1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * turbostat -- show CPU frequency and C-state residency
4  * on modern Intel and AMD processors.
5  *
6  * Copyright (c) 2023 Intel Corporation.
7  * Len Brown <len.brown@intel.com>
8  */
9 
10 #define _GNU_SOURCE
11 #include MSRHEADER
12 #include INTEL_FAMILY_HEADER
13 #include <stdarg.h>
14 #include <stdio.h>
15 #include <err.h>
16 #include <unistd.h>
17 #include <sys/types.h>
18 #include <sys/wait.h>
19 #include <sys/stat.h>
20 #include <sys/select.h>
21 #include <sys/resource.h>
22 #include <fcntl.h>
23 #include <signal.h>
24 #include <sys/time.h>
25 #include <stdlib.h>
26 #include <getopt.h>
27 #include <dirent.h>
28 #include <string.h>
29 #include <ctype.h>
30 #include <sched.h>
31 #include <time.h>
32 #include <cpuid.h>
33 #include <sys/capability.h>
34 #include <errno.h>
35 #include <math.h>
36 #include <linux/perf_event.h>
37 #include <asm/unistd.h>
38 #include <stdbool.h>
39 
40 #define UNUSED(x) (void)(x)
41 
42 /*
43  * This list matches the column headers, except
44  * 1. built-in only, the sysfs counters are not here -- we learn of those at run-time
45  * 2. Core and CPU are moved to the end, we can't have strings that contain them
46  *    matching on them for --show and --hide.
47  */
48 
49 /*
50  * buffer size used by sscanf() for added column names
51  * Usually truncated to 7 characters, but also handles 18 columns for raw 64-bit counters
52  */
53 #define	NAME_BYTES 20
54 #define PATH_BYTES 128
55 
56 #define MAX_NOFILE 0x8000
57 
58 enum counter_scope { SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE };
59 enum counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC };
60 enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT };
61 
62 struct msr_counter {
63 	unsigned int msr_num;
64 	char name[NAME_BYTES];
65 	char path[PATH_BYTES];
66 	unsigned int width;
67 	enum counter_type type;
68 	enum counter_format format;
69 	struct msr_counter *next;
70 	unsigned int flags;
71 #define	FLAGS_HIDE	(1 << 0)
72 #define	FLAGS_SHOW	(1 << 1)
73 #define	SYSFS_PERCPU	(1 << 1)
74 };
75 
76 struct msr_counter bic[] = {
77 	{ 0x0, "usec", "", 0, 0, 0, NULL, 0 },
78 	{ 0x0, "Time_Of_Day_Seconds", "", 0, 0, 0, NULL, 0 },
79 	{ 0x0, "Package", "", 0, 0, 0, NULL, 0 },
80 	{ 0x0, "Node", "", 0, 0, 0, NULL, 0 },
81 	{ 0x0, "Avg_MHz", "", 0, 0, 0, NULL, 0 },
82 	{ 0x0, "Busy%", "", 0, 0, 0, NULL, 0 },
83 	{ 0x0, "Bzy_MHz", "", 0, 0, 0, NULL, 0 },
84 	{ 0x0, "TSC_MHz", "", 0, 0, 0, NULL, 0 },
85 	{ 0x0, "IRQ", "", 0, 0, 0, NULL, 0 },
86 	{ 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL, 0 },
87 	{ 0x0, "sysfs", "", 0, 0, 0, NULL, 0 },
88 	{ 0x0, "CPU%c1", "", 0, 0, 0, NULL, 0 },
89 	{ 0x0, "CPU%c3", "", 0, 0, 0, NULL, 0 },
90 	{ 0x0, "CPU%c6", "", 0, 0, 0, NULL, 0 },
91 	{ 0x0, "CPU%c7", "", 0, 0, 0, NULL, 0 },
92 	{ 0x0, "ThreadC", "", 0, 0, 0, NULL, 0 },
93 	{ 0x0, "CoreTmp", "", 0, 0, 0, NULL, 0 },
94 	{ 0x0, "CoreCnt", "", 0, 0, 0, NULL, 0 },
95 	{ 0x0, "PkgTmp", "", 0, 0, 0, NULL, 0 },
96 	{ 0x0, "GFX%rc6", "", 0, 0, 0, NULL, 0 },
97 	{ 0x0, "GFXMHz", "", 0, 0, 0, NULL, 0 },
98 	{ 0x0, "Pkg%pc2", "", 0, 0, 0, NULL, 0 },
99 	{ 0x0, "Pkg%pc3", "", 0, 0, 0, NULL, 0 },
100 	{ 0x0, "Pkg%pc6", "", 0, 0, 0, NULL, 0 },
101 	{ 0x0, "Pkg%pc7", "", 0, 0, 0, NULL, 0 },
102 	{ 0x0, "Pkg%pc8", "", 0, 0, 0, NULL, 0 },
103 	{ 0x0, "Pkg%pc9", "", 0, 0, 0, NULL, 0 },
104 	{ 0x0, "Pk%pc10", "", 0, 0, 0, NULL, 0 },
105 	{ 0x0, "CPU%LPI", "", 0, 0, 0, NULL, 0 },
106 	{ 0x0, "SYS%LPI", "", 0, 0, 0, NULL, 0 },
107 	{ 0x0, "PkgWatt", "", 0, 0, 0, NULL, 0 },
108 	{ 0x0, "CorWatt", "", 0, 0, 0, NULL, 0 },
109 	{ 0x0, "GFXWatt", "", 0, 0, 0, NULL, 0 },
110 	{ 0x0, "PkgCnt", "", 0, 0, 0, NULL, 0 },
111 	{ 0x0, "RAMWatt", "", 0, 0, 0, NULL, 0 },
112 	{ 0x0, "PKG_%", "", 0, 0, 0, NULL, 0 },
113 	{ 0x0, "RAM_%", "", 0, 0, 0, NULL, 0 },
114 	{ 0x0, "Pkg_J", "", 0, 0, 0, NULL, 0 },
115 	{ 0x0, "Cor_J", "", 0, 0, 0, NULL, 0 },
116 	{ 0x0, "GFX_J", "", 0, 0, 0, NULL, 0 },
117 	{ 0x0, "RAM_J", "", 0, 0, 0, NULL, 0 },
118 	{ 0x0, "Mod%c6", "", 0, 0, 0, NULL, 0 },
119 	{ 0x0, "Totl%C0", "", 0, 0, 0, NULL, 0 },
120 	{ 0x0, "Any%C0", "", 0, 0, 0, NULL, 0 },
121 	{ 0x0, "GFX%C0", "", 0, 0, 0, NULL, 0 },
122 	{ 0x0, "CPUGFX%", "", 0, 0, 0, NULL, 0 },
123 	{ 0x0, "Core", "", 0, 0, 0, NULL, 0 },
124 	{ 0x0, "CPU", "", 0, 0, 0, NULL, 0 },
125 	{ 0x0, "APIC", "", 0, 0, 0, NULL, 0 },
126 	{ 0x0, "X2APIC", "", 0, 0, 0, NULL, 0 },
127 	{ 0x0, "Die", "", 0, 0, 0, NULL, 0 },
128 	{ 0x0, "GFXAMHz", "", 0, 0, 0, NULL, 0 },
129 	{ 0x0, "IPC", "", 0, 0, 0, NULL, 0 },
130 	{ 0x0, "CoreThr", "", 0, 0, 0, NULL, 0 },
131 	{ 0x0, "UncMHz", "", 0, 0, 0, NULL, 0 },
132 };
133 
134 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
135 #define	BIC_USEC	(1ULL << 0)
136 #define	BIC_TOD		(1ULL << 1)
137 #define	BIC_Package	(1ULL << 2)
138 #define	BIC_Node	(1ULL << 3)
139 #define	BIC_Avg_MHz	(1ULL << 4)
140 #define	BIC_Busy	(1ULL << 5)
141 #define	BIC_Bzy_MHz	(1ULL << 6)
142 #define	BIC_TSC_MHz	(1ULL << 7)
143 #define	BIC_IRQ		(1ULL << 8)
144 #define	BIC_SMI		(1ULL << 9)
145 #define	BIC_sysfs	(1ULL << 10)
146 #define	BIC_CPU_c1	(1ULL << 11)
147 #define	BIC_CPU_c3	(1ULL << 12)
148 #define	BIC_CPU_c6	(1ULL << 13)
149 #define	BIC_CPU_c7	(1ULL << 14)
150 #define	BIC_ThreadC	(1ULL << 15)
151 #define	BIC_CoreTmp	(1ULL << 16)
152 #define	BIC_CoreCnt	(1ULL << 17)
153 #define	BIC_PkgTmp	(1ULL << 18)
154 #define	BIC_GFX_rc6	(1ULL << 19)
155 #define	BIC_GFXMHz	(1ULL << 20)
156 #define	BIC_Pkgpc2	(1ULL << 21)
157 #define	BIC_Pkgpc3	(1ULL << 22)
158 #define	BIC_Pkgpc6	(1ULL << 23)
159 #define	BIC_Pkgpc7	(1ULL << 24)
160 #define	BIC_Pkgpc8	(1ULL << 25)
161 #define	BIC_Pkgpc9	(1ULL << 26)
162 #define	BIC_Pkgpc10	(1ULL << 27)
163 #define BIC_CPU_LPI	(1ULL << 28)
164 #define BIC_SYS_LPI	(1ULL << 29)
165 #define	BIC_PkgWatt	(1ULL << 30)
166 #define	BIC_CorWatt	(1ULL << 31)
167 #define	BIC_GFXWatt	(1ULL << 32)
168 #define	BIC_PkgCnt	(1ULL << 33)
169 #define	BIC_RAMWatt	(1ULL << 34)
170 #define	BIC_PKG__	(1ULL << 35)
171 #define	BIC_RAM__	(1ULL << 36)
172 #define	BIC_Pkg_J	(1ULL << 37)
173 #define	BIC_Cor_J	(1ULL << 38)
174 #define	BIC_GFX_J	(1ULL << 39)
175 #define	BIC_RAM_J	(1ULL << 40)
176 #define	BIC_Mod_c6	(1ULL << 41)
177 #define	BIC_Totl_c0	(1ULL << 42)
178 #define	BIC_Any_c0	(1ULL << 43)
179 #define	BIC_GFX_c0	(1ULL << 44)
180 #define	BIC_CPUGFX	(1ULL << 45)
181 #define	BIC_Core	(1ULL << 46)
182 #define	BIC_CPU		(1ULL << 47)
183 #define	BIC_APIC	(1ULL << 48)
184 #define	BIC_X2APIC	(1ULL << 49)
185 #define	BIC_Die		(1ULL << 50)
186 #define	BIC_GFXACTMHz	(1ULL << 51)
187 #define	BIC_IPC		(1ULL << 52)
188 #define	BIC_CORE_THROT_CNT	(1ULL << 53)
189 #define	BIC_UNCORE_MHZ		(1ULL << 54)
190 
191 #define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die )
192 #define BIC_THERMAL_PWR ( BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__)
193 #define BIC_FREQUENCY ( BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_UNCORE_MHZ)
194 #define BIC_IDLE ( BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX)
195 #define BIC_OTHER ( BIC_IRQ | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC)
196 
197 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
198 
199 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT);
200 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC;
201 
202 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME)
203 #define DO_BIC_READ(COUNTER_NAME) (bic_present & COUNTER_NAME)
204 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
205 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
206 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
207 #define BIC_IS_ENABLED(COUNTER_BIT) (bic_enabled & COUNTER_BIT)
208 
209 char *proc_stat = "/proc/stat";
210 FILE *outf;
211 int *fd_percpu;
212 int *fd_instr_count_percpu;
213 struct timeval interval_tv = { 5, 0 };
214 struct timespec interval_ts = { 5, 0 };
215 
216 /* Save original CPU model */
217 unsigned int model_orig;
218 
219 unsigned int num_iterations;
220 unsigned int header_iterations;
221 unsigned int debug;
222 unsigned int quiet;
223 unsigned int shown;
224 unsigned int sums_need_wide_columns;
225 unsigned int rapl_joules;
226 unsigned int summary_only;
227 unsigned int list_header_only;
228 unsigned int dump_only;
229 unsigned int do_snb_cstates;
230 unsigned int do_knl_cstates;
231 unsigned int do_slm_cstates;
232 unsigned int use_c1_residency_msr;
233 unsigned int has_aperf;
234 unsigned int has_epb;
235 unsigned int has_turbo;
236 unsigned int is_hybrid;
237 unsigned int do_irtl_snb;
238 unsigned int do_irtl_hsw;
239 unsigned int units = 1000000;	/* MHz etc */
240 unsigned int genuine_intel;
241 unsigned int authentic_amd;
242 unsigned int hygon_genuine;
243 unsigned int max_level, max_extended_level;
244 unsigned int has_invariant_tsc;
245 unsigned int do_nhm_platform_info;
246 unsigned int no_MSR_MISC_PWR_MGMT;
247 unsigned int aperf_mperf_multiplier = 1;
248 double bclk;
249 double base_hz;
250 unsigned int has_base_hz;
251 double tsc_tweak = 1.0;
252 unsigned int show_pkg_only;
253 unsigned int show_core_only;
254 char *output_buffer, *outp;
255 unsigned int do_rapl;
256 unsigned int do_dts;
257 unsigned int do_ptm;
258 unsigned int do_ipc;
259 unsigned long long gfx_cur_rc6_ms;
260 unsigned long long cpuidle_cur_cpu_lpi_us;
261 unsigned long long cpuidle_cur_sys_lpi_us;
262 unsigned int gfx_cur_mhz;
263 unsigned int gfx_act_mhz;
264 unsigned int tj_max;
265 unsigned int tj_max_override;
266 int tcc_offset_bits;
267 double rapl_power_units, rapl_time_units;
268 double rapl_dram_energy_units, rapl_energy_units;
269 double rapl_joule_counter_range;
270 unsigned int do_core_perf_limit_reasons;
271 unsigned int has_automatic_cstate_conversion;
272 unsigned int dis_cstate_prewake;
273 unsigned int do_gfx_perf_limit_reasons;
274 unsigned int do_ring_perf_limit_reasons;
275 unsigned int crystal_hz;
276 unsigned long long tsc_hz;
277 int base_cpu;
278 double discover_bclk(unsigned int family, unsigned int model);
279 unsigned int has_hwp;		/* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */
280 			/* IA32_HWP_REQUEST, IA32_HWP_STATUS */
281 unsigned int has_hwp_notify;	/* IA32_HWP_INTERRUPT */
282 unsigned int has_hwp_activity_window;	/* IA32_HWP_REQUEST[bits 41:32] */
283 unsigned int has_hwp_epp;	/* IA32_HWP_REQUEST[bits 31:24] */
284 unsigned int has_hwp_pkg;	/* IA32_HWP_REQUEST_PKG */
285 unsigned int has_misc_feature_control;
286 unsigned int first_counter_read = 1;
287 int ignore_stdin;
288 
289 #define RAPL_PKG		(1 << 0)
290 					/* 0x610 MSR_PKG_POWER_LIMIT */
291 					/* 0x611 MSR_PKG_ENERGY_STATUS */
292 #define RAPL_PKG_PERF_STATUS	(1 << 1)
293 					/* 0x613 MSR_PKG_PERF_STATUS */
294 #define RAPL_PKG_POWER_INFO	(1 << 2)
295 					/* 0x614 MSR_PKG_POWER_INFO */
296 
297 #define RAPL_DRAM		(1 << 3)
298 					/* 0x618 MSR_DRAM_POWER_LIMIT */
299 					/* 0x619 MSR_DRAM_ENERGY_STATUS */
300 #define RAPL_DRAM_PERF_STATUS	(1 << 4)
301 					/* 0x61b MSR_DRAM_PERF_STATUS */
302 #define RAPL_DRAM_POWER_INFO	(1 << 5)
303 					/* 0x61c MSR_DRAM_POWER_INFO */
304 
305 #define RAPL_CORES_POWER_LIMIT	(1 << 6)
306 					/* 0x638 MSR_PP0_POWER_LIMIT */
307 #define RAPL_CORE_POLICY	(1 << 7)
308 					/* 0x63a MSR_PP0_POLICY */
309 
310 #define RAPL_GFX		(1 << 8)
311 					/* 0x640 MSR_PP1_POWER_LIMIT */
312 					/* 0x641 MSR_PP1_ENERGY_STATUS */
313 					/* 0x642 MSR_PP1_POLICY */
314 
315 #define RAPL_CORES_ENERGY_STATUS	(1 << 9)
316 					/* 0x639 MSR_PP0_ENERGY_STATUS */
317 #define RAPL_PER_CORE_ENERGY	(1 << 10)
318 					/* Indicates cores energy collection is per-core,
319 					 * not per-package. */
320 #define RAPL_AMD_F17H		(1 << 11)
321 					/* 0xc0010299 MSR_RAPL_PWR_UNIT */
322 					/* 0xc001029a MSR_CORE_ENERGY_STAT */
323 					/* 0xc001029b MSR_PKG_ENERGY_STAT */
324 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT)
325 #define	TJMAX_DEFAULT	100
326 
327 /* MSRs that are not yet in the kernel-provided header. */
328 #define MSR_RAPL_PWR_UNIT	0xc0010299
329 #define MSR_CORE_ENERGY_STAT	0xc001029a
330 #define MSR_PKG_ENERGY_STAT	0xc001029b
331 
332 #define MAX(a, b) ((a) > (b) ? (a) : (b))
333 
334 int backwards_count;
335 char *progname;
336 
337 #define CPU_SUBSET_MAXCPUS	1024	/* need to use before probe... */
338 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset;
339 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size;
340 #define MAX_ADDED_COUNTERS 8
341 #define MAX_ADDED_THREAD_COUNTERS 24
342 #define BITMASK_SIZE 32
343 
344 struct thread_data {
345 	struct timeval tv_begin;
346 	struct timeval tv_end;
347 	struct timeval tv_delta;
348 	unsigned long long tsc;
349 	unsigned long long aperf;
350 	unsigned long long mperf;
351 	unsigned long long c1;
352 	unsigned long long instr_count;
353 	unsigned long long irq_count;
354 	unsigned int smi_count;
355 	unsigned int cpu_id;
356 	unsigned int apic_id;
357 	unsigned int x2apic_id;
358 	unsigned int flags;
359 	bool is_atom;
360 #define CPU_IS_FIRST_THREAD_IN_CORE	0x2
361 #define CPU_IS_FIRST_CORE_IN_PACKAGE	0x4
362 	unsigned long long counter[MAX_ADDED_THREAD_COUNTERS];
363 } *thread_even, *thread_odd;
364 
365 struct core_data {
366 	unsigned long long c3;
367 	unsigned long long c6;
368 	unsigned long long c7;
369 	unsigned long long mc6_us;	/* duplicate as per-core for now, even though per module */
370 	unsigned int core_temp_c;
371 	unsigned int core_energy;	/* MSR_CORE_ENERGY_STAT */
372 	unsigned int core_id;
373 	unsigned long long core_throt_cnt;
374 	unsigned long long counter[MAX_ADDED_COUNTERS];
375 } *core_even, *core_odd;
376 
377 struct pkg_data {
378 	unsigned long long pc2;
379 	unsigned long long pc3;
380 	unsigned long long pc6;
381 	unsigned long long pc7;
382 	unsigned long long pc8;
383 	unsigned long long pc9;
384 	unsigned long long pc10;
385 	unsigned long long cpu_lpi;
386 	unsigned long long sys_lpi;
387 	unsigned long long pkg_wtd_core_c0;
388 	unsigned long long pkg_any_core_c0;
389 	unsigned long long pkg_any_gfxe_c0;
390 	unsigned long long pkg_both_core_gfxe_c0;
391 	long long gfx_rc6_ms;
392 	unsigned int gfx_mhz;
393 	unsigned int gfx_act_mhz;
394 	unsigned int package_id;
395 	unsigned long long energy_pkg;	/* MSR_PKG_ENERGY_STATUS */
396 	unsigned long long energy_dram;	/* MSR_DRAM_ENERGY_STATUS */
397 	unsigned long long energy_cores;	/* MSR_PP0_ENERGY_STATUS */
398 	unsigned long long energy_gfx;	/* MSR_PP1_ENERGY_STATUS */
399 	unsigned long long rapl_pkg_perf_status;	/* MSR_PKG_PERF_STATUS */
400 	unsigned long long rapl_dram_perf_status;	/* MSR_DRAM_PERF_STATUS */
401 	unsigned int pkg_temp_c;
402 	unsigned int uncore_mhz;
403 	unsigned long long counter[MAX_ADDED_COUNTERS];
404 } *package_even, *package_odd;
405 
406 #define ODD_COUNTERS thread_odd, core_odd, package_odd
407 #define EVEN_COUNTERS thread_even, core_even, package_even
408 
409 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no)	      \
410 	((thread_base) +						      \
411 	 ((pkg_no) *							      \
412 	  topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \
413 	 ((node_no) * topo.cores_per_node * topo.threads_per_core) +	      \
414 	 ((core_no) * topo.threads_per_core) +				      \
415 	 (thread_no))
416 
417 #define GET_CORE(core_base, core_no, node_no, pkg_no)			\
418 	((core_base) +							\
419 	 ((pkg_no) *  topo.nodes_per_pkg * topo.cores_per_node) +	\
420 	 ((node_no) * topo.cores_per_node) +				\
421 	 (core_no))
422 
423 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
424 
425 /*
426  * The accumulated sum of MSR is defined as a monotonic
427  * increasing MSR, it will be accumulated periodically,
428  * despite its register's bit width.
