xref: /linux/tools/power/x86/turbostat/turbostat.c (revision 44f57d78)
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 struct timespec one_msec = {0, 1000000};
43 unsigned int num_iterations;
44 unsigned int debug;
45 unsigned int quiet;
46 unsigned int shown;
47 unsigned int sums_need_wide_columns;
48 unsigned int rapl_joules;
49 unsigned int summary_only;
50 unsigned int list_header_only;
51 unsigned int dump_only;
52 unsigned int do_snb_cstates;
53 unsigned int do_knl_cstates;
54 unsigned int do_slm_cstates;
55 unsigned int use_c1_residency_msr;
56 unsigned int has_aperf;
57 unsigned int has_epb;
58 unsigned int do_irtl_snb;
59 unsigned int do_irtl_hsw;
60 unsigned int units = 1000000;	/* MHz etc */
61 unsigned int genuine_intel;
62 unsigned int authentic_amd;
63 unsigned int max_level, max_extended_level;
64 unsigned int has_invariant_tsc;
65 unsigned int do_nhm_platform_info;
66 unsigned int no_MSR_MISC_PWR_MGMT;
67 unsigned int aperf_mperf_multiplier = 1;
68 double bclk;
69 double base_hz;
70 unsigned int has_base_hz;
71 double tsc_tweak = 1.0;
72 unsigned int show_pkg_only;
73 unsigned int show_core_only;
74 char *output_buffer, *outp;
75 unsigned int do_rapl;
76 unsigned int do_dts;
77 unsigned int do_ptm;
78 unsigned long long  gfx_cur_rc6_ms;
79 unsigned long long cpuidle_cur_cpu_lpi_us;
80 unsigned long long cpuidle_cur_sys_lpi_us;
81 unsigned int gfx_cur_mhz;
82 unsigned int tcc_activation_temp;
83 unsigned int tcc_activation_temp_override;
84 double rapl_power_units, rapl_time_units;
85 double rapl_dram_energy_units, rapl_energy_units;
86 double rapl_joule_counter_range;
87 unsigned int do_core_perf_limit_reasons;
88 unsigned int has_automatic_cstate_conversion;
89 unsigned int do_gfx_perf_limit_reasons;
90 unsigned int do_ring_perf_limit_reasons;
91 unsigned int crystal_hz;
92 unsigned long long tsc_hz;
93 int base_cpu;
94 double discover_bclk(unsigned int family, unsigned int model);
95 unsigned int has_hwp;	/* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
96 			/* IA32_HWP_REQUEST, IA32_HWP_STATUS */
97 unsigned int has_hwp_notify;		/* IA32_HWP_INTERRUPT */
98 unsigned int has_hwp_activity_window;	/* IA32_HWP_REQUEST[bits 41:32] */
99 unsigned int has_hwp_epp;		/* IA32_HWP_REQUEST[bits 31:24] */
100 unsigned int has_hwp_pkg;		/* IA32_HWP_REQUEST_PKG */
101 unsigned int has_misc_feature_control;
102 unsigned int first_counter_read = 1;
103 
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 	unsigned long long tsc;
170 	unsigned long long aperf;
171 	unsigned long long mperf;
172 	unsigned long long c1;
173 	unsigned long long  irq_count;
174 	unsigned int smi_count;
175 	unsigned int cpu_id;
176 	unsigned int apic_id;
177 	unsigned int x2apic_id;
178 	unsigned int flags;
179 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
180 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
181 	unsigned long long counter[MAX_ADDED_THREAD_COUNTERS];
182 } *thread_even, *thread_odd;
183 
184 struct core_data {
185 	unsigned long long c3;
186 	unsigned long long c6;
187 	unsigned long long c7;
188 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
189 	unsigned int core_temp_c;
190 	unsigned int core_energy;	/* MSR_CORE_ENERGY_STAT */
191 	unsigned int core_id;
192 	unsigned long long counter[MAX_ADDED_COUNTERS];
193 } *core_even, *core_odd;
194 
195 struct pkg_data {
196 	unsigned long long pc2;
197 	unsigned long long pc3;
198 	unsigned long long pc6;
199 	unsigned long long pc7;
200 	unsigned long long pc8;
201 	unsigned long long pc9;
202 	unsigned long long pc10;
203 	unsigned long long cpu_lpi;
204 	unsigned long long sys_lpi;
205 	unsigned long long pkg_wtd_core_c0;
206 	unsigned long long pkg_any_core_c0;
207 	unsigned long long pkg_any_gfxe_c0;
208 	unsigned long long pkg_both_core_gfxe_c0;
209 	long long gfx_rc6_ms;
210 	unsigned int gfx_mhz;
211 	unsigned int package_id;
212 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
213 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
214 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
215 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
216 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
217 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
218 	unsigned int pkg_temp_c;
219 	unsigned long long counter[MAX_ADDED_COUNTERS];
220 } *package_even, *package_odd;
221 
222 #define ODD_COUNTERS thread_odd, core_odd, package_odd
223 #define EVEN_COUNTERS thread_even, core_even, package_even
224 
225 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no)	      \
226 	((thread_base) +						      \
227 	 ((pkg_no) *							      \
228 	  topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \
229 	 ((node_no) * topo.cores_per_node * topo.threads_per_core) +	      \
230 	 ((core_no) * topo.threads_per_core) +				      \
231 	 (thread_no))
232 
233 #define GET_CORE(core_base, core_no, node_no, pkg_no)			\
234 	((core_base) +							\
235 	 ((pkg_no) *  topo.nodes_per_pkg * topo.cores_per_node) +	\
236 	 ((node_no) * topo.cores_per_node) +				\
237 	 (core_no))
238 
239 
240 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
241 
242 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
243 enum counter_type {COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC};
244 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
245 
246 struct msr_counter {
247 	unsigned int msr_num;
248 	char name[NAME_BYTES];
249 	char path[PATH_BYTES];
250 	unsigned int width;
251 	enum counter_type type;
252 	enum counter_format format;
253 	struct msr_counter *next;
254 	unsigned int flags;
255 #define	FLAGS_HIDE	(1 << 0)
256 #define	FLAGS_SHOW	(1 << 1)
257 #define	SYSFS_PERCPU	(1 << 1)
258 };
259 
260 struct sys_counters {
261 	unsigned int added_thread_counters;
262 	unsigned int added_core_counters;
263 	unsigned int added_package_counters;
264 	struct msr_counter *tp;
265 	struct msr_counter *cp;
266 	struct msr_counter *pp;
267 } sys;
268 
269 struct system_summary {
270 	struct thread_data threads;
271 	struct core_data cores;
272 	struct pkg_data packages;
273 } average;
274 
275 struct cpu_topology {
276 	int physical_package_id;
277 	int die_id;
278 	int logical_cpu_id;
279 	int physical_node_id;
280 	int logical_node_id;	/* 0-based count within the package */
281 	int physical_core_id;
282 	int thread_id;
283 	cpu_set_t *put_ids; /* Processing Unit/Thread IDs */
284 } *cpus;
285 
286 struct topo_params {
287 	int num_packages;
288 	int num_die;
289 	int num_cpus;
290 	int num_cores;
291 	int max_cpu_num;
292 	int max_node_num;
293 	int nodes_per_pkg;
294 	int cores_per_node;
295 	int threads_per_core;
296 } topo;
297 
298 struct timeval tv_even, tv_odd, tv_delta;
299 
300 int *irq_column_2_cpu;	/* /proc/interrupts column numbers */
301 int *irqs_per_cpu;		/* indexed by cpu_num */
302 
303 void setup_all_buffers(void);
304 
305 int cpu_is_not_present(int cpu)
306 {
307 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
308 }
309 /*
310  * run func(thread, core, package) in topology order
311  * skip non-present cpus
312  */
313 
314 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
315 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
316 {
317 	int retval, pkg_no, core_no, thread_no, node_no;
318 
319 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
320 		for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) {
321 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
322 				for (thread_no = 0; thread_no <
323 					topo.threads_per_core; ++thread_no) {
324 					struct thread_data *t;
325 					struct core_data *c;
326 					struct pkg_data *p;
327 
328 					t = GET_THREAD(thread_base, thread_no,
329 						       core_no, node_no,
330 						       pkg_no);
331 
332 					if (cpu_is_not_present(t->cpu_id))
333 						continue;
334 
335 					c = GET_CORE(core_base, core_no,
336 						     node_no, pkg_no);
337 					p = GET_PKG(pkg_base, pkg_no);
338 
339 					retval = func(t, c, p);
340 					if (retval)
341 						return retval;
342 				}
343 			}
344 		}
345 	}
346 	return 0;
347 }
348 
349 int cpu_migrate(int cpu)
350 {
351 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
352 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
353 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
354 		return -1;
355 	else
356 		return 0;
357 }
358 int get_msr_fd(int cpu)
359 {
360 	char pathname[32];
361 	int fd;
362 
363 	fd = fd_percpu[cpu];
364 
365 	if (fd)
366 		return fd;
367 
368 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
369 	fd = open(pathname, O_RDONLY);
370 	if (fd < 0)
371 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
372 
373 	fd_percpu[cpu] = fd;
374 
375 	return fd;
376 }
377 
378 int get_msr(int cpu, off_t offset, unsigned long long *msr)
379 {
380 	ssize_t retval;
381 
382 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
383 
384 	if (retval != sizeof *msr)
385 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
386 
387 	return 0;
388 }
389 
390 /*
391  * This list matches the column headers, except
392  * 1. built-in only, the sysfs counters are not here -- we learn of those at run-time
393  * 2. Core and CPU are moved to the end, we can't have strings that contain them
394  *    matching on them for --show and --hide.
395  */
396 struct msr_counter bic[] = {
397 	{ 0x0, "usec" },
398 	{ 0x0, "Time_Of_Day_Seconds" },
399 	{ 0x0, "Package" },
400 	{ 0x0, "Node" },
401 	{ 0x0, "Avg_MHz" },
402 	{ 0x0, "Busy%" },
403 	{ 0x0, "Bzy_MHz" },
404 	{ 0x0, "TSC_MHz" },
405 	{ 0x0, "IRQ" },
406 	{ 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL},
407 	{ 0x0, "sysfs" },
408 	{ 0x0, "CPU%c1" },
409 	{ 0x0, "CPU%c3" },
410 	{ 0x0, "CPU%c6" },
411 	{ 0x0, "CPU%c7" },
412 	{ 0x0, "ThreadC" },
413 	{ 0x0, "CoreTmp" },
414 	{ 0x0, "CoreCnt" },
415 	{ 0x0, "PkgTmp" },
416 	{ 0x0, "GFX%rc6" },
417 	{ 0x0, "GFXMHz" },
418 	{ 0x0, "Pkg%pc2" },
419 	{ 0x0, "Pkg%pc3" },
420 	{ 0x0, "Pkg%pc6" },
421 	{ 0x0, "Pkg%pc7" },
422 	{ 0x0, "Pkg%pc8" },
423 	{ 0x0, "Pkg%pc9" },
424 	{ 0x0, "Pk%pc10" },
425 	{ 0x0, "CPU%LPI" },
426 	{ 0x0, "SYS%LPI" },
427 	{ 0x0, "PkgWatt" },
428 	{ 0x0, "CorWatt" },
429 	{ 0x0, "GFXWatt" },
430 	{ 0x0, "PkgCnt" },
431 	{ 0x0, "RAMWatt" },
432 	{ 0x0, "PKG_%" },
433 	{ 0x0, "RAM_%" },
434 	{ 0x0, "Pkg_J" },
435 	{ 0x0, "Cor_J" },
436 	{ 0x0, "GFX_J" },
437 	{ 0x0, "RAM_J" },
438 	{ 0x0, "Mod%c6" },
439 	{ 0x0, "Totl%C0" },
440 	{ 0x0, "Any%C0" },
441 	{ 0x0, "GFX%C0" },
442 	{ 0x0, "CPUGFX%" },
443 	{ 0x0, "Core" },
444 	{ 0x0, "CPU" },
445 	{ 0x0, "APIC" },
446 	{ 0x0, "X2APIC" },
447 	{ 0x0, "Die" },
448 };
449 
450 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
451 #define	BIC_USEC	(1ULL << 0)
452 #define	BIC_TOD		(1ULL << 1)
453 #define	BIC_Package	(1ULL << 2)
454 #define	BIC_Node	(1ULL << 3)
455 #define	BIC_Avg_MHz	(1ULL << 4)
456 #define	BIC_Busy	(1ULL << 5)
457 #define	BIC_Bzy_MHz	(1ULL << 6)
458 #define	BIC_TSC_MHz	(1ULL << 7)
459 #define	BIC_IRQ		(1ULL << 8)
460 #define	BIC_SMI		(1ULL << 9)
461 #define	BIC_sysfs	(1ULL << 10)
462 #define	BIC_CPU_c1	(1ULL << 11)
463 #define	BIC_CPU_c3	(1ULL << 12)
464 #define	BIC_CPU_c6	(1ULL << 13)
465 #define	BIC_CPU_c7	(1ULL << 14)
466 #define	BIC_ThreadC	(1ULL << 15)
467 #define	BIC_CoreTmp	(1ULL << 16)
468 #define	BIC_CoreCnt	(1ULL << 17)
469 #define	BIC_PkgTmp	(1ULL << 18)
470 #define	BIC_GFX_rc6	(1ULL << 19)
471 #define	BIC_GFXMHz	(1ULL << 20)
472 #define	BIC_Pkgpc2	(1ULL << 21)
473 #define	BIC_Pkgpc3	(1ULL << 22)
474 #define	BIC_Pkgpc6	(1ULL << 23)
475 #define	BIC_Pkgpc7	(1ULL << 24)
476 #define	BIC_Pkgpc8	(1ULL << 25)
477 #define	BIC_Pkgpc9	(1ULL << 26)
478 #define	BIC_Pkgpc10	(1ULL << 27)
479 #define BIC_CPU_LPI	(1ULL << 28)
480 #define BIC_SYS_LPI	(1ULL << 29)
481 #define	BIC_PkgWatt	(1ULL << 30)
482 #define	BIC_CorWatt	(1ULL << 31)
483 #define	BIC_GFXWatt	(1ULL << 32)
484 #define	BIC_PkgCnt	(1ULL << 33)
485 #define	BIC_RAMWatt	(1ULL << 34)
486 #define	BIC_PKG__	(1ULL << 35)
487 #define	BIC_RAM__	(1ULL << 36)
488 #define	BIC_Pkg_J	(1ULL << 37)
489 #define	BIC_Cor_J	(1ULL << 38)
490 #define	BIC_GFX_J	(1ULL << 39)
491 #define	BIC_RAM_J	(1ULL << 40)
492 #define	BIC_Mod_c6	(1ULL << 41)
493 #define	BIC_Totl_c0	(1ULL << 42)
494 #define	BIC_Any_c0	(1ULL << 43)
495 #define	BIC_GFX_c0	(1ULL << 44)
496 #define	BIC_CPUGFX	(1ULL << 45)
497 #define	BIC_Core	(1ULL << 46)
498 #define	BIC_CPU		(1ULL << 47)
499 #define	BIC_APIC	(1ULL << 48)
500 #define	BIC_X2APIC	(1ULL << 49)
501 #define	BIC_Die		(1ULL << 50)
502 
503 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
504 
505 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT);
506 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC;
507 
508 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
509 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
510 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
511 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
512 
513 
514 #define MAX_DEFERRED 16
515 char *deferred_skip_names[MAX_DEFERRED];
516 int deferred_skip_index;
517 
518 /*
519  * HIDE_LIST - hide this list of counters, show the rest [default]
520  * SHOW_LIST - show this list of counters, hide the rest
521  */
522 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
523 
524 void help(void)
525 {
526 	fprintf(outf,
527 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
528 	"\n"
529 	"Turbostat forks the specified COMMAND and prints statistics\n"
530 	"when COMMAND completes.\n"
531 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
532 	"to print statistics, until interrupted.\n"
533 	"  -a, --add	add a counter\n"
534 	"		  eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
535 	"  -c, --cpu	cpu-set	limit output to summary plus cpu-set:\n"
536 	"		  {core | package | j,k,l..m,n-p }\n"
537 	"  -d, --debug	displays usec, Time_Of_Day_Seconds and more debugging\n"
538 	"  -D, --Dump	displays the raw counter values\n"
539 	"  -e, --enable	[all | column]\n"
540 	"		shows all or the specified disabled column\n"
541 	"  -H, --hide [column|column,column,...]\n"
542 	"		hide the specified column(s)\n"
543 	"  -i, --interval sec.subsec\n"
544 	"		Override default 5-second measurement interval\n"
545 	"  -J, --Joules	displays energy in Joules instead of Watts\n"
546 	"  -l, --list	list column headers only\n"
547 	"  -n, --num_iterations num\n"
548 	"		number of the measurement iterations\n"
549 	"  -o, --out file\n"
550 	"		create or truncate \"file\" for all output\n"
551 	"  -q, --quiet	skip decoding system configuration header\n"
552 	"  -s, --show [column|column,column,...]\n"
553 	"		show only the specified column(s)\n"
554 	"  -S, --Summary\n"
555 	"		limits output to 1-line system summary per interval\n"
556 	"  -T, --TCC temperature\n"
557 	"		sets the Thermal Control Circuit temperature in\n"
558 	"		  degrees Celsius\n"
559 	"  -h, --help	print this help message\n"
560 	"  -v, --version	print version information\n"
561 	"\n"
562 	"For more help, run \"man turbostat\"\n");
563 }
564 
565 /*
566  * bic_lookup
567  * for all the strings in comma separate name_list,
568  * set the approprate bit in return value.
