xref: /linux/tools/power/x86/turbostat/turbostat.c (revision f86fd32d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * turbostat -- show CPU frequency and C-state residency
4  * on modern Intel and AMD processors.
5  *
6  * Copyright (c) 2013 Intel Corporation.
7  * Len Brown <len.brown@intel.com>
8  */
9 
10 #define _GNU_SOURCE
11 #include MSRHEADER
12 #include INTEL_FAMILY_HEADER
13 #include <stdarg.h>
14 #include <stdio.h>
15 #include <err.h>
16 #include <unistd.h>
17 #include <sys/types.h>
18 #include <sys/wait.h>
19 #include <sys/stat.h>
20 #include <sys/select.h>
21 #include <sys/resource.h>
22 #include <fcntl.h>
23 #include <signal.h>
24 #include <sys/time.h>
25 #include <stdlib.h>
26 #include <getopt.h>
27 #include <dirent.h>
28 #include <string.h>
29 #include <ctype.h>
30 #include <sched.h>
31 #include <time.h>
32 #include <cpuid.h>
33 #include <linux/capability.h>
34 #include <errno.h>
35 #include <math.h>
36 
37 char *proc_stat = "/proc/stat";
38 FILE *outf;
39 int *fd_percpu;
40 struct timeval interval_tv = {5, 0};
41 struct timespec interval_ts = {5, 0};
42 unsigned int num_iterations;
43 unsigned int debug;
44 unsigned int quiet;
45 unsigned int shown;
46 unsigned int sums_need_wide_columns;
47 unsigned int rapl_joules;
48 unsigned int summary_only;
49 unsigned int list_header_only;
50 unsigned int dump_only;
51 unsigned int do_snb_cstates;
52 unsigned int do_knl_cstates;
53 unsigned int do_slm_cstates;
54 unsigned int use_c1_residency_msr;
55 unsigned int has_aperf;
56 unsigned int has_epb;
57 unsigned int do_irtl_snb;
58 unsigned int do_irtl_hsw;
59 unsigned int units = 1000000;	/* MHz etc */
60 unsigned int genuine_intel;
61 unsigned int authentic_amd;
62 unsigned int hygon_genuine;
63 unsigned int max_level, max_extended_level;
64 unsigned int has_invariant_tsc;
65 unsigned int do_nhm_platform_info;
66 unsigned int no_MSR_MISC_PWR_MGMT;
67 unsigned int aperf_mperf_multiplier = 1;
68 double bclk;
69 double base_hz;
70 unsigned int has_base_hz;
71 double tsc_tweak = 1.0;
72 unsigned int show_pkg_only;
73 unsigned int show_core_only;
74 char *output_buffer, *outp;
75 unsigned int do_rapl;
76 unsigned int do_dts;
77 unsigned int do_ptm;
78 unsigned long long  gfx_cur_rc6_ms;
79 unsigned long long cpuidle_cur_cpu_lpi_us;
80 unsigned long long cpuidle_cur_sys_lpi_us;
81 unsigned int gfx_cur_mhz;
82 unsigned int tcc_activation_temp;
83 unsigned int tcc_activation_temp_override;
84 double rapl_power_units, rapl_time_units;
85 double rapl_dram_energy_units, rapl_energy_units;
86 double rapl_joule_counter_range;
87 unsigned int do_core_perf_limit_reasons;
88 unsigned int has_automatic_cstate_conversion;
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 unsigned int first_counter_read = 1;
103 int ignore_stdin;
104 
105 #define RAPL_PKG		(1 << 0)
106 					/* 0x610 MSR_PKG_POWER_LIMIT */
107 					/* 0x611 MSR_PKG_ENERGY_STATUS */
108 #define RAPL_PKG_PERF_STATUS	(1 << 1)
109 					/* 0x613 MSR_PKG_PERF_STATUS */
110 #define RAPL_PKG_POWER_INFO	(1 << 2)
111 					/* 0x614 MSR_PKG_POWER_INFO */
112 
113 #define RAPL_DRAM		(1 << 3)
114 					/* 0x618 MSR_DRAM_POWER_LIMIT */
115 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
116 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
117 					/* 0x61b MSR_DRAM_PERF_STATUS */
118 #define RAPL_DRAM_POWER_INFO	(1 << 5)
119 					/* 0x61c MSR_DRAM_POWER_INFO */
120 
121 #define RAPL_CORES_POWER_LIMIT	(1 << 6)
122 					/* 0x638 MSR_PP0_POWER_LIMIT */
123 #define RAPL_CORE_POLICY	(1 << 7)
124 					/* 0x63a MSR_PP0_POLICY */
125 
126 #define RAPL_GFX		(1 << 8)
127 					/* 0x640 MSR_PP1_POWER_LIMIT */
128 					/* 0x641 MSR_PP1_ENERGY_STATUS */
129 					/* 0x642 MSR_PP1_POLICY */
130 
131 #define RAPL_CORES_ENERGY_STATUS	(1 << 9)
132 					/* 0x639 MSR_PP0_ENERGY_STATUS */
133 #define RAPL_PER_CORE_ENERGY	(1 << 10)
134 					/* Indicates cores energy collection is per-core,
135 					 * not per-package. */
136 #define RAPL_AMD_F17H		(1 << 11)
137 					/* 0xc0010299 MSR_RAPL_PWR_UNIT */
138 					/* 0xc001029a MSR_CORE_ENERGY_STAT */
139 					/* 0xc001029b MSR_PKG_ENERGY_STAT */
140 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
141 #define	TJMAX_DEFAULT	100
142 
143 /* MSRs that are not yet in the kernel-provided header. */
144 #define MSR_RAPL_PWR_UNIT	0xc0010299
145 #define MSR_CORE_ENERGY_STAT	0xc001029a
146 #define MSR_PKG_ENERGY_STAT	0xc001029b
147 
148 #define MAX(a, b) ((a) > (b) ? (a) : (b))
149 
150 /*
151  * buffer size used by sscanf() for added column names
152  * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
153  */
154 #define	NAME_BYTES 20
155 #define PATH_BYTES 128
156 
157 int backwards_count;
158 char *progname;
159 
160 #define CPU_SUBSET_MAXCPUS	1024	/* need to use before probe... */
161 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset;
162 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size;
163 #define MAX_ADDED_COUNTERS 8
164 #define MAX_ADDED_THREAD_COUNTERS 24
165 #define BITMASK_SIZE 32
166 
167 struct thread_data {
168 	struct timeval tv_begin;
169 	struct timeval tv_end;
170 	struct timeval tv_delta;
171 	unsigned long long tsc;
172 	unsigned long long aperf;
173 	unsigned long long mperf;
174 	unsigned long long c1;
175 	unsigned long long  irq_count;
176 	unsigned int smi_count;
177 	unsigned int cpu_id;
178 	unsigned int apic_id;
179 	unsigned int x2apic_id;
180 	unsigned int flags;
181 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
182 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
183 	unsigned long long counter[MAX_ADDED_THREAD_COUNTERS];
184 } *thread_even, *thread_odd;
185 
186 struct core_data {
187 	unsigned long long c3;
188 	unsigned long long c6;
189 	unsigned long long c7;
190 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
191 	unsigned int core_temp_c;
192 	unsigned int core_energy;	/* MSR_CORE_ENERGY_STAT */
193 	unsigned int core_id;
194 	unsigned long long counter[MAX_ADDED_COUNTERS];
195 } *core_even, *core_odd;
196 
197 struct pkg_data {
198 	unsigned long long pc2;
199 	unsigned long long pc3;
200 	unsigned long long pc6;
201 	unsigned long long pc7;
202 	unsigned long long pc8;
203 	unsigned long long pc9;
204 	unsigned long long pc10;
205 	unsigned long long cpu_lpi;
206 	unsigned long long sys_lpi;
207 	unsigned long long pkg_wtd_core_c0;
208 	unsigned long long pkg_any_core_c0;
209 	unsigned long long pkg_any_gfxe_c0;
210 	unsigned long long pkg_both_core_gfxe_c0;
211 	long long gfx_rc6_ms;
212 	unsigned int gfx_mhz;
213 	unsigned int package_id;
214 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
215 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
216 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
217 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
218 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
219 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
220 	unsigned int pkg_temp_c;
221 	unsigned long long counter[MAX_ADDED_COUNTERS];
222 } *package_even, *package_odd;
223 
224 #define ODD_COUNTERS thread_odd, core_odd, package_odd
225 #define EVEN_COUNTERS thread_even, core_even, package_even
226 
227 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no)	      \
228 	((thread_base) +						      \
229 	 ((pkg_no) *							      \
230 	  topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \
231 	 ((node_no) * topo.cores_per_node * topo.threads_per_core) +	      \
232 	 ((core_no) * topo.threads_per_core) +				      \
233 	 (thread_no))
234 
235 #define GET_CORE(core_base, core_no, node_no, pkg_no)			\
236 	((core_base) +							\
237 	 ((pkg_no) *  topo.nodes_per_pkg * topo.cores_per_node) +	\
238 	 ((node_no) * topo.cores_per_node) +				\
239 	 (core_no))
240 
241 
242 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
243 
244 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
245 enum counter_type {COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC};
246 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
247 
248 struct msr_counter {
249 	unsigned int msr_num;
250 	char name[NAME_BYTES];
251 	char path[PATH_BYTES];
252 	unsigned int width;
253 	enum counter_type type;
254 	enum counter_format format;
255 	struct msr_counter *next;
256 	unsigned int flags;
257 #define	FLAGS_HIDE	(1 << 0)
258 #define	FLAGS_SHOW	(1 << 1)
259 #define	SYSFS_PERCPU	(1 << 1)
260 };
261 
262 struct sys_counters {
263 	unsigned int added_thread_counters;
264 	unsigned int added_core_counters;
265 	unsigned int added_package_counters;
266 	struct msr_counter *tp;
267 	struct msr_counter *cp;
268 	struct msr_counter *pp;
269 } sys;
270 
271 struct system_summary {
272 	struct thread_data threads;
273 	struct core_data cores;
274 	struct pkg_data packages;
275 } average;
276 
277 struct cpu_topology {
278 	int physical_package_id;
279 	int die_id;
280 	int logical_cpu_id;
281 	int physical_node_id;
282 	int logical_node_id;	/* 0-based count within the package */
283 	int physical_core_id;
284 	int thread_id;
285 	cpu_set_t *put_ids; /* Processing Unit/Thread IDs */
286 } *cpus;
287 
288 struct topo_params {
289 	int num_packages;
290 	int num_die;
291 	int num_cpus;
292 	int num_cores;
293 	int max_cpu_num;
294 	int max_node_num;
295 	int nodes_per_pkg;
296 	int cores_per_node;
297 	int threads_per_core;
298 } topo;
299 
300 struct timeval tv_even, tv_odd, tv_delta;
301 
302 int *irq_column_2_cpu;	/* /proc/interrupts column numbers */
303 int *irqs_per_cpu;		/* indexed by cpu_num */
304 
305 void setup_all_buffers(void);
306 
307 int cpu_is_not_present(int cpu)
308 {
309 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
310 }
311 /*
312  * run func(thread, core, package) in topology order
313  * skip non-present cpus
314  */
315 
316 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
317 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
318 {
319 	int retval, pkg_no, core_no, thread_no, node_no;
320 
321 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
322 		for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) {
323 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
324 				for (thread_no = 0; thread_no <
325 					topo.threads_per_core; ++thread_no) {
326 					struct thread_data *t;
327 					struct core_data *c;
328 					struct pkg_data *p;
329 
330 					t = GET_THREAD(thread_base, thread_no,
331 						       core_no, node_no,
332 						       pkg_no);
333 
334 					if (cpu_is_not_present(t->cpu_id))
335 						continue;
336 
337 					c = GET_CORE(core_base, core_no,
338 						     node_no, pkg_no);
339 					p = GET_PKG(pkg_base, pkg_no);
340 
341 					retval = func(t, c, p);
342 					if (retval)
343 						return retval;
344 				}
345 			}
346 		}
347 	}
348 	return 0;
349 }
350 
351 int cpu_migrate(int cpu)
352 {
353 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
354 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
355 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
356 		return -1;
357 	else
358 		return 0;
359 }
360 int get_msr_fd(int cpu)
361 {
362 	char pathname[32];
363 	int fd;
364 
365 	fd = fd_percpu[cpu];
366 
367 	if (fd)
368 		return fd;
369 
370 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
371 	fd = open(pathname, O_RDONLY);
372 	if (fd < 0)
373 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
374 
375 	fd_percpu[cpu] = fd;
376 
377 	return fd;
378 }
379 
380 int get_msr(int cpu, off_t offset, unsigned long long *msr)
381 {
382 	ssize_t retval;
383 
384 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
385 
386 	if (retval != sizeof *msr)
387 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
388 
389 	return 0;
390 }
391 
392 /*
393  * This list matches the column headers, except
394  * 1. built-in only, the sysfs counters are not here -- we learn of those at run-time
395  * 2. Core and CPU are moved to the end, we can't have strings that contain them
396  *    matching on them for --show and --hide.
397  */
398 struct msr_counter bic[] = {
399 	{ 0x0, "usec" },
400 	{ 0x0, "Time_Of_Day_Seconds" },
401 	{ 0x0, "Package" },
402 	{ 0x0, "Node" },
403 	{ 0x0, "Avg_MHz" },
404 	{ 0x0, "Busy%" },
405 	{ 0x0, "Bzy_MHz" },
406 	{ 0x0, "TSC_MHz" },
407 	{ 0x0, "IRQ" },
408 	{ 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL},
409 	{ 0x0, "sysfs" },
410 	{ 0x0, "CPU%c1" },
411 	{ 0x0, "CPU%c3" },
412 	{ 0x0, "CPU%c6" },
413 	{ 0x0, "CPU%c7" },
414 	{ 0x0, "ThreadC" },
415 	{ 0x0, "CoreTmp" },
416 	{ 0x0, "CoreCnt" },
417 	{ 0x0, "PkgTmp" },
418 	{ 0x0, "GFX%rc6" },
419 	{ 0x0, "GFXMHz" },
420 	{ 0x0, "Pkg%pc2" },
421 	{ 0x0, "Pkg%pc3" },
422 	{ 0x0, "Pkg%pc6" },
423 	{ 0x0, "Pkg%pc7" },
424 	{ 0x0, "Pkg%pc8" },
425 	{ 0x0, "Pkg%pc9" },
426 	{ 0x0, "Pk%pc10" },
427 	{ 0x0, "CPU%LPI" },
428 	{ 0x0, "SYS%LPI" },
429 	{ 0x0, "PkgWatt" },
430 	{ 0x0, "CorWatt" },
431 	{ 0x0, "GFXWatt" },
432 	{ 0x0, "PkgCnt" },
433 	{ 0x0, "RAMWatt" },
434 	{ 0x0, "PKG_%" },
435 	{ 0x0, "RAM_%" },
436 	{ 0x0, "Pkg_J" },
437 	{ 0x0, "Cor_J" },
438 	{ 0x0, "GFX_J" },
439 	{ 0x0, "RAM_J" },
440 	{ 0x0, "Mod%c6" },
441 	{ 0x0, "Totl%C0" },
442 	{ 0x0, "Any%C0" },
443 	{ 0x0, "GFX%C0" },
444 	{ 0x0, "CPUGFX%" },
445 	{ 0x0, "Core" },
446 	{ 0x0, "CPU" },
447 	{ 0x0, "APIC" },
448 	{ 0x0, "X2APIC" },
449 	{ 0x0, "Die" },
450 };
451 
452 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
453 #define	BIC_USEC	(1ULL << 0)
454 #define	BIC_TOD		(1ULL << 1)
455 #define	BIC_Package	(1ULL << 2)
456 #define	BIC_Node	(1ULL << 3)
457 #define	BIC_Avg_MHz	(1ULL << 4)
458 #define	BIC_Busy	(1ULL << 5)
459 #define	BIC_Bzy_MHz	(1ULL << 6)
460 #define	BIC_TSC_MHz	(1ULL << 7)
461 #define	BIC_IRQ		(1ULL << 8)
462 #define	BIC_SMI		(1ULL << 9)
463 #define	BIC_sysfs	(1ULL << 10)
464 #define	BIC_CPU_c1	(1ULL << 11)
465 #define	BIC_CPU_c3	(1ULL << 12)
466 #define	BIC_CPU_c6	(1ULL << 13)
467 #define	BIC_CPU_c7	(1ULL << 14)
468 #define	BIC_ThreadC	(1ULL << 15)
469 #define	BIC_CoreTmp	(1ULL << 16)
470 #define	BIC_CoreCnt	(1ULL << 17)
471 #define	BIC_PkgTmp	(1ULL << 18)
472 #define	BIC_GFX_rc6	(1ULL << 19)
473 #define	BIC_GFXMHz	(1ULL << 20)
474 #define	BIC_Pkgpc2	(1ULL << 21)
475 #define	BIC_Pkgpc3	(1ULL << 22)
476 #define	BIC_Pkgpc6	(1ULL << 23)
477 #define	BIC_Pkgpc7	(1ULL << 24)
478 #define	BIC_Pkgpc8	(1ULL << 25)
479 #define	BIC_Pkgpc9	(1ULL << 26)
480 #define	BIC_Pkgpc10	(1ULL << 27)
481 #define BIC_CPU_LPI	(1ULL << 28)
482 #define BIC_SYS_LPI	(1ULL << 29)
483 #define	BIC_PkgWatt	(1ULL << 30)
484 #define	BIC_CorWatt	(1ULL << 31)
485 #define	BIC_GFXWatt	(1ULL << 32)
486 #define	BIC_PkgCnt	(1ULL << 33)
487 #define	BIC_RAMWatt	(1ULL << 34)
488 #define	BIC_PKG__	(1ULL << 35)
489 #define	BIC_RAM__	(1ULL << 36)
490 #define	BIC_Pkg_J	(1ULL << 37)
491 #define	BIC_Cor_J	(1ULL << 38)
492 #define	BIC_GFX_J	(1ULL << 39)
493 #define	BIC_RAM_J	(1ULL << 40)
494 #define	BIC_Mod_c6	(1ULL << 41)
495 #define	BIC_Totl_c0	(1ULL << 42)
496 #define	BIC_Any_c0	(1ULL << 43)
497 #define	BIC_GFX_c0	(1ULL << 44)
498 #define	BIC_CPUGFX	(1ULL << 45)
499 #define	BIC_Core	(1ULL << 46)
500 #define	BIC_CPU		(1ULL << 47)
501 #define	BIC_APIC	(1ULL << 48)
502 #define	BIC_X2APIC	(1ULL << 49)
503 #define	BIC_Die		(1ULL << 50)
504 
505 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
506 
507 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT);
508 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC;
509 
510 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
511 #define DO_BIC_READ(COUNTER_NAME) (bic_present & COUNTER_NAME)
512 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
513 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
514 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
515 
516 
517 #define MAX_DEFERRED 16
518 char *deferred_skip_names[MAX_DEFERRED];
519 int deferred_skip_index;
520 
521 /*
522  * HIDE_LIST - hide this list of counters, show the rest [default]
523  * SHOW_LIST - show this list of counters, hide the rest
524  */
525 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
526 
527 void help(void)
528 {
529 	fprintf(outf,
530 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
531 	"\n"
532 	"Turbostat forks the specified COMMAND and prints statistics\n"
533 	"when COMMAND completes.\n"
534 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
535 	"to print statistics, until interrupted.\n"
536 	"  -a, --add	add a counter\n"
537 	"		  eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
538 	"  -c, --cpu	cpu-set	limit output to summary plus cpu-set:\n"
539 	"		  {core | package | j,k,l..m,n-p }\n"
540 	"  -d, --debug	displays usec, Time_Of_Day_Seconds and more debugging\n"
541 	"  -D, --Dump	displays the raw counter values\n"
542 	"  -e, --enable	[all | column]\n"
543 	"		shows all or the specified disabled column\n"
544 	"  -H, --hide [column|column,column,...]\n"
545 	"		hide the specified column(s)\n"
546 	"  -i, --interval sec.subsec\n"
547 	"		Override default 5-second measurement interval\n"
548 	"  -J, --Joules	displays energy in Joules instead of Watts\n"
549 	"  -l, --list	list column headers only\n"
550 	"  -n, --num_iterations num\n"
551 	"		number of the measurement iterations\n"
552 	"  -o, --out file\n"
553 	"		create or truncate \"file\" for all output\n"
554 	"  -q, --quiet	skip decoding system configuration header\n"
555 	"  -s, --show [column|column,column,...]\n"
556 	"		show only the specified column(s)\n"
557 	"  -S, --Summary\n"
558 	"		limits output to 1-line system summary per interval\n"
559 	"  -T, --TCC temperature\n"
560 	"		sets the Thermal Control Circuit temperature in\n"
561 	"		  degrees Celsius\n"
562 	"  -h, --help	print this help message\n"
563 	"  -v, --version	print version information\n"
564 	"\n"
565 	"For more help, run \"man turbostat\"\n");
566 }
567 
568 /*
569  * bic_lookup
570  * for all the strings in comma separate name_list,
571  * set the approprate bit in return value.
