (do_rapl & RAPL_PKG_POWER_INFO)
3019 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3020 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3021 
3022 	switch (model) {
3023 	case INTEL_FAM6_ATOM_SILVERMONT1:
3024 	case INTEL_FAM6_ATOM_SILVERMONT2:
3025 		return 30.0;
3026 	default:
3027 		return 135.0;
3028 	}
3029 }
3030 
3031 /*
3032  * rapl_dram_energy_units_probe()
3033  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3034  */
3035 static double
3036 rapl_dram_energy_units_probe(int  model, double rapl_energy_units)
3037 {
3038 	/* only called for genuine_intel, family 6 */
3039 
3040 	switch (model) {
3041 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3042 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3043 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
3044 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3045 	case INTEL_FAM6_XEON_PHI_KNM:
3046 		return (rapl_dram_energy_units = 15.3 / 1000000);
3047 	default:
3048 		return (rapl_energy_units);
3049 	}
3050 }
3051 
3052 
3053 /*
3054  * rapl_probe()
3055  *
3056  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
3057  */
3058 void rapl_probe(unsigned int family, unsigned int model)
3059 {
3060 	unsigned long long msr;
3061 	unsigned int time_unit;
3062 	double tdp;
3063 
3064 	if (!genuine_intel)
3065 		return;
3066 
3067 	if (family != 6)
3068 		return;
3069 
3070 	switch (model) {
3071 	case INTEL_FAM6_SANDYBRIDGE:
3072 	case INTEL_FAM6_IVYBRIDGE:
3073 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3074 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3075 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3076 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3077 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3078 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3079 		if (rapl_joules) {
3080 			BIC_PRESENT(BIC_Pkg_J);
3081 			BIC_PRESENT(BIC_Cor_J);
3082 			BIC_PRESENT(BIC_GFX_J);
3083 		} else {
3084 			BIC_PRESENT(BIC_PkgWatt);
3085 			BIC_PRESENT(BIC_CorWatt);
3086 			BIC_PRESENT(BIC_GFXWatt);
3087 		}
3088 		break;
3089 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3090 	case INTEL_FAM6_ATOM_GEMINI_LAKE:
3091 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3092 		if (rapl_joules)
3093 			BIC_PRESENT(BIC_Pkg_J);
3094 		else
3095 			BIC_PRESENT(BIC_PkgWatt);
3096 		break;
3097 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3098 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
3099 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
3100 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
3101 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3102 		BIC_PRESENT(BIC_PKG__);
3103 		BIC_PRESENT(BIC_RAM__);
3104 		if (rapl_joules) {
3105 			BIC_PRESENT(BIC_Pkg_J);
3106 			BIC_PRESENT(BIC_Cor_J);
3107 			BIC_PRESENT(BIC_RAM_J);
3108 		} else {
3109 			BIC_PRESENT(BIC_PkgWatt);
3110 			BIC_PRESENT(BIC_CorWatt);
3111 			BIC_PRESENT(BIC_RAMWatt);
3112 		}
3113 		break;
3114 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3115 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3116 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
3117 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3118 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3119 	case INTEL_FAM6_XEON_PHI_KNM:
3120 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3121 		BIC_PRESENT(BIC_PKG__);
3122 		BIC_PRESENT(BIC_RAM__);
3123 		if (rapl_joules) {
3124 			BIC_PRESENT(BIC_Pkg_J);
3125 			BIC_PRESENT(BIC_RAM_J);
3126 		} else {
3127 			BIC_PRESENT(BIC_PkgWatt);
3128 			BIC_PRESENT(BIC_RAMWatt);
3129 		}
3130 		break;
3131 	case INTEL_FAM6_SANDYBRIDGE_X:
3132 	case INTEL_FAM6_IVYBRIDGE_X:
3133 		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;
3134 		BIC_PRESENT(BIC_PKG__);
3135 		BIC_PRESENT(BIC_RAM__);
3136 		if (rapl_joules) {
3137 			BIC_PRESENT(BIC_Pkg_J);
3138 			BIC_PRESENT(BIC_Cor_J);
3139 			BIC_PRESENT(BIC_RAM_J);
3140 		} else {
3141 			BIC_PRESENT(BIC_PkgWatt);
3142 			BIC_PRESENT(BIC_CorWatt);
3143 			BIC_PRESENT(BIC_RAMWatt);
3144 		}
3145 		break;
3146 	case INTEL_FAM6_ATOM_SILVERMONT1:	/* BYT */
3147 	case INTEL_FAM6_ATOM_SILVERMONT2:	/* AVN */
3148 		do_rapl = RAPL_PKG | RAPL_CORES;
3149 		if (rapl_joules) {
3150 			BIC_PRESENT(BIC_Pkg_J);
3151 			BIC_PRESENT(BIC_Cor_J);
3152 		} else {
3153 			BIC_PRESENT(BIC_PkgWatt);
3154 			BIC_PRESENT(BIC_CorWatt);
3155 		}
3156 		break;
3157 	case INTEL_FAM6_ATOM_DENVERTON:	/* DNV */
3158 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO | RAPL_CORES_ENERGY_STATUS;
3159 		BIC_PRESENT(BIC_PKG__);
3160 		BIC_PRESENT(BIC_RAM__);
3161 		if (rapl_joules) {
3162 			BIC_PRESENT(BIC_Pkg_J);
3163 			BIC_PRESENT(BIC_Cor_J);
3164 			BIC_PRESENT(BIC_RAM_J);
3165 		} else {
3166 			BIC_PRESENT(BIC_PkgWatt);
3167 			BIC_PRESENT(BIC_CorWatt);
3168 			BIC_PRESENT(BIC_RAMWatt);
3169 		}
3170 		break;
3171 	default:
3172 		return;
3173 	}
3174 
3175 	/* units on package 0, verify later other packages match */
3176 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3177 		return;
3178 
3179 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
3180 	if (model == INTEL_FAM6_ATOM_SILVERMONT1)
3181 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3182 	else
3183 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3184 
3185 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3186 
3187 	time_unit = msr >> 16 & 0xF;
3188 	if (time_unit == 0)
3189 		time_unit = 0xA;
3190 
3191 	rapl_time_units = 1.0 / (1 << (time_unit));
3192 
3193 	tdp = get_tdp(model);
3194 
3195 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3196 	if (!quiet)
3197 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3198 
3199 	return;
3200 }
3201 
3202 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
3203 {
3204 	if (!genuine_intel)
3205 		return;
3206 
3207 	if (family != 6)
3208 		return;
3209 
3210 	switch (model) {
3211 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3212 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3213 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3214 		do_gfx_perf_limit_reasons = 1;
3215 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3216 		do_core_perf_limit_reasons = 1;
3217 		do_ring_perf_limit_reasons = 1;
3218 	default:
3219 		return;
3220 	}
3221 }
3222 
3223 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3224 {
3225 	unsigned long long msr;
3226 	unsigned int dts;
3227 	int cpu;
3228 
3229 	if (!(do_dts || do_ptm))
3230 		return 0;
3231 
3232 	cpu = t->cpu_id;
3233 
3234 	/* DTS is per-core, no need to print for each thread */
3235 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
3236 		return 0;
3237 
3238 	if (cpu_migrate(cpu)) {
3239 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3240 		return -1;
3241 	}
3242 
3243 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
3244 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
3245 			return 0;
3246 
3247 		dts = (msr >> 16) & 0x7F;