429  */
430 enum {
431 	IDX_PKG_ENERGY,
432 	IDX_DRAM_ENERGY,
433 	IDX_PP0_ENERGY,
434 	IDX_PP1_ENERGY,
435 	IDX_PKG_PERF,
436 	IDX_DRAM_PERF,
437 	IDX_COUNT,
438 };
439 
440 int get_msr_sum(int cpu, off_t offset, unsigned long long *msr);
441 
442 struct msr_sum_array {
443 	/* get_msr_sum() = sum + (get_msr() - last) */
444 	struct {
445 		/*The accumulated MSR value is updated by the timer */
446 		unsigned long long sum;
447 		/*The MSR footprint recorded in last timer */
448 		unsigned long long last;
449 	} entries[IDX_COUNT];
450 };
451 
452 /* The percpu MSR sum array.*/
453 struct msr_sum_array *per_cpu_msr_sum;
454 
455 off_t idx_to_offset(int idx)
456 {
457 	off_t offset;
458 
459 	switch (idx) {
460 	case IDX_PKG_ENERGY:
461 		if (do_rapl & RAPL_AMD_F17H)
462 			offset = MSR_PKG_ENERGY_STAT;
463 		else
464 			offset = MSR_PKG_ENERGY_STATUS;
465 		break;
466 	case IDX_DRAM_ENERGY:
467 		offset = MSR_DRAM_ENERGY_STATUS;
468 		break;
469 	case IDX_PP0_ENERGY:
470 		offset = MSR_PP0_ENERGY_STATUS;
471 		break;
472 	case IDX_PP1_ENERGY:
473 		offset = MSR_PP1_ENERGY_STATUS;
474 		break;
475 	case IDX_PKG_PERF:
476 		offset = MSR_PKG_PERF_STATUS;
477 		break;
478 	case IDX_DRAM_PERF:
479 		offset = MSR_DRAM_PERF_STATUS;
480 		break;
481 	default:
482 		offset = -1;
483 	}
484 	return offset;
485 }
486 
487 int offset_to_idx(off_t offset)
488 {
489 	int idx;
490 
491 	switch (offset) {
492 	case MSR_PKG_ENERGY_STATUS:
493 	case MSR_PKG_ENERGY_STAT:
494 		idx = IDX_PKG_ENERGY;
495 		break;
496 	case MSR_DRAM_ENERGY_STATUS:
497 		idx = IDX_DRAM_ENERGY;
498 		break;
499 	case MSR_PP0_ENERGY_STATUS:
500 		idx = IDX_PP0_ENERGY;
501 		break;
502 	case MSR_PP1_ENERGY_STATUS:
503 		idx = IDX_PP1_ENERGY;
504 		break;
505 	case MSR_PKG_PERF_STATUS:
506 		idx = IDX_PKG_PERF;
507 		break;
508 	case MSR_DRAM_PERF_STATUS:
509 		idx = IDX_DRAM_PERF;
510 		break;
511 	default:
512 		idx = -1;
513 	}
514 	return idx;
515 }
516 
517 int idx_valid(int idx)
518 {
519 	switch (idx) {
520 	case IDX_PKG_ENERGY:
521 		return do_rapl & (RAPL_PKG | RAPL_AMD_F17H);
522 	case IDX_DRAM_ENERGY:
523 		return do_rapl & RAPL_DRAM;
524 	case IDX_PP0_ENERGY:
525 		return do_rapl & RAPL_CORES_ENERGY_STATUS;
526 	case IDX_PP1_ENERGY:
527 		return do_rapl & RAPL_GFX;
528 	case IDX_PKG_PERF:
529 		return do_rapl & RAPL_PKG_PERF_STATUS;
530 	case IDX_DRAM_PERF:
531 		return do_rapl & RAPL_DRAM_PERF_STATUS;
532 	default:
533 		return 0;
534 	}
535 }
536 
537 struct sys_counters {
538 	unsigned int added_thread_counters;
539 	unsigned int added_core_counters;
540 	unsigned int added_package_counters;
541 	struct msr_counter *tp;
542 	struct msr_counter *cp;
543 	struct msr_counter *pp;
544 } sys;
545 
546 struct system_summary {
547 	struct thread_data threads;
548 	struct core_data cores;
549 	struct pkg_data packages;
550 } average;
551 
552 struct cpu_topology {
553 	int physical_package_id;
554 	int die_id;
555 	int logical_cpu_id;
556 	int physical_node_id;
557 	int logical_node_id;	/* 0-based count within the package */
558 	int physical_core_id;
559 	int thread_id;
560 	cpu_set_t *put_ids;	/* Processing Unit/Thread IDs */
561 } *cpus;
562 
563 struct topo_params {
564 	int num_packages;
565 	int num_die;
566 	int num_cpus;
567 	int num_cores;
568 	int max_cpu_num;
569 	int max_node_num;
570 	int nodes_per_pkg;
571 	int cores_per_node;
572 	int threads_per_core;
573 } topo;
574 
575 struct timeval tv_even, tv_odd, tv_delta;
576 
577 int *irq_column_2_cpu;		/* /proc/interrupts column numbers */
578 int *irqs_per_cpu;		/* indexed by cpu_num */
579 
580 void setup_all_buffers(void);
581 
582 char *sys_lpi_file;
583 char *sys_lpi_file_sysfs = "/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us";
584 char *sys_lpi_file_debugfs = "/sys/kernel/debug/pmc_core/slp_s0_residency_usec";
585 
586 int cpu_is_not_present(int cpu)
587 {
588 	return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
589 }
590 
591 /*
592  * run func(thread, core, package) in topology order
593  * skip non-present cpus
594  */
595 
596 int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pkg_data *),
597 		 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
598 {
599 	int retval, pkg_no, core_no, thread_no, node_no;
600 
601 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
602 		for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) {
603 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
604 				for (thread_no = 0; thread_no < topo.threads_per_core; ++thread_no) {
605 					struct thread_data *t;
606 					struct core_data *c;
607 					struct pkg_data *p;
608 
609 					t = GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no);
610 
611 					if (cpu_is_not_present(t->cpu_id))
612 						continue;
613 
614 					c = GET_CORE(core_base, core_no, node_no, pkg_no);
615 					p = GET_PKG(pkg_base, pkg_no);
616 
617 					retval = func(t, c, p);
618 					if (retval)
619 						return retval;
620 				}
621 			}
622 		}
623 	}
624 	return 0;
625 }
626 
627 int cpu_migrate(int cpu)
628 {
629 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
630 	CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
631 	if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
632 		return -1;
633 	else
634 		return 0;
635 }
636 
637 int get_msr_fd(int cpu)
638 {
639 	char pathname[32];
640 	int fd;
641 
642 	fd = fd_percpu[cpu];
643 
644 	if (fd)
645 		return fd;
646 
647 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
648 	fd = open(pathname, O_RDONLY);
649 	if (fd < 0)
650 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname);
651 
652 	fd_percpu[cpu] = fd;
653 
654 	return fd;
655 }
656 
657 static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags)
658 {
659 	return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags);
660 }
661 
662 static int perf_instr_count_open(int cpu_num)
663 {
664 	struct perf_event_attr pea;
665 	int fd;
666 
667 	memset(&pea, 0, sizeof(struct perf_event_attr));
668 	pea.type = PERF_TYPE_HARDWARE;
669 	pea.size = sizeof(struct perf_event_attr);
670 	pea.config = PERF_COUNT_HW_INSTRUCTIONS;
671 
672 	/* counter for cpu_num, including user + kernel and all processes */
673 	fd = perf_event_open(&pea, -1, cpu_num, -1, 0);
674 	if (fd == -1) {
675 		warnx("capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\"", progname);
676 		BIC_NOT_PRESENT(BIC_IPC);
677 	}
678 
679 	return fd;
680 }
681 
682 int get_instr_count_fd(int cpu)
683 {
684 	if (fd_instr_count_percpu[cpu])
685 		return fd_instr_count_percpu[cpu];
686 
687 	fd_instr_count_percpu[cpu] = perf_instr_count_open(cpu);
688 
689 	return fd_instr_count_percpu[cpu];
690 }
691 
692 int get_msr(int cpu, off_t offset, unsigned long long *msr)
693 {
694 	ssize_t retval;
695 
696 	retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset);
697 
698 	if (retval != sizeof *msr)
699 		err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset);
700 
701 	return 0;
702 }
703 
704 #define MAX_DEFERRED 16
705 char *deferred_add_names[MAX_DEFERRED];
706 char *deferred_skip_names[MAX_DEFERRED];
707 int deferred_add_index;
708 int deferred_skip_index;
709 
710 /*
711  * HIDE_LIST - hide this list of counters, show the rest [default]
712  * SHOW_LIST - show this list of counters, hide the rest
713  */
714 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST;
715 
716 void help(void)
717 {
718 	fprintf(outf,
719 		"Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n"
720 		"\n"
721 		"Turbostat forks the specified COMMAND and prints statistics\n"
722 		"when COMMAND completes.\n"
723 		"If no COMMAND is specified, turbostat wakes every 5-seconds\n"
724 		"to print statistics, until interrupted.\n"
725 		"  -a, --add	add a counter\n"
726 		"		  eg. --add msr0x10,u64,cpu,delta,MY_TSC\n"
727 		"  -c, --cpu	cpu-set	limit output to summary plus cpu-set:\n"
728 		"		  {core | package | j,k,l..m,n-p }\n"
729 		"  -d, --debug	displays usec, Time_Of_Day_Seconds and more debugging\n"
730 		"  -D, --Dump	displays the raw counter values\n"
731 		"  -e, --enable	[all | column]\n"
732 		"		shows all or the specified disabled column\n"
733 		"  -H, --hide [column|column,column,...]\n"
734 		"		hide the specified column(s)\n"
735 		"  -i, --interval sec.subsec\n"
736 		"		Override default 5-second measurement interval\n"
737 		"  -J, --Joules	displays energy in Joules instead of Watts\n"
738 		"  -l, --list	list column headers only\n"
739 		"  -n, --num_iterations num\n"
740 		"		number of the measurement iterations\n"
741 		"  -N, --header_iterations num\n"
742 		"		print header every num iterations\n"
743 		"  -o, --out file\n"
744 		"		create or truncate \"file\" for all output\n"
745 		"  -q, --quiet	skip decoding system configuration header\n"
746 		"  -s, --show [column|column,column,...]\n"
747 		"		show only the specified column(s)\n"
748 		"  -S, --Summary\n"
749 		"		limits output to 1-line system summary per interval\n"
750 		"  -T, --TCC temperature\n"
751 		"		sets the Thermal Control Circuit temperature in\n"
752 		"		  degrees Celsius\n"
753 		"  -h, --help	print this help message\n"
754 		"  -v, --version	print version information\n" "\n" "For more help, run \"man turbostat\"\n");
755 }
756 
757 /*
758  * bic_lookup
759  * for all the strings in comma separate name_list,
760  * set the approprate bit in return value.
761  */
762 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode)
763 {
764 	unsigned int i;
765 	unsigned long long retval = 0;
766 
767 	while (name_list) {
768 		char *comma;
769 
770 		comma = strchr(name_list, ',');
771 
772 		if (comma)
773 			*comma = '\0';
774 
775 		for (i = 0; i < MAX_BIC; ++i) {
776 			if (!strcmp(name_list, bic[i].name)) {
777 				retval |= (1ULL << i);
778 				break;
779 			}
780 			if (!strcmp(name_list, "all")) {
781 				retval |= ~0;
782 				break;
783 			} else if (!strcmp(name_list, "topology")) {
784 				retval |= BIC_TOPOLOGY;
785 				break;
786 			} else if (!strcmp(name_list, "power")) {
787 				retval |= BIC_THERMAL_PWR;
788 				break;
789 			} else if (!strcmp(name_list, "idle")) {
790 				retval |= BIC_IDLE;
791 				break;
792 			} else if (!strcmp(name_list, "frequency")) {
793 				retval |= BIC_FREQUENCY;
794 				break;
795 			} else if (!strcmp(name_list, "other")) {
796 				retval |= BIC_OTHER;
797 				break;
798 			}
799 
800 		}
801 		if (i == MAX_BIC) {
802 			if (mode == SHOW_LIST) {
803 				deferred_add_names[deferred_add_index++] = name_list;
804 				if (deferred_add_index >= MAX_DEFERRED) {
805 					fprintf(stderr, "More than max %d un-recognized --add options '%s'\n",
806 						MAX_DEFERRED, name_list);
807 					help();
808 					exit(1);
809 				}
810 			} else {
811 				deferred_skip_names[deferred_skip_index++] = name_list;
812 				if (debug)
813 					fprintf(stderr, "deferred \"%s\"\n", name_list);
814 				if (deferred_skip_index >= MAX_DEFERRED) {
815 					fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n",
816 						MAX_DEFERRED, name_list);
817 					help();
818 					exit(1);
819 				}
820 			}
821 		}
822 
823 		name_list = comma;
824 		if (name_list)
825 			name_list++;
826 
827 	}
828 	return retval;
829 }
830 
831 void print_header(char *delim)
832 {
833 	struct msr_counter *mp;
834 	int printed = 0;
835 
836 	if (DO_BIC(BIC_USEC))
837 		outp += sprintf(outp, "%susec", (printed++ ? delim : ""));
838 	if (DO_BIC(BIC_TOD))
839 		outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : ""));
840 	if (DO_BIC(BIC_Package))
841 		outp += sprintf(outp, "%sPackage", (printed++ ? delim : ""));
842 	if (DO_BIC(BIC_Die))
843 		outp += sprintf(outp, "%sDie", (printed++ ? delim : ""));
844 	if (DO_BIC(BIC_Node))
845 		outp += sprintf(outp, "%sNode", (printed++ ? delim : ""));
846 	if (DO_BIC(BIC_Core))
847 		outp += sprintf(outp, "%sCore", (printed++ ? delim : ""));
848 	if (DO_BIC(BIC_CPU))
849 		outp += sprintf(outp, "%sCPU", (printed++ ? delim : ""));
850 	if (DO_BIC(BIC_APIC))
851 		outp += sprintf(outp, "%sAPIC", (printed++ ? delim : ""));
852 	if (DO_BIC(BIC_X2APIC))
853 		outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : ""));
854 	if (DO_BIC(BIC_Avg_MHz))
855 		outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : ""));
856 	if (DO_BIC(BIC_Busy))
857 		outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : ""));
858 	if (DO_BIC(BIC_Bzy_MHz))
859 		outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : ""));
860 	if (DO_BIC(BIC_TSC_MHz))
861 		outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
862 
863 	if (DO_BIC(BIC_IPC))
864 		outp += sprintf(outp, "%sIPC", (printed++ ? delim : ""));
865 
866 	if (DO_BIC(BIC_IRQ)) {
867 		if (sums_need_wide_columns)
868 			outp += sprintf(outp, "%s     IRQ", (printed++ ? delim : ""));
869 		else
870 			outp += sprintf(outp, "%sIRQ", (printed++ ? delim : ""));
871 	}
872 
873 	if (DO_BIC(BIC_SMI))
874 		outp += sprintf(outp, "%sSMI", (printed++ ? delim : ""));
875 
876 	for (mp = sys.tp; mp; mp = mp->next) {
877 
878 		if (mp->format == FORMAT_RAW) {
879 			if (mp->width == 64)
880 				outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name);
881 			else
882 				outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name);
883 		} else {
884 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
885 				outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name);
886 			else
887 				outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name);
888 		}
889 	}
890 
891 	if (DO_BIC(BIC_CPU_c1))
892 		outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : ""));
893 	if (DO_BIC(BIC_CPU_c3))
894 		outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : ""));
895 	if (DO_BIC(BIC_CPU_c6))
896 		outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : ""));
897 	if (DO_BIC(BIC_CPU_c7))
898 		outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : ""));
899 
900 	if (DO_BIC(BIC_Mod_c6))
901 		outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : ""));
902 
903 	if (DO_BIC(BIC_CoreTmp))
904 		outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : ""));
905 
906 	if (DO_BIC(BIC_CORE_THROT_CNT))
907 		outp += sprintf(outp, "%sCoreThr", (printed++ ? delim : ""));
908 
909 	if (do_rapl && !rapl_joules) {
910 		if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
911 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
912 	} else if (do_rapl && rapl_joules) {
913 		if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
914 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
915 	}
916 
917 	for (mp = sys.cp; mp; mp = mp->next) {
918 		if (mp->format == FORMAT_RAW) {
919 			if (mp->width == 64)
920 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
921 			else
922 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
923 		} else {
924 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
925 				outp += sprintf(outp, "%s%8s", delim, mp->name);
926 			else
927 				outp += sprintf(outp, "%s%s", delim, mp->name);
928 		}
929 	}
930 
931 	if (DO_BIC(BIC_PkgTmp))
932 		outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : ""));
933 
934 	if (DO_BIC(BIC_GFX_rc6))
935 		outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : ""));
936 
937 	if (DO_BIC(BIC_GFXMHz))
938 		outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : ""));
939 
940 	if (DO_BIC(BIC_GFXACTMHz))
941 		outp += sprintf(outp, "%sGFXAMHz", (printed++ ? delim : ""));
942 
943 	if (DO_BIC(BIC_Totl_c0))
944 		outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : ""));
945 	if (DO_BIC(BIC_Any_c0))
946 		outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : ""));
947 	if (DO_BIC(BIC_GFX_c0))
948 		outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : ""));
949 	if (DO_BIC(BIC_CPUGFX))
950 		outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : ""));
951 
952 	if (DO_BIC(BIC_Pkgpc2))
953 		outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : ""));
954 	if (DO_BIC(BIC_Pkgpc3))
955 		outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : ""));
956 	if (DO_BIC(BIC_Pkgpc6))
957 		outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : ""));
958 	if (DO_BIC(BIC_Pkgpc7))
959 		outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : ""));
960 	if (DO_BIC(BIC_Pkgpc8))
961 		outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : ""));
962 	if (DO_BIC(BIC_Pkgpc9))
963 		outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : ""));
964 	if (DO_BIC(BIC_Pkgpc10))
965 		outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : ""));
966 	if (DO_BIC(BIC_CPU_LPI))
967 		outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : ""));
968 	if (DO_BIC(BIC_SYS_LPI))
969 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
970 
971 	if (do_rapl && !rapl_joules) {
972 		if (DO_BIC(BIC_PkgWatt))
973 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
974 		if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
975 			outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : ""));
976 		if (DO_BIC(BIC_GFXWatt))
977 			outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : ""));
978 		if (DO_BIC(BIC_RAMWatt))
979 			outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : ""));
980 		if (DO_BIC(BIC_PKG__))
981 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
982 		if (DO_BIC(BIC_RAM__))
983 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
984 	} else if (do_rapl && rapl_joules) {
985 		if (DO_BIC(BIC_Pkg_J))
986 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
987 		if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
988 			outp += sprintf(outp, "%sCor_J", (printed++ ? delim : ""));
989 		if (DO_BIC(BIC_GFX_J))
990 			outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : ""));
991 		if (DO_BIC(BIC_RAM_J))
992 			outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : ""));
993 		if (DO_BIC(BIC_PKG__))
994 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
995 		if (DO_BIC(BIC_RAM__))
996 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
997 	}
998 	if (DO_BIC(BIC_UNCORE_MHZ))
999 		outp += sprintf(outp, "%sUncMHz", (printed++ ? delim : ""));
1000 
1001 	for (mp = sys.pp; mp; mp = mp->next) {
1002 		if (mp->format == FORMAT_RAW) {
1003 			if (mp->width == 64)
1004 				outp += sprintf(outp, "%s%18.18s", delim, mp->name);
1005 			else
1006 				outp += sprintf(outp, "%s%10.10s", delim, mp->name);
1007 		} else {
1008 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1009 				outp += sprintf(outp, "%s%8s", delim, mp->name);
1010 			else
1011 				outp += sprintf(outp, "%s%s", delim, mp->name);
1012 		}
1013 	}
1014 
1015 	outp += sprintf(outp, "\n");
1016 }
1017 
1018 int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1019 {
1020 	int i;
1021 	struct msr_counter *mp;
1022 
1023 	outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p);
1024 
1025 	if (t) {
1026 		outp += sprintf(outp, "CPU: %d flags 0x%x\n", t->cpu_id, t->flags);
1027 		outp += sprintf(outp, "TSC: %016llX\n", t->tsc);
1028 		outp += sprintf(outp, "aperf: %016llX\n", t->aperf);
1029 		outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
1030 		outp += sprintf(outp, "c1: %016llX\n", t->c1);
1031 
1032 		if (DO_BIC(BIC_IPC))
1033 			outp += sprintf(outp, "IPC: %lld\n", t->instr_count);
1034 
1035 		if (DO_BIC(BIC_IRQ))
1036 			outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
1037 		if (DO_BIC(BIC_SMI))
1038 			outp += sprintf(outp, "SMI: %d\n", t->smi_count);
1039 
1040 		for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1041 			outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, t->counter[i]);
1042 		}
1043 	}
1044 
1045 	if (c) {
1046 		outp += sprintf(outp, "core: %d\n", c->core_id);
1047 		outp += sprintf(outp, "c3: %016llX\n", c->c3);
1048 		outp += sprintf(outp, "c6: %016llX\n", c->c6);
1049 		outp += sprintf(outp, "c7: %016llX\n", c->c7);
1050 		outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c);
1051 		outp += sprintf(outp, "cpu_throt_count: %016llX\n", c->core_throt_cnt);
1052 		outp += sprintf(outp, "Joules: %0X\n", c->core_energy);
1053 
1054 		for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1055 			outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, c->counter[i]);
1056 		}
1057 		outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us);
1058 	}
1059 
1060 	if (p) {
1061 		outp += sprintf(outp, "package: %d\n", p->package_id);
1062 
1063 		outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0);
1064 		outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0);
1065 		outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0);
1066 		outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0);
1067 
1068 		outp += sprintf(outp, "pc2: %016llX\n", p->pc2);
1069 		if (DO_BIC(BIC_Pkgpc3))
1070 			outp += sprintf(outp, "pc3: %016llX\n", p->pc3);
1071 		if (DO_BIC(BIC_Pkgpc6))
1072 			outp += sprintf(outp, "pc6: %016llX\n", p->pc6);
1073 		if (DO_BIC(BIC_Pkgpc7))
1074 			outp += sprintf(outp, "pc7: %016llX\n", p->pc7);
1075 		outp += sprintf(outp, "pc8: %016llX\n", p->pc8);
1076 		outp += sprintf(outp, "pc9: %016llX\n", p->pc9);
1077 		outp += sprintf(outp, "pc10: %016llX\n", p->pc10);
1078 		outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi);
1079 		outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi);
1080 		outp += sprintf(outp, "Joules PKG: %0llX\n", p->energy_pkg);
1081 		outp += sprintf(outp, "Joules COR: %0llX\n", p->energy_cores);
1082 		outp += sprintf(outp, "Joules GFX: %0llX\n", p->energy_gfx);
1083 		outp += sprintf(outp, "Joules RAM: %0llX\n", p->energy_dram);
1084 		outp += sprintf(outp, "Throttle PKG: %0llX\n", p->rapl_pkg_perf_status);
1085 		outp += sprintf(outp, "Throttle RAM: %0llX\n", p->rapl_dram_perf_status);
1086 		outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c);
1087 
1088 		for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1089 			outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, p->counter[i]);
1090 		}
1091 	}
1092 
1093 	outp += sprintf(outp, "\n");
1094 
1095 	return 0;
1096 }
1097 
1098 /*
1099  * column formatting convention & formats
1100  */
1101 int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1102 {
1103 	double interval_float, tsc;
1104 	char *fmt8;
1105 	int i;
1106 	struct msr_counter *mp;
1107 	char *delim = "\t";
1108 	int printed = 0;
1109 
1110 	/* if showing only 1st thread in core and this isn't one, bail out */
1111 	if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1112 		return 0;
1113 
1114 	/* if showing only 1st thread in pkg and this isn't one, bail out */
1115 	if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1116 		return 0;
1117 
1118 	/*if not summary line and --cpu is used */
1119 	if ((t != &average.threads) && (cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset)))
1120 		return 0;
1121 
1122 	if (DO_BIC(BIC_USEC)) {
1123 		/* on each row, print how many usec each timestamp took to gather */
1124 		struct timeval tv;
1125 
1126 		timersub(&t->tv_end, &t->tv_begin, &tv);
1127 		outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec);
1128 	}
1129 
1130 	/* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */
1131 	if (DO_BIC(BIC_TOD))
1132 		outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec);
1133 
1134 	interval_float = t->tv_delta.tv_sec + t->tv_delta.tv_usec / 1000000.0;
1135 
1136 	tsc = t->tsc * tsc_tweak;
1137 
1138 	/* topo columns, print blanks on 1st (average) line */
1139 	if (t == &average.threads) {
1140 		if (DO_BIC(BIC_Package))
1141 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1142 		if (DO_BIC(BIC_Die))
1143 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1144 		if (DO_BIC(BIC_Node))
1145 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1146 		if (DO_BIC(BIC_Core))
1147 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1148 		if (DO_BIC(BIC_CPU))
1149 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1150 		if (DO_BIC(BIC_APIC))
1151 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1152 		if (DO_BIC(BIC_X2APIC))
1153 			outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1154 	} else {
1155 		if (DO_BIC(BIC_Package)) {
1156 			if (p)
1157 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id);
1158 			else
1159 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1160 		}
1161 		if (DO_BIC(BIC_Die)) {
1162 			if (c)
1163 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id);
1164 			else
1165 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1166 		}
1167 		if (DO_BIC(BIC_Node)) {
1168 			if (t)
1169 				outp += sprintf(outp, "%s%d",
1170 						(printed++ ? delim : ""), cpus[t->cpu_id].physical_node_id);
1171 			else
1172 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1173 		}
1174 		if (DO_BIC(BIC_Core)) {
1175 			if (c)
1176 				outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id);
1177 			else
1178 				outp += sprintf(outp, "%s-", (printed++ ? delim : ""));
1179 		}
1180 		if (DO_BIC(BIC_CPU))
1181 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id);
1182 		if (DO_BIC(BIC_APIC))
1183 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id);
1184 		if (DO_BIC(BIC_X2APIC))
1185 			outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id);
1186 	}
1187 
1188 	if (DO_BIC(BIC_Avg_MHz))
1189 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 / units * t->aperf / interval_float);
1190 
1191 	if (DO_BIC(BIC_Busy))
1192 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf / tsc);
1193 
1194 	if (DO_BIC(BIC_Bzy_MHz)) {
1195 		if (has_base_hz)
1196 			outp +=
1197 			    sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf);
1198 		else
1199 			outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""),
1200 					tsc / units * t->aperf / t->mperf / interval_float);
1201 	}
1202 
1203 	if (DO_BIC(BIC_TSC_MHz))
1204 		outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc / units / interval_float);
1205 
1206 	if (DO_BIC(BIC_IPC))
1207 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 1.0 * t->instr_count / t->aperf);
1208 
1209 	/* IRQ */
1210 	if (DO_BIC(BIC_IRQ)) {
1211 		if (sums_need_wide_columns)
1212 			outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count);
1213 		else
1214 			outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count);
1215 	}
1216 
1217 	/* SMI */
1218 	if (DO_BIC(BIC_SMI))
1219 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count);
1220 
1221 	/* Added counters */
1222 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1223 		if (mp->format == FORMAT_RAW) {
1224 			if (mp->width == 32)
1225 				outp +=
1226 				    sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)t->counter[i]);
1227 			else
1228 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]);
1229 		} else if (mp->format == FORMAT_DELTA) {