569  */
570 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
571 {
572 	int i;
573 	unsigned long long retval = 0;
574 
575 	while (name_list) {
576 		char *comma;
577 
578 		comma = strchr(name_list, ',');
579 
580 		if (comma)
581 			*comma = '\0';
582 
583 		if (!strcmp(name_list, "all"))
584 			return ~0;
585 
586 		for (i = 0; i < MAX_BIC; ++i) {
587 			if (!strcmp(name_list, bic[i].name)) {
588 				retval |= (1ULL << i);
589 				break;
590 			}
591 		}
592 		if (i == MAX_BIC) {
593 			if (mode == SHOW_LIST) {
594 				fprintf(stderr, "Invalid counter name: %s\n", name_list);
595 				exit(-1);
596 			}
597 			deferred_skip_names[deferred_skip_index++] = name_list;
598 			if (debug)
599 				fprintf(stderr, "deferred \"%s\"\n", name_list);
600 			if (deferred_skip_index >= MAX_DEFERRED) {
601 				fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
602 					MAX_DEFERRED, name_list);
603 				help();
604 				exit(1);
605 			}
606 		}
607 
608 		name_list = comma;
609 		if (name_list)
610 			name_list++;
611 
612 	}
613 	return retval;
614 }
615 
616 
617 void print_header(char *delim)
618 {
619 	struct msr_counter *mp;
620 	int printed = 0;
621 
622 	if (DO_BIC(BIC_USEC))
623 		outp += sprintf(outp, "%susec", (printed++ ? delim : ""));
624 	if (DO_BIC(BIC_TOD))
625 		outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : ""));
626 	if (DO_BIC(BIC_Package))
627 		outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
628 	if (DO_BIC(BIC_Die))
629 		outp += sprintf(outp, "%sDie", (printed++ ? delim : ""));
630 	if (DO_BIC(BIC_Node))
631 		outp += sprintf(outp, "%sNode", (printed++ ? delim : ""));
632 	if (DO_BIC(BIC_Core))
633 		outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
634 	if (DO_BIC(BIC_CPU))
635 		outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
636 	if (DO_BIC(BIC_APIC))
637 		outp += sprintf(outp, "%sAPIC", (printed++ ? delim : ""));
638 	if (DO_BIC(BIC_X2APIC))
639 		outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : ""));
640 	if (DO_BIC(BIC_Avg_MHz))
641 		outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
642 	if (DO_BIC(BIC_Busy))
643 		outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
644 	if (DO_BIC(BIC_Bzy_MHz))
645 		outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
646 	if (DO_BIC(BIC_TSC_MHz))
647 		outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
648 
649 	if (DO_BIC(BIC_IRQ)) {
650 		if (sums_need_wide_columns)
651 			outp += sprintf(outp, "%s     IRQ", (printed++ ? delim : ""));
652 		else
653 			outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
654 	}
655 
656 	if (DO_BIC(BIC_SMI))
657 		outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
658 
659 	for (mp = sys.tp; mp; mp = mp->next) {
660 
661 		if (mp->format == FORMAT_RAW) {
662 			if (mp->width == 64)
663 				outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
664 			else
665 				outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
666 		} else {
667 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
668 				outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
669 			else
670 				outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
671 		}
672 	}
673 
674 	if (DO_BIC(BIC_CPU_c1))
675 		outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
676 	if (DO_BIC(BIC_CPU_c3))
677 		outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
678 	if (DO_BIC(BIC_CPU_c6))
679 		outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
680 	if (DO_BIC(BIC_CPU_c7))
681 		outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
682 
683 	if (DO_BIC(BIC_Mod_c6))
684 		outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
685 
686 	if (DO_BIC(BIC_CoreTmp))
687 		outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
688 
689 	if (do_rapl && !rapl_joules) {
690 		if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
691 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
692 	} else if (do_rapl && rapl_joules) {
693 		if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
694 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
695 	}
696 
697 	for (mp = sys.cp; mp; mp = mp->next) {
698 		if (mp->format == FORMAT_RAW) {
699 			if (mp->width == 64)
700 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
701 			else
702 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
703 		} else {
704 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
705 				outp += sprintf(outp, "%s%8s", delim, mp->name);
706 			else
707 				outp += sprintf(outp, "%s%s", delim, mp->name);
708 		}
709 	}
710 
711 	if (DO_BIC(BIC_PkgTmp))
712 		outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
713 
714 	if (DO_BIC(BIC_GFX_rc6))
715 		outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
716 
717 	if (DO_BIC(BIC_GFXMHz))
718 		outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
719 
720 	if (DO_BIC(BIC_Totl_c0))
721 		outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
722 	if (DO_BIC(BIC_Any_c0))
723 		outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
724 	if (DO_BIC(BIC_GFX_c0))
725 		outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
726 	if (DO_BIC(BIC_CPUGFX))
727 		outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
728 
729 	if (DO_BIC(BIC_Pkgpc2))
730 		outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
731 	if (DO_BIC(BIC_Pkgpc3))
732 		outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
733 	if (DO_BIC(BIC_Pkgpc6))
734 		outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
735 	if (DO_BIC(BIC_Pkgpc7))
736 		outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
737 	if (DO_BIC(BIC_Pkgpc8))
738 		outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
739 	if (DO_BIC(BIC_Pkgpc9))
740 		outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
741 	if (DO_BIC(BIC_Pkgpc10))
742 		outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
743 	if (DO_BIC(BIC_CPU_LPI))
744 		outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : ""));
745 	if (DO_BIC(BIC_SYS_LPI))
746 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
747 
748 	if (do_rapl && !rapl_joules) {
749 		if (DO_BIC(BIC_PkgWatt))
750 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
751 		if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
752 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
753 		if (DO_BIC(BIC_GFXWatt))
754 			outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
755 		if (DO_BIC(BIC_RAMWatt))
756 			outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
757 		if (DO_BIC(BIC_PKG__))
758 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
759 		if (DO_BIC(BIC_RAM__))
760 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
761 	} else if (do_rapl && rapl_joules) {
762 		if (DO_BIC(BIC_Pkg_J))
763 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
764 		if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
765 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
766 		if (DO_BIC(BIC_GFX_J))
767 			outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
768 		if (DO_BIC(BIC_RAM_J))
769 			outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
770 		if (DO_BIC(BIC_PKG__))
771 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
772 		if (DO_BIC(BIC_RAM__))
773 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
774 	}
775 	for (mp = sys.pp; mp; mp = mp->next) {
776 		if (mp->format == FORMAT_RAW) {
777 			if (mp->width == 64)
778 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
779 			else
780 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
781 		} else {
782 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
783 				outp += sprintf(outp, "%s%8s", delim, mp->name);
784 			else
785 				outp += sprintf(outp, "%s%s", delim, mp->name);
786 		}
787 	}
788 
789 	outp += sprintf(outp, "\n");
790 }
791 
792 int dump_counters(struct thread_data *t, struct core_data *c,
793 	struct pkg_data *p)
794 {
795 	int i;
796 	struct msr_counter *mp;
797 
798 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
799 
800 	if (t) {
801 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
802 			t->cpu_id, t->flags);
803 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
804 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
805 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
806 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
807 
808 		if (DO_BIC(BIC_IRQ))
809 			outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
810 		if (DO_BIC(BIC_SMI))
811 			outp += sprintf(outp, "SMI: %d\n", t->smi_count);
812 
813 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
814 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
815 				i, mp->msr_num, t->counter[i]);
816 		}
817 	}
818 
819 	if (c) {
820 		outp += sprintf(outp, "core: %d\n", c->core_id);
821 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
822 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
823 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
824 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
825 		outp += sprintf(outp, "Joules: %0X\n", c->core_energy);
826 
827 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
828 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
829 				i, mp->msr_num, c->counter[i]);
830 		}
831 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
832 	}
833 
834 	if (p) {
835 		outp += sprintf(outp, "package: %d\n", p->package_id);
836 
837 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
838 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
839 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
840 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
841 
842 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
843 		if (DO_BIC(BIC_Pkgpc3))
844 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
845 		if (DO_BIC(BIC_Pkgpc6))
846 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
847 		if (DO_BIC(BIC_Pkgpc7))
848 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
849 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
850 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
851 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
852 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
853 		outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi);
854 		outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi);
855 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
856 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
857 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
858 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
859 		outp += sprintf(outp, "Throttle PKG: %0X\n",
860 			p->rapl_pkg_perf_status);
861 		outp += sprintf(outp, "Throttle RAM: %0X\n",
862 			p->rapl_dram_perf_status);
863 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
864 
865 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
866 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
867 				i, mp->msr_num, p->counter[i]);
868 		}
869 	}
870 
871 	outp += sprintf(outp, "\n");
872 
873 	return 0;
874 }
875 
876 /*
877  * column formatting convention & formats
878  */
879 int format_counters(struct thread_data *t, struct core_data *c,
880 	struct pkg_data *p)
881 {
882 	double interval_float, tsc;
883 	char *fmt8;
884 	int i;
885 	struct msr_counter *mp;
886 	char *delim = "\t";
887 	int printed = 0;
888 
889 	 /* if showing only 1st thread in core and this isn't one, bail out */
890 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
891 		return 0;
892 
893 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
894 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
895 		return 0;
896 
897 	/*if not summary line and --cpu is used */
898 	if ((t != &average.threads) &&
899 		(cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
900 		return 0;
901 
902 	if (DO_BIC(BIC_USEC)) {
903 		/* on each row, print how many usec each timestamp took to gather */
904 		struct timeval tv;
905 
906 		timersub(&t->tv_end, &t->tv_begin, &tv);
907 		outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec);
908 	}
909 
910 	/* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */
911 	if (DO_BIC(BIC_TOD))
912 		outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec);
913 
914 	interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
915 
916 	tsc = t->tsc * tsc_tweak;
917 
918 	/* topo columns, print blanks on 1st (average) line */
919 	if (t == &average.threads) {
920 		if (DO_BIC(BIC_Package))
921 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
922 		if (DO_BIC(BIC_Die))
923 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
924 		if (DO_BIC(BIC_Node))
925 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
926 		if (DO_BIC(BIC_Core))
927 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
928 		if (DO_BIC(BIC_CPU))
929 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
930 		if (DO_BIC(BIC_APIC))
931 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
932 		if (DO_BIC(BIC_X2APIC))
933 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
934 	} else {
935 		if (DO_BIC(BIC_Package)) {
936 			if (p)
937 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
938 			else
939 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
940 		}
941 		if (DO_BIC(BIC_Die)) {
942 			if (c)
943 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id);
944 			else
945 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
946 		}
947 		if (DO_BIC(BIC_Node)) {
948 			if (t)
949 				outp += sprintf(outp, "%s%d",
950 						(printed++ ? delim : ""),
951 					      cpus[t->cpu_id].physical_node_id);
952 			else
953 				outp += sprintf(outp, "%s-",
954 						(printed++ ? delim : ""));
955 		}
956 		if (DO_BIC(BIC_Core)) {
957 			if (c)
958 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
959 			else
960 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
961 		}
962 		if (DO_BIC(BIC_CPU))
963 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
964 		if (DO_BIC(BIC_APIC))
965 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id);
966 		if (DO_BIC(BIC_X2APIC))
967 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id);
968 	}
969 
970 	if (DO_BIC(BIC_Avg_MHz))
971 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
972 			1.0 / units * t->aperf / interval_float);
973 
974 	if (DO_BIC(BIC_Busy))
975 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf/tsc);
976 
977 	if (DO_BIC(BIC_Bzy_MHz)) {
978 		if (has_base_hz)
979 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
980 		else
981 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
982 				tsc / units * t->aperf / t->mperf / interval_float);
983 	}
984 
985 	if (DO_BIC(BIC_TSC_MHz))
986 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
987 
988 	/* IRQ */
989 	if (DO_BIC(BIC_IRQ)) {
990 		if (sums_need_wide_columns)
991 			outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
992 		else
993 			outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
994 	}
995 
996 	/* SMI */
997 	if (DO_BIC(BIC_SMI))
998 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
999 
1000 	/* Added counters */
1001 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1002 		if (mp->format == FORMAT_RAW) {
1003 			if (mp->width == 32)
1004 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) t->counter[i]);
1005 			else
1006 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
1007 		} else if (mp->format == FORMAT_DELTA) {
1008 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1009 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
1010 			else
1011 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
1012 		} else if (mp->format == FORMAT_PERCENT) {
1013 			if (mp->type == COUNTER_USEC)
1014 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), t->counter[i]/interval_float/10000);
1015 			else
1016 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i]/tsc);
1017 		}
1018 	}
1019 
1020 	/* C1 */
1021 	if (DO_BIC(BIC_CPU_c1))
1022 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1/tsc);
1023 
1024 
1025 	/* print per-core data only for 1st thread in core */
1026 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1027 		goto done;
1028 
1029 	if (DO_BIC(BIC_CPU_c3))
1030 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3/tsc);
1031 	if (DO_BIC(BIC_CPU_c6))
1032 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6/tsc);
1033 	if (DO_BIC(BIC_CPU_c7))
1034 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7/tsc);
1035 
1036 	/* Mod%c6 */
1037 	if (DO_BIC(BIC_Mod_c6))
1038 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
1039 
1040 	if (DO_BIC(BIC_CoreTmp))
1041 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
1042 
1043 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1044 		if (mp->format == FORMAT_RAW) {
1045 			if (mp->width == 32)
1046 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) c->counter[i]);
1047 			else
1048 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
1049 		} else if (mp->format == FORMAT_DELTA) {
1050 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1051 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
1052 			else
1053 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
1054 		} else if (mp->format == FORMAT_PERCENT) {
1055 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i]/tsc);
1056 		}
1057 	}
1058 
1059 	/*
1060 	 * If measurement interval exceeds minimum RAPL Joule Counter range,
1061 	 * indicate that results are suspect by printing "**" in fraction place.
1062 	 */
1063 	if (interval_float < rapl_joule_counter_range)
1064 		fmt8 = "%s%.2f";
1065 	else
1066 		fmt8 = "%6.0f**";
1067 
1068 	if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
1069 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float);
1070 	if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
1071 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units);
1072 
1073 	/* print per-package data only for 1st core in package */
1074 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1075 		goto done;
1076 
1077 	/* PkgTmp */
1078 	if (DO_BIC(BIC_PkgTmp))
1079 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
1080 
1081 	/* GFXrc6 */
1082 	if (DO_BIC(BIC_GFX_rc6)) {
1083 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
1084 			outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
1085 		} else {
1086 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1087 				p->gfx_rc6_ms / 10.0 / interval_float);
1088 		}
1089 	}
1090 
1091 	/* GFXMHz */
1092 	if (DO_BIC(BIC_GFXMHz))
1093 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
1094 
1095 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
1096 	if (DO_BIC(BIC_Totl_c0))
1097 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0/tsc);
1098 	if (DO_BIC(BIC_Any_c0))
1099 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0/tsc);
1100 	if (DO_BIC(BIC_GFX_c0))
1101 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0/tsc);
1102 	if (DO_BIC(BIC_CPUGFX))
1103 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0/tsc);
1104 
1105 	if (DO_BIC(BIC_Pkgpc2))
1106 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2/tsc);
1107 	if (DO_BIC(BIC_Pkgpc3))
1108 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3/tsc);
1109 	if (DO_BIC(BIC_Pkgpc6))
1110 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6/tsc);
1111 	if (DO_BIC(BIC_Pkgpc7))
1112 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7/tsc);
1113 	if (DO_BIC(BIC_Pkgpc8))
1114 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8/tsc);
1115 	if (DO_BIC(BIC_Pkgpc9))
1116 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9/tsc);
1117 	if (DO_BIC(BIC_Pkgpc10))
1118 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10/tsc);
1119 
1120 	if (DO_BIC(BIC_CPU_LPI))
1121 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float);
1122 	if (DO_BIC(BIC_SYS_LPI))
1123 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float);
1124 
1125 	if (DO_BIC(BIC_PkgWatt))
1126 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
1127 	if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1128 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
1129 	if (DO_BIC(BIC_GFXWatt))
1130 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
1131 	if (DO_BIC(BIC_RAMWatt))
1132 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units / interval_float);
1133 	if (DO_BIC(BIC_Pkg_J))
1134 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
1135 	if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1136 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
1137 	if (DO_BIC(BIC_GFX_J))
1138 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
1139 	if (DO_BIC(BIC_RAM_J))
1140 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
1141 	if (DO_BIC(BIC_PKG__))
1142 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
1143 	if (DO_BIC(BIC_RAM__))
1144 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
1145 
1146 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1147 		if (mp->format == FORMAT_RAW) {
1148 			if (mp->width == 32)
1149 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) p->counter[i]);
1150 			else
1151 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
1152 		} else if (mp->format == FORMAT_DELTA) {
1153 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1154 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
1155 			else
1156 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
1157 		} else if (mp->format == FORMAT_PERCENT) {
1158 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i]/tsc);
1159 		}
1160 	}
1161 
1162 done:
1163 	if (*(outp - 1) != '\n')
1164 		outp += sprintf(outp, "\n");
1165 
1166 	return 0;
1167 }
1168 
1169 void flush_output_stdout(void)
1170 {
1171 	FILE *filep;
1172 
1173 	if (outf == stderr)
1174 		filep = stdout;
1175 	else
1176 		filep = outf;
1177 
1178 	fputs(output_buffer, filep);
1179 	fflush(filep);
1180 
1181 	outp = output_buffer;
1182 }
1183 void flush_output_stderr(void)
1184 {
1185 	fputs(output_buffer, outf);
1186 	fflush(outf);
1187 	outp = output_buffer;
1188 }
1189 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1190 {
1191 	static int printed;
1192 
1193 	if (!printed || !summary_only)
1194 		print_header("\t");
1195 
1196 	format_counters(&average.threads, &average.cores, &average.packages);
1197 
1198 	printed = 1;
1199 
1200 	if (summary_only)
1201 		return;
1202 
1203 	for_all_cpus(format_counters, t, c, p);
1204 }
1205 
1206 #define DELTA_WRAP32(new, old)			\
1207 	if (new > old) {			\
1208 		old = new - old;		\
1209 	} else {				\
1210 		old = 0x100000000 + new - old;	\
1211 	}
1212 
1213 int
1214 delta_package(struct pkg_data *new, struct pkg_data *old)
1215 {
1216 	int i;
1217 	struct msr_counter *mp;
1218 
1219 
1220 	if (DO_BIC(BIC_Totl_c0))
1221 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1222 	if (DO_BIC(BIC_Any_c0))
1223 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1224 	if (DO_BIC(BIC_GFX_c0))
1225 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1226 	if (DO_BIC(BIC_CPUGFX))
1227 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1228 
1229 	old->pc2 = new->pc2 - old->pc2;
1230 	if (DO_BIC(BIC_Pkgpc3))
1231 		old->pc3 = new->pc3 - old->pc3;
1232 	if (DO_BIC(BIC_Pkgpc6))
1233 		old->pc6 = new->pc6 - old->pc6;
1234 	if (DO_BIC(BIC_Pkgpc7))
1235 		old->pc7 = new->pc7 - old->pc7;
1236 	old->pc8 = new->pc8 - old->pc8;
1237 	old->pc9 = new->pc9 - old->pc9;
1238 	old->pc10 = new->pc10 - old->pc10;
1239 	old->cpu_lpi = new->cpu_lpi - old->cpu_lpi;
1240 	old->sys_lpi = new->sys_lpi - old->sys_lpi;
1241 	old->pkg_temp_c = new->pkg_temp_c;
1242 
1243 	/* flag an error when rc6 counter resets/wraps */
1244 	if (old->gfx_rc6_ms >  new->gfx_rc6_ms)
1245 		old->gfx_rc6_ms = -1;
1246 	else
1247 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1248 
1249 	old->gfx_mhz = new->gfx_mhz;
1250 
1251 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
1252 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
1253 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
1254 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
1255 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
1256 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
1257 
1258 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1259 		if (mp->format == FORMAT_RAW)
1260 			old->counter[i] = new->counter[i];
1261 		else
1262 			old->counter[i] = new->counter[i] - old->counter[i];
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 void
1269 delta_core(struct core_data *new, struct core_data *old)
1270 {
1271 	int i;
1272 	struct msr_counter *mp;
1273 
1274 	old->c3 = new->c3 - old->c3;
1275 	old->c6 = new->c6 - old->c6;
1276 	old->c7 = new->c7 - old->c7;
1277 	old->core_temp_c = new->core_temp_c;
1278 	old->mc6_us = new->mc6_us - old->mc6_us;
1279 
1280 	DELTA_WRAP32(new->core_energy, old->core_energy);
1281 
1282 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1283 		if (mp->format == FORMAT_RAW)
1284 			old->counter[i] = new->counter[i];
1285 		else
1286 			old->counter[i] = new->counter[i] - old->counter[i];
1287 	}
1288 }
1289 
1290 /*
1291  * old = new - old
1292  */
1293 int
1294 delta_thread(struct thread_data *new, struct thread_data *old,
1295 	struct core_data *core_delta)
1296 {
1297 	int i;
1298 	struct msr_counter *mp;
1299 
1300 	/* we run cpuid just the 1st time, copy the results */
1301 	if (DO_BIC(BIC_APIC))
1302 		new->apic_id = old->apic_id;
1303 	if (DO_BIC(BIC_X2APIC))
1304 		new->x2apic_id = old->x2apic_id;
1305 
1306 	/*
1307 	 * the timestamps from start of measurement interval are in "old"
1308 	 * the timestamp from end of measurement interval are in "new"
1309 	 * over-write old w/ new so we can print end of interval values
1310 	 */
1311 
1312 	old->tv_begin = new->tv_begin;
1313 	old->tv_end = new->tv_end;
1314 
1315 	old->tsc = new->tsc - old->tsc;
1316 
1317 	/* check for TSC < 1 Mcycles over interval */
1318 	if (old->tsc < (1000 * 1000))
1319 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1320 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1321 		     "or just the deep ones with \"processor.max_cstate=1\"");
1322 
1323 	old->c1 = new->c1 - old->c1;
1324 
1325 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1326 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1327 			old->aperf = new->aperf - old->aperf;
1328 			old->mperf = new->mperf - old->mperf;
1329 		} else {
1330 			return -1;
1331 		}
1332 	}
1333 
1334 
1335 	if (use_c1_residency_msr) {
1336 		/*
1337 		 * Some models have a dedicated C1 residency MSR,
1338 		 * which should be more accurate than the derivation below.
