1 /*
2  * turbostat -- show CPU frequency and C-state residency
3  * on modern Intel and AMD processors.
4  *
5  * Copyright (c) 2013 Intel Corporation.
6  * Len Brown <len.brown@intel.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program; if not, write to the Free Software Foundation, Inc.,
19  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20  */
21 
22 #define _GNU_SOURCE
23 #include MSRHEADER
24 #include INTEL_FAMILY_HEADER
25 #include <stdarg.h>
26 #include <stdio.h>
27 #include <err.h>
28 #include <unistd.h>
29 #include <sys/types.h>
30 #include <sys/wait.h>
31 #include <sys/stat.h>
32 #include <sys/select.h>
33 #include <sys/resource.h>
34 #include <fcntl.h>
35 #include <signal.h>
36 #include <sys/time.h>
37 #include <stdlib.h>
38 #include <getopt.h>
39 #include <dirent.h>
40 #include <string.h>
41 #include <ctype.h>
42 #include <sched.h>
43 #include <time.h>
44 #include <cpuid.h>
45 #include <linux/capability.h>
46 #include <errno.h>
47 #include <math.h>
48 
49 char *proc_stat = "/proc/stat";
50 FILE *outf;
51 int *fd_percpu;
52 struct timeval interval_tv = {5, 0};
53 struct timespec interval_ts = {5, 0};
54 struct timespec one_msec = {0, 1000000};
55 unsigned int num_iterations;
56 unsigned int debug;
57 unsigned int quiet;
58 unsigned int shown;
59 unsigned int sums_need_wide_columns;
60 unsigned int rapl_joules;
61 unsigned int summary_only;
62 unsigned int list_header_only;
63 unsigned int dump_only;
64 unsigned int do_snb_cstates;
65 unsigned int do_knl_cstates;
66 unsigned int do_slm_cstates;
67 unsigned int use_c1_residency_msr;
68 unsigned int has_aperf;
69 unsigned int has_epb;
70 unsigned int do_irtl_snb;
71 unsigned int do_irtl_hsw;
72 unsigned int units = 1000000;	/* MHz etc */
73 unsigned int genuine_intel;
74 unsigned int authentic_amd;
75 unsigned int max_level, max_extended_level;
76 unsigned int has_invariant_tsc;
77 unsigned int do_nhm_platform_info;
78 unsigned int no_MSR_MISC_PWR_MGMT;
79 unsigned int aperf_mperf_multiplier = 1;
80 double bclk;
81 double base_hz;
82 unsigned int has_base_hz;
83 double tsc_tweak = 1.0;
84 unsigned int show_pkg_only;
85 unsigned int show_core_only;
86 char *output_buffer, *outp;
87 unsigned int do_rapl;
88 unsigned int do_dts;
89 unsigned int do_ptm;
90 unsigned long long  gfx_cur_rc6_ms;
91 unsigned long long cpuidle_cur_cpu_lpi_us;
92 unsigned long long cpuidle_cur_sys_lpi_us;
93 unsigned int gfx_cur_mhz;
94 unsigned int tcc_activation_temp;
95 unsigned int tcc_activation_temp_override;
96 double rapl_power_units, rapl_time_units;
97 double rapl_dram_energy_units, rapl_energy_units;
98 double rapl_joule_counter_range;
99 unsigned int do_core_perf_limit_reasons;
100 unsigned int has_automatic_cstate_conversion;
101 unsigned int do_gfx_perf_limit_reasons;
102 unsigned int do_ring_perf_limit_reasons;
103 unsigned int crystal_hz;
104 unsigned long long tsc_hz;
105 int base_cpu;
106 double discover_bclk(unsigned int family, unsigned int model);
107 unsigned int has_hwp;	/* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
108 			/* IA32_HWP_REQUEST, IA32_HWP_STATUS */
109 unsigned int has_hwp_notify;		/* IA32_HWP_INTERRUPT */
110 unsigned int has_hwp_activity_window;	/* IA32_HWP_REQUEST[bits 41:32] */
111 unsigned int has_hwp_epp;		/* IA32_HWP_REQUEST[bits 31:24] */
112 unsigned int has_hwp_pkg;		/* IA32_HWP_REQUEST_PKG */
113 unsigned int has_misc_feature_control;
114 unsigned int first_counter_read = 1;
115 
116 #define RAPL_PKG		(1 << 0)
117 					/* 0x610 MSR_PKG_POWER_LIMIT */
118 					/* 0x611 MSR_PKG_ENERGY_STATUS */
119 #define RAPL_PKG_PERF_STATUS	(1 << 1)
120 					/* 0x613 MSR_PKG_PERF_STATUS */
121 #define RAPL_PKG_POWER_INFO	(1 << 2)
122 					/* 0x614 MSR_PKG_POWER_INFO */
123 
124 #define RAPL_DRAM		(1 << 3)
125 					/* 0x618 MSR_DRAM_POWER_LIMIT */
126 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
127 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
128 					/* 0x61b MSR_DRAM_PERF_STATUS */
129 #define RAPL_DRAM_POWER_INFO	(1 << 5)
130 					/* 0x61c MSR_DRAM_POWER_INFO */
131 
132 #define RAPL_CORES_POWER_LIMIT	(1 << 6)
133 					/* 0x638 MSR_PP0_POWER_LIMIT */
134 #define RAPL_CORE_POLICY	(1 << 7)
135 					/* 0x63a MSR_PP0_POLICY */
136 
137 #define RAPL_GFX		(1 << 8)
138 					/* 0x640 MSR_PP1_POWER_LIMIT */
139 					/* 0x641 MSR_PP1_ENERGY_STATUS */
140 					/* 0x642 MSR_PP1_POLICY */
141 
142 #define RAPL_CORES_ENERGY_STATUS	(1 << 9)
143 					/* 0x639 MSR_PP0_ENERGY_STATUS */
144 #define RAPL_PER_CORE_ENERGY	(1 << 10)
145 					/* Indicates cores energy collection is per-core,
146 					 * not per-package. */
147 #define RAPL_AMD_F17H		(1 << 11)
148 					/* 0xc0010299 MSR_RAPL_PWR_UNIT */
149 					/* 0xc001029a MSR_CORE_ENERGY_STAT */
150 					/* 0xc001029b MSR_PKG_ENERGY_STAT */
151 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
152 #define	TJMAX_DEFAULT	100
153 
154 /* MSRs that are not yet in the kernel-provided header. */
155 #define MSR_RAPL_PWR_UNIT	0xc0010299
156 #define MSR_CORE_ENERGY_STAT	0xc001029a
157 #define MSR_PKG_ENERGY_STAT	0xc001029b
158 
159 #define MAX(a, b) ((a) > (b) ? (a) : (b))
160 
161 /*
162  * buffer size used by sscanf() for added column names
163  * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
164  */
165 #define	NAME_BYTES 20
166 #define PATH_BYTES 128
167 
168 int backwards_count;
169 char *progname;
170 
171 #define CPU_SUBSET_MAXCPUS	1024	/* need to use before probe... */
172 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset;
173 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size;
174 #define MAX_ADDED_COUNTERS 8
175 #define MAX_ADDED_THREAD_COUNTERS 24
176 #define BITMASK_SIZE 32
177 
178 struct thread_data {
179 	struct timeval tv_begin;
180 	struct timeval tv_end;
181 	unsigned long long tsc;
182 	unsigned long long aperf;
183 	unsigned long long mperf;
184 	unsigned long long c1;
185 	unsigned long long  irq_count;
186 	unsigned int smi_count;
187 	unsigned int cpu_id;
188 	unsigned int apic_id;
189 	unsigned int x2apic_id;
190 	unsigned int flags;
191 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
192 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
193 	unsigned long long counter[MAX_ADDED_THREAD_COUNTERS];
194 } *thread_even, *thread_odd;
195 
196 struct core_data {
197 	unsigned long long c3;
198 	unsigned long long c6;
199 	unsigned long long c7;
200 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
201 	unsigned int core_temp_c;
202 	unsigned int core_energy;	/* MSR_CORE_ENERGY_STAT */
203 	unsigned int core_id;
204 	unsigned long long counter[MAX_ADDED_COUNTERS];
205 } *core_even, *core_odd;
206 
207 struct pkg_data {
208 	unsigned long long pc2;
209 	unsigned long long pc3;
210 	unsigned long long pc6;
211 	unsigned long long pc7;
212 	unsigned long long pc8;
213 	unsigned long long pc9;
214 	unsigned long long pc10;
215 	unsigned long long cpu_lpi;
216 	unsigned long long sys_lpi;
217 	unsigned long long pkg_wtd_core_c0;
218 	unsigned long long pkg_any_core_c0;
219 	unsigned long long pkg_any_gfxe_c0;
220 	unsigned long long pkg_both_core_gfxe_c0;
221 	long long gfx_rc6_ms;
222 	unsigned int gfx_mhz;
223 	unsigned int package_id;
224 	unsigned int energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
225 	unsigned int energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
226 	unsigned int energy_cores;	/* MSR_PP0_ENERGY_STATUS */
227 	unsigned int energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
228 	unsigned int rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
229 	unsigned int rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
230 	unsigned int pkg_temp_c;
231 	unsigned long long counter[MAX_ADDED_COUNTERS];
232 } *package_even, *package_odd;
233 
234 #define ODD_COUNTERS thread_odd, core_odd, package_odd
235 #define EVEN_COUNTERS thread_even, core_even, package_even
236 
237 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no)	      \
238 	((thread_base) +						      \
239 	 ((pkg_no) *							      \
240 	  topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \
241 	 ((node_no) * topo.cores_per_node * topo.threads_per_core) +	      \
242 	 ((core_no) * topo.threads_per_core) +				      \
243 	 (thread_no))
244 
245 #define GET_CORE(core_base, core_no, node_no, pkg_no)			\
246 	((core_base) +							\
247 	 ((pkg_no) *  topo.nodes_per_pkg * topo.cores_per_node) +	\
248 	 ((node_no) * topo.cores_per_node) +				\
249 	 (core_no))
250 
251 
252 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
253 
254 enum counter_scope {SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE};
255 enum counter_type {COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC};
256 enum counter_format {FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT};
257 
258 struct msr_counter {
259 	unsigned int msr_num;
260 	char name[NAME_BYTES];
261 	char path[PATH_BYTES];
262 	unsigned int width;
263 	enum counter_type type;
264 	enum counter_format format;
265 	struct msr_counter *next;
266 	unsigned int flags;
267 #define	FLAGS_HIDE	(1 << 0)
268 #define	FLAGS_SHOW	(1 << 1)
269 #define	SYSFS_PERCPU	(1 << 1)
270 };
271 
272 struct sys_counters {
273 	unsigned int added_thread_counters;
274 	unsigned int added_core_counters;
275 	unsigned int added_package_counters;
276 	struct msr_counter *tp;
277 	struct msr_counter *cp;
278 	struct msr_counter *pp;
279 } sys;
280 
281 struct system_summary {
282 	struct thread_data threads;
283 	struct core_data cores;
284 	struct pkg_data packages;
285 } average;
286 
287 struct cpu_topology {
288 	int physical_package_id;
289 	int die_id;
290 	int logical_cpu_id;
291 	int physical_node_id;
292 	int logical_node_id;	/* 0-based count within the package */
293 	int physical_core_id;
294 	int thread_id;
295 	cpu_set_t *put_ids; /* Processing Unit/Thread IDs */
296 } *cpus;
297 
298 struct topo_params {
299 	int num_packages;
300 	int num_die;
301 	int num_cpus;
302 	int num_cores;
303 	int max_cpu_num;
304 	int max_node_num;
305 	int nodes_per_pkg;
306 	int cores_per_node;
307 	int threads_per_core;
308 } topo;
309 
310 struct timeval tv_even, tv_odd, tv_delta;
311 
312 int *irq_column_2_cpu;	/* /proc/interrupts column numbers */
313 int *irqs_per_cpu;		/* indexed by cpu_num */
314 
315 void setup_all_buffers(void);
316 
317 int cpu_is_not_present(int cpu)
318 {
319 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
320 }
321 /*
322  * run func(thread, core, package) in topology order
323  * skip non-present cpus
324  */
325 
326 int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
327 	struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
328 {
329 	int retval, pkg_no, core_no, thread_no, node_no;
330 
331 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
332 		for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) {
333 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
334 				for (thread_no = 0; thread_no <
335 					topo.threads_per_core; ++thread_no) {
336 					struct thread_data *t;
337 					struct core_data *c;
338 					struct pkg_data *p;
339 
340 					t = GET_THREAD(thread_base, thread_no,
341 						       core_no, node_no,
342 						       pkg_no);
343 
344 					if (cpu_is_not_present(t->cpu_id))
345 						continue;
346 
347 					c = GET_CORE(core_base, core_no,
348 						     node_no, pkg_no);
349 					p = GET_PKG(pkg_base, pkg_no);
350 
351 					retval = func(t, c, p);
352 					if (retval)
353 						return retval;
354 				}
355 			}
356 		}
357 	}
358 	return 0;
359 }
360 
361 int cpu_migrate(int cpu)
362 {
363 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
364 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
365 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
366 		return -1;
367 	else
368 		return 0;
369 }
370 int get_msr_fd(int cpu)
371 {
372 	char pathname[32];
373 	int fd;
374 
375 	fd = fd_percpu[cpu];
376 
377 	if (fd)
378 		return fd;
379 
380 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
381 	fd = open(pathname, O_RDONLY);
382 	if (fd < 0)
383 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
384 
385 	fd_percpu[cpu] = fd;
386 
387 	return fd;
388 }
389 
390 int get_msr(int cpu, off_t offset, unsigned long long *msr)
391 {
392 	ssize_t retval;
393 
394 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
395 
396 	if (retval != sizeof *msr)
397 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
398 
399 	return 0;
400 }
401 
402 /*
403  * This list matches the column headers, except
404  * 1. built-in only, the sysfs counters are not here -- we learn of those at run-time
405  * 2. Core and CPU are moved to the end, we can't have strings that contain them
406  *    matching on them for --show and --hide.
407  */
408 struct msr_counter bic[] = {
409 	{ 0x0, "usec" },
410 	{ 0x0, "Time_Of_Day_Seconds" },
411 	{ 0x0, "Package" },
412 	{ 0x0, "Node" },
413 	{ 0x0, "Avg_MHz" },
414 	{ 0x0, "Busy%" },
415 	{ 0x0, "Bzy_MHz" },
416 	{ 0x0, "TSC_MHz" },
417 	{ 0x0, "IRQ" },
418 	{ 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL},
419 	{ 0x0, "sysfs" },
420 	{ 0x0, "CPU%c1" },
421 	{ 0x0, "CPU%c3" },
422 	{ 0x0, "CPU%c6" },
423 	{ 0x0, "CPU%c7" },
424 	{ 0x0, "ThreadC" },
425 	{ 0x0, "CoreTmp" },
426 	{ 0x0, "CoreCnt" },
427 	{ 0x0, "PkgTmp" },
428 	{ 0x0, "GFX%rc6" },
429 	{ 0x0, "GFXMHz" },
430 	{ 0x0, "Pkg%pc2" },
431 	{ 0x0, "Pkg%pc3" },
432 	{ 0x0, "Pkg%pc6" },
433 	{ 0x0, "Pkg%pc7" },
434 	{ 0x0, "Pkg%pc8" },
435 	{ 0x0, "Pkg%pc9" },
436 	{ 0x0, "Pk%pc10" },
437 	{ 0x0, "CPU%LPI" },
438 	{ 0x0, "SYS%LPI" },
439 	{ 0x0, "PkgWatt" },
440 	{ 0x0, "CorWatt" },
441 	{ 0x0, "GFXWatt" },
442 	{ 0x0, "PkgCnt" },
443 	{ 0x0, "RAMWatt" },
444 	{ 0x0, "PKG_%" },
445 	{ 0x0, "RAM_%" },
446 	{ 0x0, "Pkg_J" },
447 	{ 0x0, "Cor_J" },
448 	{ 0x0, "GFX_J" },
449 	{ 0x0, "RAM_J" },
450 	{ 0x0, "Mod%c6" },
451 	{ 0x0, "Totl%C0" },
452 	{ 0x0, "Any%C0" },
453 	{ 0x0, "GFX%C0" },
454 	{ 0x0, "CPUGFX%" },
455 	{ 0x0, "Core" },
456 	{ 0x0, "CPU" },
457 	{ 0x0, "APIC" },
458 	{ 0x0, "X2APIC" },
459 	{ 0x0, "Die" },
460 };
461 
462 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
463 #define	BIC_USEC	(1ULL << 0)
464 #define	BIC_TOD		(1ULL << 1)
465 #define	BIC_Package	(1ULL << 2)
466 #define	BIC_Node	(1ULL << 3)
467 #define	BIC_Avg_MHz	(1ULL << 4)
468 #define	BIC_Busy	(1ULL << 5)
469 #define	BIC_Bzy_MHz	(1ULL << 6)
470 #define	BIC_TSC_MHz	(1ULL << 7)
471 #define	BIC_IRQ		(1ULL << 8)
472 #define	BIC_SMI		(1ULL << 9)
473 #define	BIC_sysfs	(1ULL << 10)
474 #define	BIC_CPU_c1	(1ULL << 11)
475 #define	BIC_CPU_c3	(1ULL << 12)
476 #define	BIC_CPU_c6	(1ULL << 13)
477 #define	BIC_CPU_c7	(1ULL << 14)
478 #define	BIC_ThreadC	(1ULL << 15)
479 #define	BIC_CoreTmp	(1ULL << 16)
480 #define	BIC_CoreCnt	(1ULL << 17)
481 #define	BIC_PkgTmp	(1ULL << 18)
482 #define	BIC_GFX_rc6	(1ULL << 19)
483 #define	BIC_GFXMHz	(1ULL << 20)
484 #define	BIC_Pkgpc2	(1ULL << 21)
485 #define	BIC_Pkgpc3	(1ULL << 22)
486 #define	BIC_Pkgpc6	(1ULL << 23)
487 #define	BIC_Pkgpc7	(1ULL << 24)
488 #define	BIC_Pkgpc8	(1ULL << 25)
489 #define	BIC_Pkgpc9	(1ULL << 26)
490 #define	BIC_Pkgpc10	(1ULL << 27)
491 #define BIC_CPU_LPI	(1ULL << 28)
492 #define BIC_SYS_LPI	(1ULL << 29)
493 #define	BIC_PkgWatt	(1ULL << 30)
494 #define	BIC_CorWatt	(1ULL << 31)
495 #define	BIC_GFXWatt	(1ULL << 32)
496 #define	BIC_PkgCnt	(1ULL << 33)
497 #define	BIC_RAMWatt	(1ULL << 34)
498 #define	BIC_PKG__	(1ULL << 35)
499 #define	BIC_RAM__	(1ULL << 36)
500 #define	BIC_Pkg_J	(1ULL << 37)
501 #define	BIC_Cor_J	(1ULL << 38)
502 #define	BIC_GFX_J	(1ULL << 39)
503 #define	BIC_RAM_J	(1ULL << 40)
504 #define	BIC_Mod_c6	(1ULL << 41)
505 #define	BIC_Totl_c0	(1ULL << 42)
506 #define	BIC_Any_c0	(1ULL << 43)
507 #define	BIC_GFX_c0	(1ULL << 44)
508 #define	BIC_CPUGFX	(1ULL << 45)
509 #define	BIC_Core	(1ULL << 46)
510 #define	BIC_CPU		(1ULL << 47)
511 #define	BIC_APIC	(1ULL << 48)
512 #define	BIC_X2APIC	(1ULL << 49)
513 #define	BIC_Die		(1ULL << 50)
514 
515 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
516 
517 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT);
518 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC;
519 
520 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
521 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
522 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
523 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
524 
525 
526 #define MAX_DEFERRED 16
527 char *deferred_skip_names[MAX_DEFERRED];
528 int deferred_skip_index;
529 
530 /*
531  * HIDE_LIST - hide this list of counters, show the rest [default]
532  * SHOW_LIST - show this list of counters, hide the rest
533  */
534 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
535 
536 void help(void)
537 {
538 	fprintf(outf,
539 	"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
540 	"\n"
541 	"Turbostat forks the specified COMMAND and prints statistics\n"
542 	"when COMMAND completes.\n"
543 	"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
544 	"to print statistics, until interrupted.\n"
545 	"  -a, --add	add a counter\n"
546 	"		  eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
547 	"  -c, --cpu	cpu-set	limit output to summary plus cpu-set:\n"
548 	"		  {core | package | j,k,l..m,n-p }\n"
549 	"  -d, --debug	displays usec, Time_Of_Day_Seconds and more debugging\n"
550 	"  -D, --Dump	displays the raw counter values\n"
551 	"  -e, --enable	[all | column]\n"
552 	"		shows all or the specified disabled column\n"
553 	"  -H, --hide [column|column,column,...]\n"
554 	"		hide the specified column(s)\n"
555 	"  -i, --interval sec.subsec\n"
556 	"		Override default 5-second measurement interval\n"
557 	"  -J, --Joules	displays energy in Joules instead of Watts\n"
558 	"  -l, --list	list column headers only\n"
559 	"  -n, --num_iterations num\n"
560 	"		number of the measurement iterations\n"
561 	"  -o, --out file\n"
562 	"		create or truncate \"file\" for all output\n"
563 	"  -q, --quiet	skip decoding system configuration header\n"
564 	"  -s, --show [column|column,column,...]\n"
565 	"		show only the specified column(s)\n"
566 	"  -S, --Summary\n"
567 	"		limits output to 1-line system summary per interval\n"
568 	"  -T, --TCC temperature\n"
569 	"		sets the Thermal Control Circuit temperature in\n"
570 	"		  degrees Celsius\n"
571 	"  -h, --help	print this help message\n"
572 	"  -v, --version	print version information\n"
573 	"\n"
574 	"For more help, run \"man turbostat\"\n");
575 }
576 
577 /*
578  * bic_lookup
579  * for all the strings in comma separate name_list,
580  * set the approprate bit in return value.
