xref: /openbmc/linux/tools/power/x86/turbostat/turbostat.c (revision 9cfa8e042f7cbb1994cc5923e46c78b36f6054f4)
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 max_level, max_extended_level;
63 unsigned int has_invariant_tsc;
64 unsigned int do_nhm_platform_info;
65 unsigned int no_MSR_MISC_PWR_MGMT;
66 unsigned int aperf_mperf_multiplier = 1;
67 double bclk;
68 double base_hz;
69 unsigned int has_base_hz;
70 double tsc_tweak = 1.0;
71 unsigned int show_pkg_only;
72 unsigned int show_core_only;
73 char *output_buffer, *outp;
74 unsigned int do_rapl;
75 unsigned int do_dts;
76 unsigned int do_ptm;
77 unsigned long long  gfx_cur_rc6_ms;
78 unsigned long long cpuidle_cur_cpu_lpi_us;
79 unsigned long long cpuidle_cur_sys_lpi_us;
80 unsigned int gfx_cur_mhz;
81 unsigned int tcc_activation_temp;
82 unsigned int tcc_activation_temp_override;
83 double rapl_power_units, rapl_time_units;
84 double rapl_dram_energy_units, rapl_energy_units;
85 double rapl_joule_counter_range;
86 unsigned int do_core_perf_limit_reasons;
87 unsigned int has_automatic_cstate_conversion;
88 unsigned int do_gfx_perf_limit_reasons;
89 unsigned int do_ring_perf_limit_reasons;
90 unsigned int crystal_hz;
91 unsigned long long tsc_hz;
92 int base_cpu;
93 double discover_bclk(unsigned int family, unsigned int model);
94 unsigned int has_hwp;	/* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
95 			/* IA32_HWP_REQUEST, IA32_HWP_STATUS */
96 unsigned int has_hwp_notify;		/* IA32_HWP_INTERRUPT */
97 unsigned int has_hwp_activity_window;	/* IA32_HWP_REQUEST[bits 41:32] */
98 unsigned int has_hwp_epp;		/* IA32_HWP_REQUEST[bits 31:24] */
99 unsigned int has_hwp_pkg;		/* IA32_HWP_REQUEST_PKG */
100 unsigned int has_misc_feature_control;
101 unsigned int first_counter_read = 1;
102 int ignore_stdin;
103 
104 #define RAPL_PKG		(1 << 0)
105 					/* 0x610 MSR_PKG_POWER_LIMIT */
106 					/* 0x611 MSR_PKG_ENERGY_STATUS */
107 #define RAPL_PKG_PERF_STATUS	(1 << 1)
108 					/* 0x613 MSR_PKG_PERF_STATUS */
109 #define RAPL_PKG_POWER_INFO	(1 << 2)
110 					/* 0x614 MSR_PKG_POWER_INFO */
111 
112 #define RAPL_DRAM		(1 << 3)
113 					/* 0x618 MSR_DRAM_POWER_LIMIT */
114 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
115 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
116 					/* 0x61b MSR_DRAM_PERF_STATUS */
117 #define RAPL_DRAM_POWER_INFO	(1 << 5)
118 					/* 0x61c MSR_DRAM_POWER_INFO */
119 
120 #define RAPL_CORES_POWER_LIMIT	(1 << 6)
121 					/* 0x638 MSR_PP0_POWER_LIMIT */
122 #define RAPL_CORE_POLICY	(1 << 7)
123 					/* 0x63a MSR_PP0_POLICY */
124 
125 #define RAPL_GFX		(1 << 8)
126 					/* 0x640 MSR_PP1_POWER_LIMIT */
127 					/* 0x641 MSR_PP1_ENERGY_STATUS */
128 					/* 0x642 MSR_PP1_POLICY */
129 
130 #define RAPL_CORES_ENERGY_STATUS	(1 << 9)
131 					/* 0x639 MSR_PP0_ENERGY_STATUS */
132 #define RAPL_PER_CORE_ENERGY	(1 << 10)
133 					/* Indicates cores energy collection is per-core,
134 					 * not per-package. */
135 #define RAPL_AMD_F17H		(1 << 11)
136 					/* 0xc0010299 MSR_RAPL_PWR_UNIT */
137 					/* 0xc001029a MSR_CORE_ENERGY_STAT */
138 					/* 0xc001029b MSR_PKG_ENERGY_STAT */
139 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
140 #define	TJMAX_DEFAULT	100
141 
142 /* MSRs that are not yet in the kernel-provided header. */
143 #define MSR_RAPL_PWR_UNIT	0xc0010299
144 #define MSR_CORE_ENERGY_STAT	0xc001029a
145 #define MSR_PKG_ENERGY_STAT	0xc001029b
146 
147 #define MAX(a, b) ((a) > (b) ? (a) : (b))
148 
149 /*
150  * buffer size used by sscanf() for added column names
151  * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
152  */
153 #define	NAME_BYTES 20
154 #define PATH_BYTES 128
155 
156 int backwards_count;
157 char *progname;
158 
159 #define CPU_SUBSET_MAXCPUS	1024	/* need to use before probe... */
160 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset;
161 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size;
162 #define MAX_ADDED_COUNTERS 8
163 #define MAX_ADDED_THREAD_COUNTERS 24
164 #define BITMASK_SIZE 32
165 
166 struct thread_data {
167 	struct timeval tv_begin;
168 	struct timeval tv_end;
169 	struct timeval tv_delta;
170 	unsigned long long tsc;
171 	unsigned long long aperf;
172 	unsigned long long mperf;
173 	unsigned long long c1;
174 	unsigned long long  irq_count;
175 	unsigned int smi_count;
176 	unsigned int cpu_id;
177 	unsigned int apic_id;
178 	unsigned int x2apic_id;
179 	unsigned int flags;
180 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
181 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
182 	unsigned long long counter[MAX_ADDED_THREAD_COUNTERS];
183 } *thread_even, *thread_odd;
184 
185 struct core_data {
186 	unsigned long long c3;
187 	unsigned long long c6;
188 	unsigned long long c7;
189 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
190 	unsigned int core_temp_c;
191 	unsigned int core_energy;	/* MSR_CORE_ENERGY_STAT */
192 	unsigned int core_id;
193 	unsigned long long counter[MAX_ADDED_COUNTERS];
194 } *core_even, *core_odd;
195 
196 struct pkg_data {
197 	unsigned long long pc2;
198 	unsigned long long pc3;
199 	unsigned long long pc6;
200 	unsigned long long pc7;
201 	unsigned long long pc8;
202 	unsigned long long pc9;
203 	unsigned long long pc10;
204 	unsigned long long cpu_lpi;
205 	unsigned long long sys_lpi;
206 	unsigned long long pkg_wtd_core_c0;
207 	unsigned long long pkg_any_core_c0;
208 	unsigned long long pkg_any_gfxe_c0;
209 	unsigned long long pkg_both_core_gfxe_c0;
210 	long long gfx_rc6_ms;
211 	unsigned int gfx_mhz;
212 	unsigned int package_id;
213 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
214 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
215 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
216 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
217 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
218 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
219 	unsigned int pkg_temp_c;
220 	unsigned long long counter[MAX_ADDED_COUNTERS];
221 } *package_even, *package_odd;
222 
223 #define ODD_COUNTERS thread_odd, core_odd, package_odd
224 #define EVEN_COUNTERS thread_even, core_even, package_even
225 
226 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no)	      \
227 	((thread_base) +						      \
228 	 ((pkg_no) *							      \
229 	  topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \
230 	 ((node_no) * topo.cores_per_node * topo.threads_per_core) +	      \
231 	 ((core_no) * topo.threads_per_core) +				      \
232 	 (thread_no))
233 
234 #define GET_CORE(core_base, core_no, node_no, pkg_no)			\
235 	((core_base) +							\
236 	 ((pkg_no) *  topo.nodes_per_pkg * topo.cores_per_node) +	\
237 	 ((node_no) * topo.cores_per_node) +				\
238 	 (core_no))
239 
240 
241 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
242 
243 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
244 enum counter_type {COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC};
245 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
246 
247 struct msr_counter {
248 	unsigned int msr_num;
249 	char name[NAME_BYTES];
250 	char path[PATH_BYTES];
251 	unsigned int width;
252 	enum counter_type type;
253 	enum counter_format format;
254 	struct msr_counter *next;
255 	unsigned int flags;
256 #define	FLAGS_HIDE	(1 << 0)
257 #define	FLAGS_SHOW	(1 << 1)
258 #define	SYSFS_PERCPU	(1 << 1)
259 };
260 
261 struct sys_counters {
262 	unsigned int added_thread_counters;
263 	unsigned int added_core_counters;
264 	unsigned int added_package_counters;
265 	struct msr_counter *tp;
266 	struct msr_counter *cp;
267 	struct msr_counter *pp;
268 } sys;
269 
270 struct system_summary {
271 	struct thread_data threads;
272 	struct core_data cores;
273 	struct pkg_data packages;
274 } average;
275 
276 struct cpu_topology {
277 	int physical_package_id;
278 	int die_id;
279 	int logical_cpu_id;
280 	int physical_node_id;
281 	int logical_node_id;	/* 0-based count within the package */
282 	int physical_core_id;
283 	int thread_id;
284 	cpu_set_t *put_ids; /* Processing Unit/Thread IDs */
285 } *cpus;
286 
287 struct topo_params {
288 	int num_packages;
289 	int num_die;
290 	int num_cpus;
291 	int num_cores;
292 	int max_cpu_num;
293 	int max_node_num;
294 	int nodes_per_pkg;
295 	int cores_per_node;
296 	int threads_per_core;
297 } topo;
298 
299 struct timeval tv_even, tv_odd, tv_delta;
300 
301 int *irq_column_2_cpu;	/* /proc/interrupts column numbers */
302 int *irqs_per_cpu;		/* indexed by cpu_num */
303 
304 void setup_all_buffers(void);
305 
306 int cpu_is_not_present(int cpu)
307 {
308 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
309 }
310 /*
311  * run func(thread, core, package) in topology order
312  * skip non-present cpus
313  */
314 
315 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
316 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
317 {
318 	int retval, pkg_no, core_no, thread_no, node_no;
319 
320 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
321 		for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) {
322 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
323 				for (thread_no = 0; thread_no <
324 					topo.threads_per_core; ++thread_no) {
325 					struct thread_data *t;
326 					struct core_data *c;
327 					struct pkg_data *p;
328 
329 					t = GET_THREAD(thread_base, thread_no,
330 						       core_no, node_no,
331 						       pkg_no);
332 
333 					if (cpu_is_not_present(t->cpu_id))
334 						continue;
335 
336 					c = GET_CORE(core_base, core_no,
337 						     node_no, pkg_no);
338 					p = GET_PKG(pkg_base, pkg_no);
339 
340 					retval = func(t, c, p);
341 					if (retval)
342 						return retval;
343 				}
344 			}
345 		}
346 	}
347 	return 0;
348 }
349 
350 int cpu_migrate(int cpu)
351 {
352 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
353 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
354 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
355 		return -1;
356 	else
357 		return 0;
358 }
359 int get_msr_fd(int cpu)
360 {
361 	char pathname[32];
362 	int fd;
363 
364 	fd = fd_percpu[cpu];
365 
366 	if (fd)
367 		return fd;
368 
369 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
370 	fd = open(pathname, O_RDONLY);
371 	if (fd < 0)
372 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
373 
374 	fd_percpu[cpu] = fd;
375 
376 	return fd;
377 }
378 
379 int get_msr(int cpu, off_t offset, unsigned long long *msr)
380 {
381 	ssize_t retval;
382 
383 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
384 
385 	if (retval != sizeof *msr)
386 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
387 
388 	return 0;
389 }
390 
391 /*
392  * This list matches the column headers, except
393  * 1. built-in only, the sysfs counters are not here -- we learn of those at run-time
394  * 2. Core and CPU are moved to the end, we can't have strings that contain them
395  *    matching on them for --show and --hide.
396  */
397 struct msr_counter bic[] = {
398 	{ 0x0, "usec" },
399 	{ 0x0, "Time_Of_Day_Seconds" },
400 	{ 0x0, "Package" },
401 	{ 0x0, "Node" },
402 	{ 0x0, "Avg_MHz" },
403 	{ 0x0, "Busy%" },
404 	{ 0x0, "Bzy_MHz" },
405 	{ 0x0, "TSC_MHz" },
406 	{ 0x0, "IRQ" },
407 	{ 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL},
408 	{ 0x0, "sysfs" },
409 	{ 0x0, "CPU%c1" },
410 	{ 0x0, "CPU%c3" },
411 	{ 0x0, "CPU%c6" },
412 	{ 0x0, "CPU%c7" },
413 	{ 0x0, "ThreadC" },
414 	{ 0x0, "CoreTmp" },
415 	{ 0x0, "CoreCnt" },
416 	{ 0x0, "PkgTmp" },
417 	{ 0x0, "GFX%rc6" },
418 	{ 0x0, "GFXMHz" },
419 	{ 0x0, "Pkg%pc2" },
420 	{ 0x0, "Pkg%pc3" },
421 	{ 0x0, "Pkg%pc6" },
422 	{ 0x0, "Pkg%pc7" },
423 	{ 0x0, "Pkg%pc8" },
424 	{ 0x0, "Pkg%pc9" },
425 	{ 0x0, "Pk%pc10" },
426 	{ 0x0, "CPU%LPI" },
427 	{ 0x0, "SYS%LPI" },
428 	{ 0x0, "PkgWatt" },
429 	{ 0x0, "CorWatt" },
430 	{ 0x0, "GFXWatt" },
431 	{ 0x0, "PkgCnt" },
432 	{ 0x0, "RAMWatt" },
433 	{ 0x0, "PKG_%" },
434 	{ 0x0, "RAM_%" },
435 	{ 0x0, "Pkg_J" },
436 	{ 0x0, "Cor_J" },
437 	{ 0x0, "GFX_J" },
438 	{ 0x0, "RAM_J" },
439 	{ 0x0, "Mod%c6" },
440 	{ 0x0, "Totl%C0" },
441 	{ 0x0, "Any%C0" },
442 	{ 0x0, "GFX%C0" },
443 	{ 0x0, "CPUGFX%" },
444 	{ 0x0, "Core" },
445 	{ 0x0, "CPU" },
446 	{ 0x0, "APIC" },
447 	{ 0x0, "X2APIC" },
448 	{ 0x0, "Die" },
449 };
450 
451 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
452 #define	BIC_USEC	(1ULL << 0)
453 #define	BIC_TOD		(1ULL << 1)
454 #define	BIC_Package	(1ULL << 2)
455 #define	BIC_Node	(1ULL << 3)
456 #define	BIC_Avg_MHz	(1ULL << 4)
457 #define	BIC_Busy	(1ULL << 5)
458 #define	BIC_Bzy_MHz	(1ULL << 6)
459 #define	BIC_TSC_MHz	(1ULL << 7)
460 #define	BIC_IRQ		(1ULL << 8)
461 #define	BIC_SMI		(1ULL << 9)
462 #define	BIC_sysfs	(1ULL << 10)
463 #define	BIC_CPU_c1	(1ULL << 11)
464 #define	BIC_CPU_c3	(1ULL << 12)
465 #define	BIC_CPU_c6	(1ULL << 13)
466 #define	BIC_CPU_c7	(1ULL << 14)
467 #define	BIC_ThreadC	(1ULL << 15)
468 #define	BIC_CoreTmp	(1ULL << 16)
469 #define	BIC_CoreCnt	(1ULL << 17)
470 #define	BIC_PkgTmp	(1ULL << 18)
471 #define	BIC_GFX_rc6	(1ULL << 19)
472 #define	BIC_GFXMHz	(1ULL << 20)
473 #define	BIC_Pkgpc2	(1ULL << 21)
474 #define	BIC_Pkgpc3	(1ULL << 22)
475 #define	BIC_Pkgpc6	(1ULL << 23)
476 #define	BIC_Pkgpc7	(1ULL << 24)
477 #define	BIC_Pkgpc8	(1ULL << 25)
478 #define	BIC_Pkgpc9	(1ULL << 26)
479 #define	BIC_Pkgpc10	(1ULL << 27)
480 #define BIC_CPU_LPI	(1ULL << 28)
481 #define BIC_SYS_LPI	(1ULL << 29)
482 #define	BIC_PkgWatt	(1ULL << 30)
483 #define	BIC_CorWatt	(1ULL << 31)
484 #define	BIC_GFXWatt	(1ULL << 32)
485 #define	BIC_PkgCnt	(1ULL << 33)
486 #define	BIC_RAMWatt	(1ULL << 34)
487 #define	BIC_PKG__	(1ULL << 35)
488 #define	BIC_RAM__	(1ULL << 36)
489 #define	BIC_Pkg_J	(1ULL << 37)
490 #define	BIC_Cor_J	(1ULL << 38)
491 #define	BIC_GFX_J	(1ULL << 39)
492 #define	BIC_RAM_J	(1ULL << 40)
493 #define	BIC_Mod_c6	(1ULL << 41)
494 #define	BIC_Totl_c0	(1ULL << 42)
495 #define	BIC_Any_c0	(1ULL << 43)
496 #define	BIC_GFX_c0	(1ULL << 44)
497 #define	BIC_CPUGFX	(1ULL << 45)
498 #define	BIC_Core	(1ULL << 46)
499 #define	BIC_CPU		(1ULL << 47)
500 #define	BIC_APIC	(1ULL << 48)
501 #define	BIC_X2APIC	(1ULL << 49)
502 #define	BIC_Die		(1ULL << 50)
503 
504 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
505 
506 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT);
507 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC;
508 
509 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
510 #define DO_BIC_READ(COUNTER_NAME) (bic_present & COUNTER_NAME)
511 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
512 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
513 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
514 
515 
516 #define MAX_DEFERRED 16
517 char *deferred_skip_names[MAX_DEFERRED];
518 int deferred_skip_index;
519 
520 /*
521  * HIDE_LIST - hide this list of counters, show the rest [default]
522  * SHOW_LIST - show this list of counters, hide the rest
523  */
524 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
525 
526 void help(void)
527 {
528 	fprintf(outf,
529 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
530 	"\n"
531 	"Turbostat forks the specified COMMAND and prints statistics\n"
532 	"when COMMAND completes.\n"
533 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
534 	"to print statistics, until interrupted.\n"
535 	"  -a, --add	add a counter\n"
536 	"		  eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
537 	"  -c, --cpu	cpu-set	limit output to summary plus cpu-set:\n"
538 	"		  {core | package | j,k,l..m,n-p }\n"
539 	"  -d, --debug	displays usec, Time_Of_Day_Seconds and more debugging\n"
540 	"  -D, --Dump	displays the raw counter values\n"
541 	"  -e, --enable	[all | column]\n"
542 	"		shows all or the specified disabled column\n"
543 	"  -H, --hide [column|column,column,...]\n"
544 	"		hide the specified column(s)\n"
545 	"  -i, --interval sec.subsec\n"
546 	"		Override default 5-second measurement interval\n"
547 	"  -J, --Joules	displays energy in Joules instead of Watts\n"
548 	"  -l, --list	list column headers only\n"
549 	"  -n, --num_iterations num\n"
550 	"		number of the measurement iterations\n"
551 	"  -o, --out file\n"
552 	"		create or truncate \"file\" for all output\n"
553 	"  -q, --quiet	skip decoding system configuration header\n"
554 	"  -s, --show [column|column,column,...]\n"
555 	"		show only the specified column(s)\n"
556 	"  -S, --Summary\n"
557 	"		limits output to 1-line system summary per interval\n"
558 	"  -T, --TCC temperature\n"
559 	"		sets the Thermal Control Circuit temperature in\n"
560 	"		  degrees Celsius\n"
561 	"  -h, --help	print this help message\n"
562 	"  -v, --version	print version information\n"
563 	"\n"
564 	"For more help, run \"man turbostat\"\n");
565 }
566 
567 /*
568  * bic_lookup
569  * for all the strings in comma separate name_list,
570  * set the approprate bit in return value.