1339 		 */
1340 	} else {
1341 		/*
1342 		 * As counter collection is not atomic,
1343 		 * it is possible for mperf's non-halted cycles + idle states
1344 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1345 		 */
1346 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1347 			old->c1 = 0;
1348 		else {
1349 			/* normal case, derive c1 */
1350 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1351 				- core_delta->c6 - core_delta->c7;
1352 		}
1353 	}
1354 
1355 	if (old->mperf == 0) {
1356 		if (debug > 1)
1357 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1358 		old->mperf = 1;	/* divide by 0 protection */
1359 	}
1360 
1361 	if (DO_BIC(BIC_IRQ))
1362 		old->irq_count = new->irq_count - old->irq_count;
1363 
1364 	if (DO_BIC(BIC_SMI))
1365 		old->smi_count = new->smi_count - old->smi_count;
1366 
1367 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1368 		if (mp->format == FORMAT_RAW)
1369 			old->counter[i] = new->counter[i];
1370 		else
1371 			old->counter[i] = new->counter[i] - old->counter[i];
1372 	}
1373 	return 0;
1374 }
1375 
1376 int delta_cpu(struct thread_data *t, struct core_data *c,
1377 	struct pkg_data *p, struct thread_data *t2,
1378 	struct core_data *c2, struct pkg_data *p2)
1379 {
1380 	int retval = 0;
1381 
1382 	/* calculate core delta only for 1st thread in core */
1383 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1384 		delta_core(c, c2);
1385 
1386 	/* always calculate thread delta */
1387 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1388 	if (retval)
1389 		return retval;
1390 
1391 	/* calculate package delta only for 1st core in package */
1392 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1393 		retval = delta_package(p, p2);
1394 
1395 	return retval;
1396 }
1397 
1398 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1399 {
1400 	int i;
1401 	struct msr_counter  *mp;
1402 
1403 	t->tv_begin.tv_sec = 0;
1404 	t->tv_begin.tv_usec = 0;
1405 	t->tv_end.tv_sec = 0;
1406 	t->tv_end.tv_usec = 0;
1407 
1408 	t->tsc = 0;
1409 	t->aperf = 0;
1410 	t->mperf = 0;
1411 	t->c1 = 0;
1412 
1413 	t->irq_count = 0;
1414 	t->smi_count = 0;
1415 
1416 	/* tells format_counters to dump all fields from this set */
1417 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1418 
1419 	c->c3 = 0;
1420 	c->c6 = 0;
1421 	c->c7 = 0;
1422 	c->mc6_us = 0;
1423 	c->core_temp_c = 0;
1424 	c->core_energy = 0;
1425 
1426 	p->pkg_wtd_core_c0 = 0;
1427 	p->pkg_any_core_c0 = 0;
1428 	p->pkg_any_gfxe_c0 = 0;
1429 	p->pkg_both_core_gfxe_c0 = 0;
1430 
1431 	p->pc2 = 0;
1432 	if (DO_BIC(BIC_Pkgpc3))
1433 		p->pc3 = 0;
1434 	if (DO_BIC(BIC_Pkgpc6))
1435 		p->pc6 = 0;
1436 	if (DO_BIC(BIC_Pkgpc7))
1437 		p->pc7 = 0;
1438 	p->pc8 = 0;
1439 	p->pc9 = 0;
1440 	p->pc10 = 0;
1441 	p->cpu_lpi = 0;
1442 	p->sys_lpi = 0;
1443 
1444 	p->energy_pkg = 0;
1445 	p->energy_dram = 0;
1446 	p->energy_cores = 0;
1447 	p->energy_gfx = 0;
1448 	p->rapl_pkg_perf_status = 0;
1449 	p->rapl_dram_perf_status = 0;
1450 	p->pkg_temp_c = 0;
1451 
1452 	p->gfx_rc6_ms = 0;
1453 	p->gfx_mhz = 0;
1454 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1455 		t->counter[i] = 0;
1456 
1457 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1458 		c->counter[i] = 0;
1459 
1460 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1461 		p->counter[i] = 0;
1462 }
1463 int sum_counters(struct thread_data *t, struct core_data *c,
1464 	struct pkg_data *p)
1465 {
1466 	int i;
1467 	struct msr_counter *mp;
1468 
1469 	/* copy un-changing apic_id's */
1470 	if (DO_BIC(BIC_APIC))
1471 		average.threads.apic_id = t->apic_id;
1472 	if (DO_BIC(BIC_X2APIC))
1473 		average.threads.x2apic_id = t->x2apic_id;
1474 
1475 	/* remember first tv_begin */
1476 	if (average.threads.tv_begin.tv_sec == 0)
1477 		average.threads.tv_begin = t->tv_begin;
1478 
1479 	/* remember last tv_end */
1480 	average.threads.tv_end = t->tv_end;
1481 
1482 	average.threads.tsc += t->tsc;
1483 	average.threads.aperf += t->aperf;
1484 	average.threads.mperf += t->mperf;
1485 	average.threads.c1 += t->c1;
1486 
1487 	average.threads.irq_count += t->irq_count;
1488 	average.threads.smi_count += t->smi_count;
1489 
1490 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1491 		if (mp->format == FORMAT_RAW)
1492 			continue;
1493 		average.threads.counter[i] += t->counter[i];
1494 	}
1495 
1496 	/* sum per-core values only for 1st thread in core */
1497 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1498 		return 0;
1499 
1500 	average.cores.c3 += c->c3;
1501 	average.cores.c6 += c->c6;
1502 	average.cores.c7 += c->c7;
1503 	average.cores.mc6_us += c->mc6_us;
1504 
1505 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1506 
1507 	average.cores.core_energy += c->core_energy;
1508 
1509 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1510 		if (mp->format == FORMAT_RAW)
1511 			continue;
1512 		average.cores.counter[i] += c->counter[i];
1513 	}
1514 
1515 	/* sum per-pkg values only for 1st core in pkg */
1516 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1517 		return 0;
1518 
1519 	if (DO_BIC(BIC_Totl_c0))
1520 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1521 	if (DO_BIC(BIC_Any_c0))
1522 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1523 	if (DO_BIC(BIC_GFX_c0))
1524 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1525 	if (DO_BIC(BIC_CPUGFX))
1526 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1527 
1528 	average.packages.pc2 += p->pc2;
1529 	if (DO_BIC(BIC_Pkgpc3))
1530 		average.packages.pc3 += p->pc3;
1531 	if (DO_BIC(BIC_Pkgpc6))
1532 		average.packages.pc6 += p->pc6;
1533 	if (DO_BIC(BIC_Pkgpc7))
1534 		average.packages.pc7 += p->pc7;
1535 	average.packages.pc8 += p->pc8;
1536 	average.packages.pc9 += p->pc9;
1537 	average.packages.pc10 += p->pc10;
1538 
1539 	average.packages.cpu_lpi = p->cpu_lpi;
1540 	average.packages.sys_lpi = p->sys_lpi;
1541 
1542 	average.packages.energy_pkg += p->energy_pkg;
1543 	average.packages.energy_dram += p->energy_dram;
1544 	average.packages.energy_cores += p->energy_cores;
1545 	average.packages.energy_gfx += p->energy_gfx;
1546 
1547 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1548 	average.packages.gfx_mhz = p->gfx_mhz;
1549 
1550 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1551 
1552 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1553 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1554 
1555 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1556 		if (mp->format == FORMAT_RAW)
1557 			continue;
1558 		average.packages.counter[i] += p->counter[i];
1559 	}
1560 	return 0;
1561 }
1562 /*
1563  * sum the counters for all cpus in the system
1564  * compute the weighted average
1565  */
1566 void compute_average(struct thread_data *t, struct core_data *c,
1567 	struct pkg_data *p)
1568 {
1569 	int i;
1570 	struct msr_counter *mp;
1571 
1572 	clear_counters(&average.threads, &average.cores, &average.packages);
1573 
1574 	for_all_cpus(sum_counters, t, c, p);
1575 
1576 	average.threads.tsc /= topo.num_cpus;
1577 	average.threads.aperf /= topo.num_cpus;
1578 	average.threads.mperf /= topo.num_cpus;
1579 	average.threads.c1 /= topo.num_cpus;
1580 
1581 	if (average.threads.irq_count > 9999999)
1582 		sums_need_wide_columns = 1;
1583 
1584 	average.cores.c3 /= topo.num_cores;
1585 	average.cores.c6 /= topo.num_cores;
1586 	average.cores.c7 /= topo.num_cores;
1587 	average.cores.mc6_us /= topo.num_cores;
1588 
1589 	if (DO_BIC(BIC_Totl_c0))
1590 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1591 	if (DO_BIC(BIC_Any_c0))
1592 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1593 	if (DO_BIC(BIC_GFX_c0))
1594 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1595 	if (DO_BIC(BIC_CPUGFX))
1596 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1597 
1598 	average.packages.pc2 /= topo.num_packages;
1599 	if (DO_BIC(BIC_Pkgpc3))
1600 		average.packages.pc3 /= topo.num_packages;
1601 	if (DO_BIC(BIC_Pkgpc6))
1602 		average.packages.pc6 /= topo.num_packages;
1603 	if (DO_BIC(BIC_Pkgpc7))
1604 		average.packages.pc7 /= topo.num_packages;
1605 
1606 	average.packages.pc8 /= topo.num_packages;
1607 	average.packages.pc9 /= topo.num_packages;
1608 	average.packages.pc10 /= topo.num_packages;
1609 
1610 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1611 		if (mp->format == FORMAT_RAW)
1612 			continue;
1613 		if (mp->type == COUNTER_ITEMS) {
1614 			if (average.threads.counter[i] > 9999999)
1615 				sums_need_wide_columns = 1;
1616 			continue;
1617 		}
1618 		average.threads.counter[i] /= topo.num_cpus;
1619 	}
1620 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1621 		if (mp->format == FORMAT_RAW)
1622 			continue;
1623 		if (mp->type == COUNTER_ITEMS) {
1624 			if (average.cores.counter[i] > 9999999)
1625 				sums_need_wide_columns = 1;
1626 		}
1627 		average.cores.counter[i] /= topo.num_cores;
1628 	}
1629 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1630 		if (mp->format == FORMAT_RAW)
1631 			continue;
1632 		if (mp->type == COUNTER_ITEMS) {
1633 			if (average.packages.counter[i] > 9999999)
1634 				sums_need_wide_columns = 1;
1635 		}
1636 		average.packages.counter[i] /= topo.num_packages;
1637 	}
1638 }
1639 
1640 static unsigned long long rdtsc(void)
1641 {
1642 	unsigned int low, high;
1643 
1644 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
1645 
1646 	return low | ((unsigned long long)high) << 32;
1647 }
1648 
1649 /*
1650  * Open a file, and exit on failure
1651  */
1652 FILE *fopen_or_die(const char *path, const char *mode)
1653 {
1654 	FILE *filep = fopen(path, mode);
1655 
1656 	if (!filep)
1657 		err(1, "%s: open failed", path);
1658 	return filep;
1659 }
1660 /*
1661  * snapshot_sysfs_counter()
1662  *
1663  * return snapshot of given counter
1664  */
1665 unsigned long long snapshot_sysfs_counter(char *path)
1666 {
1667 	FILE *fp;
1668 	int retval;
1669 	unsigned long long counter;
1670 
1671 	fp = fopen_or_die(path, "r");
1672 
1673 	retval = fscanf(fp, "%lld", &counter);
1674 	if (retval != 1)
1675 		err(1, "snapshot_sysfs_counter(%s)", path);
1676 
1677 	fclose(fp);
1678 
1679 	return counter;
1680 }
1681 
1682 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1683 {
1684 	if (mp->msr_num != 0) {
1685 		if (get_msr(cpu, mp->msr_num, counterp))
1686 			return -1;
1687 	} else {
1688 		char path[128 + PATH_BYTES];
1689 
1690 		if (mp->flags & SYSFS_PERCPU) {
1691 			sprintf(path, "/sys/devices/system/cpu/cpu%d/%s",
1692 				 cpu, mp->path);
1693 
1694 			*counterp = snapshot_sysfs_counter(path);
1695 		} else {
1696 			*counterp = snapshot_sysfs_counter(mp->path);
1697 		}
1698 	}
1699 
1700 	return 0;
1701 }
1702 
1703 void get_apic_id(struct thread_data *t)
1704 {
1705 	unsigned int eax, ebx, ecx, edx;
1706 
1707 	if (DO_BIC(BIC_APIC)) {
1708 		eax = ebx = ecx = edx = 0;
1709 		__cpuid(1, eax, ebx, ecx, edx);
1710 
1711 		t->apic_id = (ebx >> 24) & 0xff;
1712 	}
1713 
1714 	if (!DO_BIC(BIC_X2APIC))
1715 		return;
1716 
1717 	if (authentic_amd) {
1718 		unsigned int topology_extensions;
1719 
1720 		if (max_extended_level < 0x8000001e)
1721 			return;
1722 
1723 		eax = ebx = ecx = edx = 0;
1724 		__cpuid(0x80000001, eax, ebx, ecx, edx);
1725 			topology_extensions = ecx & (1 << 22);
1726 
1727 		if (topology_extensions == 0)
1728 			return;
1729 
1730 		eax = ebx = ecx = edx = 0;
1731 		__cpuid(0x8000001e, eax, ebx, ecx, edx);
1732 
1733 		t->x2apic_id = eax;
1734 		return;
1735 	}
1736 
1737 	if (!genuine_intel)
1738 		return;
1739 
1740 	if (max_level < 0xb)
1741 		return;
1742 
1743 	ecx = 0;
1744 	__cpuid(0xb, eax, ebx, ecx, edx);
1745 	t->x2apic_id = edx;
1746 
1747 	if (debug && (t->apic_id != (t->x2apic_id & 0xff)))
1748 		fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n",
1749 				t->cpu_id, t->apic_id, t->x2apic_id);
1750 }
1751 
1752 /*
1753  * get_counters(...)
1754  * migrate to cpu
1755  * acquire and record local counters for that cpu
1756  */
1757 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1758 {
1759 	int cpu = t->cpu_id;
1760 	unsigned long long msr;
1761 	int aperf_mperf_retry_count = 0;
1762 	struct msr_counter *mp;
1763 	int i;
1764 
1765 	gettimeofday(&t->tv_begin, (struct timezone *)NULL);
1766 
1767 	if (cpu_migrate(cpu)) {
1768 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1769 		return -1;
1770 	}
1771 
1772 	if (first_counter_read)
1773 		get_apic_id(t);
1774 retry:
1775 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
1776 
1777 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1778 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1779 
1780 		/*
1781 		 * The TSC, APERF and MPERF must be read together for
1782 		 * APERF/MPERF and MPERF/TSC to give accurate results.
1783 		 *
1784 		 * Unfortunately, APERF and MPERF are read by
1785 		 * individual system call, so delays may occur
1786 		 * between them.  If the time to read them
1787 		 * varies by a large amount, we re-read them.
1788 		 */
1789 
1790 		/*
1791 		 * This initial dummy APERF read has been seen to
1792 		 * reduce jitter in the subsequent reads.
1793 		 */
1794 
1795 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1796 			return -3;
1797 
1798 		t->tsc = rdtsc();	/* re-read close to APERF */
1799 
1800 		tsc_before = t->tsc;
1801 
1802 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1803 			return -3;
1804 
1805 		tsc_between = rdtsc();
1806 
1807 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1808 			return -4;
1809 
1810 		tsc_after = rdtsc();
1811 
1812 		aperf_time = tsc_between - tsc_before;
1813 		mperf_time = tsc_after - tsc_between;
1814 
1815 		/*
1816 		 * If the system call latency to read APERF and MPERF
1817 		 * differ by more than 2x, then try again.
1818 		 */
1819 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1820 			aperf_mperf_retry_count++;
1821 			if (aperf_mperf_retry_count < 5)
1822 				goto retry;
1823 			else
1824 				warnx("cpu%d jitter %lld %lld",
1825 					cpu, aperf_time, mperf_time);
1826 		}
1827 		aperf_mperf_retry_count = 0;
1828 
1829 		t->aperf = t->aperf * aperf_mperf_multiplier;
1830 		t->mperf = t->mperf * aperf_mperf_multiplier;
1831 	}
1832 
1833 	if (DO_BIC(BIC_IRQ))
1834 		t->irq_count = irqs_per_cpu[cpu];
1835 	if (DO_BIC(BIC_SMI)) {
1836 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1837 			return -5;
1838 		t->smi_count = msr & 0xFFFFFFFF;
1839 	}
1840 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1841 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1842 			return -6;
1843 	}
1844 
1845 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1846 		if (get_mp(cpu, mp, &t->counter[i]))
1847 			return -10;
1848 	}
1849 
1850 	/* collect core counters only for 1st thread in core */
1851 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1852 		goto done;
1853 
1854 	if (DO_BIC(BIC_CPU_c3)) {
1855 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1856 			return -6;
1857 	}
1858 
1859 	if (DO_BIC(BIC_CPU_c6) && !do_knl_cstates) {
1860 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1861 			return -7;
1862 	} else if (do_knl_cstates) {
1863 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1864 			return -7;
1865 	}
1866 
1867 	if (DO_BIC(BIC_CPU_c7))
1868 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1869 			return -8;
1870 
1871 	if (DO_BIC(BIC_Mod_c6))
1872 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1873 			return -8;
1874 
1875 	if (DO_BIC(BIC_CoreTmp)) {
1876 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1877 			return -9;
1878 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1879 	}
1880 
1881 	if (do_rapl & RAPL_AMD_F17H) {
1882 		if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr))
1883 			return -14;
1884 		c->core_energy = msr & 0xFFFFFFFF;
1885 	}
1886 
1887 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1888 		if (get_mp(cpu, mp, &c->counter[i]))
1889 			return -10;
1890 	}
1891 
1892 	/* collect package counters only for 1st core in package */
1893 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1894 		goto done;
1895 
1896 	if (DO_BIC(BIC_Totl_c0)) {
1897 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1898 			return -10;
1899 	}
1900 	if (DO_BIC(BIC_Any_c0)) {
1901 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1902 			return -11;
1903 	}
1904 	if (DO_BIC(BIC_GFX_c0)) {
1905 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1906 			return -12;
1907 	}
1908 	if (DO_BIC(BIC_CPUGFX)) {
1909 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1910 			return -13;
1911 	}
1912 	if (DO_BIC(BIC_Pkgpc3))
1913 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1914 			return -9;
1915 	if (DO_BIC(BIC_Pkgpc6)) {
1916 		if (do_slm_cstates) {
1917 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1918 				return -10;
1919 		} else {
1920 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1921 				return -10;
1922 		}
1923 	}
1924 
1925 	if (DO_BIC(BIC_Pkgpc2))
1926 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1927 			return -11;
1928 	if (DO_BIC(BIC_Pkgpc7))
1929 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1930 			return -12;
1931 	if (DO_BIC(BIC_Pkgpc8))
1932 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1933 			return -13;
1934 	if (DO_BIC(BIC_Pkgpc9))
1935 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1936 			return -13;
1937 	if (DO_BIC(BIC_Pkgpc10))
1938 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1939 			return -13;
1940 
1941 	if (DO_BIC(BIC_CPU_LPI))
1942 		p->cpu_lpi = cpuidle_cur_cpu_lpi_us;
1943 	if (DO_BIC(BIC_SYS_LPI))
1944 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
1945 
1946 	if (do_rapl & RAPL_PKG) {
1947 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1948 			return -13;
1949 		p->energy_pkg = msr & 0xFFFFFFFF;
1950 	}
1951 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1952 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1953 			return -14;
1954 		p->energy_cores = msr & 0xFFFFFFFF;
1955 	}
1956 	if (do_rapl & RAPL_DRAM) {
1957 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1958 			return -15;
1959 		p->energy_dram = msr & 0xFFFFFFFF;
1960 	}
1961 	if (do_rapl & RAPL_GFX) {
1962 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1963 			return -16;
1964 		p->energy_gfx = msr & 0xFFFFFFFF;
1965 	}
1966 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1967 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1968 			return -16;
1969 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1970 	}
1971 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1972 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1973 			return -16;
1974 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1975 	}
1976 	if (do_rapl & RAPL_AMD_F17H) {
1977 		if (get_msr(cpu, MSR_PKG_ENERGY_STAT, &msr))
1978 			return -13;
1979 		p->energy_pkg = msr & 0xFFFFFFFF;
1980 	}
1981 	if (DO_BIC(BIC_PkgTmp)) {
1982 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1983 			return -17;
1984 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1985 	}
1986 
1987 	if (DO_BIC(BIC_GFX_rc6))
1988 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
1989 
1990 	if (DO_BIC(BIC_GFXMHz))
1991 		p->gfx_mhz = gfx_cur_mhz;
1992 
1993 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1994 		if (get_mp(cpu, mp, &p->counter[i]))
1995 			return -10;
1996 	}
1997 done:
1998 	gettimeofday(&t->tv_end, (struct timezone *)NULL);
1999 
2000 	return 0;
2001 }
2002 
2003 /*
2004  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
2005  * If you change the values, note they are used both in comparisons
2006  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
2007  */
2008 
2009 #define PCLUKN 0 /* Unknown */
2010 #define PCLRSV 1 /* Reserved */
2011 #define PCL__0 2 /* PC0 */
2012 #define PCL__1 3 /* PC1 */
2013 #define PCL__2 4 /* PC2 */
2014 #define PCL__3 5 /* PC3 */
2015 #define PCL__4 6 /* PC4 */
2016 #define PCL__6 7 /* PC6 */
2017 #define PCL_6N 8 /* PC6 No Retention */
2018 #define PCL_6R 9 /* PC6 Retention */
2019 #define PCL__7 10 /* PC7 */
2020 #define PCL_7S 11 /* PC7 Shrink */
2021 #define PCL__8 12 /* PC8 */
2022 #define PCL__9 13 /* PC9 */
2023 #define PCL_10 14 /* PC10 */
2024 #define PCLUNL 15 /* Unlimited */
2025 
2026 int pkg_cstate_limit = PCLUKN;
2027 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
2028 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited"};
2029 
2030 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};
2031 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};
2032 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};
2033 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};
2034 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};
2035 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};
2036 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};
2037 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};
2038 
2039 
2040 static void
2041 calculate_tsc_tweak()
2042 {
2043 	tsc_tweak = base_hz / tsc_hz;
2044 }
2045 
2046 static void
2047 dump_nhm_platform_info(void)
2048 {
2049 	unsigned long long msr;
2050 	unsigned int ratio;
2051 
2052 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2053 
2054 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
2055 
2056 	ratio = (msr >> 40) & 0xFF;
2057 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
2058 		ratio, bclk, ratio * bclk);
2059 
2060 	ratio = (msr >> 8) & 0xFF;
2061 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2062 		ratio, bclk, ratio * bclk);
2063 
2064 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
2065 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
2066 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
2067 
2068 	return;
2069 }
2070 
2071 static void
2072 dump_hsw_turbo_ratio_limits(void)
2073 {
2074 	unsigned long long msr;
2075 	unsigned int ratio;
2076 
2077 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
2078 
2079 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
2080 
2081 	ratio = (msr >> 8) & 0xFF;
2082 	if (ratio)
2083 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
2084 			ratio, bclk, ratio * bclk);
2085 
2086 	ratio = (msr >> 0) & 0xFF;
2087 	if (ratio)
2088 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
2089 			ratio, bclk, ratio * bclk);
2090 	return;
2091 }
2092 
2093 static void
2094 dump_ivt_turbo_ratio_limits(void)
2095 {
2096 	unsigned long long msr;
2097 	unsigned int ratio;
2098 
2099 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
2100 
2101 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
2102 
2103 	ratio = (msr >> 56) & 0xFF;
2104 	if (ratio)
2105 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
2106 			ratio, bclk, ratio * bclk);
2107 
2108 	ratio = (msr >> 48) & 0xFF;
2109 	if (ratio)
2110 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
2111 			ratio, bclk, ratio * bclk);
2112 
2113 	ratio = (msr >> 40) & 0xFF;
2114 	if (ratio)
2115 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
2116 			ratio, bclk, ratio * bclk);
2117 
2118 	ratio = (msr >> 32) & 0xFF;
2119 	if (ratio)
2120 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
2121 			ratio, bclk, ratio * bclk);
2122 
2123 	ratio = (msr >> 24) & 0xFF;
2124 	if (ratio)
2125 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
2126 			ratio, bclk, ratio * bclk);
2127 
2128 	ratio = (msr >> 16) & 0xFF;
2129 	if (ratio)
2130 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
2131 			ratio, bclk, ratio * bclk);
2132 
2133 	ratio = (msr >> 8) & 0xFF;
2134 	if (ratio)
2135 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
2136 			ratio, bclk, ratio * bclk);
2137 
2138 	ratio = (msr >> 0) & 0xFF;
2139 	if (ratio)
2140 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
2141 			ratio, bclk, ratio * bclk);
2142 	return;
2143 }
2144 int has_turbo_ratio_group_limits(int family, int model)
2145 {
2146 
2147 	if (!genuine_intel)
2148 		return 0;
2149 
2150 	switch (model) {
2151 	case INTEL_FAM6_ATOM_GOLDMONT:
2152 	case INTEL_FAM6_SKYLAKE_X:
2153 	case INTEL_FAM6_ATOM_GOLDMONT_X:
2154 		return 1;
2155 	}
2156 	return 0;
2157 }
2158 
2159 static void
2160 dump_turbo_ratio_limits(int family, int model)
2161 {
2162 	unsigned long long msr, core_counts;
2163 	unsigned int ratio, group_size;
2164 
2165 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2166 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
2167 
2168 	if (has_turbo_ratio_group_limits(family, model)) {
2169 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
2170 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
2171 	} else {
2172 		core_counts = 0x0807060504030201;
2173 	}
2174 
2175 	ratio = (msr >> 56) & 0xFF;
2176 	group_size = (core_counts >> 56) & 0xFF;
2177 	if (ratio)
2178 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2179 			ratio, bclk, ratio * bclk, group_size);
2180 
2181 	ratio = (msr >> 48) & 0xFF;
2182 	group_size = (core_counts >> 48) & 0xFF;
2183 	if (ratio)
2184 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2185 			ratio, bclk, ratio * bclk, group_size);
2186 
2187 	ratio = (msr >> 40) & 0xFF;
2188 	group_size = (core_counts >> 40) & 0xFF;
2189 	if (ratio)
2190 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2191 			ratio, bclk, ratio * bclk, group_size);
2192 
2193 	ratio = (msr >> 32) & 0xFF;
2194 	group_size = (core_counts >> 32) & 0xFF;
2195 	if (ratio)
2196 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2197 			ratio, bclk, ratio * bclk, group_size);
2198 
2199 	ratio = (msr >> 24) & 0xFF;
2200 	group_size = (core_counts >> 24) & 0xFF;
2201 	if (ratio)
2202 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2203 			ratio, bclk, ratio * bclk, group_size);
2204 
2205 	ratio = (msr >> 16) & 0xFF;
2206 	group_size = (core_counts >> 16) & 0xFF;
2207 	if (ratio)
2208 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2209 			ratio, bclk, ratio * bclk, group_size);
2210 
2211 	ratio = (msr >> 8) & 0xFF;
2212 	group_size = (core_counts >> 8) & 0xFF;
2213 	if (ratio)
2214 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2215 			ratio, bclk, ratio * bclk, group_size);
2216 
2217 	ratio = (msr >> 0) & 0xFF;
2218 	group_size = (core_counts >> 0) & 0xFF;
2219 	if (ratio)
2220 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2221 			ratio, bclk, ratio * bclk, group_size);
2222 	return;
2223 }
2224 
2225 static void
2226 dump_atom_turbo_ratio_limits(void)
2227 {
2228 	unsigned long long msr;
2229 	unsigned int ratio;
2230 
2231 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
2232 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2233 
2234 	ratio = (msr >> 0) & 0x3F;
2235 	if (ratio)
2236 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
2237 			ratio, bclk, ratio * bclk);
2238 
2239 	ratio = (msr >> 8) & 0x3F;
2240 	if (ratio)
2241 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
2242 			ratio, bclk, ratio * bclk);
2243 
2244 	ratio = (msr >> 16) & 0x3F;
2245 	if (ratio)
2246 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2247 			ratio, bclk, ratio * bclk);
2248 
2249 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
2250 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2251 
2252 	ratio = (msr >> 24) & 0x3F;
2253 	if (ratio)
2254 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
2255 			ratio, bclk, ratio * bclk);
2256 
2257 	ratio = (msr >> 16) & 0x3F;
2258 	if (ratio)
2259 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
2260 			ratio, bclk, ratio * bclk);
2261 
2262 	ratio = (msr >> 8) & 0x3F;
2263 	if (ratio)
2264 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
2265 			ratio, bclk, ratio * bclk);
2266 
2267 	ratio = (msr >> 0) & 0x3F;
2268 	if (ratio)
2269 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
2270 			ratio, bclk, ratio * bclk);
2271 }
2272 
2273 static void
2274 dump_knl_turbo_ratio_limits(void)
2275 {
2276 	const unsigned int buckets_no = 7;
2277 
2278 	unsigned long long msr;
2279 	int delta_cores, delta_ratio;
2280 	int i, b_nr;
2281 	unsigned int cores[buckets_no];
2282 	unsigned int ratio[buckets_no];
2283 
2284 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2285 
2286 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
2287 		base_cpu, msr);
2288 
2289 	/**
2290 	 * Turbo encoding in KNL is as follows:
2291 	 * [0] -- Reserved
2292 	 * [7:1] -- Base value of number of active cores of bucket 1.