3248 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
3249 			cpu, msr, tcc_activation_temp - dts);
3250 
3251 #ifdef	THERM_DEBUG
3252 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
3253 			return 0;
3254 
3255 		dts = (msr >> 16) & 0x7F;
3256 		dts2 = (msr >> 8) & 0x7F;
3257 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
3258 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
3259 #endif
3260 	}
3261 
3262 
3263 	if (do_dts) {
3264 		unsigned int resolution;
3265 
3266 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
3267 			return 0;
3268 
3269 		dts = (msr >> 16) & 0x7F;
3270 		resolution = (msr >> 27) & 0xF;
3271 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
3272 			cpu, msr, tcc_activation_temp - dts, resolution);
3273 
3274 #ifdef THERM_DEBUG
3275 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
3276 			return 0;
3277 
3278 		dts = (msr >> 16) & 0x7F;
3279 		dts2 = (msr >> 8) & 0x7F;
3280 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
3281 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
3282 #endif
3283 	}
3284 
3285 	return 0;
3286 }
3287 
3288 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
3289 {
3290 	fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
3291 		cpu, label,
3292 		((msr >> 15) & 1) ? "EN" : "DIS",
3293 		((msr >> 0) & 0x7FFF) * rapl_power_units,
3294 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
3295 		(((msr >> 16) & 1) ? "EN" : "DIS"));
3296 
3297 	return;
3298 }
3299 
3300 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3301 {
3302 	unsigned long long msr;
3303 	int cpu;
3304 
3305 	if (!do_rapl)
3306 		return 0;
3307 
3308 	/* RAPL counters are per package, so print only for 1st thread/package */
3309 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3310 		return 0;
3311 
3312 	cpu = t->cpu_id;
3313 	if (cpu_migrate(cpu)) {
3314 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3315 		return -1;
3316 	}
3317 
3318 	if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
3319 		return -1;
3320 
3321 	fprintf(outf, "cpu%d: MSR_RAPL_POWER_UNIT: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr,
3322 		rapl_power_units, rapl_energy_units, rapl_time_units);
3323 
3324 	if (do_rapl & RAPL_PKG_POWER_INFO) {
3325 
3326 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
3327                 	return -5;
3328 
3329 
3330 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
3331 			cpu, msr,
3332 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3333 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3334 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3335 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
3336 
3337 	}
3338 	if (do_rapl & RAPL_PKG) {
3339 
3340 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
3341 			return -9;
3342 
3343 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
3344 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
3345 
3346 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
3347 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
3348 			cpu,
3349 			((msr >> 47) & 1) ? "EN" : "DIS",
3350 			((msr >> 32) & 0x7FFF) * rapl_power_units,
3351 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
3352 			((msr >> 48) & 1) ? "EN" : "DIS");
3353 	}
3354 
3355 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
3356 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
3357                 	return -6;
3358 
3359 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
3360 			cpu, msr,
3361 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3362 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3363 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
3364 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
3365 	}
3366 	if (do_rapl & RAPL_DRAM) {
3367 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
3368 			return -9;
3369 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
3370 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3371 
3372 		print_power_limit_msr(cpu, msr, "DRAM Limit");
3373 	}
3374 	if (do_rapl & RAPL_CORE_POLICY) {
3375 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
3376 			return -7;
3377 
3378 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
3379 	}
3380 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
3381 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
3382 			return -9;
3383 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
3384 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3385 		print_power_limit_msr(cpu, msr, "Cores Limit");
3386 	}
3387 	if (do_rapl & RAPL_GFX) {
3388 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
3389 			return -8;
3390 
3391 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
3392 
3393 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
3394 			return -9;
3395 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
3396 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
3397 		print_power_limit_msr(cpu, msr, "GFX Limit");
3398 	}
3399 	return 0;
3400 }
3401 
3402 /*
3403  * SNB adds support for additional MSRs:
3404  *
3405  * MSR_PKG_C7_RESIDENCY            0x000003fa
3406  * MSR_CORE_C7_RESIDENCY           0x000003fe
3407  * MSR_PKG_C2_RESIDENCY            0x0000060d
3408  */
3409 
3410 int has_snb_msrs(unsigned int family, unsigned int model)
3411 {
3412 	if (!genuine_intel)
3413 		return 0;
3414 
3415 	switch (model) {
3416 	case INTEL_FAM6_SANDYBRIDGE:
3417 	case INTEL_FAM6_SANDYBRIDGE_X:
3418 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3419 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3420 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3421 	case INTEL_FAM6_HASWELL_X:	/* HSW */
3422 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3423 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3424 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3425 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3426 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3427 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
3428 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3429 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
3430 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
3431 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
3432 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3433 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3434 	case INTEL_FAM6_ATOM_GEMINI_LAKE:
3435 	case INTEL_FAM6_ATOM_DENVERTON:	/* DNV */
3436 		return 1;
3437 	}
3438 	return 0;
3439 }
3440 
3441 /*
3442  * HSW adds support for additional MSRs:
3443  *
3444  * MSR_PKG_C8_RESIDENCY		0x00000630
3445  * MSR_PKG_C9_RESIDENCY		0x00000631
3446  * MSR_PKG_C10_RESIDENCY	0x00000632
3447  *
3448  * MSR_PKGC8_IRTL		0x00000633
3449  * MSR_PKGC9_IRTL		0x00000634
3450  * MSR_PKGC10_IRTL		0x00000635
3451  *
3452  */
3453 int has_hsw_msrs(unsigned int family, unsigned int model)
3454 {
3455 	if (!genuine_intel)
3456 		return 0;
3457 
3458 	switch (model) {