571  */
572 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
573 {
574 	int i;
575 	unsigned long long retval = 0;
576 
577 	while (name_list) {
578 		char *comma;
579 
580 		comma = strchr(name_list, ',');
581 
582 		if (comma)
583 			*comma = '\0';
584 
585 		if (!strcmp(name_list, "all"))
586 			return ~0;
587 
588 		for (i = 0; i < MAX_BIC; ++i) {
589 			if (!strcmp(name_list, bic[i].name)) {
590 				retval |= (1ULL << i);
591 				break;
592 			}
593 		}
594 		if (i == MAX_BIC) {
595 			if (mode == SHOW_LIST) {
596 				fprintf(stderr, "Invalid counter name: %s\n", name_list);
597 				exit(-1);
598 			}
599 			deferred_skip_names[deferred_skip_index++] = name_list;
600 			if (debug)
601 				fprintf(stderr, "deferred \"%s\"\n", name_list);
602 			if (deferred_skip_index >= MAX_DEFERRED) {
603 				fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
604 					MAX_DEFERRED, name_list);
605 				help();
606 				exit(1);
607 			}
608 		}
609 
610 		name_list = comma;
611 		if (name_list)
612 			name_list++;
613 
614 	}
615 	return retval;
616 }
617 
618 
619 void print_header(char *delim)
620 {
621 	struct msr_counter *mp;
622 	int printed = 0;
623 
624 	if (DO_BIC(BIC_USEC))
625 		outp += sprintf(outp, "%susec", (printed++ ? delim : ""));
626 	if (DO_BIC(BIC_TOD))
627 		outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : ""));
628 	if (DO_BIC(BIC_Package))
629 		outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
630 	if (DO_BIC(BIC_Die))
631 		outp += sprintf(outp, "%sDie", (printed++ ? delim : ""));
632 	if (DO_BIC(BIC_Node))
633 		outp += sprintf(outp, "%sNode", (printed++ ? delim : ""));
634 	if (DO_BIC(BIC_Core))
635 		outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
636 	if (DO_BIC(BIC_CPU))
637 		outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
638 	if (DO_BIC(BIC_APIC))
639 		outp += sprintf(outp, "%sAPIC", (printed++ ? delim : ""));
640 	if (DO_BIC(BIC_X2APIC))
641 		outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : ""));
642 	if (DO_BIC(BIC_Avg_MHz))
643 		outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
644 	if (DO_BIC(BIC_Busy))
645 		outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
646 	if (DO_BIC(BIC_Bzy_MHz))
647 		outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
648 	if (DO_BIC(BIC_TSC_MHz))
649 		outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
650 
651 	if (DO_BIC(BIC_IRQ)) {
652 		if (sums_need_wide_columns)
653 			outp += sprintf(outp, "%s     IRQ", (printed++ ? delim : ""));
654 		else
655 			outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
656 	}
657 
658 	if (DO_BIC(BIC_SMI))
659 		outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
660 
661 	for (mp = sys.tp; mp; mp = mp->next) {
662 
663 		if (mp->format == FORMAT_RAW) {
664 			if (mp->width == 64)
665 				outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
666 			else
667 				outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
668 		} else {
669 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
670 				outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
671 			else
672 				outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
673 		}
674 	}
675 
676 	if (DO_BIC(BIC_CPU_c1))
677 		outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
678 	if (DO_BIC(BIC_CPU_c3))
679 		outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
680 	if (DO_BIC(BIC_CPU_c6))
681 		outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
682 	if (DO_BIC(BIC_CPU_c7))
683 		outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
684 
685 	if (DO_BIC(BIC_Mod_c6))
686 		outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
687 
688 	if (DO_BIC(BIC_CoreTmp))
689 		outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
690 
691 	if (do_rapl && !rapl_joules) {
692 		if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
693 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
694 	} else if (do_rapl && rapl_joules) {
695 		if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
696 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
697 	}
698 
699 	for (mp = sys.cp; mp; mp = mp->next) {
700 		if (mp->format == FORMAT_RAW) {
701 			if (mp->width == 64)
702 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
703 			else
704 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
705 		} else {
706 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
707 				outp += sprintf(outp, "%s%8s", delim, mp->name);
708 			else
709 				outp += sprintf(outp, "%s%s", delim, mp->name);
710 		}
711 	}
712 
713 	if (DO_BIC(BIC_PkgTmp))
714 		outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
715 
716 	if (DO_BIC(BIC_GFX_rc6))
717 		outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
718 
719 	if (DO_BIC(BIC_GFXMHz))
720 		outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
721 
722 	if (DO_BIC(BIC_Totl_c0))
723 		outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
724 	if (DO_BIC(BIC_Any_c0))
725 		outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
726 	if (DO_BIC(BIC_GFX_c0))
727 		outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
728 	if (DO_BIC(BIC_CPUGFX))
729 		outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
730 
731 	if (DO_BIC(BIC_Pkgpc2))
732 		outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
733 	if (DO_BIC(BIC_Pkgpc3))
734 		outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
735 	if (DO_BIC(BIC_Pkgpc6))
736 		outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
737 	if (DO_BIC(BIC_Pkgpc7))
738 		outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
739 	if (DO_BIC(BIC_Pkgpc8))
740 		outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
741 	if (DO_BIC(BIC_Pkgpc9))
742 		outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
743 	if (DO_BIC(BIC_Pkgpc10))
744 		outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
745 	if (DO_BIC(BIC_CPU_LPI))
746 		outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : ""));
747 	if (DO_BIC(BIC_SYS_LPI))
748 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
749 
750 	if (do_rapl && !rapl_joules) {
751 		if (DO_BIC(BIC_PkgWatt))
752 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
753 		if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
754 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
755 		if (DO_BIC(BIC_GFXWatt))
756 			outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
757 		if (DO_BIC(BIC_RAMWatt))
758 			outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
759 		if (DO_BIC(BIC_PKG__))
760 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
761 		if (DO_BIC(BIC_RAM__))
762 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
763 	} else if (do_rapl && rapl_joules) {
764 		if (DO_BIC(BIC_Pkg_J))
765 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
766 		if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
767 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
768 		if (DO_BIC(BIC_GFX_J))
769 			outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
770 		if (DO_BIC(BIC_RAM_J))
771 			outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
772 		if (DO_BIC(BIC_PKG__))
773 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
774 		if (DO_BIC(BIC_RAM__))
775 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
776 	}
777 	for (mp = sys.pp; mp; mp = mp->next) {
778 		if (mp->format == FORMAT_RAW) {
779 			if (mp->width == 64)
780 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
781 			else
782 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
783 		} else {
784 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
785 				outp += sprintf(outp, "%s%8s", delim, mp->name);
786 			else
787 				outp += sprintf(outp, "%s%s", delim, mp->name);
788 		}
789 	}
790 
791 	outp += sprintf(outp, "\n");
792 }
793 
794 int dump_counters(struct thread_data *t, struct core_data *c,
795 	struct pkg_data *p)
796 {
797 	int i;
798 	struct msr_counter *mp;
799 
800 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
801 
802 	if (t) {
803 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
804 			t->cpu_id, t->flags);
805 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
806 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
807 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
808 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
809 
810 		if (DO_BIC(BIC_IRQ))
811 			outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
812 		if (DO_BIC(BIC_SMI))
813 			outp += sprintf(outp, "SMI: %d\n", t->smi_count);
814 
815 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
816 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
817 				i, mp->msr_num, t->counter[i]);
818 		}
819 	}
820 
821 	if (c) {
822 		outp += sprintf(outp, "core: %d\n", c->core_id);
823 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
824 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
825 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
826 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
827 		outp += sprintf(outp, "Joules: %0X\n", c->core_energy);
828 
829 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
830 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
831 				i, mp->msr_num, c->counter[i]);
832 		}
833 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
834 	}
835 
836 	if (p) {
837 		outp += sprintf(outp, "package: %d\n", p->package_id);
838 
839 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
840 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
841 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
842 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
843 
844 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
845 		if (DO_BIC(BIC_Pkgpc3))
846 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
847 		if (DO_BIC(BIC_Pkgpc6))
848 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
849 		if (DO_BIC(BIC_Pkgpc7))
850 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
851 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
852 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
853 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
854 		outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi);
855 		outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi);
856 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
857 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
858 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
859 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
860 		outp += sprintf(outp, "Throttle PKG: %0X\n",
861 			p->rapl_pkg_perf_status);
862 		outp += sprintf(outp, "Throttle RAM: %0X\n",
863 			p->rapl_dram_perf_status);
864 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
865 
866 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
867 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
868 				i, mp->msr_num, p->counter[i]);
869 		}
870 	}
871 
872 	outp += sprintf(outp, "\n");
873 
874 	return 0;
875 }
876 
877 /*
878  * column formatting convention & formats
879  */
880 int format_counters(struct thread_data *t, struct core_data *c,
881 	struct pkg_data *p)
882 {
883 	double interval_float, tsc;
884 	char *fmt8;
885 	int i;
886 	struct msr_counter *mp;
887 	char *delim = "\t";
888 	int printed = 0;
889 
890 	 /* if showing only 1st thread in core and this isn't one, bail out */
891 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
892 		return 0;
893 
894 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
895 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
896 		return 0;
897 
898 	/*if not summary line and --cpu is used */
899 	if ((t != &average.threads) &&
900 		(cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
901 		return 0;
902 
903 	if (DO_BIC(BIC_USEC)) {
904 		/* on each row, print how many usec each timestamp took to gather */
905 		struct timeval tv;
906 
907 		timersub(&t->tv_end, &t->tv_begin, &tv);
908 		outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec);
909 	}
910 
911 	/* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */
912 	if (DO_BIC(BIC_TOD))
913 		outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec);
914 
915 	interval_float = t->tv_delta.tv_sec + t->tv_delta.tv_usec/1000000.0;
916 
917 	tsc = t->tsc * tsc_tweak;
918 
919 	/* topo columns, print blanks on 1st (average) line */
920 	if (t == &average.threads) {
921 		if (DO_BIC(BIC_Package))
922 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
923 		if (DO_BIC(BIC_Die))
924 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
925 		if (DO_BIC(BIC_Node))
926 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
927 		if (DO_BIC(BIC_Core))
928 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
929 		if (DO_BIC(BIC_CPU))
930 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
931 		if (DO_BIC(BIC_APIC))
932 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
933 		if (DO_BIC(BIC_X2APIC))
934 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
935 	} else {
936 		if (DO_BIC(BIC_Package)) {
937 			if (p)
938 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
939 			else
940 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
941 		}
942 		if (DO_BIC(BIC_Die)) {
943 			if (c)
944 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id);
945 			else
946 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
947 		}
948 		if (DO_BIC(BIC_Node)) {
949 			if (t)
950 				outp += sprintf(outp, "%s%d",
951 						(printed++ ? delim : ""),
952 					      cpus[t->cpu_id].physical_node_id);
953 			else
954 				outp += sprintf(outp, "%s-",
955 						(printed++ ? delim : ""));
956 		}
957 		if (DO_BIC(BIC_Core)) {
958 			if (c)
959 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
960 			else
961 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
962 		}
963 		if (DO_BIC(BIC_CPU))
964 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
965 		if (DO_BIC(BIC_APIC))
966 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id);
967 		if (DO_BIC(BIC_X2APIC))
968 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id);
969 	}
970 
971 	if (DO_BIC(BIC_Avg_MHz))
972 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
973 			1.0 / units * t->aperf / interval_float);
974 
975 	if (DO_BIC(BIC_Busy))
976 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf/tsc);
977 
978 	if (DO_BIC(BIC_Bzy_MHz)) {
979 		if (has_base_hz)
980 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
981 		else
982 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
983 				tsc / units * t->aperf / t->mperf / interval_float);
984 	}
985 
986 	if (DO_BIC(BIC_TSC_MHz))
987 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
988 
989 	/* IRQ */
990 	if (DO_BIC(BIC_IRQ)) {
991 		if (sums_need_wide_columns)
992 			outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
993 		else
994 			outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
995 	}
996 
997 	/* SMI */
998 	if (DO_BIC(BIC_SMI))
999 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
1000 
1001 	/* Added counters */
1002 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1003 		if (mp->format == FORMAT_RAW) {
1004 			if (mp->width == 32)
1005 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) t->counter[i]);
1006 			else
1007 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
1008 		} else if (mp->format == FORMAT_DELTA) {
1009 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1010 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
1011 			else
1012 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
1013 		} else if (mp->format == FORMAT_PERCENT) {
1014 			if (mp->type == COUNTER_USEC)
1015 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), t->counter[i]/interval_float/10000);
1016 			else
1017 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i]/tsc);
1018 		}
1019 	}
1020 
1021 	/* C1 */
1022 	if (DO_BIC(BIC_CPU_c1))
1023 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1/tsc);
1024 
1025 
1026 	/* print per-core data only for 1st thread in core */
1027 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1028 		goto done;
1029 
1030 	if (DO_BIC(BIC_CPU_c3))
1031 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3/tsc);
1032 	if (DO_BIC(BIC_CPU_c6))
1033 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6/tsc);
1034 	if (DO_BIC(BIC_CPU_c7))
1035 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7/tsc);
1036 
1037 	/* Mod%c6 */
1038 	if (DO_BIC(BIC_Mod_c6))
1039 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
1040 
1041 	if (DO_BIC(BIC_CoreTmp))
1042 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
1043 
1044 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1045 		if (mp->format == FORMAT_RAW) {
1046 			if (mp->width == 32)
1047 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) c->counter[i]);
1048 			else
1049 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
1050 		} else if (mp->format == FORMAT_DELTA) {
1051 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1052 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
1053 			else
1054 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
1055 		} else if (mp->format == FORMAT_PERCENT) {
1056 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i]/tsc);
1057 		}
1058 	}
1059 
1060 	/*
1061 	 * If measurement interval exceeds minimum RAPL Joule Counter range,
1062 	 * indicate that results are suspect by printing "**" in fraction place.
1063 	 */
1064 	if (interval_float < rapl_joule_counter_range)
1065 		fmt8 = "%s%.2f";
1066 	else
1067 		fmt8 = "%6.0f**";
1068 
1069 	if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
1070 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float);
1071 	if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
1072 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units);
1073 
1074 	/* print per-package data only for 1st core in package */
1075 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1076 		goto done;
1077 
1078 	/* PkgTmp */
1079 	if (DO_BIC(BIC_PkgTmp))
1080 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
1081 
1082 	/* GFXrc6 */
1083 	if (DO_BIC(BIC_GFX_rc6)) {
1084 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
1085 			outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
1086 		} else {
1087 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1088 				p->gfx_rc6_ms / 10.0 / interval_float);
1089 		}
1090 	}
1091 
1092 	/* GFXMHz */
1093 	if (DO_BIC(BIC_GFXMHz))
1094 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
1095 
1096 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
1097 	if (DO_BIC(BIC_Totl_c0))
1098 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0/tsc);
1099 	if (DO_BIC(BIC_Any_c0))
1100 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0/tsc);
1101 	if (DO_BIC(BIC_GFX_c0))
1102 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0/tsc);
1103 	if (DO_BIC(BIC_CPUGFX))
1104 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0/tsc);
1105 
1106 	if (DO_BIC(BIC_Pkgpc2))
1107 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2/tsc);
1108 	if (DO_BIC(BIC_Pkgpc3))
1109 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3/tsc);
1110 	if (DO_BIC(BIC_Pkgpc6))
1111 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6/tsc);
1112 	if (DO_BIC(BIC_Pkgpc7))
1113 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7/tsc);
1114 	if (DO_BIC(BIC_Pkgpc8))
1115 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8/tsc);
1116 	if (DO_BIC(BIC_Pkgpc9))
1117 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9/tsc);
1118 	if (DO_BIC(BIC_Pkgpc10))
1119 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10/tsc);
1120 
1121 	if (DO_BIC(BIC_CPU_LPI))
1122 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float);
1123 	if (DO_BIC(BIC_SYS_LPI))
1124 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float);
1125 
1126 	if (DO_BIC(BIC_PkgWatt))
1127 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
1128 	if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1129 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
1130 	if (DO_BIC(BIC_GFXWatt))
1131 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
1132 	if (DO_BIC(BIC_RAMWatt))
1133 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units / interval_float);
1134 	if (DO_BIC(BIC_Pkg_J))
1135 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
1136 	if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1137 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
1138 	if (DO_BIC(BIC_GFX_J))
1139 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
1140 	if (DO_BIC(BIC_RAM_J))
1141 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
1142 	if (DO_BIC(BIC_PKG__))
1143 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
1144 	if (DO_BIC(BIC_RAM__))
1145 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
1146 
1147 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1148 		if (mp->format == FORMAT_RAW) {
1149 			if (mp->width == 32)
1150 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) p->counter[i]);
1151 			else
1152 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
1153 		} else if (mp->format == FORMAT_DELTA) {
1154 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1155 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
1156 			else
1157 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
1158 		} else if (mp->format == FORMAT_PERCENT) {
1159 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i]/tsc);
1160 		}
1161 	}
1162 
1163 done:
1164 	if (*(outp - 1) != '\n')
1165 		outp += sprintf(outp, "\n");
1166 
1167 	return 0;
1168 }
1169 
1170 void flush_output_stdout(void)
1171 {
1172 	FILE *filep;
1173 
1174 	if (outf == stderr)
1175 		filep = stdout;
1176 	else
1177 		filep = outf;
1178 
1179 	fputs(output_buffer, filep);
1180 	fflush(filep);
1181 
1182 	outp = output_buffer;
1183 }
1184 void flush_output_stderr(void)
1185 {
1186 	fputs(output_buffer, outf);
1187 	fflush(outf);
1188 	outp = output_buffer;
1189 }
1190 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1191 {
1192 	static int printed;
1193 
1194 	if (!printed || !summary_only)
1195 		print_header("\t");
1196 
1197 	format_counters(&average.threads, &average.cores, &average.packages);
1198 
1199 	printed = 1;
1200 
1201 	if (summary_only)
1202 		return;
1203 
1204 	for_all_cpus(format_counters, t, c, p);
1205 }
1206 
1207 #define DELTA_WRAP32(new, old)			\
1208 	if (new > old) {			\
1209 		old = new - old;		\
1210 	} else {				\
1211 		old = 0x100000000 + new - old;	\
1212 	}
1213 
1214 int
1215 delta_package(struct pkg_data *new, struct pkg_data *old)
1216 {
1217 	int i;
1218 	struct msr_counter *mp;
1219 
1220 
1221 	if (DO_BIC(BIC_Totl_c0))
1222 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1223 	if (DO_BIC(BIC_Any_c0))
1224 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1225 	if (DO_BIC(BIC_GFX_c0))
1226 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1227 	if (DO_BIC(BIC_CPUGFX))
1228 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1229 
1230 	old->pc2 = new->pc2 - old->pc2;
1231 	if (DO_BIC(BIC_Pkgpc3))
1232 		old->pc3 = new->pc3 - old->pc3;
1233 	if (DO_BIC(BIC_Pkgpc6))
1234 		old->pc6 = new->pc6 - old->pc6;
1235 	if (DO_BIC(BIC_Pkgpc7))
1236 		old->pc7 = new->pc7 - old->pc7;
1237 	old->pc8 = new->pc8 - old->pc8;
1238 	old->pc9 = new->pc9 - old->pc9;
1239 	old->pc10 = new->pc10 - old->pc10;
1240 	old->cpu_lpi = new->cpu_lpi - old->cpu_lpi;
1241 	old->sys_lpi = new->sys_lpi - old->sys_lpi;
1242 	old->pkg_temp_c = new->pkg_temp_c;
1243 
1244 	/* flag an error when rc6 counter resets/wraps */
1245 	if (old->gfx_rc6_ms >  new->gfx_rc6_ms)
1246 		old->gfx_rc6_ms = -1;
1247 	else
1248 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1249 
1250 	old->gfx_mhz = new->gfx_mhz;
1251 
1252 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
1253 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
1254 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
1255 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
1256 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
1257 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
1258 
1259 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1260 		if (mp->format == FORMAT_RAW)
1261 			old->counter[i] = new->counter[i];
1262 		else
1263 			old->counter[i] = new->counter[i] - old->counter[i];
1264 	}
1265 
1266 	return 0;
1267 }
1268 
1269 void
1270 delta_core(struct core_data *new, struct core_data *old)
1271 {
1272 	int i;
1273 	struct msr_counter *mp;
1274 
1275 	old->c3 = new->c3 - old->c3;
1276 	old->c6 = new->c6 - old->c6;
1277 	old->c7 = new->c7 - old->c7;
1278 	old->core_temp_c = new->core_temp_c;
1279 	old->mc6_us = new->mc6_us - old->mc6_us;
1280 
1281 	DELTA_WRAP32(new->core_energy, old->core_energy);
1282 
1283 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1284 		if (mp->format == FORMAT_RAW)
1285 			old->counter[i] = new->counter[i];
1286 		else
1287 			old->counter[i] = new->counter[i] - old->counter[i];
1288 	}
1289 }
1290 
1291 int soft_c1_residency_display(int bic)
1292 {
1293 	if (!DO_BIC(BIC_CPU_c1) || use_c1_residency_msr)
1294 		return 0;
1295 
1296 	return DO_BIC_READ(bic);
1297 }
1298 
1299 /*
1300  * old = new - old
1301  */
1302 int
1303 delta_thread(struct thread_data *new, struct thread_data *old,
1304 	struct core_data *core_delta)
1305 {
1306 	int i;
1307 	struct msr_counter *mp;
1308 
1309 	/* we run cpuid just the 1st time, copy the results */
1310 	if (DO_BIC(BIC_APIC))
1311 		new->apic_id = old->apic_id;
1312 	if (DO_BIC(BIC_X2APIC))
1313 		new->x2apic_id = old->x2apic_id;
1314 
1315 	/*
1316 	 * the timestamps from start of measurement interval are in "old"
1317 	 * the timestamp from end of measurement interval are in "new"
1318 	 * over-write old w/ new so we can print end of interval values
1319 	 */
1320 
1321 	timersub(&new->tv_begin, &old->tv_begin, &old->tv_delta);
1322 	old->tv_begin = new->tv_begin;
1323 	old->tv_end = new->tv_end;
1324 
1325 	old->tsc = new->tsc - old->tsc;
1326 
1327 	/* check for TSC < 1 Mcycles over interval */
1328 	if (old->tsc < (1000 * 1000))
1329 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1330 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1331 		     "or just the deep ones with \"processor.max_cstate=1\"");
1332 
1333 	old->c1 = new->c1 - old->c1;
1334 
1335 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) ||
1336 	    soft_c1_residency_display(BIC_Avg_MHz)) {
1337 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1338 			old->aperf = new->aperf - old->aperf;
1339 			old->mperf = new->mperf - old->mperf;
1340 		} else {
1341 			return -1;
1342 		}
1343 	}
1344 
1345 
1346 	if (use_c1_residency_msr) {
1347 		/*
1348 		 * Some models have a dedicated C1 residency MSR,
1349 		 * which should be more accurate than the derivation below.