1230 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1231 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]);
1232 			else
1233 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]);
1234 		} else if (mp->format == FORMAT_PERCENT) {
1235 			if (mp->type == COUNTER_USEC)
1236 				outp +=
1237 				    sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1238 					    t->counter[i] / interval_float / 10000);
1239 			else
1240 				outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i] / tsc);
1241 		}
1242 	}
1243 
1244 	/* C1 */
1245 	if (DO_BIC(BIC_CPU_c1))
1246 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1 / tsc);
1247 
1248 	/* print per-core data only for 1st thread in core */
1249 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1250 		goto done;
1251 
1252 	if (DO_BIC(BIC_CPU_c3))
1253 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3 / tsc);
1254 	if (DO_BIC(BIC_CPU_c6))
1255 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6 / tsc);
1256 	if (DO_BIC(BIC_CPU_c7))
1257 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7 / tsc);
1258 
1259 	/* Mod%c6 */
1260 	if (DO_BIC(BIC_Mod_c6))
1261 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc);
1262 
1263 	if (DO_BIC(BIC_CoreTmp))
1264 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c);
1265 
1266 	/* Core throttle count */
1267 	if (DO_BIC(BIC_CORE_THROT_CNT))
1268 		outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->core_throt_cnt);
1269 
1270 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1271 		if (mp->format == FORMAT_RAW) {
1272 			if (mp->width == 32)
1273 				outp +=
1274 				    sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)c->counter[i]);
1275 			else
1276 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]);
1277 		} else if (mp->format == FORMAT_DELTA) {
1278 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1279 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]);
1280 			else
1281 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]);
1282 		} else if (mp->format == FORMAT_PERCENT) {
1283 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i] / tsc);
1284 		}
1285 	}
1286 
1287 	fmt8 = "%s%.2f";
1288 
1289 	if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY))
1290 		outp +=
1291 		    sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float);
1292 	if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY))
1293 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units);
1294 
1295 	/* print per-package data only for 1st core in package */
1296 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1297 		goto done;
1298 
1299 	/* PkgTmp */
1300 	if (DO_BIC(BIC_PkgTmp))
1301 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c);
1302 
1303 	/* GFXrc6 */
1304 	if (DO_BIC(BIC_GFX_rc6)) {
1305 		if (p->gfx_rc6_ms == -1) {	/* detect GFX counter reset */
1306 			outp += sprintf(outp, "%s**.**", (printed++ ? delim : ""));
1307 		} else {
1308 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""),
1309 					p->gfx_rc6_ms / 10.0 / interval_float);
1310 		}
1311 	}
1312 
1313 	/* GFXMHz */
1314 	if (DO_BIC(BIC_GFXMHz))
1315 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz);
1316 
1317 	/* GFXACTMHz */
1318 	if (DO_BIC(BIC_GFXACTMHz))
1319 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_act_mhz);
1320 
1321 	/* Totl%C0, Any%C0 GFX%C0 CPUGFX% */
1322 	if (DO_BIC(BIC_Totl_c0))
1323 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0 / tsc);
1324 	if (DO_BIC(BIC_Any_c0))
1325 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0 / tsc);
1326 	if (DO_BIC(BIC_GFX_c0))
1327 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0 / tsc);
1328 	if (DO_BIC(BIC_CPUGFX))
1329 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0 / tsc);
1330 
1331 	if (DO_BIC(BIC_Pkgpc2))
1332 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2 / tsc);
1333 	if (DO_BIC(BIC_Pkgpc3))
1334 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3 / tsc);
1335 	if (DO_BIC(BIC_Pkgpc6))
1336 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6 / tsc);
1337 	if (DO_BIC(BIC_Pkgpc7))
1338 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7 / tsc);
1339 	if (DO_BIC(BIC_Pkgpc8))
1340 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8 / tsc);
1341 	if (DO_BIC(BIC_Pkgpc9))
1342 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9 / tsc);
1343 	if (DO_BIC(BIC_Pkgpc10))
1344 		outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10 / tsc);
1345 
1346 	if (DO_BIC(BIC_CPU_LPI))
1347 		outp +=
1348 		    sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float);
1349 	if (DO_BIC(BIC_SYS_LPI))
1350 		outp +=
1351 		    sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float);
1352 
1353 	if (DO_BIC(BIC_PkgWatt))
1354 		outp +=
1355 		    sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float);
1356 
1357 	if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1358 		outp +=
1359 		    sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float);
1360 	if (DO_BIC(BIC_GFXWatt))
1361 		outp +=
1362 		    sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float);
1363 	if (DO_BIC(BIC_RAMWatt))
1364 		outp +=
1365 		    sprintf(outp, fmt8, (printed++ ? delim : ""),
1366 			    p->energy_dram * rapl_dram_energy_units / interval_float);
1367 	if (DO_BIC(BIC_Pkg_J))
1368 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units);
1369 	if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY))
1370 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units);
1371 	if (DO_BIC(BIC_GFX_J))
1372 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units);
1373 	if (DO_BIC(BIC_RAM_J))
1374 		outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units);
1375 	if (DO_BIC(BIC_PKG__))
1376 		outp +=
1377 		    sprintf(outp, fmt8, (printed++ ? delim : ""),
1378 			    100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float);
1379 	if (DO_BIC(BIC_RAM__))
1380 		outp +=
1381 		    sprintf(outp, fmt8, (printed++ ? delim : ""),
1382 			    100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float);
1383 	/* UncMHz */
1384 	if (DO_BIC(BIC_UNCORE_MHZ))
1385 		outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->uncore_mhz);
1386 
1387 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1388 		if (mp->format == FORMAT_RAW) {
1389 			if (mp->width == 32)
1390 				outp +=
1391 				    sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)p->counter[i]);
1392 			else
1393 				outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]);
1394 		} else if (mp->format == FORMAT_DELTA) {
1395 			if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns)
1396 				outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]);
1397 			else
1398 				outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]);
1399 		} else if (mp->format == FORMAT_PERCENT) {
1400 			outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i] / tsc);
1401 		}
1402 	}
1403 
1404 done:
1405 	if (*(outp - 1) != '\n')
1406 		outp += sprintf(outp, "\n");
1407 
1408 	return 0;
1409 }
1410 
1411 void flush_output_stdout(void)
1412 {
1413 	FILE *filep;
1414 
1415 	if (outf == stderr)
1416 		filep = stdout;
1417 	else
1418 		filep = outf;
1419 
1420 	fputs(output_buffer, filep);
1421 	fflush(filep);
1422 
1423 	outp = output_buffer;
1424 }
1425 
1426 void flush_output_stderr(void)
1427 {
1428 	fputs(output_buffer, outf);
1429 	fflush(outf);
1430 	outp = output_buffer;
1431 }
1432 
1433 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1434 {
1435 	static int count;
1436 
1437 	if ((!count || (header_iterations && !(count % header_iterations))) || !summary_only)
1438 		print_header("\t");
1439 
1440 	format_counters(&average.threads, &average.cores, &average.packages);
1441 
1442 	count++;
1443 
1444 	if (summary_only)
1445 		return;
1446 
1447 	for_all_cpus(format_counters, t, c, p);
1448 }
1449 
1450 #define DELTA_WRAP32(new, old)			\
1451 	old = ((((unsigned long long)new << 32) - ((unsigned long long)old << 32)) >> 32);
1452 
1453 int delta_package(struct pkg_data *new, struct pkg_data *old)
1454 {
1455 	int i;
1456 	struct msr_counter *mp;
1457 
1458 	if (DO_BIC(BIC_Totl_c0))
1459 		old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0;
1460 	if (DO_BIC(BIC_Any_c0))
1461 		old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0;
1462 	if (DO_BIC(BIC_GFX_c0))
1463 		old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0;
1464 	if (DO_BIC(BIC_CPUGFX))
1465 		old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0;
1466 
1467 	old->pc2 = new->pc2 - old->pc2;
1468 	if (DO_BIC(BIC_Pkgpc3))
1469 		old->pc3 = new->pc3 - old->pc3;
1470 	if (DO_BIC(BIC_Pkgpc6))
1471 		old->pc6 = new->pc6 - old->pc6;
1472 	if (DO_BIC(BIC_Pkgpc7))
1473 		old->pc7 = new->pc7 - old->pc7;
1474 	old->pc8 = new->pc8 - old->pc8;
1475 	old->pc9 = new->pc9 - old->pc9;
1476 	old->pc10 = new->pc10 - old->pc10;
1477 	old->cpu_lpi = new->cpu_lpi - old->cpu_lpi;
1478 	old->sys_lpi = new->sys_lpi - old->sys_lpi;
1479 	old->pkg_temp_c = new->pkg_temp_c;
1480 
1481 	/* flag an error when rc6 counter resets/wraps */
1482 	if (old->gfx_rc6_ms > new->gfx_rc6_ms)
1483 		old->gfx_rc6_ms = -1;
1484 	else
1485 		old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms;
1486 
1487 	old->uncore_mhz = new->uncore_mhz;
1488 	old->gfx_mhz = new->gfx_mhz;
1489 	old->gfx_act_mhz = new->gfx_act_mhz;
1490 
1491 	old->energy_pkg = new->energy_pkg - old->energy_pkg;
1492 	old->energy_cores = new->energy_cores - old->energy_cores;
1493 	old->energy_gfx = new->energy_gfx - old->energy_gfx;
1494 	old->energy_dram = new->energy_dram - old->energy_dram;
1495 	old->rapl_pkg_perf_status = new->rapl_pkg_perf_status - old->rapl_pkg_perf_status;
1496 	old->rapl_dram_perf_status = new->rapl_dram_perf_status - old->rapl_dram_perf_status;
1497 
1498 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1499 		if (mp->format == FORMAT_RAW)
1500 			old->counter[i] = new->counter[i];
1501 		else
1502 			old->counter[i] = new->counter[i] - old->counter[i];
1503 	}
1504 
1505 	return 0;
1506 }
1507 
1508 void delta_core(struct core_data *new, struct core_data *old)
1509 {
1510 	int i;
1511 	struct msr_counter *mp;
1512 
1513 	old->c3 = new->c3 - old->c3;
1514 	old->c6 = new->c6 - old->c6;
1515 	old->c7 = new->c7 - old->c7;
1516 	old->core_temp_c = new->core_temp_c;
1517 	old->core_throt_cnt = new->core_throt_cnt;
1518 	old->mc6_us = new->mc6_us - old->mc6_us;
1519 
1520 	DELTA_WRAP32(new->core_energy, old->core_energy);
1521 
1522 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1523 		if (mp->format == FORMAT_RAW)
1524 			old->counter[i] = new->counter[i];
1525 		else
1526 			old->counter[i] = new->counter[i] - old->counter[i];
1527 	}
1528 }
1529 
1530 int soft_c1_residency_display(int bic)
1531 {
1532 	if (!DO_BIC(BIC_CPU_c1) || use_c1_residency_msr)
1533 		return 0;
1534 
1535 	return DO_BIC_READ(bic);
1536 }
1537 
1538 /*
1539  * old = new - old
1540  */
1541 int delta_thread(struct thread_data *new, struct thread_data *old, struct core_data *core_delta)
1542 {
1543 	int i;
1544 	struct msr_counter *mp;
1545 
1546 	/* we run cpuid just the 1st time, copy the results */
1547 	if (DO_BIC(BIC_APIC))
1548 		new->apic_id = old->apic_id;
1549 	if (DO_BIC(BIC_X2APIC))
1550 		new->x2apic_id = old->x2apic_id;
1551 
1552 	/*
1553 	 * the timestamps from start of measurement interval are in "old"
1554 	 * the timestamp from end of measurement interval are in "new"
1555 	 * over-write old w/ new so we can print end of interval values
1556 	 */
1557 
1558 	timersub(&new->tv_begin, &old->tv_begin, &old->tv_delta);
1559 	old->tv_begin = new->tv_begin;
1560 	old->tv_end = new->tv_end;
1561 
1562 	old->tsc = new->tsc - old->tsc;
1563 
1564 	/* check for TSC < 1 Mcycles over interval */
1565 	if (old->tsc < (1000 * 1000))
1566 		errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n"
1567 		     "You can disable all c-states by booting with \"idle=poll\"\n"
1568 		     "or just the deep ones with \"processor.max_cstate=1\"");
1569 
1570 	old->c1 = new->c1 - old->c1;
1571 
1572 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) || soft_c1_residency_display(BIC_Avg_MHz)) {
1573 		if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
1574 			old->aperf = new->aperf - old->aperf;
1575 			old->mperf = new->mperf - old->mperf;
1576 		} else {
1577 			return -1;
1578 		}
1579 	}
1580 
1581 	if (use_c1_residency_msr) {
1582 		/*
1583 		 * Some models have a dedicated C1 residency MSR,
1584 		 * which should be more accurate than the derivation below.
1585 		 */
1586 	} else {
1587 		/*
1588 		 * As counter collection is not atomic,
1589 		 * it is possible for mperf's non-halted cycles + idle states
1590 		 * to exceed TSC's all cycles: show c1 = 0% in that case.
1591 		 */
1592 		if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak))
1593 			old->c1 = 0;
1594 		else {
1595 			/* normal case, derive c1 */
1596 			old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3
1597 			    - core_delta->c6 - core_delta->c7;
1598 		}
1599 	}
1600 
1601 	if (old->mperf == 0) {
1602 		if (debug > 1)
1603 			fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id);
1604 		old->mperf = 1;	/* divide by 0 protection */
1605 	}
1606 
1607 	if (DO_BIC(BIC_IPC))
1608 		old->instr_count = new->instr_count - old->instr_count;
1609 
1610 	if (DO_BIC(BIC_IRQ))
1611 		old->irq_count = new->irq_count - old->irq_count;
1612 
1613 	if (DO_BIC(BIC_SMI))
1614 		old->smi_count = new->smi_count - old->smi_count;
1615 
1616 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1617 		if (mp->format == FORMAT_RAW)
1618 			old->counter[i] = new->counter[i];
1619 		else
1620 			old->counter[i] = new->counter[i] - old->counter[i];
1621 	}
1622 	return 0;
1623 }
1624 
1625 int delta_cpu(struct thread_data *t, struct core_data *c,
1626 	      struct pkg_data *p, struct thread_data *t2, struct core_data *c2, struct pkg_data *p2)
1627 {
1628 	int retval = 0;
1629 
1630 	/* calculate core delta only for 1st thread in core */
1631 	if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
1632 		delta_core(c, c2);
1633 
1634 	/* always calculate thread delta */
1635 	retval = delta_thread(t, t2, c2);	/* c2 is core delta */
1636 	if (retval)
1637 		return retval;
1638 
1639 	/* calculate package delta only for 1st core in package */
1640 	if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
1641 		retval = delta_package(p, p2);
1642 
1643 	return retval;
1644 }
1645 
1646 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1647 {
1648 	int i;
1649 	struct msr_counter *mp;
1650 
1651 	t->tv_begin.tv_sec = 0;
1652 	t->tv_begin.tv_usec = 0;
1653 	t->tv_end.tv_sec = 0;
1654 	t->tv_end.tv_usec = 0;
1655 	t->tv_delta.tv_sec = 0;
1656 	t->tv_delta.tv_usec = 0;
1657 
1658 	t->tsc = 0;
1659 	t->aperf = 0;
1660 	t->mperf = 0;
1661 	t->c1 = 0;
1662 
1663 	t->instr_count = 0;
1664 
1665 	t->irq_count = 0;
1666 	t->smi_count = 0;
1667 
1668 	/* tells format_counters to dump all fields from this set */
1669 	t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE;
1670 
1671 	c->c3 = 0;
1672 	c->c6 = 0;
1673 	c->c7 = 0;
1674 	c->mc6_us = 0;
1675 	c->core_temp_c = 0;
1676 	c->core_energy = 0;
1677 	c->core_throt_cnt = 0;
1678 
1679 	p->pkg_wtd_core_c0 = 0;
1680 	p->pkg_any_core_c0 = 0;
1681 	p->pkg_any_gfxe_c0 = 0;
1682 	p->pkg_both_core_gfxe_c0 = 0;
1683 
1684 	p->pc2 = 0;
1685 	if (DO_BIC(BIC_Pkgpc3))
1686 		p->pc3 = 0;
1687 	if (DO_BIC(BIC_Pkgpc6))
1688 		p->pc6 = 0;
1689 	if (DO_BIC(BIC_Pkgpc7))
1690 		p->pc7 = 0;
1691 	p->pc8 = 0;
1692 	p->pc9 = 0;
1693 	p->pc10 = 0;
1694 	p->cpu_lpi = 0;
1695 	p->sys_lpi = 0;
1696 
1697 	p->energy_pkg = 0;
1698 	p->energy_dram = 0;
1699 	p->energy_cores = 0;
1700 	p->energy_gfx = 0;
1701 	p->rapl_pkg_perf_status = 0;
1702 	p->rapl_dram_perf_status = 0;
1703 	p->pkg_temp_c = 0;
1704 
1705 	p->gfx_rc6_ms = 0;
1706 	p->uncore_mhz = 0;
1707 	p->gfx_mhz = 0;
1708 	p->gfx_act_mhz = 0;
1709 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next)
1710 		t->counter[i] = 0;
1711 
1712 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next)
1713 		c->counter[i] = 0;
1714 
1715 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next)
1716 		p->counter[i] = 0;
1717 }
1718 
1719 int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1720 {
1721 	int i;
1722 	struct msr_counter *mp;
1723 
1724 	/* copy un-changing apic_id's */
1725 	if (DO_BIC(BIC_APIC))
1726 		average.threads.apic_id = t->apic_id;
1727 	if (DO_BIC(BIC_X2APIC))
1728 		average.threads.x2apic_id = t->x2apic_id;
1729 
1730 	/* remember first tv_begin */
1731 	if (average.threads.tv_begin.tv_sec == 0)
1732 		average.threads.tv_begin = t->tv_begin;
1733 
1734 	/* remember last tv_end */
1735 	average.threads.tv_end = t->tv_end;
1736 
1737 	average.threads.tsc += t->tsc;
1738 	average.threads.aperf += t->aperf;
1739 	average.threads.mperf += t->mperf;
1740 	average.threads.c1 += t->c1;
1741 
1742 	average.threads.instr_count += t->instr_count;
1743 
1744 	average.threads.irq_count += t->irq_count;
1745 	average.threads.smi_count += t->smi_count;
1746 
1747 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1748 		if (mp->format == FORMAT_RAW)
1749 			continue;
1750 		average.threads.counter[i] += t->counter[i];
1751 	}
1752 
1753 	/* sum per-core values only for 1st thread in core */
1754 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
1755 		return 0;
1756 
1757 	average.cores.c3 += c->c3;
1758 	average.cores.c6 += c->c6;
1759 	average.cores.c7 += c->c7;
1760 	average.cores.mc6_us += c->mc6_us;
1761 
1762 	average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c);
1763 	average.cores.core_throt_cnt = MAX(average.cores.core_throt_cnt, c->core_throt_cnt);
1764 
1765 	average.cores.core_energy += c->core_energy;
1766 
1767 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1768 		if (mp->format == FORMAT_RAW)
1769 			continue;
1770 		average.cores.counter[i] += c->counter[i];
1771 	}
1772 
1773 	/* sum per-pkg values only for 1st core in pkg */
1774 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1775 		return 0;
1776 
1777 	if (DO_BIC(BIC_Totl_c0))
1778 		average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0;
1779 	if (DO_BIC(BIC_Any_c0))
1780 		average.packages.pkg_any_core_c0 += p->pkg_any_core_c0;
1781 	if (DO_BIC(BIC_GFX_c0))
1782 		average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0;
1783 	if (DO_BIC(BIC_CPUGFX))
1784 		average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0;
1785 
1786 	average.packages.pc2 += p->pc2;
1787 	if (DO_BIC(BIC_Pkgpc3))
1788 		average.packages.pc3 += p->pc3;
1789 	if (DO_BIC(BIC_Pkgpc6))
1790 		average.packages.pc6 += p->pc6;
1791 	if (DO_BIC(BIC_Pkgpc7))
1792 		average.packages.pc7 += p->pc7;
1793 	average.packages.pc8 += p->pc8;
1794 	average.packages.pc9 += p->pc9;
1795 	average.packages.pc10 += p->pc10;
1796 
1797 	average.packages.cpu_lpi = p->cpu_lpi;
1798 	average.packages.sys_lpi = p->sys_lpi;
1799 
1800 	average.packages.energy_pkg += p->energy_pkg;
1801 	average.packages.energy_dram += p->energy_dram;
1802 	average.packages.energy_cores += p->energy_cores;
1803 	average.packages.energy_gfx += p->energy_gfx;
1804 
1805 	average.packages.gfx_rc6_ms = p->gfx_rc6_ms;
1806 	average.packages.uncore_mhz = p->uncore_mhz;
1807 	average.packages.gfx_mhz = p->gfx_mhz;
1808 	average.packages.gfx_act_mhz = p->gfx_act_mhz;
1809 
1810 	average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c);
1811 
1812 	average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status;
1813 	average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status;
1814 
1815 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1816 		if ((mp->format == FORMAT_RAW) && (topo.num_packages == 0))
1817 			average.packages.counter[i] = p->counter[i];
1818 		else
1819 			average.packages.counter[i] += p->counter[i];
1820 	}
1821 	return 0;
1822 }
1823 
1824 /*
1825  * sum the counters for all cpus in the system
1826  * compute the weighted average
1827  */
1828 void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1829 {
1830 	int i;
1831 	struct msr_counter *mp;
1832 
1833 	clear_counters(&average.threads, &average.cores, &average.packages);
1834 
1835 	for_all_cpus(sum_counters, t, c, p);
1836 
1837 	/* Use the global time delta for the average. */
1838 	average.threads.tv_delta = tv_delta;
1839 
1840 	average.threads.tsc /= topo.num_cpus;
1841 	average.threads.aperf /= topo.num_cpus;
1842 	average.threads.mperf /= topo.num_cpus;
1843 	average.threads.instr_count /= topo.num_cpus;
1844 	average.threads.c1 /= topo.num_cpus;
1845 
1846 	if (average.threads.irq_count > 9999999)
1847 		sums_need_wide_columns = 1;
1848 
1849 	average.cores.c3 /= topo.num_cores;
1850 	average.cores.c6 /= topo.num_cores;
1851 	average.cores.c7 /= topo.num_cores;
1852 	average.cores.mc6_us /= topo.num_cores;
1853 
1854 	if (DO_BIC(BIC_Totl_c0))
1855 		average.packages.pkg_wtd_core_c0 /= topo.num_packages;
1856 	if (DO_BIC(BIC_Any_c0))
1857 		average.packages.pkg_any_core_c0 /= topo.num_packages;
1858 	if (DO_BIC(BIC_GFX_c0))
1859 		average.packages.pkg_any_gfxe_c0 /= topo.num_packages;
1860 	if (DO_BIC(BIC_CPUGFX))
1861 		average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages;
1862 
1863 	average.packages.pc2 /= topo.num_packages;
1864 	if (DO_BIC(BIC_Pkgpc3))
1865 		average.packages.pc3 /= topo.num_packages;
1866 	if (DO_BIC(BIC_Pkgpc6))
1867 		average.packages.pc6 /= topo.num_packages;
1868 	if (DO_BIC(BIC_Pkgpc7))
1869 		average.packages.pc7 /= topo.num_packages;
1870 
1871 	average.packages.pc8 /= topo.num_packages;
1872 	average.packages.pc9 /= topo.num_packages;
1873 	average.packages.pc10 /= topo.num_packages;
1874 
1875 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
1876 		if (mp->format == FORMAT_RAW)
1877 			continue;
1878 		if (mp->type == COUNTER_ITEMS) {
1879 			if (average.threads.counter[i] > 9999999)
1880 				sums_need_wide_columns = 1;
1881 			continue;
1882 		}
1883 		average.threads.counter[i] /= topo.num_cpus;
1884 	}
1885 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
1886 		if (mp->format == FORMAT_RAW)
1887 			continue;
1888 		if (mp->type == COUNTER_ITEMS) {
1889 			if (average.cores.counter[i] > 9999999)
1890 				sums_need_wide_columns = 1;
1891 		}
1892 		average.cores.counter[i] /= topo.num_cores;
1893 	}
1894 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
1895 		if (mp->format == FORMAT_RAW)
1896 			continue;
1897 		if (mp->type == COUNTER_ITEMS) {
1898 			if (average.packages.counter[i] > 9999999)
1899 				sums_need_wide_columns = 1;
1900 		}
1901 		average.packages.counter[i] /= topo.num_packages;
1902 	}
1903 }
1904 
1905 static unsigned long long rdtsc(void)
1906 {
1907 	unsigned int low, high;
1908 
1909 	asm volatile ("rdtsc":"=a" (low), "=d"(high));
1910 
1911 	return low | ((unsigned long long)high) << 32;
1912 }
1913 
1914 /*
1915  * Open a file, and exit on failure
1916  */
1917 FILE *fopen_or_die(const char *path, const char *mode)
1918 {
1919 	FILE *filep = fopen(path, mode);
1920 
1921 	if (!filep)
1922 		err(1, "%s: open failed", path);
1923 	return filep;
1924 }
1925 
1926 /*
1927  * snapshot_sysfs_counter()
1928  *
1929  * return snapshot of given counter
1930  */
1931 unsigned long long snapshot_sysfs_counter(char *path)
1932 {
1933 	FILE *fp;
1934 	int retval;
1935 	unsigned long long counter;
1936 
1937 	fp = fopen_or_die(path, "r");
1938 
1939 	retval = fscanf(fp, "%lld", &counter);
1940 	if (retval != 1)
1941 		err(1, "snapshot_sysfs_counter(%s)", path);
1942 
1943 	fclose(fp);
1944 
1945 	return counter;
1946 }
1947 
1948 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp)
1949 {
1950 	if (mp->msr_num != 0) {
1951 		if (get_msr(cpu, mp->msr_num, counterp))
1952 			return -1;
1953 	} else {
1954 		char path[128 + PATH_BYTES];
1955 
1956 		if (mp->flags & SYSFS_PERCPU) {
1957 			sprintf(path, "/sys/devices/system/cpu/cpu%d/%s", cpu, mp->path);
1958 
1959 			*counterp = snapshot_sysfs_counter(path);
1960 		} else {
1961 			*counterp = snapshot_sysfs_counter(mp->path);
1962 		}
1963 	}
1964 
1965 	return 0;
1966 }
1967 
1968 unsigned long long get_uncore_mhz(int package, int die)
1969 {
1970 	char path[128];
1971 
1972 	sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_%02d_die_%02d/current_freq_khz", package,
1973 		die);
1974 
1975 	return (snapshot_sysfs_counter(path) / 1000);
1976 }
1977 
1978 int get_epb(int cpu)
1979 {
1980 	char path[128 + PATH_BYTES];
1981 	unsigned long long msr;
1982 	int ret, epb = -1;
1983 	FILE *fp;
1984 
1985 	sprintf(path, "/sys/devices/system/cpu/cpu%d/power/energy_perf_bias", cpu);
1986 
1987 	fp = fopen(path, "r");
1988 	if (!fp)
1989 		goto msr_fallback;
1990 
1991 	ret = fscanf(fp, "%d", &epb);
1992 	if (ret != 1)
1993 		err(1, "%s(%s)", __func__, path);
1994 
1995 	fclose(fp);
1996 
1997 	return epb;
1998 
1999 msr_fallback:
2000 	get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr);
2001 
2002 	return msr & 0xf;
2003 }
2004 
2005 void get_apic_id(struct thread_data *t)
2006 {
2007 	unsigned int eax, ebx, ecx, edx;
2008 
2009 	if (DO_BIC(BIC_APIC)) {
2010 		eax = ebx = ecx = edx = 0;
2011 		__cpuid(1, eax, ebx, ecx, edx);
2012 
2013 		t->apic_id = (ebx >> 24) & 0xff;
2014 	}
2015 
2016 	if (!DO_BIC(BIC_X2APIC))
2017 		return;
2018 
2019 	if (authentic_amd || hygon_genuine) {
2020 		unsigned int topology_extensions;
2021 
2022 		if (max_extended_level < 0x8000001e)
2023 			return;
2024 
2025 		eax = ebx = ecx = edx = 0;
2026 		__cpuid(0x80000001, eax, ebx, ecx, edx);
2027 		topology_extensions = ecx & (1 << 22);
2028 
2029 		if (topology_extensions == 0)
2030 			return;
2031 
2032 		eax = ebx = ecx = edx = 0;
2033 		__cpuid(0x8000001e, eax, ebx, ecx, edx);
2034 
2035 		t->x2apic_id = eax;
2036 		return;
2037 	}
2038 
2039 	if (!genuine_intel)
2040 		return;
2041 
2042 	if (max_level < 0xb)
2043 		return;
2044 
2045 	ecx = 0;
2046 	__cpuid(0xb, eax, ebx, ecx, edx);
2047 	t->x2apic_id = edx;
2048 
2049 	if (debug && (t->apic_id != (t->x2apic_id & 0xff)))
2050 		fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n", t->cpu_id, t->apic_id, t->x2apic_id);
2051 }
2052 
2053 int get_core_throt_cnt(int cpu, unsigned long long *cnt)
2054 {
2055 	char path[128 + PATH_BYTES];
2056 	unsigned long long tmp;
2057 	FILE *fp;
2058 	int ret;
2059 
2060 	sprintf(path, "/sys/devices/system/cpu/cpu%d/thermal_throttle/core_throttle_count", cpu);
2061 	fp = fopen(path, "r");
2062 	if (!fp)
2063 		return -1;
2064 	ret = fscanf(fp, "%lld", &tmp);
2065 	fclose(fp);
2066 	if (ret != 1)
2067 		return -1;
2068 	*cnt = tmp;
2069 
2070 	return 0;
2071 }
2072 
2073 /*
2074  * get_counters(...)