3459 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3460 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3461 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3462 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
3463 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
3464 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
3465 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3466 	case INTEL_FAM6_ATOM_GEMINI_LAKE:
3467 		return 1;
3468 	}
3469 	return 0;
3470 }
3471 
3472 /*
3473  * SKL adds support for additional MSRS:
3474  *
3475  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
3476  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
3477  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
3478  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
3479  */
3480 int has_skl_msrs(unsigned int family, unsigned int model)
3481 {
3482 	if (!genuine_intel)
3483 		return 0;
3484 
3485 	switch (model) {
3486 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3487 	case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
3488 	case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
3489 	case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
3490 		return 1;
3491 	}
3492 	return 0;
3493 }
3494 
3495 int is_slm(unsigned int family, unsigned int model)
3496 {
3497 	if (!genuine_intel)
3498 		return 0;
3499 	switch (model) {
3500 	case INTEL_FAM6_ATOM_SILVERMONT1:	/* BYT */
3501 	case INTEL_FAM6_ATOM_SILVERMONT2:	/* AVN */
3502 		return 1;
3503 	}
3504 	return 0;
3505 }
3506 
3507 int is_knl(unsigned int family, unsigned int model)
3508 {
3509 	if (!genuine_intel)
3510 		return 0;
3511 	switch (model) {
3512 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3513 	case INTEL_FAM6_XEON_PHI_KNM:
3514 		return 1;
3515 	}
3516 	return 0;
3517 }
3518 
3519 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
3520 {
3521 	if (is_knl(family, model))
3522 		return 1024;
3523 	return 1;
3524 }
3525 
3526 #define SLM_BCLK_FREQS 5
3527 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
3528 
3529 double slm_bclk(void)
3530 {
3531 	unsigned long long msr = 3;
3532 	unsigned int i;
3533 	double freq;
3534 
3535 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
3536 		fprintf(outf, "SLM BCLK: unknown\n");
3537 
3538 	i = msr & 0xf;
3539 	if (i >= SLM_BCLK_FREQS) {
3540 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
3541 		i = 3;
3542 	}
3543 	freq = slm_freq_table[i];
3544 
3545 	if (!quiet)
3546 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
3547 
3548 	return freq;
3549 }
3550 
3551 double discover_bclk(unsigned int family, unsigned int model)
3552 {
3553 	if (has_snb_msrs(family, model) || is_knl(family, model))
3554 		return 100.00;
3555 	else if (is_slm(family, model))
3556 		return slm_bclk();
3557 	else
3558 		return 133.33;
3559 }
3560 
3561 /*
3562  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
3563  * the Thermal Control Circuit (TCC) activates.
3564  * This is usually equal to tjMax.
3565  *
3566  * Older processors do not have this MSR, so there we guess,
3567  * but also allow cmdline over-ride with -T.
3568  *
3569  * Several MSR temperature values are in units of degrees-C
3570  * below this value, including the Digital Thermal Sensor (DTS),
3571  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
3572  */
3573 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3574 {
3575 	unsigned long long msr;
3576 	unsigned int target_c_local;
3577 	int cpu;
3578 
3579 	/* tcc_activation_temp is used only for dts or ptm */
3580 	if (!(do_dts || do_ptm))
3581 		return 0;
3582 
3583 	/* this is a per-package concept */
3584 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3585 		return 0;
3586 
3587 	cpu = t->cpu_id;
3588 	if (cpu_migrate(cpu)) {
3589 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3590 		return -1;
3591 	}
3592 
3593 	if (tcc_activation_temp_override != 0) {
3594 		tcc_activation_temp = tcc_activation_temp_override;
3595 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
3596 			cpu, tcc_activation_temp);
3597 		return 0;
3598 	}
3599 
3600 	/* Temperature Target MSR is Nehalem and newer only */
3601 	if (!do_nhm_platform_info)
3602 		goto guess;
3603 
3604 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
3605 		goto guess;
3606 
3607 	target_c_local = (msr >> 16) & 0xFF;
3608 
3609 	if (!quiet)
3610 		fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
3611 			cpu, msr, target_c_local);
3612 
3613 	if (!target_c_local)
3614 		goto guess;
3615 
3616 	tcc_activation_temp = target_c_local;
3617 
3618 	return 0;
3619 
3620 guess:
3621 	tcc_activation_temp = TJMAX_DEFAULT;
3622 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
3623 		cpu, tcc_activation_temp);
3624 
3625 	return 0;
3626 }
3627 
3628 void decode_feature_control_msr(void)
3629 {
3630 	unsigned long long msr;
3631 
3632 	if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
3633 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
3634 			base_cpu, msr,
3635 			msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
3636 			msr & (1 << 18) ? "SGX" : "");
3637 }
3638 
3639 void decode_misc_enable_msr(void)
3640 {
3641 	unsigned long long msr;
3642 
3643 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
3644 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
3645 			base_cpu, msr,
3646 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
3647 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
3648 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "No-" : "",
3649 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
3650 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
3651 }
3652 
3653 void decode_misc_feature_control(void)
3654 {
3655 	unsigned long long msr;
3656 
3657 	if (!has_misc_feature_control)
3658 		return;
3659 
3660 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
3661 		fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
3662 			base_cpu, msr,
3663 			msr & (0 << 0) ? "No-" : "",
3664 			msr & (1 << 0) ? "No-" : "",
3665 			msr & (2 << 0) ? "No-" : "",
3666 			msr & (3 << 0) ? "No-" : "");
3667 }
3668 /*
3669  * Decode MSR_MISC_PWR_MGMT
3670  *
3671  * Decode the bits according to the Nehalem documentation
3672  * bit[0] seems to continue to have same meaning going forward
3673  * bit[1] less so...
3674  */
3675 void decode_misc_pwr_mgmt_msr(void)
3676 {
3677 	unsigned long long msr;
3678 
3679 	if (!do_nhm_platform_info)
3680 		return;
3681 
3682 	if (no_MSR_MISC_PWR_MGMT)
3683 		return;
3684 
3685 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
3686 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
3687 			base_cpu, msr,
3688 			msr & (1 << 0) ? "DIS" : "EN",
3689 			msr & (1 << 1) ? "EN" : "DIS",
3690 			msr & (1 << 8) ? "EN" : "DIS");
3691 }
3692 /*
3693  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
3694  *
3695  * This MSRs are present on Silvermont processors,
3696  * Intel Atom processor E3000 series (Baytrail), and friends.