1350 		 */
1351 	} else {
1352 		/*
1353 		 * As counter collection is not atomic,
1354 		 * it is possible for mperf's non-halted cycles + idle states
1355 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1356 		 */
1357 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1358 			old->c1 = 0;
1359 		else {
1360 			/* normal case, derive c1 */
1361 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1362 				- core_delta->c6 - core_delta->c7;
1363 		}
1364 	}
1365 
1366 	if (old->mperf == 0) {
1367 		if (debug > 1)
1368 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1369 		old->mperf = 1;	/* divide by 0 protection */
1370 	}
1371 
1372 	if (DO_BIC(BIC_IRQ))
1373 		old->irq_count = new->irq_count - old->irq_count;
1374 
1375 	if (DO_BIC(BIC_SMI))
1376 		old->smi_count = new->smi_count - old->smi_count;
1377 
1378 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1379 		if (mp->format == FORMAT_RAW)
1380 			old->counter[i] = new->counter[i];
1381 		else
1382 			old->counter[i] = new->counter[i] - old->counter[i];
1383 	}
1384 	return 0;
1385 }
1386 
1387 int delta_cpu(struct thread_data *t, struct core_data *c,
1388 	struct pkg_data *p, struct thread_data *t2,
1389 	struct core_data *c2, struct pkg_data *p2)
1390 {
1391 	int retval = 0;
1392 
1393 	/* calculate core delta only for 1st thread in core */
1394 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1395 		delta_core(c, c2);
1396 
1397 	/* always calculate thread delta */
1398 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1399 	if (retval)
1400 		return retval;
1401 
1402 	/* calculate package delta only for 1st core in package */
1403 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1404 		retval = delta_package(p, p2);
1405 
1406 	return retval;
1407 }
1408 
1409 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1410 {
1411 	int i;
1412 	struct msr_counter  *mp;
1413 
1414 	t->tv_begin.tv_sec = 0;
1415 	t->tv_begin.tv_usec = 0;
1416 	t->tv_end.tv_sec = 0;
1417 	t->tv_end.tv_usec = 0;
1418 	t->tv_delta.tv_sec = 0;
1419 	t->tv_delta.tv_usec = 0;
1420 
1421 	t->tsc = 0;
1422 	t->aperf = 0;
1423 	t->mperf = 0;
1424 	t->c1 = 0;
1425 
1426 	t->irq_count = 0;
1427 	t->smi_count = 0;
1428 
1429 	/* tells format_counters to dump all fields from this set */
1430 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1431 
1432 	c->c3 = 0;
1433 	c->c6 = 0;
1434 	c->c7 = 0;
1435 	c->mc6_us = 0;
1436 	c->core_temp_c = 0;
1437 	c->core_energy = 0;
1438 
1439 	p->pkg_wtd_core_c0 = 0;
1440 	p->pkg_any_core_c0 = 0;
1441 	p->pkg_any_gfxe_c0 = 0;
1442 	p->pkg_both_core_gfxe_c0 = 0;
1443 
1444 	p->pc2 = 0;
1445 	if (DO_BIC(BIC_Pkgpc3))
1446 		p->pc3 = 0;
1447 	if (DO_BIC(BIC_Pkgpc6))
1448 		p->pc6 = 0;
1449 	if (DO_BIC(BIC_Pkgpc7))
1450 		p->pc7 = 0;
1451 	p->pc8 = 0;
1452 	p->pc9 = 0;
1453 	p->pc10 = 0;
1454 	p->cpu_lpi = 0;
1455 	p->sys_lpi = 0;
1456 
1457 	p->energy_pkg = 0;
1458 	p->energy_dram = 0;
1459 	p->energy_cores = 0;
1460 	p->energy_gfx = 0;
1461 	p->rapl_pkg_perf_status = 0;
1462 	p->rapl_dram_perf_status = 0;
1463 	p->pkg_temp_c = 0;
1464 
1465 	p->gfx_rc6_ms = 0;
1466 	p->gfx_mhz = 0;
1467 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1468 		t->counter[i] = 0;
1469 
1470 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1471 		c->counter[i] = 0;
1472 
1473 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1474 		p->counter[i] = 0;
1475 }
1476 int sum_counters(struct thread_data *t, struct core_data *c,
1477 	struct pkg_data *p)
1478 {
1479 	int i;
1480 	struct msr_counter *mp;
1481 
1482 	/* copy un-changing apic_id's */
1483 	if (DO_BIC(BIC_APIC))
1484 		average.threads.apic_id = t->apic_id;
1485 	if (DO_BIC(BIC_X2APIC))
1486 		average.threads.x2apic_id = t->x2apic_id;
1487 
1488 	/* remember first tv_begin */
1489 	if (average.threads.tv_begin.tv_sec == 0)
1490 		average.threads.tv_begin = t->tv_begin;
1491 
1492 	/* remember last tv_end */
1493 	average.threads.tv_end = t->tv_end;
1494 
1495 	average.threads.tsc += t->tsc;
1496 	average.threads.aperf += t->aperf;
1497 	average.threads.mperf += t->mperf;
1498 	average.threads.c1 += t->c1;
1499 
1500 	average.threads.irq_count += t->irq_count;
1501 	average.threads.smi_count += t->smi_count;
1502 
1503 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1504 		if (mp->format == FORMAT_RAW)
1505 			continue;
1506 		average.threads.counter[i] += t->counter[i];
1507 	}
1508 
1509 	/* sum per-core values only for 1st thread in core */
1510 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1511 		return 0;
1512 
1513 	average.cores.c3 += c->c3;
1514 	average.cores.c6 += c->c6;
1515 	average.cores.c7 += c->c7;
1516 	average.cores.mc6_us += c->mc6_us;
1517 
1518 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1519 
1520 	average.cores.core_energy += c->core_energy;
1521 
1522 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1523 		if (mp->format == FORMAT_RAW)
1524 			continue;
1525 		average.cores.counter[i] += c->counter[i];
1526 	}
1527 
1528 	/* sum per-pkg values only for 1st core in pkg */
1529 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1530 		return 0;
1531 
1532 	if (DO_BIC(BIC_Totl_c0))
1533 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1534 	if (DO_BIC(BIC_Any_c0))
1535 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1536 	if (DO_BIC(BIC_GFX_c0))
1537 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1538 	if (DO_BIC(BIC_CPUGFX))
1539 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1540 
1541 	average.packages.pc2 += p->pc2;
1542 	if (DO_BIC(BIC_Pkgpc3))
1543 		average.packages.pc3 += p->pc3;
1544 	if (DO_BIC(BIC_Pkgpc6))
1545 		average.packages.pc6 += p->pc6;
1546 	if (DO_BIC(BIC_Pkgpc7))
1547 		average.packages.pc7 += p->pc7;
1548 	average.packages.pc8 += p->pc8;
1549 	average.packages.pc9 += p->pc9;
1550 	average.packages.pc10 += p->pc10;
1551 
1552 	average.packages.cpu_lpi = p->cpu_lpi;
1553 	average.packages.sys_lpi = p->sys_lpi;
1554 
1555 	average.packages.energy_pkg += p->energy_pkg;
1556 	average.packages.energy_dram += p->energy_dram;
1557 	average.packages.energy_cores += p->energy_cores;
1558 	average.packages.energy_gfx += p->energy_gfx;
1559 
1560 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1561 	average.packages.gfx_mhz = p->gfx_mhz;
1562 
1563 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1564 
1565 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1566 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1567 
1568 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1569 		if (mp->format == FORMAT_RAW)
1570 			continue;
1571 		average.packages.counter[i] += p->counter[i];
1572 	}
1573 	return 0;
1574 }
1575 /*
1576  * sum the counters for all cpus in the system
1577  * compute the weighted average
1578  */
1579 void compute_average(struct thread_data *t, struct core_data *c,
1580 	struct pkg_data *p)
1581 {
1582 	int i;
1583 	struct msr_counter *mp;
1584 
1585 	clear_counters(&average.threads, &average.cores, &average.packages);
1586 
1587 	for_all_cpus(sum_counters, t, c, p);
1588 
1589 	/* Use the global time delta for the average. */
1590 	average.threads.tv_delta = tv_delta;
1591 
1592 	average.threads.tsc /= topo.num_cpus;
1593 	average.threads.aperf /= topo.num_cpus;
1594 	average.threads.mperf /= topo.num_cpus;
1595 	average.threads.c1 /= topo.num_cpus;
1596 
1597 	if (average.threads.irq_count > 9999999)
1598 		sums_need_wide_columns = 1;
1599 
1600 	average.cores.c3 /= topo.num_cores;
1601 	average.cores.c6 /= topo.num_cores;
1602 	average.cores.c7 /= topo.num_cores;
1603 	average.cores.mc6_us /= topo.num_cores;
1604 
1605 	if (DO_BIC(BIC_Totl_c0))
1606 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1607 	if (DO_BIC(BIC_Any_c0))
1608 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1609 	if (DO_BIC(BIC_GFX_c0))
1610 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1611 	if (DO_BIC(BIC_CPUGFX))
1612 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1613 
1614 	average.packages.pc2 /= topo.num_packages;
1615 	if (DO_BIC(BIC_Pkgpc3))
1616 		average.packages.pc3 /= topo.num_packages;
1617 	if (DO_BIC(BIC_Pkgpc6))
1618 		average.packages.pc6 /= topo.num_packages;
1619 	if (DO_BIC(BIC_Pkgpc7))
1620 		average.packages.pc7 /= topo.num_packages;
1621 
1622 	average.packages.pc8 /= topo.num_packages;
1623 	average.packages.pc9 /= topo.num_packages;
1624 	average.packages.pc10 /= topo.num_packages;
1625 
1626 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1627 		if (mp->format == FORMAT_RAW)
1628 			continue;
1629 		if (mp->type == COUNTER_ITEMS) {
1630 			if (average.threads.counter[i] > 9999999)
1631 				sums_need_wide_columns = 1;
1632 			continue;
1633 		}
1634 		average.threads.counter[i] /= topo.num_cpus;
1635 	}
1636 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1637 		if (mp->format == FORMAT_RAW)
1638 			continue;
1639 		if (mp->type == COUNTER_ITEMS) {
1640 			if (average.cores.counter[i] > 9999999)
1641 				sums_need_wide_columns = 1;
1642 		}
1643 		average.cores.counter[i] /= topo.num_cores;
1644 	}
1645 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1646 		if (mp->format == FORMAT_RAW)
1647 			continue;
1648 		if (mp->type == COUNTER_ITEMS) {
1649 			if (average.packages.counter[i] > 9999999)
1650 				sums_need_wide_columns = 1;
1651 		}
1652 		average.packages.counter[i] /= topo.num_packages;
1653 	}
1654 }
1655 
1656 static unsigned long long rdtsc(void)
1657 {
1658 	unsigned int low, high;
1659 
1660 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
1661 
1662 	return low | ((unsigned long long)high) << 32;
1663 }
1664 
1665 /*
1666  * Open a file, and exit on failure
1667  */
1668 FILE *fopen_or_die(const char *path, const char *mode)
1669 {
1670 	FILE *filep = fopen(path, mode);
1671 
1672 	if (!filep)
1673 		err(1, "%s: open failed", path);
1674 	return filep;
1675 }
1676 /*
1677  * snapshot_sysfs_counter()
1678  *
1679  * return snapshot of given counter
1680  */
1681 unsigned long long snapshot_sysfs_counter(char *path)
1682 {
1683 	FILE *fp;
1684 	int retval;
1685 	unsigned long long counter;
1686 
1687 	fp = fopen_or_die(path, "r");
1688 
1689 	retval = fscanf(fp, "%lld", &counter);
1690 	if (retval != 1)
1691 		err(1, "snapshot_sysfs_counter(%s)", path);
1692 
1693 	fclose(fp);
1694 
1695 	return counter;
1696 }
1697 
1698 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1699 {
1700 	if (mp->msr_num != 0) {
1701 		if (get_msr(cpu, mp->msr_num, counterp))
1702 			return -1;
1703 	} else {
1704 		char path[128 + PATH_BYTES];
1705 
1706 		if (mp->flags & SYSFS_PERCPU) {
1707 			sprintf(path, "/sys/devices/system/cpu/cpu%d/%s",
1708 				 cpu, mp->path);
1709 
1710 			*counterp = snapshot_sysfs_counter(path);
1711 		} else {
1712 			*counterp = snapshot_sysfs_counter(mp->path);
1713 		}
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 void get_apic_id(struct thread_data *t)
1720 {
1721 	unsigned int eax, ebx, ecx, edx;
1722 
1723 	if (DO_BIC(BIC_APIC)) {
1724 		eax = ebx = ecx = edx = 0;
1725 		__cpuid(1, eax, ebx, ecx, edx);
1726 
1727 		t->apic_id = (ebx >> 24) & 0xff;
1728 	}
1729 
1730 	if (!DO_BIC(BIC_X2APIC))
1731 		return;
1732 
1733 	if (authentic_amd) {
1734 		unsigned int topology_extensions;
1735 
1736 		if (max_extended_level < 0x8000001e)
1737 			return;
1738 
1739 		eax = ebx = ecx = edx = 0;
1740 		__cpuid(0x80000001, eax, ebx, ecx, edx);
1741 			topology_extensions = ecx & (1 << 22);
1742 
1743 		if (topology_extensions == 0)
1744 			return;
1745 
1746 		eax = ebx = ecx = edx = 0;
1747 		__cpuid(0x8000001e, eax, ebx, ecx, edx);
1748 
1749 		t->x2apic_id = eax;
1750 		return;
1751 	}
1752 
1753 	if (!genuine_intel)
1754 		return;
1755 
1756 	if (max_level < 0xb)
1757 		return;
1758 
1759 	ecx = 0;
1760 	__cpuid(0xb, eax, ebx, ecx, edx);
1761 	t->x2apic_id = edx;
1762 
1763 	if (debug && (t->apic_id != (t->x2apic_id & 0xff)))
1764 		fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n",
1765 				t->cpu_id, t->apic_id, t->x2apic_id);
1766 }
1767 
1768 /*
1769  * get_counters(...)
1770  * migrate to cpu
1771  * acquire and record local counters for that cpu
1772  */
1773 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1774 {
1775 	int cpu = t->cpu_id;
1776 	unsigned long long msr;
1777 	int aperf_mperf_retry_count = 0;
1778 	struct msr_counter *mp;
1779 	int i;
1780 
1781 	if (cpu_migrate(cpu)) {
1782 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1783 		return -1;
1784 	}
1785 
1786 	gettimeofday(&t->tv_begin, (struct timezone *)NULL);
1787 
1788 	if (first_counter_read)
1789 		get_apic_id(t);
1790 retry:
1791 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
1792 
1793 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) ||
1794 	    soft_c1_residency_display(BIC_Avg_MHz)) {
1795 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1796 
1797 		/*
1798 		 * The TSC, APERF and MPERF must be read together for
1799 		 * APERF/MPERF and MPERF/TSC to give accurate results.
1800 		 *
1801 		 * Unfortunately, APERF and MPERF are read by
1802 		 * individual system call, so delays may occur
1803 		 * between them.  If the time to read them
1804 		 * varies by a large amount, we re-read them.
1805 		 */
1806 
1807 		/*
1808 		 * This initial dummy APERF read has been seen to
1809 		 * reduce jitter in the subsequent reads.
1810 		 */
1811 
1812 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1813 			return -3;
1814 
1815 		t->tsc = rdtsc();	/* re-read close to APERF */
1816 
1817 		tsc_before = t->tsc;
1818 
1819 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1820 			return -3;
1821 
1822 		tsc_between = rdtsc();
1823 
1824 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1825 			return -4;
1826 
1827 		tsc_after = rdtsc();
1828 
1829 		aperf_time = tsc_between - tsc_before;
1830 		mperf_time = tsc_after - tsc_between;
1831 
1832 		/*
1833 		 * If the system call latency to read APERF and MPERF
1834 		 * differ by more than 2x, then try again.
1835 		 */
1836 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1837 			aperf_mperf_retry_count++;
1838 			if (aperf_mperf_retry_count < 5)
1839 				goto retry;
1840 			else
1841 				warnx("cpu%d jitter %lld %lld",
1842 					cpu, aperf_time, mperf_time);
1843 		}
1844 		aperf_mperf_retry_count = 0;
1845 
1846 		t->aperf = t->aperf * aperf_mperf_multiplier;
1847 		t->mperf = t->mperf * aperf_mperf_multiplier;
1848 	}
1849 
1850 	if (DO_BIC(BIC_IRQ))
1851 		t->irq_count = irqs_per_cpu[cpu];
1852 	if (DO_BIC(BIC_SMI)) {
1853 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1854 			return -5;
1855 		t->smi_count = msr & 0xFFFFFFFF;
1856 	}
1857 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1858 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1859 			return -6;
1860 	}
1861 
1862 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1863 		if (get_mp(cpu, mp, &t->counter[i]))
1864 			return -10;
1865 	}
1866 
1867 	/* collect core counters only for 1st thread in core */
1868 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1869 		goto done;
1870 
1871 	if (DO_BIC(BIC_CPU_c3) || soft_c1_residency_display(BIC_CPU_c3)) {
1872 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1873 			return -6;
1874 	}
1875 
1876 	if ((DO_BIC(BIC_CPU_c6) || soft_c1_residency_display(BIC_CPU_c6)) && !do_knl_cstates) {
1877 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1878 			return -7;
1879 	} else if (do_knl_cstates || soft_c1_residency_display(BIC_CPU_c6)) {
1880 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1881 			return -7;
1882 	}
1883 
1884 	if (DO_BIC(BIC_CPU_c7) || soft_c1_residency_display(BIC_CPU_c7))
1885 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1886 			return -8;
1887 
1888 	if (DO_BIC(BIC_Mod_c6))
1889 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1890 			return -8;
1891 
1892 	if (DO_BIC(BIC_CoreTmp)) {
1893 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1894 			return -9;
1895 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1896 	}
1897 
1898 	if (do_rapl & RAPL_AMD_F17H) {
1899 		if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr))
1900 			return -14;
1901 		c->core_energy = msr & 0xFFFFFFFF;
1902 	}
1903 
1904 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1905 		if (get_mp(cpu, mp, &c->counter[i]))
1906 			return -10;
1907 	}
1908 
1909 	/* collect package counters only for 1st core in package */
1910 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1911 		goto done;
1912 
1913 	if (DO_BIC(BIC_Totl_c0)) {
1914 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1915 			return -10;
1916 	}
1917 	if (DO_BIC(BIC_Any_c0)) {
1918 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1919 			return -11;
1920 	}
1921 	if (DO_BIC(BIC_GFX_c0)) {
1922 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1923 			return -12;
1924 	}
1925 	if (DO_BIC(BIC_CPUGFX)) {
1926 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1927 			return -13;
1928 	}
1929 	if (DO_BIC(BIC_Pkgpc3))
1930 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1931 			return -9;
1932 	if (DO_BIC(BIC_Pkgpc6)) {
1933 		if (do_slm_cstates) {
1934 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1935 				return -10;
1936 		} else {
1937 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1938 				return -10;
1939 		}
1940 	}
1941 
1942 	if (DO_BIC(BIC_Pkgpc2))
1943 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1944 			return -11;
1945 	if (DO_BIC(BIC_Pkgpc7))
1946 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1947 			return -12;
1948 	if (DO_BIC(BIC_Pkgpc8))
1949 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1950 			return -13;
1951 	if (DO_BIC(BIC_Pkgpc9))
1952 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1953 			return -13;
1954 	if (DO_BIC(BIC_Pkgpc10))
1955 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1956 			return -13;
1957 
1958 	if (DO_BIC(BIC_CPU_LPI))
1959 		p->cpu_lpi = cpuidle_cur_cpu_lpi_us;
1960 	if (DO_BIC(BIC_SYS_LPI))
1961 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
1962 
1963 	if (do_rapl & RAPL_PKG) {
1964 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1965 			return -13;
1966 		p->energy_pkg = msr & 0xFFFFFFFF;
1967 	}
1968 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1969 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1970 			return -14;
1971 		p->energy_cores = msr & 0xFFFFFFFF;
1972 	}
1973 	if (do_rapl & RAPL_DRAM) {
1974 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1975 			return -15;
1976 		p->energy_dram = msr & 0xFFFFFFFF;
1977 	}
1978 	if (do_rapl & RAPL_GFX) {
1979 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1980 			return -16;
1981 		p->energy_gfx = msr & 0xFFFFFFFF;
1982 	}
1983 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1984 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1985 			return -16;
1986 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1987 	}
1988 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1989 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1990 			return -16;
1991 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1992 	}
1993 	if (do_rapl & RAPL_AMD_F17H) {
1994 		if (get_msr(cpu, MSR_PKG_ENERGY_STAT, &msr))
1995 			return -13;
1996 		p->energy_pkg = msr & 0xFFFFFFFF;
1997 	}
1998 	if (DO_BIC(BIC_PkgTmp)) {
1999 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
2000 			return -17;
2001 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
2002 	}
2003 
2004 	if (DO_BIC(BIC_GFX_rc6))
2005 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
2006 
2007 	if (DO_BIC(BIC_GFXMHz))
2008 		p->gfx_mhz = gfx_cur_mhz;
2009 
2010 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
2011 		if (get_mp(cpu, mp, &p->counter[i]))
2012 			return -10;
2013 	}
2014 done:
2015 	gettimeofday(&t->tv_end, (struct timezone *)NULL);
2016 
2017 	return 0;
2018 }
2019 
2020 /*
2021  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
2022  * If you change the values, note they are used both in comparisons
2023  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
2024  */
2025 
2026 #define PCLUKN 0 /* Unknown */
2027 #define PCLRSV 1 /* Reserved */
2028 #define PCL__0 2 /* PC0 */
2029 #define PCL__1 3 /* PC1 */
2030 #define PCL__2 4 /* PC2 */
2031 #define PCL__3 5 /* PC3 */
2032 #define PCL__4 6 /* PC4 */
2033 #define PCL__6 7 /* PC6 */
2034 #define PCL_6N 8 /* PC6 No Retention */
2035 #define PCL_6R 9 /* PC6 Retention */
2036 #define PCL__7 10 /* PC7 */
2037 #define PCL_7S 11 /* PC7 Shrink */
2038 #define PCL__8 12 /* PC8 */
2039 #define PCL__9 13 /* PC9 */
2040 #define PCL_10 14 /* PC10 */
2041 #define PCLUNL 15 /* Unlimited */
2042 
2043 int pkg_cstate_limit = PCLUKN;
2044 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
2045 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited"};
2046 
2047 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};
2048 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};
2049 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};
2050 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};
2051 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};
2052 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};
2053 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};
2054 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};
2055 
2056 
2057 static void
2058 calculate_tsc_tweak()
2059 {
2060 	tsc_tweak = base_hz / tsc_hz;
2061 }
2062 
2063 static void
2064 dump_nhm_platform_info(void)
2065 {
2066 	unsigned long long msr;
2067 	unsigned int ratio;
2068 
2069 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2070 
2071 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
2072 
2073 	ratio = (msr >> 40) & 0xFF;
2074 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
2075 		ratio, bclk, ratio * bclk);
2076 
2077 	ratio = (msr >> 8) & 0xFF;
2078 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2079 		ratio, bclk, ratio * bclk);
2080 
2081 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
2082 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
2083 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
2084 
2085 	return;
2086 }
2087 
2088 static void
2089 dump_hsw_turbo_ratio_limits(void)
2090 {
2091 	unsigned long long msr;
2092 	unsigned int ratio;
2093 
2094 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
2095 
2096 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
2097 
2098 	ratio = (msr >> 8) & 0xFF;
2099 	if (ratio)
2100 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
2101 			ratio, bclk, ratio * bclk);
2102 
2103 	ratio = (msr >> 0) & 0xFF;
2104 	if (ratio)
2105 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
2106 			ratio, bclk, ratio * bclk);
2107 	return;
2108 }
2109 
2110 static void
2111 dump_ivt_turbo_ratio_limits(void)
2112 {
2113 	unsigned long long msr;
2114 	unsigned int ratio;
2115 
2116 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
2117 
2118 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
2119 
2120 	ratio = (msr >> 56) & 0xFF;
2121 	if (ratio)
2122 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
2123 			ratio, bclk, ratio * bclk);
2124 
2125 	ratio = (msr >> 48) & 0xFF;
2126 	if (ratio)
2127 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
2128 			ratio, bclk, ratio * bclk);
2129 
2130 	ratio = (msr >> 40) & 0xFF;
2131 	if (ratio)
2132 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
2133 			ratio, bclk, ratio * bclk);
2134 
2135 	ratio = (msr >> 32) & 0xFF;
2136 	if (ratio)
2137 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
2138 			ratio, bclk, ratio * bclk);
2139 
2140 	ratio = (msr >> 24) & 0xFF;
2141 	if (ratio)
2142 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
2143 			ratio, bclk, ratio * bclk);
2144 
2145 	ratio = (msr >> 16) & 0xFF;
2146 	if (ratio)
2147 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
2148 			ratio, bclk, ratio * bclk);
2149 
2150 	ratio = (msr >> 8) & 0xFF;
2151 	if (ratio)
2152 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
2153 			ratio, bclk, ratio * bclk);
2154 
2155 	ratio = (msr >> 0) & 0xFF;
2156 	if (ratio)
2157 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
2158 			ratio, bclk, ratio * bclk);
2159 	return;
2160 }
2161 int has_turbo_ratio_group_limits(int family, int model)
2162 {
2163 
2164 	if (!genuine_intel)
2165 		return 0;
2166 
2167 	switch (model) {
2168 	case INTEL_FAM6_ATOM_GOLDMONT:
2169 	case INTEL_FAM6_SKYLAKE_X:
2170 	case INTEL_FAM6_ATOM_GOLDMONT_X:
2171 		return 1;
2172 	}
2173 	return 0;
2174 }
2175 
2176 static void
2177 dump_turbo_ratio_limits(int family, int model)
2178 {
2179 	unsigned long long msr, core_counts;