581  */
582 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
583 {
584 	int i;
585 	unsigned long long retval = 0;
586 
587 	while (name_list) {
588 		char *comma;
589 
590 		comma = strchr(name_list, ',');
591 
592 		if (comma)
593 			*comma = '\0';
594 
595 		if (!strcmp(name_list, "all"))
596 			return ~0;
597 
598 		for (i = 0; i < MAX_BIC; ++i) {
599 			if (!strcmp(name_list, bic[i].name)) {
600 				retval |= (1ULL << i);
601 				break;
602 			}
603 		}
604 		if (i == MAX_BIC) {
605 			if (mode == SHOW_LIST) {
606 				fprintf(stderr, "Invalid counter name: %s\n", name_list);
607 				exit(-1);
608 			}
609 			deferred_skip_names[deferred_skip_index++] = name_list;
610 			if (debug)
611 				fprintf(stderr, "deferred \"%s\"\n", name_list);
612 			if (deferred_skip_index >= MAX_DEFERRED) {
613 				fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
614 					MAX_DEFERRED, name_list);
615 				help();
616 				exit(1);
617 			}
618 		}
619 
620 		name_list = comma;
621 		if (name_list)
622 			name_list++;
623 
624 	}
625 	return retval;
626 }
627 
628 
629 void print_header(char *delim)
630 {
631 	struct msr_counter *mp;
632 	int printed = 0;
633 
634 	if (DO_BIC(BIC_USEC))
635 		outp += sprintf(outp, "%susec", (printed++ ? delim : ""));
636 	if (DO_BIC(BIC_TOD))
637 		outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : ""));
638 	if (DO_BIC(BIC_Package))
639 		outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
640 	if (DO_BIC(BIC_Die))
641 		outp += sprintf(outp, "%sDie", (printed++ ? delim : ""));
642 	if (DO_BIC(BIC_Node))
643 		outp += sprintf(outp, "%sNode", (printed++ ? delim : ""));
644 	if (DO_BIC(BIC_Core))
645 		outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
646 	if (DO_BIC(BIC_CPU))
647 		outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
648 	if (DO_BIC(BIC_APIC))
649 		outp += sprintf(outp, "%sAPIC", (printed++ ? delim : ""));
650 	if (DO_BIC(BIC_X2APIC))
651 		outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : ""));
652 	if (DO_BIC(BIC_Avg_MHz))
653 		outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
654 	if (DO_BIC(BIC_Busy))
655 		outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
656 	if (DO_BIC(BIC_Bzy_MHz))
657 		outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
658 	if (DO_BIC(BIC_TSC_MHz))
659 		outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
660 
661 	if (DO_BIC(BIC_IRQ)) {
662 		if (sums_need_wide_columns)
663 			outp += sprintf(outp, "%s     IRQ", (printed++ ? delim : ""));
664 		else
665 			outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
666 	}
667 
668 	if (DO_BIC(BIC_SMI))
669 		outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
670 
671 	for (mp = sys.tp; mp; mp = mp->next) {
672 
673 		if (mp->format == FORMAT_RAW) {
674 			if (mp->width == 64)
675 				outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
676 			else
677 				outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
678 		} else {
679 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
680 				outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
681 			else
682 				outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
683 		}
684 	}
685 
686 	if (DO_BIC(BIC_CPU_c1))
687 		outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
688 	if (DO_BIC(BIC_CPU_c3))
689 		outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
690 	if (DO_BIC(BIC_CPU_c6))
691 		outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
692 	if (DO_BIC(BIC_CPU_c7))
693 		outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
694 
695 	if (DO_BIC(BIC_Mod_c6))
696 		outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
697 
698 	if (DO_BIC(BIC_CoreTmp))
699 		outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
700 
701 	if (do_rapl && !rapl_joules) {
702 		if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
703 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
704 	} else if (do_rapl && rapl_joules) {
705 		if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
706 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
707 	}
708 
709 	for (mp = sys.cp; mp; mp = mp->next) {
710 		if (mp->format == FORMAT_RAW) {
711 			if (mp->width == 64)
712 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
713 			else
714 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
715 		} else {
716 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
717 				outp += sprintf(outp, "%s%8s", delim, mp->name);
718 			else
719 				outp += sprintf(outp, "%s%s", delim, mp->name);
720 		}
721 	}
722 
723 	if (DO_BIC(BIC_PkgTmp))
724 		outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
725 
726 	if (DO_BIC(BIC_GFX_rc6))
727 		outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
728 
729 	if (DO_BIC(BIC_GFXMHz))
730 		outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
731 
732 	if (DO_BIC(BIC_Totl_c0))
733 		outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
734 	if (DO_BIC(BIC_Any_c0))
735 		outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
736 	if (DO_BIC(BIC_GFX_c0))
737 		outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
738 	if (DO_BIC(BIC_CPUGFX))
739 		outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
740 
741 	if (DO_BIC(BIC_Pkgpc2))
742 		outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
743 	if (DO_BIC(BIC_Pkgpc3))
744 		outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
745 	if (DO_BIC(BIC_Pkgpc6))
746 		outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
747 	if (DO_BIC(BIC_Pkgpc7))
748 		outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
749 	if (DO_BIC(BIC_Pkgpc8))
750 		outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
751 	if (DO_BIC(BIC_Pkgpc9))
752 		outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
753 	if (DO_BIC(BIC_Pkgpc10))
754 		outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
755 	if (DO_BIC(BIC_CPU_LPI))
756 		outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : ""));
757 	if (DO_BIC(BIC_SYS_LPI))
758 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
759 
760 	if (do_rapl && !rapl_joules) {
761 		if (DO_BIC(BIC_PkgWatt))
762 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
763 		if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
764 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
765 		if (DO_BIC(BIC_GFXWatt))
766 			outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
767 		if (DO_BIC(BIC_RAMWatt))
768 			outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
769 		if (DO_BIC(BIC_PKG__))
770 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
771 		if (DO_BIC(BIC_RAM__))
772 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
773 	} else if (do_rapl && rapl_joules) {
774 		if (DO_BIC(BIC_Pkg_J))
775 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
776 		if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
777 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
778 		if (DO_BIC(BIC_GFX_J))
779 			outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
780 		if (DO_BIC(BIC_RAM_J))
781 			outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
782 		if (DO_BIC(BIC_PKG__))
783 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
784 		if (DO_BIC(BIC_RAM__))
785 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
786 	}
787 	for (mp = sys.pp; mp; mp = mp->next) {
788 		if (mp->format == FORMAT_RAW) {
789 			if (mp->width == 64)
790 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
791 			else
792 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
793 		} else {
794 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
795 				outp += sprintf(outp, "%s%8s", delim, mp->name);
796 			else
797 				outp += sprintf(outp, "%s%s", delim, mp->name);
798 		}
799 	}
800 
801 	outp += sprintf(outp, "\n");
802 }
803 
804 int dump_counters(struct thread_data *t, struct core_data *c,
805 	struct pkg_data *p)
806 {
807 	int i;
808 	struct msr_counter *mp;
809 
810 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
811 
812 	if (t) {
813 		outp += sprintf(outp, "CPU: %d flags 0x%x\n",
814 			t->cpu_id, t->flags);
815 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
816 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
817 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
818 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
819 
820 		if (DO_BIC(BIC_IRQ))
821 			outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
822 		if (DO_BIC(BIC_SMI))
823 			outp += sprintf(outp, "SMI: %d\n", t->smi_count);
824 
825 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
826 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n",
827 				i, mp->msr_num, t->counter[i]);
828 		}
829 	}
830 
831 	if (c) {
832 		outp += sprintf(outp, "core: %d\n", c->core_id);
833 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
834 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
835 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
836 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
837 		outp += sprintf(outp, "Joules: %0X\n", c->core_energy);
838 
839 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
840 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n",
841 				i, mp->msr_num, c->counter[i]);
842 		}
843 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
844 	}
845 
846 	if (p) {
847 		outp += sprintf(outp, "package: %d\n", p->package_id);
848 
849 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
850 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
851 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
852 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
853 
854 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
855 		if (DO_BIC(BIC_Pkgpc3))
856 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
857 		if (DO_BIC(BIC_Pkgpc6))
858 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
859 		if (DO_BIC(BIC_Pkgpc7))
860 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
861 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
862 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
863 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
864 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
865 		outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi);
866 		outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi);
867 		outp += sprintf(outp, "Joules PKG: %0X\n", p->energy_pkg);
868 		outp += sprintf(outp, "Joules COR: %0X\n", p->energy_cores);
869 		outp += sprintf(outp, "Joules GFX: %0X\n", p->energy_gfx);
870 		outp += sprintf(outp, "Joules RAM: %0X\n", p->energy_dram);
871 		outp += sprintf(outp, "Throttle PKG: %0X\n",
872 			p->rapl_pkg_perf_status);
873 		outp += sprintf(outp, "Throttle RAM: %0X\n",
874 			p->rapl_dram_perf_status);
875 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
876 
877 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
878 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n",
879 				i, mp->msr_num, p->counter[i]);
880 		}
881 	}
882 
883 	outp += sprintf(outp, "\n");
884 
885 	return 0;
886 }
887 
888 /*
889  * column formatting convention & formats
890  */
891 int format_counters(struct thread_data *t, struct core_data *c,
892 	struct pkg_data *p)
893 {
894 	double interval_float, tsc;
895 	char *fmt8;
896 	int i;
897 	struct msr_counter *mp;
898 	char *delim = "\t";
899 	int printed = 0;
900 
901 	 /* if showing only 1st thread in core and this isn't one, bail out */
902 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
903 		return 0;
904 
905 	 /* if showing only 1st thread in pkg and this isn't one, bail out */
906 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
907 		return 0;
908 
909 	/*if not summary line and --cpu is used */
910 	if ((t != &average.threads) &&
911 		(cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
912 		return 0;
913 
914 	if (DO_BIC(BIC_USEC)) {
915 		/* on each row, print how many usec each timestamp took to gather */
916 		struct timeval tv;
917 
918 		timersub(&t->tv_end, &t->tv_begin, &tv);
919 		outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec);
920 	}
921 
922 	/* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */
923 	if (DO_BIC(BIC_TOD))
924 		outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec);
925 
926 	interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
927 
928 	tsc = t->tsc * tsc_tweak;
929 
930 	/* topo columns, print blanks on 1st (average) line */
931 	if (t == &average.threads) {
932 		if (DO_BIC(BIC_Package))
933 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
934 		if (DO_BIC(BIC_Die))
935 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
936 		if (DO_BIC(BIC_Node))
937 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
938 		if (DO_BIC(BIC_Core))
939 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
940 		if (DO_BIC(BIC_CPU))
941 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
942 		if (DO_BIC(BIC_APIC))
943 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
944 		if (DO_BIC(BIC_X2APIC))
945 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
946 	} else {
947 		if (DO_BIC(BIC_Package)) {
948 			if (p)
949 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
950 			else
951 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
952 		}
953 		if (DO_BIC(BIC_Die)) {
954 			if (c)
955 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id);
956 			else
957 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
958 		}
959 		if (DO_BIC(BIC_Node)) {
960 			if (t)
961 				outp += sprintf(outp, "%s%d",
962 						(printed++ ? delim : ""),
963 					      cpus[t->cpu_id].physical_node_id);
964 			else
965 				outp += sprintf(outp, "%s-",
966 						(printed++ ? delim : ""));
967 		}
968 		if (DO_BIC(BIC_Core)) {
969 			if (c)
970 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
971 			else
972 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
973 		}
974 		if (DO_BIC(BIC_CPU))
975 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
976 		if (DO_BIC(BIC_APIC))
977 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id);
978 		if (DO_BIC(BIC_X2APIC))
979 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id);
980 	}
981 
982 	if (DO_BIC(BIC_Avg_MHz))
983 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
984 			1.0 / units * t->aperf / interval_float);
985 
986 	if (DO_BIC(BIC_Busy))
987 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf/tsc);
988 
989 	if (DO_BIC(BIC_Bzy_MHz)) {
990 		if (has_base_hz)
991 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
992 		else
993 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
994 				tsc / units * t->aperf / t->mperf / interval_float);
995 	}
996 
997 	if (DO_BIC(BIC_TSC_MHz))
998 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
999 
1000 	/* IRQ */
1001 	if (DO_BIC(BIC_IRQ)) {
1002 		if (sums_need_wide_columns)
1003 			outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
1004 		else
1005 			outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
1006 	}
1007 
1008 	/* SMI */
1009 	if (DO_BIC(BIC_SMI))
1010 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
1011 
1012 	/* Added counters */
1013 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1014 		if (mp->format == FORMAT_RAW) {
1015 			if (mp->width == 32)
1016 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) t->counter[i]);
1017 			else
1018 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
1019 		} else if (mp->format == FORMAT_DELTA) {
1020 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1021 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
1022 			else
1023 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
1024 		} else if (mp->format == FORMAT_PERCENT) {
1025 			if (mp->type == COUNTER_USEC)
1026 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), t->counter[i]/interval_float/10000);
1027 			else
1028 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i]/tsc);
1029 		}
1030 	}
1031 
1032 	/* C1 */
1033 	if (DO_BIC(BIC_CPU_c1))
1034 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1/tsc);
1035 
1036 
1037 	/* print per-core data only for 1st thread in core */
1038 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1039 		goto done;
1040 
1041 	if (DO_BIC(BIC_CPU_c3))
1042 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3/tsc);
1043 	if (DO_BIC(BIC_CPU_c6))
1044 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6/tsc);
1045 	if (DO_BIC(BIC_CPU_c7))
1046 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7/tsc);
1047 
1048 	/* Mod%c6 */
1049 	if (DO_BIC(BIC_Mod_c6))
1050 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
1051 
1052 	if (DO_BIC(BIC_CoreTmp))
1053 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
1054 
1055 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1056 		if (mp->format == FORMAT_RAW) {
1057 			if (mp->width == 32)
1058 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) c->counter[i]);
1059 			else
1060 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
1061 		} else if (mp->format == FORMAT_DELTA) {
1062 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1063 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
1064 			else
1065 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
1066 		} else if (mp->format == FORMAT_PERCENT) {
1067 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i]/tsc);
1068 		}
1069 	}
1070 
1071 	/*
1072 	 * If measurement interval exceeds minimum RAPL Joule Counter range,
1073 	 * indicate that results are suspect by printing "**" in fraction place.
1074 	 */
1075 	if (interval_float < rapl_joule_counter_range)
1076 		fmt8 = "%s%.2f";
1077 	else
1078 		fmt8 = "%6.0f**";
1079 
1080 	if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
1081 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float);
1082 	if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
1083 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units);
1084 
1085 	/* print per-package data only for 1st core in package */
1086 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1087 		goto done;
1088 
1089 	/* PkgTmp */
1090 	if (DO_BIC(BIC_PkgTmp))
1091 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
1092 
1093 	/* GFXrc6 */
1094 	if (DO_BIC(BIC_GFX_rc6)) {
1095 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
1096 			outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
1097 		} else {
1098 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1099 				p->gfx_rc6_ms / 10.0 / interval_float);
1100 		}
1101 	}
1102 
1103 	/* GFXMHz */
1104 	if (DO_BIC(BIC_GFXMHz))
1105 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
1106 
1107 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
1108 	if (DO_BIC(BIC_Totl_c0))
1109 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0/tsc);
1110 	if (DO_BIC(BIC_Any_c0))
1111 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0/tsc);
1112 	if (DO_BIC(BIC_GFX_c0))
1113 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0/tsc);
1114 	if (DO_BIC(BIC_CPUGFX))
1115 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0/tsc);
1116 
1117 	if (DO_BIC(BIC_Pkgpc2))
1118 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2/tsc);
1119 	if (DO_BIC(BIC_Pkgpc3))
1120 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3/tsc);
1121 	if (DO_BIC(BIC_Pkgpc6))
1122 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6/tsc);
1123 	if (DO_BIC(BIC_Pkgpc7))
1124 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7/tsc);
1125 	if (DO_BIC(BIC_Pkgpc8))
1126 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8/tsc);
1127 	if (DO_BIC(BIC_Pkgpc9))
1128 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9/tsc);
1129 	if (DO_BIC(BIC_Pkgpc10))
1130 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10/tsc);
1131 
1132 	if (DO_BIC(BIC_CPU_LPI))
1133 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float);
1134 	if (DO_BIC(BIC_SYS_LPI))
1135 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float);
1136 
1137 	if (DO_BIC(BIC_PkgWatt))
1138 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
1139 	if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1140 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
1141 	if (DO_BIC(BIC_GFXWatt))
1142 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
1143 	if (DO_BIC(BIC_RAMWatt))
1144 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units / interval_float);
1145 	if (DO_BIC(BIC_Pkg_J))
1146 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
1147 	if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1148 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
1149 	if (DO_BIC(BIC_GFX_J))
1150 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
1151 	if (DO_BIC(BIC_RAM_J))
1152 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
1153 	if (DO_BIC(BIC_PKG__))
1154 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
1155 	if (DO_BIC(BIC_RAM__))
1156 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
1157 
1158 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1159 		if (mp->format == FORMAT_RAW) {
1160 			if (mp->width == 32)
1161 				outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int) p->counter[i]);
1162 			else
1163 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
1164 		} else if (mp->format == FORMAT_DELTA) {
1165 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1166 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
1167 			else
1168 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
1169 		} else if (mp->format == FORMAT_PERCENT) {
1170 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i]/tsc);
1171 		}
1172 	}
1173 
1174 done:
1175 	if (*(outp - 1) != '\n')
1176 		outp += sprintf(outp, "\n");
1177 
1178 	return 0;
1179 }
1180 
1181 void flush_output_stdout(void)
1182 {
1183 	FILE *filep;
1184 
1185 	if (outf == stderr)
1186 		filep = stdout;
1187 	else
1188 		filep = outf;
1189 
1190 	fputs(output_buffer, filep);
1191 	fflush(filep);
1192 
1193 	outp = output_buffer;
1194 }
1195 void flush_output_stderr(void)
1196 {
1197 	fputs(output_buffer, outf);
1198 	fflush(outf);
1199 	outp = output_buffer;
1200 }
1201 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1202 {
1203 	static int printed;
1204 
1205 	if (!printed || !summary_only)
1206 		print_header("\t");
1207 
1208 	format_counters(&average.threads, &average.cores, &average.packages);
1209 
1210 	printed = 1;
1211 
1212 	if (summary_only)
1213 		return;
1214 
1215 	for_all_cpus(format_counters, t, c, p);
1216 }
1217 
1218 #define DELTA_WRAP32(new, old)			\
1219 	if (new > old) {			\
1220 		old = new - old;		\
1221 	} else {				\
1222 		old = 0x100000000 + new - old;	\
1223 	}
1224 
1225 int
1226 delta_package(struct pkg_data *new, struct pkg_data *old)
1227 {
1228 	int i;
1229 	struct msr_counter *mp;
1230 
1231 
1232 	if (DO_BIC(BIC_Totl_c0))
1233 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1234 	if (DO_BIC(BIC_Any_c0))
1235 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1236 	if (DO_BIC(BIC_GFX_c0))
1237 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1238 	if (DO_BIC(BIC_CPUGFX))
1239 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1240 
1241 	old->pc2 = new->pc2 - old->pc2;
1242 	if (DO_BIC(BIC_Pkgpc3))
1243 		old->pc3 = new->pc3 - old->pc3;
1244 	if (DO_BIC(BIC_Pkgpc6))
1245 		old->pc6 = new->pc6 - old->pc6;
1246 	if (DO_BIC(BIC_Pkgpc7))
1247 		old->pc7 = new->pc7 - old->pc7;
1248 	old->pc8 = new->pc8 - old->pc8;
1249 	old->pc9 = new->pc9 - old->pc9;
1250 	old->pc10 = new->pc10 - old->pc10;
1251 	old->cpu_lpi = new->cpu_lpi - old->cpu_lpi;
1252 	old->sys_lpi = new->sys_lpi - old->sys_lpi;
1253 	old->pkg_temp_c = new->pkg_temp_c;
1254 
1255 	/* flag an error when rc6 counter resets/wraps */
1256 	if (old->gfx_rc6_ms >  new->gfx_rc6_ms)
1257 		old->gfx_rc6_ms = -1;
1258 	else
1259 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1260 
1261 	old->gfx_mhz = new->gfx_mhz;
1262 
1263 	DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
1264 	DELTA_WRAP32(new->energy_cores, old->energy_cores);
1265 	DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
1266 	DELTA_WRAP32(new->energy_dram, old->energy_dram);
1267 	DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
1268 	DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
1269 
1270 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1271 		if (mp->format == FORMAT_RAW)
1272 			old->counter[i] = new->counter[i];
1273 		else
1274 			old->counter[i] = new->counter[i] - old->counter[i];
1275 	}
1276 
1277 	return 0;
1278 }
1279 
1280 void
1281 delta_core(struct core_data *new, struct core_data *old)
1282 {
1283 	int i;
1284 	struct msr_counter *mp;
1285 
1286 	old->c3 = new->c3 - old->c3;
1287 	old->c6 = new->c6 - old->c6;
1288 	old->c7 = new->c7 - old->c7;
1289 	old->core_temp_c = new->core_temp_c;
1290 	old->mc6_us = new->mc6_us - old->mc6_us;
1291 
1292 	DELTA_WRAP32(new->core_energy, old->core_energy);
1293 
1294 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1295 		if (mp->format == FORMAT_RAW)
1296 			old->counter[i] = new->counter[i];
1297 		else
1298 			old->counter[i] = new->counter[i] - old->counter[i];
1299 	}
1300 }
1301 
1302 /*
1303  * old = new - old
1304  */
1305 int
1306 delta_thread(struct thread_data *new, struct thread_data *old,
1307 	struct core_data *core_delta)
1308 {
1309 	int i;
1310 	struct msr_counter *mp;
1311 
1312 	/* we run cpuid just the 1st time, copy the results */
1313 	if (DO_BIC(BIC_APIC))
1314 		new->apic_id = old->apic_id;
1315 	if (DO_BIC(BIC_X2APIC))
1316 		new->x2apic_id = old->x2apic_id;
1317 
1318 	/*
1319 	 * the timestamps from start of measurement interval are in "old"
1320 	 * the timestamp from end of measurement interval are in "new"
1321 	 * over-write old w/ new so we can print end of interval values
1322 	 */
1323 
1324 	old->tv_begin = new->tv_begin;
1325 	old->tv_end = new->tv_end;
1326 
1327 	old->tsc = new->tsc - old->tsc;
1328 
1329 	/* check for TSC < 1 Mcycles over interval */
1330 	if (old->tsc < (1000 * 1000))
1331 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1332 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1333 		     "or just the deep ones with \"processor.max_cstate=1\"");
1334 
1335 	old->c1 = new->c1 - old->c1;
1336 
1337 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1338 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1339 			old->aperf = new->aperf - old->aperf;
1340 			old->mperf = new->mperf - old->mperf;
1341 		} else {
1342 			return -1;
1343 		}
1344 	}
1345 
1346 
1347 	if (use_c1_residency_msr) {
1348 		/*
1349 		 * Some models have a dedicated C1 residency MSR,
1350 		 * which should be more accurate than the derivation below.