572  */
573 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
574 {
575 	int i;
576 	unsigned long long retval = 0;
577 
578 	while (name_list) {
579 		char *comma;
580 
581 		comma = strchr(name_list, ',');
582 
583 		if (comma)
584 			*comma = '\0';
585 
586 		if (!strcmp(name_list, "all"))
587 			return ~0;
588 
589 		for (i = 0; i < MAX_BIC; ++i) {
590 			if (!strcmp(name_list, bic[i].name)) {
591 				retval |= (1ULL << i);
592 				break;
593 			}
594 		}
595 		if (i == MAX_BIC) {
596 			if (mode == SHOW_LIST) {
597 				fprintf(stderr, "Invalid counter name: %s\n", name_list);
598 				exit(-1);
599 			}
600 			deferred_skip_names[deferred_skip_index++] = name_list;
601 			if (debug)
602 				fprintf(stderr, "deferred \"%s\"\n", name_list);
603 			if (deferred_skip_index >= MAX_DEFERRED) {
604 				fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
605 					MAX_DEFERRED, name_list);
606 				help();
607 				exit(1);
608 			}
609 		}
610 
611 		name_list = comma;
612 		if (name_list)
613 			name_list++;
614 
615 	}
616 	return retval;
617 }
618 
619 
620 void print_header(char *delim)
621 {
622 	struct msr_counter *mp;
623 	int printed = 0;
624 
625 	if (DO_BIC(BIC_USEC))
626 		outp += sprintf(outp, "%susec", (printed++ ? delim : ""));
627 	if (DO_BIC(BIC_TOD))
628 		outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : ""));
629 	if (DO_BIC(BIC_Package))
630 		outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
631 	if (DO_BIC(BIC_Die))
632 		outp += sprintf(outp, "%sDie", (printed++ ? delim : ""));
633 	if (DO_BIC(BIC_Node))
634 		outp += sprintf(outp, "%sNode", (printed++ ? delim : ""));
635 	if (DO_BIC(BIC_Core))
636 		outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
637 	if (DO_BIC(BIC_CPU))
638 		outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
639 	if (DO_BIC(BIC_APIC))
640 		outp += sprintf(outp, "%sAPIC", (printed++ ? delim : ""));
641 	if (DO_BIC(BIC_X2APIC))
642 		outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : ""));
643 	if (DO_BIC(BIC_Avg_MHz))
644 		outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
645 	if (DO_BIC(BIC_Busy))
646 		outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
647 	if (DO_BIC(BIC_Bzy_MHz))
648 		outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
649 	if (DO_BIC(BIC_TSC_MHz))
650 		outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
651 
652 	if (DO_BIC(BIC_IRQ)) {
653 		if (sums_need_wide_columns)
654 			outp += sprintf(outp, "%s     IRQ", (printed++ ? delim : ""));
655 		else
656 			outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
657 	}
658 
659 	if (DO_BIC(BIC_SMI))
660 		outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
661 
662 	for (mp = sys.tp; mp; mp = mp->next) {
663 
664 		if (mp->format == FORMAT_RAW) {
665 			if (mp->width == 64)
666 				outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
667 			else
668 				outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
669 		} else {
670 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
671 				outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
672 			else
673 				outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
674 		}
675 	}
676 
677 	if (DO_BIC(BIC_CPU_c1))
678 		outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
679 	if (DO_BIC(BIC_CPU_c3))
680 		outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
681 	if (DO_BIC(BIC_CPU_c6))
682 		outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
683 	if (DO_BIC(BIC_CPU_c7))
684 		outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
685 
686 	if (DO_BIC(BIC_Mod_c6))
687 		outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
688 
689 	if (DO_BIC(BIC_CoreTmp))
690 		outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
691 
692 	if (do_rapl && !rapl_joules) {
693 		if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
694 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
695 	} else if (do_rapl && rapl_joules) {
696 		if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
697 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
698 	}
699 
700 	for (mp = sys.cp; mp; mp = mp->next) {
701 		if (mp->format == FORMAT_RAW) {
702 			if (mp->width == 64)
703 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
704 			else
705 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
706 		} else {
707 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
708 				outp += sprintf(outp, "%s%8s", delim, mp->name);
709 			else
710 				outp += sprintf(outp, "%s%s", delim, mp->name);
711 		}
712 	}
713 
714 	if (DO_BIC(BIC_PkgTmp))
715 		outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
716 
717 	if (DO_BIC(BIC_GFX_rc6))
718 		outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
719 
720 	if (DO_BIC(BIC_GFXMHz))
721 		outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
722 
723 	if (DO_BIC(BIC_Totl_c0))
724 		outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
725 	if (DO_BIC(BIC_Any_c0))
726 		outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
727 	if (DO_BIC(BIC_GFX_c0))
728 		outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
729 	if (DO_BIC(BIC_CPUGFX))
730 		outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
731 
732 	if (DO_BIC(BIC_Pkgpc2))
733 		outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
734 	if (DO_BIC(BIC_Pkgpc3))
735 		outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
736 	if (DO_BIC(BIC_Pkgpc6))
737 		outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
738 	if (DO_BIC(BIC_Pkgpc7))
739 		outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
740 	if (DO_BIC(BIC_Pkgpc8))
741 		outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
742 	if (DO_BIC(BIC_Pkgpc9))
743 		outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
744 	if (DO_BIC(BIC_Pkgpc10))
745 		outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
746 	if (DO_BIC(BIC_CPU_LPI))
747 		outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : ""));
748 	if (DO_BIC(BIC_SYS_LPI))
749 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
750 
751 	if (do_rapl && !rapl_joules) {
752 		if (DO_BIC(BIC_PkgWatt))
753 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
754 		if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
755 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
756 		if (DO_BIC(BIC_GFXWatt))
757 			outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
758 		if (DO_BIC(BIC_RAMWatt))
759 			outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
760 		if (DO_BIC(BIC_PKG__))
761 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
762 		if (DO_BIC(BIC_RAM__))
763 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
764 	} else if (do_rapl && rapl_joules) {
765 		if (DO_BIC(BIC_Pkg_J))
766 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
767 		if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
768 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
769 		if (DO_BIC(BIC_GFX_J))
770 			outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
771 		if (DO_BIC(BIC_RAM_J))
772 			outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
773 		if (DO_BIC(BIC_PKG__))
774 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
775 		if (DO_BIC(BIC_RAM__))
776 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
777 	}
778 	for (mp = sys.pp; mp; mp = mp->next) {
779 		if (mp->format == FORMAT_RAW) {
780 			if (mp->width == 64)
781 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
782 			else
783 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
784 		} else {
785 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
786 				outp += sprintf(outp, "%s%8s", delim, mp->name);
787 			else
788 				outp += sprintf(outp, "%s%s", delim, mp->name);
789 		}
790 	}
791 
792 	outp += sprintf(outp, "\n");
793 }
794 
795 int dump_counters(struct thread_data *t, struct core_data *c,
796 	struct pkg_data *p)
797 {
798 	int i;
799 	struct msr_counter *mp;
800 
801 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
802 
803 	if (t) {
804 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
805 			t->cpu_id, t->flags);
806 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
807 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
808 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
809 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
810 
811 		if (DO_BIC(BIC_IRQ))
812 			outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
813 		if (DO_BIC(BIC_SMI))
814 			outp += sprintf(outp, "SMI: %d\n", t->smi_count);
815 
816 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
817 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
818 				i, mp->msr_num, t->counter[i]);
819 		}
820 	}
821 
822 	if (c) {
823 		outp += sprintf(outp, "core: %d\n", c->core_id);
824 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
825 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
826 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
827 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
828 		outp += sprintf(outp, "Joules: %0X\n", c->core_energy);
829 
830 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
831 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
832 				i, mp->msr_num, c->counter[i]);
833 		}
834 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
835 	}
836 
837 	if (p) {
838 		outp += sprintf(outp, "package: %d\n", p->package_id);
839 
840 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
841 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
842 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
843 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
844 
845 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
846 		if (DO_BIC(BIC_Pkgpc3))
847 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
848 		if (DO_BIC(BIC_Pkgpc6))
849 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
850 		if (DO_BIC(BIC_Pkgpc7))
851 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
852 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
853 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
854 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
855 		outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi);
856 		outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi);
857 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
858 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
859 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
860 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
861 		outp += sprintf(outp, "Throttle PKG: %0X\n",
862 			p->rapl_pkg_perf_status);
863 		outp += sprintf(outp, "Throttle RAM: %0X\n",
864 			p->rapl_dram_perf_status);
865 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
866 
867 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
868 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
869 				i, mp->msr_num, p->counter[i]);
870 		}
871 	}
872 
873 	outp += sprintf(outp, "\n");
874 
875 	return 0;
876 }
877 
878 /*
879  * column formatting convention & formats
880  */
881 int format_counters(struct thread_data *t, struct core_data *c,
882 	struct pkg_data *p)
883 {
884 	double interval_float, tsc;
885 	char *fmt8;
886 	int i;
887 	struct msr_counter *mp;
888 	char *delim = "\t";
889 	int printed = 0;
890 
891 	 /* if showing only 1st thread in core and this isn't one, bail out */
892 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
893 		return 0;
894 
895 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
896 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
897 		return 0;
898 
899 	/*if not summary line and --cpu is used */
900 	if ((t != &average.threads) &&
901 		(cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
902 		return 0;
903 
904 	if (DO_BIC(BIC_USEC)) {
905 		/* on each row, print how many usec each timestamp took to gather */
906 		struct timeval tv;
907 
908 		timersub(&t->tv_end, &t->tv_begin, &tv);
909 		outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec);
910 	}
911 
912 	/* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */
913 	if (DO_BIC(BIC_TOD))
914 		outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec);
915 
916 	interval_float = t->tv_delta.tv_sec + t->tv_delta.tv_usec/1000000.0;
917 
918 	tsc = t->tsc * tsc_tweak;
919 
920 	/* topo columns, print blanks on 1st (average) line */
921 	if (t == &average.threads) {
922 		if (DO_BIC(BIC_Package))
923 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
924 		if (DO_BIC(BIC_Die))
925 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
926 		if (DO_BIC(BIC_Node))
927 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
928 		if (DO_BIC(BIC_Core))
929 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
930 		if (DO_BIC(BIC_CPU))
931 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
932 		if (DO_BIC(BIC_APIC))
933 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
934 		if (DO_BIC(BIC_X2APIC))
935 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
936 	} else {
937 		if (DO_BIC(BIC_Package)) {
938 			if (p)
939 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
940 			else
941 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
942 		}
943 		if (DO_BIC(BIC_Die)) {
944 			if (c)
945 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id);
946 			else
947 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
948 		}
949 		if (DO_BIC(BIC_Node)) {
950 			if (t)
951 				outp += sprintf(outp, "%s%d",
952 						(printed++ ? delim : ""),
953 					      cpus[t->cpu_id].physical_node_id);
954 			else
955 				outp += sprintf(outp, "%s-",
956 						(printed++ ? delim : ""));
957 		}
958 		if (DO_BIC(BIC_Core)) {
959 			if (c)
960 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
961 			else
962 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
963 		}
964 		if (DO_BIC(BIC_CPU))
965 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
966 		if (DO_BIC(BIC_APIC))
967 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id);
968 		if (DO_BIC(BIC_X2APIC))
969 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id);
970 	}
971 
972 	if (DO_BIC(BIC_Avg_MHz))
973 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
974 			1.0 / units * t->aperf / interval_float);
975 
976 	if (DO_BIC(BIC_Busy))
977 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf/tsc);
978 
979 	if (DO_BIC(BIC_Bzy_MHz)) {
980 		if (has_base_hz)
981 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
982 		else
983 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
984 				tsc / units * t->aperf / t->mperf / interval_float);
985 	}
986 
987 	if (DO_BIC(BIC_TSC_MHz))
988 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
989 
990 	/* IRQ */
991 	if (DO_BIC(BIC_IRQ)) {
992 		if (sums_need_wide_columns)
993 			outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
994 		else
995 			outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
996 	}
997 
998 	/* SMI */
999 	if (DO_BIC(BIC_SMI))
1000 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
1001 
1002 	/* Added counters */
1003 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1004 		if (mp->format == FORMAT_RAW) {
1005 			if (mp->width == 32)
1006 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) t->counter[i]);
1007 			else
1008 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
1009 		} else if (mp->format == FORMAT_DELTA) {
1010 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1011 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
1012 			else
1013 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
1014 		} else if (mp->format == FORMAT_PERCENT) {
1015 			if (mp->type == COUNTER_USEC)
1016 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), t->counter[i]/interval_float/10000);
1017 			else
1018 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i]/tsc);
1019 		}
1020 	}
1021 
1022 	/* C1 */
1023 	if (DO_BIC(BIC_CPU_c1))
1024 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1/tsc);
1025 
1026 
1027 	/* print per-core data only for 1st thread in core */
1028 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1029 		goto done;
1030 
1031 	if (DO_BIC(BIC_CPU_c3))
1032 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3/tsc);
1033 	if (DO_BIC(BIC_CPU_c6))
1034 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6/tsc);
1035 	if (DO_BIC(BIC_CPU_c7))
1036 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7/tsc);
1037 
1038 	/* Mod%c6 */
1039 	if (DO_BIC(BIC_Mod_c6))
1040 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
1041 
1042 	if (DO_BIC(BIC_CoreTmp))
1043 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
1044 
1045 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1046 		if (mp->format == FORMAT_RAW) {
1047 			if (mp->width == 32)
1048 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) c->counter[i]);
1049 			else
1050 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
1051 		} else if (mp->format == FORMAT_DELTA) {
1052 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1053 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
1054 			else
1055 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
1056 		} else if (mp->format == FORMAT_PERCENT) {
1057 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i]/tsc);
1058 		}
1059 	}
1060 
1061 	/*
1062 	 * If measurement interval exceeds minimum RAPL Joule Counter range,
1063 	 * indicate that results are suspect by printing "**" in fraction place.
1064 	 */
1065 	if (interval_float < rapl_joule_counter_range)
1066 		fmt8 = "%s%.2f";
1067 	else
1068 		fmt8 = "%6.0f**";
1069 
1070 	if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
1071 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float);
1072 	if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
1073 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units);
1074 
1075 	/* print per-package data only for 1st core in package */
1076 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1077 		goto done;
1078 
1079 	/* PkgTmp */
1080 	if (DO_BIC(BIC_PkgTmp))
1081 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
1082 
1083 	/* GFXrc6 */
1084 	if (DO_BIC(BIC_GFX_rc6)) {
1085 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
1086 			outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
1087 		} else {
1088 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1089 				p->gfx_rc6_ms / 10.0 / interval_float);
1090 		}
1091 	}
1092 
1093 	/* GFXMHz */
1094 	if (DO_BIC(BIC_GFXMHz))
1095 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
1096 
1097 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
1098 	if (DO_BIC(BIC_Totl_c0))
1099 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0/tsc);
1100 	if (DO_BIC(BIC_Any_c0))
1101 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0/tsc);
1102 	if (DO_BIC(BIC_GFX_c0))
1103 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0/tsc);
1104 	if (DO_BIC(BIC_CPUGFX))
1105 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0/tsc);
1106 
1107 	if (DO_BIC(BIC_Pkgpc2))
1108 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2/tsc);
1109 	if (DO_BIC(BIC_Pkgpc3))
1110 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3/tsc);
1111 	if (DO_BIC(BIC_Pkgpc6))
1112 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6/tsc);
1113 	if (DO_BIC(BIC_Pkgpc7))
1114 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7/tsc);
1115 	if (DO_BIC(BIC_Pkgpc8))
1116 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8/tsc);
1117 	if (DO_BIC(BIC_Pkgpc9))
1118 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9/tsc);
1119 	if (DO_BIC(BIC_Pkgpc10))
1120 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10/tsc);
1121 
1122 	if (DO_BIC(BIC_CPU_LPI))
1123 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float);
1124 	if (DO_BIC(BIC_SYS_LPI))
1125 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float);
1126 
1127 	if (DO_BIC(BIC_PkgWatt))
1128 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
1129 	if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1130 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
1131 	if (DO_BIC(BIC_GFXWatt))
1132 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
1133 	if (DO_BIC(BIC_RAMWatt))
1134 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units / interval_float);
1135 	if (DO_BIC(BIC_Pkg_J))
1136 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
1137 	if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1138 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
1139 	if (DO_BIC(BIC_GFX_J))
1140 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
1141 	if (DO_BIC(BIC_RAM_J))
1142 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
1143 	if (DO_BIC(BIC_PKG__))
1144 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
1145 	if (DO_BIC(BIC_RAM__))
1146 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
1147 
1148 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1149 		if (mp->format == FORMAT_RAW) {
1150 			if (mp->width == 32)
1151 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) p->counter[i]);
1152 			else
1153 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
1154 		} else if (mp->format == FORMAT_DELTA) {
1155 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1156 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
1157 			else
1158 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
1159 		} else if (mp->format == FORMAT_PERCENT) {
1160 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i]/tsc);
1161 		}
1162 	}
1163 
1164 done:
1165 	if (*(outp - 1) != '\n')
1166 		outp += sprintf(outp, "\n");
1167 
1168 	return 0;
1169 }
1170 
1171 void flush_output_stdout(void)
1172 {
1173 	FILE *filep;
1174 
1175 	if (outf == stderr)
1176 		filep = stdout;
1177 	else
1178 		filep = outf;
1179 
1180 	fputs(output_buffer, filep);
1181 	fflush(filep);
1182 
1183 	outp = output_buffer;
1184 }
1185 void flush_output_stderr(void)
1186 {
1187 	fputs(output_buffer, outf);
1188 	fflush(outf);
1189 	outp = output_buffer;
1190 }
1191 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1192 {
1193 	static int printed;
1194 
1195 	if (!printed || !summary_only)
1196 		print_header("\t");
1197 
1198 	format_counters(&average.threads, &average.cores, &average.packages);
1199 
1200 	printed = 1;
1201 
1202 	if (summary_only)
1203 		return;
1204 
1205 	for_all_cpus(format_counters, t, c, p);
1206 }
1207 
1208 #define DELTA_WRAP32(new, old)			\
1209 	if (new > old) {			\
1210 		old = new - old;		\
1211 	} else {				\
1212 		old = 0x100000000 + new - old;	\
1213 	}
1214 
1215 int
1216 delta_package(struct pkg_data *new, struct pkg_data *old)
1217 {
1218 	int i;
1219 	struct msr_counter *mp;
1220 
1221 
1222 	if (DO_BIC(BIC_Totl_c0))
1223 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1224 	if (DO_BIC(BIC_Any_c0))
1225 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1226 	if (DO_BIC(BIC_GFX_c0))
1227 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1228 	if (DO_BIC(BIC_CPUGFX))
1229 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1230 
1231 	old->pc2 = new->pc2 - old->pc2;
1232 	if (DO_BIC(BIC_Pkgpc3))
1233 		old->pc3 = new->pc3 - old->pc3;
1234 	if (DO_BIC(BIC_Pkgpc6))
1235 		old->pc6 = new->pc6 - old->pc6;
1236 	if (DO_BIC(BIC_Pkgpc7))
1237 		old->pc7 = new->pc7 - old->pc7;
1238 	old->pc8 = new->pc8 - old->pc8;
1239 	old->pc9 = new->pc9 - old->pc9;
1240 	old->pc10 = new->pc10 - old->pc10;
1241 	old->cpu_lpi = new->cpu_lpi - old->cpu_lpi;
1242 	old->sys_lpi = new->sys_lpi - old->sys_lpi;
1243 	old->pkg_temp_c = new->pkg_temp_c;
1244 
1245 	/* flag an error when rc6 counter resets/wraps */
1246 	if (old->gfx_rc6_ms >  new->gfx_rc6_ms)
1247 		old->gfx_rc6_ms = -1;
1248 	else
1249 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1250 
1251 	old->gfx_mhz = new->gfx_mhz;
1252 
1253 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
1254 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
1255 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
1256 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
1257 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
1258 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
1259 
1260 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1261 		if (mp->format == FORMAT_RAW)
1262 			old->counter[i] = new->counter[i];
1263 		else
1264 			old->counter[i] = new->counter[i] - old->counter[i];
1265 	}
1266 
1267 	return 0;
1268 }
1269 
1270 void
1271 delta_core(struct core_data *new, struct core_data *old)
1272 {
1273 	int i;
1274 	struct msr_counter *mp;
1275 
1276 	old->c3 = new->c3 - old->c3;
1277 	old->c6 = new->c6 - old->c6;
1278 	old->c7 = new->c7 - old->c7;
1279 	old->core_temp_c = new->core_temp_c;
1280 	old->mc6_us = new->mc6_us - old->mc6_us;
1281 
1282 	DELTA_WRAP32(new->core_energy, old->core_energy);
1283 
1284 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1285 		if (mp->format == FORMAT_RAW)
1286 			old->counter[i] = new->counter[i];
1287 		else
1288 			old->counter[i] = new->counter[i] - old->counter[i];
1289 	}
1290 }
1291 
1292 int soft_c1_residency_display(int bic)
1293 {
1294 	if (!DO_BIC(BIC_CPU_c1) || use_c1_residency_msr)
1295 		return 0;
1296 
1297 	return DO_BIC_READ(bic);
1298 }
1299 
1300 /*
1301  * old = new - old
1302  */
1303 int
1304 delta_thread(struct thread_data *new, struct thread_data *old,
1305 	struct core_data *core_delta)
1306 {
1307 	int i;
1308 	struct msr_counter *mp;
1309 
1310 	/* we run cpuid just the 1st time, copy the results */
1311 	if (DO_BIC(BIC_APIC))
1312 		new->apic_id = old->apic_id;
1313 	if (DO_BIC(BIC_X2APIC))
1314 		new->x2apic_id = old->x2apic_id;
1315 
1316 	/*
1317 	 * the timestamps from start of measurement interval are in "old"
1318 	 * the timestamp from end of measurement interval are in "new"
1319 	 * over-write old w/ new so we can print end of interval values
1320 	 */
1321 
1322 	timersub(&new->tv_begin, &old->tv_begin, &old->tv_delta);
1323 	old->tv_begin = new->tv_begin;
1324 	old->tv_end = new->tv_end;
1325 
1326 	old->tsc = new->tsc - old->tsc;
1327 
1328 	/* check for TSC < 1 Mcycles over interval */
1329 	if (old->tsc < (1000 * 1000))
1330 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1331 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1332 		     "or just the deep ones with \"processor.max_cstate=1\"");
1333 
1334 	old->c1 = new->c1 - old->c1;
1335 
1336 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) ||
1337 	    soft_c1_residency_display(BIC_Avg_MHz)) {
1338 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1339 			old->aperf = new->aperf - old->aperf;
1340 			old->mperf = new->mperf - old->mperf;
1341 		} else {
1342 			return -1;
1343 		}
1344 	}
1345 
1346 
1347 	if (use_c1_residency_msr) {
1348 		/*
1349 		 * Some models have a dedicated C1 residency MSR,
1350 		 * which should be more accurate than the derivation below.