2180 	unsigned int ratio, group_size;
2181 
2182 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2183 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
2184 
2185 	if (has_turbo_ratio_group_limits(family, model)) {
2186 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
2187 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
2188 	} else {
2189 		core_counts = 0x0807060504030201;
2190 	}
2191 
2192 	ratio = (msr >> 56) & 0xFF;
2193 	group_size = (core_counts >> 56) & 0xFF;
2194 	if (ratio)
2195 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2196 			ratio, bclk, ratio * bclk, group_size);
2197 
2198 	ratio = (msr >> 48) & 0xFF;
2199 	group_size = (core_counts >> 48) & 0xFF;
2200 	if (ratio)
2201 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2202 			ratio, bclk, ratio * bclk, group_size);
2203 
2204 	ratio = (msr >> 40) & 0xFF;
2205 	group_size = (core_counts >> 40) & 0xFF;
2206 	if (ratio)
2207 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2208 			ratio, bclk, ratio * bclk, group_size);
2209 
2210 	ratio = (msr >> 32) & 0xFF;
2211 	group_size = (core_counts >> 32) & 0xFF;
2212 	if (ratio)
2213 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2214 			ratio, bclk, ratio * bclk, group_size);
2215 
2216 	ratio = (msr >> 24) & 0xFF;
2217 	group_size = (core_counts >> 24) & 0xFF;
2218 	if (ratio)
2219 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2220 			ratio, bclk, ratio * bclk, group_size);
2221 
2222 	ratio = (msr >> 16) & 0xFF;
2223 	group_size = (core_counts >> 16) & 0xFF;
2224 	if (ratio)
2225 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2226 			ratio, bclk, ratio * bclk, group_size);
2227 
2228 	ratio = (msr >> 8) & 0xFF;
2229 	group_size = (core_counts >> 8) & 0xFF;
2230 	if (ratio)
2231 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2232 			ratio, bclk, ratio * bclk, group_size);
2233 
2234 	ratio = (msr >> 0) & 0xFF;
2235 	group_size = (core_counts >> 0) & 0xFF;
2236 	if (ratio)
2237 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2238 			ratio, bclk, ratio * bclk, group_size);
2239 	return;
2240 }
2241 
2242 static void
2243 dump_atom_turbo_ratio_limits(void)
2244 {
2245 	unsigned long long msr;
2246 	unsigned int ratio;
2247 
2248 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
2249 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2250 
2251 	ratio = (msr >> 0) & 0x3F;
2252 	if (ratio)
2253 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
2254 			ratio, bclk, ratio * bclk);
2255 
2256 	ratio = (msr >> 8) & 0x3F;
2257 	if (ratio)
2258 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
2259 			ratio, bclk, ratio * bclk);
2260 
2261 	ratio = (msr >> 16) & 0x3F;
2262 	if (ratio)
2263 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2264 			ratio, bclk, ratio * bclk);
2265 
2266 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
2267 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2268 
2269 	ratio = (msr >> 24) & 0x3F;
2270 	if (ratio)
2271 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
2272 			ratio, bclk, ratio * bclk);
2273 
2274 	ratio = (msr >> 16) & 0x3F;
2275 	if (ratio)
2276 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
2277 			ratio, bclk, ratio * bclk);
2278 
2279 	ratio = (msr >> 8) & 0x3F;
2280 	if (ratio)
2281 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
2282 			ratio, bclk, ratio * bclk);
2283 
2284 	ratio = (msr >> 0) & 0x3F;
2285 	if (ratio)
2286 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
2287 			ratio, bclk, ratio * bclk);
2288 }
2289 
2290 static void
2291 dump_knl_turbo_ratio_limits(void)
2292 {
2293 	const unsigned int buckets_no = 7;
2294 
2295 	unsigned long long msr;
2296 	int delta_cores, delta_ratio;
2297 	int i, b_nr;
2298 	unsigned int cores[buckets_no];
2299 	unsigned int ratio[buckets_no];
2300 
2301 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2302 
2303 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
2304 		base_cpu, msr);
2305 
2306 	/**
2307 	 * Turbo encoding in KNL is as follows:
2308 	 * [0] -- Reserved
2309 	 * [7:1] -- Base value of number of active cores of bucket 1.
2310 	 * [15:8] -- Base value of freq ratio of bucket 1.
2311 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
2312 	 * i.e. active cores of bucket 2 =
2313 	 * active cores of bucket 1 + delta
2314 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
2315 	 * i.e. freq ratio of bucket 2 =
2316 	 * freq ratio of bucket 1 - delta
2317 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
2318 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
2319 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
2320 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
2321 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
2322 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
2323 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
2324 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
2325 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
2326 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
2327 	 */
2328 
2329 	b_nr = 0;
2330 	cores[b_nr] = (msr & 0xFF) >> 1;
2331 	ratio[b_nr] = (msr >> 8) & 0xFF;
2332 
2333 	for (i = 16; i < 64; i += 8) {
2334 		delta_cores = (msr >> i) & 0x1F;
2335 		delta_ratio = (msr >> (i + 5)) & 0x7;
2336 
2337 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
2338 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2339 		b_nr++;
2340 	}
2341 
2342 	for (i = buckets_no - 1; i >= 0; i--)
2343 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2344 			fprintf(outf,
2345 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
2346 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
2347 }
2348 
2349 static void
2350 dump_nhm_cst_cfg(void)
2351 {
2352 	unsigned long long msr;
2353 
2354 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2355 
2356 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2357 
2358 	fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)",
2359 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2360 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2361 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2362 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2363 		(msr & (1 << 15)) ? "" : "UN",
2364 		(unsigned int)msr & 0xF,
2365 		pkg_cstate_limit_strings[pkg_cstate_limit]);
2366 
2367 #define AUTOMATIC_CSTATE_CONVERSION		(1UL << 16)
2368 	if (has_automatic_cstate_conversion) {
2369 		fprintf(outf, ", automatic c-state conversion=%s",
2370 			(msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off");
2371 	}
2372 
2373 	fprintf(outf, ")\n");
2374 
2375 	return;
2376 }
2377 
2378 static void
2379 dump_config_tdp(void)
2380 {
2381 	unsigned long long msr;
2382 
2383 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2384 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2385 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2386 
2387 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2388 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2389 	if (msr) {
2390 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2391 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2392 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2393 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2394 	}
2395 	fprintf(outf, ")\n");
2396 
2397 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2398 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2399 	if (msr) {
2400 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2401 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2402 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2403 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2404 	}
2405 	fprintf(outf, ")\n");
2406 
2407 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2408 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2409 	if ((msr) & 0x3)
2410 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2411 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2412 	fprintf(outf, ")\n");
2413 
2414 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2415 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2416 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2417 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2418 	fprintf(outf, ")\n");
2419 }
2420 
2421 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2422 
2423 void print_irtl(void)
2424 {
2425 	unsigned long long msr;
2426 
2427 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2428 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2429 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2430 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2431 
2432 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2433 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2434 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2435 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2436 
2437 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2438 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2439 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2440 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2441 
2442 	if (!do_irtl_hsw)
2443 		return;
2444 
2445 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2446 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2447 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2448 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2449 
2450 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2451 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2452 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2453 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2454 
2455 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2456 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2457 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2458 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2459 
2460 }
2461 void free_fd_percpu(void)
2462 {
2463 	int i;
2464 
2465 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2466 		if (fd_percpu[i] != 0)
2467 			close(fd_percpu[i]);
2468 	}
2469 
2470 	free(fd_percpu);
2471 }
2472 
2473 void free_all_buffers(void)
2474 {
2475 	int i;
2476 
2477 	CPU_FREE(cpu_present_set);
2478 	cpu_present_set = NULL;
2479 	cpu_present_setsize = 0;
2480 
2481 	CPU_FREE(cpu_affinity_set);
2482 	cpu_affinity_set = NULL;
2483 	cpu_affinity_setsize = 0;
2484 
2485 	free(thread_even);
2486 	free(core_even);
2487 	free(package_even);
2488 
2489 	thread_even = NULL;
2490 	core_even = NULL;
2491 	package_even = NULL;
2492 
2493 	free(thread_odd);
2494 	free(core_odd);
2495 	free(package_odd);
2496 
2497 	thread_odd = NULL;
2498 	core_odd = NULL;
2499 	package_odd = NULL;
2500 
2501 	free(output_buffer);
2502 	output_buffer = NULL;
2503 	outp = NULL;
2504 
2505 	free_fd_percpu();
2506 
2507 	free(irq_column_2_cpu);
2508 	free(irqs_per_cpu);
2509 
2510 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2511 		if (cpus[i].put_ids)
2512 			CPU_FREE(cpus[i].put_ids);
2513 	}
2514 	free(cpus);
2515 }
2516 
2517 
2518 /*
2519  * Parse a file containing a single int.
2520  * Return 0 if file can not be opened
2521  * Exit if file can be opened, but can not be parsed
2522  */
2523 int parse_int_file(const char *fmt, ...)
2524 {
2525 	va_list args;
2526 	char path[PATH_MAX];
2527 	FILE *filep;
2528 	int value;
2529 
2530 	va_start(args, fmt);
2531 	vsnprintf(path, sizeof(path), fmt, args);
2532 	va_end(args);
2533 	filep = fopen(path, "r");
2534 	if (!filep)
2535 		return 0;
2536 	if (fscanf(filep, "%d", &value) != 1)
2537 		err(1, "%s: failed to parse number from file", path);
2538 	fclose(filep);
2539 	return value;
2540 }
2541 
2542 /*
2543  * cpu_is_first_core_in_package(cpu)
2544  * return 1 if given CPU is 1st core in package
2545  */
2546 int cpu_is_first_core_in_package(int cpu)
2547 {
2548 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2549 }
2550 
2551 int get_physical_package_id(int cpu)
2552 {
2553 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2554 }
2555 
2556 int get_die_id(int cpu)
2557 {
2558 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu);
2559 }
2560 
2561 int get_core_id(int cpu)
2562 {
2563 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2564 }
2565 
2566 void set_node_data(void)
2567 {
2568 	int pkg, node, lnode, cpu, cpux;
2569 	int cpu_count;
2570 
2571 	/* initialize logical_node_id */
2572 	for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu)
2573 		cpus[cpu].logical_node_id = -1;
2574 
2575 	cpu_count = 0;
2576 	for (pkg = 0; pkg < topo.num_packages; pkg++) {
2577 		lnode = 0;
2578 		for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) {
2579 			if (cpus[cpu].physical_package_id != pkg)
2580 				continue;
2581 			/* find a cpu with an unset logical_node_id */
2582 			if (cpus[cpu].logical_node_id != -1)
2583 				continue;
2584 			cpus[cpu].logical_node_id = lnode;
2585 			node = cpus[cpu].physical_node_id;
2586 			cpu_count++;
2587 			/*
2588 			 * find all matching cpus on this pkg and set
2589 			 * the logical_node_id
2590 			 */
2591 			for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) {
2592 				if ((cpus[cpux].physical_package_id == pkg) &&
2593 				   (cpus[cpux].physical_node_id == node)) {
2594 					cpus[cpux].logical_node_id = lnode;
2595 					cpu_count++;
2596 				}
2597 			}
2598 			lnode++;
2599 			if (lnode > topo.nodes_per_pkg)
2600 				topo.nodes_per_pkg = lnode;
2601 		}
2602 		if (cpu_count >= topo.max_cpu_num)
2603 			break;
2604 	}
2605 }
2606 
2607 int get_physical_node_id(struct cpu_topology *thiscpu)
2608 {
2609 	char path[80];
2610 	FILE *filep;
2611 	int i;
2612 	int cpu = thiscpu->logical_cpu_id;
2613 
2614 	for (i = 0; i <= topo.max_cpu_num; i++) {
2615 		sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist",
2616 			cpu, i);
2617 		filep = fopen(path, "r");
2618 		if (!filep)
2619 			continue;
2620 		fclose(filep);
2621 		return i;
2622 	}
2623 	return -1;
2624 }
2625 
2626 int get_thread_siblings(struct cpu_topology *thiscpu)
2627 {
2628 	char path[80], character;
2629 	FILE *filep;
2630 	unsigned long map;
2631 	int so, shift, sib_core;
2632 	int cpu = thiscpu->logical_cpu_id;
2633 	int offset = topo.max_cpu_num + 1;
2634 	size_t size;
2635 	int thread_id = 0;
2636 
2637 	thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1));
2638 	if (thiscpu->thread_id < 0)
2639 		thiscpu->thread_id = thread_id++;
2640 	if (!thiscpu->put_ids)
2641 		return -1;
2642 
2643 	size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2644 	CPU_ZERO_S(size, thiscpu->put_ids);
2645 
2646 	sprintf(path,
2647 		"/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu);
2648 	filep = fopen_or_die(path, "r");
2649 	do {
2650 		offset -= BITMASK_SIZE;
2651 		if (fscanf(filep, "%lx%c", &map, &character) != 2)
2652 			err(1, "%s: failed to parse file", path);
2653 		for (shift = 0; shift < BITMASK_SIZE; shift++) {
2654 			if ((map >> shift) & 0x1) {
2655 				so = shift + offset;
2656 				sib_core = get_core_id(so);
2657 				if (sib_core == thiscpu->physical_core_id) {
2658 					CPU_SET_S(so, size, thiscpu->put_ids);
2659 					if ((so != cpu) &&
2660 					    (cpus[so].thread_id < 0))
2661 						cpus[so].thread_id =
2662 								    thread_id++;
2663 				}
2664 			}
2665 		}
2666 	} while (!strncmp(&character, ",", 1));
2667 	fclose(filep);
2668 
2669 	return CPU_COUNT_S(size, thiscpu->put_ids);
2670 }
2671 
2672 /*
2673  * run func(thread, core, package) in topology order
2674  * skip non-present cpus
2675  */
2676 
2677 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2678 	struct pkg_data *, struct thread_data *, struct core_data *,
2679 	struct pkg_data *), struct thread_data *thread_base,
2680 	struct core_data *core_base, struct pkg_data *pkg_base,
2681 	struct thread_data *thread_base2, struct core_data *core_base2,
2682 	struct pkg_data *pkg_base2)
2683 {
2684 	int retval, pkg_no, node_no, core_no, thread_no;
2685 
2686 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2687 		for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) {
2688 			for (core_no = 0; core_no < topo.cores_per_node;
2689 			     ++core_no) {
2690 				for (thread_no = 0; thread_no <
2691 					topo.threads_per_core; ++thread_no) {
2692 					struct thread_data *t, *t2;
2693 					struct core_data *c, *c2;
2694 					struct pkg_data *p, *p2;
2695 
2696 					t = GET_THREAD(thread_base, thread_no,
2697 						       core_no, node_no,
2698 						       pkg_no);
2699 
2700 					if (cpu_is_not_present(t->cpu_id))
2701 						continue;
2702 
2703 					t2 = GET_THREAD(thread_base2, thread_no,
2704 							core_no, node_no,
2705 							pkg_no);
2706 
2707 					c = GET_CORE(core_base, core_no,
2708 						     node_no, pkg_no);
2709 					c2 = GET_CORE(core_base2, core_no,
2710 						      node_no,
2711 						      pkg_no);
2712 
2713 					p = GET_PKG(pkg_base, pkg_no);
2714 					p2 = GET_PKG(pkg_base2, pkg_no);
2715 
2716 					retval = func(t, c, p, t2, c2, p2);
2717 					if (retval)
2718 						return retval;
2719 				}
2720 			}
2721 		}
2722 	}
2723 	return 0;
2724 }
2725 
2726 /*
2727  * run func(cpu) on every cpu in /proc/stat
2728  * return max_cpu number
2729  */
2730 int for_all_proc_cpus(int (func)(int))
2731 {
2732 	FILE *fp;
2733 	int cpu_num;
2734 	int retval;
2735 
2736 	fp = fopen_or_die(proc_stat, "r");
2737 
2738 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2739 	if (retval != 0)
2740 		err(1, "%s: failed to parse format", proc_stat);
2741 
2742 	while (1) {
2743 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2744 		if (retval != 1)
2745 			break;
2746 
2747 		retval = func(cpu_num);
2748 		if (retval) {
2749 			fclose(fp);
2750 			return(retval);
2751 		}
2752 	}
2753 	fclose(fp);
2754 	return 0;
2755 }
2756 
2757 void re_initialize(void)
2758 {
2759 	free_all_buffers();
2760 	setup_all_buffers();
2761 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2762 }
2763 
2764 void set_max_cpu_num(void)
2765 {
2766 	FILE *filep;
2767 	unsigned long dummy;
2768 
2769 	topo.max_cpu_num = 0;
2770 	filep = fopen_or_die(
2771 			"/sys/devices/system/cpu/cpu0/topology/thread_siblings",
2772 			"r");
2773 	while (fscanf(filep, "%lx,", &dummy) == 1)
2774 		topo.max_cpu_num += BITMASK_SIZE;
2775 	fclose(filep);
2776 	topo.max_cpu_num--; /* 0 based */
2777 }
2778 
2779 /*
2780  * count_cpus()
2781  * remember the last one seen, it will be the max
2782  */
2783 int count_cpus(int cpu)
2784 {
2785 	topo.num_cpus++;
2786 	return 0;
2787 }
2788 int mark_cpu_present(int cpu)
2789 {
2790 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2791 	return 0;
2792 }
2793 
2794 int init_thread_id(int cpu)
2795 {
2796 	cpus[cpu].thread_id = -1;
2797 	return 0;
2798 }
2799 
2800 /*
2801  * snapshot_proc_interrupts()
2802  *
2803  * read and record summary of /proc/interrupts
2804  *
2805  * return 1 if config change requires a restart, else return 0
2806  */
2807 int snapshot_proc_interrupts(void)
2808 {
2809 	static FILE *fp;
2810 	int column, retval;
2811 
2812 	if (fp == NULL)
2813 		fp = fopen_or_die("/proc/interrupts", "r");
2814 	else
2815 		rewind(fp);
2816 
2817 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
2818 	for (column = 0; column < topo.num_cpus; ++column) {
2819 		int cpu_number;
2820 
2821 		retval = fscanf(fp, " CPU%d", &cpu_number);
2822 		if (retval != 1)
2823 			break;
2824 
2825 		if (cpu_number > topo.max_cpu_num) {
2826 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2827 			return 1;
2828 		}
2829 
2830 		irq_column_2_cpu[column] = cpu_number;
2831 		irqs_per_cpu[cpu_number] = 0;
2832 	}
2833 
2834 	/* read /proc/interrupt count lines and sum up irqs per cpu */
2835 	while (1) {
2836 		int column;
2837 		char buf[64];
2838 
2839 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
2840 		if (retval != 1)
2841 			break;
2842 
2843 		/* read the count per cpu */
2844 		for (column = 0; column < topo.num_cpus; ++column) {
2845 
2846 			int cpu_number, irq_count;
2847 
2848 			retval = fscanf(fp, " %d", &irq_count);
2849 			if (retval != 1)
2850 				break;
2851 
2852 			cpu_number = irq_column_2_cpu[column];
2853 			irqs_per_cpu[cpu_number] += irq_count;
2854 
2855 		}
2856 
2857 		while (getc(fp) != '\n')
2858 			;	/* flush interrupt description */
2859 
2860 	}
2861 	return 0;
2862 }
2863 /*
2864  * snapshot_gfx_rc6_ms()
2865  *
2866  * record snapshot of
2867  * /sys/class/drm/card0/power/rc6_residency_ms
2868  *
2869  * return 1 if config change requires a restart, else return 0
2870  */
2871 int snapshot_gfx_rc6_ms(void)
2872 {
2873 	FILE *fp;
2874 	int retval;
2875 
2876 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2877 
2878 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2879 	if (retval != 1)
2880 		err(1, "GFX rc6");
2881 
2882 	fclose(fp);
2883 
2884 	return 0;
2885 }
2886 /*
2887  * snapshot_gfx_mhz()
2888  *
2889  * record snapshot of
2890  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2891  *
2892  * return 1 if config change requires a restart, else return 0
2893  */
2894 int snapshot_gfx_mhz(void)
2895 {
2896 	static FILE *fp;
2897 	int retval;
2898 
2899 	if (fp == NULL)
2900 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2901 	else {
2902 		rewind(fp);
2903 		fflush(fp);
2904 	}
2905 
2906 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
2907 	if (retval != 1)
2908 		err(1, "GFX MHz");
2909 
2910 	return 0;
2911 }
2912 
2913 /*
2914  * snapshot_cpu_lpi()
2915  *
2916  * record snapshot of
2917  * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us
2918  *
2919  * return 1 if config change requires a restart, else return 0
2920  */
2921 int snapshot_cpu_lpi_us(void)
2922 {
2923 	FILE *fp;
2924 	int retval;
2925 
2926 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r");
2927 
2928 	retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us);
2929 	if (retval != 1) {
2930 		fprintf(stderr, "Disabling Low Power Idle CPU output\n");
2931 		BIC_NOT_PRESENT(BIC_CPU_LPI);
2932 		fclose(fp);
2933 		return -1;
2934 	}
2935 
2936 	fclose(fp);
2937 
2938 	return 0;
2939 }
2940 /*
2941  * snapshot_sys_lpi()
2942  *
2943  * record snapshot of
2944  * /sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us
2945  *
2946  * return 1 if config change requires a restart, else return 0
2947  */
2948 int snapshot_sys_lpi_us(void)
2949 {
2950 	FILE *fp;
2951 	int retval;
2952 
2953 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", "r");
2954 
2955 	retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us);
2956 	if (retval != 1) {
2957 		fprintf(stderr, "Disabling Low Power Idle System output\n");
2958 		BIC_NOT_PRESENT(BIC_SYS_LPI);
2959 		fclose(fp);
2960 		return -1;
2961 	}
2962 	fclose(fp);
2963 
2964 	return 0;
2965 }
2966 /*
2967  * snapshot /proc and /sys files
2968  *
2969  * return 1 if configuration restart needed, else return 0
2970  */
2971 int snapshot_proc_sysfs_files(void)
2972 {
2973 	if (DO_BIC(BIC_IRQ))
2974 		if (snapshot_proc_interrupts())
2975 			return 1;
2976 
2977 	if (DO_BIC(BIC_GFX_rc6))
2978 		snapshot_gfx_rc6_ms();
2979 
2980 	if (DO_BIC(BIC_GFXMHz))
2981 		snapshot_gfx_mhz();
2982 
2983 	if (DO_BIC(BIC_CPU_LPI))
2984 		snapshot_cpu_lpi_us();
2985 
2986 	if (DO_BIC(BIC_SYS_LPI))
2987 		snapshot_sys_lpi_us();
2988 
2989 	return 0;
2990 }
2991 
2992 int exit_requested;
2993 
2994 static void signal_handler (int signal)
2995 {
2996 	switch (signal) {
2997 	case SIGINT:
2998 		exit_requested = 1;
2999 		if (debug)
3000 			fprintf(stderr, " SIGINT\n");
3001 		break;
3002 	case SIGUSR1:
3003 		if (debug > 1)
3004 			fprintf(stderr, "SIGUSR1\n");
3005 		break;
3006 	}
3007 }
3008 
3009 void setup_signal_handler(void)
3010 {
3011 	struct sigaction sa;
3012 
3013 	memset(&sa, 0, sizeof(sa));
3014 
3015 	sa.sa_handler = &signal_handler;
3016 
3017 	if (sigaction(SIGINT, &sa, NULL) < 0)
3018 		err(1, "sigaction SIGINT");
3019 	if (sigaction(SIGUSR1, &sa, NULL) < 0)
3020 		err(1, "sigaction SIGUSR1");
3021 }
3022 
3023 void do_sleep(void)
3024 {
3025 	struct timeval tout;
3026 	struct timespec rest;
3027 	fd_set readfds;
3028 	int retval;
3029 
3030 	FD_ZERO(&readfds);
3031 	FD_SET(0, &readfds);
3032 
3033 	if (ignore_stdin) {
3034 		nanosleep(&interval_ts, NULL);
3035 		return;
3036 	}
3037 
3038 	tout = interval_tv;
3039 	retval = select(1, &readfds, NULL, NULL, &tout);
3040 
3041 	if (retval == 1) {
3042 		switch (getc(stdin)) {
3043 		case 'q':
3044 			exit_requested = 1;
3045 			break;
3046 		case EOF:
3047 			/*
3048 			 * 'stdin' is a pipe closed on the other end. There
3049 			 * won't be any further input.