2293 	 * [15:8] -- Base value of freq ratio of bucket 1.
2294 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
2295 	 * i.e. active cores of bucket 2 =
2296 	 * active cores of bucket 1 + delta
2297 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
2298 	 * i.e. freq ratio of bucket 2 =
2299 	 * freq ratio of bucket 1 - delta
2300 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
2301 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
2302 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
2303 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
2304 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
2305 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
2306 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
2307 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
2308 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
2309 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
2310 	 */
2311 
2312 	b_nr = 0;
2313 	cores[b_nr] = (msr & 0xFF) >> 1;
2314 	ratio[b_nr] = (msr >> 8) & 0xFF;
2315 
2316 	for (i = 16; i < 64; i += 8) {
2317 		delta_cores = (msr >> i) & 0x1F;
2318 		delta_ratio = (msr >> (i + 5)) & 0x7;
2319 
2320 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
2321 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2322 		b_nr++;
2323 	}
2324 
2325 	for (i = buckets_no - 1; i >= 0; i--)
2326 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2327 			fprintf(outf,
2328 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
2329 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
2330 }
2331 
2332 static void
2333 dump_nhm_cst_cfg(void)
2334 {
2335 	unsigned long long msr;
2336 
2337 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2338 
2339 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2340 
2341 	fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)",
2342 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2343 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2344 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2345 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2346 		(msr & (1 << 15)) ? "" : "UN",
2347 		(unsigned int)msr & 0xF,
2348 		pkg_cstate_limit_strings[pkg_cstate_limit]);
2349 
2350 #define AUTOMATIC_CSTATE_CONVERSION		(1UL << 16)
2351 	if (has_automatic_cstate_conversion) {
2352 		fprintf(outf, ", automatic c-state conversion=%s",
2353 			(msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off");
2354 	}
2355 
2356 	fprintf(outf, ")\n");
2357 
2358 	return;
2359 }
2360 
2361 static void
2362 dump_config_tdp(void)
2363 {
2364 	unsigned long long msr;
2365 
2366 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2367 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2368 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2369 
2370 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2371 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2372 	if (msr) {
2373 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2374 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2375 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2376 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2377 	}
2378 	fprintf(outf, ")\n");
2379 
2380 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2381 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2382 	if (msr) {
2383 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2384 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2385 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2386 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2387 	}
2388 	fprintf(outf, ")\n");
2389 
2390 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2391 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2392 	if ((msr) & 0x3)
2393 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2394 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2395 	fprintf(outf, ")\n");
2396 
2397 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2398 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2399 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2400 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2401 	fprintf(outf, ")\n");
2402 }
2403 
2404 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2405 
2406 void print_irtl(void)
2407 {
2408 	unsigned long long msr;
2409 
2410 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2411 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2412 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2413 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2414 
2415 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2416 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2417 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2418 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2419 
2420 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2421 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2422 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2423 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2424 
2425 	if (!do_irtl_hsw)
2426 		return;
2427 
2428 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2429 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2430 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2431 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2432 
2433 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2434 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2435 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2436 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2437 
2438 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2439 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2440 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2441 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2442 
2443 }
2444 void free_fd_percpu(void)
2445 {
2446 	int i;
2447 
2448 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2449 		if (fd_percpu[i] != 0)
2450 			close(fd_percpu[i]);
2451 	}
2452 
2453 	free(fd_percpu);
2454 }
2455 
2456 void free_all_buffers(void)
2457 {
2458 	int i;
2459 
2460 	CPU_FREE(cpu_present_set);
2461 	cpu_present_set = NULL;
2462 	cpu_present_setsize = 0;
2463 
2464 	CPU_FREE(cpu_affinity_set);
2465 	cpu_affinity_set = NULL;
2466 	cpu_affinity_setsize = 0;
2467 
2468 	free(thread_even);
2469 	free(core_even);
2470 	free(package_even);
2471 
2472 	thread_even = NULL;
2473 	core_even = NULL;
2474 	package_even = NULL;
2475 
2476 	free(thread_odd);
2477 	free(core_odd);
2478 	free(package_odd);
2479 
2480 	thread_odd = NULL;
2481 	core_odd = NULL;
2482 	package_odd = NULL;
2483 
2484 	free(output_buffer);
2485 	output_buffer = NULL;
2486 	outp = NULL;
2487 
2488 	free_fd_percpu();
2489 
2490 	free(irq_column_2_cpu);
2491 	free(irqs_per_cpu);
2492 
2493 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2494 		if (cpus[i].put_ids)
2495 			CPU_FREE(cpus[i].put_ids);
2496 	}
2497 	free(cpus);
2498 }
2499 
2500 
2501 /*
2502  * Parse a file containing a single int.
2503  * Return 0 if file can not be opened
2504  * Exit if file can be opened, but can not be parsed
2505  */
2506 int parse_int_file(const char *fmt, ...)
2507 {
2508 	va_list args;
2509 	char path[PATH_MAX];
2510 	FILE *filep;
2511 	int value;
2512 
2513 	va_start(args, fmt);
2514 	vsnprintf(path, sizeof(path), fmt, args);
2515 	va_end(args);
2516 	filep = fopen(path, "r");
2517 	if (!filep)
2518 		return 0;
2519 	if (fscanf(filep, "%d", &value) != 1)
2520 		err(1, "%s: failed to parse number from file", path);
2521 	fclose(filep);
2522 	return value;
2523 }
2524 
2525 /*
2526  * cpu_is_first_core_in_package(cpu)
2527  * return 1 if given CPU is 1st core in package
2528  */
2529 int cpu_is_first_core_in_package(int cpu)
2530 {
2531 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2532 }
2533 
2534 int get_physical_package_id(int cpu)
2535 {
2536 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2537 }
2538 
2539 int get_die_id(int cpu)
2540 {
2541 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu);
2542 }
2543 
2544 int get_core_id(int cpu)
2545 {
2546 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2547 }
2548 
2549 void set_node_data(void)
2550 {
2551 	int pkg, node, lnode, cpu, cpux;
2552 	int cpu_count;
2553 
2554 	/* initialize logical_node_id */
2555 	for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu)
2556 		cpus[cpu].logical_node_id = -1;
2557 
2558 	cpu_count = 0;
2559 	for (pkg = 0; pkg < topo.num_packages; pkg++) {
2560 		lnode = 0;
2561 		for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) {
2562 			if (cpus[cpu].physical_package_id != pkg)
2563 				continue;
2564 			/* find a cpu with an unset logical_node_id */
2565 			if (cpus[cpu].logical_node_id != -1)
2566 				continue;
2567 			cpus[cpu].logical_node_id = lnode;
2568 			node = cpus[cpu].physical_node_id;
2569 			cpu_count++;
2570 			/*
2571 			 * find all matching cpus on this pkg and set
2572 			 * the logical_node_id
2573 			 */
2574 			for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) {
2575 				if ((cpus[cpux].physical_package_id == pkg) &&
2576 				   (cpus[cpux].physical_node_id == node)) {
2577 					cpus[cpux].logical_node_id = lnode;
2578 					cpu_count++;
2579 				}
2580 			}
2581 			lnode++;
2582 			if (lnode > topo.nodes_per_pkg)
2583 				topo.nodes_per_pkg = lnode;
2584 		}
2585 		if (cpu_count >= topo.max_cpu_num)
2586 			break;
2587 	}
2588 }
2589 
2590 int get_physical_node_id(struct cpu_topology *thiscpu)
2591 {
2592 	char path[80];
2593 	FILE *filep;
2594 	int i;
2595 	int cpu = thiscpu->logical_cpu_id;
2596 
2597 	for (i = 0; i <= topo.max_cpu_num; i++) {
2598 		sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist",
2599 			cpu, i);
2600 		filep = fopen(path, "r");
2601 		if (!filep)
2602 			continue;
2603 		fclose(filep);
2604 		return i;
2605 	}
2606 	return -1;
2607 }
2608 
2609 int get_thread_siblings(struct cpu_topology *thiscpu)
2610 {
2611 	char path[80], character;
2612 	FILE *filep;
2613 	unsigned long map;
2614 	int so, shift, sib_core;
2615 	int cpu = thiscpu->logical_cpu_id;
2616 	int offset = topo.max_cpu_num + 1;
2617 	size_t size;
2618 	int thread_id = 0;
2619 
2620 	thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1));
2621 	if (thiscpu->thread_id < 0)
2622 		thiscpu->thread_id = thread_id++;
2623 	if (!thiscpu->put_ids)
2624 		return -1;
2625 
2626 	size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2627 	CPU_ZERO_S(size, thiscpu->put_ids);
2628 
2629 	sprintf(path,
2630 		"/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu);
2631 	filep = fopen_or_die(path, "r");
2632 	do {
2633 		offset -= BITMASK_SIZE;
2634 		if (fscanf(filep, "%lx%c", &map, &character) != 2)
2635 			err(1, "%s: failed to parse file", path);
2636 		for (shift = 0; shift < BITMASK_SIZE; shift++) {
2637 			if ((map >> shift) & 0x1) {
2638 				so = shift + offset;
2639 				sib_core = get_core_id(so);
2640 				if (sib_core == thiscpu->physical_core_id) {
2641 					CPU_SET_S(so, size, thiscpu->put_ids);
2642 					if ((so != cpu) &&
2643 					    (cpus[so].thread_id < 0))
2644 						cpus[so].thread_id =
2645 								    thread_id++;
2646 				}
2647 			}
2648 		}
2649 	} while (!strncmp(&character, ",", 1));
2650 	fclose(filep);
2651 
2652 	return CPU_COUNT_S(size, thiscpu->put_ids);
2653 }
2654 
2655 /*
2656  * run func(thread, core, package) in topology order
2657  * skip non-present cpus
2658  */
2659 
2660 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2661 	struct pkg_data *, struct thread_data *, struct core_data *,
2662 	struct pkg_data *), struct thread_data *thread_base,
2663 	struct core_data *core_base, struct pkg_data *pkg_base,
2664 	struct thread_data *thread_base2, struct core_data *core_base2,
2665 	struct pkg_data *pkg_base2)
2666 {
2667 	int retval, pkg_no, node_no, core_no, thread_no;
2668 
2669 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2670 		for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) {
2671 			for (core_no = 0; core_no < topo.cores_per_node;
2672 			     ++core_no) {
2673 				for (thread_no = 0; thread_no <
2674 					topo.threads_per_core; ++thread_no) {
2675 					struct thread_data *t, *t2;
2676 					struct core_data *c, *c2;
2677 					struct pkg_data *p, *p2;
2678 
2679 					t = GET_THREAD(thread_base, thread_no,
2680 						       core_no, node_no,
2681 						       pkg_no);
2682 
2683 					if (cpu_is_not_present(t->cpu_id))
2684 						continue;
2685 
2686 					t2 = GET_THREAD(thread_base2, thread_no,
2687 							core_no, node_no,
2688 							pkg_no);
2689 
2690 					c = GET_CORE(core_base, core_no,
2691 						     node_no, pkg_no);
2692 					c2 = GET_CORE(core_base2, core_no,
2693 						      node_no,
2694 						      pkg_no);
2695 
2696 					p = GET_PKG(pkg_base, pkg_no);
2697 					p2 = GET_PKG(pkg_base2, pkg_no);
2698 
2699 					retval = func(t, c, p, t2, c2, p2);
2700 					if (retval)
2701 						return retval;
2702 				}
2703 			}
2704 		}
2705 	}
2706 	return 0;
2707 }
2708 
2709 /*
2710  * run func(cpu) on every cpu in /proc/stat
2711  * return max_cpu number
2712  */
2713 int for_all_proc_cpus(int (func)(int))
2714 {
2715 	FILE *fp;
2716 	int cpu_num;
2717 	int retval;
2718 
2719 	fp = fopen_or_die(proc_stat, "r");
2720 
2721 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2722 	if (retval != 0)
2723 		err(1, "%s: failed to parse format", proc_stat);
2724 
2725 	while (1) {
2726 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2727 		if (retval != 1)
2728 			break;
2729 
2730 		retval = func(cpu_num);
2731 		if (retval) {
2732 			fclose(fp);
2733 			return(retval);
2734 		}
2735 	}
2736 	fclose(fp);
2737 	return 0;
2738 }
2739 
2740 void re_initialize(void)
2741 {
2742 	free_all_buffers();
2743 	setup_all_buffers();
2744 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2745 }
2746 
2747 void set_max_cpu_num(void)
2748 {
2749 	FILE *filep;
2750 	unsigned long dummy;
2751 
2752 	topo.max_cpu_num = 0;
2753 	filep = fopen_or_die(
2754 			"/sys/devices/system/cpu/cpu0/topology/thread_siblings",
2755 			"r");
2756 	while (fscanf(filep, "%lx,", &dummy) == 1)
2757 		topo.max_cpu_num += BITMASK_SIZE;
2758 	fclose(filep);
2759 	topo.max_cpu_num--; /* 0 based */
2760 }
2761 
2762 /*
2763  * count_cpus()
2764  * remember the last one seen, it will be the max
2765  */
2766 int count_cpus(int cpu)
2767 {
2768 	topo.num_cpus++;
2769 	return 0;
2770 }
2771 int mark_cpu_present(int cpu)
2772 {
2773 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2774 	return 0;
2775 }
2776 
2777 int init_thread_id(int cpu)
2778 {
2779 	cpus[cpu].thread_id = -1;
2780 	return 0;
2781 }
2782 
2783 /*
2784  * snapshot_proc_interrupts()
2785  *
2786  * read and record summary of /proc/interrupts
2787  *
2788  * return 1 if config change requires a restart, else return 0
2789  */
2790 int snapshot_proc_interrupts(void)
2791 {
2792 	static FILE *fp;
2793 	int column, retval;
2794 
2795 	if (fp == NULL)
2796 		fp = fopen_or_die("/proc/interrupts", "r");
2797 	else
2798 		rewind(fp);
2799 
2800 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
2801 	for (column = 0; column < topo.num_cpus; ++column) {
2802 		int cpu_number;
2803 
2804 		retval = fscanf(fp, " CPU%d", &cpu_number);
2805 		if (retval != 1)
2806 			break;
2807 
2808 		if (cpu_number > topo.max_cpu_num) {
2809 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2810 			return 1;
2811 		}
2812 
2813 		irq_column_2_cpu[column] = cpu_number;
2814 		irqs_per_cpu[cpu_number] = 0;
2815 	}
2816 
2817 	/* read /proc/interrupt count lines and sum up irqs per cpu */
2818 	while (1) {
2819 		int column;
2820 		char buf[64];
2821 
2822 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
2823 		if (retval != 1)
2824 			break;
2825 
2826 		/* read the count per cpu */
2827 		for (column = 0; column < topo.num_cpus; ++column) {
2828 
2829 			int cpu_number, irq_count;
2830 
2831 			retval = fscanf(fp, " %d", &irq_count);
2832 			if (retval != 1)
2833 				break;
2834 
2835 			cpu_number = irq_column_2_cpu[column];
2836 			irqs_per_cpu[cpu_number] += irq_count;
2837 
2838 		}
2839 
2840 		while (getc(fp) != '\n')
2841 			;	/* flush interrupt description */
2842 
2843 	}
2844 	return 0;
2845 }
2846 /*
2847  * snapshot_gfx_rc6_ms()
2848  *
2849  * record snapshot of
2850  * /sys/class/drm/card0/power/rc6_residency_ms
2851  *
2852  * return 1 if config change requires a restart, else return 0
2853  */
2854 int snapshot_gfx_rc6_ms(void)
2855 {
2856 	FILE *fp;
2857 	int retval;
2858 
2859 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2860 
2861 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2862 	if (retval != 1)
2863 		err(1, "GFX rc6");
2864 
2865 	fclose(fp);
2866 
2867 	return 0;
2868 }
2869 /*
2870  * snapshot_gfx_mhz()
2871  *
2872  * record snapshot of
2873  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2874  *
2875  * return 1 if config change requires a restart, else return 0
2876  */
2877 int snapshot_gfx_mhz(void)
2878 {
2879 	static FILE *fp;
2880 	int retval;
2881 
2882 	if (fp == NULL)
2883 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2884 	else {
2885 		rewind(fp);
2886 		fflush(fp);
2887 	}
2888 
2889 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
2890 	if (retval != 1)
2891 		err(1, "GFX MHz");
2892 
2893 	return 0;
2894 }
2895 
2896 /*
2897  * snapshot_cpu_lpi()
2898  *
2899  * record snapshot of
2900  * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us
2901  *
2902  * return 1 if config change requires a restart, else return 0
2903  */
2904 int snapshot_cpu_lpi_us(void)
2905 {
2906 	FILE *fp;
2907 	int retval;
2908 
2909 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r");
2910 
2911 	retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us);
2912 	if (retval != 1) {
2913 		fprintf(stderr, "Disabling Low Power Idle CPU output\n");
2914 		BIC_NOT_PRESENT(BIC_CPU_LPI);
2915 		return -1;
2916 	}
2917 
2918 	fclose(fp);
2919 
2920 	return 0;
2921 }
2922 /*
2923  * snapshot_sys_lpi()
2924  *
2925  * record snapshot of
2926  * /sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us
2927  *
2928  * return 1 if config change requires a restart, else return 0
2929  */
2930 int snapshot_sys_lpi_us(void)
2931 {
2932 	FILE *fp;
2933 	int retval;
2934 
2935 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", "r");
2936 
2937 	retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us);
2938 	if (retval != 1) {
2939 		fprintf(stderr, "Disabling Low Power Idle System output\n");
2940 		BIC_NOT_PRESENT(BIC_SYS_LPI);
2941 		return -1;
2942 	}
2943 	fclose(fp);
2944 
2945 	return 0;
2946 }
2947 /*
2948  * snapshot /proc and /sys files
2949  *
2950  * return 1 if configuration restart needed, else return 0
2951  */
2952 int snapshot_proc_sysfs_files(void)
2953 {
2954 	if (DO_BIC(BIC_IRQ))
2955 		if (snapshot_proc_interrupts())
2956 			return 1;
2957 
2958 	if (DO_BIC(BIC_GFX_rc6))
2959 		snapshot_gfx_rc6_ms();
2960 
2961 	if (DO_BIC(BIC_GFXMHz))
2962 		snapshot_gfx_mhz();
2963 
2964 	if (DO_BIC(BIC_CPU_LPI))
2965 		snapshot_cpu_lpi_us();
2966 
2967 	if (DO_BIC(BIC_SYS_LPI))
2968 		snapshot_sys_lpi_us();
2969 
2970 	return 0;
2971 }
2972 
2973 int exit_requested;
2974 
2975 static void signal_handler (int signal)
2976 {
2977 	switch (signal) {
2978 	case SIGINT:
2979 		exit_requested = 1;
2980 		if (debug)
2981 			fprintf(stderr, " SIGINT\n");
2982 		break;
2983 	case SIGUSR1:
2984 		if (debug > 1)
2985 			fprintf(stderr, "SIGUSR1\n");
2986 		break;
2987 	}
2988 	/* make sure this manually-invoked interval is at least 1ms long */
2989 	nanosleep(&one_msec, NULL);
2990 }
2991 
2992 void setup_signal_handler(void)
2993 {
2994 	struct sigaction sa;
2995 
2996 	memset(&sa, 0, sizeof(sa));
2997 
2998 	sa.sa_handler = &signal_handler;
2999 
3000 	if (sigaction(SIGINT, &sa, NULL) < 0)
3001 		err(1, "sigaction SIGINT");
3002 	if (sigaction(SIGUSR1, &sa, NULL) < 0)
3003 		err(1, "sigaction SIGUSR1");
3004 }
3005 
3006 void do_sleep(void)
3007 {
3008 	struct timeval select_timeout;
3009 	fd_set readfds;
3010 	int retval;
3011 
3012 	FD_ZERO(&readfds);
3013 	FD_SET(0, &readfds);
3014 
3015 	if (!isatty(fileno(stdin))) {
3016 		nanosleep(&interval_ts, NULL);
3017 		return;
3018 	}
3019 
3020 	select_timeout = interval_tv;
3021 	retval = select(1, &readfds, NULL, NULL, &select_timeout);
3022 
3023 	if (retval == 1) {
3024 		switch (getc(stdin)) {
3025 		case 'q':
3026 			exit_requested = 1;
3027 			break;
3028 		}
3029 		/* make sure this manually-invoked interval is at least 1ms long */
3030 		nanosleep(&one_msec, NULL);