1351 		 */
1352 	} else {
1353 		/*
1354 		 * As counter collection is not atomic,
1355 		 * it is possible for mperf's non-halted cycles + idle states
1356 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1357 		 */
1358 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1359 			old->c1 = 0;
1360 		else {
1361 			/* normal case, derive c1 */
1362 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1363 				- core_delta->c6 - core_delta->c7;
1364 		}
1365 	}
1366 
1367 	if (old->mperf == 0) {
1368 		if (debug > 1)
1369 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1370 		old->mperf = 1;	/* divide by 0 protection */
1371 	}
1372 
1373 	if (DO_BIC(BIC_IRQ))
1374 		old->irq_count = new->irq_count - old->irq_count;
1375 
1376 	if (DO_BIC(BIC_SMI))
1377 		old->smi_count = new->smi_count - old->smi_count;
1378 
1379 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1380 		if (mp->format == FORMAT_RAW)
1381 			old->counter[i] = new->counter[i];
1382 		else
1383 			old->counter[i] = new->counter[i] - old->counter[i];
1384 	}
1385 	return 0;
1386 }
1387 
1388 int delta_cpu(struct thread_data *t, struct core_data *c,
1389 	struct pkg_data *p, struct thread_data *t2,
1390 	struct core_data *c2, struct pkg_data *p2)
1391 {
1392 	int retval = 0;
1393 
1394 	/* calculate core delta only for 1st thread in core */
1395 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1396 		delta_core(c, c2);
1397 
1398 	/* always calculate thread delta */
1399 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1400 	if (retval)
1401 		return retval;
1402 
1403 	/* calculate package delta only for 1st core in package */
1404 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1405 		retval = delta_package(p, p2);
1406 
1407 	return retval;
1408 }
1409 
1410 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1411 {
1412 	int i;
1413 	struct msr_counter  *mp;
1414 
1415 	t->tv_begin.tv_sec = 0;
1416 	t->tv_begin.tv_usec = 0;
1417 	t->tv_end.tv_sec = 0;
1418 	t->tv_end.tv_usec = 0;
1419 
1420 	t->tsc = 0;
1421 	t->aperf = 0;
1422 	t->mperf = 0;
1423 	t->c1 = 0;
1424 
1425 	t->irq_count = 0;
1426 	t->smi_count = 0;
1427 
1428 	/* tells format_counters to dump all fields from this set */
1429 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1430 
1431 	c->c3 = 0;
1432 	c->c6 = 0;
1433 	c->c7 = 0;
1434 	c->mc6_us = 0;
1435 	c->core_temp_c = 0;
1436 	c->core_energy = 0;
1437 
1438 	p->pkg_wtd_core_c0 = 0;
1439 	p->pkg_any_core_c0 = 0;
1440 	p->pkg_any_gfxe_c0 = 0;
1441 	p->pkg_both_core_gfxe_c0 = 0;
1442 
1443 	p->pc2 = 0;
1444 	if (DO_BIC(BIC_Pkgpc3))
1445 		p->pc3 = 0;
1446 	if (DO_BIC(BIC_Pkgpc6))
1447 		p->pc6 = 0;
1448 	if (DO_BIC(BIC_Pkgpc7))
1449 		p->pc7 = 0;
1450 	p->pc8 = 0;
1451 	p->pc9 = 0;
1452 	p->pc10 = 0;
1453 	p->cpu_lpi = 0;
1454 	p->sys_lpi = 0;
1455 
1456 	p->energy_pkg = 0;
1457 	p->energy_dram = 0;
1458 	p->energy_cores = 0;
1459 	p->energy_gfx = 0;
1460 	p->rapl_pkg_perf_status = 0;
1461 	p->rapl_dram_perf_status = 0;
1462 	p->pkg_temp_c = 0;
1463 
1464 	p->gfx_rc6_ms = 0;
1465 	p->gfx_mhz = 0;
1466 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1467 		t->counter[i] = 0;
1468 
1469 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1470 		c->counter[i] = 0;
1471 
1472 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1473 		p->counter[i] = 0;
1474 }
1475 int sum_counters(struct thread_data *t, struct core_data *c,
1476 	struct pkg_data *p)
1477 {
1478 	int i;
1479 	struct msr_counter *mp;
1480 
1481 	/* copy un-changing apic_id's */
1482 	if (DO_BIC(BIC_APIC))
1483 		average.threads.apic_id = t->apic_id;
1484 	if (DO_BIC(BIC_X2APIC))
1485 		average.threads.x2apic_id = t->x2apic_id;
1486 
1487 	/* remember first tv_begin */
1488 	if (average.threads.tv_begin.tv_sec == 0)
1489 		average.threads.tv_begin = t->tv_begin;
1490 
1491 	/* remember last tv_end */
1492 	average.threads.tv_end = t->tv_end;
1493 
1494 	average.threads.tsc += t->tsc;
1495 	average.threads.aperf += t->aperf;
1496 	average.threads.mperf += t->mperf;
1497 	average.threads.c1 += t->c1;
1498 
1499 	average.threads.irq_count += t->irq_count;
1500 	average.threads.smi_count += t->smi_count;
1501 
1502 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1503 		if (mp->format == FORMAT_RAW)
1504 			continue;
1505 		average.threads.counter[i] += t->counter[i];
1506 	}
1507 
1508 	/* sum per-core values only for 1st thread in core */
1509 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1510 		return 0;
1511 
1512 	average.cores.c3 += c->c3;
1513 	average.cores.c6 += c->c6;
1514 	average.cores.c7 += c->c7;
1515 	average.cores.mc6_us += c->mc6_us;
1516 
1517 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1518 
1519 	average.cores.core_energy += c->core_energy;
1520 
1521 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1522 		if (mp->format == FORMAT_RAW)
1523 			continue;
1524 		average.cores.counter[i] += c->counter[i];
1525 	}
1526 
1527 	/* sum per-pkg values only for 1st core in pkg */
1528 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1529 		return 0;
1530 
1531 	if (DO_BIC(BIC_Totl_c0))
1532 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1533 	if (DO_BIC(BIC_Any_c0))
1534 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1535 	if (DO_BIC(BIC_GFX_c0))
1536 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1537 	if (DO_BIC(BIC_CPUGFX))
1538 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1539 
1540 	average.packages.pc2 += p->pc2;
1541 	if (DO_BIC(BIC_Pkgpc3))
1542 		average.packages.pc3 += p->pc3;
1543 	if (DO_BIC(BIC_Pkgpc6))
1544 		average.packages.pc6 += p->pc6;
1545 	if (DO_BIC(BIC_Pkgpc7))
1546 		average.packages.pc7 += p->pc7;
1547 	average.packages.pc8 += p->pc8;
1548 	average.packages.pc9 += p->pc9;
1549 	average.packages.pc10 += p->pc10;
1550 
1551 	average.packages.cpu_lpi = p->cpu_lpi;
1552 	average.packages.sys_lpi = p->sys_lpi;
1553 
1554 	average.packages.energy_pkg += p->energy_pkg;
1555 	average.packages.energy_dram += p->energy_dram;
1556 	average.packages.energy_cores += p->energy_cores;
1557 	average.packages.energy_gfx += p->energy_gfx;
1558 
1559 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1560 	average.packages.gfx_mhz = p->gfx_mhz;
1561 
1562 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1563 
1564 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1565 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1566 
1567 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1568 		if (mp->format == FORMAT_RAW)
1569 			continue;
1570 		average.packages.counter[i] += p->counter[i];
1571 	}
1572 	return 0;
1573 }
1574 /*
1575  * sum the counters for all cpus in the system
1576  * compute the weighted average
1577  */
1578 void compute_average(struct thread_data *t, struct core_data *c,
1579 	struct pkg_data *p)
1580 {
1581 	int i;
1582 	struct msr_counter *mp;
1583 
1584 	clear_counters(&average.threads, &average.cores, &average.packages);
1585 
1586 	for_all_cpus(sum_counters, t, c, p);
1587 
1588 	average.threads.tsc /= topo.num_cpus;
1589 	average.threads.aperf /= topo.num_cpus;
1590 	average.threads.mperf /= topo.num_cpus;
1591 	average.threads.c1 /= topo.num_cpus;
1592 
1593 	if (average.threads.irq_count > 9999999)
1594 		sums_need_wide_columns = 1;
1595 
1596 	average.cores.c3 /= topo.num_cores;
1597 	average.cores.c6 /= topo.num_cores;
1598 	average.cores.c7 /= topo.num_cores;
1599 	average.cores.mc6_us /= topo.num_cores;
1600 
1601 	if (DO_BIC(BIC_Totl_c0))
1602 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1603 	if (DO_BIC(BIC_Any_c0))
1604 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1605 	if (DO_BIC(BIC_GFX_c0))
1606 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1607 	if (DO_BIC(BIC_CPUGFX))
1608 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1609 
1610 	average.packages.pc2 /= topo.num_packages;
1611 	if (DO_BIC(BIC_Pkgpc3))
1612 		average.packages.pc3 /= topo.num_packages;
1613 	if (DO_BIC(BIC_Pkgpc6))
1614 		average.packages.pc6 /= topo.num_packages;
1615 	if (DO_BIC(BIC_Pkgpc7))
1616 		average.packages.pc7 /= topo.num_packages;
1617 
1618 	average.packages.pc8 /= topo.num_packages;
1619 	average.packages.pc9 /= topo.num_packages;
1620 	average.packages.pc10 /= topo.num_packages;
1621 
1622 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1623 		if (mp->format == FORMAT_RAW)
1624 			continue;
1625 		if (mp->type == COUNTER_ITEMS) {
1626 			if (average.threads.counter[i] > 9999999)
1627 				sums_need_wide_columns = 1;
1628 			continue;
1629 		}
1630 		average.threads.counter[i] /= topo.num_cpus;
1631 	}
1632 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1633 		if (mp->format == FORMAT_RAW)
1634 			continue;
1635 		if (mp->type == COUNTER_ITEMS) {
1636 			if (average.cores.counter[i] > 9999999)
1637 				sums_need_wide_columns = 1;
1638 		}
1639 		average.cores.counter[i] /= topo.num_cores;
1640 	}
1641 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1642 		if (mp->format == FORMAT_RAW)
1643 			continue;
1644 		if (mp->type == COUNTER_ITEMS) {
1645 			if (average.packages.counter[i] > 9999999)
1646 				sums_need_wide_columns = 1;
1647 		}
1648 		average.packages.counter[i] /= topo.num_packages;
1649 	}
1650 }
1651 
1652 static unsigned long long rdtsc(void)
1653 {
1654 	unsigned int low, high;
1655 
1656 	asm volatile("rdtsc" : "=a" (low), "=d" (high));
1657 
1658 	return low | ((unsigned long long)high) << 32;
1659 }
1660 
1661 /*
1662  * Open a file, and exit on failure
1663  */
1664 FILE *fopen_or_die(const char *path, const char *mode)
1665 {
1666 	FILE *filep = fopen(path, mode);
1667 
1668 	if (!filep)
1669 		err(1, "%s: open failed", path);
1670 	return filep;
1671 }
1672 /*
1673  * snapshot_sysfs_counter()
1674  *
1675  * return snapshot of given counter
1676  */
1677 unsigned long long snapshot_sysfs_counter(char *path)
1678 {
1679 	FILE *fp;
1680 	int retval;
1681 	unsigned long long counter;
1682 
1683 	fp = fopen_or_die(path, "r");
1684 
1685 	retval = fscanf(fp, "%lld", &counter);
1686 	if (retval != 1)
1687 		err(1, "snapshot_sysfs_counter(%s)", path);
1688 
1689 	fclose(fp);
1690 
1691 	return counter;
1692 }
1693 
1694 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1695 {
1696 	if (mp->msr_num != 0) {
1697 		if (get_msr(cpu, mp->msr_num, counterp))
1698 			return -1;
1699 	} else {
1700 		char path[128 + PATH_BYTES];
1701 
1702 		if (mp->flags & SYSFS_PERCPU) {
1703 			sprintf(path, "/sys/devices/system/cpu/cpu%d/%s",
1704 				 cpu, mp->path);
1705 
1706 			*counterp = snapshot_sysfs_counter(path);
1707 		} else {
1708 			*counterp = snapshot_sysfs_counter(mp->path);
1709 		}
1710 	}
1711 
1712 	return 0;
1713 }
1714 
1715 void get_apic_id(struct thread_data *t)
1716 {
1717 	unsigned int eax, ebx, ecx, edx;
1718 
1719 	if (DO_BIC(BIC_APIC)) {
1720 		eax = ebx = ecx = edx = 0;
1721 		__cpuid(1, eax, ebx, ecx, edx);
1722 
1723 		t->apic_id = (ebx >> 24) & 0xff;
1724 	}
1725 
1726 	if (!DO_BIC(BIC_X2APIC))
1727 		return;
1728 
1729 	if (authentic_amd) {
1730 		unsigned int topology_extensions;
1731 
1732 		if (max_extended_level < 0x8000001e)
1733 			return;
1734 
1735 		eax = ebx = ecx = edx = 0;
1736 		__cpuid(0x80000001, eax, ebx, ecx, edx);
1737 			topology_extensions = ecx & (1 << 22);
1738 
1739 		if (topology_extensions == 0)
1740 			return;
1741 
1742 		eax = ebx = ecx = edx = 0;
1743 		__cpuid(0x8000001e, eax, ebx, ecx, edx);
1744 
1745 		t->x2apic_id = eax;
1746 		return;
1747 	}
1748 
1749 	if (!genuine_intel)
1750 		return;
1751 
1752 	if (max_level < 0xb)
1753 		return;
1754 
1755 	ecx = 0;
1756 	__cpuid(0xb, eax, ebx, ecx, edx);
1757 	t->x2apic_id = edx;
1758 
1759 	if (debug && (t->apic_id != (t->x2apic_id & 0xff)))
1760 		fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n",
1761 				t->cpu_id, t->apic_id, t->x2apic_id);
1762 }
1763 
1764 /*
1765  * get_counters(...)
1766  * migrate to cpu
1767  * acquire and record local counters for that cpu
1768  */
1769 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1770 {
1771 	int cpu = t->cpu_id;
1772 	unsigned long long msr;
1773 	int aperf_mperf_retry_count = 0;
1774 	struct msr_counter *mp;
1775 	int i;
1776 
1777 	gettimeofday(&t->tv_begin, (struct timezone *)NULL);
1778 
1779 	if (cpu_migrate(cpu)) {
1780 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
1781 		return -1;
1782 	}
1783 
1784 	if (first_counter_read)
1785 		get_apic_id(t);
1786 retry:
1787 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
1788 
1789 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz)) {
1790 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
1791 
1792 		/*
1793 		 * The TSC, APERF and MPERF must be read together for
1794 		 * APERF/MPERF and MPERF/TSC to give accurate results.
1795 		 *
1796 		 * Unfortunately, APERF and MPERF are read by
1797 		 * individual system call, so delays may occur
1798 		 * between them.  If the time to read them
1799 		 * varies by a large amount, we re-read them.
1800 		 */
1801 
1802 		/*
1803 		 * This initial dummy APERF read has been seen to
1804 		 * reduce jitter in the subsequent reads.
1805 		 */
1806 
1807 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1808 			return -3;
1809 
1810 		t->tsc = rdtsc();	/* re-read close to APERF */
1811 
1812 		tsc_before = t->tsc;
1813 
1814 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
1815 			return -3;
1816 
1817 		tsc_between = rdtsc();
1818 
1819 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
1820 			return -4;
1821 
1822 		tsc_after = rdtsc();
1823 
1824 		aperf_time = tsc_between - tsc_before;
1825 		mperf_time = tsc_after - tsc_between;
1826 
1827 		/*
1828 		 * If the system call latency to read APERF and MPERF
1829 		 * differ by more than 2x, then try again.