1351 		 */
1352 	} else {
1353 		/*
1354 		 * As counter collection is not atomic,
1355 		 * it is possible for mperf's non-halted cycles + idle states
1356 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1357 		 */
1358 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1359 			old->c1 = 0;
1360 		else {
1361 			/* normal case, derive c1 */
1362 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1363 				- core_delta->c6 - core_delta->c7;
1364 		}
1365 	}
1366 
1367 	if (old->mperf == 0) {
1368 		if (debug > 1)
1369 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1370 		old->mperf = 1;	/* divide by 0 protection */
1371 	}
1372 
1373 	if (DO_BIC(BIC_IRQ))
1374 		old->irq_count = new->irq_count - old->irq_count;
1375 
1376 	if (DO_BIC(BIC_SMI))
1377 		old->smi_count = new->smi_count - old->smi_count;
1378 
1379 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1380 		if (mp->format == FORMAT_RAW)
1381 			old->counter[i] = new->counter[i];
1382 		else
1383 			old->counter[i] = new->counter[i] - old->counter[i];
1384 	}
1385 	return 0;
1386 }
1387 
1388 int delta_cpu(struct thread_data *t, struct core_data *c,
1389 	struct pkg_data *p, struct thread_data *t2,
1390 	struct core_data *c2, struct pkg_data *p2)
1391 {
1392 	int retval = 0;
1393 
1394 	/* calculate core delta only for 1st thread in core */
1395 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1396 		delta_core(c, c2);
1397 
1398 	/* always calculate thread delta */
1399 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1400 	if (retval)
1401 		return retval;
1402 
1403 	/* calculate package delta only for 1st core in package */
1404 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1405 		retval = delta_package(p, p2);
1406 
1407 	return retval;
1408 }
1409 
1410 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1411 {
1412 	int i;
1413 	struct msr_counter  *mp;
1414 
1415 	t->tv_begin.tv_sec = 0;
1416 	t->tv_begin.tv_usec = 0;
1417 	t->tv_end.tv_sec = 0;
1418 	t->tv_end.tv_usec = 0;
1419 	t->tv_delta.tv_sec = 0;
1420 	t->tv_delta.tv_usec = 0;
1421 
1422 	t->tsc = 0;
1423 	t->aperf = 0;
1424 	t->mperf = 0;
1425 	t->c1 = 0;
1426 
1427 	t->irq_count = 0;
1428 	t->smi_count = 0;
1429 
1430 	/* tells format_counters to dump all fields from this set */
1431 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1432 
1433 	c->c3 = 0;
1434 	c->c6 = 0;
1435 	c->c7 = 0;
1436 	c->mc6_us = 0;
1437 	c->core_temp_c = 0;
1438 	c->core_energy = 0;
1439 
1440 	p->pkg_wtd_core_c0 = 0;
1441 	p->pkg_any_core_c0 = 0;
1442 	p->pkg_any_gfxe_c0 = 0;
1443 	p->pkg_both_core_gfxe_c0 = 0;
1444 
1445 	p->pc2 = 0;
1446 	if (DO_BIC(BIC_Pkgpc3))
1447 		p->pc3 = 0;
1448 	if (DO_BIC(BIC_Pkgpc6))
1449 		p->pc6 = 0;
1450 	if (DO_BIC(BIC_Pkgpc7))
1451 		p->pc7 = 0;
1452 	p->pc8 = 0;
1453 	p->pc9 = 0;
1454 	p->pc10 = 0;
1455 	p->cpu_lpi = 0;
1456 	p->sys_lpi = 0;
1457 
1458 	p->energy_pkg = 0;
1459 	p->energy_dram = 0;
1460 	p->energy_cores = 0;
1461 	p->energy_gfx = 0;
1462 	p->rapl_pkg_perf_status = 0;
1463 	p->rapl_dram_perf_status = 0;
1464 	p->pkg_temp_c = 0;
1465 
1466 	p->gfx_rc6_ms = 0;
1467 	p->gfx_mhz = 0;
1468 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1469 		t->counter[i] = 0;
1470 
1471 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1472 		c->counter[i] = 0;
1473 
1474 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1475 		p->counter[i] = 0;
1476 }
1477 int sum_counters(struct thread_data *t, struct core_data *c,
1478 	struct pkg_data *p)
1479 {
1480 	int i;
1481 	struct msr_counter *mp;
1482 
1483 	/* copy un-changing apic_id's */
1484 	if (DO_BIC(BIC_APIC))
1485 		average.threads.apic_id = t->apic_id;
1486 	if (DO_BIC(BIC_X2APIC))
1487 		average.threads.x2apic_id = t->x2apic_id;
1488 
1489 	/* remember first tv_begin */
1490 	if (average.threads.tv_begin.tv_sec == 0)
1491 		average.threads.tv_begin = t->tv_begin;
1492 
1493 	/* remember last tv_end */
1494 	average.threads.tv_end = t->tv_end;
1495 
1496 	average.threads.tsc += t->tsc;
1497 	average.threads.aperf += t->aperf;
1498 	average.threads.mperf += t->mperf;
1499 	average.threads.c1 += t->c1;
1500 
1501 	average.threads.irq_count += t->irq_count;
1502 	average.threads.smi_count += t->smi_count;
1503 
1504 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1505 		if (mp->format == FORMAT_RAW)
1506 			continue;
1507 		average.threads.counter[i] += t->counter[i];
1508 	}
1509 
1510 	/* sum per-core values only for 1st thread in core */
1511 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1512 		return 0;
1513 
1514 	average.cores.c3 += c->c3;
1515 	average.cores.c6 += c->c6;
1516 	average.cores.c7 += c->c7;
1517 	average.cores.mc6_us += c->mc6_us;
1518 
1519 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1520 
1521 	average.cores.core_energy += c->core_energy;
1522 
1523 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1524 		if (mp->format == FORMAT_RAW)
1525 			continue;
1526 		average.cores.counter[i] += c->counter[i];
1527 	}
1528 
1529 	/* sum per-pkg values only for 1st core in pkg */
1530 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1531 		return 0;
1532 
1533 	if (DO_BIC(BIC_Totl_c0))
1534 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1535 	if (DO_BIC(BIC_Any_c0))
1536 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1537 	if (DO_BIC(BIC_GFX_c0))
1538 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1539 	if (DO_BIC(BIC_CPUGFX))
1540 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1541 
1542 	average.packages.pc2 += p->pc2;
1543 	if (DO_BIC(BIC_Pkgpc3))
1544 		average.packages.pc3 += p->pc3;
1545 	if (DO_BIC(BIC_Pkgpc6))
1546 		average.packages.pc6 += p->pc6;
1547 	if (DO_BIC(BIC_Pkgpc7))
1548 		average.packages.pc7 += p->pc7;
1549 	average.packages.pc8 += p->pc8;
1550 	average.packages.pc9 += p->pc9;
1551 	average.packages.pc10 += p->pc10;
1552 
1553 	average.packages.cpu_lpi = p->cpu_lpi;
1554 	average.packages.sys_lpi = p->sys_lpi;
1555 
1556 	average.packages.energy_pkg += p->energy_pkg;
1557 	average.packages.energy_dram += p->energy_dram;
1558 	average.packages.energy_cores += p->energy_cores;
1559 	average.packages.energy_gfx += p->energy_gfx;
1560 
1561 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1562 	average.packages.gfx_mhz = p->gfx_mhz;
1563 
1564 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1565 
1566 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1567 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1568 
1569 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1570 		if (mp->format == FORMAT_RAW)
1571 			continue;
1572 		average.packages.counter[i] += p->counter[i];
1573 	}
1574 	return 0;
1575 }
1576 /*
1577  * sum the counters for all cpus in the system
1578  * compute the weighted average
1579  */
1580 void compute_average(struct thread_data *t, struct core_data *c,
1581 	struct pkg_data *p)
1582 {
1583 	int i;
1584 	struct msr_counter *mp;
1585 
1586 	clear_counters(&average.threads, &average.cores, &average.packages);
1587 
1588 	for_all_cpus(sum_counters, t, c, p);
1589 
1590 	/* Use the global time delta for the average. */
1591 	average.threads.tv_delta = tv_delta;
1592 
1593 	average.threads.tsc /= topo.num_cpus;
1594 	average.threads.aperf /= topo.num_cpus;
1595 	average.threads.mperf /= topo.num_cpus;
1596 	average.threads.c1 /= topo.num_cpus;
1597 
1598 	if (average.threads.irq_count > 9999999)
1599 		sums_need_wide_columns = 1;
1600 
1601 	average.cores.c3 /= topo.num_cores;
1602 	average.cores.c6 /= topo.num_cores;
1603 	average.cores.c7 /= topo.num_cores;
1604 	average.cores.mc6_us /= topo.num_cores;
1605 
1606 	if (DO_BIC(BIC_Totl_c0))
1607 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1608 	if (DO_BIC(BIC_Any_c0))
1609 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1610 	if (DO_BIC(BIC_GFX_c0))
1611 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1612 	if (DO_BIC(BIC_CPUGFX))
1613 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1614 
1615 	average.packages.pc2 /= topo.num_packages;
1616 	if (DO_BIC(BIC_Pkgpc3))
1617 		average.packages.pc3 /= topo.num_packages;
1618 	if (DO_BIC(BIC_Pkgpc6))
1619 		average.packages.pc6 /= topo.num_packages;
1620 	if (DO_BIC(BIC_Pkgpc7))
1621 		average.packages.pc7 /= topo.num_packages;
1622 
1623 	average.packages.pc8 /= topo.num_packages;
1624 	average.packages.pc9 /= topo.num_packages;
1625 	average.packages.pc10 /= topo.num_packages;
1626 
1627 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1628 		if (mp->format == FORMAT_RAW)
1629 			continue;
1630 		if (mp->type == COUNTER_ITEMS) {
1631 			if (average.threads.counter[i] > 9999999)
1632 				sums_need_wide_columns = 1;
1633 			continue;
1634 		}
1635 		average.threads.counter[i] /= topo.num_cpus;
1636 	}
1637 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1638 		if (mp->format == FORMAT_RAW)
1639 			continue;
1640 		if (mp->type == COUNTER_ITEMS) {
1641 			if (average.cores.counter[i] > 9999999)
1642 				sums_need_wide_columns = 1;
1643 		}
1644 		average.cores.counter[i] /= topo.num_cores;
1645 	}
1646 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1647 		if (mp->format == FORMAT_RAW)
1648 			continue;
1649 		if (mp->type == COUNTER_ITEMS) {
1650 			if (average.packages.counter[i] > 9999999)
1651 				sums_need_wide_columns = 1;
1652 		}
1653 		average.packages.counter[i] /= topo.num_packages;
1654 	}
1655 }
1656 
1657 static unsigned long long rdtsc(void)
1658 {
1659 	unsigned int low, high;
1660 
1661 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
1662 
1663 	return low | ((unsigned long long)high) << 32;
1664 }
1665 
1666 /*
1667  * Open a file, and exit on failure
1668  */
1669 FILE *fopen_or_die(const char *path, const char *mode)
1670 {
1671 	FILE *filep = fopen(path, mode);
1672 
1673 	if (!filep)
1674 		err(1, "%s: open failed", path);
1675 	return filep;
1676 }
1677 /*
1678  * snapshot_sysfs_counter()
1679  *
1680  * return snapshot of given counter
1681  */
1682 unsigned long long snapshot_sysfs_counter(char *path)
1683 {
1684 	FILE *fp;
1685 	int retval;
1686 	unsigned long long counter;
1687 
1688 	fp = fopen_or_die(path, "r");
1689 
1690 	retval = fscanf(fp, "%lld", &counter);
1691 	if (retval != 1)
1692 		err(1, "snapshot_sysfs_counter(%s)", path);
1693 
1694 	fclose(fp);
1695 
1696 	return counter;
1697 }
1698 
1699 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1700 {
1701 	if (mp->msr_num != 0) {
1702 		if (get_msr(cpu, mp->msr_num, counterp))
1703 			return -1;
1704 	} else {
1705 		char path[128 + PATH_BYTES];
1706 
1707 		if (mp->flags & SYSFS_PERCPU) {
1708 			sprintf(path, "/sys/devices/system/cpu/cpu%d/%s",
1709 				 cpu, mp->path);
1710 
1711 			*counterp = snapshot_sysfs_counter(path);
1712 		} else {
1713 			*counterp = snapshot_sysfs_counter(mp->path);
1714 		}
1715 	}
1716 
1717 	return 0;
1718 }
1719 
1720 void get_apic_id(struct thread_data *t)
1721 {
1722 	unsigned int eax, ebx, ecx, edx;
1723 
1724 	if (DO_BIC(BIC_APIC)) {
1725 		eax = ebx = ecx = edx = 0;
1726 		__cpuid(1, eax, ebx, ecx, edx);
1727 
1728 		t->apic_id = (ebx >> 24) & 0xff;
1729 	}
1730 
1731 	if (!DO_BIC(BIC_X2APIC))
1732 		return;
1733 
1734 	if (authentic_amd || hygon_genuine) {
1735 		unsigned int topology_extensions;
1736 
1737 		if (max_extended_level < 0x8000001e)
1738 			return;
1739 
1740 		eax = ebx = ecx = edx = 0;
1741 		__cpuid(0x80000001, eax, ebx, ecx, edx);
1742 			topology_extensions = ecx & (1 << 22);
1743 
1744 		if (topology_extensions == 0)
1745 			return;
1746 
1747 		eax = ebx = ecx = edx = 0;
1748 		__cpuid(0x8000001e, eax, ebx, ecx, edx);
1749 
1750 		t->x2apic_id = eax;
1751 		return;
1752 	}
1753 
1754 	if (!genuine_intel)
1755 		return;
1756 
1757 	if (max_level < 0xb)
1758 		return;
1759 
1760 	ecx = 0;
1761 	__cpuid(0xb, eax, ebx, ecx, edx);
1762 	t->x2apic_id = edx;
1763 
1764 	if (debug && (t->apic_id != (t->x2apic_id & 0xff)))
1765 		fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n",
1766 				t->cpu_id, t->apic_id, t->x2apic_id);
1767 }
1768 
1769 /*
1770  * get_counters(...)
1771  * migrate to cpu
1772  * acquire and record local counters for that cpu
1773  */
1774 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1775 {
1776 	int cpu = t->cpu_id;
1777 	unsigned long long msr;
1778 	int aperf_mperf_retry_count = 0;
1779 	struct msr_counter *mp;
1780 	int i;
1781 
1782 	if (cpu_migrate(cpu)) {
1783 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1784 		return -1;
1785 	}
1786 
1787 	gettimeofday(&t->tv_begin, (struct timezone *)NULL);
1788 
1789 	if (first_counter_read)
1790 		get_apic_id(t);
1791 retry:
1792 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
1793 
1794 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) ||
1795 	    soft_c1_residency_display(BIC_Avg_MHz)) {
1796 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1797 
1798 		/*
1799 		 * The TSC, APERF and MPERF must be read together for
1800 		 * APERF/MPERF and MPERF/TSC to give accurate results.
1801 		 *
1802 		 * Unfortunately, APERF and MPERF are read by
1803 		 * individual system call, so delays may occur
1804 		 * between them.  If the time to read them
1805 		 * varies by a large amount, we re-read them.
1806 		 */
1807 
1808 		/*
1809 		 * This initial dummy APERF read has been seen to
1810 		 * reduce jitter in the subsequent reads.
1811 		 */
1812 
1813 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1814 			return -3;
1815 
1816 		t->tsc = rdtsc();	/* re-read close to APERF */
1817 
1818 		tsc_before = t->tsc;
1819 
1820 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1821 			return -3;
1822 
1823 		tsc_between = rdtsc();
1824 
1825 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1826 			return -4;
1827 
1828 		tsc_after = rdtsc();
1829 
1830 		aperf_time = tsc_between - tsc_before;
1831 		mperf_time = tsc_after - tsc_between;
1832 
1833 		/*
1834 		 * If the system call latency to read APERF and MPERF
1835 		 * differ by more than 2x, then try again.
1836 		 */
1837 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1838 			aperf_mperf_retry_count++;
1839 			if (aperf_mperf_retry_count < 5)
1840 				goto retry;
1841 			else
1842 				warnx("cpu%d jitter %lld %lld",
1843 					cpu, aperf_time, mperf_time);
1844 		}
1845 		aperf_mperf_retry_count = 0;
1846 
1847 		t->aperf = t->aperf * aperf_mperf_multiplier;
1848 		t->mperf = t->mperf * aperf_mperf_multiplier;
1849 	}
1850 
1851 	if (DO_BIC(BIC_IRQ))
1852 		t->irq_count = irqs_per_cpu[cpu];
1853 	if (DO_BIC(BIC_SMI)) {
1854 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1855 			return -5;
1856 		t->smi_count = msr & 0xFFFFFFFF;
1857 	}
1858 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1859 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1860 			return -6;
1861 	}
1862 
1863 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1864 		if (get_mp(cpu, mp, &t->counter[i]))
1865 			return -10;
1866 	}
1867 
1868 	/* collect core counters only for 1st thread in core */
1869 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1870 		goto done;
1871 
1872 	if (DO_BIC(BIC_CPU_c3) || soft_c1_residency_display(BIC_CPU_c3)) {
1873 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1874 			return -6;
1875 	}
1876 
1877 	if ((DO_BIC(BIC_CPU_c6) || soft_c1_residency_display(BIC_CPU_c6)) && !do_knl_cstates) {
1878 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1879 			return -7;
1880 	} else if (do_knl_cstates || soft_c1_residency_display(BIC_CPU_c6)) {
1881 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1882 			return -7;
1883 	}
1884 
1885 	if (DO_BIC(BIC_CPU_c7) || soft_c1_residency_display(BIC_CPU_c7))
1886 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1887 			return -8;
1888 
1889 	if (DO_BIC(BIC_Mod_c6))
1890 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1891 			return -8;
1892 
1893 	if (DO_BIC(BIC_CoreTmp)) {
1894 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1895 			return -9;
1896 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1897 	}
1898 
1899 	if (do_rapl & RAPL_AMD_F17H) {
1900 		if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr))
1901 			return -14;
1902 		c->core_energy = msr & 0xFFFFFFFF;
1903 	}
1904 
1905 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1906 		if (get_mp(cpu, mp, &c->counter[i]))
1907 			return -10;
1908 	}
1909 
1910 	/* collect package counters only for 1st core in package */
1911 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1912 		goto done;
1913 
1914 	if (DO_BIC(BIC_Totl_c0)) {
1915 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1916 			return -10;
1917 	}
1918 	if (DO_BIC(BIC_Any_c0)) {
1919 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1920 			return -11;
1921 	}
1922 	if (DO_BIC(BIC_GFX_c0)) {
1923 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1924 			return -12;
1925 	}
1926 	if (DO_BIC(BIC_CPUGFX)) {
1927 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1928 			return -13;
1929 	}
1930 	if (DO_BIC(BIC_Pkgpc3))
1931 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1932 			return -9;
1933 	if (DO_BIC(BIC_Pkgpc6)) {
1934 		if (do_slm_cstates) {
1935 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1936 				return -10;
1937 		} else {
1938 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1939 				return -10;
1940 		}
1941 	}
1942 
1943 	if (DO_BIC(BIC_Pkgpc2))
1944 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1945 			return -11;
1946 	if (DO_BIC(BIC_Pkgpc7))
1947 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1948 			return -12;
1949 	if (DO_BIC(BIC_Pkgpc8))
1950 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1951 			return -13;
1952 	if (DO_BIC(BIC_Pkgpc9))
1953 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1954 			return -13;
1955 	if (DO_BIC(BIC_Pkgpc10))
1956 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1957 			return -13;
1958 
1959 	if (DO_BIC(BIC_CPU_LPI))
1960 		p->cpu_lpi = cpuidle_cur_cpu_lpi_us;
1961 	if (DO_BIC(BIC_SYS_LPI))
1962 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
1963 
1964 	if (do_rapl & RAPL_PKG) {
1965 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1966 			return -13;
1967 		p->energy_pkg = msr & 0xFFFFFFFF;
1968 	}
1969 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1970 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1971 			return -14;
1972 		p->energy_cores = msr & 0xFFFFFFFF;
1973 	}
1974 	if (do_rapl & RAPL_DRAM) {
1975 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1976 			return -15;
1977 		p->energy_dram = msr & 0xFFFFFFFF;
1978 	}
1979 	if (do_rapl & RAPL_GFX) {
1980 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1981 			return -16;
1982 		p->energy_gfx = msr & 0xFFFFFFFF;
1983 	}
1984 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1985 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1986 			return -16;
1987 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1988 	}
1989 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1990 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1991 			return -16;
1992 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1993 	}
1994 	if (do_rapl & RAPL_AMD_F17H) {
1995 		if (get_msr(cpu, MSR_PKG_ENERGY_STAT, &msr))
1996 			return -13;
1997 		p->energy_pkg = msr & 0xFFFFFFFF;
1998 	}
1999 	if (DO_BIC(BIC_PkgTmp)) {
2000 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
2001 			return -17;
2002 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
2003 	}
2004 
2005 	if (DO_BIC(BIC_GFX_rc6))
2006 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
2007 
2008 	if (DO_BIC(BIC_GFXMHz))
2009 		p->gfx_mhz = gfx_cur_mhz;
2010 
2011 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
2012 		if (get_mp(cpu, mp, &p->counter[i]))
2013 			return -10;
2014 	}
2015 done:
2016 	gettimeofday(&t->tv_end, (struct timezone *)NULL);
2017 
2018 	return 0;
2019 }
2020 
2021 /*
2022  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
2023  * If you change the values, note they are used both in comparisons
2024  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
2025  */
2026 
2027 #define PCLUKN 0 /* Unknown */
2028 #define PCLRSV 1 /* Reserved */
2029 #define PCL__0 2 /* PC0 */
2030 #define PCL__1 3 /* PC1 */
2031 #define PCL__2 4 /* PC2 */
2032 #define PCL__3 5 /* PC3 */
2033 #define PCL__4 6 /* PC4 */
2034 #define PCL__6 7 /* PC6 */
2035 #define PCL_6N 8 /* PC6 No Retention */
2036 #define PCL_6R 9 /* PC6 Retention */
2037 #define PCL__7 10 /* PC7 */
2038 #define PCL_7S 11 /* PC7 Shrink */
2039 #define PCL__8 12 /* PC8 */
2040 #define PCL__9 13 /* PC9 */
2041 #define PCL_10 14 /* PC10 */
2042 #define PCLUNL 15 /* Unlimited */
2043 
2044 int pkg_cstate_limit = PCLUKN;
2045 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
2046 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited"};
2047 
2048 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};
2049 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};
2050 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};
2051 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};
2052 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};
2053 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};
2054 int glm_pkg_cstate_limits[16] = {PCLUNL, PCL__1, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCL_10, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
2055 int skx_pkg_cstate_limits[16] = {PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV};
2056 
2057 
2058 static void
2059 calculate_tsc_tweak()
2060 {
2061 	tsc_tweak = base_hz / tsc_hz;
2062 }
2063 
2064 static void
2065 dump_nhm_platform_info(void)
2066 {
2067 	unsigned long long msr;
2068 	unsigned int ratio;
2069 
2070 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2071 
2072 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
2073 
2074 	ratio = (msr >> 40) & 0xFF;
2075 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
2076 		ratio, bclk, ratio * bclk);
2077 
2078 	ratio = (msr >> 8) & 0xFF;
2079 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2080 		ratio, bclk, ratio * bclk);
2081 
2082 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
2083 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
2084 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
2085 
2086 	return;
2087 }
2088 
2089 static void
2090 dump_hsw_turbo_ratio_limits(void)
2091 {
2092 	unsigned long long msr;
2093 	unsigned int ratio;
2094 
2095 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
2096 
2097 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
2098 
2099 	ratio = (msr >> 8) & 0xFF;
2100 	if (ratio)
2101 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
2102 			ratio, bclk, ratio * bclk);
2103 
2104 	ratio = (msr >> 0) & 0xFF;
2105 	if (ratio)
2106 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
2107 			ratio, bclk, ratio * bclk);
2108 	return;
2109 }
2110 
2111 static void
2112 dump_ivt_turbo_ratio_limits(void)
2113 {
2114 	unsigned long long msr;
2115 	unsigned int ratio;
2116 
2117 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
2118 
2119 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
2120 
2121 	ratio = (msr >> 56) & 0xFF;
2122 	if (ratio)
2123 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
2124 			ratio, bclk, ratio * bclk);
2125 
2126 	ratio = (msr >> 48) & 0xFF;
2127 	if (ratio)
2128 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
2129 			ratio, bclk, ratio * bclk);
2130 
2131 	ratio = (msr >> 40) & 0xFF;
2132 	if (ratio)
2133 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
2134 			ratio, bclk, ratio * bclk);
2135 
2136 	ratio = (msr >> 32) & 0xFF;
2137 	if (ratio)
2138 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
2139 			ratio, bclk, ratio * bclk);
2140 
2141 	ratio = (msr >> 24) & 0xFF;
2142 	if (ratio)
2143 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
2144 			ratio, bclk, ratio * bclk);
2145 
2146 	ratio = (msr >> 16) & 0xFF;
2147 	if (ratio)
2148 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
2149 			ratio, bclk, ratio * bclk);
2150 
2151 	ratio = (msr >> 8) & 0xFF;
2152 	if (ratio)
2153 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
2154 			ratio, bclk, ratio * bclk);
2155 
2156 	ratio = (msr >> 0) & 0xFF;
2157 	if (ratio)
2158 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
2159 			ratio, bclk, ratio * bclk);
2160 	return;
2161 }
2162 int has_turbo_ratio_group_limits(int family, int model)
2163 {
2164 
2165 	if (!genuine_intel)
2166 		return 0;
2167 
2168 	switch (model) {
2169 	case INTEL_FAM6_ATOM_GOLDMONT:
2170 	case INTEL_FAM6_SKYLAKE_X:
2171 	case INTEL_FAM6_ATOM_GOLDMONT_D:
2172 		return 1;
2173 	}
2174 	return 0;
2175 }
2176 
2177 static void