3031 	}
3032 }
3033 
3034 
3035 void turbostat_loop()
3036 {
3037 	int retval;
3038 	int restarted = 0;
3039 	int done_iters = 0;
3040 
3041 	setup_signal_handler();
3042 
3043 restart:
3044 	restarted++;
3045 
3046 	snapshot_proc_sysfs_files();
3047 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3048 	first_counter_read = 0;
3049 	if (retval < -1) {
3050 		exit(retval);
3051 	} else if (retval == -1) {
3052 		if (restarted > 1) {
3053 			exit(retval);
3054 		}
3055 		re_initialize();
3056 		goto restart;
3057 	}
3058 	restarted = 0;
3059 	done_iters = 0;
3060 	gettimeofday(&tv_even, (struct timezone *)NULL);
3061 
3062 	while (1) {
3063 		if (for_all_proc_cpus(cpu_is_not_present)) {
3064 			re_initialize();
3065 			goto restart;
3066 		}
3067 		do_sleep();
3068 		if (snapshot_proc_sysfs_files())
3069 			goto restart;
3070 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
3071 		if (retval < -1) {
3072 			exit(retval);
3073 		} else if (retval == -1) {
3074 			re_initialize();
3075 			goto restart;
3076 		}
3077 		gettimeofday(&tv_odd, (struct timezone *)NULL);
3078 		timersub(&tv_odd, &tv_even, &tv_delta);
3079 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
3080 			re_initialize();
3081 			goto restart;
3082 		}
3083 		compute_average(EVEN_COUNTERS);
3084 		format_all_counters(EVEN_COUNTERS);
3085 		flush_output_stdout();
3086 		if (exit_requested)
3087 			break;
3088 		if (num_iterations && ++done_iters >= num_iterations)
3089 			break;
3090 		do_sleep();
3091 		if (snapshot_proc_sysfs_files())
3092 			goto restart;
3093 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3094 		if (retval < -1) {
3095 			exit(retval);
3096 		} else if (retval == -1) {
3097 			re_initialize();
3098 			goto restart;
3099 		}
3100 		gettimeofday(&tv_even, (struct timezone *)NULL);
3101 		timersub(&tv_even, &tv_odd, &tv_delta);
3102 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
3103 			re_initialize();
3104 			goto restart;
3105 		}
3106 		compute_average(ODD_COUNTERS);
3107 		format_all_counters(ODD_COUNTERS);
3108 		flush_output_stdout();
3109 		if (exit_requested)
3110 			break;
3111 		if (num_iterations && ++done_iters >= num_iterations)
3112 			break;
3113 	}
3114 }
3115 
3116 void check_dev_msr()
3117 {
3118 	struct stat sb;
3119 	char pathname[32];
3120 
3121 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3122 	if (stat(pathname, &sb))
3123  		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
3124 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
3125 }
3126 
3127 void check_permissions()
3128 {
3129 	struct __user_cap_header_struct cap_header_data;
3130 	cap_user_header_t cap_header = &cap_header_data;
3131 	struct __user_cap_data_struct cap_data_data;
3132 	cap_user_data_t cap_data = &cap_data_data;
3133 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
3134 	int do_exit = 0;
3135 	char pathname[32];
3136 
3137 	/* check for CAP_SYS_RAWIO */
3138 	cap_header->pid = getpid();
3139 	cap_header->version = _LINUX_CAPABILITY_VERSION;
3140 	if (capget(cap_header, cap_data) < 0)
3141 		err(-6, "capget(2) failed");
3142 
3143 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
3144 		do_exit++;
3145 		warnx("capget(CAP_SYS_RAWIO) failed,"
3146 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
3147 	}
3148 
3149 	/* test file permissions */
3150 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3151 	if (euidaccess(pathname, R_OK)) {
3152 		do_exit++;
3153 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
3154 	}
3155 
3156 	/* if all else fails, thell them to be root */
3157 	if (do_exit)
3158 		if (getuid() != 0)
3159 			warnx("... or simply run as root");
3160 
3161 	if (do_exit)
3162 		exit(-6);
3163 }
3164 
3165 /*
3166  * NHM adds support for additional MSRs:
3167  *
3168  * MSR_SMI_COUNT                   0x00000034
3169  *
3170  * MSR_PLATFORM_INFO               0x000000ce
3171  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
3172  *
3173  * MSR_MISC_PWR_MGMT               0x000001aa
3174  *
3175  * MSR_PKG_C3_RESIDENCY            0x000003f8
3176  * MSR_PKG_C6_RESIDENCY            0x000003f9
3177  * MSR_CORE_C3_RESIDENCY           0x000003fc
3178  * MSR_CORE_C6_RESIDENCY           0x000003fd
3179  *
3180  * Side effect:
3181  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
3182  * sets has_misc_feature_control
3183  */
3184 int probe_nhm_msrs(unsigned int family, unsigned int model)
3185 {
3186 	unsigned long long msr;
3187 	unsigned int base_ratio;
3188 	int *pkg_cstate_limits;
3189 
3190 	if (!genuine_intel)
3191 		return 0;
3192 
3193 	if (family != 6)
3194 		return 0;
3195 
3196 	bclk = discover_bclk(family, model);
3197 
3198 	switch (model) {
3199 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
3200 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3201 		pkg_cstate_limits = nhm_pkg_cstate_limits;
3202 		break;
3203 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
3204 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
3205 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3206 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3207 		pkg_cstate_limits = snb_pkg_cstate_limits;
3208 		has_misc_feature_control = 1;
3209 		break;
3210 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3211 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3212 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3213 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3214 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3215 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3216 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3217 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3218 		pkg_cstate_limits = hsw_pkg_cstate_limits;
3219 		has_misc_feature_control = 1;
3220 		break;
3221 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3222 		pkg_cstate_limits = skx_pkg_cstate_limits;
3223 		has_misc_feature_control = 1;
3224 		break;
3225 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3226 		no_MSR_MISC_PWR_MGMT = 1;
3227 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3228 		pkg_cstate_limits = slv_pkg_cstate_limits;
3229 		break;
3230 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
3231 		pkg_cstate_limits = amt_pkg_cstate_limits;
3232 		no_MSR_MISC_PWR_MGMT = 1;
3233 		break;
3234 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
3235 		pkg_cstate_limits = phi_pkg_cstate_limits;
3236 		break;
3237 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3238 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3239 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3240 		pkg_cstate_limits = glm_pkg_cstate_limits;
3241 		break;
3242 	default:
3243 		return 0;
3244 	}
3245 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
3246 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
3247 
3248 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
3249 	base_ratio = (msr >> 8) & 0xFF;
3250 
3251 	base_hz = base_ratio * bclk * 1000000;
3252 	has_base_hz = 1;
3253 	return 1;
3254 }
3255 /*
3256  * SLV client has support for unique MSRs:
3257  *
3258  * MSR_CC6_DEMOTION_POLICY_CONFIG
3259  * MSR_MC6_DEMOTION_POLICY_CONFIG
3260  */
3261 
3262 int has_slv_msrs(unsigned int family, unsigned int model)
3263 {
3264 	if (!genuine_intel)
3265 		return 0;
3266 
3267 	switch (model) {
3268 	case INTEL_FAM6_ATOM_SILVERMONT:
3269 	case INTEL_FAM6_ATOM_SILVERMONT_MID:
3270 	case INTEL_FAM6_ATOM_AIRMONT_MID:
3271 		return 1;
3272 	}
3273 	return 0;
3274 }
3275 int is_dnv(unsigned int family, unsigned int model)
3276 {
3277 
3278 	if (!genuine_intel)
3279 		return 0;
3280 
3281 	switch (model) {
3282 	case INTEL_FAM6_ATOM_GOLDMONT_X:
3283 		return 1;
3284 	}
3285 	return 0;
3286 }
3287 int is_bdx(unsigned int family, unsigned int model)
3288 {
3289 
3290 	if (!genuine_intel)
3291 		return 0;
3292 
3293 	switch (model) {
3294 	case INTEL_FAM6_BROADWELL_X:
3295 		return 1;
3296 	}
3297 	return 0;
3298 }
3299 int is_skx(unsigned int family, unsigned int model)
3300 {
3301 
3302 	if (!genuine_intel)
3303 		return 0;
3304 
3305 	switch (model) {
3306 	case INTEL_FAM6_SKYLAKE_X:
3307 		return 1;
3308 	}
3309 	return 0;
3310 }
3311 
3312 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
3313 {
3314 	if (has_slv_msrs(family, model))
3315 		return 0;
3316 
3317 	switch (model) {
3318 	/* Nehalem compatible, but do not include turbo-ratio limit support */
3319 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3320 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
3321 		return 0;
3322 	default:
3323 		return 1;
3324 	}
3325 }
3326 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
3327 {
3328 	if (has_slv_msrs(family, model))
3329 		return 1;
3330 
3331 	return 0;
3332 }
3333 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
3334 {
3335 	if (!genuine_intel)
3336 		return 0;
3337 
3338 	if (family != 6)
3339 		return 0;
3340 
3341 	switch (model) {
3342 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3343 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3344 		return 1;
3345 	default:
3346 		return 0;
3347 	}
3348 }
3349 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
3350 {
3351 	if (!genuine_intel)
3352 		return 0;
3353 
3354 	if (family != 6)
3355 		return 0;
3356 
3357 	switch (model) {
3358 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3359 		return 1;
3360 	default:
3361 		return 0;
3362 	}
3363 }
3364 
3365 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
3366 {
3367 	if (!genuine_intel)
3368 		return 0;
3369 
3370 	if (family != 6)
3371 		return 0;
3372 
3373 	switch (model) {
3374 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3375 		return 1;
3376 	default:
3377 		return 0;
3378 	}
3379 }
3380 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
3381 {
3382 	if (!genuine_intel)
3383 		return 0;
3384 
3385 	if (family != 6)
3386 		return 0;
3387 
3388 	switch (model) {
3389 	case INTEL_FAM6_ATOM_GOLDMONT:
3390 	case INTEL_FAM6_SKYLAKE_X:
3391 		return 1;
3392 	default:
3393 		return 0;
3394 	}
3395 }
3396 int has_config_tdp(unsigned int family, unsigned int model)
3397 {
3398 	if (!genuine_intel)
3399 		return 0;
3400 
3401 	if (family != 6)
3402 		return 0;
3403 
3404 	switch (model) {
3405 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3406 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3407 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3408 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3409 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3410 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3411 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3412 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3413 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3414 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3415 
3416 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3417 		return 1;
3418 	default:
3419 		return 0;
3420 	}
3421 }
3422 
3423 static void
3424 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
3425 {
3426 	if (!do_nhm_platform_info)
3427 		return;
3428 
3429 	dump_nhm_platform_info();
3430 
3431 	if (has_hsw_turbo_ratio_limit(family, model))
3432 		dump_hsw_turbo_ratio_limits();
3433 
3434 	if (has_ivt_turbo_ratio_limit(family, model))
3435 		dump_ivt_turbo_ratio_limits();
3436 
3437 	if (has_turbo_ratio_limit(family, model))
3438 		dump_turbo_ratio_limits(family, model);
3439 
3440 	if (has_atom_turbo_ratio_limit(family, model))
3441 		dump_atom_turbo_ratio_limits();
3442 
3443 	if (has_knl_turbo_ratio_limit(family, model))
3444 		dump_knl_turbo_ratio_limits();
3445 
3446 	if (has_config_tdp(family, model))
3447 		dump_config_tdp();
3448 
3449 	dump_nhm_cst_cfg();
3450 }
3451 
3452 static void
3453 dump_sysfs_cstate_config(void)
3454 {
3455 	char path[64];
3456 	char name_buf[16];
3457 	char desc[64];
3458 	FILE *input;
3459 	int state;
3460 	char *sp;
3461 
3462 	if (!DO_BIC(BIC_sysfs))
3463 		return;
3464 
3465 	for (state = 0; state < 10; ++state) {
3466 
3467 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
3468 			base_cpu, state);
3469 		input = fopen(path, "r");
3470 		if (input == NULL)
3471 			continue;
3472 		if (!fgets(name_buf, sizeof(name_buf), input))
3473 			err(1, "%s: failed to read file", path);
3474 
3475 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
3476 		sp = strchr(name_buf, '-');
3477 		if (!sp)
3478 			sp = strchrnul(name_buf, '\n');
3479 		*sp = '\0';
3480 		fclose(input);
3481 
3482 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc",
3483 			base_cpu, state);
3484 		input = fopen(path, "r");
3485 		if (input == NULL)
3486 			continue;
3487 		if (!fgets(desc, sizeof(desc), input))
3488 			err(1, "%s: failed to read file", path);
3489 
3490 		fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
3491 		fclose(input);
3492 	}
3493 }
3494 static void
3495 dump_sysfs_pstate_config(void)
3496 {
3497 	char path[64];
3498 	char driver_buf[64];
3499 	char governor_buf[64];
3500 	FILE *input;
3501 	int turbo;
3502 
3503 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver",
3504 			base_cpu);
3505 	input = fopen(path, "r");
3506 	if (input == NULL) {
3507 		fprintf(outf, "NSFOD %s\n", path);
3508 		return;
3509 	}
3510 	if (!fgets(driver_buf, sizeof(driver_buf), input))
3511 		err(1, "%s: failed to read file", path);
3512 	fclose(input);
3513 
3514 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor",
3515 			base_cpu);
3516 	input = fopen(path, "r");
3517 	if (input == NULL) {
3518 		fprintf(outf, "NSFOD %s\n", path);
3519 		return;
3520 	}
3521 	if (!fgets(governor_buf, sizeof(governor_buf), input))
3522 		err(1, "%s: failed to read file", path);
3523 	fclose(input);
3524 
3525 	fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
3526 	fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
3527 
3528 	sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
3529 	input = fopen(path, "r");
3530 	if (input != NULL) {
3531 		if (fscanf(input, "%d", &turbo) != 1)
3532 			err(1, "%s: failed to parse number from file", path);
3533 		fprintf(outf, "cpufreq boost: %d\n", turbo);
3534 		fclose(input);
3535 	}
3536 
3537 	sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
3538 	input = fopen(path, "r");
3539 	if (input != NULL) {
3540 		if (fscanf(input, "%d", &turbo) != 1)
3541 			err(1, "%s: failed to parse number from file", path);
3542 		fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
3543 		fclose(input);
3544 	}
3545 }
3546 
3547 
3548 /*
3549  * print_epb()
3550  * Decode the ENERGY_PERF_BIAS MSR
3551  */
3552 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3553 {
3554 	unsigned long long msr;
3555 	char *epb_string;
3556 	int cpu;
3557 
3558 	if (!has_epb)
3559 		return 0;
3560 
3561 	cpu = t->cpu_id;
3562 
3563 	/* EPB is per-package */
3564 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3565 		return 0;
3566 
3567 	if (cpu_migrate(cpu)) {
3568 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3569 		return -1;
3570 	}
3571 
3572 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
3573 		return 0;
3574 
3575 	switch (msr & 0xF) {
3576 	case ENERGY_PERF_BIAS_PERFORMANCE:
3577 		epb_string = "performance";
3578 		break;
3579 	case ENERGY_PERF_BIAS_NORMAL:
3580 		epb_string = "balanced";
3581 		break;
3582 	case ENERGY_PERF_BIAS_POWERSAVE:
3583 		epb_string = "powersave";
3584 		break;
3585 	default:
3586 		epb_string = "custom";
3587 		break;
3588 	}
3589 	fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
3590 
3591 	return 0;
3592 }
3593 /*
3594  * print_hwp()
3595  * Decode the MSR_HWP_CAPABILITIES
3596  */
3597 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3598 {
3599 	unsigned long long msr;
3600 	int cpu;
3601 
3602 	if (!has_hwp)
3603 		return 0;
3604 
3605 	cpu = t->cpu_id;
3606 
3607 	/* MSR_HWP_CAPABILITIES is per-package */
3608 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3609 		return 0;
3610 
3611 	if (cpu_migrate(cpu)) {
3612 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3613 		return -1;
3614 	}
3615 
3616 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
3617 		return 0;
3618 
3619 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
3620 		cpu, msr, (msr & (1 << 0)) ? "" : "No-");
3621 
3622 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
3623 	if ((msr & (1 << 0)) == 0)
3624 		return 0;
3625 
3626 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
3627 		return 0;
3628 
3629 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
3630 			"(high %d guar %d eff %d low %d)\n",
3631 			cpu, msr,
3632 			(unsigned int)HWP_HIGHEST_PERF(msr),
3633 			(unsigned int)HWP_GUARANTEED_PERF(msr),
3634 			(unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
3635 			(unsigned int)HWP_LOWEST_PERF(msr));
3636 
3637 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
3638 		return 0;
3639 
3640 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
3641 			"(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
3642 			cpu, msr,
3643 			(unsigned int)(((msr) >> 0) & 0xff),
3644 			(unsigned int)(((msr) >> 8) & 0xff),
3645 			(unsigned int)(((msr) >> 16) & 0xff),
3646 			(unsigned int)(((msr) >> 24) & 0xff),
3647 			(unsigned int)(((msr) >> 32) & 0xff3),
3648 			(unsigned int)(((msr) >> 42) & 0x1));
3649 
3650 	if (has_hwp_pkg) {
3651 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
3652 			return 0;
3653 
3654 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
3655 			"(min %d max %d des %d epp 0x%x window 0x%x)\n",
3656 			cpu, msr,
3657 			(unsigned int)(((msr) >> 0) & 0xff),
3658 			(unsigned int)(((msr) >> 8) & 0xff),
3659 			(unsigned int)(((msr) >> 16) & 0xff),
3660 			(unsigned int)(((msr) >> 24) & 0xff),
3661 			(unsigned int)(((msr) >> 32) & 0xff3));
3662 	}
3663 	if (has_hwp_notify) {
3664 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
3665 			return 0;
3666 
3667 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
3668 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
3669 			cpu, msr,
3670 			((msr) & 0x1) ? "EN" : "Dis",
3671 			((msr) & 0x2) ? "EN" : "Dis");
3672 	}
3673 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
3674 		return 0;
3675 
3676 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
3677 			"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
3678 			cpu, msr,
3679 			((msr) & 0x1) ? "" : "No-",
3680 			((msr) & 0x2) ? "" : "No-");
3681 
3682 	return 0;
3683 }
3684 
3685 /*
3686  * print_perf_limit()
3687  */
3688 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3689 {
3690 	unsigned long long msr;
3691 	int cpu;
3692 
3693 	cpu = t->cpu_id;
3694 
3695 	/* per-package */
3696 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3697 		return 0;
3698 
3699 	if (cpu_migrate(cpu)) {
3700 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3701 		return -1;
3702 	}
3703 
3704 	if (do_core_perf_limit_reasons) {
3705 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
3706 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3707 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
3708 			(msr & 1 << 15) ? "bit15, " : "",
3709 			(msr & 1 << 14) ? "bit14, " : "",
3710 			(msr & 1 << 13) ? "Transitions, " : "",
3711 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
3712 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
3713 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3714 			(msr & 1 << 9) ? "CorePwr, " : "",
3715 			(msr & 1 << 8) ? "Amps, " : "",
3716 			(msr & 1 << 6) ? "VR-Therm, " : "",
3717 			(msr & 1 << 5) ? "Auto-HWP, " : "",
3718 			(msr & 1 << 4) ? "Graphics, " : "",
3719 			(msr & 1 << 2) ? "bit2, " : "",
3720 			(msr & 1 << 1) ? "ThermStatus, " : "",
3721 			(msr & 1 << 0) ? "PROCHOT, " : "");
3722 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
3723 			(msr & 1 << 31) ? "bit31, " : "",
3724 			(msr & 1 << 30) ? "bit30, " : "",
3725 			(msr & 1 << 29) ? "Transitions, " : "",
3726 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
3727 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
3728 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3729 			(msr & 1 << 25) ? "CorePwr, " : "",
3730 			(msr & 1 << 24) ? "Amps, " : "",
3731 			(msr & 1 << 22) ? "VR-Therm, " : "",
3732 			(msr & 1 << 21) ? "Auto-HWP, " : "",
3733 			(msr & 1 << 20) ? "Graphics, " : "",
3734 			(msr & 1 << 18) ? "bit18, " : "",
3735 			(msr & 1 << 17) ? "ThermStatus, " : "",
3736 			(msr & 1 << 16) ? "PROCHOT, " : "");
3737 
3738 	}
3739 	if (do_gfx_perf_limit_reasons) {
3740 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