1830 		 */
1831 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
1832 			aperf_mperf_retry_count++;
1833 			if (aperf_mperf_retry_count < 5)
1834 				goto retry;
1835 			else
1836 				warnx("cpu%d jitter %lld %lld",
1837 					cpu, aperf_time, mperf_time);
1838 		}
1839 		aperf_mperf_retry_count = 0;
1840 
1841 		t->aperf = t->aperf * aperf_mperf_multiplier;
1842 		t->mperf = t->mperf * aperf_mperf_multiplier;
1843 	}
1844 
1845 	if (DO_BIC(BIC_IRQ))
1846 		t->irq_count = irqs_per_cpu[cpu];
1847 	if (DO_BIC(BIC_SMI)) {
1848 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
1849 			return -5;
1850 		t->smi_count = msr & 0xFFFFFFFF;
1851 	}
1852 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
1853 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
1854 			return -6;
1855 	}
1856 
1857 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1858 		if (get_mp(cpu, mp, &t->counter[i]))
1859 			return -10;
1860 	}
1861 
1862 	/* collect core counters only for 1st thread in core */
1863 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1864 		goto done;
1865 
1866 	if (DO_BIC(BIC_CPU_c3)) {
1867 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
1868 			return -6;
1869 	}
1870 
1871 	if (DO_BIC(BIC_CPU_c6) && !do_knl_cstates) {
1872 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
1873 			return -7;
1874 	} else if (do_knl_cstates) {
1875 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
1876 			return -7;
1877 	}
1878 
1879 	if (DO_BIC(BIC_CPU_c7))
1880 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
1881 			return -8;
1882 
1883 	if (DO_BIC(BIC_Mod_c6))
1884 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
1885 			return -8;
1886 
1887 	if (DO_BIC(BIC_CoreTmp)) {
1888 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
1889 			return -9;
1890 		c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1891 	}
1892 
1893 	if (do_rapl & RAPL_AMD_F17H) {
1894 		if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr))
1895 			return -14;
1896 		c->core_energy = msr & 0xFFFFFFFF;
1897 	}
1898 
1899 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1900 		if (get_mp(cpu, mp, &c->counter[i]))
1901 			return -10;
1902 	}
1903 
1904 	/* collect package counters only for 1st core in package */
1905 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1906 		goto done;
1907 
1908 	if (DO_BIC(BIC_Totl_c0)) {
1909 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
1910 			return -10;
1911 	}
1912 	if (DO_BIC(BIC_Any_c0)) {
1913 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
1914 			return -11;
1915 	}
1916 	if (DO_BIC(BIC_GFX_c0)) {
1917 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
1918 			return -12;
1919 	}
1920 	if (DO_BIC(BIC_CPUGFX)) {
1921 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
1922 			return -13;
1923 	}
1924 	if (DO_BIC(BIC_Pkgpc3))
1925 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
1926 			return -9;
1927 	if (DO_BIC(BIC_Pkgpc6)) {
1928 		if (do_slm_cstates) {
1929 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
1930 				return -10;
1931 		} else {
1932 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
1933 				return -10;
1934 		}
1935 	}
1936 
1937 	if (DO_BIC(BIC_Pkgpc2))
1938 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
1939 			return -11;
1940 	if (DO_BIC(BIC_Pkgpc7))
1941 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
1942 			return -12;
1943 	if (DO_BIC(BIC_Pkgpc8))
1944 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
1945 			return -13;
1946 	if (DO_BIC(BIC_Pkgpc9))
1947 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
1948 			return -13;
1949 	if (DO_BIC(BIC_Pkgpc10))
1950 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
1951 			return -13;
1952 
1953 	if (DO_BIC(BIC_CPU_LPI))
1954 		p->cpu_lpi = cpuidle_cur_cpu_lpi_us;
1955 	if (DO_BIC(BIC_SYS_LPI))
1956 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
1957 
1958 	if (do_rapl & RAPL_PKG) {
1959 		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
1960 			return -13;
1961 		p->energy_pkg = msr & 0xFFFFFFFF;
1962 	}
1963 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
1964 		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
1965 			return -14;
1966 		p->energy_cores = msr & 0xFFFFFFFF;
1967 	}
1968 	if (do_rapl & RAPL_DRAM) {
1969 		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
1970 			return -15;
1971 		p->energy_dram = msr & 0xFFFFFFFF;
1972 	}
1973 	if (do_rapl & RAPL_GFX) {
1974 		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
1975 			return -16;
1976 		p->energy_gfx = msr & 0xFFFFFFFF;
1977 	}
1978 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
1979 		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
1980 			return -16;
1981 		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
1982 	}
1983 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
1984 		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
1985 			return -16;
1986 		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
1987 	}
1988 	if (do_rapl & RAPL_AMD_F17H) {
1989 		if (get_msr(cpu, MSR_PKG_ENERGY_STAT, &msr))
1990 			return -13;
1991 		p->energy_pkg = msr & 0xFFFFFFFF;
1992 	}
1993 	if (DO_BIC(BIC_PkgTmp)) {
1994 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
1995 			return -17;
1996 		p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
1997 	}
1998 
1999 	if (DO_BIC(BIC_GFX_rc6))
2000 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
2001 
2002 	if (DO_BIC(BIC_GFXMHz))
2003 		p->gfx_mhz = gfx_cur_mhz;
2004 
2005 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
2006 		if (get_mp(cpu, mp, &p->counter[i]))
2007 			return -10;
2008 	}
2009 done:
2010 	gettimeofday(&t->tv_end, (struct timezone *)NULL);
2011 
2012 	return 0;
2013 }
2014 
2015 /*
2016  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
2017  * If you change the values, note they are used both in comparisons
2018  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
2019  */
2020 
2021 #define PCLUKN 0 /* Unknown */
2022 #define PCLRSV 1 /* Reserved */
2023 #define PCL__0 2 /* PC0 */
2024 #define PCL__1 3 /* PC1 */
2025 #define PCL__2 4 /* PC2 */
2026 #define PCL__3 5 /* PC3 */
2027 #define PCL__4 6 /* PC4 */
2028 #define PCL__6 7 /* PC6 */
2029 #define PCL_6N 8 /* PC6 No Retention */
2030 #define PCL_6R 9 /* PC6 Retention */
2031 #define PCL__7 10 /* PC7 */
2032 #define PCL_7S 11 /* PC7 Shrink */
2033 #define PCL__8 12 /* PC8 */
2034 #define PCL__9 13 /* PC9 */
2035 #define PCL_10 14 /* PC10 */
2036 #define PCLUNL 15 /* Unlimited */
2037 
2038 int pkg_cstate_limit = PCLUKN;
2039 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
2040 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited"};
2041 
2042 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};
2043 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};
2044 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};
2045 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};
2046 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};
2047 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};
2048 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};
2049 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};
2050 
2051 
2052 static void
2053 calculate_tsc_tweak()
2054 {
2055 	tsc_tweak = base_hz / tsc_hz;
2056 }
2057 
2058 static void
2059 dump_nhm_platform_info(void)
2060 {
2061 	unsigned long long msr;
2062 	unsigned int ratio;
2063 
2064 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2065 
2066 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
2067 
2068 	ratio = (msr >> 40) & 0xFF;
2069 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n",
2070 		ratio, bclk, ratio * bclk);
2071 
2072 	ratio = (msr >> 8) & 0xFF;
2073 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2074 		ratio, bclk, ratio * bclk);
2075 
2076 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
2077 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
2078 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
2079 
2080 	return;
2081 }
2082 
2083 static void
2084 dump_hsw_turbo_ratio_limits(void)
2085 {
2086 	unsigned long long msr;
2087 	unsigned int ratio;
2088 
2089 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
2090 
2091 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
2092 
2093 	ratio = (msr >> 8) & 0xFF;
2094 	if (ratio)
2095 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n",
2096 			ratio, bclk, ratio * bclk);
2097 
2098 	ratio = (msr >> 0) & 0xFF;
2099 	if (ratio)
2100 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n",
2101 			ratio, bclk, ratio * bclk);
2102 	return;
2103 }
2104 
2105 static void
2106 dump_ivt_turbo_ratio_limits(void)
2107 {
2108 	unsigned long long msr;
2109 	unsigned int ratio;
2110 
2111 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
2112 
2113 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
2114 
2115 	ratio = (msr >> 56) & 0xFF;
2116 	if (ratio)
2117 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n",
2118 			ratio, bclk, ratio * bclk);
2119 
2120 	ratio = (msr >> 48) & 0xFF;
2121 	if (ratio)
2122 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n",
2123 			ratio, bclk, ratio * bclk);
2124 
2125 	ratio = (msr >> 40) & 0xFF;
2126 	if (ratio)
2127 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n",
2128 			ratio, bclk, ratio * bclk);
2129 
2130 	ratio = (msr >> 32) & 0xFF;
2131 	if (ratio)
2132 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n",
2133 			ratio, bclk, ratio * bclk);
2134 
2135 	ratio = (msr >> 24) & 0xFF;
2136 	if (ratio)
2137 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n",
2138 			ratio, bclk, ratio * bclk);
2139 
2140 	ratio = (msr >> 16) & 0xFF;
2141 	if (ratio)
2142 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n",
2143 			ratio, bclk, ratio * bclk);
2144 
2145 	ratio = (msr >> 8) & 0xFF;
2146 	if (ratio)
2147 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n",
2148 			ratio, bclk, ratio * bclk);
2149 
2150 	ratio = (msr >> 0) & 0xFF;
2151 	if (ratio)
2152 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n",
2153 			ratio, bclk, ratio * bclk);
2154 	return;
2155 }
2156 int has_turbo_ratio_group_limits(int family, int model)
2157 {
2158 
2159 	if (!genuine_intel)
2160 		return 0;
2161 
2162 	switch (model) {
2163 	case INTEL_FAM6_ATOM_GOLDMONT:
2164 	case INTEL_FAM6_SKYLAKE_X:
2165 	case INTEL_FAM6_ATOM_GOLDMONT_X:
2166 		return 1;
2167 	}
2168 	return 0;
2169 }
2170 
2171 static void
2172 dump_turbo_ratio_limits(int family, int model)
2173 {
2174 	unsigned long long msr, core_counts;
2175 	unsigned int ratio, group_size;
2176 
2177 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2178 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
2179 
2180 	if (has_turbo_ratio_group_limits(family, model)) {
2181 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
2182 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
2183 	} else {
2184 		core_counts = 0x0807060504030201;
2185 	}
2186 
2187 	ratio = (msr >> 56) & 0xFF;
2188 	group_size = (core_counts >> 56) & 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 >> 48) & 0xFF;
2194 	group_size = (core_counts >> 48) & 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 >> 40) & 0xFF;
2200 	group_size = (core_counts >> 40) & 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 >> 32) & 0xFF;
2206 	group_size = (core_counts >> 32) & 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 >> 24) & 0xFF;
2212 	group_size = (core_counts >> 24) & 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 >> 16) & 0xFF;
2218 	group_size = (core_counts >> 16) & 0xFF;
2219 	if (ratio)
2220 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2221 			ratio, bclk, ratio * bclk, group_size);
2222 
2223 	ratio = (msr >> 8) & 0xFF;
2224 	group_size = (core_counts >> 8) & 0xFF;
2225 	if (ratio)
2226 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2227 			ratio, bclk, ratio * bclk, group_size);
2228 
2229 	ratio = (msr >> 0) & 0xFF;
2230 	group_size = (core_counts >> 0) & 0xFF;
2231 	if (ratio)
2232 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2233 			ratio, bclk, ratio * bclk, group_size);
2234 	return;
2235 }
2236 
2237 static void
2238 dump_atom_turbo_ratio_limits(void)
2239 {
2240 	unsigned long long msr;
2241 	unsigned int ratio;
2242 
2243 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
2244 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2245 
2246 	ratio = (msr >> 0) & 0x3F;
2247 	if (ratio)
2248 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n",
2249 			ratio, bclk, ratio * bclk);
2250 
2251 	ratio = (msr >> 8) & 0x3F;
2252 	if (ratio)
2253 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n",
2254 			ratio, bclk, ratio * bclk);
2255 
2256 	ratio = (msr >> 16) & 0x3F;
2257 	if (ratio)
2258 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n",
2259 			ratio, bclk, ratio * bclk);
2260 
2261 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
2262 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2263 
2264 	ratio = (msr >> 24) & 0x3F;
2265 	if (ratio)
2266 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n",
2267 			ratio, bclk, ratio * bclk);
2268 
2269 	ratio = (msr >> 16) & 0x3F;
2270 	if (ratio)
2271 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n",
2272 			ratio, bclk, ratio * bclk);
2273 
2274 	ratio = (msr >> 8) & 0x3F;
2275 	if (ratio)
2276 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n",
2277 			ratio, bclk, ratio * bclk);
2278 
2279 	ratio = (msr >> 0) & 0x3F;
2280 	if (ratio)
2281 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n",
2282 			ratio, bclk, ratio * bclk);
2283 }
2284 
2285 static void
2286 dump_knl_turbo_ratio_limits(void)
2287 {
2288 	const unsigned int buckets_no = 7;
2289 
2290 	unsigned long long msr;
2291 	int delta_cores, delta_ratio;
2292 	int i, b_nr;
2293 	unsigned int cores[buckets_no];
2294 	unsigned int ratio[buckets_no];
2295 
2296 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2297 
2298 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n",
2299 		base_cpu, msr);
2300 
2301 	/**
2302 	 * Turbo encoding in KNL is as follows:
2303 	 * [0] -- Reserved
2304 	 * [7:1] -- Base value of number of active cores of bucket 1.
2305 	 * [15:8] -- Base value of freq ratio of bucket 1.
2306 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
2307 	 * i.e. active cores of bucket 2 =
2308 	 * active cores of bucket 1 + delta
2309 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
2310 	 * i.e. freq ratio of bucket 2 =
2311 	 * freq ratio of bucket 1 - delta
2312 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
2313 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
2314 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
2315 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
2316 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
2317 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
2318 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
2319 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
2320 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
2321 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
2322 	 */
2323 
2324 	b_nr = 0;
2325 	cores[b_nr] = (msr & 0xFF) >> 1;
2326 	ratio[b_nr] = (msr >> 8) & 0xFF;
2327 
2328 	for (i = 16; i < 64; i += 8) {
2329 		delta_cores = (msr >> i) & 0x1F;
2330 		delta_ratio = (msr >> (i + 5)) & 0x7;
2331 
2332 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
2333 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2334 		b_nr++;
2335 	}
2336 
2337 	for (i = buckets_no - 1; i >= 0; i--)
2338 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2339 			fprintf(outf,
2340 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
2341 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
2342 }
2343 
2344 static void
2345 dump_nhm_cst_cfg(void)
2346 {
2347 	unsigned long long msr;
2348 
2349 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2350 
2351 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2352 
2353 	fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)",
2354 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2355 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2356 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2357 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2358 		(msr & (1 << 15)) ? "" : "UN",
2359 		(unsigned int)msr & 0xF,
2360 		pkg_cstate_limit_strings[pkg_cstate_limit]);
2361 
2362 #define AUTOMATIC_CSTATE_CONVERSION		(1UL << 16)
2363 	if (has_automatic_cstate_conversion) {
2364 		fprintf(outf, ", automatic c-state conversion=%s",
2365 			(msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off");
2366 	}
2367 
2368 	fprintf(outf, ")\n");
2369 
2370 	return;
2371 }
2372 
2373 static void
2374 dump_config_tdp(void)
2375 {
2376 	unsigned long long msr;
2377 
2378 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2379 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2380 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2381 
2382 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2383 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2384 	if (msr) {
2385 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2386 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2387 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2388 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2389 	}
2390 	fprintf(outf, ")\n");
2391 
2392 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2393 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2394 	if (msr) {
2395 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2396 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2397 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2398 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2399 	}
2400 	fprintf(outf, ")\n");
2401 
2402 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2403 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2404 	if ((msr) & 0x3)
2405 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2406 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2407 	fprintf(outf, ")\n");
2408 
2409 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2410 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2411 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2412 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2413 	fprintf(outf, ")\n");
2414 }
2415 
2416 unsigned int irtl_time_units[] = {1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2417 
2418 void print_irtl(void)
2419 {
2420 	unsigned long long msr;
2421 
2422 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2423 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2424 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2425 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2426 
2427 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2428 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2429 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2430 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2431 
2432 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2433 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2434 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2435 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2436 
2437 	if (!do_irtl_hsw)
2438 		return;
2439 
2440 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2441 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2442 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2443 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2444 
2445 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2446 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2447 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2448 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2449 
2450 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2451 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2452 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2453 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2454 
2455 }
2456 void free_fd_percpu(void)
2457 {
2458 	int i;
2459 
2460 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2461 		if (fd_percpu[i] != 0)
2462 			close(fd_percpu[i]);
2463 	}
2464 
2465 	free(fd_percpu);
2466 }
2467 
2468 void free_all_buffers(void)
2469 {
2470 	int i;
2471 
2472 	CPU_FREE(cpu_present_set);
2473 	cpu_present_set = NULL;
2474 	cpu_present_setsize = 0;
2475 
2476 	CPU_FREE(cpu_affinity_set);
2477 	cpu_affinity_set = NULL;
2478 	cpu_affinity_setsize = 0;
2479 
2480 	free(thread_even);
2481 	free(core_even);
2482 	free(package_even);
2483 
2484 	thread_even = NULL;
2485 	core_even = NULL;
2486 	package_even = NULL;
2487 
2488 	free(thread_odd);
2489 	free(core_odd);
2490 	free(package_odd);
2491 
2492 	thread_odd = NULL;
2493 	core_odd = NULL;
2494 	package_odd = NULL;
2495 
2496 	free(output_buffer);
2497 	output_buffer = NULL;
2498 	outp = NULL;
2499 
2500 	free_fd_percpu();
2501 
2502 	free(irq_column_2_cpu);
2503 	free(irqs_per_cpu);
2504 
2505 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2506 		if (cpus[i].put_ids)
2507 			CPU_FREE(cpus[i].put_ids);
2508 	}
2509 	free(cpus);
2510 }
2511 
2512 
2513 /*
2514  * Parse a file containing a single int.
2515  * Return 0 if file can not be opened
2516  * Exit if file can be opened, but can not be parsed
2517  */
2518 int parse_int_file(const char *fmt, ...)
2519 {
2520 	va_list args;
2521 	char path[PATH_MAX];
2522 	FILE *filep;
2523 	int value;
2524 
2525 	va_start(args, fmt);
2526 	vsnprintf(path, sizeof(path), fmt, args);
2527 	va_end(args);
2528 	filep = fopen(path, "r");
2529 	if (!filep)
2530 		return 0;
2531 	if (fscanf(filep, "%d", &value) != 1)
2532 		err(1, "%s: failed to parse number from file", path);
2533 	fclose(filep);
2534 	return value;
2535 }
2536 
2537 /*
2538  * cpu_is_first_core_in_package(cpu)
2539  * return 1 if given CPU is 1st core in package
2540  */
2541 int cpu_is_first_core_in_package(int cpu)
2542 {
2543 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2544 }
2545 
2546 int get_physical_package_id(int cpu)
2547 {
2548 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2549 }
2550 
2551 int get_die_id(int cpu)
2552 {
2553 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu);
2554 }
2555 
2556 int get_core_id(int cpu)
2557 {
2558 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2559 }
2560 
2561 void set_node_data(void)
2562 {
2563 	int pkg, node, lnode, cpu, cpux;
2564 	int cpu_count;
2565 
2566 	/* initialize logical_node_id */
2567 	for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu)
2568 		cpus[cpu].logical_node_id = -1;
2569 
2570 	cpu_count = 0;
2571 	for (pkg = 0; pkg < topo.num_packages; pkg++) {
2572 		lnode = 0;
2573 		for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) {
2574 			if (cpus[cpu].physical_package_id != pkg)
2575 				continue;
2576 			/* find a cpu with an unset logical_node_id */
2577 			if (cpus[cpu].logical_node_id != -1)
2578 				continue;
2579 			cpus[cpu].logical_node_id = lnode;
2580 			node = cpus[cpu].physical_node_id;
2581 			cpu_count++;
2582 			/*
2583 			 * find all matching cpus on this pkg and set
2584 			 * the logical_node_id
2585 			 */
2586 			for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) {
2587 				if ((cpus[cpux].physical_package_id == pkg) &&
2588 				   (cpus[cpux].physical_node_id == node)) {
2589 					cpus[cpux].logical_node_id = lnode;
2590 					cpu_count++;
2591 				}
2592 			}
2593 			lnode++;
2594 			if (lnode > topo.nodes_per_pkg)
2595 				topo.nodes_per_pkg = lnode;
2596 		}
2597 		if (cpu_count >= topo.max_cpu_num)
2598 			break;
2599 	}
2600 }
2601 
2602 int get_physical_node_id(struct cpu_topology *thiscpu)
2603 {
2604 	char path[80];
2605 	FILE *filep;
2606 	int i;
2607 	int cpu = thiscpu->logical_cpu_id;
2608 
2609 	for (i = 0; i <= topo.max_cpu_num; i++) {
2610 		sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist",
2611 			cpu, i);
2612 		filep = fopen(path, "r");
2613 		if (!filep)
2614 			continue;
2615 		fclose(filep);
2616 		return i;
2617 	}
2618 	return -1;
2619 }
2620 
2621 int get_thread_siblings(struct cpu_topology *thiscpu)
2622 {
2623 	char path[80], character;
2624 	FILE *filep;
2625 	unsigned long map;
2626 	int so, shift, sib_core;
2627 	int cpu = thiscpu->logical_cpu_id;
2628 	int offset = topo.max_cpu_num + 1;
2629 	size_t size;
2630 	int thread_id = 0;
2631 
2632 	thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1));
2633 	if (thiscpu->thread_id < 0)
2634 		thiscpu->thread_id = thread_id++;
2635 	if (!thiscpu->put_ids)
2636 		return -1;
2637 
2638 	size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2639 	CPU_ZERO_S(size, thiscpu->put_ids);
2640 
2641 	sprintf(path,
2642 		"/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu);
2643 	filep = fopen_or_die(path, "r");
2644 	do {
2645 		offset -= BITMASK_SIZE;
2646 		if (fscanf(filep, "%lx%c", &map, &character) != 2)
2647 			err(1, "%s: failed to parse file", path);
2648 		for (shift = 0; shift < BITMASK_SIZE; shift++) {
2649 			if ((map >> shift) & 0x1) {
2650 				so = shift + offset;
2651 				sib_core = get_core_id(so);
2652 				if (sib_core == thiscpu->physical_core_id) {
2653 					CPU_SET_S(so, size, thiscpu->put_ids);
2654 					if ((so != cpu) &&