2075  * migrate to cpu
2076  * acquire and record local counters for that cpu
2077  */
2078 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
2079 {
2080 	int cpu = t->cpu_id;
2081 	unsigned long long msr;
2082 	int aperf_mperf_retry_count = 0;
2083 	struct msr_counter *mp;
2084 	int i;
2085 
2086 	if (cpu_migrate(cpu)) {
2087 		fprintf(outf, "get_counters: Could not migrate to CPU %d\n", cpu);
2088 		return -1;
2089 	}
2090 
2091 	gettimeofday(&t->tv_begin, (struct timezone *)NULL);
2092 
2093 	if (first_counter_read)
2094 		get_apic_id(t);
2095 retry:
2096 	t->tsc = rdtsc();	/* we are running on local CPU of interest */
2097 
2098 	if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) || soft_c1_residency_display(BIC_Avg_MHz)) {
2099 		unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time;
2100 
2101 		/*
2102 		 * The TSC, APERF and MPERF must be read together for
2103 		 * APERF/MPERF and MPERF/TSC to give accurate results.
2104 		 *
2105 		 * Unfortunately, APERF and MPERF are read by
2106 		 * individual system call, so delays may occur
2107 		 * between them.  If the time to read them
2108 		 * varies by a large amount, we re-read them.
2109 		 */
2110 
2111 		/*
2112 		 * This initial dummy APERF read has been seen to
2113 		 * reduce jitter in the subsequent reads.
2114 		 */
2115 
2116 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
2117 			return -3;
2118 
2119 		t->tsc = rdtsc();	/* re-read close to APERF */
2120 
2121 		tsc_before = t->tsc;
2122 
2123 		if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
2124 			return -3;
2125 
2126 		tsc_between = rdtsc();
2127 
2128 		if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
2129 			return -4;
2130 
2131 		tsc_after = rdtsc();
2132 
2133 		aperf_time = tsc_between - tsc_before;
2134 		mperf_time = tsc_after - tsc_between;
2135 
2136 		/*
2137 		 * If the system call latency to read APERF and MPERF
2138 		 * differ by more than 2x, then try again.
2139 		 */
2140 		if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) {
2141 			aperf_mperf_retry_count++;
2142 			if (aperf_mperf_retry_count < 5)
2143 				goto retry;
2144 			else
2145 				warnx("cpu%d jitter %lld %lld", cpu, aperf_time, mperf_time);
2146 		}
2147 		aperf_mperf_retry_count = 0;
2148 
2149 		t->aperf = t->aperf * aperf_mperf_multiplier;
2150 		t->mperf = t->mperf * aperf_mperf_multiplier;
2151 	}
2152 
2153 	if (DO_BIC(BIC_IPC))
2154 		if (read(get_instr_count_fd(cpu), &t->instr_count, sizeof(long long)) != sizeof(long long))
2155 			return -4;
2156 
2157 	if (DO_BIC(BIC_IRQ))
2158 		t->irq_count = irqs_per_cpu[cpu];
2159 	if (DO_BIC(BIC_SMI)) {
2160 		if (get_msr(cpu, MSR_SMI_COUNT, &msr))
2161 			return -5;
2162 		t->smi_count = msr & 0xFFFFFFFF;
2163 	}
2164 	if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) {
2165 		if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1))
2166 			return -6;
2167 	}
2168 
2169 	for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) {
2170 		if (get_mp(cpu, mp, &t->counter[i]))
2171 			return -10;
2172 	}
2173 
2174 	/* collect core counters only for 1st thread in core */
2175 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
2176 		goto done;
2177 
2178 	if (DO_BIC(BIC_CPU_c3) || soft_c1_residency_display(BIC_CPU_c3)) {
2179 		if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
2180 			return -6;
2181 	}
2182 
2183 	if ((DO_BIC(BIC_CPU_c6) || soft_c1_residency_display(BIC_CPU_c6)) && !do_knl_cstates) {
2184 		if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
2185 			return -7;
2186 	} else if (do_knl_cstates && soft_c1_residency_display(BIC_CPU_c6)) {
2187 		if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6))
2188 			return -7;
2189 	}
2190 
2191 	if (DO_BIC(BIC_CPU_c7) || soft_c1_residency_display(BIC_CPU_c7)) {
2192 		if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
2193 			return -8;
2194 		else if (t->is_atom) {
2195 			/*
2196 			 * For Atom CPUs that has core cstate deeper than c6,
2197 			 * MSR_CORE_C6_RESIDENCY returns residency of cc6 and deeper.
2198 			 * Minus CC7 (and deeper cstates) residency to get
2199 			 * accturate cc6 residency.
2200 			 */
2201 			c->c6 -= c->c7;
2202 		}
2203 	}
2204 
2205 	if (DO_BIC(BIC_Mod_c6))
2206 		if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us))
2207 			return -8;
2208 
2209 	if (DO_BIC(BIC_CoreTmp)) {
2210 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
2211 			return -9;
2212 		c->core_temp_c = tj_max - ((msr >> 16) & 0x7F);
2213 	}
2214 
2215 	if (DO_BIC(BIC_CORE_THROT_CNT))
2216 		get_core_throt_cnt(cpu, &c->core_throt_cnt);
2217 
2218 	if (do_rapl & RAPL_AMD_F17H) {
2219 		if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr))
2220 			return -14;
2221 		c->core_energy = msr & 0xFFFFFFFF;
2222 	}
2223 
2224 	for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) {
2225 		if (get_mp(cpu, mp, &c->counter[i]))
2226 			return -10;
2227 	}
2228 
2229 	/* collect package counters only for 1st core in package */
2230 	if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
2231 		goto done;
2232 
2233 	if (DO_BIC(BIC_Totl_c0)) {
2234 		if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0))
2235 			return -10;
2236 	}
2237 	if (DO_BIC(BIC_Any_c0)) {
2238 		if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0))
2239 			return -11;
2240 	}
2241 	if (DO_BIC(BIC_GFX_c0)) {
2242 		if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0))
2243 			return -12;
2244 	}
2245 	if (DO_BIC(BIC_CPUGFX)) {
2246 		if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0))
2247 			return -13;
2248 	}
2249 	if (DO_BIC(BIC_Pkgpc3))
2250 		if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
2251 			return -9;
2252 	if (DO_BIC(BIC_Pkgpc6)) {
2253 		if (do_slm_cstates) {
2254 			if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6))
2255 				return -10;
2256 		} else {
2257 			if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
2258 				return -10;
2259 		}
2260 	}
2261 
2262 	if (DO_BIC(BIC_Pkgpc2))
2263 		if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
2264 			return -11;
2265 	if (DO_BIC(BIC_Pkgpc7))
2266 		if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
2267 			return -12;
2268 	if (DO_BIC(BIC_Pkgpc8))
2269 		if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
2270 			return -13;
2271 	if (DO_BIC(BIC_Pkgpc9))
2272 		if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
2273 			return -13;
2274 	if (DO_BIC(BIC_Pkgpc10))
2275 		if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
2276 			return -13;
2277 
2278 	if (DO_BIC(BIC_CPU_LPI))
2279 		p->cpu_lpi = cpuidle_cur_cpu_lpi_us;
2280 	if (DO_BIC(BIC_SYS_LPI))
2281 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
2282 
2283 	if (do_rapl & RAPL_PKG) {
2284 		if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr))
2285 			return -13;
2286 		p->energy_pkg = msr;
2287 	}
2288 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
2289 		if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr))
2290 			return -14;
2291 		p->energy_cores = msr;
2292 	}
2293 	if (do_rapl & RAPL_DRAM) {
2294 		if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
2295 			return -15;
2296 		p->energy_dram = msr;
2297 	}
2298 	if (do_rapl & RAPL_GFX) {
2299 		if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr))
2300 			return -16;
2301 		p->energy_gfx = msr;
2302 	}
2303 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
2304 		if (get_msr_sum(cpu, MSR_PKG_PERF_STATUS, &msr))
2305 			return -16;
2306 		p->rapl_pkg_perf_status = msr;
2307 	}
2308 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
2309 		if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr))
2310 			return -16;
2311 		p->rapl_dram_perf_status = msr;
2312 	}
2313 	if (do_rapl & RAPL_AMD_F17H) {
2314 		if (get_msr_sum(cpu, MSR_PKG_ENERGY_STAT, &msr))
2315 			return -13;
2316 		p->energy_pkg = msr;
2317 	}
2318 	if (DO_BIC(BIC_PkgTmp)) {
2319 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
2320 			return -17;
2321 		p->pkg_temp_c = tj_max - ((msr >> 16) & 0x7F);
2322 	}
2323 
2324 	if (DO_BIC(BIC_GFX_rc6))
2325 		p->gfx_rc6_ms = gfx_cur_rc6_ms;
2326 
2327 	/* n.b. assume die0 uncore frequency applies to whole package */
2328 	if (DO_BIC(BIC_UNCORE_MHZ))
2329 		p->uncore_mhz = get_uncore_mhz(p->package_id, 0);
2330 
2331 	if (DO_BIC(BIC_GFXMHz))
2332 		p->gfx_mhz = gfx_cur_mhz;
2333 
2334 	if (DO_BIC(BIC_GFXACTMHz))
2335 		p->gfx_act_mhz = gfx_act_mhz;
2336 
2337 	for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) {
2338 		if (get_mp(cpu, mp, &p->counter[i]))
2339 			return -10;
2340 	}
2341 done:
2342 	gettimeofday(&t->tv_end, (struct timezone *)NULL);
2343 
2344 	return 0;
2345 }
2346 
2347 /*
2348  * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit:
2349  * If you change the values, note they are used both in comparisons
2350  * (>= PCL__7) and to index pkg_cstate_limit_strings[].
2351  */
2352 
2353 #define PCLUKN 0		/* Unknown */
2354 #define PCLRSV 1		/* Reserved */
2355 #define PCL__0 2		/* PC0 */
2356 #define PCL__1 3		/* PC1 */
2357 #define PCL__2 4		/* PC2 */
2358 #define PCL__3 5		/* PC3 */
2359 #define PCL__4 6		/* PC4 */
2360 #define PCL__6 7		/* PC6 */
2361 #define PCL_6N 8		/* PC6 No Retention */
2362 #define PCL_6R 9		/* PC6 Retention */
2363 #define PCL__7 10		/* PC7 */
2364 #define PCL_7S 11		/* PC7 Shrink */
2365 #define PCL__8 12		/* PC8 */
2366 #define PCL__9 13		/* PC9 */
2367 #define PCL_10 14		/* PC10 */
2368 #define PCLUNL 15		/* Unlimited */
2369 
2370 int pkg_cstate_limit = PCLUKN;
2371 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2",
2372 	"pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited"
2373 };
2374 
2375 int nhm_pkg_cstate_limits[16] =
2376     { PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2377 	PCLRSV, PCLRSV
2378 };
2379 
2380 int snb_pkg_cstate_limits[16] =
2381     { PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2382 	PCLRSV, PCLRSV
2383 };
2384 
2385 int hsw_pkg_cstate_limits[16] =
2386     { PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2387 	PCLRSV, PCLRSV
2388 };
2389 
2390 int slv_pkg_cstate_limits[16] =
2391     { PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2392 	PCL__6, PCL__7
2393 };
2394 
2395 int amt_pkg_cstate_limits[16] =
2396     { PCLUNL, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2397 	PCLRSV, PCLRSV
2398 };
2399 
2400 int phi_pkg_cstate_limits[16] =
2401     { PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2402 	PCLRSV, PCLRSV
2403 };
2404 
2405 int glm_pkg_cstate_limits[16] =
2406     { PCLUNL, PCL__1, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCL_10, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2407 	PCLRSV, PCLRSV
2408 };
2409 
2410 int skx_pkg_cstate_limits[16] =
2411     { PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2412 	PCLRSV, PCLRSV
2413 };
2414 
2415 int icx_pkg_cstate_limits[16] =
2416     { PCL__0, PCL__2, PCL__6, PCL__6, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV,
2417 	PCLRSV, PCLRSV
2418 };
2419 
2420 static void calculate_tsc_tweak()
2421 {
2422 	tsc_tweak = base_hz / tsc_hz;
2423 }
2424 
2425 void prewake_cstate_probe(unsigned int family, unsigned int model);
2426 
2427 static void dump_nhm_platform_info(void)
2428 {
2429 	unsigned long long msr;
2430 	unsigned int ratio;
2431 
2432 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
2433 
2434 	fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr);
2435 
2436 	ratio = (msr >> 40) & 0xFF;
2437 	fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n", ratio, bclk, ratio * bclk);
2438 
2439 	ratio = (msr >> 8) & 0xFF;
2440 	fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n", ratio, bclk, ratio * bclk);
2441 
2442 	get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr);
2443 	fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n",
2444 		base_cpu, msr, msr & 0x2 ? "EN" : "DIS");
2445 
2446 	/* C-state Pre-wake Disable (CSTATE_PREWAKE_DISABLE) */
2447 	if (dis_cstate_prewake)
2448 		fprintf(outf, "C-state Pre-wake: %sabled\n", msr & 0x40000000 ? "DIS" : "EN");
2449 
2450 	return;
2451 }
2452 
2453 static void dump_hsw_turbo_ratio_limits(void)
2454 {
2455 	unsigned long long msr;
2456 	unsigned int ratio;
2457 
2458 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr);
2459 
2460 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr);
2461 
2462 	ratio = (msr >> 8) & 0xFF;
2463 	if (ratio)
2464 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n", ratio, bclk, ratio * bclk);
2465 
2466 	ratio = (msr >> 0) & 0xFF;
2467 	if (ratio)
2468 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n", ratio, bclk, ratio * bclk);
2469 	return;
2470 }
2471 
2472 static void dump_ivt_turbo_ratio_limits(void)
2473 {
2474 	unsigned long long msr;
2475 	unsigned int ratio;
2476 
2477 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr);
2478 
2479 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr);
2480 
2481 	ratio = (msr >> 56) & 0xFF;
2482 	if (ratio)
2483 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n", ratio, bclk, ratio * bclk);
2484 
2485 	ratio = (msr >> 48) & 0xFF;
2486 	if (ratio)
2487 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n", ratio, bclk, ratio * bclk);
2488 
2489 	ratio = (msr >> 40) & 0xFF;
2490 	if (ratio)
2491 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n", ratio, bclk, ratio * bclk);
2492 
2493 	ratio = (msr >> 32) & 0xFF;
2494 	if (ratio)
2495 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n", ratio, bclk, ratio * bclk);
2496 
2497 	ratio = (msr >> 24) & 0xFF;
2498 	if (ratio)
2499 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n", ratio, bclk, ratio * bclk);
2500 
2501 	ratio = (msr >> 16) & 0xFF;
2502 	if (ratio)
2503 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n", ratio, bclk, ratio * bclk);
2504 
2505 	ratio = (msr >> 8) & 0xFF;
2506 	if (ratio)
2507 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n", ratio, bclk, ratio * bclk);
2508 
2509 	ratio = (msr >> 0) & 0xFF;
2510 	if (ratio)
2511 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n", ratio, bclk, ratio * bclk);
2512 	return;
2513 }
2514 
2515 int has_turbo_ratio_group_limits(int family, int model)
2516 {
2517 
2518 	if (!genuine_intel)
2519 		return 0;
2520 
2521 	if (family != 6)
2522 		return 0;
2523 
2524 	switch (model) {
2525 	case INTEL_FAM6_ATOM_GOLDMONT:
2526 	case INTEL_FAM6_SKYLAKE_X:
2527 	case INTEL_FAM6_ICELAKE_X:
2528 	case INTEL_FAM6_SAPPHIRERAPIDS_X:
2529 	case INTEL_FAM6_ATOM_GOLDMONT_D:
2530 	case INTEL_FAM6_ATOM_TREMONT_D:
2531 		return 1;
2532 	default:
2533 		return 0;
2534 	}
2535 }
2536 
2537 static void dump_turbo_ratio_limits(int trl_msr_offset, int family, int model)
2538 {
2539 	unsigned long long msr, core_counts;
2540 	int shift;
2541 
2542 	get_msr(base_cpu, trl_msr_offset, &msr);
2543 	fprintf(outf, "cpu%d: MSR_%sTURBO_RATIO_LIMIT: 0x%08llx\n",
2544 		base_cpu, trl_msr_offset == MSR_SECONDARY_TURBO_RATIO_LIMIT ? "SECONDARY_" : "", msr);
2545 
2546 	if (has_turbo_ratio_group_limits(family, model)) {
2547 		get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts);
2548 		fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts);
2549 	} else {
2550 		core_counts = 0x0807060504030201;
2551 	}
2552 
2553 	for (shift = 56; shift >= 0; shift -= 8) {
2554 		unsigned int ratio, group_size;
2555 
2556 		ratio = (msr >> shift) & 0xFF;
2557 		group_size = (core_counts >> shift) & 0xFF;
2558 		if (ratio)
2559 			fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n",
2560 				ratio, bclk, ratio * bclk, group_size);
2561 	}
2562 
2563 	return;
2564 }
2565 
2566 static void dump_atom_turbo_ratio_limits(void)
2567 {
2568 	unsigned long long msr;
2569 	unsigned int ratio;
2570 
2571 	get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr);
2572 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2573 
2574 	ratio = (msr >> 0) & 0x3F;
2575 	if (ratio)
2576 		fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n", ratio, bclk, ratio * bclk);
2577 
2578 	ratio = (msr >> 8) & 0x3F;
2579 	if (ratio)
2580 		fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n", ratio, bclk, ratio * bclk);
2581 
2582 	ratio = (msr >> 16) & 0x3F;
2583 	if (ratio)
2584 		fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n", ratio, bclk, ratio * bclk);
2585 
2586 	get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr);
2587 	fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF);
2588 
2589 	ratio = (msr >> 24) & 0x3F;
2590 	if (ratio)
2591 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n", ratio, bclk, ratio * bclk);
2592 
2593 	ratio = (msr >> 16) & 0x3F;
2594 	if (ratio)
2595 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n", ratio, bclk, ratio * bclk);
2596 
2597 	ratio = (msr >> 8) & 0x3F;
2598 	if (ratio)
2599 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n", ratio, bclk, ratio * bclk);
2600 
2601 	ratio = (msr >> 0) & 0x3F;
2602 	if (ratio)
2603 		fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n", ratio, bclk, ratio * bclk);
2604 }
2605 
2606 static void dump_knl_turbo_ratio_limits(void)
2607 {
2608 	const unsigned int buckets_no = 7;
2609 
2610 	unsigned long long msr;
2611 	int delta_cores, delta_ratio;
2612 	int i, b_nr;
2613 	unsigned int cores[buckets_no];
2614 	unsigned int ratio[buckets_no];
2615 
2616 	get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr);
2617 
2618 	fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr);
2619 
2620 	/*
2621 	 * Turbo encoding in KNL is as follows:
2622 	 * [0] -- Reserved
2623 	 * [7:1] -- Base value of number of active cores of bucket 1.