2178 dump_turbo_ratio_limits(int family, int model)
2179 {
2180 	unsigned long long msr, core_counts;
2181 	unsigned int ratio, group_size;
2182 
2183 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2184 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
2185 
2186 	if (has_turbo_ratio_group_limits(family, model)) {
2187 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
2188 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
2189 	} else {
2190 		core_counts = 0x0807060504030201;
2191 	}
2192 
2193 	ratio = (msr >> 56) & 0xFF;
2194 	group_size = (core_counts >> 56) & 0xFF;
2195 	if (ratio)
2196 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2197 			ratio, bclk, ratio * bclk, group_size);
2198 
2199 	ratio = (msr >> 48) & 0xFF;
2200 	group_size = (core_counts >> 48) & 0xFF;
2201 	if (ratio)
2202 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2203 			ratio, bclk, ratio * bclk, group_size);
2204 
2205 	ratio = (msr >> 40) & 0xFF;
2206 	group_size = (core_counts >> 40) & 0xFF;
2207 	if (ratio)
2208 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2209 			ratio, bclk, ratio * bclk, group_size);
2210 
2211 	ratio = (msr >> 32) & 0xFF;
2212 	group_size = (core_counts >> 32) & 0xFF;
2213 	if (ratio)
2214 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2215 			ratio, bclk, ratio * bclk, group_size);
2216 
2217 	ratio = (msr >> 24) & 0xFF;
2218 	group_size = (core_counts >> 24) & 0xFF;
2219 	if (ratio)
2220 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2221 			ratio, bclk, ratio * bclk, group_size);
2222 
2223 	ratio = (msr >> 16) & 0xFF;
2224 	group_size = (core_counts >> 16) & 0xFF;
2225 	if (ratio)
2226 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2227 			ratio, bclk, ratio * bclk, group_size);
2228 
2229 	ratio = (msr >> 8) & 0xFF;
2230 	group_size = (core_counts >> 8) & 0xFF;
2231 	if (ratio)
2232 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2233 			ratio, bclk, ratio * bclk, group_size);
2234 
2235 	ratio = (msr >> 0) & 0xFF;
2236 	group_size = (core_counts >> 0) & 0xFF;
2237 	if (ratio)
2238 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2239 			ratio, bclk, ratio * bclk, group_size);
2240 	return;
2241 }
2242 
2243 static void
2244 dump_atom_turbo_ratio_limits(void)
2245 {
2246 	unsigned long long msr;
2247 	unsigned int ratio;
2248 
2249 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
2250 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2251 
2252 	ratio = (msr >> 0) & 0x3F;
2253 	if (ratio)
2254 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
2255 			ratio, bclk, ratio * bclk);
2256 
2257 	ratio = (msr >> 8) & 0x3F;
2258 	if (ratio)
2259 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
2260 			ratio, bclk, ratio * bclk);
2261 
2262 	ratio = (msr >> 16) & 0x3F;
2263 	if (ratio)
2264 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2265 			ratio, bclk, ratio * bclk);
2266 
2267 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
2268 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2269 
2270 	ratio = (msr >> 24) & 0x3F;
2271 	if (ratio)
2272 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
2273 			ratio, bclk, ratio * bclk);
2274 
2275 	ratio = (msr >> 16) & 0x3F;
2276 	if (ratio)
2277 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
2278 			ratio, bclk, ratio * bclk);
2279 
2280 	ratio = (msr >> 8) & 0x3F;
2281 	if (ratio)
2282 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
2283 			ratio, bclk, ratio * bclk);
2284 
2285 	ratio = (msr >> 0) & 0x3F;
2286 	if (ratio)
2287 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
2288 			ratio, bclk, ratio * bclk);
2289 }
2290 
2291 static void
2292 dump_knl_turbo_ratio_limits(void)
2293 {
2294 	const unsigned int buckets_no = 7;
2295 
2296 	unsigned long long msr;
2297 	int delta_cores, delta_ratio;
2298 	int i, b_nr;
2299 	unsigned int cores[buckets_no];
2300 	unsigned int ratio[buckets_no];
2301 
2302 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2303 
2304 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
2305 		base_cpu, msr);
2306 
2307 	/**
2308 	 * Turbo encoding in KNL is as follows:
2309 	 * [0] -- Reserved
2310 	 * [7:1] -- Base value of number of active cores of bucket 1.
2311 	 * [15:8] -- Base value of freq ratio of bucket 1.
2312 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
2313 	 * i.e. active cores of bucket 2 =
2314 	 * active cores of bucket 1 + delta
2315 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
2316 	 * i.e. freq ratio of bucket 2 =
2317 	 * freq ratio of bucket 1 - delta
2318 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
2319 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
2320 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
2321 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
2322 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
2323 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
2324 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
2325 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
2326 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
2327 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
2328 	 */
2329 
2330 	b_nr = 0;
2331 	cores[b_nr] = (msr & 0xFF) >> 1;
2332 	ratio[b_nr] = (msr >> 8) & 0xFF;
2333 
2334 	for (i = 16; i < 64; i += 8) {
2335 		delta_cores = (msr >> i) & 0x1F;
2336 		delta_ratio = (msr >> (i + 5)) & 0x7;
2337 
2338 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
2339 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2340 		b_nr++;
2341 	}
2342 
2343 	for (i = buckets_no - 1; i >= 0; i--)
2344 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2345 			fprintf(outf,
2346 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
2347 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
2348 }
2349 
2350 static void
2351 dump_nhm_cst_cfg(void)
2352 {
2353 	unsigned long long msr;
2354 
2355 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2356 
2357 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2358 
2359 	fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)",
2360 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2361 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2362 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2363 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2364 		(msr & (1 << 15)) ? "" : "UN",
2365 		(unsigned int)msr & 0xF,
2366 		pkg_cstate_limit_strings[pkg_cstate_limit]);
2367 
2368 #define AUTOMATIC_CSTATE_CONVERSION		(1UL << 16)
2369 	if (has_automatic_cstate_conversion) {
2370 		fprintf(outf, ", automatic c-state conversion=%s",
2371 			(msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off");
2372 	}
2373 
2374 	fprintf(outf, ")\n");
2375 
2376 	return;
2377 }
2378 
2379 static void
2380 dump_config_tdp(void)
2381 {
2382 	unsigned long long msr;
2383 
2384 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2385 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2386 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2387 
2388 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2389 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2390 	if (msr) {
2391 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2392 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2393 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2394 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2395 	}
2396 	fprintf(outf, ")\n");
2397 
2398 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2399 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2400 	if (msr) {
2401 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2402 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2403 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2404 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2405 	}
2406 	fprintf(outf, ")\n");
2407 
2408 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2409 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2410 	if ((msr) & 0x3)
2411 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2412 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2413 	fprintf(outf, ")\n");
2414 
2415 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2416 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2417 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2418 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2419 	fprintf(outf, ")\n");
2420 }
2421 
2422 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2423 
2424 void print_irtl(void)
2425 {
2426 	unsigned long long msr;
2427 
2428 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2429 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2430 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2431 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2432 
2433 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2434 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2435 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2436 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2437 
2438 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2439 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2440 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2441 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2442 
2443 	if (!do_irtl_hsw)
2444 		return;
2445 
2446 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2447 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2448 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2449 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2450 
2451 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2452 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2453 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2454 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2455 
2456 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2457 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2458 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2459 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2460 
2461 }
2462 void free_fd_percpu(void)
2463 {
2464 	int i;
2465 
2466 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2467 		if (fd_percpu[i] != 0)
2468 			close(fd_percpu[i]);
2469 	}
2470 
2471 	free(fd_percpu);
2472 }
2473 
2474 void free_all_buffers(void)
2475 {
2476 	int i;
2477 
2478 	CPU_FREE(cpu_present_set);
2479 	cpu_present_set = NULL;
2480 	cpu_present_setsize = 0;
2481 
2482 	CPU_FREE(cpu_affinity_set);
2483 	cpu_affinity_set = NULL;
2484 	cpu_affinity_setsize = 0;
2485 
2486 	free(thread_even);
2487 	free(core_even);
2488 	free(package_even);
2489 
2490 	thread_even = NULL;
2491 	core_even = NULL;
2492 	package_even = NULL;
2493 
2494 	free(thread_odd);
2495 	free(core_odd);
2496 	free(package_odd);
2497 
2498 	thread_odd = NULL;
2499 	core_odd = NULL;
2500 	package_odd = NULL;
2501 
2502 	free(output_buffer);
2503 	output_buffer = NULL;
2504 	outp = NULL;
2505 
2506 	free_fd_percpu();
2507 
2508 	free(irq_column_2_cpu);
2509 	free(irqs_per_cpu);
2510 
2511 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2512 		if (cpus[i].put_ids)
2513 			CPU_FREE(cpus[i].put_ids);
2514 	}
2515 	free(cpus);
2516 }
2517 
2518 
2519 /*
2520  * Parse a file containing a single int.
2521  * Return 0 if file can not be opened
2522  * Exit if file can be opened, but can not be parsed
2523  */
2524 int parse_int_file(const char *fmt, ...)
2525 {
2526 	va_list args;
2527 	char path[PATH_MAX];
2528 	FILE *filep;
2529 	int value;
2530 
2531 	va_start(args, fmt);
2532 	vsnprintf(path, sizeof(path), fmt, args);
2533 	va_end(args);
2534 	filep = fopen(path, "r");
2535 	if (!filep)
2536 		return 0;
2537 	if (fscanf(filep, "%d", &value) != 1)
2538 		err(1, "%s: failed to parse number from file", path);
2539 	fclose(filep);
2540 	return value;
2541 }
2542 
2543 /*
2544  * cpu_is_first_core_in_package(cpu)
2545  * return 1 if given CPU is 1st core in package
2546  */
2547 int cpu_is_first_core_in_package(int cpu)
2548 {
2549 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2550 }
2551 
2552 int get_physical_package_id(int cpu)
2553 {
2554 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2555 }
2556 
2557 int get_die_id(int cpu)
2558 {
2559 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu);
2560 }
2561 
2562 int get_core_id(int cpu)
2563 {
2564 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2565 }
2566 
2567 void set_node_data(void)
2568 {
2569 	int pkg, node, lnode, cpu, cpux;
2570 	int cpu_count;
2571 
2572 	/* initialize logical_node_id */
2573 	for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu)
2574 		cpus[cpu].logical_node_id = -1;
2575 
2576 	cpu_count = 0;
2577 	for (pkg = 0; pkg < topo.num_packages; pkg++) {
2578 		lnode = 0;
2579 		for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) {
2580 			if (cpus[cpu].physical_package_id != pkg)
2581 				continue;
2582 			/* find a cpu with an unset logical_node_id */
2583 			if (cpus[cpu].logical_node_id != -1)
2584 				continue;
2585 			cpus[cpu].logical_node_id = lnode;
2586 			node = cpus[cpu].physical_node_id;
2587 			cpu_count++;
2588 			/*
2589 			 * find all matching cpus on this pkg and set
2590 			 * the logical_node_id
2591 			 */
2592 			for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) {
2593 				if ((cpus[cpux].physical_package_id == pkg) &&
2594 				   (cpus[cpux].physical_node_id == node)) {
2595 					cpus[cpux].logical_node_id = lnode;
2596 					cpu_count++;
2597 				}
2598 			}
2599 			lnode++;
2600 			if (lnode > topo.nodes_per_pkg)
2601 				topo.nodes_per_pkg = lnode;
2602 		}
2603 		if (cpu_count >= topo.max_cpu_num)
2604 			break;
2605 	}
2606 }
2607 
2608 int get_physical_node_id(struct cpu_topology *thiscpu)
2609 {
2610 	char path[80];
2611 	FILE *filep;
2612 	int i;
2613 	int cpu = thiscpu->logical_cpu_id;
2614 
2615 	for (i = 0; i <= topo.max_cpu_num; i++) {
2616 		sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist",
2617 			cpu, i);
2618 		filep = fopen(path, "r");
2619 		if (!filep)
2620 			continue;
2621 		fclose(filep);
2622 		return i;
2623 	}
2624 	return -1;
2625 }
2626 
2627 int get_thread_siblings(struct cpu_topology *thiscpu)
2628 {
2629 	char path[80], character;
2630 	FILE *filep;
2631 	unsigned long map;
2632 	int so, shift, sib_core;
2633 	int cpu = thiscpu->logical_cpu_id;
2634 	int offset = topo.max_cpu_num + 1;
2635 	size_t size;
2636 	int thread_id = 0;
2637 
2638 	thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1));
2639 	if (thiscpu->thread_id < 0)
2640 		thiscpu->thread_id = thread_id++;
2641 	if (!thiscpu->put_ids)
2642 		return -1;
2643 
2644 	size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2645 	CPU_ZERO_S(size, thiscpu->put_ids);
2646 
2647 	sprintf(path,
2648 		"/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu);
2649 	filep = fopen_or_die(path, "r");
2650 	do {
2651 		offset -= BITMASK_SIZE;
2652 		if (fscanf(filep, "%lx%c", &map, &character) != 2)
2653 			err(1, "%s: failed to parse file", path);
2654 		for (shift = 0; shift < BITMASK_SIZE; shift++) {
2655 			if ((map >> shift) & 0x1) {
2656 				so = shift + offset;
2657 				sib_core = get_core_id(so);
2658 				if (sib_core == thiscpu->physical_core_id) {
2659 					CPU_SET_S(so, size, thiscpu->put_ids);
2660 					if ((so != cpu) &&
2661 					    (cpus[so].thread_id < 0))
2662 						cpus[so].thread_id =
2663 								    thread_id++;
2664 				}
2665 			}
2666 		}
2667 	} while (!strncmp(&character, ",", 1));
2668 	fclose(filep);
2669 
2670 	return CPU_COUNT_S(size, thiscpu->put_ids);
2671 }
2672 
2673 /*
2674  * run func(thread, core, package) in topology order
2675  * skip non-present cpus
2676  */
2677 
2678 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2679 	struct pkg_data *, struct thread_data *, struct core_data *,
2680 	struct pkg_data *), struct thread_data *thread_base,
2681 	struct core_data *core_base, struct pkg_data *pkg_base,
2682 	struct thread_data *thread_base2, struct core_data *core_base2,
2683 	struct pkg_data *pkg_base2)
2684 {
2685 	int retval, pkg_no, node_no, core_no, thread_no;
2686 
2687 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2688 		for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) {
2689 			for (core_no = 0; core_no < topo.cores_per_node;
2690 			     ++core_no) {
2691 				for (thread_no = 0; thread_no <
2692 					topo.threads_per_core; ++thread_no) {
2693 					struct thread_data *t, *t2;
2694 					struct core_data *c, *c2;
2695 					struct pkg_data *p, *p2;
2696 
2697 					t = GET_THREAD(thread_base, thread_no,
2698 						       core_no, node_no,
2699 						       pkg_no);
2700 
2701 					if (cpu_is_not_present(t->cpu_id))
2702 						continue;
2703 
2704 					t2 = GET_THREAD(thread_base2, thread_no,
2705 							core_no, node_no,
2706 							pkg_no);
2707 
2708 					c = GET_CORE(core_base, core_no,
2709 						     node_no, pkg_no);
2710 					c2 = GET_CORE(core_base2, core_no,
2711 						      node_no,
2712 						      pkg_no);
2713 
2714 					p = GET_PKG(pkg_base, pkg_no);
2715 					p2 = GET_PKG(pkg_base2, pkg_no);
2716 
2717 					retval = func(t, c, p, t2, c2, p2);
2718 					if (retval)
2719 						return retval;
2720 				}
2721 			}
2722 		}
2723 	}
2724 	return 0;
2725 }
2726 
2727 /*
2728  * run func(cpu) on every cpu in /proc/stat
2729  * return max_cpu number
2730  */
2731 int for_all_proc_cpus(int (func)(int))
2732 {
2733 	FILE *fp;
2734 	int cpu_num;
2735 	int retval;
2736 
2737 	fp = fopen_or_die(proc_stat, "r");
2738 
2739 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2740 	if (retval != 0)
2741 		err(1, "%s: failed to parse format", proc_stat);
2742 
2743 	while (1) {
2744 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2745 		if (retval != 1)
2746 			break;
2747 
2748 		retval = func(cpu_num);
2749 		if (retval) {
2750 			fclose(fp);
2751 			return(retval);
2752 		}
2753 	}
2754 	fclose(fp);
2755 	return 0;
2756 }
2757 
2758 void re_initialize(void)
2759 {
2760 	free_all_buffers();
2761 	setup_all_buffers();
2762 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2763 }
2764 
2765 void set_max_cpu_num(void)
2766 {
2767 	FILE *filep;
2768 	unsigned long dummy;
2769 
2770 	topo.max_cpu_num = 0;
2771 	filep = fopen_or_die(
2772 			"/sys/devices/system/cpu/cpu0/topology/thread_siblings",
2773 			"r");
2774 	while (fscanf(filep, "%lx,", &dummy) == 1)
2775 		topo.max_cpu_num += BITMASK_SIZE;
2776 	fclose(filep);
2777 	topo.max_cpu_num--; /* 0 based */
2778 }
2779 
2780 /*
2781  * count_cpus()
2782  * remember the last one seen, it will be the max
2783  */
2784 int count_cpus(int cpu)
2785 {
2786 	topo.num_cpus++;
2787 	return 0;
2788 }
2789 int mark_cpu_present(int cpu)
2790 {
2791 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2792 	return 0;
2793 }
2794 
2795 int init_thread_id(int cpu)
2796 {
2797 	cpus[cpu].thread_id = -1;
2798 	return 0;
2799 }
2800 
2801 /*
2802  * snapshot_proc_interrupts()
2803  *
2804  * read and record summary of /proc/interrupts
2805  *
2806  * return 1 if config change requires a restart, else return 0
2807  */
2808 int snapshot_proc_interrupts(void)
2809 {
2810 	static FILE *fp;
2811 	int column, retval;
2812 
2813 	if (fp == NULL)
2814 		fp = fopen_or_die("/proc/interrupts", "r");
2815 	else
2816 		rewind(fp);
2817 
2818 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
2819 	for (column = 0; column < topo.num_cpus; ++column) {
2820 		int cpu_number;
2821 
2822 		retval = fscanf(fp, " CPU%d", &cpu_number);
2823 		if (retval != 1)
2824 			break;
2825 
2826 		if (cpu_number > topo.max_cpu_num) {
2827 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2828 			return 1;
2829 		}
2830 
2831 		irq_column_2_cpu[column] = cpu_number;
2832 		irqs_per_cpu[cpu_number] = 0;
2833 	}
2834 
2835 	/* read /proc/interrupt count lines and sum up irqs per cpu */
2836 	while (1) {
2837 		int column;
2838 		char buf[64];
2839 
2840 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
2841 		if (retval != 1)
2842 			break;
2843 
2844 		/* read the count per cpu */
2845 		for (column = 0; column < topo.num_cpus; ++column) {
2846 
2847 			int cpu_number, irq_count;
2848 
2849 			retval = fscanf(fp, " %d", &irq_count);
2850 			if (retval != 1)
2851 				break;
2852 
2853 			cpu_number = irq_column_2_cpu[column];
2854 			irqs_per_cpu[cpu_number] += irq_count;
2855 
2856 		}
2857 
2858 		while (getc(fp) != '\n')
2859 			;	/* flush interrupt description */
2860 
2861 	}
2862 	return 0;
2863 }
2864 /*
2865  * snapshot_gfx_rc6_ms()
2866  *
2867  * record snapshot of
2868  * /sys/class/drm/card0/power/rc6_residency_ms
2869  *
2870  * return 1 if config change requires a restart, else return 0
2871  */
2872 int snapshot_gfx_rc6_ms(void)
2873 {
2874 	FILE *fp;
2875 	int retval;
2876 
2877 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2878 
2879 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2880 	if (retval != 1)
2881 		err(1, "GFX rc6");
2882 
2883 	fclose(fp);
2884 
2885 	return 0;
2886 }
2887 /*
2888  * snapshot_gfx_mhz()
2889  *
2890  * record snapshot of
2891  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2892  *
2893  * return 1 if config change requires a restart, else return 0
2894  */
2895 int snapshot_gfx_mhz(void)
2896 {
2897 	static FILE *fp;
2898 	int retval;
2899 
2900 	if (fp == NULL)
2901 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2902 	else {
2903 		rewind(fp);
2904 		fflush(fp);
2905 	}
2906 
2907 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
2908 	if (retval != 1)
2909 		err(1, "GFX MHz");
2910 
2911 	return 0;
2912 }
2913 
2914 /*
2915  * snapshot_cpu_lpi()
2916  *
2917  * record snapshot of
2918  * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us
2919  *
2920  * return 1 if config change requires a restart, else return 0
2921  */
2922 int snapshot_cpu_lpi_us(void)
2923 {
2924 	FILE *fp;
2925 	int retval;
2926 
2927 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r");
2928 
2929 	retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us);
2930 	if (retval != 1) {
2931 		fprintf(stderr, "Disabling Low Power Idle CPU output\n");
2932 		BIC_NOT_PRESENT(BIC_CPU_LPI);
2933 		fclose(fp);
2934 		return -1;
2935 	}
2936 
2937 	fclose(fp);
2938 
2939 	return 0;
2940 }
2941 /*
2942  * snapshot_sys_lpi()
2943  *
2944  * record snapshot of
2945  * /sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us
2946  *
2947  * return 1 if config change requires a restart, else return 0
2948  */
2949 int snapshot_sys_lpi_us(void)
2950 {
2951 	FILE *fp;
2952 	int retval;
2953 
2954 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", "r");
2955 
2956 	retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us);
2957 	if (retval != 1) {
2958 		fprintf(stderr, "Disabling Low Power Idle System output\n");
2959 		BIC_NOT_PRESENT(BIC_SYS_LPI);
2960 		fclose(fp);
2961 		return -1;
2962 	}
2963 	fclose(fp);
2964 
2965 	return 0;
2966 }
2967 /*
2968  * snapshot /proc and /sys files
2969  *
2970  * return 1 if configuration restart needed, else return 0
2971  */
2972 int snapshot_proc_sysfs_files(void)
2973 {
2974 	if (DO_BIC(BIC_IRQ))
2975 		if (snapshot_proc_interrupts())
2976 			return 1;
2977 
2978 	if (DO_BIC(BIC_GFX_rc6))
2979 		snapshot_gfx_rc6_ms();
2980 
2981 	if (DO_BIC(BIC_GFXMHz))
2982 		snapshot_gfx_mhz();
2983 
2984 	if (DO_BIC(BIC_CPU_LPI))
2985 		snapshot_cpu_lpi_us();
2986 
2987 	if (DO_BIC(BIC_SYS_LPI))
2988 		snapshot_sys_lpi_us();
2989 
2990 	return 0;
2991 }
2992 
2993 int exit_requested;
2994 
2995 static void signal_handler (int signal)
2996 {
2997 	switch (signal) {
2998 	case SIGINT:
2999 		exit_requested = 1;
3000 		if (debug)
3001 			fprintf(stderr, " SIGINT\n");
3002 		break;
3003 	case SIGUSR1:
3004 		if (debug > 1)
3005 			fprintf(stderr, "SIGUSR1\n");
3006 		break;
3007 	}
3008 }
3009 
3010 void setup_signal_handler(void)
3011 {
3012 	struct sigaction sa;
3013 
3014 	memset(&sa, 0, sizeof(sa));
3015 
3016 	sa.sa_handler = &signal_handler;
3017 
3018 	if (sigaction(SIGINT, &sa, NULL) < 0)
3019 		err(1, "sigaction SIGINT");
3020 	if (sigaction(SIGUSR1, &sa, NULL) < 0)
3021 		err(1, "sigaction SIGUSR1");
3022 }
3023 
3024 void do_sleep(void)
3025 {
3026 	struct timeval tout;
3027 	struct timespec rest;
3028 	fd_set readfds;
3029 	int retval;
3030 
3031 	FD_ZERO(&readfds);
3032 	FD_SET(0, &readfds);
3033 
3034 	if (ignore_stdin) {
3035 		nanosleep(&interval_ts, NULL);
3036 		return;
3037 	}
3038 
3039 	tout = interval_tv;
3040 	retval = select(1, &readfds, NULL, NULL, &tout);
3041 
3042 	if (retval == 1) {
3043 		switch (getc(stdin)) {
3044 		case 'q':
3045 			exit_requested = 1;
3046 			break;
3047 		case EOF:
3048 			/*
3049 			 * 'stdin' is a pipe closed on the other end. There
3050 			 * won't be any further input.