2655 					    (cpus[so].thread_id < 0))
2656 						cpus[so].thread_id =
2657 								    thread_id++;
2658 				}
2659 			}
2660 		}
2661 	} while (!strncmp(&character, ",", 1));
2662 	fclose(filep);
2663 
2664 	return CPU_COUNT_S(size, thiscpu->put_ids);
2665 }
2666 
2667 /*
2668  * run func(thread, core, package) in topology order
2669  * skip non-present cpus
2670  */
2671 
2672 int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
2673 	struct pkg_data *, struct thread_data *, struct core_data *,
2674 	struct pkg_data *), struct thread_data *thread_base,
2675 	struct core_data *core_base, struct pkg_data *pkg_base,
2676 	struct thread_data *thread_base2, struct core_data *core_base2,
2677 	struct pkg_data *pkg_base2)
2678 {
2679 	int retval, pkg_no, node_no, core_no, thread_no;
2680 
2681 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2682 		for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) {
2683 			for (core_no = 0; core_no < topo.cores_per_node;
2684 			     ++core_no) {
2685 				for (thread_no = 0; thread_no <
2686 					topo.threads_per_core; ++thread_no) {
2687 					struct thread_data *t, *t2;
2688 					struct core_data *c, *c2;
2689 					struct pkg_data *p, *p2;
2690 
2691 					t = GET_THREAD(thread_base, thread_no,
2692 						       core_no, node_no,
2693 						       pkg_no);
2694 
2695 					if (cpu_is_not_present(t->cpu_id))
2696 						continue;
2697 
2698 					t2 = GET_THREAD(thread_base2, thread_no,
2699 							core_no, node_no,
2700 							pkg_no);
2701 
2702 					c = GET_CORE(core_base, core_no,
2703 						     node_no, pkg_no);
2704 					c2 = GET_CORE(core_base2, core_no,
2705 						      node_no,
2706 						      pkg_no);
2707 
2708 					p = GET_PKG(pkg_base, pkg_no);
2709 					p2 = GET_PKG(pkg_base2, pkg_no);
2710 
2711 					retval = func(t, c, p, t2, c2, p2);
2712 					if (retval)
2713 						return retval;
2714 				}
2715 			}
2716 		}
2717 	}
2718 	return 0;
2719 }
2720 
2721 /*
2722  * run func(cpu) on every cpu in /proc/stat
2723  * return max_cpu number
2724  */
2725 int for_all_proc_cpus(int (func)(int))
2726 {
2727 	FILE *fp;
2728 	int cpu_num;
2729 	int retval;
2730 
2731 	fp = fopen_or_die(proc_stat, "r");
2732 
2733 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
2734 	if (retval != 0)
2735 		err(1, "%s: failed to parse format", proc_stat);
2736 
2737 	while (1) {
2738 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
2739 		if (retval != 1)
2740 			break;
2741 
2742 		retval = func(cpu_num);
2743 		if (retval) {
2744 			fclose(fp);
2745 			return(retval);
2746 		}
2747 	}
2748 	fclose(fp);
2749 	return 0;
2750 }
2751 
2752 void re_initialize(void)
2753 {
2754 	free_all_buffers();
2755 	setup_all_buffers();
2756 	printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
2757 }
2758 
2759 void set_max_cpu_num(void)
2760 {
2761 	FILE *filep;
2762 	unsigned long dummy;
2763 
2764 	topo.max_cpu_num = 0;
2765 	filep = fopen_or_die(
2766 			"/sys/devices/system/cpu/cpu0/topology/thread_siblings",
2767 			"r");
2768 	while (fscanf(filep, "%lx,", &dummy) == 1)
2769 		topo.max_cpu_num += BITMASK_SIZE;
2770 	fclose(filep);
2771 	topo.max_cpu_num--; /* 0 based */
2772 }
2773 
2774 /*
2775  * count_cpus()
2776  * remember the last one seen, it will be the max
2777  */
2778 int count_cpus(int cpu)
2779 {
2780 	topo.num_cpus++;
2781 	return 0;
2782 }
2783 int mark_cpu_present(int cpu)
2784 {
2785 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
2786 	return 0;
2787 }
2788 
2789 int init_thread_id(int cpu)
2790 {
2791 	cpus[cpu].thread_id = -1;
2792 	return 0;
2793 }
2794 
2795 /*
2796  * snapshot_proc_interrupts()
2797  *
2798  * read and record summary of /proc/interrupts
2799  *
2800  * return 1 if config change requires a restart, else return 0
2801  */
2802 int snapshot_proc_interrupts(void)
2803 {
2804 	static FILE *fp;
2805 	int column, retval;
2806 
2807 	if (fp == NULL)
2808 		fp = fopen_or_die("/proc/interrupts", "r");
2809 	else
2810 		rewind(fp);
2811 
2812 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
2813 	for (column = 0; column < topo.num_cpus; ++column) {
2814 		int cpu_number;
2815 
2816 		retval = fscanf(fp, " CPU%d", &cpu_number);
2817 		if (retval != 1)
2818 			break;
2819 
2820 		if (cpu_number > topo.max_cpu_num) {
2821 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
2822 			return 1;
2823 		}
2824 
2825 		irq_column_2_cpu[column] = cpu_number;
2826 		irqs_per_cpu[cpu_number] = 0;
2827 	}
2828 
2829 	/* read /proc/interrupt count lines and sum up irqs per cpu */
2830 	while (1) {
2831 		int column;
2832 		char buf[64];
2833 
2834 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
2835 		if (retval != 1)
2836 			break;
2837 
2838 		/* read the count per cpu */
2839 		for (column = 0; column < topo.num_cpus; ++column) {
2840 
2841 			int cpu_number, irq_count;
2842 
2843 			retval = fscanf(fp, " %d", &irq_count);
2844 			if (retval != 1)
2845 				break;
2846 
2847 			cpu_number = irq_column_2_cpu[column];
2848 			irqs_per_cpu[cpu_number] += irq_count;
2849 
2850 		}
2851 
2852 		while (getc(fp) != '\n')
2853 			;	/* flush interrupt description */
2854 
2855 	}
2856 	return 0;
2857 }
2858 /*
2859  * snapshot_gfx_rc6_ms()
2860  *
2861  * record snapshot of
2862  * /sys/class/drm/card0/power/rc6_residency_ms
2863  *
2864  * return 1 if config change requires a restart, else return 0
2865  */
2866 int snapshot_gfx_rc6_ms(void)
2867 {
2868 	FILE *fp;
2869 	int retval;
2870 
2871 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
2872 
2873 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
2874 	if (retval != 1)
2875 		err(1, "GFX rc6");
2876 
2877 	fclose(fp);
2878 
2879 	return 0;
2880 }
2881 /*
2882  * snapshot_gfx_mhz()
2883  *
2884  * record snapshot of
2885  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
2886  *
2887  * return 1 if config change requires a restart, else return 0
2888  */
2889 int snapshot_gfx_mhz(void)
2890 {
2891 	static FILE *fp;
2892 	int retval;
2893 
2894 	if (fp == NULL)
2895 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
2896 	else {
2897 		rewind(fp);
2898 		fflush(fp);
2899 	}
2900 
2901 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
2902 	if (retval != 1)
2903 		err(1, "GFX MHz");
2904 
2905 	return 0;
2906 }
2907 
2908 /*
2909  * snapshot_cpu_lpi()
2910  *
2911  * record snapshot of
2912  * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us
2913  *
2914  * return 1 if config change requires a restart, else return 0
2915  */
2916 int snapshot_cpu_lpi_us(void)
2917 {
2918 	FILE *fp;
2919 	int retval;
2920 
2921 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r");
2922 
2923 	retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us);
2924 	if (retval != 1) {
2925 		fprintf(stderr, "Disabling Low Power Idle CPU output\n");
2926 		BIC_NOT_PRESENT(BIC_CPU_LPI);
2927 		return -1;
2928 	}
2929 
2930 	fclose(fp);
2931 
2932 	return 0;
2933 }
2934 /*
2935  * snapshot_sys_lpi()
2936  *
2937  * record snapshot of
2938  * /sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us
2939  *
2940  * return 1 if config change requires a restart, else return 0
2941  */
2942 int snapshot_sys_lpi_us(void)
2943 {
2944 	FILE *fp;
2945 	int retval;
2946 
2947 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", "r");
2948 
2949 	retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us);
2950 	if (retval != 1) {
2951 		fprintf(stderr, "Disabling Low Power Idle System output\n");
2952 		BIC_NOT_PRESENT(BIC_SYS_LPI);
2953 		return -1;
2954 	}
2955 	fclose(fp);
2956 
2957 	return 0;
2958 }
2959 /*
2960  * snapshot /proc and /sys files
2961  *
2962  * return 1 if configuration restart needed, else return 0
2963  */
2964 int snapshot_proc_sysfs_files(void)
2965 {
2966 	if (DO_BIC(BIC_IRQ))
2967 		if (snapshot_proc_interrupts())
2968 			return 1;
2969 
2970 	if (DO_BIC(BIC_GFX_rc6))
2971 		snapshot_gfx_rc6_ms();
2972 
2973 	if (DO_BIC(BIC_GFXMHz))
2974 		snapshot_gfx_mhz();
2975 
2976 	if (DO_BIC(BIC_CPU_LPI))
2977 		snapshot_cpu_lpi_us();
2978 
2979 	if (DO_BIC(BIC_SYS_LPI))
2980 		snapshot_sys_lpi_us();
2981 
2982 	return 0;
2983 }
2984 
2985 int exit_requested;
2986 
2987 static void signal_handler (int signal)
2988 {
2989 	switch (signal) {
2990 	case SIGINT:
2991 		exit_requested = 1;
2992 		if (debug)
2993 			fprintf(stderr, " SIGINT\n");
2994 		break;
2995 	case SIGUSR1:
2996 		if (debug > 1)
2997 			fprintf(stderr, "SIGUSR1\n");
2998 		break;
2999 	}
3000 	/* make sure this manually-invoked interval is at least 1ms long */
3001 	nanosleep(&one_msec, NULL);
3002 }
3003 
3004 void setup_signal_handler(void)
3005 {
3006 	struct sigaction sa;
3007 
3008 	memset(&sa, 0, sizeof(sa));
3009 
3010 	sa.sa_handler = &signal_handler;
3011 
3012 	if (sigaction(SIGINT, &sa, NULL) < 0)
3013 		err(1, "sigaction SIGINT");
3014 	if (sigaction(SIGUSR1, &sa, NULL) < 0)
3015 		err(1, "sigaction SIGUSR1");
3016 }
3017 
3018 void do_sleep(void)
3019 {
3020 	struct timeval select_timeout;
3021 	fd_set readfds;
3022 	int retval;
3023 
3024 	FD_ZERO(&readfds);
3025 	FD_SET(0, &readfds);
3026 
3027 	if (!isatty(fileno(stdin))) {
3028 		nanosleep(&interval_ts, NULL);
3029 		return;
3030 	}
3031 
3032 	select_timeout = interval_tv;
3033 	retval = select(1, &readfds, NULL, NULL, &select_timeout);
3034 
3035 	if (retval == 1) {
3036 		switch (getc(stdin)) {
3037 		case 'q':
3038 			exit_requested = 1;
3039 			break;
3040 		}
3041 		/* make sure this manually-invoked interval is at least 1ms long */
3042 		nanosleep(&one_msec, NULL);
3043 	}
3044 }
3045 
3046 
3047 void turbostat_loop()
3048 {
3049 	int retval;
3050 	int restarted = 0;
3051 	int done_iters = 0;
3052 
3053 	setup_signal_handler();
3054 
3055 restart:
3056 	restarted++;
3057 
3058 	snapshot_proc_sysfs_files();
3059 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3060 	first_counter_read = 0;
3061 	if (retval < -1) {
3062 		exit(retval);
3063 	} else if (retval == -1) {
3064 		if (restarted > 1) {
3065 			exit(retval);
3066 		}
3067 		re_initialize();
3068 		goto restart;
3069 	}
3070 	restarted = 0;
3071 	done_iters = 0;
3072 	gettimeofday(&tv_even, (struct timezone *)NULL);
3073 
3074 	while (1) {
3075 		if (for_all_proc_cpus(cpu_is_not_present)) {
3076 			re_initialize();
3077 			goto restart;
3078 		}
3079 		do_sleep();
3080 		if (snapshot_proc_sysfs_files())
3081 			goto restart;
3082 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
3083 		if (retval < -1) {
3084 			exit(retval);
3085 		} else if (retval == -1) {
3086 			re_initialize();
3087 			goto restart;
3088 		}
3089 		gettimeofday(&tv_odd, (struct timezone *)NULL);
3090 		timersub(&tv_odd, &tv_even, &tv_delta);
3091 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
3092 			re_initialize();
3093 			goto restart;
3094 		}
3095 		compute_average(EVEN_COUNTERS);
3096 		format_all_counters(EVEN_COUNTERS);
3097 		flush_output_stdout();
3098 		if (exit_requested)
3099 			break;
3100 		if (num_iterations && ++done_iters >= num_iterations)
3101 			break;
3102 		do_sleep();
3103 		if (snapshot_proc_sysfs_files())
3104 			goto restart;
3105 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3106 		if (retval < -1) {
3107 			exit(retval);
3108 		} else if (retval == -1) {
3109 			re_initialize();
3110 			goto restart;
3111 		}
3112 		gettimeofday(&tv_even, (struct timezone *)NULL);
3113 		timersub(&tv_even, &tv_odd, &tv_delta);
3114 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
3115 			re_initialize();
3116 			goto restart;
3117 		}
3118 		compute_average(ODD_COUNTERS);
3119 		format_all_counters(ODD_COUNTERS);
3120 		flush_output_stdout();
3121 		if (exit_requested)
3122 			break;
3123 		if (num_iterations && ++done_iters >= num_iterations)
3124 			break;
3125 	}
3126 }
3127 
3128 void check_dev_msr()
3129 {
3130 	struct stat sb;
3131 	char pathname[32];
3132 
3133 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3134 	if (stat(pathname, &sb))
3135  		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
3136 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
3137 }
3138 
3139 void check_permissions()
3140 {
3141 	struct __user_cap_header_struct cap_header_data;
3142 	cap_user_header_t cap_header = &cap_header_data;
3143 	struct __user_cap_data_struct cap_data_data;
3144 	cap_user_data_t cap_data = &cap_data_data;
3145 	extern int capget(cap_user_header_t hdrp, cap_user_data_t datap);
3146 	int do_exit = 0;
3147 	char pathname[32];
3148 
3149 	/* check for CAP_SYS_RAWIO */
3150 	cap_header->pid = getpid();
3151 	cap_header->version = _LINUX_CAPABILITY_VERSION;
3152 	if (capget(cap_header, cap_data) < 0)
3153 		err(-6, "capget(2) failed");
3154 
3155 	if ((cap_data->effective & (1 << CAP_SYS_RAWIO)) == 0) {
3156 		do_exit++;
3157 		warnx("capget(CAP_SYS_RAWIO) failed,"
3158 			" try \"# setcap cap_sys_rawio=ep %s\"", progname);
3159 	}
3160 
3161 	/* test file permissions */
3162 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3163 	if (euidaccess(pathname, R_OK)) {
3164 		do_exit++;
3165 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
3166 	}
3167 
3168 	/* if all else fails, thell them to be root */
3169 	if (do_exit)
3170 		if (getuid() != 0)
3171 			warnx("... or simply run as root");
3172 
3173 	if (do_exit)
3174 		exit(-6);
3175 }
3176 
3177 /*
3178  * NHM adds support for additional MSRs:
3179  *
3180  * MSR_SMI_COUNT                   0x00000034
3181  *
3182  * MSR_PLATFORM_INFO               0x000000ce
3183  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
3184  *
3185  * MSR_MISC_PWR_MGMT               0x000001aa
3186  *
3187  * MSR_PKG_C3_RESIDENCY            0x000003f8
3188  * MSR_PKG_C6_RESIDENCY            0x000003f9
3189  * MSR_CORE_C3_RESIDENCY           0x000003fc
3190  * MSR_CORE_C6_RESIDENCY           0x000003fd
3191  *
3192  * Side effect:
3193  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
3194  * sets has_misc_feature_control
3195  */
3196 int probe_nhm_msrs(unsigned int family, unsigned int model)
3197 {
3198 	unsigned long long msr;
3199 	unsigned int base_ratio;
3200 	int *pkg_cstate_limits;
3201 
3202 	if (!genuine_intel)
3203 		return 0;
3204 
3205 	if (family != 6)
3206 		return 0;
3207 
3208 	bclk = discover_bclk(family, model);
3209 
3210 	switch (model) {
3211 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
3212 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3213 		pkg_cstate_limits = nhm_pkg_cstate_limits;
3214 		break;
3215 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
3216 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
3217 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3218 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3219 		pkg_cstate_limits = snb_pkg_cstate_limits;
3220 		has_misc_feature_control = 1;
3221 		break;
3222 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3223 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3224 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3225 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3226 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3227 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3228 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3229 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3230 		pkg_cstate_limits = hsw_pkg_cstate_limits;
3231 		has_misc_feature_control = 1;
3232 		break;
3233 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3234 		pkg_cstate_limits = skx_pkg_cstate_limits;
3235 		has_misc_feature_control = 1;
3236 		break;
3237 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3238 		no_MSR_MISC_PWR_MGMT = 1;
3239 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3240 		pkg_cstate_limits = slv_pkg_cstate_limits;
3241 		break;
3242 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
3243 		pkg_cstate_limits = amt_pkg_cstate_limits;
3244 		no_MSR_MISC_PWR_MGMT = 1;
3245 		break;
3246 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
3247 		pkg_cstate_limits = phi_pkg_cstate_limits;
3248 		break;
3249 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3250 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3251 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3252 		pkg_cstate_limits = glm_pkg_cstate_limits;
3253 		break;
3254 	default:
3255 		return 0;
3256 	}
3257 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
3258 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
3259 
3260 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
3261 	base_ratio = (msr >> 8) & 0xFF;
3262 
3263 	base_hz = base_ratio * bclk * 1000000;
3264 	has_base_hz = 1;
3265 	return 1;
3266 }
3267 /*
3268  * SLV client has support for unique MSRs:
3269  *
3270  * MSR_CC6_DEMOTION_POLICY_CONFIG
3271  * MSR_MC6_DEMOTION_POLICY_CONFIG
3272  */
3273 
3274 int has_slv_msrs(unsigned int family, unsigned int model)
3275 {
3276 	if (!genuine_intel)
3277 		return 0;
3278 
3279 	switch (model) {
3280 	case INTEL_FAM6_ATOM_SILVERMONT:
3281 	case INTEL_FAM6_ATOM_SILVERMONT_MID:
3282 	case INTEL_FAM6_ATOM_AIRMONT_MID:
3283 		return 1;
3284 	}
3285 	return 0;
3286 }
3287 int is_dnv(unsigned int family, unsigned int model)
3288 {
3289 
3290 	if (!genuine_intel)
3291 		return 0;
3292 
3293 	switch (model) {
3294 	case INTEL_FAM6_ATOM_GOLDMONT_X:
3295 		return 1;
3296 	}
3297 	return 0;
3298 }
3299 int is_bdx(unsigned int family, unsigned int model)
3300 {
3301 
3302 	if (!genuine_intel)
3303 		return 0;
3304 
3305 	switch (model) {
3306 	case INTEL_FAM6_BROADWELL_X:
3307 		return 1;
3308 	}
3309 	return 0;
3310 }
3311 int is_skx(unsigned int family, unsigned int model)
3312 {
3313 
3314 	if (!genuine_intel)
3315 		return 0;
3316 
3317 	switch (model) {
3318 	case INTEL_FAM6_SKYLAKE_X:
3319 		return 1;
3320 	}
3321 	return 0;
3322 }
3323 
3324 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
3325 {
3326 	if (has_slv_msrs(family, model))
3327 		return 0;
3328 
3329 	switch (model) {
3330 	/* Nehalem compatible, but do not include turbo-ratio limit support */
3331 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3332 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
3333 		return 0;
3334 	default:
3335 		return 1;
3336 	}
3337 }
3338 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
3339 {
3340 	if (has_slv_msrs(family, model))
3341 		return 1;
3342 
3343 	return 0;
3344 }
3345 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
3346 {
3347 	if (!genuine_intel)
3348 		return 0;
3349 
3350 	if (family != 6)
3351 		return 0;
3352 
3353 	switch (model) {
3354 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3355 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3356 		return 1;
3357 	default:
3358 		return 0;
3359 	}
3360 }
3361 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
3362 {
3363 	if (!genuine_intel)
3364 		return 0;
3365 
3366 	if (family != 6)
3367 		return 0;
3368 
3369 	switch (model) {
3370 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3371 		return 1;
3372 	default:
3373 		return 0;
3374 	}
3375 }
3376 
3377 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
3378 {
3379 	if (!genuine_intel)
3380 		return 0;
3381 
3382 	if (family != 6)
3383 		return 0;
3384 
3385 	switch (model) {
3386 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3387 		return 1;
3388 	default:
3389 		return 0;
3390 	}
3391 }
3392 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
3393 {
3394 	if (!genuine_intel)
3395 		return 0;
3396 
3397 	if (family != 6)
3398 		return 0;
3399 
3400 	switch (model) {
3401 	case INTEL_FAM6_ATOM_GOLDMONT:
3402 	case INTEL_FAM6_SKYLAKE_X:
3403 		return 1;
3404 	default:
3405 		return 0;
3406 	}
3407 }
3408 int has_config_tdp(unsigned int family, unsigned int model)
3409 {
3410 	if (!genuine_intel)
3411 		return 0;
3412 
3413 	if (family != 6)
3414 		return 0;
3415 
3416 	switch (model) {
3417 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3418 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3419 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3420 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3421 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3422 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3423 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3424 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3425 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3426 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3427 
3428 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3429 		return 1;
3430 	default:
3431 		return 0;
3432 	}
3433 }
3434 
3435 static void
3436 dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
3437 {
3438 	if (!do_nhm_platform_info)
3439 		return;
3440 
3441 	dump_nhm_platform_info();
3442 
3443 	if (has_hsw_turbo_ratio_limit(family, model))
3444 		dump_hsw_turbo_ratio_limits();
3445 
3446 	if (has_ivt_turbo_ratio_limit(family, model))
3447 		dump_ivt_turbo_ratio_limits();
3448 
3449 	if (has_turbo_ratio_limit(family, model))
3450 		dump_turbo_ratio_limits(family, model);
3451 
3452 	if (has_atom_turbo_ratio_limit(family, model))
3453 		dump_atom_turbo_ratio_limits();
3454 
3455 	if (has_knl_turbo_ratio_limit(family, model))
3456 		dump_knl_turbo_ratio_limits();
3457 
3458 	if (has_config_tdp(family, model))
3459 		dump_config_tdp();
3460 
3461 	dump_nhm_cst_cfg();
3462 }
3463 
3464 static void
3465 dump_sysfs_cstate_config(void)
3466 {
3467 	char path[64];
3468 	char name_buf[16];
3469 	char desc[64];
3470 	FILE *input;
3471 	int state;
3472 	char *sp;
3473 
3474 	if (!DO_BIC(BIC_sysfs))
3475 		return;
3476 
3477 	for (state = 0; state < 10; ++state) {
3478 
3479 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
3480 			base_cpu, state);
3481 		input = fopen(path, "r");
3482 		if (input == NULL)
3483 			continue;
3484 		if (!fgets(name_buf, sizeof(name_buf), input))
3485 			err(1, "%s: failed to read file", path);
3486 
3487 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
3488 		sp = strchr(name_buf, '-');
3489 		if (!sp)
3490 			sp = strchrnul(name_buf, '\n');
3491 		*sp = '\0';
3492 		fclose(input);
3493 
3494 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc",
3495 			base_cpu, state);
3496 		input = fopen(path, "r");
3497 		if (input == NULL)
3498 			continue;
3499 		if (!fgets(desc, sizeof(desc), input))
3500 			err(1, "%s: failed to read file", path);
3501 
3502 		fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
3503 		fclose(input);
3504 	}
3505 }
3506 static void
3507 dump_sysfs_pstate_config(void)
3508 {
3509 	char path[64];
3510 	char driver_buf[64];
3511 	char governor_buf[64];
3512 	FILE *input;
3513 	int turbo;
3514 
3515 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver",
3516 			base_cpu);
3517 	input = fopen(path, "r");
3518 	if (input == NULL) {
3519 		fprintf(outf, "NSFOD %s\n", path);
3520 		return;
3521 	}
3522 	if (!fgets(driver_buf, sizeof(driver_buf), input))
3523 		err(1, "%s: failed to read file", path);
3524 	fclose(input);
3525 
3526 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor",
3527 			base_cpu);
3528 	input = fopen(path, "r");
3529 	if (input == NULL) {
3530 		fprintf(outf, "NSFOD %s\n", path);
3531 		return;
3532 	}
3533 	if (!fgets(governor_buf, sizeof(governor_buf), input))
3534 		err(1, "%s: failed to read file", path);
3535 	fclose(input);
3536 
3537 	fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
3538 	fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
3539 
3540 	sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
3541 	input = fopen(path, "r");
3542 	if (input != NULL) {
3543 		if (fscanf(input, "%d", &turbo) != 1)
3544 			err(1, "%s: failed to parse number from file", path);
3545 		fprintf(outf, "cpufreq boost: %d\n", turbo);
3546 		fclose(input);
3547 	}
3548 
3549 	sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
3550 	input = fopen(path, "r");
3551 	if (input != NULL) {
3552 		if (fscanf(input, "%d", &turbo) != 1)
3553 			err(1, "%s: failed to parse number from file", path);
3554 		fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
3555 		fclose(input);
3556 	}
3557 }
3558 
3559 
3560 /*
3561  * print_epb()
3562  * Decode the ENERGY_PERF_BIAS MSR
3563  */
3564 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3565 {
3566 	unsigned long long msr;
3567 	char *epb_string;
3568 	int cpu;
3569 
3570 	if (!has_epb)
3571 		return 0;
3572 
3573 	cpu = t->cpu_id;
3574 
3575 	/* EPB is per-package */
3576 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3577 		return 0;
3578 
3579 	if (cpu_migrate(cpu)) {
3580 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3581 		return -1;
3582 	}
3583 
3584 	if (get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr))
3585 		return 0;
3586 
3587 	switch (msr & 0xF) {
3588 	case ENERGY_PERF_BIAS_PERFORMANCE:
3589 		epb_string = "performance";
3590 		break;
3591 	case ENERGY_PERF_BIAS_NORMAL:
3592 		epb_string = "balanced";
3593 		break;
3594 	case ENERGY_PERF_BIAS_POWERSAVE:
3595 		epb_string = "powersave";
3596 		break;
3597 	default:
3598 		epb_string = "custom";
3599 		break;
3600 	}
3601 	fprintf(outf, "cpu%d: MSR_IA32_ENERGY_PERF_BIAS: 0x%08llx (%s)\n", cpu, msr, epb_string);
3602 
3603 	return 0;
3604 }
3605 /*
3606  * print_hwp()
3607  * Decode the MSR_HWP_CAPABILITIES
3608  */
3609 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3610 {
3611 	unsigned long long msr;