3050 			 */
3051 			ignore_stdin = 1;
3052 			/* Sleep the rest of the time */
3053 			rest.tv_sec = (tout.tv_sec + tout.tv_usec / 1000000);
3054 			rest.tv_nsec = (tout.tv_usec % 1000000) * 1000;
3055 			nanosleep(&rest, NULL);
3056 		}
3057 	}
3058 }
3059 
3060 
3061 void turbostat_loop()
3062 {
3063 	int retval;
3064 	int restarted = 0;
3065 	int done_iters = 0;
3066 
3067 	setup_signal_handler();
3068 
3069 restart:
3070 	restarted++;
3071 
3072 	snapshot_proc_sysfs_files();
3073 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3074 	first_counter_read = 0;
3075 	if (retval < -1) {
3076 		exit(retval);
3077 	} else if (retval == -1) {
3078 		if (restarted > 1) {
3079 			exit(retval);
3080 		}
3081 		re_initialize();
3082 		goto restart;
3083 	}
3084 	restarted = 0;
3085 	done_iters = 0;
3086 	gettimeofday(&tv_even, (struct timezone *)NULL);
3087 
3088 	while (1) {
3089 		if (for_all_proc_cpus(cpu_is_not_present)) {
3090 			re_initialize();
3091 			goto restart;
3092 		}
3093 		do_sleep();
3094 		if (snapshot_proc_sysfs_files())
3095 			goto restart;
3096 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
3097 		if (retval < -1) {
3098 			exit(retval);
3099 		} else if (retval == -1) {
3100 			re_initialize();
3101 			goto restart;
3102 		}
3103 		gettimeofday(&tv_odd, (struct timezone *)NULL);
3104 		timersub(&tv_odd, &tv_even, &tv_delta);
3105 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
3106 			re_initialize();
3107 			goto restart;
3108 		}
3109 		compute_average(EVEN_COUNTERS);
3110 		format_all_counters(EVEN_COUNTERS);
3111 		flush_output_stdout();
3112 		if (exit_requested)
3113 			break;
3114 		if (num_iterations && ++done_iters >= num_iterations)
3115 			break;
3116 		do_sleep();
3117 		if (snapshot_proc_sysfs_files())
3118 			goto restart;
3119 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3120 		if (retval < -1) {
3121 			exit(retval);
3122 		} else if (retval == -1) {
3123 			re_initialize();
3124 			goto restart;
3125 		}
3126 		gettimeofday(&tv_even, (struct timezone *)NULL);
3127 		timersub(&tv_even, &tv_odd, &tv_delta);
3128 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
3129 			re_initialize();
3130 			goto restart;
3131 		}
3132 		compute_average(ODD_COUNTERS);
3133 		format_all_counters(ODD_COUNTERS);
3134 		flush_output_stdout();
3135 		if (exit_requested)
3136 			break;
3137 		if (num_iterations && ++done_iters >= num_iterations)
3138 			break;
3139 	}
3140 }
3141 
3142 void check_dev_msr()
3143 {
3144 	struct stat sb;
3145 	char pathname[32];
3146 
3147 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3148 	if (stat(pathname, &sb))
3149  		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
3150 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
3151 }
3152 
3153 void check_permissions()
3154 {
3155 	struct __user_cap_header_struct cap_header_data;
3156 	cap_user_header_t cap_header = &cap_header_data;
3157 	struct __user_cap_data_struct cap_data_data;
3158 	cap_user_data_t cap_data = &cap_data_data;
3159 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
3160 	int do_exit = 0;
3161 	char pathname[32];
3162 
3163 	/* check for CAP_SYS_RAWIO */
3164 	cap_header->pid = getpid();
3165 	cap_header->version = _LINUX_CAPABILITY_VERSION;
3166 	if (capget(cap_header, cap_data) < 0)
3167 		err(-6, "capget(2) failed");
3168 
3169 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
3170 		do_exit++;
3171 		warnx("capget(CAP_SYS_RAWIO) failed,"
3172 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
3173 	}
3174 
3175 	/* test file permissions */
3176 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3177 	if (euidaccess(pathname, R_OK)) {
3178 		do_exit++;
3179 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
3180 	}
3181 
3182 	/* if all else fails, thell them to be root */
3183 	if (do_exit)
3184 		if (getuid() != 0)
3185 			warnx("... or simply run as root");
3186 
3187 	if (do_exit)
3188 		exit(-6);
3189 }
3190 
3191 /*
3192  * NHM adds support for additional MSRs:
3193  *
3194  * MSR_SMI_COUNT                   0x00000034
3195  *
3196  * MSR_PLATFORM_INFO               0x000000ce
3197  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
3198  *
3199  * MSR_MISC_PWR_MGMT               0x000001aa
3200  *
3201  * MSR_PKG_C3_RESIDENCY            0x000003f8
3202  * MSR_PKG_C6_RESIDENCY            0x000003f9
3203  * MSR_CORE_C3_RESIDENCY           0x000003fc
3204  * MSR_CORE_C6_RESIDENCY           0x000003fd
3205  *
3206  * Side effect:
3207  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
3208  * sets has_misc_feature_control
3209  */
3210 int probe_nhm_msrs(unsigned int family, unsigned int model)
3211 {
3212 	unsigned long long msr;
3213 	unsigned int base_ratio;
3214 	int *pkg_cstate_limits;
3215 
3216 	if (!genuine_intel)
3217 		return 0;
3218 
3219 	if (family != 6)
3220 		return 0;
3221 
3222 	bclk = discover_bclk(family, model);
3223 
3224 	switch (model) {
3225 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
3226 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3227 		pkg_cstate_limits = nhm_pkg_cstate_limits;
3228 		break;
3229 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
3230 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
3231 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3232 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3233 		pkg_cstate_limits = snb_pkg_cstate_limits;
3234 		has_misc_feature_control = 1;
3235 		break;
3236 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3237 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3238 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3239 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3240 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3241 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3242 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3243 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3244 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3245 		pkg_cstate_limits = hsw_pkg_cstate_limits;
3246 		has_misc_feature_control = 1;
3247 		break;
3248 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3249 		pkg_cstate_limits = skx_pkg_cstate_limits;
3250 		has_misc_feature_control = 1;
3251 		break;
3252 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3253 		no_MSR_MISC_PWR_MGMT = 1;
3254 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3255 		pkg_cstate_limits = slv_pkg_cstate_limits;
3256 		break;
3257 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
3258 		pkg_cstate_limits = amt_pkg_cstate_limits;
3259 		no_MSR_MISC_PWR_MGMT = 1;
3260 		break;
3261 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
3262 		pkg_cstate_limits = phi_pkg_cstate_limits;
3263 		break;
3264 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3265 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3266 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3267 		pkg_cstate_limits = glm_pkg_cstate_limits;
3268 		break;
3269 	default:
3270 		return 0;
3271 	}
3272 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
3273 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
3274 
3275 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
3276 	base_ratio = (msr >> 8) & 0xFF;
3277 
3278 	base_hz = base_ratio * bclk * 1000000;
3279 	has_base_hz = 1;
3280 	return 1;
3281 }
3282 /*
3283  * SLV client has support for unique MSRs:
3284  *
3285  * MSR_CC6_DEMOTION_POLICY_CONFIG
3286  * MSR_MC6_DEMOTION_POLICY_CONFIG
3287  */
3288 
3289 int has_slv_msrs(unsigned int family, unsigned int model)
3290 {
3291 	if (!genuine_intel)
3292 		return 0;
3293 
3294 	switch (model) {
3295 	case INTEL_FAM6_ATOM_SILVERMONT:
3296 	case INTEL_FAM6_ATOM_SILVERMONT_MID:
3297 	case INTEL_FAM6_ATOM_AIRMONT_MID:
3298 		return 1;
3299 	}
3300 	return 0;
3301 }
3302 int is_dnv(unsigned int family, unsigned int model)
3303 {
3304 
3305 	if (!genuine_intel)
3306 		return 0;
3307 
3308 	switch (model) {
3309 	case INTEL_FAM6_ATOM_GOLDMONT_X:
3310 		return 1;
3311 	}
3312 	return 0;
3313 }
3314 int is_bdx(unsigned int family, unsigned int model)
3315 {
3316 
3317 	if (!genuine_intel)
3318 		return 0;
3319 
3320 	switch (model) {
3321 	case INTEL_FAM6_BROADWELL_X:
3322 		return 1;
3323 	}
3324 	return 0;
3325 }
3326 int is_skx(unsigned int family, unsigned int model)
3327 {
3328 
3329 	if (!genuine_intel)
3330 		return 0;
3331 
3332 	switch (model) {
3333 	case INTEL_FAM6_SKYLAKE_X:
3334 		return 1;
3335 	}
3336 	return 0;
3337 }
3338 
3339 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
3340 {
3341 	if (has_slv_msrs(family, model))
3342 		return 0;
3343 
3344 	switch (model) {
3345 	/* Nehalem compatible, but do not include turbo-ratio limit support */
3346 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3347 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
3348 		return 0;
3349 	default:
3350 		return 1;
3351 	}
3352 }
3353 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
3354 {
3355 	if (has_slv_msrs(family, model))
3356 		return 1;
3357 
3358 	return 0;
3359 }
3360 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
3361 {
3362 	if (!genuine_intel)
3363 		return 0;
3364 
3365 	if (family != 6)
3366 		return 0;
3367 
3368 	switch (model) {
3369 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3370 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3371 		return 1;
3372 	default:
3373 		return 0;
3374 	}
3375 }
3376 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
3377 {
3378 	if (!genuine_intel)
3379 		return 0;
3380 
3381 	if (family != 6)
3382 		return 0;
3383 
3384 	switch (model) {
3385 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3386 		return 1;
3387 	default:
3388 		return 0;
3389 	}
3390 }
3391 
3392 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
3393 {
3394 	if (!genuine_intel)
3395 		return 0;
3396 
3397 	if (family != 6)
3398 		return 0;
3399 
3400 	switch (model) {
3401 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3402 		return 1;
3403 	default:
3404 		return 0;
3405 	}
3406 }
3407 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
3408 {
3409 	if (!genuine_intel)
3410 		return 0;
3411 
3412 	if (family != 6)
3413 		return 0;
3414 
3415 	switch (model) {
3416 	case INTEL_FAM6_ATOM_GOLDMONT:
3417 	case INTEL_FAM6_SKYLAKE_X:
3418 		return 1;
3419 	default:
3420 		return 0;
3421 	}
3422 }
3423 int has_config_tdp(unsigned int family, unsigned int model)
3424 {
3425 	if (!genuine_intel)
3426 		return 0;
3427 
3428 	if (family != 6)
3429 		return 0;
3430 
3431 	switch (model) {
3432 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3433 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3434 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3435 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3436 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3437 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3438 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3439 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3440 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3441 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3442 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3443 
3444 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3445 		return 1;
3446 	default:
3447 		return 0;
3448 	}
3449 }
3450 
3451 static void
3452 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
3453 {
3454 	if (!do_nhm_platform_info)
3455 		return;
3456 
3457 	dump_nhm_platform_info();
3458 
3459 	if (has_hsw_turbo_ratio_limit(family, model))
3460 		dump_hsw_turbo_ratio_limits();
3461 
3462 	if (has_ivt_turbo_ratio_limit(family, model))
3463 		dump_ivt_turbo_ratio_limits();
3464 
3465 	if (has_turbo_ratio_limit(family, model))
3466 		dump_turbo_ratio_limits(family, model);
3467 
3468 	if (has_atom_turbo_ratio_limit(family, model))
3469 		dump_atom_turbo_ratio_limits();
3470 
3471 	if (has_knl_turbo_ratio_limit(family, model))
3472 		dump_knl_turbo_ratio_limits();
3473 
3474 	if (has_config_tdp(family, model))
3475 		dump_config_tdp();
3476 
3477 	dump_nhm_cst_cfg();
3478 }
3479 
3480 static void
3481 dump_sysfs_cstate_config(void)
3482 {
3483 	char path[64];
3484 	char name_buf[16];
3485 	char desc[64];
3486 	FILE *input;
3487 	int state;
3488 	char *sp;
3489 
3490 	if (!DO_BIC(BIC_sysfs))
3491 		return;
3492 
3493 	for (state = 0; state < 10; ++state) {
3494 
3495 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
3496 			base_cpu, state);
3497 		input = fopen(path, "r");
3498 		if (input == NULL)
3499 			continue;
3500 		if (!fgets(name_buf, sizeof(name_buf), input))
3501 			err(1, "%s: failed to read file", path);
3502 
3503 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
3504 		sp = strchr(name_buf, '-');
3505 		if (!sp)
3506 			sp = strchrnul(name_buf, '\n');
3507 		*sp = '\0';
3508 		fclose(input);
3509 
3510 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc",
3511 			base_cpu, state);
3512 		input = fopen(path, "r");
3513 		if (input == NULL)
3514 			continue;
3515 		if (!fgets(desc, sizeof(desc), input))
3516 			err(1, "%s: failed to read file", path);
3517 
3518 		fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
3519 		fclose(input);
3520 	}
3521 }
3522 static void
3523 dump_sysfs_pstate_config(void)
3524 {
3525 	char path[64];
3526 	char driver_buf[64];
3527 	char governor_buf[64];
3528 	FILE *input;
3529 	int turbo;
3530 
3531 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver",
3532 			base_cpu);
3533 	input = fopen(path, "r");
3534 	if (input == NULL) {
3535 		fprintf(outf, "NSFOD %s\n", path);
3536 		return;
3537 	}
3538 	if (!fgets(driver_buf, sizeof(driver_buf), input))
3539 		err(1, "%s: failed to read file", path);
3540 	fclose(input);
3541 
3542 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor",
3543 			base_cpu);
3544 	input = fopen(path, "r");
3545 	if (input == NULL) {
3546 		fprintf(outf, "NSFOD %s\n", path);
3547 		return;
3548 	}
3549 	if (!fgets(governor_buf, sizeof(governor_buf), input))
3550 		err(1, "%s: failed to read file", path);
3551 	fclose(input);
3552 
3553 	fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
3554 	fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
3555 
3556 	sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
3557 	input = fopen(path, "r");
3558 	if (input != NULL) {
3559 		if (fscanf(input, "%d", &turbo) != 1)
3560 			err(1, "%s: failed to parse number from file", path);
3561 		fprintf(outf, "cpufreq boost: %d\n", turbo);
3562 		fclose(input);
3563 	}
3564 
3565 	sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
3566 	input = fopen(path, "r");
3567 	if (input != NULL) {
3568 		if (fscanf(input, "%d", &turbo) != 1)
3569 			err(1, "%s: failed to parse number from file", path);
3570 		fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
3571 		fclose(input);
3572 	}
3573 }
3574 
3575 
3576 /*
3577  * print_epb()
3578  * Decode the ENERGY_PERF_BIAS MSR
3579  */
3580 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3581 {
3582 	unsigned long long msr;
3583 	char *epb_string;
3584 	int cpu;
3585 
3586 	if (!has_epb)
3587 		return 0;
3588 
3589 	cpu = t->cpu_id;
3590 
3591 	/* EPB is per-package */
3592 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3593 		return 0;
3594 
3595 	if (cpu_migrate(cpu)) {
3596 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3597 		return -1;
3598 	}
3599 
3600 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
3601 		return 0;
3602 
3603 	switch (msr & 0xF) {
3604 	case ENERGY_PERF_BIAS_PERFORMANCE:
3605 		epb_string = "performance";
3606 		break;
3607 	case ENERGY_PERF_BIAS_NORMAL:
3608 		epb_string = "balanced";
3609 		break;
3610 	case ENERGY_PERF_BIAS_POWERSAVE:
3611 		epb_string = "powersave";
3612 		break;
3613 	default:
3614 		epb_string = "custom";
3615 		break;
3616 	}
3617 	fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
3618 
3619 	return 0;
3620 }
3621 /*
3622  * print_hwp()
3623  * Decode the MSR_HWP_CAPABILITIES
3624  */
3625 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3626 {
3627 	unsigned long long msr;
3628 	int cpu;
3629 
3630 	if (!has_hwp)
3631 		return 0;
3632 
3633 	cpu = t->cpu_id;
3634 
3635 	/* MSR_HWP_CAPABILITIES is per-package */
3636 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3637 		return 0;
3638 
3639 	if (cpu_migrate(cpu)) {
3640 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3641 		return -1;
3642 	}
3643 
3644 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
3645 		return 0;
3646 
3647 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
3648 		cpu, msr, (msr & (1 << 0)) ? "" : "No-");
3649 
3650 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
3651 	if ((msr & (1 << 0)) == 0)
3652 		return 0;
3653 
3654 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
3655 		return 0;
3656 
3657 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
3658 			"(high %d guar %d eff %d low %d)\n",
3659 			cpu, msr,
3660 			(unsigned int)HWP_HIGHEST_PERF(msr),
3661 			(unsigned int)HWP_GUARANTEED_PERF(msr),
3662 			(unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
3663 			(unsigned int)HWP_LOWEST_PERF(msr));
3664 
3665 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
3666 		return 0;
3667 
3668 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
3669 			"(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
3670 			cpu, msr,
3671 			(unsigned int)(((msr) >> 0) & 0xff),
3672 			(unsigned int)(((msr) >> 8) & 0xff),
3673 			(unsigned int)(((msr) >> 16) & 0xff),
3674 			(unsigned int)(((msr) >> 24) & 0xff),
3675 			(unsigned int)(((msr) >> 32) & 0xff3),
3676 			(unsigned int)(((msr) >> 42) & 0x1));
3677 
3678 	if (has_hwp_pkg) {
3679 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
3680 			return 0;
3681 
3682 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
3683 			"(min %d max %d des %d epp 0x%x window 0x%x)\n",
3684 			cpu, msr,
3685 			(unsigned int)(((msr) >> 0) & 0xff),
3686 			(unsigned int)(((msr) >> 8) & 0xff),
3687 			(unsigned int)(((msr) >> 16) & 0xff),
3688 			(unsigned int)(((msr) >> 24) & 0xff),
3689 			(unsigned int)(((msr) >> 32) & 0xff3));
3690 	}
3691 	if (has_hwp_notify) {
3692 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
3693 			return 0;
3694 
3695 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
3696 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
3697 			cpu, msr,
3698 			((msr) & 0x1) ? "EN" : "Dis",
3699 			((msr) & 0x2) ? "EN" : "Dis");
3700 	}
3701 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
3702 		return 0;
3703 
3704 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
3705 			"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
3706 			cpu, msr,
3707 			((msr) & 0x1) ? "" : "No-",
3708 			((msr) & 0x2) ? "" : "No-");
3709 
3710 	return 0;
3711 }
3712 
3713 /*
3714  * print_perf_limit()
3715  */
3716 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3717 {
3718 	unsigned long long msr;
3719 	int cpu;
3720 
3721 	cpu = t->cpu_id;
3722 
3723 	/* per-package */
3724 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3725 		return 0;
3726 
3727 	if (cpu_migrate(cpu)) {
3728 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3729 		return -1;
3730 	}
3731 
3732 	if (do_core_perf_limit_reasons) {
3733 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
3734 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3735 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
3736 			(msr & 1 << 15) ? "bit15, " : "",
3737 			(msr & 1 << 14) ? "bit14, " : "",
3738 			(msr & 1 << 13) ? "Transitions, " : "",
3739 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
3740 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
3741 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3742 			(msr & 1 << 9) ? "CorePwr, " : "",
3743 			(msr & 1 << 8) ? "Amps, " : "",
3744 			(msr & 1 << 6) ? "VR-Therm, " : "",
3745 			(msr & 1 << 5) ? "Auto-HWP, " : "",
3746 			(msr & 1 << 4) ? "Graphics, " : "",
3747 			(msr & 1 << 2) ? "bit2, " : "",
3748 			(msr & 1 << 1) ? "ThermStatus, " : "",
3749 			(msr & 1 << 0) ? "PROCHOT, " : "");
3750 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
3751 			(msr & 1 << 31) ? "bit31, " : "",
3752 			(msr & 1 << 30) ? "bit30, " : "",
3753 			(msr & 1 << 29) ? "Transitions, " : "",
3754 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
3755 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
3756 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3757 			(msr & 1 << 25) ? "CorePwr, " : "",
3758 			(msr & 1 << 24) ? "Amps, " : "",
3759 			(msr & 1 << 22) ? "VR-Therm, " : "",
3760 			(msr & 1 << 21) ? "Auto-HWP, " : "",
3761 			(msr & 1 << 20) ? "Graphics, " : "",
3762 			(msr & 1 << 18) ? "bit18, " : "",
3763 			(msr & 1 << 17) ? "ThermStatus, " : "",