3741 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3742 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
3743 			(msr & 1 << 0) ? "PROCHOT, " : "",
3744 			(msr & 1 << 1) ? "ThermStatus, " : "",
3745 			(msr & 1 << 4) ? "Graphics, " : "",
3746 			(msr & 1 << 6) ? "VR-Therm, " : "",
3747 			(msr & 1 << 8) ? "Amps, " : "",
3748 			(msr & 1 << 9) ? "GFXPwr, " : "",
3749 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3750 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3751 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
3752 			(msr & 1 << 16) ? "PROCHOT, " : "",
3753 			(msr & 1 << 17) ? "ThermStatus, " : "",
3754 			(msr & 1 << 20) ? "Graphics, " : "",
3755 			(msr & 1 << 22) ? "VR-Therm, " : "",
3756 			(msr & 1 << 24) ? "Amps, " : "",
3757 			(msr & 1 << 25) ? "GFXPwr, " : "",
3758 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3759 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3760 	}
3761 	if (do_ring_perf_limit_reasons) {
3762 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
3763 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3764 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
3765 			(msr & 1 << 0) ? "PROCHOT, " : "",
3766 			(msr & 1 << 1) ? "ThermStatus, " : "",
3767 			(msr & 1 << 6) ? "VR-Therm, " : "",
3768 			(msr & 1 << 8) ? "Amps, " : "",
3769 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3770 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3771 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3772 			(msr & 1 << 16) ? "PROCHOT, " : "",
3773 			(msr & 1 << 17) ? "ThermStatus, " : "",
3774 			(msr & 1 << 22) ? "VR-Therm, " : "",
3775 			(msr & 1 << 24) ? "Amps, " : "",
3776 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3777 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3778 	}
3779 	return 0;
3780 }
3781 
3782 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
3783 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
3784 
3785 double get_tdp_intel(unsigned int model)
3786 {
3787 	unsigned long long msr;
3788 
3789 	if (do_rapl & RAPL_PKG_POWER_INFO)
3790 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3791 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3792 
3793 	switch (model) {
3794 	case INTEL_FAM6_ATOM_SILVERMONT:
3795 	case INTEL_FAM6_ATOM_SILVERMONT_X:
3796 		return 30.0;
3797 	default:
3798 		return 135.0;
3799 	}
3800 }
3801 
3802 double get_tdp_amd(unsigned int family)
3803 {
3804 	switch (family) {
3805 	case 0x17:
3806 	default:
3807 		/* This is the max stock TDP of HEDT/Server Fam17h chips */
3808 		return 250.0;
3809 	}
3810 }
3811 
3812 /*
3813  * rapl_dram_energy_units_probe()
3814  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3815  */
3816 static double
3817 rapl_dram_energy_units_probe(int  model, double rapl_energy_units)
3818 {
3819 	/* only called for genuine_intel, family 6 */
3820 
3821 	switch (model) {
3822 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3823 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3824 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3825 		return (rapl_dram_energy_units = 15.3 / 1000000);
3826 	default:
3827 		return (rapl_energy_units);
3828 	}
3829 }
3830 
3831 void rapl_probe_intel(unsigned int family, unsigned int model)
3832 {
3833 	unsigned long long msr;
3834 	unsigned int time_unit;
3835 	double tdp;
3836 
3837 	if (family != 6)
3838 		return;
3839 
3840 	switch (model) {
3841 	case INTEL_FAM6_SANDYBRIDGE:
3842 	case INTEL_FAM6_IVYBRIDGE:
3843 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3844 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3845 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3846 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3847 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3848 		if (rapl_joules) {
3849 			BIC_PRESENT(BIC_Pkg_J);
3850 			BIC_PRESENT(BIC_Cor_J);
3851 			BIC_PRESENT(BIC_GFX_J);
3852 		} else {
3853 			BIC_PRESENT(BIC_PkgWatt);
3854 			BIC_PRESENT(BIC_CorWatt);
3855 			BIC_PRESENT(BIC_GFXWatt);
3856 		}
3857 		break;
3858 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3859 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3860 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3861 		if (rapl_joules)
3862 			BIC_PRESENT(BIC_Pkg_J);
3863 		else
3864 			BIC_PRESENT(BIC_PkgWatt);
3865 		break;
3866 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3867 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3868 		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;
3869 		BIC_PRESENT(BIC_PKG__);
3870 		BIC_PRESENT(BIC_RAM__);
3871 		if (rapl_joules) {
3872 			BIC_PRESENT(BIC_Pkg_J);
3873 			BIC_PRESENT(BIC_Cor_J);
3874 			BIC_PRESENT(BIC_RAM_J);
3875 			BIC_PRESENT(BIC_GFX_J);
3876 		} else {
3877 			BIC_PRESENT(BIC_PkgWatt);
3878 			BIC_PRESENT(BIC_CorWatt);
3879 			BIC_PRESENT(BIC_RAMWatt);
3880 			BIC_PRESENT(BIC_GFXWatt);
3881 		}
3882 		break;
3883 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3884 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3885 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3886 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3887 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3888 		BIC_PRESENT(BIC_PKG__);
3889 		BIC_PRESENT(BIC_RAM__);
3890 		if (rapl_joules) {
3891 			BIC_PRESENT(BIC_Pkg_J);
3892 			BIC_PRESENT(BIC_RAM_J);
3893 		} else {
3894 			BIC_PRESENT(BIC_PkgWatt);
3895 			BIC_PRESENT(BIC_RAMWatt);
3896 		}
3897 		break;
3898 	case INTEL_FAM6_SANDYBRIDGE_X:
3899 	case INTEL_FAM6_IVYBRIDGE_X:
3900 		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;
3901 		BIC_PRESENT(BIC_PKG__);
3902 		BIC_PRESENT(BIC_RAM__);
3903 		if (rapl_joules) {
3904 			BIC_PRESENT(BIC_Pkg_J);
3905 			BIC_PRESENT(BIC_Cor_J);
3906 			BIC_PRESENT(BIC_RAM_J);
3907 		} else {
3908 			BIC_PRESENT(BIC_PkgWatt);
3909 			BIC_PRESENT(BIC_CorWatt);
3910 			BIC_PRESENT(BIC_RAMWatt);
3911 		}
3912 		break;
3913 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3914 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3915 		do_rapl = RAPL_PKG | RAPL_CORES;
3916 		if (rapl_joules) {
3917 			BIC_PRESENT(BIC_Pkg_J);
3918 			BIC_PRESENT(BIC_Cor_J);
3919 		} else {
3920 			BIC_PRESENT(BIC_PkgWatt);
3921 			BIC_PRESENT(BIC_CorWatt);
3922 		}
3923 		break;
3924 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3925 		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;
3926 		BIC_PRESENT(BIC_PKG__);
3927 		BIC_PRESENT(BIC_RAM__);
3928 		if (rapl_joules) {
3929 			BIC_PRESENT(BIC_Pkg_J);
3930 			BIC_PRESENT(BIC_Cor_J);
3931 			BIC_PRESENT(BIC_RAM_J);
3932 		} else {
3933 			BIC_PRESENT(BIC_PkgWatt);
3934 			BIC_PRESENT(BIC_CorWatt);
3935 			BIC_PRESENT(BIC_RAMWatt);
3936 		}
3937 		break;
3938 	default:
3939 		return;
3940 	}
3941 
3942 	/* units on package 0, verify later other packages match */
3943 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3944 		return;
3945 
3946 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
3947 	if (model == INTEL_FAM6_ATOM_SILVERMONT)
3948 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3949 	else
3950 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3951 
3952 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3953 
3954 	time_unit = msr >> 16 & 0xF;
3955 	if (time_unit == 0)
3956 		time_unit = 0xA;
3957 
3958 	rapl_time_units = 1.0 / (1 << (time_unit));
3959 
3960 	tdp = get_tdp_intel(model);
3961 
3962 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3963 	if (!quiet)
3964 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3965 }
3966 
3967 void rapl_probe_amd(unsigned int family, unsigned int model)
3968 {
3969 	unsigned long long msr;
3970 	unsigned int eax, ebx, ecx, edx;
3971 	unsigned int has_rapl = 0;
3972 	double tdp;
3973 
3974 	if (max_extended_level >= 0x80000007) {
3975 		__cpuid(0x80000007, eax, ebx, ecx, edx);
3976 		/* RAPL (Fam 17h) */
3977 		has_rapl = edx & (1 << 14);
3978 	}
3979 
3980 	if (!has_rapl)
3981 		return;
3982 
3983 	switch (family) {
3984 	case 0x17: /* Zen, Zen+ */
3985 		do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY;
3986 		if (rapl_joules) {
3987 			BIC_PRESENT(BIC_Pkg_J);
3988 			BIC_PRESENT(BIC_Cor_J);
3989 		} else {
3990 			BIC_PRESENT(BIC_PkgWatt);
3991 			BIC_PRESENT(BIC_CorWatt);
3992 		}
3993 		break;
3994 	default:
3995 		return;
3996 	}
3997 
3998 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
3999 		return;
4000 
4001 	rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf));
4002 	rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f));
4003 	rapl_power_units = ldexp(1.0, -(msr & 0xf));
4004 
4005 	tdp = get_tdp_amd(model);
4006 
4007 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4008 	if (!quiet)
4009 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4010 }
4011 
4012 /*
4013  * rapl_probe()
4014  *
4015  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
4016  */
4017 void rapl_probe(unsigned int family, unsigned int model)
4018 {
4019 	if (genuine_intel)
4020 		rapl_probe_intel(family, model);
4021 	if (authentic_amd)
4022 		rapl_probe_amd(family, model);
4023 }
4024 
4025 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
4026 {
4027 	if (!genuine_intel)
4028 		return;
4029 
4030 	if (family != 6)
4031 		return;
4032 
4033 	switch (model) {
4034 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4035 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4036 		do_gfx_perf_limit_reasons = 1;
4037 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4038 		do_core_perf_limit_reasons = 1;
4039 		do_ring_perf_limit_reasons = 1;
4040 	default:
4041 		return;
4042 	}
4043 }
4044 
4045 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model)
4046 {
4047 	if (is_skx(family, model) || is_bdx(family, model))
4048 		has_automatic_cstate_conversion = 1;
4049 }
4050 
4051 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4052 {
4053 	unsigned long long msr;
4054 	unsigned int dts, dts2;
4055 	int cpu;
4056 
4057 	if (!(do_dts || do_ptm))
4058 		return 0;
4059 
4060 	cpu = t->cpu_id;
4061 
4062 	/* DTS is per-core, no need to print for each thread */
4063 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
4064 		return 0;
4065 
4066 	if (cpu_migrate(cpu)) {
4067 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4068 		return -1;
4069 	}
4070 
4071 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
4072 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
4073 			return 0;
4074 
4075 		dts = (msr >> 16) & 0x7F;
4076 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
4077 			cpu, msr, tcc_activation_temp - dts);
4078 
4079 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
4080 			return 0;
4081 
4082 		dts = (msr >> 16) & 0x7F;
4083 		dts2 = (msr >> 8) & 0x7F;
4084 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4085 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4086 	}
4087 
4088 
4089 	if (do_dts && debug) {
4090 		unsigned int resolution;
4091 
4092 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
4093 			return 0;
4094 
4095 		dts = (msr >> 16) & 0x7F;
4096 		resolution = (msr >> 27) & 0xF;
4097 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
4098 			cpu, msr, tcc_activation_temp - dts, resolution);
4099 
4100 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
4101 			return 0;
4102 
4103 		dts = (msr >> 16) & 0x7F;
4104 		dts2 = (msr >> 8) & 0x7F;
4105 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4106 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4107 	}
4108 
4109 	return 0;
4110 }
4111 
4112 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
4113 {
4114 	fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
4115 		cpu, label,
4116 		((msr >> 15) & 1) ? "EN" : "DIS",
4117 		((msr >> 0) & 0x7FFF) * rapl_power_units,
4118 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
4119 		(((msr >> 16) & 1) ? "EN" : "DIS"));
4120 
4121 	return;
4122 }
4123 
4124 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4125 {
4126 	unsigned long long msr;
4127 	const char *msr_name;
4128 	int cpu;
4129 
4130 	if (!do_rapl)
4131 		return 0;
4132 
4133 	/* RAPL counters are per package, so print only for 1st thread/package */
4134 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4135 		return 0;
4136 
4137 	cpu = t->cpu_id;
4138 	if (cpu_migrate(cpu)) {
4139 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4140 		return -1;
4141 	}
4142 
4143 	if (do_rapl & RAPL_AMD_F17H) {
4144 		msr_name = "MSR_RAPL_PWR_UNIT";
4145 		if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr))
4146 			return -1;
4147 	} else {
4148 		msr_name = "MSR_RAPL_POWER_UNIT";
4149 		if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
4150 			return -1;
4151 	}
4152 
4153 	fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr,
4154 		rapl_power_units, rapl_energy_units, rapl_time_units);
4155 
4156 	if (do_rapl & RAPL_PKG_POWER_INFO) {
4157 
4158 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
4159                 	return -5;
4160 
4161 
4162 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4163 			cpu, msr,
4164 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4165 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4166 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4167 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4168 
4169 	}
4170 	if (do_rapl & RAPL_PKG) {
4171 
4172 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
4173 			return -9;
4174 
4175 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
4176 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
4177 
4178 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
4179 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
4180 			cpu,
4181 			((msr >> 47) & 1) ? "EN" : "DIS",
4182 			((msr >> 32) & 0x7FFF) * rapl_power_units,
4183 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
4184 			((msr >> 48) & 1) ? "EN" : "DIS");
4185 	}
4186 
4187 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
4188 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
4189                 	return -6;
4190 
4191 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4192 			cpu, msr,
4193 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4194 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4195 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4196 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4197 	}
4198 	if (do_rapl & RAPL_DRAM) {
4199 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
4200 			return -9;
4201 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
4202 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4203 
4204 		print_power_limit_msr(cpu, msr, "DRAM Limit");
4205 	}
4206 	if (do_rapl & RAPL_CORE_POLICY) {
4207 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
4208 			return -7;
4209 
4210 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
4211 	}
4212 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
4213 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
4214 			return -9;
4215 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
4216 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4217 		print_power_limit_msr(cpu, msr, "Cores Limit");
4218 	}
4219 	if (do_rapl & RAPL_GFX) {
4220 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
4221 			return -8;
4222 
4223 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
4224 
4225 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
4226 			return -9;
4227 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
4228 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4229 		print_power_limit_msr(cpu, msr, "GFX Limit");
4230 	}
4231 	return 0;
4232 }
4233 
4234 /*
4235  * SNB adds support for additional MSRs:
4236  *
4237  * MSR_PKG_C7_RESIDENCY            0x000003fa
4238  * MSR_CORE_C7_RESIDENCY           0x000003fe
4239  * MSR_PKG_C2_RESIDENCY            0x0000060d
4240  */
4241 
4242 int has_snb_msrs(unsigned int family, unsigned int model)
4243 {
4244 	if (!genuine_intel)
4245 		return 0;
4246 
4247 	switch (model) {
4248 	case INTEL_FAM6_SANDYBRIDGE:
4249 	case INTEL_FAM6_SANDYBRIDGE_X:
4250 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
4251 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
4252 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4253 	case INTEL_FAM6_HASWELL_X:	/* HSW */
4254 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4255 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4256 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
4257 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4258 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4259 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4260 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4261 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4262 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4263 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4264 		return 1;
4265 	}
4266 	return 0;
4267 }
4268 
4269 /*
4270  * HSW adds support for additional MSRs:
4271  *
4272  * MSR_PKG_C8_RESIDENCY		0x00000630
4273  * MSR_PKG_C9_RESIDENCY		0x00000631
4274  * MSR_PKG_C10_RESIDENCY	0x00000632
4275  *
4276  * MSR_PKGC8_IRTL		0x00000633
4277  * MSR_PKGC9_IRTL		0x00000634
4278  * MSR_PKGC10_IRTL		0x00000635
4279  *
4280  */
4281 int has_hsw_msrs(unsigned int family, unsigned int model)
4282 {
4283 	if (!genuine_intel)
4284 		return 0;
4285 
4286 	switch (model) {
4287 	case INTEL_FAM6_HASWELL_CORE:
4288 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4289 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4290 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4291 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4292 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4293 		return 1;
4294 	}
4295 	return 0;
4296 }
4297 
4298 /*
4299  * SKL adds support for additional MSRS:
4300  *
4301  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
4302  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
4303  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
4304  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
4305  */
4306 int has_skl_msrs(unsigned int family, unsigned int model)
4307 {
4308 	if (!genuine_intel)
4309 		return 0;
4310 
4311 	switch (model) {
4312 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4313 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4314 		return 1;
4315 	}
4316 	return 0;
4317 }
4318 
4319 int is_slm(unsigned int family, unsigned int model)
4320 {
4321 	if (!genuine_intel)
4322 		return 0;
4323 	switch (model) {
4324 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
4325 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
4326 		return 1;
4327 	}
4328 	return 0;
4329 }
4330 
4331 int is_knl(unsigned int family, unsigned int model)
4332 {
4333 	if (!genuine_intel)
4334 		return 0;
4335 	switch (model) {
4336 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4337 		return 1;
4338 	}
4339 	return 0;
4340 }
4341 
4342 int is_cnl(unsigned int family, unsigned int model)
4343 {
4344 	if (!genuine_intel)
4345 		return 0;
4346 
4347 	switch (model) {
4348 	case INTEL_FAM6_CANNONLAKE_MOBILE: /* CNL */
4349 		return 1;
4350 	}
4351 
4352 	return 0;
4353 }
4354 
4355 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
4356 {
4357 	if (is_knl(family, model))
4358 		return 1024;
4359 	return 1;
4360 }
4361 
4362 #define SLM_BCLK_FREQS 5
4363 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
4364 
4365 double slm_bclk(void)
4366 {
4367 	unsigned long long msr = 3;
4368 	unsigned int i;
4369 	double freq;
4370 
4371 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
4372 		fprintf(outf, "SLM BCLK: unknown\n");
4373 
4374 	i = msr & 0xf;
4375 	if (i >= SLM_BCLK_FREQS) {
4376 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
4377 		i = 3;
4378 	}
4379 	freq = slm_freq_table[i];
4380 
4381 	if (!quiet)
4382 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
4383 
4384 	return freq;
4385 }
4386 
4387 double discover_bclk(unsigned int family, unsigned int model)
4388 {
4389 	if (has_snb_msrs(family, model) || is_knl(family, model))
4390 		return 100.00;
4391 	else if (is_slm(family, model))
4392 		return slm_bclk();
4393 	else
4394 		return 133.33;
4395 }
4396 
4397 /*
4398  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
4399  * the Thermal Control Circuit (TCC) activates.
4400  * This is usually equal to tjMax.
4401  *
4402  * Older processors do not have this MSR, so there we guess,
4403  * but also allow cmdline over-ride with -T.