3612 	int cpu;
3613 
3614 	if (!has_hwp)
3615 		return 0;
3616 
3617 	cpu = t->cpu_id;
3618 
3619 	/* MSR_HWP_CAPABILITIES is per-package */
3620 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3621 		return 0;
3622 
3623 	if (cpu_migrate(cpu)) {
3624 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3625 		return -1;
3626 	}
3627 
3628 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
3629 		return 0;
3630 
3631 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n",
3632 		cpu, msr, (msr & (1 << 0)) ? "" : "No-");
3633 
3634 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
3635 	if ((msr & (1 << 0)) == 0)
3636 		return 0;
3637 
3638 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
3639 		return 0;
3640 
3641 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
3642 			"(high %d guar %d eff %d low %d)\n",
3643 			cpu, msr,
3644 			(unsigned int)HWP_HIGHEST_PERF(msr),
3645 			(unsigned int)HWP_GUARANTEED_PERF(msr),
3646 			(unsigned int)HWP_MOSTEFFICIENT_PERF(msr),
3647 			(unsigned int)HWP_LOWEST_PERF(msr));
3648 
3649 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
3650 		return 0;
3651 
3652 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
3653 			"(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
3654 			cpu, msr,
3655 			(unsigned int)(((msr) >> 0) & 0xff),
3656 			(unsigned int)(((msr) >> 8) & 0xff),
3657 			(unsigned int)(((msr) >> 16) & 0xff),
3658 			(unsigned int)(((msr) >> 24) & 0xff),
3659 			(unsigned int)(((msr) >> 32) & 0xff3),
3660 			(unsigned int)(((msr) >> 42) & 0x1));
3661 
3662 	if (has_hwp_pkg) {
3663 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
3664 			return 0;
3665 
3666 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
3667 			"(min %d max %d des %d epp 0x%x window 0x%x)\n",
3668 			cpu, msr,
3669 			(unsigned int)(((msr) >> 0) & 0xff),
3670 			(unsigned int)(((msr) >> 8) & 0xff),
3671 			(unsigned int)(((msr) >> 16) & 0xff),
3672 			(unsigned int)(((msr) >> 24) & 0xff),
3673 			(unsigned int)(((msr) >> 32) & 0xff3));
3674 	}
3675 	if (has_hwp_notify) {
3676 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
3677 			return 0;
3678 
3679 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
3680 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
3681 			cpu, msr,
3682 			((msr) & 0x1) ? "EN" : "Dis",
3683 			((msr) & 0x2) ? "EN" : "Dis");
3684 	}
3685 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
3686 		return 0;
3687 
3688 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
3689 			"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
3690 			cpu, msr,
3691 			((msr) & 0x1) ? "" : "No-",
3692 			((msr) & 0x2) ? "" : "No-");
3693 
3694 	return 0;
3695 }
3696 
3697 /*
3698  * print_perf_limit()
3699  */
3700 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3701 {
3702 	unsigned long long msr;
3703 	int cpu;
3704 
3705 	cpu = t->cpu_id;
3706 
3707 	/* per-package */
3708 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
3709 		return 0;
3710 
3711 	if (cpu_migrate(cpu)) {
3712 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
3713 		return -1;
3714 	}
3715 
3716 	if (do_core_perf_limit_reasons) {
3717 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
3718 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3719 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
3720 			(msr & 1 << 15) ? "bit15, " : "",
3721 			(msr & 1 << 14) ? "bit14, " : "",
3722 			(msr & 1 << 13) ? "Transitions, " : "",
3723 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
3724 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
3725 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3726 			(msr & 1 << 9) ? "CorePwr, " : "",
3727 			(msr & 1 << 8) ? "Amps, " : "",
3728 			(msr & 1 << 6) ? "VR-Therm, " : "",
3729 			(msr & 1 << 5) ? "Auto-HWP, " : "",
3730 			(msr & 1 << 4) ? "Graphics, " : "",
3731 			(msr & 1 << 2) ? "bit2, " : "",
3732 			(msr & 1 << 1) ? "ThermStatus, " : "",
3733 			(msr & 1 << 0) ? "PROCHOT, " : "");
3734 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
3735 			(msr & 1 << 31) ? "bit31, " : "",
3736 			(msr & 1 << 30) ? "bit30, " : "",
3737 			(msr & 1 << 29) ? "Transitions, " : "",
3738 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
3739 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
3740 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3741 			(msr & 1 << 25) ? "CorePwr, " : "",
3742 			(msr & 1 << 24) ? "Amps, " : "",
3743 			(msr & 1 << 22) ? "VR-Therm, " : "",
3744 			(msr & 1 << 21) ? "Auto-HWP, " : "",
3745 			(msr & 1 << 20) ? "Graphics, " : "",
3746 			(msr & 1 << 18) ? "bit18, " : "",
3747 			(msr & 1 << 17) ? "ThermStatus, " : "",
3748 			(msr & 1 << 16) ? "PROCHOT, " : "");
3749 
3750 	}
3751 	if (do_gfx_perf_limit_reasons) {
3752 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
3753 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3754 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
3755 			(msr & 1 << 0) ? "PROCHOT, " : "",
3756 			(msr & 1 << 1) ? "ThermStatus, " : "",
3757 			(msr & 1 << 4) ? "Graphics, " : "",
3758 			(msr & 1 << 6) ? "VR-Therm, " : "",
3759 			(msr & 1 << 8) ? "Amps, " : "",
3760 			(msr & 1 << 9) ? "GFXPwr, " : "",
3761 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3762 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3763 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
3764 			(msr & 1 << 16) ? "PROCHOT, " : "",
3765 			(msr & 1 << 17) ? "ThermStatus, " : "",
3766 			(msr & 1 << 20) ? "Graphics, " : "",
3767 			(msr & 1 << 22) ? "VR-Therm, " : "",
3768 			(msr & 1 << 24) ? "Amps, " : "",
3769 			(msr & 1 << 25) ? "GFXPwr, " : "",
3770 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3771 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3772 	}
3773 	if (do_ring_perf_limit_reasons) {
3774 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
3775 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
3776 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
3777 			(msr & 1 << 0) ? "PROCHOT, " : "",
3778 			(msr & 1 << 1) ? "ThermStatus, " : "",
3779 			(msr & 1 << 6) ? "VR-Therm, " : "",
3780 			(msr & 1 << 8) ? "Amps, " : "",
3781 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
3782 			(msr & 1 << 11) ? "PkgPwrL2, " : "");
3783 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
3784 			(msr & 1 << 16) ? "PROCHOT, " : "",
3785 			(msr & 1 << 17) ? "ThermStatus, " : "",
3786 			(msr & 1 << 22) ? "VR-Therm, " : "",
3787 			(msr & 1 << 24) ? "Amps, " : "",
3788 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
3789 			(msr & 1 << 27) ? "PkgPwrL2, " : "");
3790 	}
3791 	return 0;
3792 }
3793 
3794 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
3795 #define	RAPL_TIME_GRANULARITY	0x3F /* 6 bit time granularity */
3796 
3797 double get_tdp_intel(unsigned int model)
3798 {
3799 	unsigned long long msr;
3800 
3801 	if (do_rapl & RAPL_PKG_POWER_INFO)
3802 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
3803 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
3804 
3805 	switch (model) {
3806 	case INTEL_FAM6_ATOM_SILVERMONT:
3807 	case INTEL_FAM6_ATOM_SILVERMONT_X:
3808 		return 30.0;
3809 	default:
3810 		return 135.0;
3811 	}
3812 }
3813 
3814 double get_tdp_amd(unsigned int family)
3815 {
3816 	switch (family) {
3817 	case 0x17:
3818 	default:
3819 		/* This is the max stock TDP of HEDT/Server Fam17h chips */
3820 		return 250.0;
3821 	}
3822 }
3823 
3824 /*
3825  * rapl_dram_energy_units_probe()
3826  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
3827  */
3828 static double
3829 rapl_dram_energy_units_probe(int  model, double rapl_energy_units)
3830 {
3831 	/* only called for genuine_intel, family 6 */
3832 
3833 	switch (model) {
3834 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3835 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3836 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3837 		return (rapl_dram_energy_units = 15.3 / 1000000);
3838 	default:
3839 		return (rapl_energy_units);
3840 	}
3841 }
3842 
3843 void rapl_probe_intel(unsigned int family, unsigned int model)
3844 {
3845 	unsigned long long msr;
3846 	unsigned int time_unit;
3847 	double tdp;
3848 
3849 	if (family != 6)
3850 		return;
3851 
3852 	switch (model) {
3853 	case INTEL_FAM6_SANDYBRIDGE:
3854 	case INTEL_FAM6_IVYBRIDGE:
3855 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
3856 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
3857 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
3858 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
3859 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
3860 		if (rapl_joules) {
3861 			BIC_PRESENT(BIC_Pkg_J);
3862 			BIC_PRESENT(BIC_Cor_J);
3863 			BIC_PRESENT(BIC_GFX_J);
3864 		} else {
3865 			BIC_PRESENT(BIC_PkgWatt);
3866 			BIC_PRESENT(BIC_CorWatt);
3867 			BIC_PRESENT(BIC_GFXWatt);
3868 		}
3869 		break;
3870 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3871 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3872 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
3873 		if (rapl_joules)
3874 			BIC_PRESENT(BIC_Pkg_J);
3875 		else
3876 			BIC_PRESENT(BIC_PkgWatt);
3877 		break;
3878 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
3879 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
3880 		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;
3881 		BIC_PRESENT(BIC_PKG__);
3882 		BIC_PRESENT(BIC_RAM__);
3883 		if (rapl_joules) {
3884 			BIC_PRESENT(BIC_Pkg_J);
3885 			BIC_PRESENT(BIC_Cor_J);
3886 			BIC_PRESENT(BIC_RAM_J);
3887 			BIC_PRESENT(BIC_GFX_J);
3888 		} else {
3889 			BIC_PRESENT(BIC_PkgWatt);
3890 			BIC_PRESENT(BIC_CorWatt);
3891 			BIC_PRESENT(BIC_RAMWatt);
3892 			BIC_PRESENT(BIC_GFXWatt);
3893 		}
3894 		break;
3895 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3896 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3897 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3898 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
3899 		do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
3900 		BIC_PRESENT(BIC_PKG__);
3901 		BIC_PRESENT(BIC_RAM__);
3902 		if (rapl_joules) {
3903 			BIC_PRESENT(BIC_Pkg_J);
3904 			BIC_PRESENT(BIC_RAM_J);
3905 		} else {
3906 			BIC_PRESENT(BIC_PkgWatt);
3907 			BIC_PRESENT(BIC_RAMWatt);
3908 		}
3909 		break;
3910 	case INTEL_FAM6_SANDYBRIDGE_X:
3911 	case INTEL_FAM6_IVYBRIDGE_X:
3912 		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;
3913 		BIC_PRESENT(BIC_PKG__);
3914 		BIC_PRESENT(BIC_RAM__);
3915 		if (rapl_joules) {
3916 			BIC_PRESENT(BIC_Pkg_J);
3917 			BIC_PRESENT(BIC_Cor_J);
3918 			BIC_PRESENT(BIC_RAM_J);
3919 		} else {
3920 			BIC_PRESENT(BIC_PkgWatt);
3921 			BIC_PRESENT(BIC_CorWatt);
3922 			BIC_PRESENT(BIC_RAMWatt);
3923 		}
3924 		break;
3925 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3926 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
3927 		do_rapl = RAPL_PKG | RAPL_CORES;
3928 		if (rapl_joules) {
3929 			BIC_PRESENT(BIC_Pkg_J);
3930 			BIC_PRESENT(BIC_Cor_J);
3931 		} else {
3932 			BIC_PRESENT(BIC_PkgWatt);
3933 			BIC_PRESENT(BIC_CorWatt);
3934 		}
3935 		break;
3936 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
3937 		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;
3938 		BIC_PRESENT(BIC_PKG__);
3939 		BIC_PRESENT(BIC_RAM__);
3940 		if (rapl_joules) {
3941 			BIC_PRESENT(BIC_Pkg_J);
3942 			BIC_PRESENT(BIC_Cor_J);
3943 			BIC_PRESENT(BIC_RAM_J);
3944 		} else {
3945 			BIC_PRESENT(BIC_PkgWatt);
3946 			BIC_PRESENT(BIC_CorWatt);
3947 			BIC_PRESENT(BIC_RAMWatt);
3948 		}
3949 		break;
3950 	default:
3951 		return;
3952 	}
3953 
3954 	/* units on package 0, verify later other packages match */
3955 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
3956 		return;
3957 
3958 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
3959 	if (model == INTEL_FAM6_ATOM_SILVERMONT)
3960 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
3961 	else
3962 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
3963 
3964 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
3965 
3966 	time_unit = msr >> 16 & 0xF;
3967 	if (time_unit == 0)
3968 		time_unit = 0xA;
3969 
3970 	rapl_time_units = 1.0 / (1 << (time_unit));
3971 
3972 	tdp = get_tdp_intel(model);
3973 
3974 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
3975 	if (!quiet)
3976 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
3977 }
3978 
3979 void rapl_probe_amd(unsigned int family, unsigned int model)
3980 {
3981 	unsigned long long msr;
3982 	unsigned int eax, ebx, ecx, edx;
3983 	unsigned int has_rapl = 0;
3984 	double tdp;
3985 
3986 	if (max_extended_level >= 0x80000007) {
3987 		__cpuid(0x80000007, eax, ebx, ecx, edx);
3988 		/* RAPL (Fam 17h) */
3989 		has_rapl = edx & (1 << 14);
3990 	}
3991 
3992 	if (!has_rapl)
3993 		return;
3994 
3995 	switch (family) {
3996 	case 0x17: /* Zen, Zen+ */
3997 		do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY;
3998 		if (rapl_joules) {
3999 			BIC_PRESENT(BIC_Pkg_J);
4000 			BIC_PRESENT(BIC_Cor_J);
4001 		} else {
4002 			BIC_PRESENT(BIC_PkgWatt);
4003 			BIC_PRESENT(BIC_CorWatt);
4004 		}
4005 		break;
4006 	default:
4007 		return;
4008 	}
4009 
4010 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
4011 		return;
4012 
4013 	rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf));
4014 	rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f));
4015 	rapl_power_units = ldexp(1.0, -(msr & 0xf));
4016 
4017 	tdp = get_tdp_amd(model);
4018 
4019 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4020 	if (!quiet)
4021 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4022 }
4023 
4024 /*
4025  * rapl_probe()
4026  *
4027  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
4028  */
4029 void rapl_probe(unsigned int family, unsigned int model)
4030 {
4031 	if (genuine_intel)
4032 		rapl_probe_intel(family, model);
4033 	if (authentic_amd)
4034 		rapl_probe_amd(family, model);
4035 }
4036 
4037 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
4038 {
4039 	if (!genuine_intel)
4040 		return;
4041 
4042 	if (family != 6)
4043 		return;
4044 
4045 	switch (model) {
4046 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4047 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4048 		do_gfx_perf_limit_reasons = 1;
4049 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4050 		do_core_perf_limit_reasons = 1;
4051 		do_ring_perf_limit_reasons = 1;
4052 	default:
4053 		return;
4054 	}
4055 }
4056 
4057 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model)
4058 {
4059 	if (is_skx(family, model) || is_bdx(family, model))
4060 		has_automatic_cstate_conversion = 1;
4061 }
4062 
4063 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4064 {
4065 	unsigned long long msr;
4066 	unsigned int dts, dts2;
4067 	int cpu;
4068 
4069 	if (!(do_dts || do_ptm))
4070 		return 0;
4071 
4072 	cpu = t->cpu_id;
4073 
4074 	/* DTS is per-core, no need to print for each thread */
4075 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
4076 		return 0;
4077 
4078 	if (cpu_migrate(cpu)) {
4079 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4080 		return -1;
4081 	}
4082 
4083 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
4084 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
4085 			return 0;
4086 
4087 		dts = (msr >> 16) & 0x7F;
4088 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n",
4089 			cpu, msr, tcc_activation_temp - dts);
4090 
4091 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
4092 			return 0;
4093 
4094 		dts = (msr >> 16) & 0x7F;
4095 		dts2 = (msr >> 8) & 0x7F;
4096 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4097 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4098 	}
4099 
4100 
4101 	if (do_dts && debug) {
4102 		unsigned int resolution;
4103 
4104 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
4105 			return 0;
4106 
4107 		dts = (msr >> 16) & 0x7F;
4108 		resolution = (msr >> 27) & 0xF;
4109 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
4110 			cpu, msr, tcc_activation_temp - dts, resolution);
4111 
4112 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
4113 			return 0;
4114 
4115 		dts = (msr >> 16) & 0x7F;
4116 		dts2 = (msr >> 8) & 0x7F;
4117 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4118 			cpu, msr, tcc_activation_temp - dts, tcc_activation_temp - dts2);
4119 	}
4120 
4121 	return 0;
4122 }
4123 
4124 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
4125 {
4126 	fprintf(outf, "cpu%d: %s: %sabled (%f Watts, %f sec, clamp %sabled)\n",
4127 		cpu, label,
4128 		((msr >> 15) & 1) ? "EN" : "DIS",
4129 		((msr >> 0) & 0x7FFF) * rapl_power_units,
4130 		(1.0 + (((msr >> 22) & 0x3)/4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
4131 		(((msr >> 16) & 1) ? "EN" : "DIS"));
4132 
4133 	return;
4134 }
4135 
4136 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4137 {
4138 	unsigned long long msr;
4139 	const char *msr_name;
4140 	int cpu;
4141 
4142 	if (!do_rapl)
4143 		return 0;
4144 
4145 	/* RAPL counters are per package, so print only for 1st thread/package */
4146 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4147 		return 0;
4148 
4149 	cpu = t->cpu_id;
4150 	if (cpu_migrate(cpu)) {
4151 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4152 		return -1;
4153 	}
4154 
4155 	if (do_rapl & RAPL_AMD_F17H) {
4156 		msr_name = "MSR_RAPL_PWR_UNIT";
4157 		if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr))
4158 			return -1;
4159 	} else {
4160 		msr_name = "MSR_RAPL_POWER_UNIT";
4161 		if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
4162 			return -1;
4163 	}
4164 
4165 	fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr,
4166 		rapl_power_units, rapl_energy_units, rapl_time_units);
4167 
4168 	if (do_rapl & RAPL_PKG_POWER_INFO) {
4169 
4170 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
4171                 	return -5;
4172 
4173 
4174 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4175 			cpu, msr,
4176 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4177 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4178 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4179 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4180 
4181 	}
4182 	if (do_rapl & RAPL_PKG) {
4183 
4184 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
4185 			return -9;
4186 
4187 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
4188 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
4189 
4190 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
4191 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%f Watts, %f* sec, clamp %sabled)\n",
4192 			cpu,
4193 			((msr >> 47) & 1) ? "EN" : "DIS",
4194 			((msr >> 32) & 0x7FFF) * rapl_power_units,
4195 			(1.0 + (((msr >> 54) & 0x3)/4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
4196 			((msr >> 48) & 1) ? "EN" : "DIS");
4197 	}
4198 
4199 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
4200 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
4201                 	return -6;
4202 
4203 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4204 			cpu, msr,
4205 			((msr >>  0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4206 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4207 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4208 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4209 	}
4210 	if (do_rapl & RAPL_DRAM) {
4211 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
4212 			return -9;
4213 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
4214 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4215 
4216 		print_power_limit_msr(cpu, msr, "DRAM Limit");
4217 	}
4218 	if (do_rapl & RAPL_CORE_POLICY) {
4219 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
4220 			return -7;
4221 
4222 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
4223 	}
4224 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
4225 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
4226 			return -9;
4227 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
4228 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4229 		print_power_limit_msr(cpu, msr, "Cores Limit");
4230 	}
4231 	if (do_rapl & RAPL_GFX) {
4232 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
4233 			return -8;
4234 
4235 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
4236 
4237 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
4238 			return -9;
4239 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
4240 				cpu, msr, (msr >> 31) & 1 ? "" : "UN");
4241 		print_power_limit_msr(cpu, msr, "GFX Limit");
4242 	}
4243 	return 0;
4244 }
4245 
4246 /*
4247  * SNB adds support for additional MSRs:
4248  *
4249  * MSR_PKG_C7_RESIDENCY            0x000003fa
4250  * MSR_CORE_C7_RESIDENCY           0x000003fe
4251  * MSR_PKG_C2_RESIDENCY            0x0000060d
4252  */
4253 
4254 int has_snb_msrs(unsigned int family, unsigned int model)
4255 {
4256 	if (!genuine_intel)
4257 		return 0;
4258 
4259 	switch (model) {
4260 	case INTEL_FAM6_SANDYBRIDGE:
4261 	case INTEL_FAM6_SANDYBRIDGE_X:
4262 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
4263 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
4264 	case INTEL_FAM6_HASWELL_CORE:	/* HSW */
4265 	case INTEL_FAM6_HASWELL_X:	/* HSW */
4266 	case INTEL_FAM6_HASWELL_GT3E:	/* HSW */
4267 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4268 	case INTEL_FAM6_BROADWELL_GT3E:	/* BDW */
4269 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4270 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4271 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4272 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4273 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4274 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4275 	case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4276 		return 1;
4277 	}
4278 	return 0;
4279 }
4280 
4281 /*
4282  * HSW adds support for additional MSRs:
4283  *
4284  * MSR_PKG_C8_RESIDENCY		0x00000630
4285  * MSR_PKG_C9_RESIDENCY		0x00000631
4286  * MSR_PKG_C10_RESIDENCY	0x00000632
4287  *
4288  * MSR_PKGC8_IRTL		0x00000633
4289  * MSR_PKGC9_IRTL		0x00000634
4290  * MSR_PKGC10_IRTL		0x00000635
4291  *
4292  */
4293 int has_hsw_msrs(unsigned int family, unsigned int model)
4294 {
4295 	if (!genuine_intel)
4296 		return 0;
4297 
4298 	switch (model) {
4299 	case INTEL_FAM6_HASWELL_CORE:
4300 	case INTEL_FAM6_BROADWELL_CORE:	/* BDW */
4301 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4302 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4303 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4304 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4305 		return 1;
4306 	}
4307 	return 0;
4308 }
4309 
4310 /*
4311  * SKL adds support for additional MSRS:
4312  *
4313  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
4314  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
4315  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
4316  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
4317  */
4318 int has_skl_msrs(unsigned int family, unsigned int model)
4319 {
4320 	if (!genuine_intel)
4321 		return 0;
4322 
4323 	switch (model) {
4324 	case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4325 	case INTEL_FAM6_CANNONLAKE_MOBILE:	/* CNL */
4326 		return 1;
4327 	}
4328 	return 0;
4329 }
4330 
4331 int is_slm(unsigned int family, unsigned int model)
4332 {
4333 	if (!genuine_intel)
4334 		return 0;
4335 	switch (model) {
4336 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
4337 	case INTEL_FAM6_ATOM_SILVERMONT_X:	/* AVN */
4338 		return 1;
4339 	}
4340 	return 0;
4341 }
4342 
4343 int is_knl(unsigned int family, unsigned int model)
4344 {
4345 	if (!genuine_intel)
4346 		return 0;
4347 	switch (model) {
4348 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4349 		return 1;
4350 	}
4351 	return 0;
4352 }
4353 
4354 int is_cnl(unsigned int family, unsigned int model)
4355 {
4356 	if (!genuine_intel)
4357 		return 0;
4358 
4359 	switch (model) {
4360 	case INTEL_FAM6_CANNONLAKE_MOBILE: /* CNL */
4361 		return 1;
4362 	}
4363 
4364 	return 0;
4365 }
4366 
4367 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
4368 {
4369 	if (is_knl(family, model))
4370 		return 1024;
4371 	return 1;
4372 }
4373 
4374 #define SLM_BCLK_FREQS 5
4375 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0};
4376 
4377 double slm_bclk(void)
4378 {
4379 	unsigned long long msr = 3;
4380 	unsigned int i;
4381 	double freq;
4382 
4383 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
4384 		fprintf(outf, "SLM BCLK: unknown\n");
4385 
4386 	i = msr & 0xf;
4387 	if (i >= SLM_BCLK_FREQS) {
4388 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
4389 		i = 3;
4390 	}
4391 	freq = slm_freq_table[i];
4392 
4393 	if (!quiet)
4394 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
4395 
4396 	return freq;
4397 }
4398 
4399 double discover_bclk(unsigned int family, unsigned int model)
4400 {
4401 	if (has_snb_msrs(family, model) || is_knl(family, model))
4402 		return 100.00;
4403 	else if (is_slm(family, model))
4404 		return slm_bclk();
4405 	else
4406 		return 133.33;
4407 }
4408 
4409 /*
4410  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
4411  * the Thermal Control Circuit (TCC) activates.