3697  */
3698 void decode_c6_demotion_policy_msr(void)
3699 {
3700 	unsigned long long msr;
3701 
3702 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
3703 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
3704 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
3705 
3706 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
3707 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
3708 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
3709 }
3710 
3711 void process_cpuid()
3712 {
3713 	unsigned int eax, ebx, ecx, edx, max_level, max_extended_level;
3714 	unsigned int fms, family, model, stepping;
3715 	unsigned int has_turbo;
3716 
3717 	eax = ebx = ecx = edx = 0;
3718 
3719 	__cpuid(0, max_level, ebx, ecx, edx);
3720 
3721 	if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
3722 		genuine_intel = 1;
3723 
3724 	if (!quiet)
3725 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
3726 			(char *)&ebx, (char *)&edx, (char *)&ecx);
3727 
3728 	__cpuid(1, fms, ebx, ecx, edx);
3729 	family = (fms >> 8) & 0xf;
3730 	model = (fms >> 4) & 0xf;
3731 	stepping = fms & 0xf;
3732 	if (family == 6 || family == 0xf)
3733 		model += ((fms >> 16) & 0xf) << 4;
3734 
3735 	if (!quiet) {
3736 		fprintf(outf, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
3737 			max_level, family, model, stepping, family, model, stepping);
3738 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s\n",
3739 			ecx & (1 << 0) ? "SSE3" : "-",
3740 			ecx & (1 << 3) ? "MONITOR" : "-",
3741 			ecx & (1 << 6) ? "SMX" : "-",
3742 			ecx & (1 << 7) ? "EIST" : "-",
3743 			ecx & (1 << 8) ? "TM2" : "-",
3744 			edx & (1 << 4) ? "TSC" : "-",
3745 			edx & (1 << 5) ? "MSR" : "-",
3746 			edx & (1 << 22) ? "ACPI-TM" : "-",
3747 			edx & (1 << 29) ? "TM" : "-");
3748 	}
3749 
3750 	if (!(edx & (1 << 5)))
3751 		errx(1, "CPUID: no MSR");
3752 
3753 	/*
3754 	 * check max extended function levels of CPUID.
3755 	 * This is needed to check for invariant TSC.
3756 	 * This check is valid for both Intel and AMD.
3757 	 */
3758 	ebx = ecx = edx = 0;
3759 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
3760 
3761 	if (max_extended_level >= 0x80000007) {
3762 
3763 		/*
3764 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
3765 		 * this check is valid for both Intel and AMD
3766 		 */
3767 		__cpuid(0x80000007, eax, ebx, ecx, edx);
3768 		has_invariant_tsc = edx & (1 << 8);
3769 	}
3770 
3771 	/*
3772 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
3773 	 * this check is valid for both Intel and AMD
3774 	 */
3775 
3776 	__cpuid(0x6, eax, ebx, ecx, edx);
3777 	has_aperf = ecx & (1 << 0);
3778 	if (has_aperf) {
3779 		BIC_PRESENT(BIC_Avg_MHz);
3780 		BIC_PRESENT(BIC_Busy);
3781 		BIC_PRESENT(BIC_Bzy_MHz);
3782 	}
3783 	do_dts = eax & (1 << 0);
3784 	if (do_dts)
3785 		BIC_PRESENT(BIC_CoreTmp);
3786 	has_turbo = eax & (1 << 1);
3787 	do_ptm = eax & (1 << 6);
3788 	if (do_ptm)
3789 		BIC_PRESENT(BIC_PkgTmp);
3790 	has_hwp = eax & (1 << 7);
3791 	has_hwp_notify = eax & (1 << 8);
3792 	has_hwp_activity_window = eax & (1 << 9);
3793 	has_hwp_epp = eax & (1 << 10);
3794 	has_hwp_pkg = eax & (1 << 11);
3795 	has_epb = ecx & (1 << 3);
3796 
3797 	if (!quiet)
3798 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
3799 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
3800 			has_aperf ? "" : "No-",
3801 			has_turbo ? "" : "No-",
3802 			do_dts ? "" : "No-",
3803 			do_ptm ? "" : "No-",
3804 			has_hwp ? "" : "No-",
3805 			has_hwp_notify ? "" : "No-",
3806 			has_hwp_activity_window ? "" : "No-",
3807 			has_hwp_epp ? "" : "No-",
3808 			has_hwp_pkg ? "" : "No-",
3809 			has_epb ? "" : "No-");
3810 
3811 	if (!quiet)
3812 		decode_misc_enable_msr();
3813 
3814 
3815 	if (max_level >= 0x7 && !quiet) {
3816 		int has_sgx;
3817 
3818 		ecx = 0;
3819 
3820 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
3821 
3822 		has_sgx = ebx & (1 << 2);
3823 		fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
3824 
3825 		if (has_sgx)
3826 			decode_feature_control_msr();
3827 	}
3828 
3829 	if (max_level >= 0x15) {
3830 		unsigned int eax_crystal;
3831 		unsigned int ebx_tsc;
3832 
3833 		/*
3834 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
3835 		 */
3836 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
3837 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
3838 
3839 		if (ebx_tsc != 0) {
3840 
3841 			if (!quiet && (ebx != 0))
3842 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
3843 					eax_crystal, ebx_tsc, crystal_hz);
3844 
3845 			if (crystal_hz == 0)
3846 				switch(model) {
3847 				case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3848 				case INTEL_FAM6_SKYLAKE_DESKTOP:	/* SKL */