4404  *
4405  * Several MSR temperature values are in units of degrees-C
4406  * below this value, including the Digital Thermal Sensor (DTS),
4407  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
4408  */
4409 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4410 {
4411 	unsigned long long msr;
4412 	unsigned int target_c_local;
4413 	int cpu;
4414 
4415 	/* tcc_activation_temp is used only for dts or ptm */
4416 	if (!(do_dts || do_ptm))
4417 		return 0;
4418 
4419 	/* this is a per-package concept */
4420 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4421 		return 0;
4422 
4423 	cpu = t->cpu_id;
4424 	if (cpu_migrate(cpu)) {
4425 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4426 		return -1;
4427 	}
4428 
4429 	if (tcc_activation_temp_override != 0) {
4430 		tcc_activation_temp = tcc_activation_temp_override;
4431 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
4432 			cpu, tcc_activation_temp);
4433 		return 0;
4434 	}
4435 
4436 	/* Temperature Target MSR is Nehalem and newer only */
4437 	if (!do_nhm_platform_info)
4438 		goto guess;
4439 
4440 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
4441 		goto guess;
4442 
4443 	target_c_local = (msr >> 16) & 0xFF;
4444 
4445 	if (!quiet)
4446 		fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
4447 			cpu, msr, target_c_local);
4448 
4449 	if (!target_c_local)
4450 		goto guess;
4451 
4452 	tcc_activation_temp = target_c_local;
4453 
4454 	return 0;
4455 
4456 guess:
4457 	tcc_activation_temp = TJMAX_DEFAULT;
4458 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
4459 		cpu, tcc_activation_temp);
4460 
4461 	return 0;
4462 }
4463 
4464 void decode_feature_control_msr(void)
4465 {
4466 	unsigned long long msr;
4467 
4468 	if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
4469 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
4470 			base_cpu, msr,
4471 			msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
4472 			msr & (1 << 18) ? "SGX" : "");
4473 }
4474 
4475 void decode_misc_enable_msr(void)
4476 {
4477 	unsigned long long msr;
4478 
4479 	if (!genuine_intel)
4480 		return;
4481 
4482 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
4483 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
4484 			base_cpu, msr,
4485 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
4486 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
4487 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-",
4488 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
4489 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
4490 }
4491 
4492 void decode_misc_feature_control(void)
4493 {
4494 	unsigned long long msr;
4495 
4496 	if (!has_misc_feature_control)
4497 		return;
4498 
4499 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
4500 		fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
4501 			base_cpu, msr,
4502 			msr & (0 << 0) ? "No-" : "",
4503 			msr & (1 << 0) ? "No-" : "",
4504 			msr & (2 << 0) ? "No-" : "",
4505 			msr & (3 << 0) ? "No-" : "");
4506 }
4507 /*
4508  * Decode MSR_MISC_PWR_MGMT
4509  *
4510  * Decode the bits according to the Nehalem documentation
4511  * bit[0] seems to continue to have same meaning going forward
4512  * bit[1] less so...
4513  */
4514 void decode_misc_pwr_mgmt_msr(void)
4515 {
4516 	unsigned long long msr;
4517 
4518 	if (!do_nhm_platform_info)
4519 		return;
4520 
4521 	if (no_MSR_MISC_PWR_MGMT)
4522 		return;
4523 
4524 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
4525 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
4526 			base_cpu, msr,
4527 			msr & (1 << 0) ? "DIS" : "EN",
4528 			msr & (1 << 1) ? "EN" : "DIS",
4529 			msr & (1 << 8) ? "EN" : "DIS");
4530 }
4531 /*
4532  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
4533  *
4534  * This MSRs are present on Silvermont processors,
4535  * Intel Atom processor E3000 series (Baytrail), and friends.
4536  */
4537 void decode_c6_demotion_policy_msr(void)
4538 {
4539 	unsigned long long msr;
4540 
4541 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
4542 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
4543 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4544 
4545 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
4546 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
4547 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4548 }
4549 
4550 /*
4551  * When models are the same, for the purpose of turbostat, reuse
4552  */
4553 unsigned int intel_model_duplicates(unsigned int model)
4554 {
4555 
4556 	switch(model) {
4557 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
4558 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
4559 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
4560 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
4561 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
4562 		return INTEL_FAM6_NEHALEM;
4563 
4564 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
4565 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
4566 		return INTEL_FAM6_NEHALEM_EX;
4567 
4568 	case INTEL_FAM6_XEON_PHI_KNM:
4569 		return INTEL_FAM6_XEON_PHI_KNL;
4570 
4571 	case INTEL_FAM6_HASWELL_ULT:
4572 		return INTEL_FAM6_HASWELL_CORE;
4573 
4574 	case INTEL_FAM6_BROADWELL_X:
4575 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
4576 		return INTEL_FAM6_BROADWELL_X;
4577 
4578 	case INTEL_FAM6_SKYLAKE_MOBILE:
4579 	case INTEL_FAM6_SKYLAKE_DESKTOP:
4580 	case INTEL_FAM6_KABYLAKE_MOBILE:
4581 	case INTEL_FAM6_KABYLAKE_DESKTOP:
4582 		return INTEL_FAM6_SKYLAKE_MOBILE;
4583 
4584 	case INTEL_FAM6_ICELAKE_MOBILE:
4585 		return INTEL_FAM6_CANNONLAKE_MOBILE;
4586 	}
4587 	return model;
4588 }
4589 void process_cpuid()
4590 {
4591 	unsigned int eax, ebx, ecx, edx;
4592 	unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
4593 	unsigned int has_turbo;
4594 
4595 	eax = ebx = ecx = edx = 0;
4596 
4597 	__cpuid(0, max_level, ebx, ecx, edx);
4598 
4599 	if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
4600 		genuine_intel = 1;
4601 	else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
4602 		authentic_amd = 1;
4603 
4604 	if (!quiet)
4605 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
4606 			(char *)&ebx, (char *)&edx, (char *)&ecx);
4607 
4608 	__cpuid(1, fms, ebx, ecx, edx);
4609 	family = (fms >> 8) & 0xf;
4610 	model = (fms >> 4) & 0xf;
4611 	stepping = fms & 0xf;
4612 	if (family == 0xf)
4613 		family += (fms >> 20) & 0xff;
4614 	if (family >= 6)
4615 		model += ((fms >> 16) & 0xf) << 4;
4616 	ecx_flags = ecx;
4617 	edx_flags = edx;
4618 
4619 	/*
4620 	 * check max extended function levels of CPUID.
4621 	 * This is needed to check for invariant TSC.
4622 	 * This check is valid for both Intel and AMD.
4623 	 */
4624 	ebx = ecx = edx = 0;
4625 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
4626 
4627 	if (!quiet) {
4628 		fprintf(outf, "0x%x CPUID levels; 0x%x xlevels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
4629 			max_level, max_extended_level, family, model, stepping, family, model, stepping);
4630 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n",
4631 			ecx_flags & (1 << 0) ? "SSE3" : "-",
4632 			ecx_flags & (1 << 3) ? "MONITOR" : "-",
4633 			ecx_flags & (1 << 6) ? "SMX" : "-",
4634 			ecx_flags & (1 << 7) ? "EIST" : "-",
4635 			ecx_flags & (1 << 8) ? "TM2" : "-",
4636 			edx_flags & (1 << 4) ? "TSC" : "-",
4637 			edx_flags & (1 << 5) ? "MSR" : "-",
4638 			edx_flags & (1 << 22) ? "ACPI-TM" : "-",
4639 			edx_flags & (1 << 28) ? "HT" : "-",
4640 			edx_flags & (1 << 29) ? "TM" : "-");
4641 	}
4642 	if (genuine_intel)
4643 		model = intel_model_duplicates(model);
4644 
4645 	if (!(edx_flags & (1 << 5)))
4646 		errx(1, "CPUID: no MSR");
4647 
4648 	if (max_extended_level >= 0x80000007) {
4649 
4650 		/*
4651 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
4652 		 * this check is valid for both Intel and AMD
4653 		 */
4654 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4655 		has_invariant_tsc = edx & (1 << 8);
4656 	}
4657 
4658 	/*
4659 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
4660 	 * this check is valid for both Intel and AMD
4661 	 */
4662 
4663 	__cpuid(0x6, eax, ebx, ecx, edx);
4664 	has_aperf = ecx & (1 << 0);
4665 	if (has_aperf) {
4666 		BIC_PRESENT(BIC_Avg_MHz);
4667 		BIC_PRESENT(BIC_Busy);
4668 		BIC_PRESENT(BIC_Bzy_MHz);
4669 	}
4670 	do_dts = eax & (1 << 0);
4671 	if (do_dts)
4672 		BIC_PRESENT(BIC_CoreTmp);
4673 	has_turbo = eax & (1 << 1);
4674 	do_ptm = eax & (1 << 6);
4675 	if (do_ptm)
4676 		BIC_PRESENT(BIC_PkgTmp);
4677 	has_hwp = eax & (1 << 7);
4678 	has_hwp_notify = eax & (1 << 8);
4679 	has_hwp_activity_window = eax & (1 << 9);
4680 	has_hwp_epp = eax & (1 << 10);
4681 	has_hwp_pkg = eax & (1 << 11);
4682 	has_epb = ecx & (1 << 3);
4683 
4684 	if (!quiet)
4685 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
4686 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
4687 			has_aperf ? "" : "No-",
4688 			has_turbo ? "" : "No-",
4689 			do_dts ? "" : "No-",
4690 			do_ptm ? "" : "No-",
4691 			has_hwp ? "" : "No-",
4692 			has_hwp_notify ? "" : "No-",
4693 			has_hwp_activity_window ? "" : "No-",
4694 			has_hwp_epp ? "" : "No-",
4695 			has_hwp_pkg ? "" : "No-",
4696 			has_epb ? "" : "No-");
4697 
4698 	if (!quiet)
4699 		decode_misc_enable_msr();
4700 
4701 
4702 	if (max_level >= 0x7 && !quiet) {
4703 		int has_sgx;
4704 
4705 		ecx = 0;
4706 
4707 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
4708 
4709 		has_sgx = ebx & (1 << 2);
4710 		fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
4711 
4712 		if (has_sgx)
4713 			decode_feature_control_msr();
4714 	}
4715 
4716 	if (max_level >= 0x15) {
4717 		unsigned int eax_crystal;
4718 		unsigned int ebx_tsc;
4719 
4720 		/*
4721 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
4722 		 */
4723 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
4724 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
4725 
4726 		if (ebx_tsc != 0) {
4727 
4728 			if (!quiet && (ebx != 0))
4729 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
4730 					eax_crystal, ebx_tsc, crystal_hz);
4731 
4732 			if (crystal_hz == 0)
4733 				switch(model) {
4734 				case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4735 					crystal_hz = 24000000;	/* 24.0 MHz */
4736 					break;
4737 				case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4738 					crystal_hz = 25000000;	/* 25.0 MHz */
4739 					break;
4740 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4741 				case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4742 					crystal_hz = 19200000;	/* 19.2 MHz */
4743 					break;
4744 				default:
4745 					crystal_hz = 0;
4746 			}
4747 
4748 			if (crystal_hz) {
4749 				tsc_hz =  (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
4750 				if (!quiet)
4751 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
4752 						tsc_hz / 1000000, crystal_hz, ebx_tsc,  eax_crystal);
4753 			}
4754 		}
4755 	}
4756 	if (max_level >= 0x16) {
4757 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
4758 
4759 		/*
4760 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
4761 		 */
4762 		base_mhz = max_mhz = bus_mhz = edx = 0;
4763 
4764 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
4765 		if (!quiet)
4766 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
4767 				base_mhz, max_mhz, bus_mhz);
4768 	}
4769 
4770 	if (has_aperf)
4771 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
4772 
4773 	BIC_PRESENT(BIC_IRQ);
4774 	BIC_PRESENT(BIC_TSC_MHz);
4775 
4776 	if (probe_nhm_msrs(family, model)) {
4777 		do_nhm_platform_info = 1;
4778 		BIC_PRESENT(BIC_CPU_c1);
4779 		BIC_PRESENT(BIC_CPU_c3);
4780 		BIC_PRESENT(BIC_CPU_c6);
4781 		BIC_PRESENT(BIC_SMI);
4782 	}
4783 	do_snb_cstates = has_snb_msrs(family, model);
4784 
4785 	if (do_snb_cstates)
4786 		BIC_PRESENT(BIC_CPU_c7);
4787 
4788 	do_irtl_snb = has_snb_msrs(family, model);
4789 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
4790 		BIC_PRESENT(BIC_Pkgpc2);
4791 	if (pkg_cstate_limit >= PCL__3)
4792 		BIC_PRESENT(BIC_Pkgpc3);
4793 	if (pkg_cstate_limit >= PCL__6)
4794 		BIC_PRESENT(BIC_Pkgpc6);
4795 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
4796 		BIC_PRESENT(BIC_Pkgpc7);
4797 	if (has_slv_msrs(family, model)) {
4798 		BIC_NOT_PRESENT(BIC_Pkgpc2);
4799 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4800 		BIC_PRESENT(BIC_Pkgpc6);
4801 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4802 		BIC_PRESENT(BIC_Mod_c6);
4803 		use_c1_residency_msr = 1;
4804 	}
4805 	if (is_dnv(family, model)) {
4806 		BIC_PRESENT(BIC_CPU_c1);
4807 		BIC_NOT_PRESENT(BIC_CPU_c3);
4808 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4809 		BIC_NOT_PRESENT(BIC_CPU_c7);
4810 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4811 		use_c1_residency_msr = 1;
4812 	}
4813 	if (is_skx(family, model)) {
4814 		BIC_NOT_PRESENT(BIC_CPU_c3);
4815 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4816 		BIC_NOT_PRESENT(BIC_CPU_c7);
4817 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4818 	}
4819 	if (is_bdx(family, model)) {
4820 		BIC_NOT_PRESENT(BIC_CPU_c7);
4821 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4822 	}
4823 	if (has_hsw_msrs(family, model)) {
4824 		BIC_PRESENT(BIC_Pkgpc8);
4825 		BIC_PRESENT(BIC_Pkgpc9);
4826 		BIC_PRESENT(BIC_Pkgpc10);
4827 	}
4828 	do_irtl_hsw = has_hsw_msrs(family, model);
4829 	if (has_skl_msrs(family, model)) {
4830 		BIC_PRESENT(BIC_Totl_c0);
4831 		BIC_PRESENT(BIC_Any_c0);
4832 		BIC_PRESENT(BIC_GFX_c0);
4833 		BIC_PRESENT(BIC_CPUGFX);
4834 	}
4835 	do_slm_cstates = is_slm(family, model);
4836 	do_knl_cstates  = is_knl(family, model);
4837 
4838 	if (do_slm_cstates || do_knl_cstates || is_cnl(family, model))
4839 		BIC_NOT_PRESENT(BIC_CPU_c3);
4840 
4841 	if (!quiet)
4842 		decode_misc_pwr_mgmt_msr();
4843 
4844 	if (!quiet && has_slv_msrs(family, model))
4845 		decode_c6_demotion_policy_msr();
4846 
4847 	rapl_probe(family, model);
4848 	perf_limit_reasons_probe(family, model);
4849 	automatic_cstate_conversion_probe(family, model);
4850 
4851 	if (!quiet)
4852 		dump_cstate_pstate_config_info(family, model);
4853 
4854 	if (!quiet)
4855 		dump_sysfs_cstate_config();
4856 	if (!quiet)
4857 		dump_sysfs_pstate_config();
4858 
4859 	if (has_skl_msrs(family, model))
4860 		calculate_tsc_tweak();
4861 
4862 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
4863 		BIC_PRESENT(BIC_GFX_rc6);
4864 
4865 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
4866 		BIC_PRESENT(BIC_GFXMHz);
4867 
4868 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK))
4869 		BIC_PRESENT(BIC_CPU_LPI);
4870 	else
4871 		BIC_NOT_PRESENT(BIC_CPU_LPI);
4872 
4873 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", R_OK))
4874 		BIC_PRESENT(BIC_SYS_LPI);
4875 	else
4876 		BIC_NOT_PRESENT(BIC_SYS_LPI);
4877 
4878 	if (!quiet)
4879 		decode_misc_feature_control();
4880 
4881 	return;
4882 }
4883 
4884 /*
4885  * in /dev/cpu/ return success for names that are numbers
4886  * ie. filter out ".", "..", "microcode".