2624 	 * [15:8] -- Base value of freq ratio of bucket 1.
2625 	 * [20:16] -- +ve delta of number of active cores of bucket 2.
2626 	 * i.e. active cores of bucket 2 =
2627 	 * active cores of bucket 1 + delta
2628 	 * [23:21] -- Negative delta of freq ratio of bucket 2.
2629 	 * i.e. freq ratio of bucket 2 =
2630 	 * freq ratio of bucket 1 - delta
2631 	 * [28:24]-- +ve delta of number of active cores of bucket 3.
2632 	 * [31:29]-- -ve delta of freq ratio of bucket 3.
2633 	 * [36:32]-- +ve delta of number of active cores of bucket 4.
2634 	 * [39:37]-- -ve delta of freq ratio of bucket 4.
2635 	 * [44:40]-- +ve delta of number of active cores of bucket 5.
2636 	 * [47:45]-- -ve delta of freq ratio of bucket 5.
2637 	 * [52:48]-- +ve delta of number of active cores of bucket 6.
2638 	 * [55:53]-- -ve delta of freq ratio of bucket 6.
2639 	 * [60:56]-- +ve delta of number of active cores of bucket 7.
2640 	 * [63:61]-- -ve delta of freq ratio of bucket 7.
2641 	 */
2642 
2643 	b_nr = 0;
2644 	cores[b_nr] = (msr & 0xFF) >> 1;
2645 	ratio[b_nr] = (msr >> 8) & 0xFF;
2646 
2647 	for (i = 16; i < 64; i += 8) {
2648 		delta_cores = (msr >> i) & 0x1F;
2649 		delta_ratio = (msr >> (i + 5)) & 0x7;
2650 
2651 		cores[b_nr + 1] = cores[b_nr] + delta_cores;
2652 		ratio[b_nr + 1] = ratio[b_nr] - delta_ratio;
2653 		b_nr++;
2654 	}
2655 
2656 	for (i = buckets_no - 1; i >= 0; i--)
2657 		if (i > 0 ? ratio[i] != ratio[i - 1] : 1)
2658 			fprintf(outf,
2659 				"%d * %.1f = %.1f MHz max turbo %d active cores\n",
2660 				ratio[i], bclk, ratio[i] * bclk, cores[i]);
2661 }
2662 
2663 static void dump_nhm_cst_cfg(void)
2664 {
2665 	unsigned long long msr;
2666 
2667 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
2668 
2669 	fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr);
2670 
2671 	fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)",
2672 		(msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "",
2673 		(msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "",
2674 		(msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "",
2675 		(msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "",
2676 		(msr & (1 << 15)) ? "" : "UN", (unsigned int)msr & 0xF, pkg_cstate_limit_strings[pkg_cstate_limit]);
2677 
2678 #define AUTOMATIC_CSTATE_CONVERSION		(1UL << 16)
2679 	if (has_automatic_cstate_conversion) {
2680 		fprintf(outf, ", automatic c-state conversion=%s", (msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off");
2681 	}
2682 
2683 	fprintf(outf, ")\n");
2684 
2685 	return;
2686 }
2687 
2688 static void dump_config_tdp(void)
2689 {
2690 	unsigned long long msr;
2691 
2692 	get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr);
2693 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr);
2694 	fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF);
2695 
2696 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr);
2697 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr);
2698 	if (msr) {
2699 		fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2700 		fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2701 		fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2702 		fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF);
2703 	}
2704 	fprintf(outf, ")\n");
2705 
2706 	get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr);
2707 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr);
2708 	if (msr) {
2709 		fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF);
2710 		fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF);
2711 		fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF);
2712 		fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF);
2713 	}
2714 	fprintf(outf, ")\n");
2715 
2716 	get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr);
2717 	fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr);
2718 	if ((msr) & 0x3)
2719 		fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3);
2720 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2721 	fprintf(outf, ")\n");
2722 
2723 	get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr);
2724 	fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr);
2725 	fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF);
2726 	fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1);
2727 	fprintf(outf, ")\n");
2728 }
2729 
2730 unsigned int irtl_time_units[] = { 1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
2731 
2732 void print_irtl(void)
2733 {
2734 	unsigned long long msr;
2735 
2736 	get_msr(base_cpu, MSR_PKGC3_IRTL, &msr);
2737 	fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr);
2738 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2739 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2740 
2741 	get_msr(base_cpu, MSR_PKGC6_IRTL, &msr);
2742 	fprintf(outf, "cpu%d: MSR_PKGC6_IRTL: 0x%08llx (", base_cpu, msr);
2743 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2744 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2745 
2746 	get_msr(base_cpu, MSR_PKGC7_IRTL, &msr);
2747 	fprintf(outf, "cpu%d: MSR_PKGC7_IRTL: 0x%08llx (", base_cpu, msr);
2748 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2749 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2750 
2751 	if (!do_irtl_hsw)
2752 		return;
2753 
2754 	get_msr(base_cpu, MSR_PKGC8_IRTL, &msr);
2755 	fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr);
2756 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2757 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2758 
2759 	get_msr(base_cpu, MSR_PKGC9_IRTL, &msr);
2760 	fprintf(outf, "cpu%d: MSR_PKGC9_IRTL: 0x%08llx (", base_cpu, msr);
2761 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2762 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2763 
2764 	get_msr(base_cpu, MSR_PKGC10_IRTL, &msr);
2765 	fprintf(outf, "cpu%d: MSR_PKGC10_IRTL: 0x%08llx (", base_cpu, msr);
2766 	fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT",
2767 		(msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]);
2768 
2769 }
2770 
2771 void free_fd_percpu(void)
2772 {
2773 	int i;
2774 
2775 	for (i = 0; i < topo.max_cpu_num + 1; ++i) {
2776 		if (fd_percpu[i] != 0)
2777 			close(fd_percpu[i]);
2778 	}
2779 
2780 	free(fd_percpu);
2781 }
2782 
2783 void free_all_buffers(void)
2784 {
2785 	int i;
2786 
2787 	CPU_FREE(cpu_present_set);
2788 	cpu_present_set = NULL;
2789 	cpu_present_setsize = 0;
2790 
2791 	CPU_FREE(cpu_affinity_set);
2792 	cpu_affinity_set = NULL;
2793 	cpu_affinity_setsize = 0;
2794 
2795 	free(thread_even);
2796 	free(core_even);
2797 	free(package_even);
2798 
2799 	thread_even = NULL;
2800 	core_even = NULL;
2801 	package_even = NULL;
2802 
2803 	free(thread_odd);
2804 	free(core_odd);
2805 	free(package_odd);
2806 
2807 	thread_odd = NULL;
2808 	core_odd = NULL;
2809 	package_odd = NULL;
2810 
2811 	free(output_buffer);
2812 	output_buffer = NULL;
2813 	outp = NULL;
2814 
2815 	free_fd_percpu();
2816 
2817 	free(irq_column_2_cpu);
2818 	free(irqs_per_cpu);
2819 
2820 	for (i = 0; i <= topo.max_cpu_num; ++i) {
2821 		if (cpus[i].put_ids)
2822 			CPU_FREE(cpus[i].put_ids);
2823 	}
2824 	free(cpus);
2825 }
2826 
2827 /*
2828  * Parse a file containing a single int.
2829  * Return 0 if file can not be opened
2830  * Exit if file can be opened, but can not be parsed
2831  */
2832 int parse_int_file(const char *fmt, ...)
2833 {
2834 	va_list args;
2835 	char path[PATH_MAX];
2836 	FILE *filep;
2837 	int value;
2838 
2839 	va_start(args, fmt);
2840 	vsnprintf(path, sizeof(path), fmt, args);
2841 	va_end(args);
2842 	filep = fopen(path, "r");
2843 	if (!filep)
2844 		return 0;
2845 	if (fscanf(filep, "%d", &value) != 1)
2846 		err(1, "%s: failed to parse number from file", path);
2847 	fclose(filep);
2848 	return value;
2849 }
2850 
2851 /*
2852  * cpu_is_first_core_in_package(cpu)
2853  * return 1 if given CPU is 1st core in package
2854  */
2855 int cpu_is_first_core_in_package(int cpu)
2856 {
2857 	return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
2858 }
2859 
2860 int get_physical_package_id(int cpu)
2861 {
2862 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
2863 }
2864 
2865 int get_die_id(int cpu)
2866 {
2867 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu);
2868 }
2869 
2870 int get_core_id(int cpu)
2871 {
2872 	return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
2873 }
2874 
2875 void set_node_data(void)
2876 {
2877 	int pkg, node, lnode, cpu, cpux;
2878 	int cpu_count;
2879 
2880 	/* initialize logical_node_id */
2881 	for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu)
2882 		cpus[cpu].logical_node_id = -1;
2883 
2884 	cpu_count = 0;
2885 	for (pkg = 0; pkg < topo.num_packages; pkg++) {
2886 		lnode = 0;
2887 		for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) {
2888 			if (cpus[cpu].physical_package_id != pkg)
2889 				continue;
2890 			/* find a cpu with an unset logical_node_id */
2891 			if (cpus[cpu].logical_node_id != -1)
2892 				continue;
2893 			cpus[cpu].logical_node_id = lnode;
2894 			node = cpus[cpu].physical_node_id;
2895 			cpu_count++;
2896 			/*
2897 			 * find all matching cpus on this pkg and set
2898 			 * the logical_node_id
2899 			 */
2900 			for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) {
2901 				if ((cpus[cpux].physical_package_id == pkg) && (cpus[cpux].physical_node_id == node)) {
2902 					cpus[cpux].logical_node_id = lnode;
2903 					cpu_count++;
2904 				}
2905 			}
2906 			lnode++;
2907 			if (lnode > topo.nodes_per_pkg)
2908 				topo.nodes_per_pkg = lnode;
2909 		}
2910 		if (cpu_count >= topo.max_cpu_num)
2911 			break;
2912 	}
2913 }
2914 
2915 int get_physical_node_id(struct cpu_topology *thiscpu)
2916 {
2917 	char path[80];
2918 	FILE *filep;
2919 	int i;
2920 	int cpu = thiscpu->logical_cpu_id;
2921 
2922 	for (i = 0; i <= topo.max_cpu_num; i++) {
2923 		sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist", cpu, i);
2924 		filep = fopen(path, "r");
2925 		if (!filep)
2926 			continue;
2927 		fclose(filep);
2928 		return i;
2929 	}
2930 	return -1;
2931 }
2932 
2933 int get_thread_siblings(struct cpu_topology *thiscpu)
2934 {
2935 	char path[80], character;
2936 	FILE *filep;
2937 	unsigned long map;
2938 	int so, shift, sib_core;
2939 	int cpu = thiscpu->logical_cpu_id;
2940 	int offset = topo.max_cpu_num + 1;
2941 	size_t size;
2942 	int thread_id = 0;
2943 
2944 	thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1));
2945 	if (thiscpu->thread_id < 0)
2946 		thiscpu->thread_id = thread_id++;
2947 	if (!thiscpu->put_ids)
2948 		return -1;
2949 
2950 	size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
2951 	CPU_ZERO_S(size, thiscpu->put_ids);
2952 
2953 	sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu);
2954 	filep = fopen(path, "r");
2955 
2956 	if (!filep) {
2957 		warnx("%s: open failed", path);
2958 		return -1;
2959 	}
2960 	do {
2961 		offset -= BITMASK_SIZE;
2962 		if (fscanf(filep, "%lx%c", &map, &character) != 2)
2963 			err(1, "%s: failed to parse file", path);
2964 		for (shift = 0; shift < BITMASK_SIZE; shift++) {
2965 			if ((map >> shift) & 0x1) {
2966 				so = shift + offset;
2967 				sib_core = get_core_id(so);
2968 				if (sib_core == thiscpu->physical_core_id) {
2969 					CPU_SET_S(so, size, thiscpu->put_ids);
2970 					if ((so != cpu) && (cpus[so].thread_id < 0))
2971 						cpus[so].thread_id = thread_id++;
2972 				}
2973 			}
2974 		}
2975 	} while (character == ',');
2976 	fclose(filep);
2977 
2978 	return CPU_COUNT_S(size, thiscpu->put_ids);
2979 }
2980 
2981 /*
2982  * run func(thread, core, package) in topology order
2983  * skip non-present cpus
2984  */
2985 
2986 int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *,
2987 			       struct pkg_data *, struct thread_data *, struct core_data *,
2988 			       struct pkg_data *), struct thread_data *thread_base,
2989 		   struct core_data *core_base, struct pkg_data *pkg_base,
2990 		   struct thread_data *thread_base2, struct core_data *core_base2, struct pkg_data *pkg_base2)
2991 {
2992 	int retval, pkg_no, node_no, core_no, thread_no;
2993 
2994 	for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
2995 		for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) {
2996 			for (core_no = 0; core_no < topo.cores_per_node; ++core_no) {
2997 				for (thread_no = 0; thread_no < topo.threads_per_core; ++thread_no) {
2998 					struct thread_data *t, *t2;
2999 					struct core_data *c, *c2;
3000 					struct pkg_data *p, *p2;
3001 
3002 					t = GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no);
3003 
3004 					if (cpu_is_not_present(t->cpu_id))
3005 						continue;
3006 
3007 					t2 = GET_THREAD(thread_base2, thread_no, core_no, node_no, pkg_no);
3008 
3009 					c = GET_CORE(core_base, core_no, node_no, pkg_no);
3010 					c2 = GET_CORE(core_base2, core_no, node_no, pkg_no);
3011 
3012 					p = GET_PKG(pkg_base, pkg_no);
3013 					p2 = GET_PKG(pkg_base2, pkg_no);
3014 
3015 					retval = func(t, c, p, t2, c2, p2);
3016 					if (retval)
3017 						return retval;
3018 				}
3019 			}
3020 		}
3021 	}
3022 	return 0;
3023 }
3024 
3025 /*
3026  * run func(cpu) on every cpu in /proc/stat
3027  * return max_cpu number
3028  */
3029 int for_all_proc_cpus(int (func) (int))
3030 {
3031 	FILE *fp;
3032 	int cpu_num;
3033 	int retval;
3034 
3035 	fp = fopen_or_die(proc_stat, "r");
3036 
3037 	retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
3038 	if (retval != 0)
3039 		err(1, "%s: failed to parse format", proc_stat);
3040 
3041 	while (1) {
3042 		retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
3043 		if (retval != 1)
3044 			break;
3045 
3046 		retval = func(cpu_num);
3047 		if (retval) {
3048 			fclose(fp);
3049 			return (retval);
3050 		}
3051 	}
3052 	fclose(fp);
3053 	return 0;
3054 }
3055 
3056 void re_initialize(void)
3057 {
3058 	free_all_buffers();
3059 	setup_all_buffers();
3060 	fprintf(outf, "turbostat: re-initialized with num_cpus %d\n", topo.num_cpus);
3061 }
3062 
3063 void set_max_cpu_num(void)
3064 {
3065 	FILE *filep;
3066 	int base_cpu;
3067 	unsigned long dummy;
3068 	char pathname[64];
3069 
3070 	base_cpu = sched_getcpu();
3071 	if (base_cpu < 0)
3072 		err(1, "cannot find calling cpu ID");
3073 	sprintf(pathname, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings", base_cpu);
3074 
3075 	filep = fopen_or_die(pathname, "r");
3076 	topo.max_cpu_num = 0;
3077 	while (fscanf(filep, "%lx,", &dummy) == 1)
3078 		topo.max_cpu_num += BITMASK_SIZE;
3079 	fclose(filep);
3080 	topo.max_cpu_num--;	/* 0 based */
3081 }
3082 
3083 /*
3084  * count_cpus()
3085  * remember the last one seen, it will be the max
3086  */
3087 int count_cpus(int cpu)
3088 {
3089 	UNUSED(cpu);
3090 
3091 	topo.num_cpus++;
3092 	return 0;
3093 }
3094 
3095 int mark_cpu_present(int cpu)
3096 {
3097 	CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
3098 	return 0;
3099 }
3100 
3101 int init_thread_id(int cpu)
3102 {
3103 	cpus[cpu].thread_id = -1;
3104 	return 0;
3105 }
3106 
3107 /*
3108  * snapshot_proc_interrupts()
3109  *
3110  * read and record summary of /proc/interrupts
3111  *
3112  * return 1 if config change requires a restart, else return 0
3113  */
3114 int snapshot_proc_interrupts(void)
3115 {
3116 	static FILE *fp;
3117 	int column, retval;
3118 
3119 	if (fp == NULL)
3120 		fp = fopen_or_die("/proc/interrupts", "r");
3121 	else
3122 		rewind(fp);
3123 
3124 	/* read 1st line of /proc/interrupts to get cpu* name for each column */
3125 	for (column = 0; column < topo.num_cpus; ++column) {
3126 		int cpu_number;
3127 
3128 		retval = fscanf(fp, " CPU%d", &cpu_number);
3129 		if (retval != 1)
3130 			break;
3131 
3132 		if (cpu_number > topo.max_cpu_num) {
3133 			warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num);
3134 			return 1;
3135 		}
3136 
3137 		irq_column_2_cpu[column] = cpu_number;
3138 		irqs_per_cpu[cpu_number] = 0;
3139 	}
3140 
3141 	/* read /proc/interrupt count lines and sum up irqs per cpu */
3142 	while (1) {
3143 		int column;
3144 		char buf[64];
3145 
3146 		retval = fscanf(fp, " %s:", buf);	/* flush irq# "N:" */
3147 		if (retval != 1)
3148 			break;
3149 
3150 		/* read the count per cpu */
3151 		for (column = 0; column < topo.num_cpus; ++column) {
3152 
3153 			int cpu_number, irq_count;
3154 
3155 			retval = fscanf(fp, " %d", &irq_count);
3156 			if (retval != 1)
3157 				break;
3158 
3159 			cpu_number = irq_column_2_cpu[column];
3160 			irqs_per_cpu[cpu_number] += irq_count;
3161 
3162 		}
3163 
3164 		while (getc(fp) != '\n') ;	/* flush interrupt description */
3165 
3166 	}
3167 	return 0;
3168 }
3169 
3170 /*
3171  * snapshot_gfx_rc6_ms()
3172  *
3173  * record snapshot of
3174  * /sys/class/drm/card0/power/rc6_residency_ms
3175  *
3176  * return 1 if config change requires a restart, else return 0
3177  */
3178 int snapshot_gfx_rc6_ms(void)
3179 {
3180 	FILE *fp;
3181 	int retval;
3182 
3183 	fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r");
3184 
3185 	retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms);
3186 	if (retval != 1)
3187 		err(1, "GFX rc6");
3188 
3189 	fclose(fp);
3190 
3191 	return 0;
3192 }
3193 
3194 /*
3195  * snapshot_gfx_mhz()
3196  *
3197  * record snapshot of
3198  * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz
3199  *
3200  * return 1 if config change requires a restart, else return 0
3201  */
3202 int snapshot_gfx_mhz(void)
3203 {
3204 	static FILE *fp;
3205 	int retval;
3206 
3207 	if (fp == NULL)
3208 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r");
3209 	else {
3210 		rewind(fp);
3211 		fflush(fp);
3212 	}
3213 
3214 	retval = fscanf(fp, "%d", &gfx_cur_mhz);
3215 	if (retval != 1)
3216 		err(1, "GFX MHz");
3217 
3218 	return 0;
3219 }
3220 
3221 /*
3222  * snapshot_gfx_cur_mhz()
3223  *
3224  * record snapshot of
3225  * /sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz
3226  *
3227  * return 1 if config change requires a restart, else return 0
3228  */
3229 int snapshot_gfx_act_mhz(void)
3230 {
3231 	static FILE *fp;
3232 	int retval;
3233 
3234 	if (fp == NULL)
3235 		fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", "r");
3236 	else {
3237 		rewind(fp);
3238 		fflush(fp);
3239 	}
3240 
3241 	retval = fscanf(fp, "%d", &gfx_act_mhz);
3242 	if (retval != 1)
3243 		err(1, "GFX ACT MHz");
3244 
3245 	return 0;
3246 }
3247 
3248 /*
3249  * snapshot_cpu_lpi()
3250  *
3251  * record snapshot of
3252  * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us
3253  */
3254 int snapshot_cpu_lpi_us(void)
3255 {
3256 	FILE *fp;
3257 	int retval;
3258 
3259 	fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r");
3260 
3261 	retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us);
3262 	if (retval != 1) {
3263 		fprintf(stderr, "Disabling Low Power Idle CPU output\n");
3264 		BIC_NOT_PRESENT(BIC_CPU_LPI);
3265 		fclose(fp);
3266 		return -1;
3267 	}
3268 
3269 	fclose(fp);
3270 
3271 	return 0;
3272 }
3273 
3274 /*
3275  * snapshot_sys_lpi()
3276  *
3277  * record snapshot of sys_lpi_file
3278  */
3279 int snapshot_sys_lpi_us(void)
3280 {
3281 	FILE *fp;
3282 	int retval;
3283 
3284 	fp = fopen_or_die(sys_lpi_file, "r");
3285 
3286 	retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us);
3287 	if (retval != 1) {
3288 		fprintf(stderr, "Disabling Low Power Idle System output\n");
3289 		BIC_NOT_PRESENT(BIC_SYS_LPI);
3290 		fclose(fp);
3291 		return -1;
3292 	}
3293 	fclose(fp);
3294 
3295 	return 0;
3296 }
3297 
3298 /*
3299  * snapshot /proc and /sys files
3300  *
3301  * return 1 if configuration restart needed, else return 0
3302  */
3303 int snapshot_proc_sysfs_files(void)
3304 {
3305 	if (DO_BIC(BIC_IRQ))
3306 		if (snapshot_proc_interrupts())
3307 			return 1;
3308 
3309 	if (DO_BIC(BIC_GFX_rc6))
3310 		snapshot_gfx_rc6_ms();
3311 
3312 	if (DO_BIC(BIC_GFXMHz))
3313 		snapshot_gfx_mhz();
3314 
3315 	if (DO_BIC(BIC_GFXACTMHz))
3316 		snapshot_gfx_act_mhz();
3317 
3318 	if (DO_BIC(BIC_CPU_LPI))
3319 		snapshot_cpu_lpi_us();
3320 
3321 	if (DO_BIC(BIC_SYS_LPI))
3322 		snapshot_sys_lpi_us();
3323 
3324 	return 0;
3325 }
3326 
3327 int exit_requested;
3328 
3329 static void signal_handler(int signal)
3330 {
3331 	switch (signal) {
3332 	case SIGINT:
3333 		exit_requested = 1;
3334 		if (debug)
3335 			fprintf(stderr, " SIGINT\n");
3336 		break;
3337 	case SIGUSR1:
3338 		if (debug > 1)
3339 			fprintf(stderr, "SIGUSR1\n");
3340 		break;
3341 	}
3342 }
3343 
3344 void setup_signal_handler(void)
3345 {
3346 	struct sigaction sa;
3347 
3348 	memset(&sa, 0, sizeof(sa));
3349 
3350 	sa.sa_handler = &signal_handler;
3351 
3352 	if (sigaction(SIGINT, &sa, NULL) < 0)
3353 		err(1, "sigaction SIGINT");
3354 	if (sigaction(SIGUSR1, &sa, NULL) < 0)
3355 		err(1, "sigaction SIGUSR1");
3356 }
3357 
3358 void do_sleep(void)
3359 {
3360 	struct timeval tout;
3361 	struct timespec rest;
3362 	fd_set readfds;
3363 	int retval;
3364 
3365 	FD_ZERO(&readfds);
3366 	FD_SET(0, &readfds);
3367 
3368 	if (ignore_stdin) {
3369 		nanosleep(&interval_ts, NULL);
3370 		return;
3371 	}
3372 
3373 	tout = interval_tv;
3374 	retval = select(1, &readfds, NULL, NULL, &tout);
3375 
3376 	if (retval == 1) {
3377 		switch (getc(stdin)) {
3378 		case 'q':
3379 			exit_requested = 1;
3380 			break;
3381 		case EOF:
3382 			/*
3383 			 * 'stdin' is a pipe closed on the other end. There
3384 			 * won't be any further input.