3849 				case INTEL_FAM6_KABYLAKE_MOBILE:	/* KBL */
3850 				case INTEL_FAM6_KABYLAKE_DESKTOP:	/* KBL */
3851 					crystal_hz = 24000000;	/* 24.0 MHz */
3852 					break;
3853 				case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3854 				case INTEL_FAM6_ATOM_DENVERTON:	/* DNV */
3855 					crystal_hz = 25000000;	/* 25.0 MHz */
3856 					break;
3857 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3858 				case INTEL_FAM6_ATOM_GEMINI_LAKE:
3859 					crystal_hz = 19200000;	/* 19.2 MHz */
3860 					break;
3861 				default:
3862 					crystal_hz = 0;
3863 			}
3864 
3865 			if (crystal_hz) {
3866 				tsc_hz =  (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
3867 				if (!quiet)
3868 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
3869 						tsc_hz / 1000000, crystal_hz, ebx_tsc,  eax_crystal);
3870 			}
3871 		}
3872 	}
3873 	if (max_level >= 0x16) {
3874 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
3875 
3876 		/*
3877 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
3878 		 */
3879 		base_mhz = max_mhz = bus_mhz = edx = 0;
3880 
3881 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
3882 		if (!quiet)
3883 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
3884 				base_mhz, max_mhz, bus_mhz);
3885 	}
3886 
3887 	if (has_aperf)
3888 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
3889 
3890 	BIC_PRESENT(BIC_IRQ);
3891 	BIC_PRESENT(BIC_TSC_MHz);
3892 
3893 	if (probe_nhm_msrs(family, model)) {
3894 		do_nhm_platform_info = 1;
3895 		BIC_PRESENT(BIC_CPU_c1);
3896 		BIC_PRESENT(BIC_CPU_c3);
3897 		BIC_PRESENT(BIC_CPU_c6);
3898 		BIC_PRESENT(BIC_SMI);
3899 	}
3900 	do_snb_cstates = has_snb_msrs(family, model);
3901 
3902 	if (do_snb_cstates)
3903 		BIC_PRESENT(BIC_CPU_c7);
3904 
3905 	do_irtl_snb = has_snb_msrs(family, model);
3906 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
3907 		BIC_PRESENT(BIC_Pkgpc2);
3908 	if (pkg_cstate_limit >= PCL__3)
3909 		BIC_PRESENT(BIC_Pkgpc3);
3910 	if (pkg_cstate_limit >= PCL__6)
3911 		BIC_PRESENT(BIC_Pkgpc6);
3912 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
3913 		BIC_PRESENT(BIC_Pkgpc7);
3914 	if (has_slv_msrs(family, model)) {
3915 		BIC_NOT_PRESENT(BIC_Pkgpc2);
3916 		BIC_NOT_PRESENT(BIC_Pkgpc3);
3917 		BIC_PRESENT(BIC_Pkgpc6);
3918 		BIC_NOT_PRESENT(BIC_Pkgpc7);
3919 		BIC_PRESENT(BIC_Mod_c6);
3920 		use_c1_residency_msr = 1;
3921 	}
3922 	if (is_dnv(family, model)) {
3923 		BIC_PRESENT(BIC_CPU_c1);
3924 		BIC_NOT_PRESENT(BIC_CPU_c3);
3925 		BIC_NOT_PRESENT(BIC_Pkgpc3);
3926 		BIC_NOT_PRESENT(BIC_CPU_c7);
3927 		BIC_NOT_PRESENT(BIC_Pkgpc7);
3928 		use_c1_residency_msr = 1;
3929 	}
3930 	if (is_skx(family, model)) {
3931 		BIC_NOT_PRESENT(BIC_CPU_c3);
3932 		BIC_NOT_PRESENT(BIC_Pkgpc3);
3933 		BIC_NOT_PRESENT(BIC_CPU_c7);
3934 		BIC_NOT_PRESENT(BIC_Pkgpc7);
3935 	}
3936 	if (has_hsw_msrs(family, model)) {
3937 		BIC_PRESENT(BIC_Pkgpc8);
3938 		BIC_PRESENT(BIC_Pkgpc9);
3939 		BIC_PRESENT(BIC_Pkgpc10);
3940 	}
3941 	do_irtl_hsw = has_hsw_msrs(family, model);
3942 	do_skl_residency = has_skl_msrs(family, model);
3943 	do_slm_cstates = is_slm(family, model);
3944 	do_knl_cstates  = is_knl(family, model);
3945 
3946 	if (!quiet)
3947 		decode_misc_pwr_mgmt_msr();
3948 
3949 	if (!quiet && has_slv_msrs(family, model))
3950 		decode_c6_demotion_policy_msr();
3951 
3952 	rapl_probe(family, model);
3953 	perf_limit_reasons_probe(family, model);
3954 
3955 	if (!quiet)
3956 		dump_cstate_pstate_config_info(family, model);
3957 
3958 	if (has_skl_msrs(family, model))
3959 		calculate_tsc_tweak();
3960 
3961 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
3962 		BIC_PRESENT(BIC_GFX_rc6);
3963 
3964 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
3965 		BIC_PRESENT(BIC_GFXMHz);
3966 
3967 	if (!quiet)
3968 		decode_misc_feature_control();
3969 
3970 	return;
3971 }
3972 
3973 void help()
3974 {
3975 	fprintf(outf,
3976 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
3977 	"\n"
3978 	"Turbostat forks the specified COMMAND and prints statistics\n"
3979 	"when COMMAND completes.\n"
3980 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
3981 	"to print statistics, until interrupted.\n"
3982 	"--add		add a counter\n"
3983 	"		eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
3984 	"--quiet	skip decoding system configuration header\n"
3985 	"--interval sec	Override default 5-second measurement interval\n"
3986 	"--help		print this help message\n"
3987 	"--out file	create or truncate \"file\" for all output\n"
3988 	"--version	print version information\n"
3989 	"\n"
3990 	"For more help, run \"man turbostat\"\n");
3991 }
3992 
3993 
3994 /*
3995  * in /dev/cpu/ return success for names that are numbers
3996  * ie. filter out ".", "..", "microcode".