3051 			 */
3052 			ignore_stdin = 1;
3053 			/* Sleep the rest of the time */
3054 			rest.tv_sec = (tout.tv_sec + tout.tv_usec / 1000000);
3055 			rest.tv_nsec = (tout.tv_usec % 1000000) * 1000;
3056 			nanosleep(&rest, NULL);
3057 		}
3058 	}
3059 }
3060 
3061 
3062 void turbostat_loop()
3063 {
3064 	int retval;
3065 	int restarted = 0;
3066 	int done_iters = 0;
3067 
3068 	setup_signal_handler();
3069 
3070 restart:
3071 	restarted++;
3072 
3073 	snapshot_proc_sysfs_files();
3074 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3075 	first_counter_read = 0;
3076 	if (retval < -1) {
3077 		exit(retval);
3078 	} else if (retval == -1) {
3079 		if (restarted > 1) {
3080 			exit(retval);
3081 		}
3082 		re_initialize();
3083 		goto restart;
3084 	}
3085 	restarted = 0;
3086 	done_iters = 0;
3087 	gettimeofday(&tv_even, (struct timezone *)NULL);
3088 
3089 	while (1) {
3090 		if (for_all_proc_cpus(cpu_is_not_present)) {
3091 			re_initialize();
3092 			goto restart;
3093 		}
3094 		do_sleep();
3095 		if (snapshot_proc_sysfs_files())
3096 			goto restart;
3097 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
3098 		if (retval < -1) {
3099 			exit(retval);
3100 		} else if (retval == -1) {
3101 			re_initialize();
3102 			goto restart;
3103 		}
3104 		gettimeofday(&tv_odd, (struct timezone *)NULL);
3105 		timersub(&tv_odd, &tv_even, &tv_delta);
3106 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
3107 			re_initialize();
3108 			goto restart;
3109 		}
3110 		compute_average(EVEN_COUNTERS);
3111 		format_all_counters(EVEN_COUNTERS);
3112 		flush_output_stdout();
3113 		if (exit_requested)
3114 			break;
3115 		if (num_iterations && ++done_iters >= num_iterations)
3116 			break;
3117 		do_sleep();
3118 		if (snapshot_proc_sysfs_files())
3119 			goto restart;
3120 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3121 		if (retval < -1) {
3122 			exit(retval);
3123 		} else if (retval == -1) {
3124 			re_initialize();
3125 			goto restart;
3126 		}
3127 		gettimeofday(&tv_even, (struct timezone *)NULL);
3128 		timersub(&tv_even, &tv_odd, &tv_delta);
3129 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
3130 			re_initialize();
3131 			goto restart;
3132 		}
3133 		compute_average(ODD_COUNTERS);
3134 		format_all_counters(ODD_COUNTERS);
3135 		flush_output_stdout();
3136 		if (exit_requested)
3137 			break;
3138 		if (num_iterations && ++done_iters >= num_iterations)
3139 			break;
3140 	}
3141 }
3142 
3143 void check_dev_msr()
3144 {
3145 	struct stat sb;
3146 	char pathname[32];
3147 
3148 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3149 	if (stat(pathname, &sb))
3150  		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
3151 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
3152 }
3153 
3154 void check_permissions()
3155 {
3156 	struct __user_cap_header_struct cap_header_data;
3157 	cap_user_header_t cap_header = &cap_header_data;
3158 	struct __user_cap_data_struct cap_data_data;
3159 	cap_user_data_t cap_data = &cap_data_data;
3160 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
3161 	int do_exit = 0;
3162 	char pathname[32];
3163 
3164 	/* check for CAP_SYS_RAWIO */
3165 	cap_header->pid = getpid();
3166 	cap_header->version = _LINUX_CAPABILITY_VERSION;
3167 	if (capget(cap_header, cap_data) < 0)
3168 		err(-6, "capget(2) failed");
3169 
3170 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
3171 		do_exit++;
3172 		warnx("capget(CAP_SYS_RAWIO) failed,"
3173 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
3174 	}
3175 
3176 	/* test file permissions */
3177 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3178 	if (euidaccess(pathname, R_OK)) {
3179 		do_exit++;
3180 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
3181 	}
3182 
3183 	/* if all else fails, thell them to be root */
3184 	if (do_exit)
3185 		if (getuid() != 0)
3186 			warnx("... or simply run as root");
3187 
3188 	if (do_exit)
3189 		exit(-6);
3190 }
3191 
3192 /*
3193  * NHM adds support for additional MSRs:
3194  *
3195  * MSR_SMI_COUNT                   0x00000034
3196  *
3197  * MSR_PLATFORM_INFO               0x000000ce
3198  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
3199  *
3200  * MSR_MISC_PWR_MGMT               0x000001aa
3201  *
3202  * MSR_PKG_C3_RESIDENCY            0x000003f8
3203  * MSR_PKG_C6_RESIDENCY            0x000003f9
3204  * MSR_CORE_C3_RESIDENCY           0x000003fc
3205  * MSR_CORE_C6_RESIDENCY           0x000003fd
3206  *
3207  * Side effect:
3208  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
3209  * sets has_misc_feature_control
3210  */
3211 int probe_nhm_msrs(unsigned int family, unsigned int model)
3212 {
3213 	unsigned long long msr;
3214 	unsigned int base_ratio;
3215 	int *pkg_cstate_limits;
3216 
3217 	if (!genuine_intel)
3218 		return 0;
3219 
3220 	if (family != 6)
3221 		return 0;
3222 
3223 	bclk = discover_bclk(family, model);
3224 
3225 	switch (model) {
3226 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
3227 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3228 		pkg_cstate_limits = nhm_pkg_cstate_limits;
3229 		break;
3230 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
3231 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
3232 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3233 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3234 		pkg_cstate_limits = snb_pkg_cstate_limits;
3235 		has_misc_feature_control = 1;
3236 		break;
3237 	case INTEL_FAM6_HASWELL:	/* HSW */
3238 	case INTEL_FAM6_HASWELL_G:	/* HSW */
3239 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3240 	case INTEL_FAM6_HASWELL_L:	/* HSW */
3241 	case INTEL_FAM6_BROADWELL:	/* BDW */
3242 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
3243 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3244 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
3245 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
3246 		pkg_cstate_limits = hsw_pkg_cstate_limits;
3247 		has_misc_feature_control = 1;
3248 		break;
3249 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3250 		pkg_cstate_limits = skx_pkg_cstate_limits;
3251 		has_misc_feature_control = 1;
3252 		break;
3253 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3254 		no_MSR_MISC_PWR_MGMT = 1;
3255 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
3256 		pkg_cstate_limits = slv_pkg_cstate_limits;
3257 		break;
3258 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
3259 		pkg_cstate_limits = amt_pkg_cstate_limits;
3260 		no_MSR_MISC_PWR_MGMT = 1;
3261 		break;
3262 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
3263 		pkg_cstate_limits = phi_pkg_cstate_limits;
3264 		break;
3265 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3266 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3267 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
3268 		pkg_cstate_limits = glm_pkg_cstate_limits;
3269 		break;
3270 	default:
3271 		return 0;
3272 	}
3273 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
3274 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
3275 
3276 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
3277 	base_ratio = (msr >> 8) & 0xFF;
3278 
3279 	base_hz = base_ratio * bclk * 1000000;
3280 	has_base_hz = 1;
3281 	return 1;
3282 }
3283 /*
3284  * SLV client has support for unique MSRs:
3285  *
3286  * MSR_CC6_DEMOTION_POLICY_CONFIG
3287  * MSR_MC6_DEMOTION_POLICY_CONFIG
3288  */
3289 
3290 int has_slv_msrs(unsigned int family, unsigned int model)
3291 {
3292 	if (!genuine_intel)
3293 		return 0;
3294 
3295 	switch (model) {
3296 	case INTEL_FAM6_ATOM_SILVERMONT:
3297 	case INTEL_FAM6_ATOM_SILVERMONT_MID:
3298 	case INTEL_FAM6_ATOM_AIRMONT_MID:
3299 		return 1;
3300 	}
3301 	return 0;
3302 }
3303 int is_dnv(unsigned int family, unsigned int model)
3304 {
3305 
3306 	if (!genuine_intel)
3307 		return 0;
3308 
3309 	switch (model) {
3310 	case INTEL_FAM6_ATOM_GOLDMONT_D:
3311 		return 1;
3312 	}
3313 	return 0;
3314 }
3315 int is_bdx(unsigned int family, unsigned int model)
3316 {
3317 
3318 	if (!genuine_intel)
3319 		return 0;
3320 
3321 	switch (model) {
3322 	case INTEL_FAM6_BROADWELL_X:
3323 		return 1;
3324 	}
3325 	return 0;
3326 }
3327 int is_skx(unsigned int family, unsigned int model)
3328 {
3329 
3330 	if (!genuine_intel)
3331 		return 0;
3332 
3333 	switch (model) {
3334 	case INTEL_FAM6_SKYLAKE_X:
3335 		return 1;
3336 	}
3337 	return 0;
3338 }
3339 
3340 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
3341 {
3342 	if (has_slv_msrs(family, model))
3343 		return 0;
3344 
3345 	switch (model) {
3346 	/* Nehalem compatible, but do not include turbo-ratio limit support */
3347 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3348 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
3349 		return 0;
3350 	default:
3351 		return 1;
3352 	}
3353 }
3354 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
3355 {
3356 	if (has_slv_msrs(family, model))
3357 		return 1;
3358 
3359 	return 0;
3360 }
3361 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
3362 {
3363 	if (!genuine_intel)
3364 		return 0;
3365 
3366 	if (family != 6)
3367 		return 0;
3368 
3369 	switch (model) {
3370 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3371 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3372 		return 1;
3373 	default:
3374 		return 0;
3375 	}
3376 }
3377 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
3378 {
3379 	if (!genuine_intel)
3380 		return 0;
3381 
3382 	if (family != 6)
3383 		return 0;
3384 
3385 	switch (model) {
3386 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3387 		return 1;
3388 	default:
3389 		return 0;
3390 	}
3391 }
3392 
3393 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
3394 {
3395 	if (!genuine_intel)
3396 		return 0;
3397 
3398 	if (family != 6)
3399 		return 0;
3400 
3401 	switch (model) {
3402 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3403 		return 1;
3404 	default:
3405 		return 0;
3406 	}
3407 }
3408 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
3409 {
3410 	if (!genuine_intel)
3411 		return 0;
3412 
3413 	if (family != 6)
3414 		return 0;
3415 
3416 	switch (model) {
3417 	case INTEL_FAM6_ATOM_GOLDMONT:
3418 	case INTEL_FAM6_SKYLAKE_X:
3419 		return 1;
3420 	default:
3421 		return 0;
3422 	}
3423 }
3424 int has_config_tdp(unsigned int family, unsigned int model)
3425 {
3426 	if (!genuine_intel)
3427 		return 0;
3428 
3429 	if (family != 6)
3430 		return 0;
3431 
3432 	switch (model) {
3433 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3434 	case INTEL_FAM6_HASWELL:	/* HSW */
3435 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3436 	case INTEL_FAM6_HASWELL_L:	/* HSW */
3437 	case INTEL_FAM6_HASWELL_G:	/* HSW */
3438 	case INTEL_FAM6_BROADWELL:	/* BDW */
3439 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
3440 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3441 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
3442 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
3443 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3444 
3445 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3446 		return 1;
3447 	default:
3448 		return 0;
3449 	}
3450 }
3451 
3452 static void
3453 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
3454 {
3455 	if (!do_nhm_platform_info)
3456 		return;
3457 
3458 	dump_nhm_platform_info();
3459 
3460 	if (has_hsw_turbo_ratio_limit(family, model))
3461 		dump_hsw_turbo_ratio_limits();
3462 
3463 	if (has_ivt_turbo_ratio_limit(family, model))
3464 		dump_ivt_turbo_ratio_limits();
3465 
3466 	if (has_turbo_ratio_limit(family, model))
3467 		dump_turbo_ratio_limits(family, model);
3468 
3469 	if (has_atom_turbo_ratio_limit(family, model))
3470 		dump_atom_turbo_ratio_limits();
3471 
3472 	if (has_knl_turbo_ratio_limit(family, model))
3473 		dump_knl_turbo_ratio_limits();
3474 
3475 	if (has_config_tdp(family, model))
3476 		dump_config_tdp();
3477 
3478 	dump_nhm_cst_cfg();
3479 }
3480 
3481 static void
3482 dump_sysfs_cstate_config(void)
3483 {
3484 	char path[64];
3485 	char name_buf[16];
3486 	char desc[64];
3487 	FILE *input;
3488 	int state;
3489 	char *sp;
3490 
3491 	if (!DO_BIC(BIC_sysfs))
3492 		return;
3493 
3494 	for (state = 0; state < 10; ++state) {
3495 
3496 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
3497 			base_cpu, state);
3498 		input = fopen(path, "r");
3499 		if (input == NULL)
3500 			continue;
3501 		if (!fgets(name_buf, sizeof(name_buf), input))
3502 			err(1, "%s: failed to read file", path);
3503 
3504 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
3505 		sp = strchr(name_buf, '-');
3506 		if (!sp)
3507 			sp = strchrnul(name_buf, '\n');
3508 		*sp = '\0';
3509 		fclose(input);
3510 
3511 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc",
3512 			base_cpu, state);
3513 		input = fopen(path, "r");
3514 		if (input == NULL)
3515 			continue;
3516 		if (!fgets(desc, sizeof(desc), input))
3517 			err(1, "%s: failed to read file", path);
3518 
3519 		fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
3520 		fclose(input);
3521 	}
3522 }
3523 static void
3524 dump_sysfs_pstate_config(void)
3525 {
3526 	char path[64];
3527 	char driver_buf[64];
3528 	char governor_buf[64];
3529 	FILE *input;
3530 	int turbo;
3531 
3532 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver",
3533 			base_cpu);
3534 	input = fopen(path, "r");
3535 	if (input == NULL) {
3536 		fprintf(outf, "NSFOD %s\n", path);
3537 		return;
3538 	}
3539 	if (!fgets(driver_buf, sizeof(driver_buf), input))
3540 		err(1, "%s: failed to read file", path);
3541 	fclose(input);
3542 
3543 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor",
3544 			base_cpu);
3545 	input = fopen(path, "r");
3546 	if (input == NULL) {
3547 		fprintf(outf, "NSFOD %s\n", path);
3548 		return;
3549 	}
3550 	if (!fgets(governor_buf, sizeof(governor_buf), input))
3551 		err(1, "%s: failed to read file", path);
3552 	fclose(input);
3553 
3554 	fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
3555 	fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
3556 
3557 	sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
3558 	input = fopen(path, "r");
3559 	if (input != NULL) {
3560 		if (fscanf(input, "%d", &turbo) != 1)
3561 			err(1, "%s: failed to parse number from file", path);
3562 		fprintf(outf, "cpufreq boost: %d\n", turbo);
3563 		fclose(input);
3564 	}
3565 
3566 	sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
3567 	input = fopen(path, "r");
3568 	if (input != NULL) {
3569 		if (fscanf(input, "%d", &turbo) != 1)
3570 			err(1, "%s: failed to parse number from file", path);
3571 		fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
3572 		fclose(input);
3573 	}
3574 }
3575 
3576 
3577 /*
3578  * print_epb()
3579  * Decode the ENERGY_PERF_BIAS MSR
3580  */
3581 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3582 {
3583 	unsigned long long msr;
3584 	char *epb_string;
3585 	int cpu;
3586 
3587 	if (!has_epb)
3588 		return 0;
3589 
3590 	cpu = t->cpu_id;
3591 
3592 	/* EPB is per-package */
3593 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3594 		return 0;
3595 
3596 	if (cpu_migrate(cpu)) {
3597 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3598 		return -1;
3599 	}
3600 
3601 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
3602 		return 0;
3603 
3604 	switch (msr & 0xF) {
3605 	case ENERGY_PERF_BIAS_PERFORMANCE:
3606 		epb_string = "performance";
3607 		break;
3608 	case ENERGY_PERF_BIAS_NORMAL:
3609 		epb_string = "balanced";
3610 		break;
3611 	case ENERGY_PERF_BIAS_POWERSAVE:
3612 		epb_string = "powersave";
3613 		break;
3614 	default:
3615 		epb_string = "custom";
3616 		break;
3617 	}
3618 	fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
3619 
3620 	return 0;
3621 }
3622 /*
3623  * print_hwp()
3624  * Decode the MSR_HWP_CAPABILITIES
3625  */
3626 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3627 {
3628 	unsigned long long msr;
3629 	int cpu;
3630 
3631 	if (!has_hwp)
3632 		return 0;
3633 
3634 	cpu = t->cpu_id;
3635 
3636 	/* MSR_HWP_CAPABILITIES is per-package */
3637 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3638 		return 0;
3639 
3640 	if (cpu_migrate(cpu)) {
3641 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3642 		return -1;
3643 	}
3644 
3645 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
3646 		return 0;
3647 
3648 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
3649 		cpu, msr, (msr & (1 << 0)) ? "" : "No-");
3650 
3651 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
3652 	if ((msr & (1 << 0)) == 0)
3653 		return 0;
3654 
3655 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
3656 		return 0;
3657 
3658 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
3659 			"(high %d guar %d eff %d low %d)\n",
3660 			cpu, msr,
3661 			(unsigned int)HWP_HIGHEST_PERF(msr),
3662 			(unsigned int)HWP_GUARANTEED_PERF(msr),
3663 			(unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
3664 			(unsigned int)HWP_LOWEST_PERF(msr));
3665 
3666 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
3667 		return 0;
3668 
3669 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
3670 			"(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
3671 			cpu, msr,
3672 			(unsigned int)(((msr) >> 0) & 0xff),
3673 			(unsigned int)(((msr) >> 8) & 0xff),
3674 			(unsigned int)(((msr) >> 16) & 0xff),
3675 			(unsigned int)(((msr) >> 24) & 0xff),
3676 			(unsigned int)(((msr) >> 32) & 0xff3),
3677 			(unsigned int)(((msr) >> 42) & 0x1));
3678 
3679 	if (has_hwp_pkg) {
3680 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
3681 			return 0;
3682 
3683 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
3684 			"(min %d max %d des %d epp 0x%x window 0x%x)\n",
3685 			cpu, msr,
3686 			(unsigned int)(((msr) >> 0) & 0xff),
3687 			(unsigned int)(((msr) >> 8) & 0xff),
3688 			(unsigned int)(((msr) >> 16) & 0xff),
3689 			(unsigned int)(((msr) >> 24) & 0xff),
3690 			(unsigned int)(((msr) >> 32) & 0xff3));
3691 	}
3692 	if (has_hwp_notify) {
3693 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
3694 			return 0;
3695 
3696 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
3697 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
3698 			cpu, msr,
3699 			((msr) & 0x1) ? "EN" : "Dis",
3700 			((msr) & 0x2) ? "EN" : "Dis");
3701 	}
3702 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
3703 		return 0;
3704 
3705 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
3706 			"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
3707 			cpu, msr,
3708 			((msr) & 0x1) ? "" : "No-",
3709 			((msr) & 0x2) ? "" : "No-");
3710 
3711 	return 0;
3712 }
3713 
3714 /*
3715  * print_perf_limit()
3716  */
3717 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3718 {
3719 	unsigned long long msr;
3720 	int cpu;
3721 
3722 	cpu = t->cpu_id;
3723 
3724 	/* per-package */
3725 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3726 		return 0;
3727 
3728 	if (cpu_migrate(cpu)) {
3729 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3730 		return -1;
3731 	}
3732 
3733 	if (do_core_perf_limit_reasons) {
3734 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
3735 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3736 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
3737 			(msr & 1 << 15) ? "bit15, " : "",
3738 			(msr & 1 << 14) ? "bit14, " : "",
3739 			(msr & 1 << 13) ? "Transitions, " : "",
3740 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
3741 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
3742 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3743 			(msr & 1 << 9) ? "CorePwr, " : "",
3744 			(msr & 1 << 8) ? "Amps, " : "",
3745 			(msr & 1 << 6) ? "VR-Therm, " : "",
3746 			(msr & 1 << 5) ? "Auto-HWP, " : "",
3747 			(msr & 1 << 4) ? "Graphics, " : "",
3748 			(msr & 1 << 2) ? "bit2, " : "",
3749 			(msr & 1 << 1) ? "ThermStatus, " : "",
3750 			(msr & 1 << 0) ? "PROCHOT, " : "");
3751 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
3752 			(msr & 1 << 31) ? "bit31, " : "",
3753 			(msr & 1 << 30) ? "bit30, " : "",
3754 			(msr & 1 << 29) ? "Transitions, " : "",
3755 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
3756 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
3757 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3758 			(msr & 1 << 25) ? "CorePwr, " : "",
3759 			(msr & 1 << 24) ? "Amps, " : "",
3760 			(msr & 1 << 22) ? "VR-Therm, " : "",
3761 			(msr & 1 << 21) ? "Auto-HWP, " : "",
3762 			(msr & 1 << 20) ? "Graphics, " : "",
3763 			(msr & 1 << 18) ? "bit18, " : "",