3764 			(msr & 1 << 16) ? "PROCHOT, " : "");
3765 
3766 	}
3767 	if (do_gfx_perf_limit_reasons) {
3768 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
3769 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3770 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
3771 			(msr & 1 << 0) ? "PROCHOT, " : "",
3772 			(msr & 1 << 1) ? "ThermStatus, " : "",
3773 			(msr & 1 << 4) ? "Graphics, " : "",
3774 			(msr & 1 << 6) ? "VR-Therm, " : "",
3775 			(msr & 1 << 8) ? "Amps, " : "",
3776 			(msr & 1 << 9) ? "GFXPwr, " : "",
3777 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3778 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3779 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
3780 			(msr & 1 << 16) ? "PROCHOT, " : "",
3781 			(msr & 1 << 17) ? "ThermStatus, " : "",
3782 			(msr & 1 << 20) ? "Graphics, " : "",
3783 			(msr & 1 << 22) ? "VR-Therm, " : "",
3784 			(msr & 1 << 24) ? "Amps, " : "",
3785 			(msr & 1 << 25) ? "GFXPwr, " : "",
3786 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3787 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3788 	}
3789 	if (do_ring_perf_limit_reasons) {
3790 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
3791 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3792 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
3793 			(msr & 1 << 0) ? "PROCHOT, " : "",
3794 			(msr & 1 << 1) ? "ThermStatus, " : "",
3795 			(msr & 1 << 6) ? "VR-Therm, " : "",
3796 			(msr & 1 << 8) ? "Amps, " : "",
3797 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3798 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3799 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3800 			(msr & 1 << 16) ? "PROCHOT, " : "",
3801 			(msr & 1 << 17) ? "ThermStatus, " : "",
3802 			(msr & 1 << 22) ? "VR-Therm, " : "",
3803 			(msr & 1 << 24) ? "Amps, " : "",
3804 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3805 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3806 	}
3807 	return 0;
3808 }
3809 
3810 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
3811 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
3812 
3813 double get_tdp_intel(unsigned int model)
3814 {
3815 	unsigned long long msr;
3816 
3817 	if (do_rapl & RAPL_PKG_POWER_INFO)
3818 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3819 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3820 
3821 	switch (model) {
3822 	case INTEL_FAM6_ATOM_SILVERMONT:
3823 	case INTEL_FAM6_ATOM_SILVERMONT_X:
3824 		return 30.0;
3825 	default:
3826 		return 135.0;
3827 	}
3828 }
3829 
3830 double get_tdp_amd(unsigned int family)
3831 {
3832 	switch (family) {
3833 	case 0x17:
3834 	default:
3835 		/* This is the max stock TDP of HEDT/Server Fam17h chips */
3836 		return 250.0;
3837 	}
3838 }
3839 
3840 /*
3841  * rapl_dram_energy_units_probe()
3842  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3843  */
3844 static double
3845 rapl_dram_energy_units_probe(int  model, double rapl_energy_units)
3846 {
3847 	/* only called for genuine_intel, family 6 */
3848 
3849 	switch (model) {
3850 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3851 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3852 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3853 		return (rapl_dram_energy_units = 15.3 / 1000000);
3854 	default:
3855 		return (rapl_energy_units);
3856 	}
3857 }
3858 
3859 void rapl_probe_intel(unsigned int family, unsigned int model)
3860 {
3861 	unsigned long long msr;
3862 	unsigned int time_unit;
3863 	double tdp;
3864 
3865 	if (family != 6)
3866 		return;
3867 
3868 	switch (model) {
3869 	case INTEL_FAM6_SANDYBRIDGE:
3870 	case INTEL_FAM6_IVYBRIDGE:
3871 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3872 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
3873 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3874 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3875 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3876 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3877 		if (rapl_joules) {
3878 			BIC_PRESENT(BIC_Pkg_J);
3879 			BIC_PRESENT(BIC_Cor_J);
3880 			BIC_PRESENT(BIC_GFX_J);
3881 		} else {
3882 			BIC_PRESENT(BIC_PkgWatt);
3883 			BIC_PRESENT(BIC_CorWatt);
3884 			BIC_PRESENT(BIC_GFXWatt);
3885 		}
3886 		break;
3887 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3888 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3889 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3890 		if (rapl_joules)
3891 			BIC_PRESENT(BIC_Pkg_J);
3892 		else
3893 			BIC_PRESENT(BIC_PkgWatt);
3894 		break;
3895 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3896 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3897 		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;
3898 		BIC_PRESENT(BIC_PKG__);
3899 		BIC_PRESENT(BIC_RAM__);
3900 		if (rapl_joules) {
3901 			BIC_PRESENT(BIC_Pkg_J);
3902 			BIC_PRESENT(BIC_Cor_J);
3903 			BIC_PRESENT(BIC_RAM_J);
3904 			BIC_PRESENT(BIC_GFX_J);
3905 		} else {
3906 			BIC_PRESENT(BIC_PkgWatt);
3907 			BIC_PRESENT(BIC_CorWatt);
3908 			BIC_PRESENT(BIC_RAMWatt);
3909 			BIC_PRESENT(BIC_GFXWatt);
3910 		}
3911 		break;
3912 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3913 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3914 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3915 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3916 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3917 		BIC_PRESENT(BIC_PKG__);
3918 		BIC_PRESENT(BIC_RAM__);
3919 		if (rapl_joules) {
3920 			BIC_PRESENT(BIC_Pkg_J);
3921 			BIC_PRESENT(BIC_RAM_J);
3922 		} else {
3923 			BIC_PRESENT(BIC_PkgWatt);
3924 			BIC_PRESENT(BIC_RAMWatt);
3925 		}
3926 		break;
3927 	case INTEL_FAM6_SANDYBRIDGE_X:
3928 	case INTEL_FAM6_IVYBRIDGE_X:
3929 		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;
3930 		BIC_PRESENT(BIC_PKG__);
3931 		BIC_PRESENT(BIC_RAM__);
3932 		if (rapl_joules) {
3933 			BIC_PRESENT(BIC_Pkg_J);
3934 			BIC_PRESENT(BIC_Cor_J);
3935 			BIC_PRESENT(BIC_RAM_J);
3936 		} else {
3937 			BIC_PRESENT(BIC_PkgWatt);
3938 			BIC_PRESENT(BIC_CorWatt);
3939 			BIC_PRESENT(BIC_RAMWatt);
3940 		}
3941 		break;
3942 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3943 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3944 		do_rapl = RAPL_PKG | RAPL_CORES;
3945 		if (rapl_joules) {
3946 			BIC_PRESENT(BIC_Pkg_J);
3947 			BIC_PRESENT(BIC_Cor_J);
3948 		} else {
3949 			BIC_PRESENT(BIC_PkgWatt);
3950 			BIC_PRESENT(BIC_CorWatt);
3951 		}
3952 		break;
3953 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3954 		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;
3955 		BIC_PRESENT(BIC_PKG__);
3956 		BIC_PRESENT(BIC_RAM__);
3957 		if (rapl_joules) {
3958 			BIC_PRESENT(BIC_Pkg_J);
3959 			BIC_PRESENT(BIC_Cor_J);
3960 			BIC_PRESENT(BIC_RAM_J);
3961 		} else {
3962 			BIC_PRESENT(BIC_PkgWatt);
3963 			BIC_PRESENT(BIC_CorWatt);
3964 			BIC_PRESENT(BIC_RAMWatt);
3965 		}
3966 		break;
3967 	default:
3968 		return;
3969 	}
3970 
3971 	/* units on package 0, verify later other packages match */
3972 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3973 		return;
3974 
3975 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
3976 	if (model == INTEL_FAM6_ATOM_SILVERMONT)
3977 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3978 	else
3979 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3980 
3981 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3982 
3983 	time_unit = msr >> 16 & 0xF;
3984 	if (time_unit == 0)
3985 		time_unit = 0xA;
3986 
3987 	rapl_time_units = 1.0 / (1 << (time_unit));
3988 
3989 	tdp = get_tdp_intel(model);
3990 
3991 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3992 	if (!quiet)
3993 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3994 }
3995 
3996 void rapl_probe_amd(unsigned int family, unsigned int model)
3997 {
3998 	unsigned long long msr;
3999 	unsigned int eax, ebx, ecx, edx;
4000 	unsigned int has_rapl = 0;
4001 	double tdp;
4002 
4003 	if (max_extended_level >= 0x80000007) {
4004 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4005 		/* RAPL (Fam 17h) */
4006 		has_rapl = edx & (1 << 14);
4007 	}
4008 
4009 	if (!has_rapl)
4010 		return;
4011 
4012 	switch (family) {
4013 	case 0x17: /* Zen, Zen+ */
4014 		do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY;
4015 		if (rapl_joules) {
4016 			BIC_PRESENT(BIC_Pkg_J);
4017 			BIC_PRESENT(BIC_Cor_J);
4018 		} else {
4019 			BIC_PRESENT(BIC_PkgWatt);
4020 			BIC_PRESENT(BIC_CorWatt);
4021 		}
4022 		break;
4023 	default:
4024 		return;
4025 	}
4026 
4027 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
4028 		return;
4029 
4030 	rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf));
4031 	rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f));
4032 	rapl_power_units = ldexp(1.0, -(msr & 0xf));
4033 
4034 	tdp = get_tdp_amd(family);
4035 
4036 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4037 	if (!quiet)
4038 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4039 }
4040 
4041 /*
4042  * rapl_probe()
4043  *
4044  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
4045  */
4046 void rapl_probe(unsigned int family, unsigned int model)
4047 {
4048 	if (genuine_intel)
4049 		rapl_probe_intel(family, model);
4050 	if (authentic_amd)
4051 		rapl_probe_amd(family, model);
4052 }
4053 
4054 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
4055 {
4056 	if (!genuine_intel)
4057 		return;
4058 
4059 	if (family != 6)
4060 		return;
4061 
4062 	switch (model) {
4063 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4064 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
4065 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4066 		do_gfx_perf_limit_reasons = 1;
4067 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4068 		do_core_perf_limit_reasons = 1;
4069 		do_ring_perf_limit_reasons = 1;
4070 	default:
4071 		return;
4072 	}
4073 }
4074 
4075 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model)
4076 {
4077 	if (is_skx(family, model) || is_bdx(family, model))
4078 		has_automatic_cstate_conversion = 1;
4079 }
4080 
4081 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4082 {
4083 	unsigned long long msr;
4084 	unsigned int dts, dts2;
4085 	int cpu;
4086 
4087 	if (!(do_dts || do_ptm))
4088 		return 0;
4089 
4090 	cpu = t->cpu_id;
4091 
4092 	/* DTS is per-core, no need to print for each thread */
4093 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
4094 		return 0;
4095 
4096 	if (cpu_migrate(cpu)) {
4097 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4098 		return -1;
4099 	}
4100 
4101 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
4102 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
4103 			return 0;
4104 
4105 		dts = (msr >> 16) & 0x7F;
4106 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
4107 			cpu, msr, tcc_activation_temp - dts);
4108 
4109 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
4110 			return 0;
4111 
4112 		dts = (msr >> 16) & 0x7F;
4113 		dts2 = (msr >> 8) & 0x7F;
4114 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4115 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4116 	}
4117 
4118 
4119 	if (do_dts && debug) {
4120 		unsigned int resolution;
4121 
4122 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
4123 			return 0;
4124 
4125 		dts = (msr >> 16) & 0x7F;
4126 		resolution = (msr >> 27) & 0xF;
4127 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
4128 			cpu, msr, tcc_activation_temp - dts, resolution);
4129 
4130 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
4131 			return 0;
4132 
4133 		dts = (msr >> 16) & 0x7F;
4134 		dts2 = (msr >> 8) & 0x7F;
4135 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4136 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4137 	}
4138 
4139 	return 0;
4140 }
4141 
4142 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
4143 {
4144 	fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
4145 		cpu, label,
4146 		((msr >> 15) & 1) ? "EN" : "DIS",
4147 		((msr >> 0) & 0x7FFF) * rapl_power_units,
4148 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
4149 		(((msr >> 16) & 1) ? "EN" : "DIS"));
4150 
4151 	return;
4152 }
4153 
4154 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4155 {
4156 	unsigned long long msr;
4157 	const char *msr_name;
4158 	int cpu;
4159 
4160 	if (!do_rapl)
4161 		return 0;
4162 
4163 	/* RAPL counters are per package, so print only for 1st thread/package */
4164 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4165 		return 0;
4166 
4167 	cpu = t->cpu_id;
4168 	if (cpu_migrate(cpu)) {
4169 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4170 		return -1;
4171 	}
4172 
4173 	if (do_rapl & RAPL_AMD_F17H) {
4174 		msr_name = "MSR_RAPL_PWR_UNIT";
4175 		if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr))
4176 			return -1;
4177 	} else {
4178 		msr_name = "MSR_RAPL_POWER_UNIT";
4179 		if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
4180 			return -1;
4181 	}
4182 
4183 	fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr,
4184 		rapl_power_units, rapl_energy_units, rapl_time_units);
4185 
4186 	if (do_rapl & RAPL_PKG_POWER_INFO) {
4187 
4188 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
4189                 	return -5;
4190 
4191 
4192 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4193 			cpu, msr,
4194 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4195 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4196 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4197 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4198 
4199 	}
4200 	if (do_rapl & RAPL_PKG) {
4201 
4202 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
4203 			return -9;
4204 
4205 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
4206 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
4207 
4208 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
4209 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
4210 			cpu,
4211 			((msr >> 47) & 1) ? "EN" : "DIS",
4212 			((msr >> 32) & 0x7FFF) * rapl_power_units,
4213 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
4214 			((msr >> 48) & 1) ? "EN" : "DIS");
4215 	}
4216 
4217 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
4218 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
4219                 	return -6;
4220 
4221 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4222 			cpu, msr,
4223 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4224 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4225 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4226 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4227 	}
4228 	if (do_rapl & RAPL_DRAM) {
4229 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
4230 			return -9;
4231 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
4232 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4233 
4234 		print_power_limit_msr(cpu, msr, "DRAM Limit");
4235 	}
4236 	if (do_rapl & RAPL_CORE_POLICY) {
4237 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
4238 			return -7;
4239 
4240 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
4241 	}
4242 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
4243 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
4244 			return -9;
4245 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
4246 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4247 		print_power_limit_msr(cpu, msr, "Cores Limit");
4248 	}
4249 	if (do_rapl & RAPL_GFX) {
4250 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
4251 			return -8;
4252 
4253 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
4254 
4255 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
4256 			return -9;
4257 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
4258 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4259 		print_power_limit_msr(cpu, msr, "GFX Limit");
4260 	}
4261 	return 0;
4262 }
4263 
4264 /*
4265  * SNB adds support for additional MSRs:
4266  *
4267  * MSR_PKG_C7_RESIDENCY            0x000003fa
4268  * MSR_CORE_C7_RESIDENCY           0x000003fe
4269  * MSR_PKG_C2_RESIDENCY            0x0000060d
4270  */
4271 
4272 int has_snb_msrs(unsigned int family, unsigned int model)
4273 {
4274 	if (!genuine_intel)
4275 		return 0;
4276 
4277 	switch (model) {
4278 	case INTEL_FAM6_SANDYBRIDGE:
4279 	case INTEL_FAM6_SANDYBRIDGE_X:
4280 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
4281 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
4282 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4283 	case INTEL_FAM6_HASWELL_X:	/* HSW */
4284 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
4285 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4286 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4287 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
4288 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4289 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4290 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4291 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4292 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4293 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4294 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4295 		return 1;
4296 	}
4297 	return 0;
4298 }
4299 
4300 /*
4301  * HSW ULT added support for C8/C9/C10 MSRs:
4302  *
4303  * MSR_PKG_C8_RESIDENCY		0x00000630
4304  * MSR_PKG_C9_RESIDENCY		0x00000631
4305  * MSR_PKG_C10_RESIDENCY	0x00000632
4306  *
4307  * MSR_PKGC8_IRTL		0x00000633
4308  * MSR_PKGC9_IRTL		0x00000634
4309  * MSR_PKGC10_IRTL		0x00000635
4310  *
4311  */
4312 int has_c8910_msrs(unsigned int family, unsigned int model)
4313 {
4314 	if (!genuine_intel)
4315 		return 0;
4316 
4317 	switch (model) {
4318 	case INTEL_FAM6_HASWELL_ULT:	/* HSW */
4319 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4320 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4321 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4322 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4323 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4324 		return 1;
4325 	}
4326 	return 0;
4327 }
4328 
4329 /*
4330  * SKL adds support for additional MSRS:
4331  *
4332  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
4333  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
4334  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
4335  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
4336  */
4337 int has_skl_msrs(unsigned int family, unsigned int model)
4338 {
4339 	if (!genuine_intel)
4340 		return 0;
4341 
4342 	switch (model) {
4343 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4344 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4345 		return 1;
4346 	}
4347 	return 0;
4348 }
4349 
4350 int is_slm(unsigned int family, unsigned int model)
4351 {
4352 	if (!genuine_intel)
4353 		return 0;
4354 	switch (model) {
4355 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
4356 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
4357 		return 1;
4358 	}
4359 	return 0;
4360 }
4361 
4362 int is_knl(unsigned int family, unsigned int model)
4363 {
4364 	if (!genuine_intel)
4365 		return 0;
4366 	switch (model) {
4367 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4368 		return 1;
4369 	}
4370 	return 0;
4371 }
4372 
4373 int is_cnl(unsigned int family, unsigned int model)
4374 {
4375 	if (!genuine_intel)
4376 		return 0;
4377 
4378 	switch (model) {
4379 	case INTEL_FAM6_CANNONLAKE_MOBILE: /* CNL */
4380 		return 1;
4381 	}
4382 
4383 	return 0;
4384 }
4385 
4386 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
4387 {
4388 	if (is_knl(family, model))
4389 		return 1024;
4390 	return 1;
4391 }
4392 
4393 #define SLM_BCLK_FREQS 5
4394 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
4395 
4396 double slm_bclk(void)
4397 {
4398 	unsigned long long msr = 3;
4399 	unsigned int i;
4400 	double freq;
4401 
4402 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
4403 		fprintf(outf, "SLM BCLK: unknown\n");
4404 
4405 	i = msr & 0xf;
4406 	if (i >= SLM_BCLK_FREQS) {
4407 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
4408 		i = 3;
4409 	}
4410 	freq = slm_freq_table[i];
4411 
4412 	if (!quiet)
4413 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
4414 
4415 	return freq;
4416 }
4417 
4418 double discover_bclk(unsigned int family, unsigned int model)
4419 {
4420 	if (has_snb_msrs(family, model) || is_knl(family, model))
4421 		return 100.00;
4422 	else if (is_slm(family, model))
4423 		return slm_bclk();
4424 	else
4425 		return 133.33;
4426 }
4427 
4428 /*
4429  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
4430  * the Thermal Control Circuit (TCC) activates.