3997  */
3998 int dir_filter(const struct dirent *dirp)
3999 {
4000 	if (isdigit(dirp->d_name[0]))
4001 		return 1;
4002 	else
4003 		return 0;
4004 }
4005 
4006 int open_dev_cpu_msr(int dummy1)
4007 {
4008 	return 0;
4009 }
4010 
4011 void topology_probe()
4012 {
4013 	int i;
4014 	int max_core_id = 0;
4015 	int max_package_id = 0;
4016 	int max_siblings = 0;
4017 	struct cpu_topology {
4018 		int core_id;
4019 		int physical_package_id;
4020 	} *cpus;
4021 
4022 	/* Initialize num_cpus, max_cpu_num */
4023 	topo.num_cpus = 0;
4024 	topo.max_cpu_num = 0;
4025 	for_all_proc_cpus(count_cpus);
4026 	if (!summary_only && topo.num_cpus > 1)
4027 		BIC_PRESENT(BIC_CPU);
4028 
4029 	if (debug > 1)
4030 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4031 
4032 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
4033 	if (cpus == NULL)
4034 		err(1, "calloc cpus");
4035 
4036 	/*
4037 	 * Allocate and initialize cpu_present_set
4038 	 */
4039 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4040 	if (cpu_present_set == NULL)
4041 		err(3, "CPU_ALLOC");
4042 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4043 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4044 	for_all_proc_cpus(mark_cpu_present);
4045 
4046 	/*
4047 	 * Allocate and initialize cpu_affinity_set
4048 	 */
4049 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4050 	if (cpu_affinity_set == NULL)
4051 		err(3, "CPU_ALLOC");
4052 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4053 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4054 
4055 
4056 	/*
4057 	 * For online cpus
4058 	 * find max_core_id, max_package_id
4059 	 */
4060 	for (i = 0; i <= topo.max_cpu_num; ++i) {
4061 		int siblings;
4062 
4063 		if (cpu_is_not_present(i)) {
4064 			if (debug > 1)
4065 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
4066 			continue;
4067 		}
4068 		cpus[i].core_id = get_core_id(i);
4069 		if (cpus[i].core_id > max_core_id)
4070 			max_core_id = cpus[i].core_id;
4071 
4072 		cpus[i].physical_package_id = get_physical_package_id(i);
4073 		if (cpus[i].physical_package_id > max_package_id)
4074 			max_package_id = cpus[i].physical_package_id;
4075 
4076 		siblings = get_num_ht_siblings(i);
4077 		if (siblings > max_siblings)
4078 			max_siblings = siblings;
4079 		if (debug > 1)
4080 			fprintf(outf, "cpu %d pkg %d core %d\n",
4081 				i, cpus[i].physical_package_id, cpus[i].core_id);
4082 	}
4083 	topo.num_cores_per_pkg = max_core_id + 1;
4084 	if (debug > 1)
4085 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
4086 			max_core_id, topo.num_cores_per_pkg);
4087 	if (!summary_only && topo.num_cores_per_pkg > 1)
4088 		BIC_PRESENT(BIC_Core);
4089 
4090 	topo.num_packages = max_package_id + 1;
4091 	if (debug > 1)
4092 		fprintf(outf, "max_package_id %d, sizing for %d packages\n",
4093 			max_package_id, topo.num_packages);
4094 	if (debug && !summary_only && topo.num_packages > 1)
4095 		BIC_PRESENT(BIC_Package);
4096 
4097 	topo.num_threads_per_core = max_siblings;
4098 	if (debug > 1)
4099 		fprintf(outf, "max_siblings %d\n", max_siblings);
4100 
4101 	free(cpus);
4102 }
4103 
4104 void
4105 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
4106 {
4107 	int i;
4108 
4109 	*t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
4110 		topo.num_packages, sizeof(struct thread_data));
4111 	if (*t == NULL)
4112 		goto error;
4113 
4114 	for (i = 0; i < topo.num_threads_per_core *
4115 		topo.num_cores_per_pkg * topo.num_packages; i++)
4116 		(*t)[i].cpu_id = -1;
4117 
4118 	*c = calloc(topo.num_cores_per_pkg * topo.num_packages,
4119 		sizeof(struct core_data));
4120 	if (*c == NULL)
4121 		goto error;
4122 
4123 	for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
4124 		(*c)[i].core_id = -1;
4125 
4126 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
4127 	if (*p == NULL)
4128 		goto error;
4129 
4130 	for (i = 0; i < topo.num_packages; i++)
4131 		(*p)[i].package_id = i;
4132 
4133 	return;
4134 error:
4135 	err(1, "calloc counters");
4136 }
4137 /*
4138  * init_counter()
4139  *
4140  * set cpu_id, core_num, pkg_num
4141  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
4142  *
4143  * increment topo.num_cores when 1st core in pkg seen
4144  */
4145 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
4146 	struct pkg_data *pkg_base, int thread_num, int core_num,
4147 	int pkg_num, int cpu_id)
4148 {
4149 	struct thread_data *t;
4150 	struct core_data *c;
4151 	struct pkg_data *p;
4152 
4153 	t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
4154 	c = GET_CORE(core_base, core_num, pkg_num);
4155 	p = GET_PKG(pkg_base, pkg_num);
4156 
4157 	t->cpu_id = cpu_id;
4158 	if (thread_num == 0) {
4159 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
4160 		if (cpu_is_first_core_in_package(cpu_id))
4161 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
4162 	}
4163 
4164 	c->core_id = core_num;
4165 	p->package_id = pkg_num;
4166 }
4167 
4168 
4169 int initialize_counters(int cpu_id)
4170 {
4171 	int my_thread_id, my_core_id, my_package_id;
4172 
4173 	my_package_id = get_physical_package_id(cpu_id);
4174 	my_core_id = get_core_id(cpu_id);
4175 	my_thread_id = get_cpu_position_in_core(cpu_id);
4176 	if (!my_thread_id)
4177 		topo.num_cores++;
4178 
4179 	init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
4180 	init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
4181 	return 0;
4182 }
4183 
4184 void allocate_output_buffer()
4185 {
4186 	output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
4187 	outp = output_buffer;
4188 	if (outp == NULL)
4189 		err(-1, "calloc output buffer");
4190 }
4191 void allocate_fd_percpu(void)
4192 {
4193 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
4194 	if (fd_percpu == NULL)
4195 		err(-1, "calloc fd_percpu");