3385 			 */
3386 			ignore_stdin = 1;
3387 			/* Sleep the rest of the time */
3388 			rest.tv_sec = (tout.tv_sec + tout.tv_usec / 1000000);
3389 			rest.tv_nsec = (tout.tv_usec % 1000000) * 1000;
3390 			nanosleep(&rest, NULL);
3391 		}
3392 	}
3393 }
3394 
3395 int get_msr_sum(int cpu, off_t offset, unsigned long long *msr)
3396 {
3397 	int ret, idx;
3398 	unsigned long long msr_cur, msr_last;
3399 
3400 	if (!per_cpu_msr_sum)
3401 		return 1;
3402 
3403 	idx = offset_to_idx(offset);
3404 	if (idx < 0)
3405 		return idx;
3406 	/* get_msr_sum() = sum + (get_msr() - last) */
3407 	ret = get_msr(cpu, offset, &msr_cur);
3408 	if (ret)
3409 		return ret;
3410 	msr_last = per_cpu_msr_sum[cpu].entries[idx].last;
3411 	DELTA_WRAP32(msr_cur, msr_last);
3412 	*msr = msr_last + per_cpu_msr_sum[cpu].entries[idx].sum;
3413 
3414 	return 0;
3415 }
3416 
3417 timer_t timerid;
3418 
3419 /* Timer callback, update the sum of MSRs periodically. */
3420 static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg_data *p)
3421 {
3422 	int i, ret;
3423 	int cpu = t->cpu_id;
3424 
3425 	UNUSED(c);
3426 	UNUSED(p);
3427 
3428 	for (i = IDX_PKG_ENERGY; i < IDX_COUNT; i++) {
3429 		unsigned long long msr_cur, msr_last;
3430 		off_t offset;
3431 
3432 		if (!idx_valid(i))
3433 			continue;
3434 		offset = idx_to_offset(i);
3435 		if (offset < 0)
3436 			continue;
3437 		ret = get_msr(cpu, offset, &msr_cur);
3438 		if (ret) {
3439 			fprintf(outf, "Can not update msr(0x%llx)\n", (unsigned long long)offset);
3440 			continue;
3441 		}
3442 
3443 		msr_last = per_cpu_msr_sum[cpu].entries[i].last;
3444 		per_cpu_msr_sum[cpu].entries[i].last = msr_cur & 0xffffffff;
3445 
3446 		DELTA_WRAP32(msr_cur, msr_last);
3447 		per_cpu_msr_sum[cpu].entries[i].sum += msr_last;
3448 	}
3449 	return 0;
3450 }
3451 
3452 static void msr_record_handler(union sigval v)
3453 {
3454 	UNUSED(v);
3455 
3456 	for_all_cpus(update_msr_sum, EVEN_COUNTERS);
3457 }
3458 
3459 void msr_sum_record(void)
3460 {
3461 	struct itimerspec its;
3462 	struct sigevent sev;
3463 
3464 	per_cpu_msr_sum = calloc(topo.max_cpu_num + 1, sizeof(struct msr_sum_array));
3465 	if (!per_cpu_msr_sum) {
3466 		fprintf(outf, "Can not allocate memory for long time MSR.\n");
3467 		return;
3468 	}
3469 	/*
3470 	 * Signal handler might be restricted, so use thread notifier instead.
3471 	 */
3472 	memset(&sev, 0, sizeof(struct sigevent));
3473 	sev.sigev_notify = SIGEV_THREAD;
3474 	sev.sigev_notify_function = msr_record_handler;
3475 
3476 	sev.sigev_value.sival_ptr = &timerid;
3477 	if (timer_create(CLOCK_REALTIME, &sev, &timerid) == -1) {
3478 		fprintf(outf, "Can not create timer.\n");
3479 		goto release_msr;
3480 	}
3481 
3482 	its.it_value.tv_sec = 0;
3483 	its.it_value.tv_nsec = 1;
3484 	/*
3485 	 * A wraparound time has been calculated early.
3486 	 * Some sources state that the peak power for a
3487 	 * microprocessor is usually 1.5 times the TDP rating,
3488 	 * use 2 * TDP for safety.
3489 	 */
3490 	its.it_interval.tv_sec = rapl_joule_counter_range / 2;
3491 	its.it_interval.tv_nsec = 0;
3492 
3493 	if (timer_settime(timerid, 0, &its, NULL) == -1) {
3494 		fprintf(outf, "Can not set timer.\n");
3495 		goto release_timer;
3496 	}
3497 	return;
3498 
3499 release_timer:
3500 	timer_delete(timerid);
3501 release_msr:
3502 	free(per_cpu_msr_sum);
3503 }
3504 
3505 /*
3506  * set_my_sched_priority(pri)
3507  * return previous
3508  */
3509 int set_my_sched_priority(int priority)
3510 {
3511 	int retval;
3512 	int original_priority;
3513 
3514 	errno = 0;
3515 	original_priority = getpriority(PRIO_PROCESS, 0);
3516 	if (errno && (original_priority == -1))
3517 		err(errno, "getpriority");
3518 
3519 	retval = setpriority(PRIO_PROCESS, 0, priority);
3520 	if (retval)
3521 		errx(retval, "capget(CAP_SYS_NICE) failed,try \"# setcap cap_sys_nice=ep %s\"", progname);
3522 
3523 	errno = 0;
3524 	retval = getpriority(PRIO_PROCESS, 0);
3525 	if (retval != priority)
3526 		err(retval, "getpriority(%d) != setpriority(%d)", retval, priority);
3527 
3528 	return original_priority;
3529 }
3530 
3531 void turbostat_loop()
3532 {
3533 	int retval;
3534 	int restarted = 0;
3535 	unsigned int done_iters = 0;
3536 
3537 	setup_signal_handler();
3538 
3539 	/*
3540 	 * elevate own priority for interval mode
3541 	 */
3542 	set_my_sched_priority(-20);
3543 
3544 restart:
3545 	restarted++;
3546 
3547 	snapshot_proc_sysfs_files();
3548 	retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3549 	first_counter_read = 0;
3550 	if (retval < -1) {
3551 		exit(retval);
3552 	} else if (retval == -1) {
3553 		if (restarted > 10) {
3554 			exit(retval);
3555 		}
3556 		re_initialize();
3557 		goto restart;
3558 	}
3559 	restarted = 0;
3560 	done_iters = 0;
3561 	gettimeofday(&tv_even, (struct timezone *)NULL);
3562 
3563 	while (1) {
3564 		if (for_all_proc_cpus(cpu_is_not_present)) {
3565 			re_initialize();
3566 			goto restart;
3567 		}
3568 		do_sleep();
3569 		if (snapshot_proc_sysfs_files())
3570 			goto restart;
3571 		retval = for_all_cpus(get_counters, ODD_COUNTERS);
3572 		if (retval < -1) {
3573 			exit(retval);
3574 		} else if (retval == -1) {
3575 			re_initialize();
3576 			goto restart;
3577 		}
3578 		gettimeofday(&tv_odd, (struct timezone *)NULL);
3579 		timersub(&tv_odd, &tv_even, &tv_delta);
3580 		if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) {
3581 			re_initialize();
3582 			goto restart;
3583 		}
3584 		compute_average(EVEN_COUNTERS);
3585 		format_all_counters(EVEN_COUNTERS);
3586 		flush_output_stdout();
3587 		if (exit_requested)
3588 			break;
3589 		if (num_iterations && ++done_iters >= num_iterations)
3590 			break;
3591 		do_sleep();
3592 		if (snapshot_proc_sysfs_files())
3593 			goto restart;
3594 		retval = for_all_cpus(get_counters, EVEN_COUNTERS);
3595 		if (retval < -1) {
3596 			exit(retval);
3597 		} else if (retval == -1) {
3598 			re_initialize();
3599 			goto restart;
3600 		}
3601 		gettimeofday(&tv_even, (struct timezone *)NULL);
3602 		timersub(&tv_even, &tv_odd, &tv_delta);
3603 		if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) {
3604 			re_initialize();
3605 			goto restart;
3606 		}
3607 		compute_average(ODD_COUNTERS);
3608 		format_all_counters(ODD_COUNTERS);
3609 		flush_output_stdout();
3610 		if (exit_requested)
3611 			break;
3612 		if (num_iterations && ++done_iters >= num_iterations)
3613 			break;
3614 	}
3615 }
3616 
3617 void check_dev_msr()
3618 {
3619 	struct stat sb;
3620 	char pathname[32];
3621 
3622 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3623 	if (stat(pathname, &sb))
3624 		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
3625 			err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" ");
3626 }
3627 
3628 /*
3629  * check for CAP_SYS_RAWIO
3630  * return 0 on success
3631  * return 1 on fail
3632  */
3633 int check_for_cap_sys_rawio(void)
3634 {
3635 	cap_t caps;
3636 	cap_flag_value_t cap_flag_value;
3637 
3638 	caps = cap_get_proc();
3639 	if (caps == NULL)
3640 		err(-6, "cap_get_proc\n");
3641 
3642 	if (cap_get_flag(caps, CAP_SYS_RAWIO, CAP_EFFECTIVE, &cap_flag_value))
3643 		err(-6, "cap_get\n");
3644 
3645 	if (cap_flag_value != CAP_SET) {
3646 		warnx("capget(CAP_SYS_RAWIO) failed," " try \"# setcap cap_sys_rawio=ep %s\"", progname);
3647 		return 1;
3648 	}
3649 
3650 	if (cap_free(caps) == -1)
3651 		err(-6, "cap_free\n");
3652 
3653 	return 0;
3654 }
3655 
3656 void check_permissions(void)
3657 {
3658 	int do_exit = 0;
3659 	char pathname[32];
3660 
3661 	/* check for CAP_SYS_RAWIO */
3662 	do_exit += check_for_cap_sys_rawio();
3663 
3664 	/* test file permissions */
3665 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
3666 	if (euidaccess(pathname, R_OK)) {
3667 		do_exit++;
3668 		warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr");
3669 	}
3670 
3671 	/* if all else fails, thell them to be root */
3672 	if (do_exit)
3673 		if (getuid() != 0)
3674 			warnx("... or simply run as root");
3675 
3676 	if (do_exit)
3677 		exit(-6);
3678 }
3679 
3680 /*
3681  * NHM adds support for additional MSRs:
3682  *
3683  * MSR_SMI_COUNT                   0x00000034
3684  *
3685  * MSR_PLATFORM_INFO               0x000000ce
3686  * MSR_PKG_CST_CONFIG_CONTROL     0x000000e2
3687  *
3688  * MSR_MISC_PWR_MGMT               0x000001aa
3689  *
3690  * MSR_PKG_C3_RESIDENCY            0x000003f8
3691  * MSR_PKG_C6_RESIDENCY            0x000003f9
3692  * MSR_CORE_C3_RESIDENCY           0x000003fc
3693  * MSR_CORE_C6_RESIDENCY           0x000003fd
3694  *
3695  * Side effect:
3696  * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL
3697  * sets has_misc_feature_control
3698  */
3699 int probe_nhm_msrs(unsigned int family, unsigned int model)
3700 {
3701 	unsigned long long msr;
3702 	unsigned int base_ratio;
3703 	int *pkg_cstate_limits;
3704 
3705 	if (!genuine_intel)
3706 		return 0;
3707 
3708 	if (family != 6)
3709 		return 0;
3710 
3711 	bclk = discover_bclk(family, model);
3712 
3713 	switch (model) {
3714 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
3715 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3716 		pkg_cstate_limits = nhm_pkg_cstate_limits;
3717 		break;
3718 	case INTEL_FAM6_SANDYBRIDGE:	/* SNB */
3719 	case INTEL_FAM6_SANDYBRIDGE_X:	/* SNB Xeon */
3720 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
3721 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3722 		pkg_cstate_limits = snb_pkg_cstate_limits;
3723 		has_misc_feature_control = 1;
3724 		break;
3725 	case INTEL_FAM6_HASWELL:	/* HSW */
3726 	case INTEL_FAM6_HASWELL_G:	/* HSW */
3727 	case INTEL_FAM6_HASWELL_X:	/* HSX */
3728 	case INTEL_FAM6_HASWELL_L:	/* HSW */
3729 	case INTEL_FAM6_BROADWELL:	/* BDW */
3730 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
3731 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
3732 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
3733 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
3734 		pkg_cstate_limits = hsw_pkg_cstate_limits;
3735 		has_misc_feature_control = 1;
3736 		break;
3737 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
3738 	case INTEL_FAM6_SAPPHIRERAPIDS_X:	/* SPR */
3739 		pkg_cstate_limits = skx_pkg_cstate_limits;
3740 		has_misc_feature_control = 1;
3741 		break;
3742 	case INTEL_FAM6_ICELAKE_X:	/* ICX */
3743 		pkg_cstate_limits = icx_pkg_cstate_limits;
3744 		has_misc_feature_control = 1;
3745 		break;
3746 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
3747 		no_MSR_MISC_PWR_MGMT = 1;
3748 		/* FALLTHRU */
3749 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
3750 		pkg_cstate_limits = slv_pkg_cstate_limits;
3751 		break;
3752 	case INTEL_FAM6_ATOM_AIRMONT:	/* AMT */
3753 		pkg_cstate_limits = amt_pkg_cstate_limits;
3754 		no_MSR_MISC_PWR_MGMT = 1;
3755 		break;
3756 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI */
3757 		pkg_cstate_limits = phi_pkg_cstate_limits;
3758 		break;
3759 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
3760 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
3761 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
3762 	case INTEL_FAM6_ATOM_TREMONT:	/* EHL */
3763 	case INTEL_FAM6_ATOM_TREMONT_D:	/* JVL */
3764 		pkg_cstate_limits = glm_pkg_cstate_limits;
3765 		break;
3766 	default:
3767 		return 0;
3768 	}
3769 	get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr);
3770 	pkg_cstate_limit = pkg_cstate_limits[msr & 0xF];
3771 
3772 	get_msr(base_cpu, MSR_PLATFORM_INFO, &msr);
3773 	base_ratio = (msr >> 8) & 0xFF;
3774 
3775 	base_hz = base_ratio * bclk * 1000000;
3776 	has_base_hz = 1;
3777 	return 1;
3778 }
3779 
3780 /*
3781  * SLV client has support for unique MSRs:
3782  *
3783  * MSR_CC6_DEMOTION_POLICY_CONFIG
3784  * MSR_MC6_DEMOTION_POLICY_CONFIG
3785  */
3786 
3787 int has_slv_msrs(unsigned int family, unsigned int model)
3788 {
3789 	if (!genuine_intel)
3790 		return 0;
3791 
3792 	if (family != 6)
3793 		return 0;
3794 
3795 	switch (model) {
3796 	case INTEL_FAM6_ATOM_SILVERMONT:
3797 	case INTEL_FAM6_ATOM_SILVERMONT_MID:
3798 	case INTEL_FAM6_ATOM_AIRMONT_MID:
3799 		return 1;
3800 	}
3801 	return 0;
3802 }
3803 
3804 int is_dnv(unsigned int family, unsigned int model)
3805 {
3806 
3807 	if (!genuine_intel)
3808 		return 0;
3809 
3810 	if (family != 6)
3811 		return 0;
3812 
3813 	switch (model) {
3814 	case INTEL_FAM6_ATOM_GOLDMONT_D:
3815 		return 1;
3816 	}
3817 	return 0;
3818 }
3819 
3820 int is_bdx(unsigned int family, unsigned int model)
3821 {
3822 
3823 	if (!genuine_intel)
3824 		return 0;
3825 
3826 	if (family != 6)
3827 		return 0;
3828 
3829 	switch (model) {
3830 	case INTEL_FAM6_BROADWELL_X:
3831 		return 1;
3832 	}
3833 	return 0;
3834 }
3835 
3836 int is_skx(unsigned int family, unsigned int model)
3837 {
3838 
3839 	if (!genuine_intel)
3840 		return 0;
3841 
3842 	if (family != 6)
3843 		return 0;
3844 
3845 	switch (model) {
3846 	case INTEL_FAM6_SKYLAKE_X:
3847 		return 1;
3848 	}
3849 	return 0;
3850 }
3851 
3852 int is_icx(unsigned int family, unsigned int model)
3853 {
3854 
3855 	if (!genuine_intel)
3856 		return 0;
3857 
3858 	if (family != 6)
3859 		return 0;
3860 
3861 	switch (model) {
3862 	case INTEL_FAM6_ICELAKE_X:
3863 		return 1;
3864 	}
3865 	return 0;
3866 }
3867 
3868 int is_spr(unsigned int family, unsigned int model)
3869 {
3870 
3871 	if (!genuine_intel)
3872 		return 0;
3873 
3874 	if (family != 6)
3875 		return 0;
3876 
3877 	switch (model) {
3878 	case INTEL_FAM6_SAPPHIRERAPIDS_X:
3879 		return 1;
3880 	}
3881 	return 0;
3882 }
3883 
3884 int is_ehl(unsigned int family, unsigned int model)
3885 {
3886 	if (!genuine_intel)
3887 		return 0;
3888 
3889 	if (family != 6)
3890 		return 0;
3891 
3892 	switch (model) {
3893 	case INTEL_FAM6_ATOM_TREMONT:
3894 		return 1;
3895 	}
3896 	return 0;
3897 }
3898 
3899 int is_jvl(unsigned int family, unsigned int model)
3900 {
3901 	if (!genuine_intel)
3902 		return 0;
3903 
3904 	if (family != 6)
3905 		return 0;
3906 
3907 	switch (model) {
3908 	case INTEL_FAM6_ATOM_TREMONT_D:
3909 		return 1;
3910 	}
3911 	return 0;
3912 }
3913 
3914 int has_turbo_ratio_limit(unsigned int family, unsigned int model)
3915 {
3916 	if (has_slv_msrs(family, model))
3917 		return 0;
3918 
3919 	if (family != 6)
3920 		return 0;
3921 
3922 	switch (model) {
3923 		/* Nehalem compatible, but do not include turbo-ratio limit support */
3924 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
3925 	case INTEL_FAM6_XEON_PHI_KNL:	/* PHI - Knights Landing (different MSR definition) */
3926 		return 0;
3927 	default:
3928 		return 1;
3929 	}
3930 }
3931 
3932 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model)
3933 {
3934 	if (has_slv_msrs(family, model))
3935 		return 1;
3936 
3937 	return 0;
3938 }
3939 
3940 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model)
3941 {
3942 	if (!genuine_intel)
3943 		return 0;
3944 
3945 	if (family != 6)
3946 		return 0;
3947 
3948 	switch (model) {
3949 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
3950 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3951 		return 1;
3952 	default:
3953 		return 0;
3954 	}
3955 }
3956 
3957 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model)
3958 {
3959 	if (!genuine_intel)
3960 		return 0;
3961 
3962 	if (family != 6)
3963 		return 0;
3964 
3965 	switch (model) {
3966 	case INTEL_FAM6_HASWELL_X:	/* HSW Xeon */
3967 		return 1;
3968 	default:
3969 		return 0;
3970 	}
3971 }
3972 
3973 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model)
3974 {
3975 	if (!genuine_intel)
3976 		return 0;
3977 
3978 	if (family != 6)
3979 		return 0;
3980 
3981 	switch (model) {
3982 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
3983 		return 1;
3984 	default:
3985 		return 0;
3986 	}
3987 }
3988 
3989 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model)
3990 {
3991 	if (!genuine_intel)
3992 		return 0;
3993 
3994 	if (family != 6)
3995 		return 0;
3996 
3997 	switch (model) {
3998 	case INTEL_FAM6_ATOM_GOLDMONT:
3999 	case INTEL_FAM6_SKYLAKE_X:
4000 	case INTEL_FAM6_ICELAKE_X:
4001 	case INTEL_FAM6_SAPPHIRERAPIDS_X:
4002 		return 1;
4003 	default:
4004 		return 0;
4005 	}
4006 }
4007 
4008 int has_config_tdp(unsigned int family, unsigned int model)
4009 {
4010 	if (!genuine_intel)
4011 		return 0;
4012 
4013 	if (family != 6)
4014 		return 0;
4015 
4016 	switch (model) {
4017 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
4018 	case INTEL_FAM6_HASWELL:	/* HSW */
4019 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4020 	case INTEL_FAM6_HASWELL_L:	/* HSW */
4021 	case INTEL_FAM6_HASWELL_G:	/* HSW */
4022 	case INTEL_FAM6_BROADWELL:	/* BDW */