3764 			(msr & 1 << 17) ? "ThermStatus, " : "",
3765 			(msr & 1 << 16) ? "PROCHOT, " : "");
3766 
3767 	}
3768 	if (do_gfx_perf_limit_reasons) {
3769 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
3770 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3771 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
3772 			(msr & 1 << 0) ? "PROCHOT, " : "",
3773 			(msr & 1 << 1) ? "ThermStatus, " : "",
3774 			(msr & 1 << 4) ? "Graphics, " : "",
3775 			(msr & 1 << 6) ? "VR-Therm, " : "",
3776 			(msr & 1 << 8) ? "Amps, " : "",
3777 			(msr & 1 << 9) ? "GFXPwr, " : "",
3778 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3779 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3780 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
3781 			(msr & 1 << 16) ? "PROCHOT, " : "",
3782 			(msr & 1 << 17) ? "ThermStatus, " : "",
3783 			(msr & 1 << 20) ? "Graphics, " : "",
3784 			(msr & 1 << 22) ? "VR-Therm, " : "",
3785 			(msr & 1 << 24) ? "Amps, " : "",
3786 			(msr & 1 << 25) ? "GFXPwr, " : "",
3787 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3788 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3789 	}
3790 	if (do_ring_perf_limit_reasons) {
3791 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
3792 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3793 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
3794 			(msr & 1 << 0) ? "PROCHOT, " : "",
3795 			(msr & 1 << 1) ? "ThermStatus, " : "",
3796 			(msr & 1 << 6) ? "VR-Therm, " : "",
3797 			(msr & 1 << 8) ? "Amps, " : "",
3798 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3799 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3800 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3801 			(msr & 1 << 16) ? "PROCHOT, " : "",
3802 			(msr & 1 << 17) ? "ThermStatus, " : "",
3803 			(msr & 1 << 22) ? "VR-Therm, " : "",
3804 			(msr & 1 << 24) ? "Amps, " : "",
3805 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3806 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3807 	}
3808 	return 0;
3809 }
3810 
3811 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
3812 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
3813 
3814 double get_tdp_intel(unsigned int model)
3815 {
3816 	unsigned long long msr;
3817 
3818 	if (do_rapl & RAPL_PKG_POWER_INFO)
3819 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3820 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3821 
3822 	switch (model) {
3823 	case INTEL_FAM6_ATOM_SILVERMONT:
3824 	case INTEL_FAM6_ATOM_SILVERMONT_D:
3825 		return 30.0;
3826 	default:
3827 		return 135.0;
3828 	}
3829 }
3830 
3831 double get_tdp_amd(unsigned int family)
3832 {
3833 	switch (family) {
3834 	case 0x17:
3835 	case 0x18:
3836 	default:
3837 		/* This is the max stock TDP of HEDT/Server Fam17h chips */
3838 		return 250.0;
3839 	}
3840 }
3841 
3842 /*
3843  * rapl_dram_energy_units_probe()
3844  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3845  */
3846 static double
3847 rapl_dram_energy_units_probe(int  model, double rapl_energy_units)
3848 {
3849 	/* only called for genuine_intel, family 6 */
3850 
3851 	switch (model) {
3852 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3853 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3854 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3855 		return (rapl_dram_energy_units = 15.3 / 1000000);
3856 	default:
3857 		return (rapl_energy_units);
3858 	}
3859 }
3860 
3861 void rapl_probe_intel(unsigned int family, unsigned int model)
3862 {
3863 	unsigned long long msr;
3864 	unsigned int time_unit;
3865 	double tdp;
3866 
3867 	if (family != 6)
3868 		return;
3869 
3870 	switch (model) {
3871 	case INTEL_FAM6_SANDYBRIDGE:
3872 	case INTEL_FAM6_IVYBRIDGE:
3873 	case INTEL_FAM6_HASWELL:	/* HSW */
3874 	case INTEL_FAM6_HASWELL_L:	/* HSW */
3875 	case INTEL_FAM6_HASWELL_G:	/* HSW */
3876 	case INTEL_FAM6_BROADWELL:	/* BDW */
3877 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
3878 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3879 		if (rapl_joules) {
3880 			BIC_PRESENT(BIC_Pkg_J);
3881 			BIC_PRESENT(BIC_Cor_J);
3882 			BIC_PRESENT(BIC_GFX_J);
3883 		} else {
3884 			BIC_PRESENT(BIC_PkgWatt);
3885 			BIC_PRESENT(BIC_CorWatt);
3886 			BIC_PRESENT(BIC_GFXWatt);
3887 		}
3888 		break;
3889 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3890 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3891 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3892 		if (rapl_joules)
3893 			BIC_PRESENT(BIC_Pkg_J);
3894 		else
3895 			BIC_PRESENT(BIC_PkgWatt);
3896 		break;
3897 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
3898 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
3899 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_GFX | RAPL_PKG_POWER_INFO;
3900 		BIC_PRESENT(BIC_PKG__);
3901 		BIC_PRESENT(BIC_RAM__);
3902 		if (rapl_joules) {
3903 			BIC_PRESENT(BIC_Pkg_J);
3904 			BIC_PRESENT(BIC_Cor_J);
3905 			BIC_PRESENT(BIC_RAM_J);
3906 			BIC_PRESENT(BIC_GFX_J);
3907 		} else {
3908 			BIC_PRESENT(BIC_PkgWatt);
3909 			BIC_PRESENT(BIC_CorWatt);
3910 			BIC_PRESENT(BIC_RAMWatt);
3911 			BIC_PRESENT(BIC_GFXWatt);
3912 		}
3913 		break;
3914 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3915 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3916 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3917 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3918 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3919 		BIC_PRESENT(BIC_PKG__);
3920 		BIC_PRESENT(BIC_RAM__);
3921 		if (rapl_joules) {
3922 			BIC_PRESENT(BIC_Pkg_J);
3923 			BIC_PRESENT(BIC_RAM_J);
3924 		} else {
3925 			BIC_PRESENT(BIC_PkgWatt);
3926 			BIC_PRESENT(BIC_RAMWatt);
3927 		}
3928 		break;
3929 	case INTEL_FAM6_SANDYBRIDGE_X:
3930 	case INTEL_FAM6_IVYBRIDGE_X:
3931 		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;
3932 		BIC_PRESENT(BIC_PKG__);
3933 		BIC_PRESENT(BIC_RAM__);
3934 		if (rapl_joules) {
3935 			BIC_PRESENT(BIC_Pkg_J);
3936 			BIC_PRESENT(BIC_Cor_J);
3937 			BIC_PRESENT(BIC_RAM_J);
3938 		} else {
3939 			BIC_PRESENT(BIC_PkgWatt);
3940 			BIC_PRESENT(BIC_CorWatt);
3941 			BIC_PRESENT(BIC_RAMWatt);
3942 		}
3943 		break;
3944 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3945 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
3946 		do_rapl = RAPL_PKG | RAPL_CORES;
3947 		if (rapl_joules) {
3948 			BIC_PRESENT(BIC_Pkg_J);
3949 			BIC_PRESENT(BIC_Cor_J);
3950 		} else {
3951 			BIC_PRESENT(BIC_PkgWatt);
3952 			BIC_PRESENT(BIC_CorWatt);
3953 		}
3954 		break;
3955 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
3956 		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;
3957 		BIC_PRESENT(BIC_PKG__);
3958 		BIC_PRESENT(BIC_RAM__);
3959 		if (rapl_joules) {
3960 			BIC_PRESENT(BIC_Pkg_J);
3961 			BIC_PRESENT(BIC_Cor_J);
3962 			BIC_PRESENT(BIC_RAM_J);
3963 		} else {
3964 			BIC_PRESENT(BIC_PkgWatt);
3965 			BIC_PRESENT(BIC_CorWatt);
3966 			BIC_PRESENT(BIC_RAMWatt);
3967 		}
3968 		break;
3969 	default:
3970 		return;
3971 	}
3972 
3973 	/* units on package 0, verify later other packages match */
3974 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3975 		return;
3976 
3977 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
3978 	if (model == INTEL_FAM6_ATOM_SILVERMONT)
3979 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3980 	else
3981 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3982 
3983 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3984 
3985 	time_unit = msr >> 16 & 0xF;
3986 	if (time_unit == 0)
3987 		time_unit = 0xA;
3988 
3989 	rapl_time_units = 1.0 / (1 << (time_unit));
3990 
3991 	tdp = get_tdp_intel(model);
3992 
3993 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3994 	if (!quiet)
3995 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3996 }
3997 
3998 void rapl_probe_amd(unsigned int family, unsigned int model)
3999 {
4000 	unsigned long long msr;
4001 	unsigned int eax, ebx, ecx, edx;
4002 	unsigned int has_rapl = 0;
4003 	double tdp;
4004 
4005 	if (max_extended_level >= 0x80000007) {
4006 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4007 		/* RAPL (Fam 17h) */
4008 		has_rapl = edx & (1 << 14);
4009 	}
4010 
4011 	if (!has_rapl)
4012 		return;
4013 
4014 	switch (family) {
4015 	case 0x17: /* Zen, Zen+ */
4016 	case 0x18: /* Hygon Dhyana */
4017 		do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY;
4018 		if (rapl_joules) {
4019 			BIC_PRESENT(BIC_Pkg_J);
4020 			BIC_PRESENT(BIC_Cor_J);
4021 		} else {
4022 			BIC_PRESENT(BIC_PkgWatt);
4023 			BIC_PRESENT(BIC_CorWatt);
4024 		}
4025 		break;
4026 	default:
4027 		return;
4028 	}
4029 
4030 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
4031 		return;
4032 
4033 	rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf));
4034 	rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f));
4035 	rapl_power_units = ldexp(1.0, -(msr & 0xf));
4036 
4037 	tdp = get_tdp_amd(family);
4038 
4039 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4040 	if (!quiet)
4041 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4042 }
4043 
4044 /*
4045  * rapl_probe()
4046  *
4047  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
4048  */
4049 void rapl_probe(unsigned int family, unsigned int model)
4050 {
4051 	if (genuine_intel)
4052 		rapl_probe_intel(family, model);
4053 	if (authentic_amd || hygon_genuine)
4054 		rapl_probe_amd(family, model);
4055 }
4056 
4057 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
4058 {
4059 	if (!genuine_intel)
4060 		return;
4061 
4062 	if (family != 6)
4063 		return;
4064 
4065 	switch (model) {
4066 	case INTEL_FAM6_HASWELL:	/* HSW */
4067 	case INTEL_FAM6_HASWELL_L:	/* HSW */
4068 	case INTEL_FAM6_HASWELL_G:	/* HSW */
4069 		do_gfx_perf_limit_reasons = 1;
4070 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4071 		do_core_perf_limit_reasons = 1;
4072 		do_ring_perf_limit_reasons = 1;
4073 	default:
4074 		return;
4075 	}
4076 }
4077 
4078 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model)
4079 {
4080 	if (is_skx(family, model) || is_bdx(family, model))
4081 		has_automatic_cstate_conversion = 1;
4082 }
4083 
4084 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4085 {
4086 	unsigned long long msr;
4087 	unsigned int dts, dts2;
4088 	int cpu;
4089 
4090 	if (!(do_dts || do_ptm))
4091 		return 0;
4092 
4093 	cpu = t->cpu_id;
4094 
4095 	/* DTS is per-core, no need to print for each thread */
4096 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
4097 		return 0;
4098 
4099 	if (cpu_migrate(cpu)) {
4100 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4101 		return -1;
4102 	}
4103 
4104 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
4105 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
4106 			return 0;
4107 
4108 		dts = (msr >> 16) & 0x7F;
4109 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
4110 			cpu, msr, tcc_activation_temp - dts);
4111 
4112 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
4113 			return 0;
4114 
4115 		dts = (msr >> 16) & 0x7F;
4116 		dts2 = (msr >> 8) & 0x7F;
4117 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4118 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4119 	}
4120 
4121 
4122 	if (do_dts && debug) {
4123 		unsigned int resolution;
4124 
4125 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
4126 			return 0;
4127 
4128 		dts = (msr >> 16) & 0x7F;
4129 		resolution = (msr >> 27) & 0xF;
4130 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
4131 			cpu, msr, tcc_activation_temp - dts, resolution);
4132 
4133 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
4134 			return 0;
4135 
4136 		dts = (msr >> 16) & 0x7F;
4137 		dts2 = (msr >> 8) & 0x7F;
4138 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4139 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4140 	}
4141 
4142 	return 0;
4143 }
4144 
4145 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
4146 {
4147 	fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
4148 		cpu, label,
4149 		((msr >> 15) & 1) ? "EN" : "DIS",
4150 		((msr >> 0) & 0x7FFF) * rapl_power_units,
4151 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
4152 		(((msr >> 16) & 1) ? "EN" : "DIS"));
4153 
4154 	return;
4155 }
4156 
4157 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4158 {
4159 	unsigned long long msr;
4160 	const char *msr_name;
4161 	int cpu;
4162 
4163 	if (!do_rapl)
4164 		return 0;
4165 
4166 	/* RAPL counters are per package, so print only for 1st thread/package */
4167 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4168 		return 0;
4169 
4170 	cpu = t->cpu_id;
4171 	if (cpu_migrate(cpu)) {
4172 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4173 		return -1;
4174 	}
4175 
4176 	if (do_rapl & RAPL_AMD_F17H) {
4177 		msr_name = "MSR_RAPL_PWR_UNIT";
4178 		if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr))
4179 			return -1;
4180 	} else {
4181 		msr_name = "MSR_RAPL_POWER_UNIT";
4182 		if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
4183 			return -1;
4184 	}
4185 
4186 	fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr,
4187 		rapl_power_units, rapl_energy_units, rapl_time_units);
4188 
4189 	if (do_rapl & RAPL_PKG_POWER_INFO) {
4190 
4191 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
4192                 	return -5;
4193 
4194 
4195 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4196 			cpu, msr,
4197 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4198 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4199 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4200 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4201 
4202 	}
4203 	if (do_rapl & RAPL_PKG) {
4204 
4205 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
4206 			return -9;
4207 
4208 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
4209 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
4210 
4211 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
4212 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
4213 			cpu,
4214 			((msr >> 47) & 1) ? "EN" : "DIS",
4215 			((msr >> 32) & 0x7FFF) * rapl_power_units,
4216 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
4217 			((msr >> 48) & 1) ? "EN" : "DIS");
4218 	}
4219 
4220 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
4221 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
4222                 	return -6;
4223 
4224 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4225 			cpu, msr,
4226 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4227 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4228 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4229 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4230 	}
4231 	if (do_rapl & RAPL_DRAM) {
4232 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
4233 			return -9;
4234 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
4235 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4236 
4237 		print_power_limit_msr(cpu, msr, "DRAM Limit");
4238 	}
4239 	if (do_rapl & RAPL_CORE_POLICY) {
4240 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
4241 			return -7;
4242 
4243 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
4244 	}
4245 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
4246 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
4247 			return -9;
4248 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
4249 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4250 		print_power_limit_msr(cpu, msr, "Cores Limit");
4251 	}
4252 	if (do_rapl & RAPL_GFX) {
4253 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
4254 			return -8;
4255 
4256 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
4257 
4258 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
4259 			return -9;
4260 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
4261 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4262 		print_power_limit_msr(cpu, msr, "GFX Limit");
4263 	}
4264 	return 0;
4265 }
4266 
4267 /*
4268  * SNB adds support for additional MSRs:
4269  *
4270  * MSR_PKG_C7_RESIDENCY            0x000003fa
4271  * MSR_CORE_C7_RESIDENCY           0x000003fe
4272  * MSR_PKG_C2_RESIDENCY            0x0000060d
4273  */
4274 
4275 int has_snb_msrs(unsigned int family, unsigned int model)
4276 {
4277 	if (!genuine_intel)
4278 		return 0;
4279 
4280 	switch (model) {
4281 	case INTEL_FAM6_SANDYBRIDGE:
4282 	case INTEL_FAM6_SANDYBRIDGE_X:
4283 	case INTEL_FAM6_IVYBRIDGE:		/* IVB */
4284 	case INTEL_FAM6_IVYBRIDGE_X:		/* IVB Xeon */
4285 	case INTEL_FAM6_HASWELL:		/* HSW */
4286 	case INTEL_FAM6_HASWELL_X:		/* HSW */
4287 	case INTEL_FAM6_HASWELL_L:		/* HSW */
4288 	case INTEL_FAM6_HASWELL_G:		/* HSW */
4289 	case INTEL_FAM6_BROADWELL:		/* BDW */
4290 	case INTEL_FAM6_BROADWELL_G:		/* BDW */
4291 	case INTEL_FAM6_BROADWELL_X:		/* BDX */
4292 	case INTEL_FAM6_SKYLAKE_L:		/* SKL */
4293 	case INTEL_FAM6_CANNONLAKE_L:		/* CNL */
4294 	case INTEL_FAM6_SKYLAKE_X:		/* SKX */
4295 	case INTEL_FAM6_ATOM_GOLDMONT:		/* BXT */
4296 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4297 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
4298 		return 1;
4299 	}
4300 	return 0;
4301 }
4302 
4303 /*
4304  * HSW ULT added support for C8/C9/C10 MSRs:
4305  *
4306  * MSR_PKG_C8_RESIDENCY		0x00000630
4307  * MSR_PKG_C9_RESIDENCY		0x00000631
4308  * MSR_PKG_C10_RESIDENCY	0x00000632
4309  *
4310  * MSR_PKGC8_IRTL		0x00000633
4311  * MSR_PKGC9_IRTL		0x00000634
4312  * MSR_PKGC10_IRTL		0x00000635
4313  *
4314  */
4315 int has_c8910_msrs(unsigned int family, unsigned int model)
4316 {
4317 	if (!genuine_intel)
4318 		return 0;
4319 
4320 	switch (model) {
4321 	case INTEL_FAM6_HASWELL_L:	/* HSW */
4322 	case INTEL_FAM6_BROADWELL:	/* BDW */
4323 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
4324 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
4325 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4326 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4327 		return 1;
4328 	}
4329 	return 0;
4330 }
4331 
4332 /*
4333  * SKL adds support for additional MSRS:
4334  *
4335  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
4336  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
4337  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
4338  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
4339  */
4340 int has_skl_msrs(unsigned int family, unsigned int model)
4341 {
4342 	if (!genuine_intel)
4343 		return 0;
4344 
4345 	switch (model) {
4346 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
4347 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
4348 		return 1;
4349 	}
4350 	return 0;
4351 }
4352 
4353 int is_slm(unsigned int family, unsigned int model)
4354 {
4355 	if (!genuine_intel)
4356 		return 0;
4357 	switch (model) {
4358 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
4359 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
4360 		return 1;
4361 	}
4362 	return 0;
4363 }
4364 
4365 int is_knl(unsigned int family, unsigned int model)
4366 {
4367 	if (!genuine_intel)
4368 		return 0;
4369 	switch (model) {
4370 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4371 		return 1;
4372 	}
4373 	return 0;
4374 }
4375 
4376 int is_cnl(unsigned int family, unsigned int model)
4377 {
4378 	if (!genuine_intel)
4379 		return 0;
4380 
4381 	switch (model) {
4382 	case INTEL_FAM6_CANNONLAKE_L: /* CNL */
4383 		return 1;
4384 	}
4385 
4386 	return 0;
4387 }
4388 
4389 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
4390 {
4391 	if (is_knl(family, model))
4392 		return 1024;
4393 	return 1;
4394 }
4395 
4396 #define SLM_BCLK_FREQS 5
4397 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
4398 
4399 double slm_bclk(void)
4400 {
4401 	unsigned long long msr = 3;
4402 	unsigned int i;
4403 	double freq;
4404 
4405 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
4406 		fprintf(outf, "SLM BCLK: unknown\n");
4407 
4408 	i = msr & 0xf;
4409 	if (i >= SLM_BCLK_FREQS) {
4410 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
4411 		i = 3;
4412 	}
4413 	freq = slm_freq_table[i];
4414 
4415 	if (!quiet)
4416 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
4417 
4418 	return freq;
4419 }
4420 
4421 double discover_bclk(unsigned int family, unsigned int model)
4422 {
4423 	if (has_snb_msrs(family, model) || is_knl(family, model))
4424 		return 100.00;
4425 	else if (is_slm(family, model))
4426 		return slm_bclk();
4427 	else
4428 		return 133.33;
4429 }
4430 
4431 /*
4432  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
4433  * the Thermal Control Circuit (TCC) activates.