4412  * This is usually equal to tjMax.
4413  *
4414  * Older processors do not have this MSR, so there we guess,
4415  * but also allow cmdline over-ride with -T.
4416  *
4417  * Several MSR temperature values are in units of degrees-C
4418  * below this value, including the Digital Thermal Sensor (DTS),
4419  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
4420  */
4421 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4422 {
4423 	unsigned long long msr;
4424 	unsigned int target_c_local;
4425 	int cpu;
4426 
4427 	/* tcc_activation_temp is used only for dts or ptm */
4428 	if (!(do_dts || do_ptm))
4429 		return 0;
4430 
4431 	/* this is a per-package concept */
4432 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4433 		return 0;
4434 
4435 	cpu = t->cpu_id;
4436 	if (cpu_migrate(cpu)) {
4437 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
4438 		return -1;
4439 	}
4440 
4441 	if (tcc_activation_temp_override != 0) {
4442 		tcc_activation_temp = tcc_activation_temp_override;
4443 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
4444 			cpu, tcc_activation_temp);
4445 		return 0;
4446 	}
4447 
4448 	/* Temperature Target MSR is Nehalem and newer only */
4449 	if (!do_nhm_platform_info)
4450 		goto guess;
4451 
4452 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
4453 		goto guess;
4454 
4455 	target_c_local = (msr >> 16) & 0xFF;
4456 
4457 	if (!quiet)
4458 		fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
4459 			cpu, msr, target_c_local);
4460 
4461 	if (!target_c_local)
4462 		goto guess;
4463 
4464 	tcc_activation_temp = target_c_local;
4465 
4466 	return 0;
4467 
4468 guess:
4469 	tcc_activation_temp = TJMAX_DEFAULT;
4470 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
4471 		cpu, tcc_activation_temp);
4472 
4473 	return 0;
4474 }
4475 
4476 void decode_feature_control_msr(void)
4477 {
4478 	unsigned long long msr;
4479 
4480 	if (!get_msr(base_cpu, MSR_IA32_FEATURE_CONTROL, &msr))
4481 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
4482 			base_cpu, msr,
4483 			msr & FEATURE_CONTROL_LOCKED ? "" : "UN-",
4484 			msr & (1 << 18) ? "SGX" : "");
4485 }
4486 
4487 void decode_misc_enable_msr(void)
4488 {
4489 	unsigned long long msr;
4490 
4491 	if (!genuine_intel)
4492 		return;
4493 
4494 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
4495 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
4496 			base_cpu, msr,
4497 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
4498 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
4499 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-",
4500 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
4501 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
4502 }
4503 
4504 void decode_misc_feature_control(void)
4505 {
4506 	unsigned long long msr;
4507 
4508 	if (!has_misc_feature_control)
4509 		return;
4510 
4511 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
4512 		fprintf(outf, "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
4513 			base_cpu, msr,
4514 			msr & (0 << 0) ? "No-" : "",
4515 			msr & (1 << 0) ? "No-" : "",
4516 			msr & (2 << 0) ? "No-" : "",
4517 			msr & (3 << 0) ? "No-" : "");
4518 }
4519 /*
4520  * Decode MSR_MISC_PWR_MGMT
4521  *
4522  * Decode the bits according to the Nehalem documentation
4523  * bit[0] seems to continue to have same meaning going forward
4524  * bit[1] less so...
4525  */
4526 void decode_misc_pwr_mgmt_msr(void)
4527 {
4528 	unsigned long long msr;
4529 
4530 	if (!do_nhm_platform_info)
4531 		return;
4532 
4533 	if (no_MSR_MISC_PWR_MGMT)
4534 		return;
4535 
4536 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
4537 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
4538 			base_cpu, msr,
4539 			msr & (1 << 0) ? "DIS" : "EN",
4540 			msr & (1 << 1) ? "EN" : "DIS",
4541 			msr & (1 << 8) ? "EN" : "DIS");
4542 }
4543 /*
4544  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
4545  *
4546  * This MSRs are present on Silvermont processors,
4547  * Intel Atom processor E3000 series (Baytrail), and friends.
4548  */
4549 void decode_c6_demotion_policy_msr(void)
4550 {
4551 	unsigned long long msr;
4552 
4553 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
4554 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
4555 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4556 
4557 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
4558 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
4559 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
4560 }
4561 
4562 /*
4563  * When models are the same, for the purpose of turbostat, reuse
4564  */
4565 unsigned int intel_model_duplicates(unsigned int model)
4566 {
4567 
4568 	switch(model) {
4569 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
4570 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
4571 	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
4572 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
4573 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
4574 		return INTEL_FAM6_NEHALEM;
4575 
4576 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
4577 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
4578 		return INTEL_FAM6_NEHALEM_EX;
4579 
4580 	case INTEL_FAM6_XEON_PHI_KNM:
4581 		return INTEL_FAM6_XEON_PHI_KNL;
4582 
4583 	case INTEL_FAM6_HASWELL_ULT:
4584 		return INTEL_FAM6_HASWELL_CORE;
4585 
4586 	case INTEL_FAM6_BROADWELL_X:
4587 	case INTEL_FAM6_BROADWELL_XEON_D:	/* BDX-DE */
4588 		return INTEL_FAM6_BROADWELL_X;
4589 
4590 	case INTEL_FAM6_SKYLAKE_MOBILE:
4591 	case INTEL_FAM6_SKYLAKE_DESKTOP:
4592 	case INTEL_FAM6_KABYLAKE_MOBILE:
4593 	case INTEL_FAM6_KABYLAKE_DESKTOP:
4594 		return INTEL_FAM6_SKYLAKE_MOBILE;
4595 
4596 	case INTEL_FAM6_ICELAKE_MOBILE:
4597 		return INTEL_FAM6_CANNONLAKE_MOBILE;
4598 	}
4599 	return model;
4600 }
4601 void process_cpuid()
4602 {
4603 	unsigned int eax, ebx, ecx, edx;
4604 	unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
4605 	unsigned int has_turbo;
4606 
4607 	eax = ebx = ecx = edx = 0;
4608 
4609 	__cpuid(0, max_level, ebx, ecx, edx);
4610 
4611 	if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
4612 		genuine_intel = 1;
4613 	else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
4614 		authentic_amd = 1;
4615 
4616 	if (!quiet)
4617 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
4618 			(char *)&ebx, (char *)&edx, (char *)&ecx);
4619 
4620 	__cpuid(1, fms, ebx, ecx, edx);
4621 	family = (fms >> 8) & 0xf;
4622 	model = (fms >> 4) & 0xf;
4623 	stepping = fms & 0xf;
4624 	if (family == 0xf)
4625 		family += (fms >> 20) & 0xff;
4626 	if (family >= 6)
4627 		model += ((fms >> 16) & 0xf) << 4;
4628 	ecx_flags = ecx;
4629 	edx_flags = edx;
4630 
4631 	/*
4632 	 * check max extended function levels of CPUID.
4633 	 * This is needed to check for invariant TSC.
4634 	 * This check is valid for both Intel and AMD.
4635 	 */
4636 	ebx = ecx = edx = 0;
4637 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
4638 
4639 	if (!quiet) {
4640 		fprintf(outf, "0x%x CPUID levels; 0x%x xlevels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
4641 			max_level, max_extended_level, family, model, stepping, family, model, stepping);
4642 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n",
4643 			ecx_flags & (1 << 0) ? "SSE3" : "-",
4644 			ecx_flags & (1 << 3) ? "MONITOR" : "-",
4645 			ecx_flags & (1 << 6) ? "SMX" : "-",
4646 			ecx_flags & (1 << 7) ? "EIST" : "-",
4647 			ecx_flags & (1 << 8) ? "TM2" : "-",
4648 			edx_flags & (1 << 4) ? "TSC" : "-",
4649 			edx_flags & (1 << 5) ? "MSR" : "-",
4650 			edx_flags & (1 << 22) ? "ACPI-TM" : "-",
4651 			edx_flags & (1 << 28) ? "HT" : "-",
4652 			edx_flags & (1 << 29) ? "TM" : "-");
4653 	}
4654 	if (genuine_intel)
4655 		model = intel_model_duplicates(model);
4656 
4657 	if (!(edx_flags & (1 << 5)))
4658 		errx(1, "CPUID: no MSR");
4659 
4660 	if (max_extended_level >= 0x80000007) {
4661 
4662 		/*
4663 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
4664 		 * this check is valid for both Intel and AMD
4665 		 */
4666 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4667 		has_invariant_tsc = edx & (1 << 8);
4668 	}
4669 
4670 	/*
4671 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
4672 	 * this check is valid for both Intel and AMD
4673 	 */
4674 
4675 	__cpuid(0x6, eax, ebx, ecx, edx);
4676 	has_aperf = ecx & (1 << 0);
4677 	if (has_aperf) {
4678 		BIC_PRESENT(BIC_Avg_MHz);
4679 		BIC_PRESENT(BIC_Busy);
4680 		BIC_PRESENT(BIC_Bzy_MHz);
4681 	}
4682 	do_dts = eax & (1 << 0);
4683 	if (do_dts)
4684 		BIC_PRESENT(BIC_CoreTmp);
4685 	has_turbo = eax & (1 << 1);
4686 	do_ptm = eax & (1 << 6);
4687 	if (do_ptm)
4688 		BIC_PRESENT(BIC_PkgTmp);
4689 	has_hwp = eax & (1 << 7);
4690 	has_hwp_notify = eax & (1 << 8);
4691 	has_hwp_activity_window = eax & (1 << 9);
4692 	has_hwp_epp = eax & (1 << 10);
4693 	has_hwp_pkg = eax & (1 << 11);
4694 	has_epb = ecx & (1 << 3);
4695 
4696 	if (!quiet)
4697 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
4698 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
4699 			has_aperf ? "" : "No-",
4700 			has_turbo ? "" : "No-",
4701 			do_dts ? "" : "No-",
4702 			do_ptm ? "" : "No-",
4703 			has_hwp ? "" : "No-",
4704 			has_hwp_notify ? "" : "No-",
4705 			has_hwp_activity_window ? "" : "No-",
4706 			has_hwp_epp ? "" : "No-",
4707 			has_hwp_pkg ? "" : "No-",
4708 			has_epb ? "" : "No-");
4709 
4710 	if (!quiet)
4711 		decode_misc_enable_msr();
4712 
4713 
4714 	if (max_level >= 0x7 && !quiet) {
4715 		int has_sgx;
4716 
4717 		ecx = 0;
4718 
4719 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
4720 
4721 		has_sgx = ebx & (1 << 2);
4722 		fprintf(outf, "CPUID(7): %sSGX\n", has_sgx ? "" : "No-");
4723 
4724 		if (has_sgx)
4725 			decode_feature_control_msr();
4726 	}
4727 
4728 	if (max_level >= 0x15) {
4729 		unsigned int eax_crystal;
4730 		unsigned int ebx_tsc;
4731 
4732 		/*
4733 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
4734 		 */
4735 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
4736 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
4737 
4738 		if (ebx_tsc != 0) {
4739 
4740 			if (!quiet && (ebx != 0))
4741 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
4742 					eax_crystal, ebx_tsc, crystal_hz);
4743 
4744 			if (crystal_hz == 0)
4745 				switch(model) {
4746 				case INTEL_FAM6_SKYLAKE_MOBILE:	/* SKL */
4747 					crystal_hz = 24000000;	/* 24.0 MHz */
4748 					break;
4749 				case INTEL_FAM6_ATOM_GOLDMONT_X:	/* DNV */
4750 					crystal_hz = 25000000;	/* 25.0 MHz */
4751 					break;
4752 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4753 				case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4754 					crystal_hz = 19200000;	/* 19.2 MHz */
4755 					break;
4756 				default:
4757 					crystal_hz = 0;
4758 			}
4759 
4760 			if (crystal_hz) {
4761 				tsc_hz =  (unsigned long long) crystal_hz * ebx_tsc / eax_crystal;
4762 				if (!quiet)
4763 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
4764 						tsc_hz / 1000000, crystal_hz, ebx_tsc,  eax_crystal);
4765 			}
4766 		}
4767 	}
4768 	if (max_level >= 0x16) {
4769 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
4770 
4771 		/*
4772 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
4773 		 */
4774 		base_mhz = max_mhz = bus_mhz = edx = 0;
4775 
4776 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
4777 		if (!quiet)
4778 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
4779 				base_mhz, max_mhz, bus_mhz);
4780 	}
4781 
4782 	if (has_aperf)
4783 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
4784 
4785 	BIC_PRESENT(BIC_IRQ);
4786 	BIC_PRESENT(BIC_TSC_MHz);
4787 
4788 	if (probe_nhm_msrs(family, model)) {
4789 		do_nhm_platform_info = 1;
4790 		BIC_PRESENT(BIC_CPU_c1);
4791 		BIC_PRESENT(BIC_CPU_c3);
4792 		BIC_PRESENT(BIC_CPU_c6);
4793 		BIC_PRESENT(BIC_SMI);
4794 	}
4795 	do_snb_cstates = has_snb_msrs(family, model);
4796 
4797 	if (do_snb_cstates)
4798 		BIC_PRESENT(BIC_CPU_c7);
4799 
4800 	do_irtl_snb = has_snb_msrs(family, model);
4801 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
4802 		BIC_PRESENT(BIC_Pkgpc2);
4803 	if (pkg_cstate_limit >= PCL__3)
4804 		BIC_PRESENT(BIC_Pkgpc3);
4805 	if (pkg_cstate_limit >= PCL__6)
4806 		BIC_PRESENT(BIC_Pkgpc6);
4807 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
4808 		BIC_PRESENT(BIC_Pkgpc7);
4809 	if (has_slv_msrs(family, model)) {
4810 		BIC_NOT_PRESENT(BIC_Pkgpc2);
4811 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4812 		BIC_PRESENT(BIC_Pkgpc6);
4813 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4814 		BIC_PRESENT(BIC_Mod_c6);
4815 		use_c1_residency_msr = 1;
4816 	}
4817 	if (is_dnv(family, model)) {
4818 		BIC_PRESENT(BIC_CPU_c1);
4819 		BIC_NOT_PRESENT(BIC_CPU_c3);
4820 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4821 		BIC_NOT_PRESENT(BIC_CPU_c7);
4822 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4823 		use_c1_residency_msr = 1;
4824 	}
4825 	if (is_skx(family, model)) {
4826 		BIC_NOT_PRESENT(BIC_CPU_c3);
4827 		BIC_NOT_PRESENT(BIC_Pkgpc3);
4828 		BIC_NOT_PRESENT(BIC_CPU_c7);
4829 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4830 	}
4831 	if (is_bdx(family, model)) {
4832 		BIC_NOT_PRESENT(BIC_CPU_c7);
4833 		BIC_NOT_PRESENT(BIC_Pkgpc7);
4834 	}
4835 	if (has_hsw_msrs(family, model)) {
4836 		BIC_PRESENT(BIC_Pkgpc8);
4837 		BIC_PRESENT(BIC_Pkgpc9);
4838 		BIC_PRESENT(BIC_Pkgpc10);
4839 	}
4840 	do_irtl_hsw = has_hsw_msrs(family, model);
4841 	if (has_skl_msrs(family, model)) {
4842 		BIC_PRESENT(BIC_Totl_c0);
4843 		BIC_PRESENT(BIC_Any_c0);
4844 		BIC_PRESENT(BIC_GFX_c0);
4845 		BIC_PRESENT(BIC_CPUGFX);
4846 	}
4847 	do_slm_cstates = is_slm(family, model);
4848 	do_knl_cstates  = is_knl(family, model);
4849 
4850 	if (do_slm_cstates || do_knl_cstates || is_cnl(family, model))
4851 		BIC_NOT_PRESENT(BIC_CPU_c3);
4852 
4853 	if (!quiet)
4854 		decode_misc_pwr_mgmt_msr();
4855 
4856 	if (!quiet && has_slv_msrs(family, model))
4857 		decode_c6_demotion_policy_msr();
4858 
4859 	rapl_probe(family, model);
4860 	perf_limit_reasons_probe(family, model);
4861 	automatic_cstate_conversion_probe(family, model);
4862 
4863 	if (!quiet)
4864 		dump_cstate_pstate_config_info(family, model);
4865 
4866 	if (!quiet)
4867 		dump_sysfs_cstate_config();
4868 	if (!quiet)
4869 		dump_sysfs_pstate_config();
4870 
4871 	if (has_skl_msrs(family, model))
4872 		calculate_tsc_tweak();
4873 
4874 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
4875 		BIC_PRESENT(BIC_GFX_rc6);
4876 
4877 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
4878 		BIC_PRESENT(BIC_GFXMHz);
4879 
4880 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK))
4881 		BIC_PRESENT(BIC_CPU_LPI);
4882 	else
4883 		BIC_NOT_PRESENT(BIC_CPU_LPI);
4884 
4885 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us", R_OK))
4886 		BIC_PRESENT(BIC_SYS_LPI);
4887 	else
4888 		BIC_NOT_PRESENT(BIC_SYS_LPI);
4889 
4890 	if (!quiet)
4891 		decode_misc_feature_control();
4892 
4893 	return;
4894 }
4895 
4896 /*
4897  * in /dev/cpu/ return success for names that are numbers
4898  * ie. filter out ".", "..", "microcode".