4196 }
4197 void allocate_irq_buffers(void)
4198 {
4199 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
4200 	if (irq_column_2_cpu == NULL)
4201 		err(-1, "calloc %d", topo.num_cpus);
4202 
4203 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
4204 	if (irqs_per_cpu == NULL)
4205 		err(-1, "calloc %d", topo.max_cpu_num + 1);
4206 }
4207 void setup_all_buffers(void)
4208 {
4209 	topology_probe();
4210 	allocate_irq_buffers();
4211 	allocate_fd_percpu();
4212 	allocate_counters(&thread_even, &core_even, &package_even);
4213 	allocate_counters(&thread_odd, &core_odd, &package_odd);
4214 	allocate_output_buffer();
4215 	for_all_proc_cpus(initialize_counters);
4216 }
4217 
4218 void set_base_cpu(void)
4219 {
4220 	base_cpu = sched_getcpu();
4221 	if (base_cpu < 0)
4222 		err(-ENODEV, "No valid cpus found");
4223 
4224 	if (debug > 1)
4225 		fprintf(outf, "base_cpu = %d\n", base_cpu);
4226 }
4227 
4228 void turbostat_init()
4229 {
4230 	setup_all_buffers();
4231 	set_base_cpu();
4232 	check_dev_msr();
4233 	check_permissions();
4234 	process_cpuid();
4235 
4236 
4237 	if (!quiet)
4238 		for_all_cpus(print_hwp, ODD_COUNTERS);
4239 
4240 	if (!quiet)
4241 		for_all_cpus(print_epb, ODD_COUNTERS);
4242 
4243 	if (!quiet)
4244 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
4245 
4246 	if (!quiet)
4247 		for_all_cpus(print_rapl, ODD_COUNTERS);
4248 
4249 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
4250 
4251 	if (!quiet)
4252 		for_all_cpus(print_thermal, ODD_COUNTERS);
4253 
4254 	if (!quiet && do_irtl_snb)
4255 		print_irtl();
4256 }
4257 
4258 int fork_it(char **argv)
4259 {
4260 	pid_t child_pid;
4261 	int status;
4262 
4263 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
4264 	if (status)
4265 		exit(status);
4266 	/* clear affinity side-effect of get_counters() */
4267 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
4268 	gettimeofday(&tv_even, (struct timezone *)NULL);
4269 
4270 	child_pid = fork();
4271 	if (!child_pid) {
4272 		/* child */
4273 		execvp(argv[0], argv);
4274 	} else {
4275 
4276 		/* parent */
4277 		if (child_pid == -1)
4278 			err(1, "fork");
4279 
4280 		signal(SIGINT, SIG_IGN);
4281 		signal(SIGQUIT, SIG_IGN);
4282 		if (waitpid(child_pid, &status, 0) == -1)
4283 			err(status, "waitpid");
4284 	}
4285 	/*
4286 	 * n.b. fork_it() does not check for errors from for_all_cpus()
4287 	 * because re-starting is problematic when forking
4288 	 */
4289 	for_all_cpus(get_counters, ODD_COUNTERS);
4290 	gettimeofday(&tv_odd, (struct timezone *)NULL);
4291 	timersub(&tv_odd, &tv_even, &tv_delta);
4292 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
4293 		fprintf(outf, "%s: Counter reset detected\n", progname);
4294 	else {
4295 		compute_average(EVEN_COUNTERS);
4296 		format_all_counters(EVEN_COUNTERS);
4297 	}
4298 
4299 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
4300 
4301 	flush_output_stderr();
4302 
4303 	return status;
4304 }
4305 
4306 int get_and_dump_counters(void)
4307 {
4308 	int status;
4309 
4310 	status = for_all_cpus(get_counters, ODD_COUNTERS);
4311 	if (status)
4312 		return status;
4313 
4314 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
4315 	if (status)
4316 		return status;
4317 
4318 	flush_output_stdout();
4319 
4320 	return status;
4321 }
4322 
4323 void print_version() {
4324 	fprintf(outf, "turbostat version 4.17 10 Jan 2017"
4325 		" - Len Brown <lenb@kernel.org>\n");
4326 }
4327 
4328 int add_counter(unsigned int msr_num, char *name, unsigned int width,
4329 	enum counter_scope scope, enum counter_type type,
4330 	enum counter_format format)
4331 {
4332 	struct msr_counter *msrp;
4333 
4334 	msrp = calloc(1, sizeof(struct msr_counter));
4335 	if (msrp == NULL) {
4336 		perror("calloc");
4337 		exit(1);
4338 	}
4339 
4340 	msrp->msr_num = msr_num;
4341 	strncpy(msrp->name, name, NAME_BYTES);
4342 	msrp->width = width;
4343 	msrp->type = type;
4344 	msrp->format = format;
4345 
4346 	switch (scope) {
4347 
4348 	case SCOPE_CPU:
4349 		msrp->next = sys.tp;
4350 		sys.tp = msrp;
4351 		sys.added_thread_counters++;
4352 		if (sys.added_thread_counters > MAX_ADDED_COUNTERS) {
4353 			fprintf(stderr, "exceeded max %d added thread counters\n",
4354 				MAX_ADDED_COUNTERS);
4355 			exit(-1);
4356 		}
4357 		break;
4358 
4359 	case SCOPE_CORE:
4360 		msrp->next = sys.cp;
4361 		sys.cp = msrp;
4362 		sys.added_core_counters++;
4363 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
4364 			fprintf(stderr, "exceeded max %d added core counters\n",
4365 				MAX_ADDED_COUNTERS);
4366 			exit(-1);
4367 		}
4368 		break;
4369 
4370 	case SCOPE_PACKAGE:
4371 		msrp->next = sys.pp;
4372 		sys.pp = msrp;
4373 		sys.added_package_counters++;
4374 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
4375 			fprintf(stderr, "exceeded max %d added package counters\n",
4376 				MAX_ADDED_COUNTERS);
4377 			exit(-1);
4378 		}
4379 		break;
4380 	}
4381 
4382 	return 0;
4383 }
4384 
4385 void parse_add_command(char *add_command)
4386 {
4387 	int msr_num = 0;
4388 	char name_buffer[NAME_BYTES] = "";
4389 	int width = 64;
4390 	int fail = 0;
4391 	enum counter_scope scope = SCOPE_CPU;
4392 	enum counter_type type = COUNTER_CYCLES;
4393 	enum counter_format format = FORMAT_DELTA;
4394 
4395 	while (add_command) {
4396 
4397 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
4398 			goto next;
4399 
4400 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
4401 			goto next;
4402 
4403 		if (sscanf(add_command, "u%d", &width) == 1) {
4404 			if ((width == 32) || (width == 64))
4405 				goto next;
4406 			width = 64;
4407 		}
4408 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
4409 			scope = SCOPE_CPU;
4410 			goto next;
4411 		}
4412 		if (!strncmp(add_command, "core", strlen("core"))) {
4413 			scope = SCOPE_CORE;