4434  * This is usually equal to tjMax.
4435  *
4436  * Older processors do not have this MSR, so there we guess,
4437  * but also allow cmdline over-ride with -T.
4438  *
4439  * Several MSR temperature values are in units of degrees-C
4440  * below this value, including the Digital Thermal Sensor (DTS),
4441  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
4442  */
4443 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4444 {
4445 	unsigned long long msr;
4446 	unsigned int target_c_local;
4447 	int cpu;
4448 
4449 	/* tcc_activation_temp is used only for dts or ptm */
4450 	if (!(do_dts || do_ptm))
4451 		return 0;
4452 
4453 	/* this is a per-package concept */
4454 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4455 		return 0;
4456 
4457 	cpu = t->cpu_id;
4458 	if (cpu_migrate(cpu)) {
4459 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4460 		return -1;
4461 	}
4462 
4463 	if (tcc_activation_temp_override != 0) {
4464 		tcc_activation_temp = tcc_activation_temp_override;
4465 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
4466 			cpu, tcc_activation_temp);
4467 		return 0;
4468 	}
4469 
4470 	/* Temperature Target MSR is Nehalem and newer only */
4471 	if (!do_nhm_platform_info)
4472 		goto guess;
4473 
4474 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
4475 		goto guess;
4476 
4477 	target_c_local = (msr >> 16) & 0xFF;
4478 
4479 	if (!quiet)
4480 		fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
4481 			cpu, msr, target_c_local);
4482 
4483 	if (!target_c_local)
4484 		goto guess;
4485 
4486 	tcc_activation_temp = target_c_local;
4487 
4488 	return 0;
4489 
4490 guess:
4491 	tcc_activation_temp = TJMAX_DEFAULT;
4492 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
4493 		cpu, tcc_activation_temp);
4494 
4495 	return 0;
4496 }
4497 
4498 void decode_feature_control_msr(void)
4499 {
4500 	unsigned long long msr;
4501 
4502 	if (!get_msr(base_cpu, MSR_IA32_FEAT_CTL, &msr))
4503 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
4504 			base_cpu, msr,
4505 			msr & FEAT_CTL_LOCKED ? "" : "UN-",
4506 			msr & (1 << 18) ? "SGX" : "");
4507 }
4508 
4509 void decode_misc_enable_msr(void)
4510 {
4511 	unsigned long long msr;
4512 
4513 	if (!genuine_intel)
4514 		return;
4515 
4516 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
4517 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
4518 			base_cpu, msr,
4519 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
4520 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
4521 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-",
4522 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
4523 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
4524 }
4525 
4526 void decode_misc_feature_control(void)
4527 {
4528 	unsigned long long msr;
4529 
4530 	if (!has_misc_feature_control)
4531 		return;
4532 
4533 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
4534 		fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
4535 			base_cpu, msr,
4536 			msr & (0 << 0) ? "No-" : "",
4537 			msr & (1 << 0) ? "No-" : "",
4538 			msr & (2 << 0) ? "No-" : "",
4539 			msr & (3 << 0) ? "No-" : "");
4540 }
4541 /*
4542  * Decode MSR_MISC_PWR_MGMT
4543  *
4544  * Decode the bits according to the Nehalem documentation
4545  * bit[0] seems to continue to have same meaning going forward
4546  * bit[1] less so...
4547  */
4548 void decode_misc_pwr_mgmt_msr(void)
4549 {
4550 	unsigned long long msr;
4551 
4552 	if (!do_nhm_platform_info)
4553 		return;
4554 
4555 	if (no_MSR_MISC_PWR_MGMT)
4556 		return;
4557 
4558 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
4559 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
4560 			base_cpu, msr,
4561 			msr & (1 << 0) ? "DIS" : "EN",
4562 			msr & (1 << 1) ? "EN" : "DIS",
4563 			msr & (1 << 8) ? "EN" : "DIS");
4564 }
4565 /*
4566  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
4567  *
4568  * This MSRs are present on Silvermont processors,
4569  * Intel Atom processor E3000 series (Baytrail), and friends.
4570  */
4571 void decode_c6_demotion_policy_msr(void)
4572 {
4573 	unsigned long long msr;
4574 
4575 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
4576 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
4577 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4578 
4579 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
4580 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
4581 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4582 }
4583 
4584 /*
4585  * When models are the same, for the purpose of turbostat, reuse
4586  */
4587 unsigned int intel_model_duplicates(unsigned int model)
4588 {
4589 
4590 	switch(model) {
4591 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
4592 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
4593 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
4594 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
4595 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
4596 		return INTEL_FAM6_NEHALEM;
4597 
4598 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
4599 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
4600 		return INTEL_FAM6_NEHALEM_EX;
4601 
4602 	case INTEL_FAM6_XEON_PHI_KNM:
4603 		return INTEL_FAM6_XEON_PHI_KNL;
4604 
4605 	case INTEL_FAM6_BROADWELL_X:
4606 	case INTEL_FAM6_BROADWELL_D:	/* BDX-DE */
4607 		return INTEL_FAM6_BROADWELL_X;
4608 
4609 	case INTEL_FAM6_SKYLAKE_L:
4610 	case INTEL_FAM6_SKYLAKE:
4611 	case INTEL_FAM6_KABYLAKE_L:
4612 	case INTEL_FAM6_KABYLAKE:
4613 		return INTEL_FAM6_SKYLAKE_L;
4614 
4615 	case INTEL_FAM6_ICELAKE_L:
4616 	case INTEL_FAM6_ICELAKE_NNPI:
4617 		return INTEL_FAM6_CANNONLAKE_L;
4618 
4619 	case INTEL_FAM6_ATOM_TREMONT_D:
4620 		return INTEL_FAM6_ATOM_GOLDMONT_D;
4621 	}
4622 	return model;
4623 }
4624 void process_cpuid()
4625 {
4626 	unsigned int eax, ebx, ecx, edx;
4627 	unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
4628 	unsigned int has_turbo;
4629 
4630 	eax = ebx = ecx = edx = 0;
4631 
4632 	__cpuid(0, max_level, ebx, ecx, edx);
4633 
4634 	if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
4635 		genuine_intel = 1;
4636 	else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
4637 		authentic_amd = 1;
4638 	else if (ebx == 0x6f677948 && ecx == 0x656e6975 && edx == 0x6e65476e)
4639 		hygon_genuine = 1;
4640 
4641 	if (!quiet)
4642 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
4643 			(char *)&ebx, (char *)&edx, (char *)&ecx);
4644 
4645 	__cpuid(1, fms, ebx, ecx, edx);
4646 	family = (fms >> 8) & 0xf;
4647 	model = (fms >> 4) & 0xf;
4648 	stepping = fms & 0xf;
4649 	if (family == 0xf)
4650 		family += (fms >> 20) & 0xff;
4651 	if (family >= 6)
4652 		model += ((fms >> 16) & 0xf) << 4;
4653 	ecx_flags = ecx;
4654 	edx_flags = edx;
4655 
4656 	/*
4657 	 * check max extended function levels of CPUID.
4658 	 * This is needed to check for invariant TSC.
4659 	 * This check is valid for both Intel and AMD.
4660 	 */
4661 	ebx = ecx = edx = 0;
4662 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
4663 
4664 	if (!quiet) {
4665 		fprintf(outf, "0x%x CPUID levels; 0x%x xlevels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
4666 			max_level, max_extended_level, family, model, stepping, family, model, stepping);
4667 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n",
4668 			ecx_flags & (1 << 0) ? "SSE3" : "-",
4669 			ecx_flags & (1 << 3) ? "MONITOR" : "-",
4670 			ecx_flags & (1 << 6) ? "SMX" : "-",
4671 			ecx_flags & (1 << 7) ? "EIST" : "-",
4672 			ecx_flags & (1 << 8) ? "TM2" : "-",
4673 			edx_flags & (1 << 4) ? "TSC" : "-",
4674 			edx_flags & (1 << 5) ? "MSR" : "-",
4675 			edx_flags & (1 << 22) ? "ACPI-TM" : "-",
4676 			edx_flags & (1 << 28) ? "HT" : "-",
4677 			edx_flags & (1 << 29) ? "TM" : "-");
4678 	}
4679 	if (genuine_intel)
4680 		model = intel_model_duplicates(model);
4681 
4682 	if (!(edx_flags & (1 << 5)))
4683 		errx(1, "CPUID: no MSR");
4684 
4685 	if (max_extended_level >= 0x80000007) {
4686 
4687 		/*
4688 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
4689 		 * this check is valid for both Intel and AMD
4690 		 */
4691 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4692 		has_invariant_tsc = edx & (1 << 8);
4693 	}
4694 
4695 	/*
4696 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
4697 	 * this check is valid for both Intel and AMD
4698 	 */
4699 
4700 	__cpuid(0x6, eax, ebx, ecx, edx);
4701 	has_aperf = ecx & (1 << 0);
4702 	if (has_aperf) {
4703 		BIC_PRESENT(BIC_Avg_MHz);
4704 		BIC_PRESENT(BIC_Busy);
4705 		BIC_PRESENT(BIC_Bzy_MHz);
4706 	}
4707 	do_dts = eax & (1 << 0);
4708 	if (do_dts)
4709 		BIC_PRESENT(BIC_CoreTmp);
4710 	has_turbo = eax & (1 << 1);
4711 	do_ptm = eax & (1 << 6);
4712 	if (do_ptm)
4713 		BIC_PRESENT(BIC_PkgTmp);
4714 	has_hwp = eax & (1 << 7);
4715 	has_hwp_notify = eax & (1 << 8);
4716 	has_hwp_activity_window = eax & (1 << 9);
4717 	has_hwp_epp = eax & (1 << 10);
4718 	has_hwp_pkg = eax & (1 << 11);
4719 	has_epb = ecx & (1 << 3);
4720 
4721 	if (!quiet)
4722 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
4723 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
4724 			has_aperf ? "" : "No-",
4725 			has_turbo ? "" : "No-",
4726 			do_dts ? "" : "No-",
4727 			do_ptm ? "" : "No-",
4728 			has_hwp ? "" : "No-",
4729 			has_hwp_notify ? "" : "No-",
4730 			has_hwp_activity_window ? "" : "No-",
4731 			has_hwp_epp ? "" : "No-",
4732 			has_hwp_pkg ? "" : "No-",
4733 			has_epb ? "" : "No-");
4734 
4735 	if (!quiet)
4736 		decode_misc_enable_msr();
4737 
4738 
4739 	if (max_level >= 0x7 && !quiet) {
4740 		int has_sgx;
4741 
4742 		ecx = 0;
4743 
4744 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
4745 
4746 		has_sgx = ebx & (1 << 2);
4747 		fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
4748 
4749 		if (has_sgx)
4750 			decode_feature_control_msr();
4751 	}
4752 
4753 	if (max_level >= 0x15) {
4754 		unsigned int eax_crystal;
4755 		unsigned int ebx_tsc;
4756 
4757 		/*
4758 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
4759 		 */
4760 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
4761 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
4762 
4763 		if (ebx_tsc != 0) {
4764 
4765 			if (!quiet && (ebx != 0))
4766 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
4767 					eax_crystal, ebx_tsc, crystal_hz);
4768 
4769 			if (crystal_hz == 0)
4770 				switch(model) {
4771 				case INTEL_FAM6_SKYLAKE_L:	/* SKL */
4772 					crystal_hz = 24000000;	/* 24.0 MHz */
4773 					break;
4774 				case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
4775 					crystal_hz = 25000000;	/* 25.0 MHz */
4776 					break;
4777 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4778 				case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4779 					crystal_hz = 19200000;	/* 19.2 MHz */
4780 					break;
4781 				default:
4782 					crystal_hz = 0;
4783 			}
4784 
4785 			if (crystal_hz) {
4786 				tsc_hz =  (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
4787 				if (!quiet)
4788 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
4789 						tsc_hz / 1000000, crystal_hz, ebx_tsc,  eax_crystal);
4790 			}
4791 		}
4792 	}
4793 	if (max_level >= 0x16) {
4794 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
4795 
4796 		/*
4797 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
4798 		 */
4799 		base_mhz = max_mhz = bus_mhz = edx = 0;
4800 
4801 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
4802 		if (!quiet)
4803 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
4804 				base_mhz, max_mhz, bus_mhz);
4805 	}
4806 
4807 	if (has_aperf)
4808 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
4809 
4810 	BIC_PRESENT(BIC_IRQ);
4811 	BIC_PRESENT(BIC_TSC_MHz);
4812 
4813 	if (probe_nhm_msrs(family, model)) {
4814 		do_nhm_platform_info = 1;
4815 		BIC_PRESENT(BIC_CPU_c1);
4816 		BIC_PRESENT(BIC_CPU_c3);
4817 		BIC_PRESENT(BIC_CPU_c6);
4818 		BIC_PRESENT(BIC_SMI);
4819 	}
4820 	do_snb_cstates = has_snb_msrs(family, model);
4821 
4822 	if (do_snb_cstates)
4823 		BIC_PRESENT(BIC_CPU_c7);
4824 
4825 	do_irtl_snb = has_snb_msrs(family, model);
4826 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
4827 		BIC_PRESENT(BIC_Pkgpc2);
4828 	if (pkg_cstate_limit >= PCL__3)
4829 		BIC_PRESENT(BIC_Pkgpc3);
4830 	if (pkg_cstate_limit >= PCL__6)
4831 		BIC_PRESENT(BIC_Pkgpc6);
4832 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
4833 		BIC_PRESENT(BIC_Pkgpc7);
4834 	if (has_slv_msrs(family, model)) {
4835 		BIC_NOT_PRESENT(BIC_Pkgpc2);
4836 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4837 		BIC_PRESENT(BIC_Pkgpc6);
4838 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4839 		BIC_PRESENT(BIC_Mod_c6);
4840 		use_c1_residency_msr = 1;
4841 	}
4842 	if (is_dnv(family, model)) {
4843 		BIC_PRESENT(BIC_CPU_c1);
4844 		BIC_NOT_PRESENT(BIC_CPU_c3);
4845 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4846 		BIC_NOT_PRESENT(BIC_CPU_c7);
4847 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4848 		use_c1_residency_msr = 1;
4849 	}
4850 	if (is_skx(family, model)) {
4851 		BIC_NOT_PRESENT(BIC_CPU_c3);
4852 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4853 		BIC_NOT_PRESENT(BIC_CPU_c7);
4854 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4855 	}
4856 	if (is_bdx(family, model)) {
4857 		BIC_NOT_PRESENT(BIC_CPU_c7);
4858 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4859 	}
4860 	if (has_c8910_msrs(family, model)) {
4861 		BIC_PRESENT(BIC_Pkgpc8);
4862 		BIC_PRESENT(BIC_Pkgpc9);
4863 		BIC_PRESENT(BIC_Pkgpc10);
4864 	}
4865 	do_irtl_hsw = has_c8910_msrs(family, model);
4866 	if (has_skl_msrs(family, model)) {
4867 		BIC_PRESENT(BIC_Totl_c0);
4868 		BIC_PRESENT(BIC_Any_c0);
4869 		BIC_PRESENT(BIC_GFX_c0);
4870 		BIC_PRESENT(BIC_CPUGFX);
4871 	}
4872 	do_slm_cstates = is_slm(family, model);
4873 	do_knl_cstates  = is_knl(family, model);
4874 
4875 	if (do_slm_cstates || do_knl_cstates || is_cnl(family, model))
4876 		BIC_NOT_PRESENT(BIC_CPU_c3);
4877 
4878 	if (!quiet)
4879 		decode_misc_pwr_mgmt_msr();
4880 
4881 	if (!quiet && has_slv_msrs(family, model))
4882 		decode_c6_demotion_policy_msr();
4883 
4884 	rapl_probe(family, model);
4885 	perf_limit_reasons_probe(family, model);
4886 	automatic_cstate_conversion_probe(family, model);
4887 
4888 	if (!quiet)
4889 		dump_cstate_pstate_config_info(family, model);
4890 
4891 	if (!quiet)
4892 		dump_sysfs_cstate_config();
4893 	if (!quiet)
4894 		dump_sysfs_pstate_config();
4895 
4896 	if (has_skl_msrs(family, model))
4897 		calculate_tsc_tweak();
4898 
4899 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
4900 		BIC_PRESENT(BIC_GFX_rc6);
4901 
4902 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
4903 		BIC_PRESENT(BIC_GFXMHz);
4904 
4905 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK))
4906 		BIC_PRESENT(BIC_CPU_LPI);
4907 	else
4908 		BIC_NOT_PRESENT(BIC_CPU_LPI);
4909 
4910 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", R_OK))
4911 		BIC_PRESENT(BIC_SYS_LPI);
4912 	else
4913 		BIC_NOT_PRESENT(BIC_SYS_LPI);
4914 
4915 	if (!quiet)
4916 		decode_misc_feature_control();
4917 
4918 	return;
4919 }
4920 
4921 /*
4922  * in /dev/cpu/ return success for names that are numbers
4923  * ie. filter out ".", "..", "microcode".