4023 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
4024 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4025 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
4026 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
4027 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4028 	case INTEL_FAM6_ICELAKE_X:	/* ICX */
4029 	case INTEL_FAM6_SAPPHIRERAPIDS_X:	/* SPR */
4030 	case INTEL_FAM6_XEON_PHI_KNL:	/* Knights Landing */
4031 		return 1;
4032 	default:
4033 		return 0;
4034 	}
4035 }
4036 
4037 /*
4038  * tcc_offset_bits:
4039  * 0: Tcc Offset not supported (Default)
4040  * 6: Bit 29:24 of MSR_PLATFORM_INFO
4041  * 4: Bit 27:24 of MSR_PLATFORM_INFO
4042  */
4043 void check_tcc_offset(int model)
4044 {
4045 	unsigned long long msr;
4046 
4047 	if (!genuine_intel)
4048 		return;
4049 
4050 	switch (model) {
4051 	case INTEL_FAM6_SKYLAKE_L:
4052 	case INTEL_FAM6_SKYLAKE:
4053 	case INTEL_FAM6_KABYLAKE_L:
4054 	case INTEL_FAM6_KABYLAKE:
4055 	case INTEL_FAM6_ICELAKE_L:
4056 	case INTEL_FAM6_ICELAKE:
4057 	case INTEL_FAM6_TIGERLAKE_L:
4058 	case INTEL_FAM6_TIGERLAKE:
4059 	case INTEL_FAM6_COMETLAKE:
4060 		if (!get_msr(base_cpu, MSR_PLATFORM_INFO, &msr)) {
4061 			msr = (msr >> 30) & 1;
4062 			if (msr)
4063 				tcc_offset_bits = 6;
4064 		}
4065 		return;
4066 	default:
4067 		return;
4068 	}
4069 }
4070 
4071 static void remove_underbar(char *s)
4072 {
4073 	char *to = s;
4074 
4075 	while (*s) {
4076 		if (*s != '_')
4077 			*to++ = *s;
4078 		s++;
4079 	}
4080 
4081 	*to = 0;
4082 }
4083 
4084 static void dump_turbo_ratio_info(unsigned int family, unsigned int model)
4085 {
4086 	if (!has_turbo)
4087 		return;
4088 
4089 	if (has_hsw_turbo_ratio_limit(family, model))
4090 		dump_hsw_turbo_ratio_limits();
4091 
4092 	if (has_ivt_turbo_ratio_limit(family, model))
4093 		dump_ivt_turbo_ratio_limits();
4094 
4095 	if (has_turbo_ratio_limit(family, model)) {
4096 		dump_turbo_ratio_limits(MSR_TURBO_RATIO_LIMIT, family, model);
4097 
4098 		if (is_hybrid)
4099 			dump_turbo_ratio_limits(MSR_SECONDARY_TURBO_RATIO_LIMIT, family, model);
4100 	}
4101 
4102 	if (has_atom_turbo_ratio_limit(family, model))
4103 		dump_atom_turbo_ratio_limits();
4104 
4105 	if (has_knl_turbo_ratio_limit(family, model))
4106 		dump_knl_turbo_ratio_limits();
4107 
4108 	if (has_config_tdp(family, model))
4109 		dump_config_tdp();
4110 }
4111 
4112 static void dump_cstate_pstate_config_info(unsigned int family, unsigned int model)
4113 {
4114 	if (!do_nhm_platform_info)
4115 		return;
4116 
4117 	dump_nhm_platform_info();
4118 	dump_turbo_ratio_info(family, model);
4119 	dump_nhm_cst_cfg();
4120 }
4121 
4122 static int read_sysfs_int(char *path)
4123 {
4124 	FILE *input;
4125 	int retval = -1;
4126 
4127 	input = fopen(path, "r");
4128 	if (input == NULL) {
4129 		if (debug)
4130 			fprintf(outf, "NSFOD %s\n", path);
4131 		return (-1);
4132 	}
4133 	if (fscanf(input, "%d", &retval) != 1)
4134 		err(1, "%s: failed to read int from file", path);
4135 	fclose(input);
4136 
4137 	return (retval);
4138 }
4139 
4140 static void dump_sysfs_file(char *path)
4141 {
4142 	FILE *input;
4143 	char cpuidle_buf[64];
4144 
4145 	input = fopen(path, "r");
4146 	if (input == NULL) {
4147 		if (debug)
4148 			fprintf(outf, "NSFOD %s\n", path);
4149 		return;
4150 	}
4151 	if (!fgets(cpuidle_buf, sizeof(cpuidle_buf), input))
4152 		err(1, "%s: failed to read file", path);
4153 	fclose(input);
4154 
4155 	fprintf(outf, "%s: %s", strrchr(path, '/') + 1, cpuidle_buf);
4156 }
4157 
4158 static void intel_uncore_frequency_probe(void)
4159 {
4160 	int i, j;
4161 	char path[128];
4162 
4163 	if (!genuine_intel)
4164 		return;
4165 
4166 	if (access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00", R_OK))
4167 		return;
4168 
4169 	if (!access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00/current_freq_khz", R_OK))
4170 		BIC_PRESENT(BIC_UNCORE_MHZ);
4171 
4172 	if (quiet)
4173 		return;
4174 
4175 	for (i = 0; i < topo.num_packages; ++i) {
4176 		for (j = 0; j < topo.num_die; ++j) {
4177 			int k, l;
4178 
4179 			sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/min_freq_khz",
4180 				i, j);
4181 			k = read_sysfs_int(path);
4182 			sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/max_freq_khz",
4183 				i, j);
4184 			l = read_sysfs_int(path);
4185 			fprintf(outf, "Uncore Frequency pkg%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000);
4186 
4187 			sprintf(path,
4188 				"/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_min_freq_khz",
4189 				i, j);
4190 			k = read_sysfs_int(path);
4191 			sprintf(path,
4192 				"/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_max_freq_khz",
4193 				i, j);
4194 			l = read_sysfs_int(path);
4195 			fprintf(outf, "(%d - %d MHz)\n", k / 1000, l / 1000);
4196 		}
4197 	}
4198 }
4199 
4200 static void dump_sysfs_cstate_config(void)
4201 {
4202 	char path[64];
4203 	char name_buf[16];
4204 	char desc[64];
4205 	FILE *input;
4206 	int state;
4207 	char *sp;
4208 
4209 	if (access("/sys/devices/system/cpu/cpuidle", R_OK)) {
4210 		fprintf(outf, "cpuidle not loaded\n");
4211 		return;
4212 	}
4213 
4214 	dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_driver");
4215 	dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_governor");
4216 	dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_governor_ro");
4217 
4218 	for (state = 0; state < 10; ++state) {
4219 
4220 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state);
4221 		input = fopen(path, "r");
4222 		if (input == NULL)
4223 			continue;
4224 		if (!fgets(name_buf, sizeof(name_buf), input))
4225 			err(1, "%s: failed to read file", path);
4226 
4227 		/* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
4228 		sp = strchr(name_buf, '-');
4229 		if (!sp)
4230 			sp = strchrnul(name_buf, '\n');
4231 		*sp = '\0';
4232 		fclose(input);
4233 
4234 		remove_underbar(name_buf);
4235 
4236 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc", base_cpu, state);
4237 		input = fopen(path, "r");
4238 		if (input == NULL)
4239 			continue;
4240 		if (!fgets(desc, sizeof(desc), input))
4241 			err(1, "%s: failed to read file", path);
4242 
4243 		fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc);
4244 		fclose(input);
4245 	}
4246 }
4247 
4248 static void dump_sysfs_pstate_config(void)
4249 {
4250 	char path[64];
4251 	char driver_buf[64];
4252 	char governor_buf[64];
4253 	FILE *input;
4254 	int turbo;
4255 
4256 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver", base_cpu);
4257 	input = fopen(path, "r");
4258 	if (input == NULL) {
4259 		fprintf(outf, "NSFOD %s\n", path);
4260 		return;
4261 	}
4262 	if (!fgets(driver_buf, sizeof(driver_buf), input))
4263 		err(1, "%s: failed to read file", path);
4264 	fclose(input);
4265 
4266 	sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor", base_cpu);
4267 	input = fopen(path, "r");
4268 	if (input == NULL) {
4269 		fprintf(outf, "NSFOD %s\n", path);
4270 		return;
4271 	}
4272 	if (!fgets(governor_buf, sizeof(governor_buf), input))
4273 		err(1, "%s: failed to read file", path);
4274 	fclose(input);
4275 
4276 	fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf);
4277 	fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf);
4278 
4279 	sprintf(path, "/sys/devices/system/cpu/cpufreq/boost");
4280 	input = fopen(path, "r");
4281 	if (input != NULL) {
4282 		if (fscanf(input, "%d", &turbo) != 1)
4283 			err(1, "%s: failed to parse number from file", path);
4284 		fprintf(outf, "cpufreq boost: %d\n", turbo);
4285 		fclose(input);
4286 	}
4287 
4288 	sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo");
4289 	input = fopen(path, "r");
4290 	if (input != NULL) {
4291 		if (fscanf(input, "%d", &turbo) != 1)
4292 			err(1, "%s: failed to parse number from file", path);
4293 		fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo);
4294 		fclose(input);
4295 	}
4296 }
4297 
4298 /*
4299  * print_epb()
4300  * Decode the ENERGY_PERF_BIAS MSR
4301  */
4302 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4303 {
4304 	char *epb_string;
4305 	int cpu, epb;
4306 
4307 	UNUSED(c);
4308 	UNUSED(p);
4309 
4310 	if (!has_epb)
4311 		return 0;
4312 
4313 	cpu = t->cpu_id;
4314 
4315 	/* EPB is per-package */
4316 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4317 		return 0;
4318 
4319 	if (cpu_migrate(cpu)) {
4320 		fprintf(outf, "print_epb: Could not migrate to CPU %d\n", cpu);
4321 		return -1;
4322 	}
4323 
4324 	epb = get_epb(cpu);
4325 	if (epb < 0)
4326 		return 0;
4327 
4328 	switch (epb) {
4329 	case ENERGY_PERF_BIAS_PERFORMANCE:
4330 		epb_string = "performance";
4331 		break;
4332 	case ENERGY_PERF_BIAS_NORMAL:
4333 		epb_string = "balanced";
4334 		break;
4335 	case ENERGY_PERF_BIAS_POWERSAVE:
4336 		epb_string = "powersave";
4337 		break;
4338 	default:
4339 		epb_string = "custom";
4340 		break;
4341 	}
4342 	fprintf(outf, "cpu%d: EPB: %d (%s)\n", cpu, epb, epb_string);
4343 
4344 	return 0;
4345 }
4346 
4347 /*
4348  * print_hwp()
4349  * Decode the MSR_HWP_CAPABILITIES
4350  */
4351 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4352 {
4353 	unsigned long long msr;
4354 	int cpu;
4355 
4356 	UNUSED(c);
4357 	UNUSED(p);
4358 
4359 	if (!has_hwp)
4360 		return 0;
4361 
4362 	cpu = t->cpu_id;
4363 
4364 	/* MSR_HWP_CAPABILITIES is per-package */
4365 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4366 		return 0;
4367 
4368 	if (cpu_migrate(cpu)) {
4369 		fprintf(outf, "print_hwp: Could not migrate to CPU %d\n", cpu);
4370 		return -1;
4371 	}
4372 
4373 	if (get_msr(cpu, MSR_PM_ENABLE, &msr))
4374 		return 0;
4375 
4376 	fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n", cpu, msr, (msr & (1 << 0)) ? "" : "No-");
4377 
4378 	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
4379 	if ((msr & (1 << 0)) == 0)
4380 		return 0;
4381 
4382 	if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr))
4383 		return 0;
4384 
4385 	fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx "
4386 		"(high %d guar %d eff %d low %d)\n",
4387 		cpu, msr,
4388 		(unsigned int)HWP_HIGHEST_PERF(msr),
4389 		(unsigned int)HWP_GUARANTEED_PERF(msr),
4390 		(unsigned int)HWP_MOSTEFFICIENT_PERF(msr), (unsigned int)HWP_LOWEST_PERF(msr));
4391 
4392 	if (get_msr(cpu, MSR_HWP_REQUEST, &msr))
4393 		return 0;
4394 
4395 	fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx "
4396 		"(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n",
4397 		cpu, msr,
4398 		(unsigned int)(((msr) >> 0) & 0xff),
4399 		(unsigned int)(((msr) >> 8) & 0xff),
4400 		(unsigned int)(((msr) >> 16) & 0xff),
4401 		(unsigned int)(((msr) >> 24) & 0xff),
4402 		(unsigned int)(((msr) >> 32) & 0xff3), (unsigned int)(((msr) >> 42) & 0x1));
4403 
4404 	if (has_hwp_pkg) {
4405 		if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr))
4406 			return 0;
4407 
4408 		fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx "
4409 			"(min %d max %d des %d epp 0x%x window 0x%x)\n",
4410 			cpu, msr,
4411 			(unsigned int)(((msr) >> 0) & 0xff),
4412 			(unsigned int)(((msr) >> 8) & 0xff),
4413 			(unsigned int)(((msr) >> 16) & 0xff),
4414 			(unsigned int)(((msr) >> 24) & 0xff), (unsigned int)(((msr) >> 32) & 0xff3));
4415 	}
4416 	if (has_hwp_notify) {
4417 		if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr))
4418 			return 0;
4419 
4420 		fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx "
4421 			"(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n",
4422 			cpu, msr, ((msr) & 0x1) ? "EN" : "Dis", ((msr) & 0x2) ? "EN" : "Dis");
4423 	}
4424 	if (get_msr(cpu, MSR_HWP_STATUS, &msr))
4425 		return 0;
4426 
4427 	fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx "
4428 		"(%sGuaranteed_Perf_Change, %sExcursion_Min)\n",
4429 		cpu, msr, ((msr) & 0x1) ? "" : "No-", ((msr) & 0x4) ? "" : "No-");
4430 
4431 	return 0;
4432 }
4433 
4434 /*
4435  * print_perf_limit()
4436  */
4437 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4438 {
4439 	unsigned long long msr;
4440 	int cpu;
4441 
4442 	UNUSED(c);
4443 	UNUSED(p);
4444 
4445 	cpu = t->cpu_id;
4446 
4447 	/* per-package */
4448 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4449 		return 0;
4450 
4451 	if (cpu_migrate(cpu)) {
4452 		fprintf(outf, "print_perf_limit: Could not migrate to CPU %d\n", cpu);
4453 		return -1;
4454 	}
4455 
4456 	if (do_core_perf_limit_reasons) {
4457 		get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr);
4458 		fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
4459 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)",
4460 			(msr & 1 << 15) ? "bit15, " : "",
4461 			(msr & 1 << 14) ? "bit14, " : "",
4462 			(msr & 1 << 13) ? "Transitions, " : "",
4463 			(msr & 1 << 12) ? "MultiCoreTurbo, " : "",
4464 			(msr & 1 << 11) ? "PkgPwrL2, " : "",
4465 			(msr & 1 << 10) ? "PkgPwrL1, " : "",
4466 			(msr & 1 << 9) ? "CorePwr, " : "",
4467 			(msr & 1 << 8) ? "Amps, " : "",
4468 			(msr & 1 << 6) ? "VR-Therm, " : "",
4469 			(msr & 1 << 5) ? "Auto-HWP, " : "",
4470 			(msr & 1 << 4) ? "Graphics, " : "",
4471 			(msr & 1 << 2) ? "bit2, " : "",
4472 			(msr & 1 << 1) ? "ThermStatus, " : "", (msr & 1 << 0) ? "PROCHOT, " : "");
4473 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
4474 			(msr & 1 << 31) ? "bit31, " : "",
4475 			(msr & 1 << 30) ? "bit30, " : "",
4476 			(msr & 1 << 29) ? "Transitions, " : "",
4477 			(msr & 1 << 28) ? "MultiCoreTurbo, " : "",
4478 			(msr & 1 << 27) ? "PkgPwrL2, " : "",
4479 			(msr & 1 << 26) ? "PkgPwrL1, " : "",
4480 			(msr & 1 << 25) ? "CorePwr, " : "",
4481 			(msr & 1 << 24) ? "Amps, " : "",
4482 			(msr & 1 << 22) ? "VR-Therm, " : "",
4483 			(msr & 1 << 21) ? "Auto-HWP, " : "",
4484 			(msr & 1 << 20) ? "Graphics, " : "",
4485 			(msr & 1 << 18) ? "bit18, " : "",
4486 			(msr & 1 << 17) ? "ThermStatus, " : "", (msr & 1 << 16) ? "PROCHOT, " : "");
4487 
4488 	}
4489 	if (do_gfx_perf_limit_reasons) {
4490 		get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr);
4491 		fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
4492 		fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)",
4493 			(msr & 1 << 0) ? "PROCHOT, " : "",
4494 			(msr & 1 << 1) ? "ThermStatus, " : "",
4495 			(msr & 1 << 4) ? "Graphics, " : "",
4496 			(msr & 1 << 6) ? "VR-Therm, " : "",
4497 			(msr & 1 << 8) ? "Amps, " : "",
4498 			(msr & 1 << 9) ? "GFXPwr, " : "",
4499 			(msr & 1 << 10) ? "PkgPwrL1, " : "", (msr & 1 << 11) ? "PkgPwrL2, " : "");
4500 		fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n",
4501 			(msr & 1 << 16) ? "PROCHOT, " : "",
4502 			(msr & 1 << 17) ? "ThermStatus, " : "",
4503 			(msr & 1 << 20) ? "Graphics, " : "",
4504 			(msr & 1 << 22) ? "VR-Therm, " : "",
4505 			(msr & 1 << 24) ? "Amps, " : "",
4506 			(msr & 1 << 25) ? "GFXPwr, " : "",
4507 			(msr & 1 << 26) ? "PkgPwrL1, " : "", (msr & 1 << 27) ? "PkgPwrL2, " : "");
4508 	}
4509 	if (do_ring_perf_limit_reasons) {
4510 		get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr);
4511 		fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr);
4512 		fprintf(outf, " (Active: %s%s%s%s%s%s)",
4513 			(msr & 1 << 0) ? "PROCHOT, " : "",
4514 			(msr & 1 << 1) ? "ThermStatus, " : "",
4515 			(msr & 1 << 6) ? "VR-Therm, " : "",
4516 			(msr & 1 << 8) ? "Amps, " : "",
4517 			(msr & 1 << 10) ? "PkgPwrL1, " : "", (msr & 1 << 11) ? "PkgPwrL2, " : "");
4518 		fprintf(outf, " (Logged: %s%s%s%s%s%s)\n",
4519 			(msr & 1 << 16) ? "PROCHOT, " : "",
4520 			(msr & 1 << 17) ? "ThermStatus, " : "",
4521 			(msr & 1 << 22) ? "VR-Therm, " : "",
4522 			(msr & 1 << 24) ? "Amps, " : "",
4523 			(msr & 1 << 26) ? "PkgPwrL1, " : "", (msr & 1 << 27) ? "PkgPwrL2, " : "");
4524 	}
4525 	return 0;
4526 }
4527 
4528 #define	RAPL_POWER_GRANULARITY	0x7FFF	/* 15 bit power granularity */
4529 #define	RAPL_TIME_GRANULARITY	0x3F	/* 6 bit time granularity */
4530 
4531 double get_tdp_intel(unsigned int model)
4532 {
4533 	unsigned long long msr;
4534 
4535 	if (do_rapl & RAPL_PKG_POWER_INFO)
4536 		if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr))
4537 			return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
4538 
4539 	switch (model) {
4540 	case INTEL_FAM6_ATOM_SILVERMONT:
4541 	case INTEL_FAM6_ATOM_SILVERMONT_D:
4542 		return 30.0;
4543 	default:
4544 		return 135.0;
4545 	}
4546 }
4547 
4548 double get_tdp_amd(unsigned int family)
4549 {
4550 	UNUSED(family);
4551 
4552 	/* This is the max stock TDP of HEDT/Server Fam17h+ chips */
4553 	return 280.0;
4554 }
4555 
4556 /*
4557  * rapl_dram_energy_units_probe()
4558  * Energy units are either hard-coded, or come from RAPL Energy Unit MSR.