4431  * This is usually equal to tjMax.
4432  *
4433  * Older processors do not have this MSR, so there we guess,
4434  * but also allow cmdline over-ride with -T.
4435  *
4436  * Several MSR temperature values are in units of degrees-C
4437  * below this value, including the Digital Thermal Sensor (DTS),
4438  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
4439  */
4440 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4441 {
4442 	unsigned long long msr;
4443 	unsigned int target_c_local;
4444 	int cpu;
4445 
4446 	/* tcc_activation_temp is used only for dts or ptm */
4447 	if (!(do_dts || do_ptm))
4448 		return 0;
4449 
4450 	/* this is a per-package concept */
4451 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4452 		return 0;
4453 
4454 	cpu = t->cpu_id;
4455 	if (cpu_migrate(cpu)) {
4456 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4457 		return -1;
4458 	}
4459 
4460 	if (tcc_activation_temp_override != 0) {
4461 		tcc_activation_temp = tcc_activation_temp_override;
4462 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
4463 			cpu, tcc_activation_temp);
4464 		return 0;
4465 	}
4466 
4467 	/* Temperature Target MSR is Nehalem and newer only */
4468 	if (!do_nhm_platform_info)
4469 		goto guess;
4470 
4471 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
4472 		goto guess;
4473 
4474 	target_c_local = (msr >> 16) & 0xFF;
4475 
4476 	if (!quiet)
4477 		fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
4478 			cpu, msr, target_c_local);
4479 
4480 	if (!target_c_local)
4481 		goto guess;
4482 
4483 	tcc_activation_temp = target_c_local;
4484 
4485 	return 0;
4486 
4487 guess:
4488 	tcc_activation_temp = TJMAX_DEFAULT;
4489 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
4490 		cpu, tcc_activation_temp);
4491 
4492 	return 0;
4493 }
4494 
4495 void decode_feature_control_msr(void)
4496 {
4497 	unsigned long long msr;
4498 
4499 	if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
4500 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
4501 			base_cpu, msr,
4502 			msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
4503 			msr & (1 << 18) ? "SGX" : "");
4504 }
4505 
4506 void decode_misc_enable_msr(void)
4507 {
4508 	unsigned long long msr;
4509 
4510 	if (!genuine_intel)
4511 		return;
4512 
4513 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
4514 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
4515 			base_cpu, msr,
4516 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
4517 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
4518 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-",
4519 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
4520 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
4521 }
4522 
4523 void decode_misc_feature_control(void)
4524 {
4525 	unsigned long long msr;
4526 
4527 	if (!has_misc_feature_control)
4528 		return;
4529 
4530 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
4531 		fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
4532 			base_cpu, msr,
4533 			msr & (0 << 0) ? "No-" : "",
4534 			msr & (1 << 0) ? "No-" : "",
4535 			msr & (2 << 0) ? "No-" : "",
4536 			msr & (3 << 0) ? "No-" : "");
4537 }
4538 /*
4539  * Decode MSR_MISC_PWR_MGMT
4540  *
4541  * Decode the bits according to the Nehalem documentation
4542  * bit[0] seems to continue to have same meaning going forward
4543  * bit[1] less so...
4544  */
4545 void decode_misc_pwr_mgmt_msr(void)
4546 {
4547 	unsigned long long msr;
4548 
4549 	if (!do_nhm_platform_info)
4550 		return;
4551 
4552 	if (no_MSR_MISC_PWR_MGMT)
4553 		return;
4554 
4555 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
4556 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
4557 			base_cpu, msr,
4558 			msr & (1 << 0) ? "DIS" : "EN",
4559 			msr & (1 << 1) ? "EN" : "DIS",
4560 			msr & (1 << 8) ? "EN" : "DIS");
4561 }
4562 /*
4563  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
4564  *
4565  * This MSRs are present on Silvermont processors,
4566  * Intel Atom processor E3000 series (Baytrail), and friends.
4567  */
4568 void decode_c6_demotion_policy_msr(void)
4569 {
4570 	unsigned long long msr;
4571 
4572 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
4573 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
4574 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4575 
4576 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
4577 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
4578 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4579 }
4580 
4581 /*
4582  * When models are the same, for the purpose of turbostat, reuse
4583  */
4584 unsigned int intel_model_duplicates(unsigned int model)
4585 {
4586 
4587 	switch(model) {
4588 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
4589 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
4590 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
4591 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
4592 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
4593 		return INTEL_FAM6_NEHALEM;
4594 
4595 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
4596 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
4597 		return INTEL_FAM6_NEHALEM_EX;
4598 
4599 	case INTEL_FAM6_XEON_PHI_KNM:
4600 		return INTEL_FAM6_XEON_PHI_KNL;
4601 
4602 	case INTEL_FAM6_BROADWELL_X:
4603 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
4604 		return INTEL_FAM6_BROADWELL_X;
4605 
4606 	case INTEL_FAM6_SKYLAKE_MOBILE:
4607 	case INTEL_FAM6_SKYLAKE_DESKTOP:
4608 	case INTEL_FAM6_KABYLAKE_MOBILE:
4609 	case INTEL_FAM6_KABYLAKE_DESKTOP:
4610 		return INTEL_FAM6_SKYLAKE_MOBILE;
4611 
4612 	case INTEL_FAM6_ICELAKE_MOBILE:
4613 	case INTEL_FAM6_ICELAKE_NNPI:
4614 		return INTEL_FAM6_CANNONLAKE_MOBILE;
4615 
4616 	case INTEL_FAM6_ATOM_TREMONT_X:
4617 		return INTEL_FAM6_ATOM_GOLDMONT_X;
4618 	}
4619 	return model;
4620 }
4621 void process_cpuid()
4622 {
4623 	unsigned int eax, ebx, ecx, edx;
4624 	unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
4625 	unsigned int has_turbo;
4626 
4627 	eax = ebx = ecx = edx = 0;
4628 
4629 	__cpuid(0, max_level, ebx, ecx, edx);
4630 
4631 	if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
4632 		genuine_intel = 1;
4633 	else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
4634 		authentic_amd = 1;
4635 
4636 	if (!quiet)
4637 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
4638 			(char *)&ebx, (char *)&edx, (char *)&ecx);
4639 
4640 	__cpuid(1, fms, ebx, ecx, edx);
4641 	family = (fms >> 8) & 0xf;
4642 	model = (fms >> 4) & 0xf;
4643 	stepping = fms & 0xf;
4644 	if (family == 0xf)
4645 		family += (fms >> 20) & 0xff;
4646 	if (family >= 6)
4647 		model += ((fms >> 16) & 0xf) << 4;
4648 	ecx_flags = ecx;
4649 	edx_flags = edx;
4650 
4651 	/*
4652 	 * check max extended function levels of CPUID.
4653 	 * This is needed to check for invariant TSC.
4654 	 * This check is valid for both Intel and AMD.
4655 	 */
4656 	ebx = ecx = edx = 0;
4657 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
4658 
4659 	if (!quiet) {
4660 		fprintf(outf, "0x%x CPUID levels; 0x%x xlevels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
4661 			max_level, max_extended_level, family, model, stepping, family, model, stepping);
4662 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n",
4663 			ecx_flags & (1 << 0) ? "SSE3" : "-",
4664 			ecx_flags & (1 << 3) ? "MONITOR" : "-",
4665 			ecx_flags & (1 << 6) ? "SMX" : "-",
4666 			ecx_flags & (1 << 7) ? "EIST" : "-",
4667 			ecx_flags & (1 << 8) ? "TM2" : "-",
4668 			edx_flags & (1 << 4) ? "TSC" : "-",
4669 			edx_flags & (1 << 5) ? "MSR" : "-",
4670 			edx_flags & (1 << 22) ? "ACPI-TM" : "-",
4671 			edx_flags & (1 << 28) ? "HT" : "-",
4672 			edx_flags & (1 << 29) ? "TM" : "-");
4673 	}
4674 	if (genuine_intel)
4675 		model = intel_model_duplicates(model);
4676 
4677 	if (!(edx_flags & (1 << 5)))
4678 		errx(1, "CPUID: no MSR");
4679 
4680 	if (max_extended_level >= 0x80000007) {
4681 
4682 		/*
4683 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
4684 		 * this check is valid for both Intel and AMD
4685 		 */
4686 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4687 		has_invariant_tsc = edx & (1 << 8);
4688 	}
4689 
4690 	/*
4691 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
4692 	 * this check is valid for both Intel and AMD
4693 	 */
4694 
4695 	__cpuid(0x6, eax, ebx, ecx, edx);
4696 	has_aperf = ecx & (1 << 0);
4697 	if (has_aperf) {
4698 		BIC_PRESENT(BIC_Avg_MHz);
4699 		BIC_PRESENT(BIC_Busy);
4700 		BIC_PRESENT(BIC_Bzy_MHz);
4701 	}
4702 	do_dts = eax & (1 << 0);
4703 	if (do_dts)
4704 		BIC_PRESENT(BIC_CoreTmp);
4705 	has_turbo = eax & (1 << 1);
4706 	do_ptm = eax & (1 << 6);
4707 	if (do_ptm)
4708 		BIC_PRESENT(BIC_PkgTmp);
4709 	has_hwp = eax & (1 << 7);
4710 	has_hwp_notify = eax & (1 << 8);
4711 	has_hwp_activity_window = eax & (1 << 9);
4712 	has_hwp_epp = eax & (1 << 10);
4713 	has_hwp_pkg = eax & (1 << 11);
4714 	has_epb = ecx & (1 << 3);
4715 
4716 	if (!quiet)
4717 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
4718 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
4719 			has_aperf ? "" : "No-",
4720 			has_turbo ? "" : "No-",
4721 			do_dts ? "" : "No-",
4722 			do_ptm ? "" : "No-",
4723 			has_hwp ? "" : "No-",
4724 			has_hwp_notify ? "" : "No-",
4725 			has_hwp_activity_window ? "" : "No-",
4726 			has_hwp_epp ? "" : "No-",
4727 			has_hwp_pkg ? "" : "No-",
4728 			has_epb ? "" : "No-");
4729 
4730 	if (!quiet)
4731 		decode_misc_enable_msr();
4732 
4733 
4734 	if (max_level >= 0x7 && !quiet) {
4735 		int has_sgx;
4736 
4737 		ecx = 0;
4738 
4739 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
4740 
4741 		has_sgx = ebx & (1 << 2);
4742 		fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
4743 
4744 		if (has_sgx)
4745 			decode_feature_control_msr();
4746 	}
4747 
4748 	if (max_level >= 0x15) {
4749 		unsigned int eax_crystal;
4750 		unsigned int ebx_tsc;
4751 
4752 		/*
4753 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
4754 		 */
4755 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
4756 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
4757 
4758 		if (ebx_tsc != 0) {
4759 
4760 			if (!quiet && (ebx != 0))
4761 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
4762 					eax_crystal, ebx_tsc, crystal_hz);
4763 
4764 			if (crystal_hz == 0)
4765 				switch(model) {
4766 				case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4767 					crystal_hz = 24000000;	/* 24.0 MHz */
4768 					break;
4769 				case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4770 					crystal_hz = 25000000;	/* 25.0 MHz */
4771 					break;
4772 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4773 				case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4774 					crystal_hz = 19200000;	/* 19.2 MHz */
4775 					break;
4776 				default:
4777 					crystal_hz = 0;
4778 			}
4779 
4780 			if (crystal_hz) {
4781 				tsc_hz =  (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
4782 				if (!quiet)
4783 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
4784 						tsc_hz / 1000000, crystal_hz, ebx_tsc,  eax_crystal);
4785 			}
4786 		}
4787 	}
4788 	if (max_level >= 0x16) {
4789 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
4790 
4791 		/*
4792 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
4793 		 */
4794 		base_mhz = max_mhz = bus_mhz = edx = 0;
4795 
4796 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
4797 		if (!quiet)
4798 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
4799 				base_mhz, max_mhz, bus_mhz);
4800 	}
4801 
4802 	if (has_aperf)
4803 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
4804 
4805 	BIC_PRESENT(BIC_IRQ);
4806 	BIC_PRESENT(BIC_TSC_MHz);
4807 
4808 	if (probe_nhm_msrs(family, model)) {
4809 		do_nhm_platform_info = 1;
4810 		BIC_PRESENT(BIC_CPU_c1);
4811 		BIC_PRESENT(BIC_CPU_c3);
4812 		BIC_PRESENT(BIC_CPU_c6);
4813 		BIC_PRESENT(BIC_SMI);
4814 	}
4815 	do_snb_cstates = has_snb_msrs(family, model);
4816 
4817 	if (do_snb_cstates)
4818 		BIC_PRESENT(BIC_CPU_c7);
4819 
4820 	do_irtl_snb = has_snb_msrs(family, model);
4821 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
4822 		BIC_PRESENT(BIC_Pkgpc2);
4823 	if (pkg_cstate_limit >= PCL__3)
4824 		BIC_PRESENT(BIC_Pkgpc3);
4825 	if (pkg_cstate_limit >= PCL__6)
4826 		BIC_PRESENT(BIC_Pkgpc6);
4827 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
4828 		BIC_PRESENT(BIC_Pkgpc7);
4829 	if (has_slv_msrs(family, model)) {
4830 		BIC_NOT_PRESENT(BIC_Pkgpc2);
4831 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4832 		BIC_PRESENT(BIC_Pkgpc6);
4833 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4834 		BIC_PRESENT(BIC_Mod_c6);
4835 		use_c1_residency_msr = 1;
4836 	}
4837 	if (is_dnv(family, model)) {
4838 		BIC_PRESENT(BIC_CPU_c1);
4839 		BIC_NOT_PRESENT(BIC_CPU_c3);
4840 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4841 		BIC_NOT_PRESENT(BIC_CPU_c7);
4842 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4843 		use_c1_residency_msr = 1;
4844 	}
4845 	if (is_skx(family, model)) {
4846 		BIC_NOT_PRESENT(BIC_CPU_c3);
4847 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4848 		BIC_NOT_PRESENT(BIC_CPU_c7);
4849 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4850 	}
4851 	if (is_bdx(family, model)) {
4852 		BIC_NOT_PRESENT(BIC_CPU_c7);
4853 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4854 	}
4855 	if (has_c8910_msrs(family, model)) {
4856 		BIC_PRESENT(BIC_Pkgpc8);
4857 		BIC_PRESENT(BIC_Pkgpc9);
4858 		BIC_PRESENT(BIC_Pkgpc10);
4859 	}
4860 	do_irtl_hsw = has_c8910_msrs(family, model);
4861 	if (has_skl_msrs(family, model)) {
4862 		BIC_PRESENT(BIC_Totl_c0);
4863 		BIC_PRESENT(BIC_Any_c0);
4864 		BIC_PRESENT(BIC_GFX_c0);
4865 		BIC_PRESENT(BIC_CPUGFX);
4866 	}
4867 	do_slm_cstates = is_slm(family, model);
4868 	do_knl_cstates  = is_knl(family, model);
4869 
4870 	if (do_slm_cstates || do_knl_cstates || is_cnl(family, model))
4871 		BIC_NOT_PRESENT(BIC_CPU_c3);
4872 
4873 	if (!quiet)
4874 		decode_misc_pwr_mgmt_msr();
4875 
4876 	if (!quiet && has_slv_msrs(family, model))
4877 		decode_c6_demotion_policy_msr();
4878 
4879 	rapl_probe(family, model);
4880 	perf_limit_reasons_probe(family, model);
4881 	automatic_cstate_conversion_probe(family, model);
4882 
4883 	if (!quiet)
4884 		dump_cstate_pstate_config_info(family, model);
4885 
4886 	if (!quiet)
4887 		dump_sysfs_cstate_config();
4888 	if (!quiet)
4889 		dump_sysfs_pstate_config();
4890 
4891 	if (has_skl_msrs(family, model))
4892 		calculate_tsc_tweak();
4893 
4894 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
4895 		BIC_PRESENT(BIC_GFX_rc6);
4896 
4897 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
4898 		BIC_PRESENT(BIC_GFXMHz);
4899 
4900 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK))
4901 		BIC_PRESENT(BIC_CPU_LPI);
4902 	else
4903 		BIC_NOT_PRESENT(BIC_CPU_LPI);
4904 
4905 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", R_OK))
4906 		BIC_PRESENT(BIC_SYS_LPI);
4907 	else
4908 		BIC_NOT_PRESENT(BIC_SYS_LPI);
4909 
4910 	if (!quiet)
4911 		decode_misc_feature_control();
4912 
4913 	return;
4914 }
4915 
4916 /*
4917  * in /dev/cpu/ return success for names that are numbers
4918  * ie. filter out ".", "..", "microcode".