4924  */
4925 int dir_filter(const struct dirent *dirp)
4926 {
4927 	if (isdigit(dirp->d_name[0]))
4928 		return 1;
4929 	else
4930 		return 0;
4931 }
4932 
4933 int open_dev_cpu_msr(int dummy1)
4934 {
4935 	return 0;
4936 }
4937 
4938 void topology_probe()
4939 {
4940 	int i;
4941 	int max_core_id = 0;
4942 	int max_package_id = 0;
4943 	int max_die_id = 0;
4944 	int max_siblings = 0;
4945 
4946 	/* Initialize num_cpus, max_cpu_num */
4947 	set_max_cpu_num();
4948 	topo.num_cpus = 0;
4949 	for_all_proc_cpus(count_cpus);
4950 	if (!summary_only && topo.num_cpus > 1)
4951 		BIC_PRESENT(BIC_CPU);
4952 
4953 	if (debug > 1)
4954 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4955 
4956 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
4957 	if (cpus == NULL)
4958 		err(1, "calloc cpus");
4959 
4960 	/*
4961 	 * Allocate and initialize cpu_present_set
4962 	 */
4963 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4964 	if (cpu_present_set == NULL)
4965 		err(3, "CPU_ALLOC");
4966 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4967 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4968 	for_all_proc_cpus(mark_cpu_present);
4969 
4970 	/*
4971 	 * Validate that all cpus in cpu_subset are also in cpu_present_set
4972 	 */
4973 	for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
4974 		if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
4975 			if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
4976 				err(1, "cpu%d not present", i);
4977 	}
4978 
4979 	/*
4980 	 * Allocate and initialize cpu_affinity_set
4981 	 */
4982 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4983 	if (cpu_affinity_set == NULL)
4984 		err(3, "CPU_ALLOC");
4985 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4986 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4987 
4988 	for_all_proc_cpus(init_thread_id);
4989 
4990 	/*
4991 	 * For online cpus
4992 	 * find max_core_id, max_package_id
4993 	 */
4994 	for (i = 0; i <= topo.max_cpu_num; ++i) {
4995 		int siblings;
4996 
4997 		if (cpu_is_not_present(i)) {
4998 			if (debug > 1)
4999 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
5000 			continue;
5001 		}
5002 
5003 		cpus[i].logical_cpu_id = i;
5004 
5005 		/* get package information */
5006 		cpus[i].physical_package_id = get_physical_package_id(i);
5007 		if (cpus[i].physical_package_id > max_package_id)
5008 			max_package_id = cpus[i].physical_package_id;
5009 
5010 		/* get die information */
5011 		cpus[i].die_id = get_die_id(i);
5012 		if (cpus[i].die_id > max_die_id)
5013 			max_die_id = cpus[i].die_id;
5014 
5015 		/* get numa node information */
5016 		cpus[i].physical_node_id = get_physical_node_id(&cpus[i]);
5017 		if (cpus[i].physical_node_id > topo.max_node_num)
5018 			topo.max_node_num = cpus[i].physical_node_id;
5019 
5020 		/* get core information */
5021 		cpus[i].physical_core_id = get_core_id(i);
5022 		if (cpus[i].physical_core_id > max_core_id)
5023 			max_core_id = cpus[i].physical_core_id;
5024 
5025 		/* get thread information */
5026 		siblings = get_thread_siblings(&cpus[i]);
5027 		if (siblings > max_siblings)
5028 			max_siblings = siblings;
5029 		if (cpus[i].thread_id == 0)
5030 			topo.num_cores++;
5031 	}
5032 
5033 	topo.cores_per_node = max_core_id + 1;
5034 	if (debug > 1)
5035 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
5036 			max_core_id, topo.cores_per_node);
5037 	if (!summary_only && topo.cores_per_node > 1)
5038 		BIC_PRESENT(BIC_Core);
5039 
5040 	topo.num_die = max_die_id + 1;
5041 	if (debug > 1)
5042 		fprintf(outf, "max_die_id %d, sizing for %d die\n",
5043 				max_die_id, topo.num_die);
5044 	if (!summary_only && topo.num_die > 1)
5045 		BIC_PRESENT(BIC_Die);
5046 
5047 	topo.num_packages = max_package_id + 1;
5048 	if (debug > 1)
5049 		fprintf(outf, "max_package_id %d, sizing for %d packages\n",
5050 			max_package_id, topo.num_packages);
5051 	if (!summary_only && topo.num_packages > 1)
5052 		BIC_PRESENT(BIC_Package);
5053 
5054 	set_node_data();
5055 	if (debug > 1)
5056 		fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg);
5057 	if (!summary_only && topo.nodes_per_pkg > 1)
5058 		BIC_PRESENT(BIC_Node);
5059 
5060 	topo.threads_per_core = max_siblings;
5061 	if (debug > 1)
5062 		fprintf(outf, "max_siblings %d\n", max_siblings);
5063 
5064 	if (debug < 1)
5065 		return;
5066 
5067 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5068 		if (cpu_is_not_present(i))
5069 			continue;
5070 		fprintf(outf,
5071 			"cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n",
5072 			i, cpus[i].physical_package_id, cpus[i].die_id,
5073 			cpus[i].physical_node_id,
5074 			cpus[i].logical_node_id,
5075 			cpus[i].physical_core_id,
5076 			cpus[i].thread_id);
5077 	}
5078 
5079 }
5080 
5081 void
5082 allocate_counters(struct thread_data **t, struct core_data **c,
5083 		  struct pkg_data **p)
5084 {
5085 	int i;
5086 	int num_cores = topo.cores_per_node * topo.nodes_per_pkg *
5087 			topo.num_packages;
5088 	int num_threads = topo.threads_per_core * num_cores;
5089 
5090 	*t = calloc(num_threads, sizeof(struct thread_data));
5091 	if (*t == NULL)
5092 		goto error;
5093 
5094 	for (i = 0; i < num_threads; i++)
5095 		(*t)[i].cpu_id = -1;
5096 
5097 	*c = calloc(num_cores, sizeof(struct core_data));
5098 	if (*c == NULL)
5099 		goto error;
5100 
5101 	for (i = 0; i < num_cores; i++)
5102 		(*c)[i].core_id = -1;
5103 
5104 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
5105 	if (*p == NULL)
5106 		goto error;
5107 
5108 	for (i = 0; i < topo.num_packages; i++)
5109 		(*p)[i].package_id = i;
5110 
5111 	return;
5112 error:
5113 	err(1, "calloc counters");
5114 }
5115 /*
5116  * init_counter()
5117  *
5118  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
5119  */
5120 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
5121 	struct pkg_data *pkg_base, int cpu_id)
5122 {
5123 	int pkg_id = cpus[cpu_id].physical_package_id;
5124 	int node_id = cpus[cpu_id].logical_node_id;
5125 	int core_id = cpus[cpu_id].physical_core_id;
5126 	int thread_id = cpus[cpu_id].thread_id;
5127 	struct thread_data *t;
5128 	struct core_data *c;
5129 	struct pkg_data *p;
5130 
5131 
5132 	/* Workaround for systems where physical_node_id==-1
5133 	 * and logical_node_id==(-1 - topo.num_cpus)
5134 	 */
5135 	if (node_id < 0)
5136 		node_id = 0;
5137 
5138 	t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id);
5139 	c = GET_CORE(core_base, core_id, node_id, pkg_id);
5140 	p = GET_PKG(pkg_base, pkg_id);
5141 
5142 	t->cpu_id = cpu_id;
5143 	if (thread_id == 0) {
5144 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
5145 		if (cpu_is_first_core_in_package(cpu_id))
5146 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
5147 	}
5148 
5149 	c->core_id = core_id;
5150 	p->package_id = pkg_id;
5151 }
5152 
5153 
5154 int initialize_counters(int cpu_id)
5155 {
5156 	init_counter(EVEN_COUNTERS, cpu_id);
5157 	init_counter(ODD_COUNTERS, cpu_id);
5158 	return 0;
5159 }
5160 
5161 void allocate_output_buffer()
5162 {
5163 	output_buffer = calloc(1, (1 + topo.num_cpus) * 2048);
5164 	outp = output_buffer;
5165 	if (outp == NULL)
5166 		err(-1, "calloc output buffer");
5167 }
5168 void allocate_fd_percpu(void)
5169 {
5170 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5171 	if (fd_percpu == NULL)
5172 		err(-1, "calloc fd_percpu");
5173 }
5174 void allocate_irq_buffers(void)
5175 {
5176 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
5177 	if (irq_column_2_cpu == NULL)
5178 		err(-1, "calloc %d", topo.num_cpus);
5179 
5180 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5181 	if (irqs_per_cpu == NULL)
5182 		err(-1, "calloc %d", topo.max_cpu_num + 1);
5183 }
5184 void setup_all_buffers(void)
5185 {
5186 	topology_probe();
5187 	allocate_irq_buffers();
5188 	allocate_fd_percpu();
5189 	allocate_counters(&thread_even, &core_even, &package_even);
5190 	allocate_counters(&thread_odd, &core_odd, &package_odd);
5191 	allocate_output_buffer();
5192 	for_all_proc_cpus(initialize_counters);
5193 }
5194 
5195 void set_base_cpu(void)
5196 {
5197 	base_cpu = sched_getcpu();
5198 	if (base_cpu < 0)
5199 		err(-ENODEV, "No valid cpus found");
5200 
5201 	if (debug > 1)
5202 		fprintf(outf, "base_cpu = %d\n", base_cpu);
5203 }
5204 
5205 void turbostat_init()
5206 {
5207 	setup_all_buffers();
5208 	set_base_cpu();
5209 	check_dev_msr();
5210 	check_permissions();
5211 	process_cpuid();
5212 
5213 
5214 	if (!quiet)
5215 		for_all_cpus(print_hwp, ODD_COUNTERS);
5216 
5217 	if (!quiet)
5218 		for_all_cpus(print_epb, ODD_COUNTERS);
5219 
5220 	if (!quiet)
5221 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
5222 
5223 	if (!quiet)
5224 		for_all_cpus(print_rapl, ODD_COUNTERS);
5225 
5226 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
5227 
5228 	if (!quiet)
5229 		for_all_cpus(print_thermal, ODD_COUNTERS);
5230 
5231 	if (!quiet && do_irtl_snb)
5232 		print_irtl();
5233 }
5234 
5235 int fork_it(char **argv)
5236 {
5237 	pid_t child_pid;
5238 	int status;
5239 
5240 	snapshot_proc_sysfs_files();
5241 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
5242 	first_counter_read = 0;
5243 	if (status)
5244 		exit(status);
5245 	/* clear affinity side-effect of get_counters() */
5246 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
5247 	gettimeofday(&tv_even, (struct timezone *)NULL);
5248 
5249 	child_pid = fork();
5250 	if (!child_pid) {
5251 		/* child */
5252 		execvp(argv[0], argv);
5253 		err(errno, "exec %s", argv[0]);
5254 	} else {
5255 
5256 		/* parent */
5257 		if (child_pid == -1)
5258 			err(1, "fork");
5259 
5260 		signal(SIGINT, SIG_IGN);
5261 		signal(SIGQUIT, SIG_IGN);
5262 		if (waitpid(child_pid, &status, 0) == -1)
5263 			err(status, "waitpid");
5264 
5265 		if (WIFEXITED(status))
5266 			status = WEXITSTATUS(status);
5267 	}
5268 	/*
5269 	 * n.b. fork_it() does not check for errors from for_all_cpus()
5270 	 * because re-starting is problematic when forking
5271 	 */
5272 	snapshot_proc_sysfs_files();
5273 	for_all_cpus(get_counters, ODD_COUNTERS);
5274 	gettimeofday(&tv_odd, (struct timezone *)NULL);
5275 	timersub(&tv_odd, &tv_even, &tv_delta);
5276 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
5277 		fprintf(outf, "%s: Counter reset detected\n", progname);
5278 	else {
5279 		compute_average(EVEN_COUNTERS);
5280 		format_all_counters(EVEN_COUNTERS);
5281 	}
5282 
5283 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
5284 
5285 	flush_output_stderr();
5286 
5287 	return status;
5288 }
5289 
5290 int get_and_dump_counters(void)
5291 {
5292 	int status;
5293 
5294 	snapshot_proc_sysfs_files();
5295 	status = for_all_cpus(get_counters, ODD_COUNTERS);
5296 	if (status)
5297 		return status;
5298 
5299 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
5300 	if (status)
5301 		return status;
5302 
5303 	flush_output_stdout();
5304 
5305 	return status;
5306 }
5307 
5308 void print_version() {
5309 	fprintf(outf, "turbostat version 19.08.31"
5310 		" - Len Brown <lenb@kernel.org>\n");
5311 }
5312 
5313 int add_counter(unsigned int msr_num, char *path, char *name,
5314 	unsigned int width, enum counter_scope scope,
5315 	enum counter_type type, enum counter_format format, int flags)
5316 {
5317 	struct msr_counter *msrp;
5318 
5319 	msrp = calloc(1, sizeof(struct msr_counter));
5320 	if (msrp == NULL) {
5321 		perror("calloc");
5322 		exit(1);
5323 	}
5324 
5325 	msrp->msr_num = msr_num;
5326 	strncpy(msrp->name, name, NAME_BYTES);
5327 	if (path)
5328 		strncpy(msrp->path, path, PATH_BYTES);
5329 	msrp->width = width;
5330 	msrp->type = type;
5331 	msrp->format = format;
5332 	msrp->flags = flags;
5333 
5334 	switch (scope) {
5335 
5336 	case SCOPE_CPU:
5337 		msrp->next = sys.tp;
5338 		sys.tp = msrp;
5339 		sys.added_thread_counters++;
5340 		if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) {
5341 			fprintf(stderr, "exceeded max %d added thread counters\n",
5342 				MAX_ADDED_COUNTERS);
5343 			exit(-1);
5344 		}
5345 		break;
5346 
5347 	case SCOPE_CORE:
5348 		msrp->next = sys.cp;
5349 		sys.cp = msrp;
5350 		sys.added_core_counters++;
5351 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
5352 			fprintf(stderr, "exceeded max %d added core counters\n",
5353 				MAX_ADDED_COUNTERS);
5354 			exit(-1);
5355 		}
5356 		break;
5357 
5358 	case SCOPE_PACKAGE:
5359 		msrp->next = sys.pp;
5360 		sys.pp = msrp;
5361 		sys.added_package_counters++;
5362 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
5363 			fprintf(stderr, "exceeded max %d added package counters\n",
5364 				MAX_ADDED_COUNTERS);
5365 			exit(-1);
5366 		}
5367 		break;
5368 	}
5369 
5370 	return 0;
5371 }
5372 
5373 void parse_add_command(char *add_command)
5374 {
5375 	int msr_num = 0;
5376 	char *path = NULL;
5377 	char name_buffer[NAME_BYTES] = "";
5378 	int width = 64;
5379 	int fail = 0;
5380 	enum counter_scope scope = SCOPE_CPU;
5381 	enum counter_type type = COUNTER_CYCLES;
5382 	enum counter_format format = FORMAT_DELTA;
5383 
5384 	while (add_command) {
5385 
5386 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
5387 			goto next;
5388 
5389 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
5390 			goto next;
5391 
5392 		if (*add_command == '/') {
5393 			path = add_command;
5394 			goto next;
5395 		}
5396 
5397 		if (sscanf(add_command, "u%d", &width) == 1) {
5398 			if ((width == 32) || (width == 64))
5399 				goto next;
5400 			width = 64;
5401 		}
5402 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
5403 			scope = SCOPE_CPU;
5404 			goto next;
5405 		}
5406 		if (!strncmp(add_command, "core", strlen("core"))) {
5407 			scope = SCOPE_CORE;
5408 			goto next;
5409 		}
5410 		if (!strncmp(add_command, "package", strlen("package"))) {
5411 			scope = SCOPE_PACKAGE;
5412 			goto next;
5413 		}
5414 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
5415 			type = COUNTER_CYCLES;
5416 			goto next;
5417 		}
5418 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
5419 			type = COUNTER_SECONDS;
5420 			goto next;
5421 		}
5422 		if (!strncmp(add_command, "usec", strlen("usec"))) {
5423 			type = COUNTER_USEC;
5424 			goto next;
5425 		}
5426 		if (!strncmp(add_command, "raw", strlen("raw"))) {
5427 			format = FORMAT_RAW;
5428 			goto next;
5429 		}
5430 		if (!strncmp(add_command, "delta", strlen("delta"))) {
5431 			format = FORMAT_DELTA;
5432 			goto next;
5433 		}
5434 		if (!strncmp(add_command, "percent", strlen("percent"))) {
5435 			format = FORMAT_PERCENT;
5436 			goto next;
5437 		}
5438 
5439 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
5440 			char *eos;
5441 
5442 			eos = strchr(name_buffer, ',');
5443 			if (eos)
5444 				*eos = '\0';
5445 			goto next;
5446 		}
5447 
5448 next:
5449 		add_command = strchr(add_command, ',');
5450 		if (add_command) {
5451 			*add_command = '\0';
5452 			add_command++;
5453 		}
5454 
5455 	}
5456 	if ((msr_num == 0) && (path == NULL)) {
5457 		fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
5458 		fail++;
5459 	}
5460 
5461 	/* generate default column header */
5462 	if (*name_buffer == '\0') {
5463 		if (width == 32)
5464 			sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5465 		else
5466 			sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5467 	}
5468 
5469 	if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
5470 		fail++;
5471 
5472 	if (fail) {
5473 		help();
5474 		exit(1);
5475 	}
5476 }
5477 
5478 int is_deferred_skip(char *name)
5479 {
5480 	int i;
5481 
5482 	for (i = 0; i < deferred_skip_index; ++i)
5483 		if (!strcmp(name, deferred_skip_names[i]))
5484 			return 1;
5485 	return 0;
5486 }
5487 
5488 void probe_sysfs(void)
5489 {
5490 	char path[64];
5491 	char name_buf[16];
5492 	FILE *input;
5493 	int state;
5494 	char *sp;
5495 
5496 	if (!DO_BIC(BIC_sysfs))
5497 		return;
5498 
5499 	for (state = 10; state >= 0; --state) {
5500 
5501 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5502 			base_cpu, state);
5503 		input = fopen(path, "r");
5504 		if (input == NULL)
5505 			continue;
5506 		if (!fgets(name_buf, sizeof(name_buf), input))
5507 			err(1, "%s: failed to read file", path);
5508 
5509 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5510 		sp = strchr(name_buf, '-');
5511 		if (!sp)
5512 			sp = strchrnul(name_buf, '\n');
5513 		*sp = '%';
5514 		*(sp + 1) = '\0';
5515 
5516 		fclose(input);
5517 
5518 		sprintf(path, "cpuidle/state%d/time", state);
5519 
5520 		if (is_deferred_skip(name_buf))
5521 			continue;
5522 
5523 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC,
5524 				FORMAT_PERCENT, SYSFS_PERCPU);
5525 	}
5526 
5527 	for (state = 10; state >= 0; --state) {
5528 
5529 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5530 			base_cpu, state);
5531 		input = fopen(path, "r");
5532 		if (input == NULL)
5533 			continue;
5534 		if (!fgets(name_buf, sizeof(name_buf), input))
5535 			err(1, "%s: failed to read file", path);
5536 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5537 		sp = strchr(name_buf, '-');
5538 		if (!sp)
5539 			sp = strchrnul(name_buf, '\n');
5540 		*sp = '\0';
5541 		fclose(input);
5542 
5543 		sprintf(path, "cpuidle/state%d/usage", state);
5544 
5545 		if (is_deferred_skip(name_buf))
5546 			continue;
5547 
5548 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS,
5549 				FORMAT_DELTA, SYSFS_PERCPU);
5550 	}
5551 
5552 }
5553 
5554 
5555 /*
5556  * parse cpuset with following syntax
5557  * 1,2,4..6,8-10 and set bits in cpu_subset
5558  */
5559 void parse_cpu_command(char *optarg)
5560 {
5561 	unsigned int start, end;
5562 	char *next;
5563 
5564 	if (!strcmp(optarg, "core")) {
5565 		if (cpu_subset)
5566 			goto error;
5567 		show_core_only++;
5568 		return;
5569 	}
5570 	if (!strcmp(optarg, "package")) {
5571 		if (cpu_subset)
5572 			goto error;
5573 		show_pkg_only++;
5574 		return;
5575 	}
5576 	if (show_core_only || show_pkg_only)
5577 		goto error;
5578 
5579 	cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
5580 	if (cpu_subset == NULL)
5581 		err(3, "CPU_ALLOC");
5582 	cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
5583 
5584 	CPU_ZERO_S(cpu_subset_size, cpu_subset);
5585 
5586 	next = optarg;
5587 
5588 	while (next && *next) {
5589 
5590 		if (*next == '-')	/* no negative cpu numbers */
5591 			goto error;
5592 
5593 		start = strtoul(next, &next, 10);
5594 
5595 		if (start >= CPU_SUBSET_MAXCPUS)
5596 			goto error;
5597 		CPU_SET_S(start, cpu_subset_size, cpu_subset);
5598 
5599 		if (*next == '\0')
5600 			break;
5601 
5602 		if (*next == ',') {
5603 			next += 1;
5604 			continue;
5605 		}
5606 
5607 		if (*next == '-') {
5608 			next += 1;	/* start range */
5609 		} else if (*next == '.') {
5610 			next += 1;
5611 			if (*next == '.')
5612 				next += 1;	/* start range */
5613 			else
5614 				goto error;
5615 		}
5616 
5617 		end = strtoul(next, &next, 10);
5618 		if (end <= start)
5619 			goto error;
5620 
5621 		while (++start <= end) {
5622 			if (start >= CPU_SUBSET_MAXCPUS)
5623 				goto error;
5624 			CPU_SET_S(start, cpu_subset_size, cpu_subset);
5625 		}
5626 
5627 		if (*next == ',')
5628 			next += 1;
5629 		else if (*next != '\0')
5630 			goto error;
5631 	}
5632 
5633 	return;
5634 
5635 error:
5636 	fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
5637 	help();
5638 	exit(-1);
5639 }
5640 
5641 
5642 void cmdline(int argc, char **argv)
5643 {
5644 	int opt;
5645 	int option_index = 0;
5646 	static struct option long_options[] = {
5647 		{"add",		required_argument,	0, 'a'},
5648 		{"cpu",		required_argument,	0, 'c'},
5649 		{"Dump",	no_argument,		0, 'D'},
5650 		{"debug",	no_argument,		0, 'd'},	/* internal, not documented */
5651 		{"enable",	required_argument,	0, 'e'},
5652 		{"interval",	required_argument,	0, 'i'},
5653 		{"num_iterations",	required_argument,	0, 'n'},
5654 		{"help",	no_argument,		0, 'h'},
5655 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
5656 		{"Joules",	no_argument,		0, 'J'},
5657 		{"list",	no_argument,		0, 'l'},
5658 		{"out",		required_argument,	0, 'o'},
5659 		{"quiet",	no_argument,		0, 'q'},
5660 		{"show",	required_argument,	0, 's'},
5661 		{"Summary",	no_argument,		0, 'S'},
5662 		{"TCC",		required_argument,	0, 'T'},
5663 		{"version",	no_argument,		0, 'v' },
5664 		{0,		0,			0,  0 }
5665 	};
5666 
5667 	progname = argv[0];
5668 
5669 	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v",
5670 				long_options, &option_index)) != -1) {
5671 		switch (opt) {
5672 		case 'a':
5673 			parse_add_command(optarg);
5674 			break;
5675 		case 'c':
5676 			parse_cpu_command(optarg);
5677 			break;
5678 		case 'D':
5679 			dump_only++;
5680 			break;
5681 		case 'e':
5682 			/* --enable specified counter */
5683 			bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST);
5684 			break;
5685 		case 'd':
5686 			debug++;
5687 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5688 			break;
5689 		case 'H':
5690 			/*
5691 			 * --hide: do not show those specified
5692 			 *  multiple invocations simply clear more bits in enabled mask
5693 			 */
5694 			bic_enabled &= ~bic_lookup(optarg, HIDE_LIST);
5695 			break;
5696 		case 'h':
5697 		default:
5698 			help();
5699 			exit(1);
5700 		case 'i':
5701 			{
5702 				double interval = strtod(optarg, NULL);
5703 
5704 				if (interval < 0.001) {
5705 					fprintf(outf, "interval %f seconds is too small\n",
5706 						interval);
5707 					exit(2);
5708 				}
5709 
5710 				interval_tv.tv_sec = interval_ts.tv_sec = interval;
5711 				interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000;
5712 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
5713 			}
5714 			break;
5715 		case 'J':
5716 			rapl_joules++;
5717 			break;
5718 		case 'l':
5719 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5720 			list_header_only++;
5721 			quiet++;
5722 			break;
5723 		case 'o':
5724 			outf = fopen_or_die(optarg, "w");
5725 			break;
5726 		case 'q':
5727 			quiet = 1;
5728 			break;
5729 		case 'n':
5730 			num_iterations = strtod(optarg, NULL);
5731 
5732 			if (num_iterations <= 0) {
5733 				fprintf(outf, "iterations %d should be positive number\n",
5734 					num_iterations);
5735 				exit(2);
5736 			}
5737 			break;
5738 		case 's':
5739 			/*
5740 			 * --show: show only those specified
5741 			 *  The 1st invocation will clear and replace the enabled mask
5742 			 *  subsequent invocations can add to it.
5743 			 */
5744 			if (shown == 0)
5745 				bic_enabled = bic_lookup(optarg, SHOW_LIST);
5746 			else
5747 				bic_enabled |= bic_lookup(optarg, SHOW_LIST);
5748 			shown = 1;
5749 			break;
5750 		case 'S':
5751 			summary_only++;
5752 			break;
5753 		case 'T':
5754 			tcc_activation_temp_override = atoi(optarg);
5755 			break;
5756 		case 'v':
5757 			print_version();
5758 			exit(0);
5759 			break;
5760 		}
5761 	}
5762 }
5763 
5764 int main(int argc, char **argv)
5765 {
5766 	outf = stderr;
5767 	cmdline(argc, argv);
5768 
5769 	if (!quiet)
5770 		print_version();
5771 
5772 	probe_sysfs();
5773 
5774 	turbostat_init();
5775 
5776 	/* dump counters and exit */
5777 	if (dump_only)
5778 		return get_and_dump_counters();
5779 
5780 	/* list header and exit */
5781 	if (list_header_only) {
5782 		print_header(",");
5783 		flush_output_stdout();
5784 		return 0;
5785 	}
5786 
5787 	/*
5788 	 * if any params left, it must be a command to fork
5789 	 */
5790 	if (argc - optind)
5791 		return fork_it(argv + optind);
5792 	else
5793 		turbostat_loop();
5794 
5795 	return 0;
5796 }
5797