4887  */
4888 int dir_filter(const struct dirent *dirp)
4889 {
4890 	if (isdigit(dirp->d_name[0]))
4891 		return 1;
4892 	else
4893 		return 0;
4894 }
4895 
4896 int open_dev_cpu_msr(int dummy1)
4897 {
4898 	return 0;
4899 }
4900 
4901 void topology_probe()
4902 {
4903 	int i;
4904 	int max_core_id = 0;
4905 	int max_package_id = 0;
4906 	int max_die_id = 0;
4907 	int max_siblings = 0;
4908 
4909 	/* Initialize num_cpus, max_cpu_num */
4910 	set_max_cpu_num();
4911 	topo.num_cpus = 0;
4912 	for_all_proc_cpus(count_cpus);
4913 	if (!summary_only && topo.num_cpus > 1)
4914 		BIC_PRESENT(BIC_CPU);
4915 
4916 	if (debug > 1)
4917 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4918 
4919 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
4920 	if (cpus == NULL)
4921 		err(1, "calloc cpus");
4922 
4923 	/*
4924 	 * Allocate and initialize cpu_present_set
4925 	 */
4926 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4927 	if (cpu_present_set == NULL)
4928 		err(3, "CPU_ALLOC");
4929 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4930 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4931 	for_all_proc_cpus(mark_cpu_present);
4932 
4933 	/*
4934 	 * Validate that all cpus in cpu_subset are also in cpu_present_set
4935 	 */
4936 	for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
4937 		if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
4938 			if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
4939 				err(1, "cpu%d not present", i);
4940 	}
4941 
4942 	/*
4943 	 * Allocate and initialize cpu_affinity_set
4944 	 */
4945 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4946 	if (cpu_affinity_set == NULL)
4947 		err(3, "CPU_ALLOC");
4948 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4949 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4950 
4951 	for_all_proc_cpus(init_thread_id);
4952 
4953 	/*
4954 	 * For online cpus
4955 	 * find max_core_id, max_package_id
4956 	 */
4957 	for (i = 0; i <= topo.max_cpu_num; ++i) {
4958 		int siblings;
4959 
4960 		if (cpu_is_not_present(i)) {
4961 			if (debug > 1)
4962 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
4963 			continue;
4964 		}
4965 
4966 		cpus[i].logical_cpu_id = i;
4967 
4968 		/* get package information */
4969 		cpus[i].physical_package_id = get_physical_package_id(i);
4970 		if (cpus[i].physical_package_id > max_package_id)
4971 			max_package_id = cpus[i].physical_package_id;
4972 
4973 		/* get die information */
4974 		cpus[i].die_id = get_die_id(i);
4975 		if (cpus[i].die_id > max_die_id)
4976 			max_die_id = cpus[i].die_id;
4977 
4978 		/* get numa node information */
4979 		cpus[i].physical_node_id = get_physical_node_id(&cpus[i]);
4980 		if (cpus[i].physical_node_id > topo.max_node_num)
4981 			topo.max_node_num = cpus[i].physical_node_id;
4982 
4983 		/* get core information */
4984 		cpus[i].physical_core_id = get_core_id(i);
4985 		if (cpus[i].physical_core_id > max_core_id)
4986 			max_core_id = cpus[i].physical_core_id;
4987 
4988 		/* get thread information */
4989 		siblings = get_thread_siblings(&cpus[i]);
4990 		if (siblings > max_siblings)
4991 			max_siblings = siblings;
4992 		if (cpus[i].thread_id == 0)
4993 			topo.num_cores++;
4994 	}
4995 
4996 	topo.cores_per_node = max_core_id + 1;
4997 	if (debug > 1)
4998 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
4999 			max_core_id, topo.cores_per_node);
5000 	if (!summary_only && topo.cores_per_node > 1)
5001 		BIC_PRESENT(BIC_Core);
5002 
5003 	topo.num_die = max_die_id + 1;
5004 	if (debug > 1)
5005 		fprintf(outf, "max_die_id %d, sizing for %d die\n",
5006 				max_die_id, topo.num_die);
5007 	if (!summary_only && topo.num_die > 1)
5008 		BIC_PRESENT(BIC_Die);
5009 
5010 	topo.num_packages = max_package_id + 1;
5011 	if (debug > 1)
5012 		fprintf(outf, "max_package_id %d, sizing for %d packages\n",
5013 			max_package_id, topo.num_packages);
5014 	if (!summary_only && topo.num_packages > 1)
5015 		BIC_PRESENT(BIC_Package);
5016 
5017 	set_node_data();
5018 	if (debug > 1)
5019 		fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg);
5020 	if (!summary_only && topo.nodes_per_pkg > 1)
5021 		BIC_PRESENT(BIC_Node);
5022 
5023 	topo.threads_per_core = max_siblings;
5024 	if (debug > 1)
5025 		fprintf(outf, "max_siblings %d\n", max_siblings);
5026 
5027 	if (debug < 1)
5028 		return;
5029 
5030 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5031 		if (cpu_is_not_present(i))
5032 			continue;
5033 		fprintf(outf,
5034 			"cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n",
5035 			i, cpus[i].physical_package_id, cpus[i].die_id,
5036 			cpus[i].physical_node_id,
5037 			cpus[i].logical_node_id,
5038 			cpus[i].physical_core_id,
5039 			cpus[i].thread_id);
5040 	}
5041 
5042 }
5043 
5044 void
5045 allocate_counters(struct thread_data **t, struct core_data **c,
5046 		  struct pkg_data **p)
5047 {
5048 	int i;
5049 	int num_cores = topo.cores_per_node * topo.nodes_per_pkg *
5050 			topo.num_packages;
5051 	int num_threads = topo.threads_per_core * num_cores;
5052 
5053 	*t = calloc(num_threads, sizeof(struct thread_data));
5054 	if (*t == NULL)
5055 		goto error;
5056 
5057 	for (i = 0; i < num_threads; i++)
5058 		(*t)[i].cpu_id = -1;
5059 
5060 	*c = calloc(num_cores, sizeof(struct core_data));
5061 	if (*c == NULL)
5062 		goto error;
5063 
5064 	for (i = 0; i < num_cores; i++)
5065 		(*c)[i].core_id = -1;
5066 
5067 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
5068 	if (*p == NULL)
5069 		goto error;
5070 
5071 	for (i = 0; i < topo.num_packages; i++)
5072 		(*p)[i].package_id = i;
5073 
5074 	return;
5075 error:
5076 	err(1, "calloc counters");
5077 }
5078 /*
5079  * init_counter()
5080  *
5081  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
5082  */
5083 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
5084 	struct pkg_data *pkg_base, int cpu_id)
5085 {
5086 	int pkg_id = cpus[cpu_id].physical_package_id;
5087 	int node_id = cpus[cpu_id].logical_node_id;
5088 	int core_id = cpus[cpu_id].physical_core_id;
5089 	int thread_id = cpus[cpu_id].thread_id;
5090 	struct thread_data *t;
5091 	struct core_data *c;
5092 	struct pkg_data *p;
5093 
5094 
5095 	/* Workaround for systems where physical_node_id==-1
5096 	 * and logical_node_id==(-1 - topo.num_cpus)
5097 	 */
5098 	if (node_id < 0)
5099 		node_id = 0;
5100 
5101 	t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id);
5102 	c = GET_CORE(core_base, core_id, node_id, pkg_id);
5103 	p = GET_PKG(pkg_base, pkg_id);
5104 
5105 	t->cpu_id = cpu_id;
5106 	if (thread_id == 0) {
5107 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
5108 		if (cpu_is_first_core_in_package(cpu_id))
5109 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
5110 	}
5111 
5112 	c->core_id = core_id;
5113 	p->package_id = pkg_id;
5114 }
5115 
5116 
5117 int initialize_counters(int cpu_id)
5118 {
5119 	init_counter(EVEN_COUNTERS, cpu_id);
5120 	init_counter(ODD_COUNTERS, cpu_id);
5121 	return 0;
5122 }
5123 
5124 void allocate_output_buffer()
5125 {
5126 	output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
5127 	outp = output_buffer;
5128 	if (outp == NULL)
5129 		err(-1, "calloc output buffer");
5130 }
5131 void allocate_fd_percpu(void)
5132 {
5133 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5134 	if (fd_percpu == NULL)
5135 		err(-1, "calloc fd_percpu");
5136 }
5137 void allocate_irq_buffers(void)
5138 {
5139 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
5140 	if (irq_column_2_cpu == NULL)
5141 		err(-1, "calloc %d", topo.num_cpus);
5142 
5143 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5144 	if (irqs_per_cpu == NULL)
5145 		err(-1, "calloc %d", topo.max_cpu_num + 1);
5146 }
5147 void setup_all_buffers(void)
5148 {
5149 	topology_probe();
5150 	allocate_irq_buffers();
5151 	allocate_fd_percpu();
5152 	allocate_counters(&thread_even, &core_even, &package_even);
5153 	allocate_counters(&thread_odd, &core_odd, &package_odd);
5154 	allocate_output_buffer();
5155 	for_all_proc_cpus(initialize_counters);
5156 }
5157 
5158 void set_base_cpu(void)
5159 {
5160 	base_cpu = sched_getcpu();
5161 	if (base_cpu < 0)
5162 		err(-ENODEV, "No valid cpus found");
5163 
5164 	if (debug > 1)
5165 		fprintf(outf, "base_cpu = %d\n", base_cpu);
5166 }
5167 
5168 void turbostat_init()
5169 {
5170 	setup_all_buffers();
5171 	set_base_cpu();
5172 	check_dev_msr();
5173 	check_permissions();
5174 	process_cpuid();
5175 
5176 
5177 	if (!quiet)
5178 		for_all_cpus(print_hwp, ODD_COUNTERS);
5179 
5180 	if (!quiet)
5181 		for_all_cpus(print_epb, ODD_COUNTERS);
5182 
5183 	if (!quiet)
5184 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
5185 
5186 	if (!quiet)
5187 		for_all_cpus(print_rapl, ODD_COUNTERS);
5188 
5189 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
5190 
5191 	if (!quiet)
5192 		for_all_cpus(print_thermal, ODD_COUNTERS);
5193 
5194 	if (!quiet && do_irtl_snb)
5195 		print_irtl();
5196 }
5197 
5198 int fork_it(char **argv)
5199 {
5200 	pid_t child_pid;
5201 	int status;
5202 
5203 	snapshot_proc_sysfs_files();
5204 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
5205 	first_counter_read = 0;
5206 	if (status)
5207 		exit(status);
5208 	/* clear affinity side-effect of get_counters() */
5209 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
5210 	gettimeofday(&tv_even, (struct timezone *)NULL);
5211 
5212 	child_pid = fork();
5213 	if (!child_pid) {
5214 		/* child */
5215 		execvp(argv[0], argv);
5216 		err(errno, "exec %s", argv[0]);
5217 	} else {
5218 
5219 		/* parent */
5220 		if (child_pid == -1)
5221 			err(1, "fork");
5222 
5223 		signal(SIGINT, SIG_IGN);
5224 		signal(SIGQUIT, SIG_IGN);
5225 		if (waitpid(child_pid, &status, 0) == -1)
5226 			err(status, "waitpid");
5227 
5228 		if (WIFEXITED(status))
5229 			status = WEXITSTATUS(status);
5230 	}
5231 	/*
5232 	 * n.b. fork_it() does not check for errors from for_all_cpus()
5233 	 * because re-starting is problematic when forking
5234 	 */
5235 	snapshot_proc_sysfs_files();
5236 	for_all_cpus(get_counters, ODD_COUNTERS);
5237 	gettimeofday(&tv_odd, (struct timezone *)NULL);
5238 	timersub(&tv_odd, &tv_even, &tv_delta);
5239 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
5240 		fprintf(outf, "%s: Counter reset detected\n", progname);
5241 	else {
5242 		compute_average(EVEN_COUNTERS);
5243 		format_all_counters(EVEN_COUNTERS);
5244 	}
5245 
5246 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
5247 
5248 	flush_output_stderr();
5249 
5250 	return status;
5251 }
5252 
5253 int get_and_dump_counters(void)
5254 {
5255 	int status;
5256 
5257 	snapshot_proc_sysfs_files();
5258 	status = for_all_cpus(get_counters, ODD_COUNTERS);
5259 	if (status)
5260 		return status;
5261 
5262 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
5263 	if (status)
5264 		return status;
5265 
5266 	flush_output_stdout();
5267 
5268 	return status;
5269 }
5270 
5271 void print_version() {
5272 	fprintf(outf, "turbostat version 19.03.20"
5273 		" - Len Brown <lenb@kernel.org>\n");
5274 }
5275 
5276 int add_counter(unsigned int msr_num, char *path, char *name,
5277 	unsigned int width, enum counter_scope scope,
5278 	enum counter_type type, enum counter_format format, int flags)
5279 {
5280 	struct msr_counter *msrp;
5281 
5282 	msrp = calloc(1, sizeof(struct msr_counter));
5283 	if (msrp == NULL) {
5284 		perror("calloc");
5285 		exit(1);
5286 	}
5287 
5288 	msrp->msr_num = msr_num;
5289 	strncpy(msrp->name, name, NAME_BYTES);
5290 	if (path)
5291 		strncpy(msrp->path, path, PATH_BYTES);
5292 	msrp->width = width;
5293 	msrp->type = type;
5294 	msrp->format = format;
5295 	msrp->flags = flags;
5296 
5297 	switch (scope) {
5298 
5299 	case SCOPE_CPU:
5300 		msrp->next = sys.tp;
5301 		sys.tp = msrp;
5302 		sys.added_thread_counters++;
5303 		if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) {
5304 			fprintf(stderr, "exceeded max %d added thread counters\n",
5305 				MAX_ADDED_COUNTERS);
5306 			exit(-1);
5307 		}
5308 		break;
5309 
5310 	case SCOPE_CORE:
5311 		msrp->next = sys.cp;
5312 		sys.cp = msrp;
5313 		sys.added_core_counters++;
5314 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
5315 			fprintf(stderr, "exceeded max %d added core counters\n",
5316 				MAX_ADDED_COUNTERS);
5317 			exit(-1);
5318 		}
5319 		break;
5320 
5321 	case SCOPE_PACKAGE:
5322 		msrp->next = sys.pp;
5323 		sys.pp = msrp;
5324 		sys.added_package_counters++;
5325 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
5326 			fprintf(stderr, "exceeded max %d added package counters\n",
5327 				MAX_ADDED_COUNTERS);
5328 			exit(-1);
5329 		}
5330 		break;
5331 	}
5332 
5333 	return 0;
5334 }
5335 
5336 void parse_add_command(char *add_command)
5337 {
5338 	int msr_num = 0;
5339 	char *path = NULL;
5340 	char name_buffer[NAME_BYTES] = "";
5341 	int width = 64;
5342 	int fail = 0;
5343 	enum counter_scope scope = SCOPE_CPU;
5344 	enum counter_type type = COUNTER_CYCLES;
5345 	enum counter_format format = FORMAT_DELTA;
5346 
5347 	while (add_command) {
5348 
5349 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
5350 			goto next;
5351 
5352 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
5353 			goto next;
5354 
5355 		if (*add_command == '/') {
5356 			path = add_command;
5357 			goto next;
5358 		}
5359 
5360 		if (sscanf(add_command, "u%d", &width) == 1) {
5361 			if ((width == 32) || (width == 64))
5362 				goto next;
5363 			width = 64;
5364 		}
5365 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
5366 			scope = SCOPE_CPU;
5367 			goto next;
5368 		}
5369 		if (!strncmp(add_command, "core", strlen("core"))) {
5370 			scope = SCOPE_CORE;
5371 			goto next;
5372 		}
5373 		if (!strncmp(add_command, "package", strlen("package"))) {
5374 			scope = SCOPE_PACKAGE;
5375 			goto next;
5376 		}
5377 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
5378 			type = COUNTER_CYCLES;
5379 			goto next;
5380 		}
5381 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
5382 			type = COUNTER_SECONDS;
5383 			goto next;
5384 		}
5385 		if (!strncmp(add_command, "usec", strlen("usec"))) {
5386 			type = COUNTER_USEC;
5387 			goto next;
5388 		}
5389 		if (!strncmp(add_command, "raw", strlen("raw"))) {
5390 			format = FORMAT_RAW;
5391 			goto next;
5392 		}
5393 		if (!strncmp(add_command, "delta", strlen("delta"))) {
5394 			format = FORMAT_DELTA;
5395 			goto next;
5396 		}
5397 		if (!strncmp(add_command, "percent", strlen("percent"))) {
5398 			format = FORMAT_PERCENT;
5399 			goto next;
5400 		}
5401 
5402 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
5403 			char *eos;
5404 
5405 			eos = strchr(name_buffer, ',');
5406 			if (eos)
5407 				*eos = '\0';
5408 			goto next;
5409 		}
5410 
5411 next:
5412 		add_command = strchr(add_command, ',');
5413 		if (add_command) {
5414 			*add_command = '\0';
5415 			add_command++;
5416 		}
5417 
5418 	}
5419 	if ((msr_num == 0) && (path == NULL)) {
5420 		fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
5421 		fail++;
5422 	}
5423 
5424 	/* generate default column header */
5425 	if (*name_buffer == '\0') {
5426 		if (width == 32)
5427 			sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5428 		else
5429 			sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5430 	}
5431 
5432 	if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
5433 		fail++;
5434 
5435 	if (fail) {
5436 		help();
5437 		exit(1);
5438 	}
5439 }
5440 
5441 int is_deferred_skip(char *name)
5442 {
5443 	int i;
5444 
5445 	for (i = 0; i < deferred_skip_index; ++i)
5446 		if (!strcmp(name, deferred_skip_names[i]))
5447 			return 1;
5448 	return 0;
5449 }
5450 
5451 void probe_sysfs(void)
5452 {
5453 	char path[64];
5454 	char name_buf[16];
5455 	FILE *input;
5456 	int state;
5457 	char *sp;
5458 
5459 	if (!DO_BIC(BIC_sysfs))
5460 		return;
5461 
5462 	for (state = 10; state >= 0; --state) {
5463 
5464 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5465 			base_cpu, state);
5466 		input = fopen(path, "r");
5467 		if (input == NULL)
5468 			continue;
5469 		if (!fgets(name_buf, sizeof(name_buf), input))
5470 			err(1, "%s: failed to read file", path);
5471 
5472 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5473 		sp = strchr(name_buf, '-');
5474 		if (!sp)
5475 			sp = strchrnul(name_buf, '\n');
5476 		*sp = '%';
5477 		*(sp + 1) = '\0';
5478 
5479 		fclose(input);
5480 
5481 		sprintf(path, "cpuidle/state%d/time", state);
5482 
5483 		if (is_deferred_skip(name_buf))
5484 			continue;
5485 
5486 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC,
5487 				FORMAT_PERCENT, SYSFS_PERCPU);
5488 	}
5489 
5490 	for (state = 10; state >= 0; --state) {
5491 
5492 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5493 			base_cpu, state);
5494 		input = fopen(path, "r");
5495 		if (input == NULL)
5496 			continue;
5497 		if (!fgets(name_buf, sizeof(name_buf), input))
5498 			err(1, "%s: failed to read file", path);
5499 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5500 		sp = strchr(name_buf, '-');
5501 		if (!sp)
5502 			sp = strchrnul(name_buf, '\n');
5503 		*sp = '\0';
5504 		fclose(input);
5505 
5506 		sprintf(path, "cpuidle/state%d/usage", state);
5507 
5508 		if (is_deferred_skip(name_buf))
5509 			continue;
5510 
5511 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS,
5512 				FORMAT_DELTA, SYSFS_PERCPU);
5513 	}
5514 
5515 }
5516 
5517 
5518 /*
5519  * parse cpuset with following syntax
5520  * 1,2,4..6,8-10 and set bits in cpu_subset
5521  */
5522 void parse_cpu_command(char *optarg)
5523 {
5524 	unsigned int start, end;
5525 	char *next;
5526 
5527 	if (!strcmp(optarg, "core")) {
5528 		if (cpu_subset)
5529 			goto error;
5530 		show_core_only++;
5531 		return;
5532 	}
5533 	if (!strcmp(optarg, "package")) {
5534 		if (cpu_subset)
5535 			goto error;
5536 		show_pkg_only++;
5537 		return;
5538 	}
5539 	if (show_core_only || show_pkg_only)
5540 		goto error;
5541 
5542 	cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
5543 	if (cpu_subset == NULL)
5544 		err(3, "CPU_ALLOC");
5545 	cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
5546 
5547 	CPU_ZERO_S(cpu_subset_size, cpu_subset);
5548 
5549 	next = optarg;
5550 
5551 	while (next && *next) {
5552 
5553 		if (*next == '-')	/* no negative cpu numbers */
5554 			goto error;
5555 
5556 		start = strtoul(next, &next, 10);
5557 
5558 		if (start >= CPU_SUBSET_MAXCPUS)
5559 			goto error;
5560 		CPU_SET_S(start, cpu_subset_size, cpu_subset);
5561 
5562 		if (*next == '\0')
5563 			break;
5564 
5565 		if (*next == ',') {
5566 			next += 1;
5567 			continue;
5568 		}
5569 
5570 		if (*next == '-') {
5571 			next += 1;	/* start range */
5572 		} else if (*next == '.') {
5573 			next += 1;
5574 			if (*next == '.')
5575 				next += 1;	/* start range */
5576 			else
5577 				goto error;
5578 		}
5579 
5580 		end = strtoul(next, &next, 10);
5581 		if (end <= start)
5582 			goto error;
5583 
5584 		while (++start <= end) {
5585 			if (start >= CPU_SUBSET_MAXCPUS)
5586 				goto error;
5587 			CPU_SET_S(start, cpu_subset_size, cpu_subset);
5588 		}
5589 
5590 		if (*next == ',')
5591 			next += 1;
5592 		else if (*next != '\0')
5593 			goto error;
5594 	}
5595 
5596 	return;
5597 
5598 error:
5599 	fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
5600 	help();
5601 	exit(-1);
5602 }
5603 
5604 
5605 void cmdline(int argc, char **argv)
5606 {
5607 	int opt;
5608 	int option_index = 0;
5609 	static struct option long_options[] = {
5610 		{"add",		required_argument,	0, 'a'},
5611 		{"cpu",		required_argument,	0, 'c'},
5612 		{"Dump",	no_argument,		0, 'D'},
5613 		{"debug",	no_argument,		0, 'd'},	/* internal, not documented */
5614 		{"enable",	required_argument,	0, 'e'},
5615 		{"interval",	required_argument,	0, 'i'},
5616 		{"num_iterations",	required_argument,	0, 'n'},
5617 		{"help",	no_argument,		0, 'h'},
5618 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
5619 		{"Joules",	no_argument,		0, 'J'},
5620 		{"list",	no_argument,		0, 'l'},
5621 		{"out",		required_argument,	0, 'o'},
5622 		{"quiet",	no_argument,		0, 'q'},
5623 		{"show",	required_argument,	0, 's'},
5624 		{"Summary",	no_argument,		0, 'S'},
5625 		{"TCC",		required_argument,	0, 'T'},
5626 		{"version",	no_argument,		0, 'v' },
5627 		{0,		0,			0,  0 }
5628 	};
5629 
5630 	progname = argv[0];
5631 
5632 	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v",
5633 				long_options, &option_index)) != -1) {
5634 		switch (opt) {
5635 		case 'a':
5636 			parse_add_command(optarg);
5637 			break;
5638 		case 'c':
5639 			parse_cpu_command(optarg);
5640 			break;
5641 		case 'D':
5642 			dump_only++;
5643 			break;
5644 		case 'e':
5645 			/* --enable specified counter */
5646 			bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST);
5647 			break;
5648 		case 'd':
5649 			debug++;
5650 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5651 			break;
5652 		case 'H':
5653 			/*
5654 			 * --hide: do not show those specified
5655 			 *  multiple invocations simply clear more bits in enabled mask
5656 			 */
5657 			bic_enabled &= ~bic_lookup(optarg, HIDE_LIST);
5658 			break;
5659 		case 'h':
5660 		default:
5661 			help();
5662 			exit(1);
5663 		case 'i':
5664 			{
5665 				double interval = strtod(optarg, NULL);
5666 
5667 				if (interval < 0.001) {
5668 					fprintf(outf, "interval %f seconds is too small\n",
5669 						interval);
5670 					exit(2);
5671 				}
5672 
5673 				interval_tv.tv_sec = interval_ts.tv_sec = interval;
5674 				interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000;
5675 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
5676 			}
5677 			break;
5678 		case 'J':
5679 			rapl_joules++;
5680 			break;
5681 		case 'l':
5682 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5683 			list_header_only++;
5684 			quiet++;
5685 			break;
5686 		case 'o':
5687 			outf = fopen_or_die(optarg, "w");
5688 			break;
5689 		case 'q':
5690 			quiet = 1;
5691 			break;
5692 		case 'n':
5693 			num_iterations = strtod(optarg, NULL);
5694 
5695 			if (num_iterations <= 0) {
5696 				fprintf(outf, "iterations %d should be positive number\n",
5697 					num_iterations);
5698 				exit(2);
5699 			}
5700 			break;
5701 		case 's':
5702 			/*
5703 			 * --show: show only those specified
5704 			 *  The 1st invocation will clear and replace the enabled mask
5705 			 *  subsequent invocations can add to it.
5706 			 */
5707 			if (shown == 0)
5708 				bic_enabled = bic_lookup(optarg, SHOW_LIST);
5709 			else
5710 				bic_enabled |= bic_lookup(optarg, SHOW_LIST);
5711 			shown = 1;
5712 			break;
5713 		case 'S':
5714 			summary_only++;
5715 			break;
5716 		case 'T':
5717 			tcc_activation_temp_override = atoi(optarg);
5718 			break;
5719 		case 'v':
5720 			print_version();
5721 			exit(0);
5722 			break;
5723 		}
5724 	}
5725 }
5726 
5727 int main(int argc, char **argv)
5728 {
5729 	outf = stderr;
5730 	cmdline(argc, argv);
5731 
5732 	if (!quiet)
5733 		print_version();
5734 
5735 	probe_sysfs();
5736 
5737 	turbostat_init();
5738 
5739 	/* dump counters and exit */
5740 	if (dump_only)
5741 		return get_and_dump_counters();
5742 
5743 	/* list header and exit */
5744 	if (list_header_only) {
5745 		print_header(",");
5746 		flush_output_stdout();
5747 		return 0;
5748 	}
5749 
5750 	/*
5751 	 * if any params left, it must be a command to fork
5752 	 */
5753 	if (argc - optind)
5754 		return fork_it(argv + optind);
5755 	else
5756 		turbostat_loop();
5757 
5758 	return 0;
5759 }
5760