4899  */
4900 int dir_filter(const struct dirent *dirp)
4901 {
4902 	if (isdigit(dirp->d_name[0]))
4903 		return 1;
4904 	else
4905 		return 0;
4906 }
4907 
4908 int open_dev_cpu_msr(int dummy1)
4909 {
4910 	return 0;
4911 }
4912 
4913 void topology_probe()
4914 {
4915 	int i;
4916 	int max_core_id = 0;
4917 	int max_package_id = 0;
4918 	int max_die_id = 0;
4919 	int max_siblings = 0;
4920 
4921 	/* Initialize num_cpus, max_cpu_num */
4922 	set_max_cpu_num();
4923 	topo.num_cpus = 0;
4924 	for_all_proc_cpus(count_cpus);
4925 	if (!summary_only && topo.num_cpus > 1)
4926 		BIC_PRESENT(BIC_CPU);
4927 
4928 	if (debug > 1)
4929 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
4930 
4931 	cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
4932 	if (cpus == NULL)
4933 		err(1, "calloc cpus");
4934 
4935 	/*
4936 	 * Allocate and initialize cpu_present_set
4937 	 */
4938 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
4939 	if (cpu_present_set == NULL)
4940 		err(3, "CPU_ALLOC");
4941 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4942 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
4943 	for_all_proc_cpus(mark_cpu_present);
4944 
4945 	/*
4946 	 * Validate that all cpus in cpu_subset are also in cpu_present_set
4947 	 */
4948 	for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
4949 		if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
4950 			if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
4951 				err(1, "cpu%d not present", i);
4952 	}
4953 
4954 	/*
4955 	 * Allocate and initialize cpu_affinity_set
4956 	 */
4957 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
4958 	if (cpu_affinity_set == NULL)
4959 		err(3, "CPU_ALLOC");
4960 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
4961 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
4962 
4963 	for_all_proc_cpus(init_thread_id);
4964 
4965 	/*
4966 	 * For online cpus
4967 	 * find max_core_id, max_package_id
4968 	 */
4969 	for (i = 0; i <= topo.max_cpu_num; ++i) {
4970 		int siblings;
4971 
4972 		if (cpu_is_not_present(i)) {
4973 			if (debug > 1)
4974 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
4975 			continue;
4976 		}
4977 
4978 		cpus[i].logical_cpu_id = i;
4979 
4980 		/* get package information */
4981 		cpus[i].physical_package_id = get_physical_package_id(i);
4982 		if (cpus[i].physical_package_id > max_package_id)
4983 			max_package_id = cpus[i].physical_package_id;
4984 
4985 		/* get die information */
4986 		cpus[i].die_id = get_die_id(i);
4987 		if (cpus[i].die_id > max_die_id)
4988 			max_die_id = cpus[i].die_id;
4989 
4990 		/* get numa node information */
4991 		cpus[i].physical_node_id = get_physical_node_id(&cpus[i]);
4992 		if (cpus[i].physical_node_id > topo.max_node_num)
4993 			topo.max_node_num = cpus[i].physical_node_id;
4994 
4995 		/* get core information */
4996 		cpus[i].physical_core_id = get_core_id(i);
4997 		if (cpus[i].physical_core_id > max_core_id)
4998 			max_core_id = cpus[i].physical_core_id;
4999 
5000 		/* get thread information */
5001 		siblings = get_thread_siblings(&cpus[i]);
5002 		if (siblings > max_siblings)
5003 			max_siblings = siblings;
5004 		if (cpus[i].thread_id == 0)
5005 			topo.num_cores++;
5006 	}
5007 
5008 	topo.cores_per_node = max_core_id + 1;
5009 	if (debug > 1)
5010 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n",
5011 			max_core_id, topo.cores_per_node);
5012 	if (!summary_only && topo.cores_per_node > 1)
5013 		BIC_PRESENT(BIC_Core);
5014 
5015 	topo.num_die = max_die_id + 1;
5016 	if (debug > 1)
5017 		fprintf(outf, "max_die_id %d, sizing for %d die\n",
5018 				max_die_id, topo.num_die);
5019 	if (!summary_only && topo.num_die > 1)
5020 		BIC_PRESENT(BIC_Die);
5021 
5022 	topo.num_packages = max_package_id + 1;
5023 	if (debug > 1)
5024 		fprintf(outf, "max_package_id %d, sizing for %d packages\n",
5025 			max_package_id, topo.num_packages);
5026 	if (!summary_only && topo.num_packages > 1)
5027 		BIC_PRESENT(BIC_Package);
5028 
5029 	set_node_data();
5030 	if (debug > 1)
5031 		fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg);
5032 	if (!summary_only && topo.nodes_per_pkg > 1)
5033 		BIC_PRESENT(BIC_Node);
5034 
5035 	topo.threads_per_core = max_siblings;
5036 	if (debug > 1)
5037 		fprintf(outf, "max_siblings %d\n", max_siblings);
5038 
5039 	if (debug < 1)
5040 		return;
5041 
5042 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5043 		if (cpu_is_not_present(i))
5044 			continue;
5045 		fprintf(outf,
5046 			"cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n",
5047 			i, cpus[i].physical_package_id, cpus[i].die_id,
5048 			cpus[i].physical_node_id,
5049 			cpus[i].logical_node_id,
5050 			cpus[i].physical_core_id,
5051 			cpus[i].thread_id);
5052 	}
5053 
5054 }
5055 
5056 void
5057 allocate_counters(struct thread_data **t, struct core_data **c,
5058 		  struct pkg_data **p)
5059 {
5060 	int i;
5061 	int num_cores = topo.cores_per_node * topo.nodes_per_pkg *
5062 			topo.num_packages;
5063 	int num_threads = topo.threads_per_core * num_cores;
5064 
5065 	*t = calloc(num_threads, sizeof(struct thread_data));
5066 	if (*t == NULL)
5067 		goto error;
5068 
5069 	for (i = 0; i < num_threads; i++)
5070 		(*t)[i].cpu_id = -1;
5071 
5072 	*c = calloc(num_cores, sizeof(struct core_data));
5073 	if (*c == NULL)
5074 		goto error;
5075 
5076 	for (i = 0; i < num_cores; i++)
5077 		(*c)[i].core_id = -1;
5078 
5079 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
5080 	if (*p == NULL)
5081 		goto error;
5082 
5083 	for (i = 0; i < topo.num_packages; i++)
5084 		(*p)[i].package_id = i;
5085 
5086 	return;
5087 error:
5088 	err(1, "calloc counters");
5089 }
5090 /*
5091  * init_counter()
5092  *
5093  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
5094  */
5095 void init_counter(struct thread_data *thread_base, struct core_data *core_base,
5096 	struct pkg_data *pkg_base, int cpu_id)
5097 {
5098 	int pkg_id = cpus[cpu_id].physical_package_id;
5099 	int node_id = cpus[cpu_id].logical_node_id;
5100 	int core_id = cpus[cpu_id].physical_core_id;
5101 	int thread_id = cpus[cpu_id].thread_id;
5102 	struct thread_data *t;
5103 	struct core_data *c;
5104 	struct pkg_data *p;
5105 
5106 
5107 	/* Workaround for systems where physical_node_id==-1
5108 	 * and logical_node_id==(-1 - topo.num_cpus)
5109 	 */
5110 	if (node_id < 0)
5111 		node_id = 0;
5112 
5113 	t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id);
5114 	c = GET_CORE(core_base, core_id, node_id, pkg_id);
5115 	p = GET_PKG(pkg_base, pkg_id);
5116 
5117 	t->cpu_id = cpu_id;
5118 	if (thread_id == 0) {
5119 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
5120 		if (cpu_is_first_core_in_package(cpu_id))
5121 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
5122 	}
5123 
5124 	c->core_id = core_id;
5125 	p->package_id = pkg_id;
5126 }
5127 
5128 
5129 int initialize_counters(int cpu_id)
5130 {
5131 	init_counter(EVEN_COUNTERS, cpu_id);
5132 	init_counter(ODD_COUNTERS, cpu_id);
5133 	return 0;
5134 }
5135 
5136 void allocate_output_buffer()
5137 {
5138 	output_buffer = calloc(1, (1 + topo.num_cpus) * 1024);
5139 	outp = output_buffer;
5140 	if (outp == NULL)
5141 		err(-1, "calloc output buffer");
5142 }
5143 void allocate_fd_percpu(void)
5144 {
5145 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5146 	if (fd_percpu == NULL)
5147 		err(-1, "calloc fd_percpu");
5148 }
5149 void allocate_irq_buffers(void)
5150 {
5151 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
5152 	if (irq_column_2_cpu == NULL)
5153 		err(-1, "calloc %d", topo.num_cpus);
5154 
5155 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5156 	if (irqs_per_cpu == NULL)
5157 		err(-1, "calloc %d", topo.max_cpu_num + 1);
5158 }
5159 void setup_all_buffers(void)
5160 {
5161 	topology_probe();
5162 	allocate_irq_buffers();
5163 	allocate_fd_percpu();
5164 	allocate_counters(&thread_even, &core_even, &package_even);
5165 	allocate_counters(&thread_odd, &core_odd, &package_odd);
5166 	allocate_output_buffer();
5167 	for_all_proc_cpus(initialize_counters);
5168 }
5169 
5170 void set_base_cpu(void)
5171 {
5172 	base_cpu = sched_getcpu();
5173 	if (base_cpu < 0)
5174 		err(-ENODEV, "No valid cpus found");
5175 
5176 	if (debug > 1)
5177 		fprintf(outf, "base_cpu = %d\n", base_cpu);
5178 }
5179 
5180 void turbostat_init()
5181 {
5182 	setup_all_buffers();
5183 	set_base_cpu();
5184 	check_dev_msr();
5185 	check_permissions();
5186 	process_cpuid();
5187 
5188 
5189 	if (!quiet)
5190 		for_all_cpus(print_hwp, ODD_COUNTERS);
5191 
5192 	if (!quiet)
5193 		for_all_cpus(print_epb, ODD_COUNTERS);
5194 
5195 	if (!quiet)
5196 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
5197 
5198 	if (!quiet)
5199 		for_all_cpus(print_rapl, ODD_COUNTERS);
5200 
5201 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
5202 
5203 	if (!quiet)
5204 		for_all_cpus(print_thermal, ODD_COUNTERS);
5205 
5206 	if (!quiet && do_irtl_snb)
5207 		print_irtl();
5208 }
5209 
5210 int fork_it(char **argv)
5211 {
5212 	pid_t child_pid;
5213 	int status;
5214 
5215 	snapshot_proc_sysfs_files();
5216 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
5217 	first_counter_read = 0;
5218 	if (status)
5219 		exit(status);
5220 	/* clear affinity side-effect of get_counters() */
5221 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
5222 	gettimeofday(&tv_even, (struct timezone *)NULL);
5223 
5224 	child_pid = fork();
5225 	if (!child_pid) {
5226 		/* child */
5227 		execvp(argv[0], argv);
5228 		err(errno, "exec %s", argv[0]);
5229 	} else {
5230 
5231 		/* parent */
5232 		if (child_pid == -1)
5233 			err(1, "fork");
5234 
5235 		signal(SIGINT, SIG_IGN);
5236 		signal(SIGQUIT, SIG_IGN);
5237 		if (waitpid(child_pid, &status, 0) == -1)
5238 			err(status, "waitpid");
5239 
5240 		if (WIFEXITED(status))
5241 			status = WEXITSTATUS(status);
5242 	}
5243 	/*
5244 	 * n.b. fork_it() does not check for errors from for_all_cpus()
5245 	 * because re-starting is problematic when forking
5246 	 */
5247 	snapshot_proc_sysfs_files();
5248 	for_all_cpus(get_counters, ODD_COUNTERS);
5249 	gettimeofday(&tv_odd, (struct timezone *)NULL);
5250 	timersub(&tv_odd, &tv_even, &tv_delta);
5251 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
5252 		fprintf(outf, "%s: Counter reset detected\n", progname);
5253 	else {
5254 		compute_average(EVEN_COUNTERS);
5255 		format_all_counters(EVEN_COUNTERS);
5256 	}
5257 
5258 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);
5259 
5260 	flush_output_stderr();
5261 
5262 	return status;
5263 }
5264 
5265 int get_and_dump_counters(void)
5266 {
5267 	int status;
5268 
5269 	snapshot_proc_sysfs_files();
5270 	status = for_all_cpus(get_counters, ODD_COUNTERS);
5271 	if (status)
5272 		return status;
5273 
5274 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
5275 	if (status)
5276 		return status;
5277 
5278 	flush_output_stdout();
5279 
5280 	return status;
5281 }
5282 
5283 void print_version() {
5284 	fprintf(outf, "turbostat version 19.03.20"
5285 		" - Len Brown <lenb@kernel.org>\n");
5286 }
5287 
5288 int add_counter(unsigned int msr_num, char *path, char *name,
5289 	unsigned int width, enum counter_scope scope,
5290 	enum counter_type type, enum counter_format format, int flags)
5291 {
5292 	struct msr_counter *msrp;
5293 
5294 	msrp = calloc(1, sizeof(struct msr_counter));
5295 	if (msrp == NULL) {
5296 		perror("calloc");
5297 		exit(1);
5298 	}
5299 
5300 	msrp->msr_num = msr_num;
5301 	strncpy(msrp->name, name, NAME_BYTES);
5302 	if (path)
5303 		strncpy(msrp->path, path, PATH_BYTES);
5304 	msrp->width = width;
5305 	msrp->type = type;
5306 	msrp->format = format;
5307 	msrp->flags = flags;
5308 
5309 	switch (scope) {
5310 
5311 	case SCOPE_CPU:
5312 		msrp->next = sys.tp;
5313 		sys.tp = msrp;
5314 		sys.added_thread_counters++;
5315 		if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) {
5316 			fprintf(stderr, "exceeded max %d added thread counters\n",
5317 				MAX_ADDED_COUNTERS);
5318 			exit(-1);
5319 		}
5320 		break;
5321 
5322 	case SCOPE_CORE:
5323 		msrp->next = sys.cp;
5324 		sys.cp = msrp;
5325 		sys.added_core_counters++;
5326 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
5327 			fprintf(stderr, "exceeded max %d added core counters\n",
5328 				MAX_ADDED_COUNTERS);
5329 			exit(-1);
5330 		}
5331 		break;
5332 
5333 	case SCOPE_PACKAGE:
5334 		msrp->next = sys.pp;
5335 		sys.pp = msrp;
5336 		sys.added_package_counters++;
5337 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
5338 			fprintf(stderr, "exceeded max %d added package counters\n",
5339 				MAX_ADDED_COUNTERS);
5340 			exit(-1);
5341 		}
5342 		break;
5343 	}
5344 
5345 	return 0;
5346 }
5347 
5348 void parse_add_command(char *add_command)
5349 {
5350 	int msr_num = 0;
5351 	char *path = NULL;
5352 	char name_buffer[NAME_BYTES] = "";
5353 	int width = 64;
5354 	int fail = 0;
5355 	enum counter_scope scope = SCOPE_CPU;
5356 	enum counter_type type = COUNTER_CYCLES;
5357 	enum counter_format format = FORMAT_DELTA;
5358 
5359 	while (add_command) {
5360 
5361 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
5362 			goto next;
5363 
5364 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
5365 			goto next;
5366 
5367 		if (*add_command == '/') {
5368 			path = add_command;
5369 			goto next;
5370 		}
5371 
5372 		if (sscanf(add_command, "u%d", &width) == 1) {
5373 			if ((width == 32) || (width == 64))
5374 				goto next;
5375 			width = 64;
5376 		}
5377 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
5378 			scope = SCOPE_CPU;
5379 			goto next;
5380 		}
5381 		if (!strncmp(add_command, "core", strlen("core"))) {
5382 			scope = SCOPE_CORE;
5383 			goto next;
5384 		}
5385 		if (!strncmp(add_command, "package", strlen("package"))) {
5386 			scope = SCOPE_PACKAGE;
5387 			goto next;
5388 		}
5389 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
5390 			type = COUNTER_CYCLES;
5391 			goto next;
5392 		}
5393 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
5394 			type = COUNTER_SECONDS;
5395 			goto next;
5396 		}
5397 		if (!strncmp(add_command, "usec", strlen("usec"))) {
5398 			type = COUNTER_USEC;
5399 			goto next;
5400 		}
5401 		if (!strncmp(add_command, "raw", strlen("raw"))) {
5402 			format = FORMAT_RAW;
5403 			goto next;
5404 		}
5405 		if (!strncmp(add_command, "delta", strlen("delta"))) {
5406 			format = FORMAT_DELTA;
5407 			goto next;
5408 		}
5409 		if (!strncmp(add_command, "percent", strlen("percent"))) {
5410 			format = FORMAT_PERCENT;
5411 			goto next;
5412 		}
5413 
5414 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
5415 			char *eos;
5416 
5417 			eos = strchr(name_buffer, ',');
5418 			if (eos)
5419 				*eos = '\0';
5420 			goto next;
5421 		}
5422 
5423 next:
5424 		add_command = strchr(add_command, ',');
5425 		if (add_command) {
5426 			*add_command = '\0';
5427 			add_command++;
5428 		}
5429 
5430 	}
5431 	if ((msr_num == 0) && (path == NULL)) {
5432 		fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
5433 		fail++;
5434 	}
5435 
5436 	/* generate default column header */
5437 	if (*name_buffer == '\0') {
5438 		if (width == 32)
5439 			sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5440 		else
5441 			sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
5442 	}
5443 
5444 	if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
5445 		fail++;
5446 
5447 	if (fail) {
5448 		help();
5449 		exit(1);
5450 	}
5451 }
5452 
5453 int is_deferred_skip(char *name)
5454 {
5455 	int i;
5456 
5457 	for (i = 0; i < deferred_skip_index; ++i)
5458 		if (!strcmp(name, deferred_skip_names[i]))
5459 			return 1;
5460 	return 0;
5461 }
5462 
5463 void probe_sysfs(void)
5464 {
5465 	char path[64];
5466 	char name_buf[16];
5467 	FILE *input;
5468 	int state;
5469 	char *sp;
5470 
5471 	if (!DO_BIC(BIC_sysfs))
5472 		return;
5473 
5474 	for (state = 10; state >= 0; --state) {
5475 
5476 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5477 			base_cpu, state);
5478 		input = fopen(path, "r");
5479 		if (input == NULL)
5480 			continue;
5481 		if (!fgets(name_buf, sizeof(name_buf), input))
5482 			err(1, "%s: failed to read file", path);
5483 
5484 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5485 		sp = strchr(name_buf, '-');
5486 		if (!sp)
5487 			sp = strchrnul(name_buf, '\n');
5488 		*sp = '%';
5489 		*(sp + 1) = '\0';
5490 
5491 		fclose(input);
5492 
5493 		sprintf(path, "cpuidle/state%d/time", state);
5494 
5495 		if (is_deferred_skip(name_buf))
5496 			continue;
5497 
5498 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC,
5499 				FORMAT_PERCENT, SYSFS_PERCPU);
5500 	}
5501 
5502 	for (state = 10; state >= 0; --state) {
5503 
5504 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name",
5505 			base_cpu, state);
5506 		input = fopen(path, "r");
5507 		if (input == NULL)
5508 			continue;
5509 		if (!fgets(name_buf, sizeof(name_buf), input))
5510 			err(1, "%s: failed to read file", path);
5511 		 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
5512 		sp = strchr(name_buf, '-');
5513 		if (!sp)
5514 			sp = strchrnul(name_buf, '\n');
5515 		*sp = '\0';
5516 		fclose(input);
5517 
5518 		sprintf(path, "cpuidle/state%d/usage", state);
5519 
5520 		if (is_deferred_skip(name_buf))
5521 			continue;
5522 
5523 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS,
5524 				FORMAT_DELTA, SYSFS_PERCPU);
5525 	}
5526 
5527 }
5528 
5529 
5530 /*
5531  * parse cpuset with following syntax
5532  * 1,2,4..6,8-10 and set bits in cpu_subset
5533  */
5534 void parse_cpu_command(char *optarg)
5535 {
5536 	unsigned int start, end;
5537 	char *next;
5538 
5539 	if (!strcmp(optarg, "core")) {
5540 		if (cpu_subset)
5541 			goto error;
5542 		show_core_only++;
5543 		return;
5544 	}
5545 	if (!strcmp(optarg, "package")) {
5546 		if (cpu_subset)
5547 			goto error;
5548 		show_pkg_only++;
5549 		return;
5550 	}
5551 	if (show_core_only || show_pkg_only)
5552 		goto error;
5553 
5554 	cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
5555 	if (cpu_subset == NULL)
5556 		err(3, "CPU_ALLOC");
5557 	cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
5558 
5559 	CPU_ZERO_S(cpu_subset_size, cpu_subset);
5560 
5561 	next = optarg;
5562 
5563 	while (next && *next) {
5564 
5565 		if (*next == '-')	/* no negative cpu numbers */
5566 			goto error;
5567 
5568 		start = strtoul(next, &next, 10);
5569 
5570 		if (start >= CPU_SUBSET_MAXCPUS)
5571 			goto error;
5572 		CPU_SET_S(start, cpu_subset_size, cpu_subset);
5573 
5574 		if (*next == '\0')
5575 			break;
5576 
5577 		if (*next == ',') {
5578 			next += 1;
5579 			continue;
5580 		}
5581 
5582 		if (*next == '-') {
5583 			next += 1;	/* start range */
5584 		} else if (*next == '.') {
5585 			next += 1;
5586 			if (*next == '.')
5587 				next += 1;	/* start range */
5588 			else
5589 				goto error;
5590 		}
5591 
5592 		end = strtoul(next, &next, 10);
5593 		if (end <= start)
5594 			goto error;
5595 
5596 		while (++start <= end) {
5597 			if (start >= CPU_SUBSET_MAXCPUS)
5598 				goto error;
5599 			CPU_SET_S(start, cpu_subset_size, cpu_subset);
5600 		}
5601 
5602 		if (*next == ',')
5603 			next += 1;
5604 		else if (*next != '\0')
5605 			goto error;
5606 	}
5607 
5608 	return;
5609 
5610 error:
5611 	fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
5612 	help();
5613 	exit(-1);
5614 }
5615 
5616 
5617 void cmdline(int argc, char **argv)
5618 {
5619 	int opt;
5620 	int option_index = 0;
5621 	static struct option long_options[] = {
5622 		{"add",		required_argument,	0, 'a'},
5623 		{"cpu",		required_argument,	0, 'c'},
5624 		{"Dump",	no_argument,		0, 'D'},
5625 		{"debug",	no_argument,		0, 'd'},	/* internal, not documented */
5626 		{"enable",	required_argument,	0, 'e'},
5627 		{"interval",	required_argument,	0, 'i'},
5628 		{"num_iterations",	required_argument,	0, 'n'},
5629 		{"help",	no_argument,		0, 'h'},
5630 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
5631 		{"Joules",	no_argument,		0, 'J'},
5632 		{"list",	no_argument,		0, 'l'},
5633 		{"out",		required_argument,	0, 'o'},
5634 		{"quiet",	no_argument,		0, 'q'},
5635 		{"show",	required_argument,	0, 's'},
5636 		{"Summary",	no_argument,		0, 'S'},
5637 		{"TCC",		required_argument,	0, 'T'},
5638 		{"version",	no_argument,		0, 'v' },
5639 		{0,		0,			0,  0 }
5640 	};
5641 
5642 	progname = argv[0];
5643 
5644 	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v",
5645 				long_options, &option_index)) != -1) {
5646 		switch (opt) {
5647 		case 'a':
5648 			parse_add_command(optarg);
5649 			break;
5650 		case 'c':
5651 			parse_cpu_command(optarg);
5652 			break;
5653 		case 'D':
5654 			dump_only++;
5655 			break;
5656 		case 'e':
5657 			/* --enable specified counter */
5658 			bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST);
5659 			break;
5660 		case 'd':
5661 			debug++;
5662 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5663 			break;
5664 		case 'H':
5665 			/*
5666 			 * --hide: do not show those specified
5667 			 *  multiple invocations simply clear more bits in enabled mask
5668 			 */
5669 			bic_enabled &= ~bic_lookup(optarg, HIDE_LIST);
5670 			break;
5671 		case 'h':
5672 		default:
5673 			help();
5674 			exit(1);
5675 		case 'i':
5676 			{
5677 				double interval = strtod(optarg, NULL);
5678 
5679 				if (interval < 0.001) {
5680 					fprintf(outf, "interval %f seconds is too small\n",
5681 						interval);
5682 					exit(2);
5683 				}
5684 
5685 				interval_tv.tv_sec = interval_ts.tv_sec = interval;
5686 				interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000;
5687 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
5688 			}
5689 			break;
5690 		case 'J':
5691 			rapl_joules++;
5692 			break;
5693 		case 'l':
5694 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
5695 			list_header_only++;
5696 			quiet++;
5697 			break;
5698 		case 'o':
5699 			outf = fopen_or_die(optarg, "w");
5700 			break;
5701 		case 'q':
5702 			quiet = 1;
5703 			break;
5704 		case 'n':
5705 			num_iterations = strtod(optarg, NULL);
5706 
5707 			if (num_iterations <= 0) {
5708 				fprintf(outf, "iterations %d should be positive number\n",
5709 					num_iterations);
5710 				exit(2);
5711 			}
5712 			break;
5713 		case 's':
5714 			/*
5715 			 * --show: show only those specified
5716 			 *  The 1st invocation will clear and replace the enabled mask
5717 			 *  subsequent invocations can add to it.
5718 			 */
5719 			if (shown == 0)
5720 				bic_enabled = bic_lookup(optarg, SHOW_LIST);
5721 			else
5722 				bic_enabled |= bic_lookup(optarg, SHOW_LIST);
5723 			shown = 1;
5724 			break;
5725 		case 'S':
5726 			summary_only++;
5727 			break;
5728 		case 'T':
5729 			tcc_activation_temp_override = atoi(optarg);
5730 			break;
5731 		case 'v':
5732 			print_version();
5733 			exit(0);
5734 			break;
5735 		}
5736 	}
5737 }
5738 
5739 int main(int argc, char **argv)
5740 {
5741 	outf = stderr;
5742 	cmdline(argc, argv);
5743 
5744 	if (!quiet)
5745 		print_version();
5746 
5747 	probe_sysfs();
5748 
5749 	turbostat_init();
5750 
5751 	/* dump counters and exit */
5752 	if (dump_only)
5753 		return get_and_dump_counters();
5754 
5755 	/* list header and exit */
5756 	if (list_header_only) {
5757 		print_header(",");
5758 		flush_output_stdout();
5759 		return 0;
5760 	}
5761 
5762 	/*
5763 	 * if any params left, it must be a command to fork
5764 	 */
5765 	if (argc - optind)
5766 		return fork_it(argv + optind);
5767 	else
5768 		turbostat_loop();
5769 
5770 	return 0;
5771 }
5772