4414 			goto next;
4415 		}
4416 		if (!strncmp(add_command, "package", strlen("package"))) {
4417 			scope = SCOPE_PACKAGE;
4418 			goto next;
4419 		}
4420 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
4421 			type = COUNTER_CYCLES;
4422 			goto next;
4423 		}
4424 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
4425 			type = COUNTER_SECONDS;
4426 			goto next;
4427 		}
4428 		if (!strncmp(add_command, "raw", strlen("raw"))) {
4429 			format = FORMAT_RAW;
4430 			goto next;
4431 		}
4432 		if (!strncmp(add_command, "delta", strlen("delta"))) {
4433 			format = FORMAT_DELTA;
4434 			goto next;
4435 		}
4436 		if (!strncmp(add_command, "percent", strlen("percent"))) {
4437 			format = FORMAT_PERCENT;
4438 			goto next;
4439 		}
4440 
4441 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
4442 			char *eos;
4443 
4444 			eos = strchr(name_buffer, ',');
4445 			if (eos)
4446 				*eos = '\0';
4447 			goto next;
4448 		}
4449 
4450 next:
4451 		add_command = strchr(add_command, ',');
4452 		if (add_command)
4453 			add_command++;
4454 
4455 	}
4456 	if (msr_num == 0) {
4457 		fprintf(stderr, "--add: (msrDDD | msr0xXXX) required\n");
4458 		fail++;
4459 	}
4460 
4461 	/* generate default column header */
4462 	if (*name_buffer == '\0') {
4463 		if (format == FORMAT_RAW) {
4464 			if (width == 32)
4465 				sprintf(name_buffer, "msr%d", msr_num);
4466 			else
4467 				sprintf(name_buffer, "MSR%d", msr_num);
4468 		} else if (format == FORMAT_DELTA) {
4469 			if (width == 32)
4470 				sprintf(name_buffer, "cnt%d", msr_num);
4471 			else
4472 				sprintf(name_buffer, "CNT%d", msr_num);
4473 		} else if (format == FORMAT_PERCENT) {
4474 			if (width == 32)
4475 				sprintf(name_buffer, "msr%d%%", msr_num);
4476 			else
4477 				sprintf(name_buffer, "MSR%d%%", msr_num);
4478 		}
4479 	}
4480 
4481 	if (add_counter(msr_num, name_buffer, width, scope, type, format))
4482 		fail++;
4483 
4484 	if (fail) {
4485 		help();
4486 		exit(1);
4487 	}
4488 }
4489 /*
4490  * HIDE_LIST - hide this list of counters, show the rest [default]
4491  * SHOW_LIST - show this list of counters, hide the rest
4492  */
4493 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
4494 
4495 int shown;
4496 /*
4497  * parse_show_hide() - process cmdline to set default counter action
4498  */
4499 void parse_show_hide(char *optarg, enum show_hide_mode new_mode)
4500 {
4501 	/*
4502 	 * --show: show only those specified
4503 	 *  The 1st invocation will clear and replace the enabled mask
4504 	 *  subsequent invocations can add to it.
4505 	 */
4506 	if (new_mode == SHOW_LIST) {
4507 		if (shown == 0)
4508 			bic_enabled = bic_lookup(optarg);
4509 		else
4510 			bic_enabled |= bic_lookup(optarg);
4511 		shown = 1;
4512 
4513 		return;
4514 	}
4515 
4516 	/*
4517 	 * --hide: do not show those specified
4518 	 *  multiple invocations simply clear more bits in enabled mask
4519 	 */
4520 	bic_enabled &= ~bic_lookup(optarg);
4521 }
4522 
4523 void cmdline(int argc, char **argv)
4524 {
4525 	int opt;
4526 	int option_index = 0;
4527 	static struct option long_options[] = {
4528 		{"add",		required_argument,	0, 'a'},
4529 		{"Dump",	no_argument,		0, 'D'},
4530 		{"debug",	no_argument,		0, 'd'},	/* internal, not documented */
4531 		{"interval",	required_argument,	0, 'i'},
4532 		{"help",	no_argument,		0, 'h'},
4533 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
4534 		{"Joules",	no_argument,		0, 'J'},
4535 		{"out",		required_argument,	0, 'o'},
4536 		{"Package",	no_argument,		0, 'p'},
4537 		{"processor",	no_argument,		0, 'p'},
4538 		{"quiet",	no_argument,		0, 'q'},
4539 		{"show",	required_argument,	0, 's'},
4540 		{"Summary",	no_argument,		0, 'S'},
4541 		{"TCC",		required_argument,	0, 'T'},
4542 		{"version",	no_argument,		0, 'v' },
4543 		{0,		0,			0,  0 }
4544 	};
4545 
4546 	progname = argv[0];
4547 
4548 	while ((opt = getopt_long_only(argc, argv, "+C:c:Ddhi:JM:m:o:PpqST:v",
4549 				long_options, &option_index)) != -1) {
4550 		switch (opt) {
4551 		case 'a':
4552 			parse_add_command(optarg);
4553 			break;
4554 		case 'D':
4555 			dump_only++;
4556 			break;
4557 		case 'd':
4558 			debug++;
4559 			break;
4560 		case 'H':
4561 			parse_show_hide(optarg, HIDE_LIST);
4562 			break;
4563 		case 'h':
4564 		default:
4565 			help();
4566 			exit(1);
4567 		case 'i':
4568 			{
4569 				double interval = strtod(optarg, NULL);
4570 
4571 				if (interval < 0.001) {
4572 					fprintf(outf, "interval %f seconds is too small\n",
4573 						interval);
4574 					exit(2);
4575 				}
4576 
4577 				interval_ts.tv_sec = interval;
4578 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
4579 			}
4580 			break;
4581 		case 'J':
4582 			rapl_joules++;
4583 			break;
4584 		case 'o':
4585 			outf = fopen_or_die(optarg, "w");
4586 			break;
4587 		case 'P':
4588 			show_pkg_only++;
4589 			break;
4590 		case 'p':
4591 			show_core_only++;
4592 			break;
4593 		case 'q':
4594 			quiet = 1;
4595 			break;
4596 		case 's':
4597 			parse_show_hide(optarg, SHOW_LIST);
4598 			break;
4599 		case 'S':
4600 			summary_only++;
4601 			break;
4602 		case 'T':
4603 			tcc_activation_temp_override = atoi(optarg);
4604 			break;
4605 		case 'v':
4606 			print_version();
4607 			exit(0);
4608 			break;
4609 		}
4610 	}
4611 }
4612 
4613 int main(int argc, char **argv)
4614 {
4615 	outf = stderr;
4616 
4617 	cmdline(argc, argv);
4618 
4619 	if (!quiet)
4620 		print_version();
4621 
4622 	turbostat_init();
4623 
4624 	/* dump counters and exit */
4625 	if (dump_only)
4626 		return get_and_dump_counters();
4627 
4628 	/*
4629 	 * if any params left, it must be a command to fork
4630 	 */
4631 	if (argc - optind)
4632 		return fork_it(argv + optind);
4633 	else
4634 		turbostat_loop();
4635 
4636 	return 0;
4637 }
4638