4919  */
4920 int dir_filter(const struct dirent *dirp)
4921 {
4922 	if (isdigit(dirp->d_name[0]))
4923 		return 1;
4924 	else
4925 		return 0;
4926 }
4927 
4928 int open_dev_cpu_msr(int dummy1)
4929 {
4930 	return 0;
4931 }
4932 
4933 void topology_probe()
4934 {
4935 	int i;
4936 	int max_core_id = 0;
4937 	int max_package_id = 0;
4938 	int max_die_id = 0;
4939 	int max_siblings = 0;
4940 
4941 	/* Initialize num_cpus, max_cpu_num */
4942 	set_max_cpu_num();
4943 	topo.num_cpus = 0;
4944 	for_all_proc_cpus(count_cpus);
4945 	if (!summary_only && topo.num_cpus > 1)
4946 		BIC_PRESENT(BIC_CPU);
4947 
4948 	if (debug > 1)
4949 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4950 
4951 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
4952 	if (cpus == NULL)
4953 		err(1, "calloc cpus");
4954 
4955 	/*
4956 	 * Allocate and initialize cpu_present_set
4957 	 */
4958 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4959 	if (cpu_present_set == NULL)
4960 		err(3, "CPU_ALLOC");
4961 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4962 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4963 	for_all_proc_cpus(mark_cpu_present);
4964 
4965 	/*
4966 	 * Validate that all cpus in cpu_subset are also in cpu_present_set
4967 	 */
4968 	for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
4969 		if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
4970 			if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
4971 				err(1, "cpu%d not present", i);
4972 	}
4973 
4974 	/*
4975 	 * Allocate and initialize cpu_affinity_set
4976 	 */
4977 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4978 	if (cpu_affinity_set == NULL)
4979 		err(3, "CPU_ALLOC");
4980 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4981 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4982 
4983 	for_all_proc_cpus(init_thread_id);
4984 
4985 	/*
4986 	 * For online cpus
4987 	 * find max_core_id, max_package_id
4988 	 */
4989 	for (i = 0; i <= topo.max_cpu_num; ++i) {
4990 		int siblings;
4991 
4992 		if (cpu_is_not_present(i)) {
4993 			if (debug > 1)
4994 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
4995 			continue;
4996 		}
4997 
4998 		cpus[i].logical_cpu_id = i;
4999 
5000 		/* get package information */
5001 		cpus[i].physical_package_id = get_physical_package_id(i);
5002 		if (cpus[i].physical_package_id > max_package_id)
5003 			max_package_id = cpus[i].physical_package_id;
5004 
5005 		/* get die information */
5006 		cpus[i].die_id = get_die_id(i);
5007 		if (cpus[i].die_id > max_die_id)
5008 			max_die_id = cpus[i].die_id;
5009 
5010 		/* get numa node information */
5011 		cpus[i].physical_node_id = get_physical_node_id(&cpus[i]);
5012 		if (cpus[i].physical_node_id > topo.max_node_num)
5013 			topo.max_node_num = cpus[i].physical_node_id;
5014 
5015 		/* get core information */
5016 		cpus[i].physical_core_id = get_core_id(i);
5017 		if (cpus[i].physical_core_id > max_core_id)
5018 			max_core_id = cpus[i].physical_core_id;
5019 
5020 		/* get thread information */
5021 		siblings = get_thread_siblings(&cpus[i]);
5022 		if (siblings > max_siblings)
5023 			max_siblings = siblings;
5024 		if (cpus[i].thread_id == 0)
5025 			topo.num_cores++;
5026 	}
5027 
5028 	topo.cores_per_node = max_core_id + 1;
5029 	if (debug > 1)
5030 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
5031 			max_core_id, topo.cores_per_node);
5032 	if (!summary_only && topo.cores_per_node > 1)
5033 		BIC_PRESENT(BIC_Core);
5034 
5035 	topo.num_die = max_die_id + 1;
5036 	if (debug > 1)
5037 		fprintf(outf, "max_die_id %d, sizing for %d die\n",
5038 				max_die_id, topo.num_die);
5039 	if (!summary_only && topo.num_die > 1)
5040 		BIC_PRESENT(BIC_Die);
5041 
5042 	topo.num_packages = max_package_id + 1;
5043 	if (debug > 1)
5044 		fprintf(outf, "max_package_id %d, sizing for %d packages\n",
5045 			max_package_id, topo.num_packages);
5046 	if (!summary_only && topo.num_packages > 1)
5047 		BIC_PRESENT(BIC_Package);
5048 
5049 	set_node_data();
5050 	if (debug > 1)
5051 		fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg);
5052 	if (!summary_only && topo.nodes_per_pkg > 1)
5053 		BIC_PRESENT(BIC_Node);
5054 
5055 	topo.threads_per_core = max_siblings;
5056 	if (debug > 1)
5057 		fprintf(outf, "max_siblings %d\n", max_siblings);
5058 
5059 	if (debug < 1)
5060 		return;
5061 
5062 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5063 		if (cpu_is_not_present(i))
5064 			continue;
5065 		fprintf(outf,
5066 			"cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n",
5067 			i, cpus[i].physical_package_id, cpus[i].die_id,
5068 			cpus[i].physical_node_id,
5069 			cpus[i].logical_node_id,
5070 			cpus[i].physical_core_id,
5071 			cpus[i].thread_id);
5072 	}
5073 
5074 }
5075 
5076 void
5077 allocate_counters(struct thread_data **t, struct core_data **c,
5078 		  struct pkg_data **p)
5079 {
5080 	int i;
5081 	int num_cores = topo.cores_per_node * topo.nodes_per_pkg *
5082 			topo.num_packages;
5083 	int num_threads = topo.threads_per_core * num_cores;
5084 
5085 	*t = calloc(num_threads, sizeof(struct thread_data));
5086 	if (*t == NULL)
5087 		goto error;
5088 
5089 	for (i = 0; i < num_threads; i++)
5090 		(*t)[i].cpu_id = -1;
5091 
5092 	*c = calloc(num_cores, sizeof(struct core_data));
5093 	if (*c == NULL)
5094 		goto error;
5095 
5096 	for (i = 0; i < num_cores; i++)
5097 		(*c)[i].core_id = -1;
5098 
5099 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
5100 	if (*p == NULL)
5101 		goto error;
5102 
5103 	for (i = 0; i < topo.num_packages; i++)
5104 		(*p)[i].package_id = i;
5105 
5106 	return;
5107 error:
5108 	err(1, "calloc counters");
5109 }
5110 /*
5111  * init_counter()
5112  *
5113  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
5114  */
5115 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
5116 	struct pkg_data *pkg_base, int cpu_id)
5117 {
5118 	int pkg_id = cpus[cpu_id].physical_package_id;
5119 	int node_id = cpus[cpu_id].logical_node_id;
5120 	int core_id = cpus[cpu_id].physical_core_id;
5121 	int thread_id = cpus[cpu_id].thread_id;
5122 	struct thread_data *t;
5123 	struct core_data *c;
5124 	struct pkg_data *p;
5125 
5126 
5127 	/* Workaround for systems where physical_node_id==-1
5128 	 * and logical_node_id==(-1 - topo.num_cpus)
5129 	 */
5130 	if (node_id < 0)
5131 		node_id = 0;
5132 
5133 	t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id);
5134 	c = GET_CORE(core_base, core_id, node_id, pkg_id);
5135 	p = GET_PKG(pkg_base, pkg_id);
5136 
5137 	t->cpu_id = cpu_id;
5138 	if (thread_id == 0) {
5139 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
5140 		if (cpu_is_first_core_in_package(cpu_id))
5141 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
5142 	}
5143 
5144 	c->core_id = core_id;
5145 	p->package_id = pkg_id;
5146 }
5147 
5148 
5149 int initialize_counters(int cpu_id)
5150 {
5151 	init_counter(EVEN_COUNTERS, cpu_id);
5152 	init_counter(ODD_COUNTERS, cpu_id);
5153 	return 0;
5154 }
5155 
5156 void allocate_output_buffer()
5157 {
5158 	output_buffer = calloc(1, (1 + topo.num_cpus) * 2048);
5159 	outp = output_buffer;
5160 	if (outp == NULL)
5161 		err(-1, "calloc output buffer");
5162 }
5163 void allocate_fd_percpu(void)
5164 {
5165 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5166 	if (fd_percpu == NULL)
5167 		err(-1, "calloc fd_percpu");
5168 }
5169 void allocate_irq_buffers(void)
5170 {
5171 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
5172 	if (irq_column_2_cpu == NULL)
5173 		err(-1, "calloc %d", topo.num_cpus);
5174 
5175 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5176 	if (irqs_per_cpu == NULL)
5177 		err(-1, "calloc %d", topo.max_cpu_num + 1);
5178 }
5179 void setup_all_buffers(void)
5180 {
5181 	topology_probe();
5182 	allocate_irq_buffers();
5183 	allocate_fd_percpu();
5184 	allocate_counters(&thread_even, &core_even, &package_even);
5185 	allocate_counters(&thread_odd, &core_odd, &package_odd);
5186 	allocate_output_buffer();
5187 	for_all_proc_cpus(initialize_counters);
5188 }
5189 
5190 void set_base_cpu(void)
5191 {
5192 	base_cpu = sched_getcpu();
5193 	if (base_cpu < 0)
5194 		err(-ENODEV, "No valid cpus found");
5195 
5196 	if (debug > 1)
5197 		fprintf(outf, "base_cpu = %d\n", base_cpu);
5198 }
5199 
5200 void turbostat_init()
5201 {
5202 	setup_all_buffers();
5203 	set_base_cpu();
5204 	check_dev_msr();
5205 	check_permissions();
5206 	process_cpuid();
5207 
5208 
5209 	if (!quiet)
5210 		for_all_cpus(print_hwp, ODD_COUNTERS);
5211 
5212 	if (!quiet)
5213 		for_all_cpus(print_epb, ODD_COUNTERS);
5214 
5215 	if (!quiet)
5216 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
5217 
5218 	if (!quiet)
5219 		for_all_cpus(print_rapl, ODD_COUNTERS);
5220 
5221 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
5222 
5223 	if (!quiet)
5224 		for_all_cpus(print_thermal, ODD_COUNTERS);
5225 
5226 	if (!quiet && do_irtl_snb)
5227 		print_irtl();
5228 }
5229 
5230 int fork_it(char **argv)
5231 {
5232 	pid_t child_pid;
5233 	int status;
5234 
5235 	snapshot_proc_sysfs_files();
5236 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
5237 	first_counter_read = 0;
5238 	if (status)
5239 		exit(status);
5240 	/* clear affinity side-effect of get_counters() */
5241 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
5242 	gettimeofday(&tv_even, (struct timezone *)NULL);
5243 
5244 	child_pid = fork();
5245 	if (!child_pid) {
5246 		/* child */
5247 		execvp(argv[0], argv);
5248 		err(errno, "exec %s", argv[0]);
5249 	} else {
5250 
5251 		/* parent */
5252 		if (child_pid == -1)
5253 			err(1, "fork");
5254 
5255 		signal(SIGINT, SIG_IGN);
5256 		signal(SIGQUIT, SIG_IGN);
5257 		if (waitpid(child_pid, &status, 0) == -1)
5258 			err(status, "waitpid");
5259 
5260 		if (WIFEXITED(status))
5261 			status = WEXITSTATUS(status);
5262 	}
5263 	/*
5264 	 * n.b. fork_it() does not check for errors from for_all_cpus()
5265 	 * because re-starting is problematic when forking
5266 	 */
5267 	snapshot_proc_sysfs_files();
5268 	for_all_cpus(get_counters, ODD_COUNTERS);
5269 	gettimeofday(&tv_odd, (struct timezone *)NULL);
5270 	timersub(&tv_odd, &tv_even, &tv_delta);
5271 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
5272 		fprintf(outf, "%s: Counter reset detected\n", progname);
5273 	else {
5274 		compute_average(EVEN_COUNTERS);
5275 		format_all_counters(EVEN_COUNTERS);
5276 	}
5277 
5278 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
5279 
5280 	flush_output_stderr();
5281 
5282 	return status;
5283 }
5284 
5285 int get_and_dump_counters(void)
5286 {
5287 	int status;
5288 
5289 	snapshot_proc_sysfs_files();
5290 	status = for_all_cpus(get_counters, ODD_COUNTERS);
5291 	if (status)
5292 		return status;
5293 
5294 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
5295 	if (status)
5296 		return status;
5297 
5298 	flush_output_stdout();
5299 
5300 	return status;
5301 }
5302 
5303 void print_version() {
5304 	fprintf(outf, "turbostat version 19.03.20"
5305 		" - Len Brown <lenb@kernel.org>\n");
5306 }
5307 
5308 int add_counter(unsigned int msr_num, char *path, char *name,
5309 	unsigned int width, enum counter_scope scope,
5310 	enum counter_type type, enum counter_format format, int flags)
5311 {
5312 	struct msr_counter *msrp;
5313 
5314 	msrp = calloc(1, sizeof(struct msr_counter));
5315 	if (msrp == NULL) {
5316 		perror("calloc");
5317 		exit(1);
5318 	}
5319 
5320 	msrp->msr_num = msr_num;
5321 	strncpy(msrp->name, name, NAME_BYTES);
5322 	if (path)
5323 		strncpy(msrp->path, path, PATH_BYTES);
5324 	msrp->width = width;
5325 	msrp->type = type;
5326 	msrp->format = format;
5327 	msrp->flags = flags;
5328 
5329 	switch (scope) {
5330 
5331 	case SCOPE_CPU:
5332 		msrp->next = sys.tp;
5333 		sys.tp = msrp;
5334 		sys.added_thread_counters++;
5335 		if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) {
5336 			fprintf(stderr, "exceeded max %d added thread counters\n",
5337 				MAX_ADDED_COUNTERS);
5338 			exit(-1);
5339 		}
5340 		break;
5341 
5342 	case SCOPE_CORE:
5343 		msrp->next = sys.cp;
5344 		sys.cp = msrp;
5345 		sys.added_core_counters++;
5346 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
5347 			fprintf(stderr, "exceeded max %d added core counters\n",
5348 				MAX_ADDED_COUNTERS);
5349 			exit(-1);
5350 		}
5351 		break;
5352 
5353 	case SCOPE_PACKAGE:
5354 		msrp->next = sys.pp;
5355 		sys.pp = msrp;
5356 		sys.added_package_counters++;
5357 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
5358 			fprintf(stderr, "exceeded max %d added package counters\n",
5359 				MAX_ADDED_COUNTERS);
5360 			exit(-1);
5361 		}
5362 		break;
5363 	}
5364 
5365 	return 0;
5366 }
5367 
5368 void parse_add_command(char *add_command)
5369 {
5370 	int msr_num = 0;
5371 	char *path = NULL;
5372 	char name_buffer[NAME_BYTES] = "";
5373 	int width = 64;
5374 	int fail = 0;
5375 	enum counter_scope scope = SCOPE_CPU;
5376 	enum counter_type type = COUNTER_CYCLES;
5377 	enum counter_format format = FORMAT_DELTA;
5378 
5379 	while (add_command) {
5380 
5381 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
5382 			goto next;
5383 
5384 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
5385 			goto next;
5386 
5387 		if (*add_command == '/') {
5388 			path = add_command;
5389 			goto next;
5390 		}
5391 
5392 		if (sscanf(add_command, "u%d", &width) == 1) {
5393 			if ((width == 32) || (width == 64))
5394 				goto next;
5395 			width = 64;
5396 		}
5397 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
5398 			scope = SCOPE_CPU;
5399 			goto next;
5400 		}
5401 		if (!strncmp(add_command, "core", strlen("core"))) {
5402 			scope = SCOPE_CORE;
5403 			goto next;
5404 		}
5405 		if (!strncmp(add_command, "package", strlen("package"))) {
5406 			scope = SCOPE_PACKAGE;
5407 			goto next;
5408 		}
5409 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
5410 			type = COUNTER_CYCLES;
5411 			goto next;
5412 		}
5413 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
5414 			type = COUNTER_SECONDS;
5415 			goto next;
5416 		}
5417 		if (!strncmp(add_command, "usec", strlen("usec"))) {
5418 			type = COUNTER_USEC;
5419 			goto next;
5420 		}
5421 		if (!strncmp(add_command, "raw", strlen("raw"))) {
5422 			format = FORMAT_RAW;
5423 			goto next;
5424 		}
5425 		if (!strncmp(add_command, "delta", strlen("delta"))) {
5426 			format = FORMAT_DELTA;
5427 			goto next;
5428 		}
5429 		if (!strncmp(add_command, "percent", strlen("percent"))) {
5430 			format = FORMAT_PERCENT;
5431 			goto next;
5432 		}
5433 
5434 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
5435 			char *eos;
5436 
5437 			eos = strchr(name_buffer, ',');
5438 			if (eos)
5439 				*eos = '\0';
5440 			goto next;
5441 		}
5442 
5443 next:
5444 		add_command = strchr(add_command, ',');
5445 		if (add_command) {
5446 			*add_command = '\0';
5447 			add_command++;
5448 		}
5449 
5450 	}
5451 	if ((msr_num == 0) && (path == NULL)) {
5452 		fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
5453 		fail++;
5454 	}
5455 
5456 	/* generate default column header */
5457 	if (*name_buffer == '\0') {
5458 		if (width == 32)
5459 			sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5460 		else
5461 			sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5462 	}
5463 
5464 	if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
5465 		fail++;
5466 
5467 	if (fail) {
5468 		help();
5469 		exit(1);
5470 	}
5471 }
5472 
5473 int is_deferred_skip(char *name)
5474 {
5475 	int i;
5476 
5477 	for (i = 0; i < deferred_skip_index; ++i)
5478 		if (!strcmp(name, deferred_skip_names[i]))
5479 			return 1;
5480 	return 0;
5481 }
5482 
5483 void probe_sysfs(void)
5484 {
5485 	char path[64];
5486 	char name_buf[16];
5487 	FILE *input;
5488 	int state;
5489 	char *sp;
5490 
5491 	if (!DO_BIC(BIC_sysfs))
5492 		return;
5493 
5494 	for (state = 10; state >= 0; --state) {
5495 
5496 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5497 			base_cpu, state);
5498 		input = fopen(path, "r");
5499 		if (input == NULL)
5500 			continue;
5501 		if (!fgets(name_buf, sizeof(name_buf), input))
5502 			err(1, "%s: failed to read file", path);
5503 
5504 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5505 		sp = strchr(name_buf, '-');
5506 		if (!sp)
5507 			sp = strchrnul(name_buf, '\n');
5508 		*sp = '%';
5509 		*(sp + 1) = '\0';
5510 
5511 		fclose(input);
5512 
5513 		sprintf(path, "cpuidle/state%d/time", state);
5514 
5515 		if (is_deferred_skip(name_buf))
5516 			continue;
5517 
5518 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC,
5519 				FORMAT_PERCENT, SYSFS_PERCPU);
5520 	}
5521 
5522 	for (state = 10; state >= 0; --state) {
5523 
5524 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5525 			base_cpu, state);
5526 		input = fopen(path, "r");
5527 		if (input == NULL)
5528 			continue;
5529 		if (!fgets(name_buf, sizeof(name_buf), input))
5530 			err(1, "%s: failed to read file", path);
5531 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5532 		sp = strchr(name_buf, '-');
5533 		if (!sp)
5534 			sp = strchrnul(name_buf, '\n');
5535 		*sp = '\0';
5536 		fclose(input);
5537 
5538 		sprintf(path, "cpuidle/state%d/usage", state);
5539 
5540 		if (is_deferred_skip(name_buf))
5541 			continue;
5542 
5543 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS,
5544 				FORMAT_DELTA, SYSFS_PERCPU);
5545 	}
5546 
5547 }
5548 
5549 
5550 /*
5551  * parse cpuset with following syntax
5552  * 1,2,4..6,8-10 and set bits in cpu_subset
5553  */
5554 void parse_cpu_command(char *optarg)
5555 {
5556 	unsigned int start, end;
5557 	char *next;
5558 
5559 	if (!strcmp(optarg, "core")) {
5560 		if (cpu_subset)
5561 			goto error;
5562 		show_core_only++;
5563 		return;
5564 	}
5565 	if (!strcmp(optarg, "package")) {
5566 		if (cpu_subset)
5567 			goto error;
5568 		show_pkg_only++;
5569 		return;
5570 	}
5571 	if (show_core_only || show_pkg_only)
5572 		goto error;
5573 
5574 	cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
5575 	if (cpu_subset == NULL)
5576 		err(3, "CPU_ALLOC");
5577 	cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
5578 
5579 	CPU_ZERO_S(cpu_subset_size, cpu_subset);
5580 
5581 	next = optarg;
5582 
5583 	while (next && *next) {
5584 
5585 		if (*next == '-')	/* no negative cpu numbers */
5586 			goto error;
5587 
5588 		start = strtoul(next, &next, 10);
5589 
5590 		if (start >= CPU_SUBSET_MAXCPUS)
5591 			goto error;
5592 		CPU_SET_S(start, cpu_subset_size, cpu_subset);
5593 
5594 		if (*next == '\0')
5595 			break;
5596 
5597 		if (*next == ',') {
5598 			next += 1;
5599 			continue;
5600 		}
5601 
5602 		if (*next == '-') {
5603 			next += 1;	/* start range */
5604 		} else if (*next == '.') {
5605 			next += 1;
5606 			if (*next == '.')
5607 				next += 1;	/* start range */
5608 			else
5609 				goto error;
5610 		}
5611 
5612 		end = strtoul(next, &next, 10);
5613 		if (end <= start)
5614 			goto error;
5615 
5616 		while (++start <= end) {
5617 			if (start >= CPU_SUBSET_MAXCPUS)
5618 				goto error;
5619 			CPU_SET_S(start, cpu_subset_size, cpu_subset);
5620 		}
5621 
5622 		if (*next == ',')
5623 			next += 1;
5624 		else if (*next != '\0')
5625 			goto error;
5626 	}
5627 
5628 	return;
5629 
5630 error:
5631 	fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
5632 	help();
5633 	exit(-1);
5634 }
5635 
5636 
5637 void cmdline(int argc, char **argv)
5638 {
5639 	int opt;
5640 	int option_index = 0;
5641 	static struct option long_options[] = {
5642 		{"add",		required_argument,	0, 'a'},
5643 		{"cpu",		required_argument,	0, 'c'},
5644 		{"Dump",	no_argument,		0, 'D'},
5645 		{"debug",	no_argument,		0, 'd'},	/* internal, not documented */
5646 		{"enable",	required_argument,	0, 'e'},
5647 		{"interval",	required_argument,	0, 'i'},
5648 		{"num_iterations",	required_argument,	0, 'n'},
5649 		{"help",	no_argument,		0, 'h'},
5650 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
5651 		{"Joules",	no_argument,		0, 'J'},
5652 		{"list",	no_argument,		0, 'l'},
5653 		{"out",		required_argument,	0, 'o'},
5654 		{"quiet",	no_argument,		0, 'q'},
5655 		{"show",	required_argument,	0, 's'},
5656 		{"Summary",	no_argument,		0, 'S'},
5657 		{"TCC",		required_argument,	0, 'T'},
5658 		{"version",	no_argument,		0, 'v' },
5659 		{0,		0,			0,  0 }
5660 	};
5661 
5662 	progname = argv[0];
5663 
5664 	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v",
5665 				long_options, &option_index)) != -1) {
5666 		switch (opt) {
5667 		case 'a':
5668 			parse_add_command(optarg);
5669 			break;
5670 		case 'c':
5671 			parse_cpu_command(optarg);
5672 			break;
5673 		case 'D':
5674 			dump_only++;
5675 			break;
5676 		case 'e':
5677 			/* --enable specified counter */
5678 			bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST);
5679 			break;
5680 		case 'd':
5681 			debug++;
5682 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5683 			break;
5684 		case 'H':
5685 			/*
5686 			 * --hide: do not show those specified
5687 			 *  multiple invocations simply clear more bits in enabled mask
5688 			 */
5689 			bic_enabled &= ~bic_lookup(optarg, HIDE_LIST);
5690 			break;
5691 		case 'h':
5692 		default:
5693 			help();
5694 			exit(1);
5695 		case 'i':
5696 			{
5697 				double interval = strtod(optarg, NULL);
5698 
5699 				if (interval < 0.001) {
5700 					fprintf(outf, "interval %f seconds is too small\n",
5701 						interval);
5702 					exit(2);
5703 				}
5704 
5705 				interval_tv.tv_sec = interval_ts.tv_sec = interval;
5706 				interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000;
5707 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
5708 			}
5709 			break;
5710 		case 'J':
5711 			rapl_joules++;
5712 			break;
5713 		case 'l':
5714 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5715 			list_header_only++;
5716 			quiet++;
5717 			break;
5718 		case 'o':
5719 			outf = fopen_or_die(optarg, "w");
5720 			break;
5721 		case 'q':
5722 			quiet = 1;
5723 			break;
5724 		case 'n':
5725 			num_iterations = strtod(optarg, NULL);
5726 
5727 			if (num_iterations <= 0) {
5728 				fprintf(outf, "iterations %d should be positive number\n",
5729 					num_iterations);
5730 				exit(2);
5731 			}
5732 			break;
5733 		case 's':
5734 			/*
5735 			 * --show: show only those specified
5736 			 *  The 1st invocation will clear and replace the enabled mask
5737 			 *  subsequent invocations can add to it.
5738 			 */
5739 			if (shown == 0)
5740 				bic_enabled = bic_lookup(optarg, SHOW_LIST);
5741 			else
5742 				bic_enabled |= bic_lookup(optarg, SHOW_LIST);
5743 			shown = 1;
5744 			break;
5745 		case 'S':
5746 			summary_only++;
5747 			break;
5748 		case 'T':
5749 			tcc_activation_temp_override = atoi(optarg);
5750 			break;
5751 		case 'v':
5752 			print_version();
5753 			exit(0);
5754 			break;
5755 		}
5756 	}
5757 }
5758 
5759 int main(int argc, char **argv)
5760 {
5761 	outf = stderr;
5762 	cmdline(argc, argv);
5763 
5764 	if (!quiet)
5765 		print_version();
5766 
5767 	probe_sysfs();
5768 
5769 	turbostat_init();
5770 
5771 	/* dump counters and exit */
5772 	if (dump_only)
5773 		return get_and_dump_counters();
5774 
5775 	/* list header and exit */
5776 	if (list_header_only) {
5777 		print_header(",");
5778 		flush_output_stdout();
5779 		return 0;
5780 	}
5781 
5782 	/*
5783 	 * if any params left, it must be a command to fork
5784 	 */
5785 	if (argc - optind)
5786 		return fork_it(argv + optind);
5787 	else
5788 		turbostat_loop();
5789 
5790 	return 0;
5791 }
5792