4559  */
4560 static double rapl_dram_energy_units_probe(int model, double rapl_energy_units)
4561 {
4562 	/* only called for genuine_intel, family 6 */
4563 
4564 	switch (model) {
4565 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4566 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4567 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4568 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4569 	case INTEL_FAM6_ICELAKE_X:	/* ICX */
4570 		return (rapl_dram_energy_units = 15.3 / 1000000);
4571 	default:
4572 		return (rapl_energy_units);
4573 	}
4574 }
4575 
4576 void rapl_probe_intel(unsigned int family, unsigned int model)
4577 {
4578 	unsigned long long msr;
4579 	unsigned int time_unit;
4580 	double tdp;
4581 
4582 	if (family != 6)
4583 		return;
4584 
4585 	switch (model) {
4586 	case INTEL_FAM6_SANDYBRIDGE:
4587 	case INTEL_FAM6_IVYBRIDGE:
4588 	case INTEL_FAM6_HASWELL:	/* HSW */
4589 	case INTEL_FAM6_HASWELL_L:	/* HSW */
4590 	case INTEL_FAM6_HASWELL_G:	/* HSW */
4591 	case INTEL_FAM6_BROADWELL:	/* BDW */
4592 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
4593 		do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
4594 		if (rapl_joules) {
4595 			BIC_PRESENT(BIC_Pkg_J);
4596 			BIC_PRESENT(BIC_Cor_J);
4597 			BIC_PRESENT(BIC_GFX_J);
4598 		} else {
4599 			BIC_PRESENT(BIC_PkgWatt);
4600 			BIC_PRESENT(BIC_CorWatt);
4601 			BIC_PRESENT(BIC_GFXWatt);
4602 		}
4603 		break;
4604 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
4605 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
4606 		do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO;
4607 		if (rapl_joules)
4608 			BIC_PRESENT(BIC_Pkg_J);
4609 		else
4610 			BIC_PRESENT(BIC_PkgWatt);
4611 		break;
4612 	case INTEL_FAM6_ATOM_TREMONT:	/* EHL */
4613 		do_rapl =
4614 		    RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS
4615 		    | RAPL_GFX | RAPL_PKG_POWER_INFO;
4616 		if (rapl_joules) {
4617 			BIC_PRESENT(BIC_Pkg_J);
4618 			BIC_PRESENT(BIC_Cor_J);
4619 			BIC_PRESENT(BIC_RAM_J);
4620 			BIC_PRESENT(BIC_GFX_J);
4621 		} else {
4622 			BIC_PRESENT(BIC_PkgWatt);
4623 			BIC_PRESENT(BIC_CorWatt);
4624 			BIC_PRESENT(BIC_RAMWatt);
4625 			BIC_PRESENT(BIC_GFXWatt);
4626 		}
4627 		break;
4628 	case INTEL_FAM6_ATOM_TREMONT_D:	/* JVL */
4629 		do_rapl = RAPL_PKG | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
4630 		BIC_PRESENT(BIC_PKG__);
4631 		if (rapl_joules)
4632 			BIC_PRESENT(BIC_Pkg_J);
4633 		else
4634 			BIC_PRESENT(BIC_PkgWatt);
4635 		break;
4636 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
4637 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
4638 		do_rapl =
4639 		    RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS
4640 		    | RAPL_GFX | RAPL_PKG_POWER_INFO;
4641 		BIC_PRESENT(BIC_PKG__);
4642 		BIC_PRESENT(BIC_RAM__);
4643 		if (rapl_joules) {
4644 			BIC_PRESENT(BIC_Pkg_J);
4645 			BIC_PRESENT(BIC_Cor_J);
4646 			BIC_PRESENT(BIC_RAM_J);
4647 			BIC_PRESENT(BIC_GFX_J);
4648 		} else {
4649 			BIC_PRESENT(BIC_PkgWatt);
4650 			BIC_PRESENT(BIC_CorWatt);
4651 			BIC_PRESENT(BIC_RAMWatt);
4652 			BIC_PRESENT(BIC_GFXWatt);
4653 		}
4654 		break;
4655 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4656 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
4657 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
4658 	case INTEL_FAM6_ICELAKE_X:	/* ICX */
4659 	case INTEL_FAM6_SAPPHIRERAPIDS_X:	/* SPR */
4660 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
4661 		do_rapl =
4662 		    RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS |
4663 		    RAPL_PKG_POWER_INFO;
4664 		BIC_PRESENT(BIC_PKG__);
4665 		BIC_PRESENT(BIC_RAM__);
4666 		if (rapl_joules) {
4667 			BIC_PRESENT(BIC_Pkg_J);
4668 			BIC_PRESENT(BIC_RAM_J);
4669 		} else {
4670 			BIC_PRESENT(BIC_PkgWatt);
4671 			BIC_PRESENT(BIC_RAMWatt);
4672 		}
4673 		break;
4674 	case INTEL_FAM6_SANDYBRIDGE_X:
4675 	case INTEL_FAM6_IVYBRIDGE_X:
4676 		do_rapl =
4677 		    RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_PKG_PERF_STATUS |
4678 		    RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
4679 		BIC_PRESENT(BIC_PKG__);
4680 		BIC_PRESENT(BIC_RAM__);
4681 		if (rapl_joules) {
4682 			BIC_PRESENT(BIC_Pkg_J);
4683 			BIC_PRESENT(BIC_Cor_J);
4684 			BIC_PRESENT(BIC_RAM_J);
4685 		} else {
4686 			BIC_PRESENT(BIC_PkgWatt);
4687 			BIC_PRESENT(BIC_CorWatt);
4688 			BIC_PRESENT(BIC_RAMWatt);
4689 		}
4690 		break;
4691 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
4692 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
4693 		do_rapl = RAPL_PKG | RAPL_CORES;
4694 		if (rapl_joules) {
4695 			BIC_PRESENT(BIC_Pkg_J);
4696 			BIC_PRESENT(BIC_Cor_J);
4697 		} else {
4698 			BIC_PRESENT(BIC_PkgWatt);
4699 			BIC_PRESENT(BIC_CorWatt);
4700 		}
4701 		break;
4702 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
4703 		do_rapl =
4704 		    RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS |
4705 		    RAPL_PKG_POWER_INFO | RAPL_CORES_ENERGY_STATUS;
4706 		BIC_PRESENT(BIC_PKG__);
4707 		BIC_PRESENT(BIC_RAM__);
4708 		if (rapl_joules) {
4709 			BIC_PRESENT(BIC_Pkg_J);
4710 			BIC_PRESENT(BIC_Cor_J);
4711 			BIC_PRESENT(BIC_RAM_J);
4712 		} else {
4713 			BIC_PRESENT(BIC_PkgWatt);
4714 			BIC_PRESENT(BIC_CorWatt);
4715 			BIC_PRESENT(BIC_RAMWatt);
4716 		}
4717 		break;
4718 	default:
4719 		return;
4720 	}
4721 
4722 	/* units on package 0, verify later other packages match */
4723 	if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr))
4724 		return;
4725 
4726 	rapl_power_units = 1.0 / (1 << (msr & 0xF));
4727 	if (model == INTEL_FAM6_ATOM_SILVERMONT)
4728 		rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
4729 	else
4730 		rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
4731 
4732 	rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units);
4733 
4734 	time_unit = msr >> 16 & 0xF;
4735 	if (time_unit == 0)
4736 		time_unit = 0xA;
4737 
4738 	rapl_time_units = 1.0 / (1 << (time_unit));
4739 
4740 	tdp = get_tdp_intel(model);
4741 
4742 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4743 	if (!quiet)
4744 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4745 }
4746 
4747 void rapl_probe_amd(unsigned int family, unsigned int model)
4748 {
4749 	unsigned long long msr;
4750 	unsigned int eax, ebx, ecx, edx;
4751 	unsigned int has_rapl = 0;
4752 	double tdp;
4753 
4754 	UNUSED(model);
4755 
4756 	if (max_extended_level >= 0x80000007) {
4757 		__cpuid(0x80000007, eax, ebx, ecx, edx);
4758 		/* RAPL (Fam 17h+) */
4759 		has_rapl = edx & (1 << 14);
4760 	}
4761 
4762 	if (!has_rapl || family < 0x17)
4763 		return;
4764 
4765 	do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY;
4766 	if (rapl_joules) {
4767 		BIC_PRESENT(BIC_Pkg_J);
4768 		BIC_PRESENT(BIC_Cor_J);
4769 	} else {
4770 		BIC_PRESENT(BIC_PkgWatt);
4771 		BIC_PRESENT(BIC_CorWatt);
4772 	}
4773 
4774 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
4775 		return;
4776 
4777 	rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf));
4778 	rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f));
4779 	rapl_power_units = ldexp(1.0, -(msr & 0xf));
4780 
4781 	tdp = get_tdp_amd(family);
4782 
4783 	rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
4784 	if (!quiet)
4785 		fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
4786 }
4787 
4788 /*
4789  * rapl_probe()
4790  *
4791  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
4792  */
4793 void rapl_probe(unsigned int family, unsigned int model)
4794 {
4795 	if (genuine_intel)
4796 		rapl_probe_intel(family, model);
4797 	if (authentic_amd || hygon_genuine)
4798 		rapl_probe_amd(family, model);
4799 }
4800 
4801 void perf_limit_reasons_probe(unsigned int family, unsigned int model)
4802 {
4803 	if (!genuine_intel)
4804 		return;
4805 
4806 	if (family != 6)
4807 		return;
4808 
4809 	switch (model) {
4810 	case INTEL_FAM6_HASWELL:	/* HSW */
4811 	case INTEL_FAM6_HASWELL_L:	/* HSW */
4812 	case INTEL_FAM6_HASWELL_G:	/* HSW */
4813 		do_gfx_perf_limit_reasons = 1;
4814 		/* FALLTHRU */
4815 	case INTEL_FAM6_HASWELL_X:	/* HSX */
4816 		do_core_perf_limit_reasons = 1;
4817 		do_ring_perf_limit_reasons = 1;
4818 	default:
4819 		return;
4820 	}
4821 }
4822 
4823 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model)
4824 {
4825 	if (family != 6)
4826 		return;
4827 
4828 	switch (model) {
4829 	case INTEL_FAM6_BROADWELL_X:
4830 	case INTEL_FAM6_SKYLAKE_X:
4831 		has_automatic_cstate_conversion = 1;
4832 	}
4833 }
4834 
4835 void prewake_cstate_probe(unsigned int family, unsigned int model)
4836 {
4837 	if (is_icx(family, model) || is_spr(family, model))
4838 		dis_cstate_prewake = 1;
4839 }
4840 
4841 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4842 {
4843 	unsigned long long msr;
4844 	unsigned int dts, dts2;
4845 	int cpu;
4846 
4847 	UNUSED(c);
4848 	UNUSED(p);
4849 
4850 	if (!(do_dts || do_ptm))
4851 		return 0;
4852 
4853 	cpu = t->cpu_id;
4854 
4855 	/* DTS is per-core, no need to print for each thread */
4856 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
4857 		return 0;
4858 
4859 	if (cpu_migrate(cpu)) {
4860 		fprintf(outf, "print_thermal: Could not migrate to CPU %d\n", cpu);
4861 		return -1;
4862 	}
4863 
4864 	if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
4865 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
4866 			return 0;
4867 
4868 		dts = (msr >> 16) & 0x7F;
4869 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n", cpu, msr, tj_max - dts);
4870 
4871 		if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr))
4872 			return 0;
4873 
4874 		dts = (msr >> 16) & 0x7F;
4875 		dts2 = (msr >> 8) & 0x7F;
4876 		fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4877 			cpu, msr, tj_max - dts, tj_max - dts2);
4878 	}
4879 
4880 	if (do_dts && debug) {
4881 		unsigned int resolution;
4882 
4883 		if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
4884 			return 0;
4885 
4886 		dts = (msr >> 16) & 0x7F;
4887 		resolution = (msr >> 27) & 0xF;
4888 		fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n",
4889 			cpu, msr, tj_max - dts, resolution);
4890 
4891 		if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr))
4892 			return 0;
4893 
4894 		dts = (msr >> 16) & 0x7F;
4895 		dts2 = (msr >> 8) & 0x7F;
4896 		fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n",
4897 			cpu, msr, tj_max - dts, tj_max - dts2);
4898 	}
4899 
4900 	return 0;
4901 }
4902 
4903 void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
4904 {
4905 	fprintf(outf, "cpu%d: %s: %sabled (%0.3f Watts, %f sec, clamp %sabled)\n",
4906 		cpu, label,
4907 		((msr >> 15) & 1) ? "EN" : "DIS",
4908 		((msr >> 0) & 0x7FFF) * rapl_power_units,
4909 		(1.0 + (((msr >> 22) & 0x3) / 4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units,
4910 		(((msr >> 16) & 1) ? "EN" : "DIS"));
4911 
4912 	return;
4913 }
4914 
4915 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
4916 {
4917 	unsigned long long msr;
4918 	const char *msr_name;
4919 	int cpu;
4920 
4921 	UNUSED(c);
4922 	UNUSED(p);
4923 
4924 	if (!do_rapl)
4925 		return 0;
4926 
4927 	/* RAPL counters are per package, so print only for 1st thread/package */
4928 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
4929 		return 0;
4930 
4931 	cpu = t->cpu_id;
4932 	if (cpu_migrate(cpu)) {
4933 		fprintf(outf, "print_rapl: Could not migrate to CPU %d\n", cpu);
4934 		return -1;
4935 	}
4936 
4937 	if (do_rapl & RAPL_AMD_F17H) {
4938 		msr_name = "MSR_RAPL_PWR_UNIT";
4939 		if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr))
4940 			return -1;
4941 	} else {
4942 		msr_name = "MSR_RAPL_POWER_UNIT";
4943 		if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr))
4944 			return -1;
4945 	}
4946 
4947 	fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr,
4948 		rapl_power_units, rapl_energy_units, rapl_time_units);
4949 
4950 	if (do_rapl & RAPL_PKG_POWER_INFO) {
4951 
4952 		if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr))
4953 			return -5;
4954 
4955 		fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4956 			cpu, msr,
4957 			((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4958 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4959 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4960 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4961 
4962 	}
4963 	if (do_rapl & RAPL_PKG) {
4964 
4965 		if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr))
4966 			return -9;
4967 
4968 		fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n",
4969 			cpu, msr, (msr >> 63) & 1 ? "" : "UN");
4970 
4971 		print_power_limit_msr(cpu, msr, "PKG Limit #1");
4972 		fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%0.3f Watts, %f* sec, clamp %sabled)\n",
4973 			cpu,
4974 			((msr >> 47) & 1) ? "EN" : "DIS",
4975 			((msr >> 32) & 0x7FFF) * rapl_power_units,
4976 			(1.0 + (((msr >> 54) & 0x3) / 4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units,
4977 			((msr >> 48) & 1) ? "EN" : "DIS");
4978 
4979 		if (get_msr(cpu, MSR_VR_CURRENT_CONFIG, &msr))
4980 			return -9;
4981 
4982 		fprintf(outf, "cpu%d: MSR_VR_CURRENT_CONFIG: 0x%08llx\n", cpu, msr);
4983 		fprintf(outf, "cpu%d: PKG Limit #4: %f Watts (%slocked)\n",
4984 			cpu, ((msr >> 0) & 0x1FFF) * rapl_power_units, (msr >> 31) & 1 ? "" : "UN");
4985 	}
4986 
4987 	if (do_rapl & RAPL_DRAM_POWER_INFO) {
4988 		if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr))
4989 			return -6;
4990 
4991 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n",
4992 			cpu, msr,
4993 			((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4994 			((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4995 			((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units,
4996 			((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units);
4997 	}
4998 	if (do_rapl & RAPL_DRAM) {
4999 		if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr))
5000 			return -9;
5001 		fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n",
5002 			cpu, msr, (msr >> 31) & 1 ? "" : "UN");
5003 
5004 		print_power_limit_msr(cpu, msr, "DRAM Limit");
5005 	}
5006 	if (do_rapl & RAPL_CORE_POLICY) {
5007 		if (get_msr(cpu, MSR_PP0_POLICY, &msr))
5008 			return -7;
5009 
5010 		fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF);
5011 	}
5012 	if (do_rapl & RAPL_CORES_POWER_LIMIT) {
5013 		if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr))
5014 			return -9;
5015 		fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n",
5016 			cpu, msr, (msr >> 31) & 1 ? "" : "UN");
5017 		print_power_limit_msr(cpu, msr, "Cores Limit");
5018 	}
5019 	if (do_rapl & RAPL_GFX) {
5020 		if (get_msr(cpu, MSR_PP1_POLICY, &msr))
5021 			return -8;
5022 
5023 		fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF);
5024 
5025 		if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr))
5026 			return -9;
5027 		fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n",
5028 			cpu, msr, (msr >> 31) & 1 ? "" : "UN");
5029 		print_power_limit_msr(cpu, msr, "GFX Limit");
5030 	}
5031 	return 0;
5032 }
5033 
5034 /*
5035  * SNB adds support for additional MSRs:
5036  *
5037  * MSR_PKG_C7_RESIDENCY            0x000003fa
5038  * MSR_CORE_C7_RESIDENCY           0x000003fe
5039  * MSR_PKG_C2_RESIDENCY            0x0000060d
5040  */
5041 
5042 int has_snb_msrs(unsigned int family, unsigned int model)
5043 {
5044 	if (!genuine_intel)
5045 		return 0;
5046 
5047 	if (family != 6)
5048 		return 0;
5049 
5050 	switch (model) {
5051 	case INTEL_FAM6_SANDYBRIDGE:
5052 	case INTEL_FAM6_SANDYBRIDGE_X:
5053 	case INTEL_FAM6_IVYBRIDGE:	/* IVB */
5054 	case INTEL_FAM6_IVYBRIDGE_X:	/* IVB Xeon */
5055 	case INTEL_FAM6_HASWELL:	/* HSW */
5056 	case INTEL_FAM6_HASWELL_X:	/* HSW */
5057 	case INTEL_FAM6_HASWELL_L:	/* HSW */
5058 	case INTEL_FAM6_HASWELL_G:	/* HSW */
5059 	case INTEL_FAM6_BROADWELL:	/* BDW */
5060 	case INTEL_FAM6_BROADWELL_G:	/* BDW */
5061 	case INTEL_FAM6_BROADWELL_X:	/* BDX */
5062 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
5063 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
5064 	case INTEL_FAM6_SKYLAKE_X:	/* SKX */
5065 	case INTEL_FAM6_ICELAKE_X:	/* ICX */
5066 	case INTEL_FAM6_SAPPHIRERAPIDS_X:	/* SPR */
5067 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
5068 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
5069 	case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
5070 	case INTEL_FAM6_ATOM_TREMONT:	/* EHL */
5071 	case INTEL_FAM6_ATOM_TREMONT_D:	/* JVL */
5072 		return 1;
5073 	}
5074 	return 0;
5075 }
5076 
5077 /*
5078  * HSW ULT added support for C8/C9/C10 MSRs:
5079  *
5080  * MSR_PKG_C8_RESIDENCY		0x00000630
5081  * MSR_PKG_C9_RESIDENCY		0x00000631
5082  * MSR_PKG_C10_RESIDENCY	0x00000632
5083  *
5084  * MSR_PKGC8_IRTL		0x00000633
5085  * MSR_PKGC9_IRTL		0x00000634
5086  * MSR_PKGC10_IRTL		0x00000635
5087  *
5088  */
5089 int has_c8910_msrs(unsigned int family, unsigned int model)
5090 {
5091 	if (!genuine_intel)
5092 		return 0;
5093 
5094 	if (family != 6)
5095 		return 0;
5096 
5097 	switch (model) {
5098 	case INTEL_FAM6_HASWELL_L:	/* HSW */
5099 	case INTEL_FAM6_BROADWELL:	/* BDW */
5100 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
5101 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
5102 	case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
5103 	case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
5104 	case INTEL_FAM6_ATOM_TREMONT:	/* EHL */
5105 		return 1;
5106 	}
5107 	return 0;
5108 }
5109 
5110 /*
5111  * SKL adds support for additional MSRS:
5112  *
5113  * MSR_PKG_WEIGHTED_CORE_C0_RES    0x00000658
5114  * MSR_PKG_ANY_CORE_C0_RES         0x00000659
5115  * MSR_PKG_ANY_GFXE_C0_RES         0x0000065A
5116  * MSR_PKG_BOTH_CORE_GFXE_C0_RES   0x0000065B
5117  */
5118 int has_skl_msrs(unsigned int family, unsigned int model)
5119 {
5120 	if (!genuine_intel)
5121 		return 0;
5122 
5123 	if (family != 6)
5124 		return 0;
5125 
5126 	switch (model) {
5127 	case INTEL_FAM6_SKYLAKE_L:	/* SKL */
5128 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
5129 		return 1;
5130 	}
5131 	return 0;
5132 }
5133 
5134 int is_slm(unsigned int family, unsigned int model)
5135 {
5136 	if (!genuine_intel)
5137 		return 0;
5138 
5139 	if (family != 6)
5140 		return 0;
5141 
5142 	switch (model) {
5143 	case INTEL_FAM6_ATOM_SILVERMONT:	/* BYT */
5144 	case INTEL_FAM6_ATOM_SILVERMONT_D:	/* AVN */
5145 		return 1;
5146 	}
5147 	return 0;
5148 }
5149 
5150 int is_knl(unsigned int family, unsigned int model)
5151 {
5152 	if (!genuine_intel)
5153 		return 0;
5154 
5155 	if (family != 6)
5156 		return 0;
5157 
5158 	switch (model) {
5159 	case INTEL_FAM6_XEON_PHI_KNL:	/* KNL */
5160 		return 1;
5161 	}
5162 	return 0;
5163 }
5164 
5165 int is_cnl(unsigned int family, unsigned int model)
5166 {
5167 	if (!genuine_intel)
5168 		return 0;
5169 
5170 	if (family != 6)
5171 		return 0;
5172 
5173 	switch (model) {
5174 	case INTEL_FAM6_CANNONLAKE_L:	/* CNL */
5175 		return 1;
5176 	}
5177 
5178 	return 0;
5179 }
5180 
5181 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model)
5182 {
5183 	if (is_knl(family, model))
5184 		return 1024;
5185 	return 1;
5186 }
5187 
5188 #define SLM_BCLK_FREQS 5
5189 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0 };
5190 
5191 double slm_bclk(void)
5192 {
5193 	unsigned long long msr = 3;
5194 	unsigned int i;
5195 	double freq;
5196 
5197 	if (get_msr(base_cpu, MSR_FSB_FREQ, &msr))
5198 		fprintf(outf, "SLM BCLK: unknown\n");
5199 
5200 	i = msr & 0xf;
5201 	if (i >= SLM_BCLK_FREQS) {
5202 		fprintf(outf, "SLM BCLK[%d] invalid\n", i);
5203 		i = 3;
5204 	}
5205 	freq = slm_freq_table[i];
5206 
5207 	if (!quiet)
5208 		fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq);
5209 
5210 	return freq;
5211 }
5212 
5213 double discover_bclk(unsigned int family, unsigned int model)
5214 {
5215 	if (has_snb_msrs(family, model) || is_knl(family, model))
5216 		return 100.00;
5217 	else if (is_slm(family, model))
5218 		return slm_bclk();
5219 	else
5220 		return 133.33;
5221 }
5222 
5223 int get_cpu_type(struct thread_data *t, struct core_data *c, struct pkg_data *p)
5224 {
5225 	unsigned int eax, ebx, ecx, edx;
5226 
5227 	UNUSED(c);
5228 	UNUSED(p);
5229 
5230 	if (!genuine_intel)
5231 		return 0;
5232 
5233 	if (cpu_migrate(t->cpu_id)) {
5234 		fprintf(outf, "Could not migrate to CPU %d\n", t->cpu_id);
5235 		return -1;
5236 	}
5237 
5238 	if (max_level < 0x1a)
5239 		return 0;
5240 
5241 	__cpuid(0x1a, eax, ebx, ecx, edx);
5242 	eax = (eax >> 24) & 0xFF;
5243 	if (eax == 0x20)
5244 		t->is_atom = true;
5245 	return 0;
5246 }
5247 
5248 /*
5249  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
5250  * the Thermal Control Circuit (TCC) activates.
5251  * This is usually equal to tjMax.
5252  *
5253  * Older processors do not have this MSR, so there we guess,
5254  * but also allow cmdline over-ride with -T.
5255  *
5256  * Several MSR temperature values are in units of degrees-C
5257  * below this value, including the Digital Thermal Sensor (DTS),
5258  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
5259  */
5260 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
5261 {
5262 	unsigned long long msr;
5263 	unsigned int tcc_default, tcc_offset;
5264 	int cpu;
5265 
5266 	UNUSED(c);
5267 	UNUSED(p);
5268 
5269 	/* tj_max is used only for dts or ptm */
5270 	if (!(do_dts || do_ptm))
5271 		return 0;
5272 
5273 	/* this is a per-package concept */
5274 	if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
5275 		return 0;
5276 
5277 	cpu = t->cpu_id;
5278 	if (cpu_migrate(cpu)) {
5279 		fprintf(outf, "Could not migrate to CPU %d\n", cpu);
5280 		return -1;
5281 	}
5282 
5283 	if (tj_max_override != 0) {
5284 		tj_max = tj_max_override;
5285 		fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n", cpu, tj_max);
5286 		return 0;
5287 	}
5288 
5289 	/* Temperature Target MSR is Nehalem and newer only */
5290 	if (!do_nhm_platform_info)
5291 		goto guess;
5292 
5293 	if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
5294 		goto guess;
5295 
5296 	tcc_default = (msr >> 16) & 0xFF;
5297 
5298 	if (!quiet) {
5299 		switch (tcc_offset_bits) {
5300 		case 4:
5301 			tcc_offset = (msr >> 24) & 0xF;
5302 			fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C) (%d default - %d offset)\n",
5303 				cpu, msr, tcc_default - tcc_offset, tcc_default, tcc_offset);
5304 			break;
5305 		case 6:
5306 			tcc_offset = (msr >> 24) & 0x3F;
5307 			fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C) (%d default - %d offset)\n",
5308 				cpu, msr, tcc_default - tcc_offset, tcc_default, tcc_offset);
5309 			break;
5310 		default:
5311 			fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n", cpu, msr, tcc_default);
5312 			break;
5313 		}
5314 	}
5315 
5316 	if (!tcc_default)
5317 		goto guess;
5318 
5319 	tj_max = tcc_default;
5320 
5321 	return 0;
5322 
5323 guess:
5324 	tj_max = TJMAX_DEFAULT;
5325 	fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n", cpu, tj_max);
5326 
5327 	return 0;
5328 }
5329 
5330 void decode_feature_control_msr(void)
5331 {
5332 	unsigned long long msr;
5333 
5334 	if (!get_msr(base_cpu, MSR_IA32_FEAT_CTL, &msr))
5335 		fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n",
5336 			base_cpu, msr, msr & FEAT_CTL_LOCKED ? "" : "UN-", msr & (1 << 18) ? "SGX" : "");
5337 }
5338 
5339 void decode_misc_enable_msr(void)
5340 {
5341 	unsigned long long msr;
5342 
5343 	if (!genuine_intel)
5344 		return;
5345 
5346 	if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr))
5347 		fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n",
5348 			base_cpu, msr,
5349 			msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-",
5350 			msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-",
5351 			msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-",
5352 			msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "",
5353 			msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : "");
5354 }
5355 
5356 void decode_misc_feature_control(void)
5357 {
5358 	unsigned long long msr;
5359 
5360 	if (!has_misc_feature_control)
5361 		return;
5362 
5363 	if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr))
5364 		fprintf(outf,
5365 			"cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n",
5366 			base_cpu, msr, msr & (0 << 0) ? "No-" : "", msr & (1 << 0) ? "No-" : "",
5367 			msr & (2 << 0) ? "No-" : "", msr & (3 << 0) ? "No-" : "");
5368 }
5369 
5370 /*
5371  * Decode MSR_MISC_PWR_MGMT
5372  *
5373  * Decode the bits according to the Nehalem documentation
5374  * bit[0] seems to continue to have same meaning going forward
5375  * bit[1] less so...
5376  */
5377 void decode_misc_pwr_mgmt_msr(void)
5378 {
5379 	unsigned long long msr;
5380 
5381 	if (!do_nhm_platform_info)
5382 		return;
5383 
5384 	if (no_MSR_MISC_PWR_MGMT)
5385 		return;
5386 
5387 	if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr))
5388 		fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n",
5389 			base_cpu, msr,
5390 			msr & (1 << 0) ? "DIS" : "EN", msr & (1 << 1) ? "EN" : "DIS", msr & (1 << 8) ? "EN" : "DIS");
5391 }
5392 
5393 /*
5394  * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG
5395  *
5396  * This MSRs are present on Silvermont processors,
5397  * Intel Atom processor E3000 series (Baytrail), and friends.
5398  */
5399 void decode_c6_demotion_policy_msr(void)
5400 {
5401 	unsigned long long msr;
5402 
5403 	if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr))
5404 		fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n",
5405 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
5406 
5407 	if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr))
5408 		fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n",
5409 			base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS");
5410 }
5411 
5412 /*
5413  * When models are the same, for the purpose of turbostat, reuse
5414  */
5415 unsigned int intel_model_duplicates(unsigned int model)
5416 {
5417 
5418 	switch (model) {
5419 	case INTEL_FAM6_NEHALEM_EP:	/* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
5420 	case INTEL_FAM6_NEHALEM:	/* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
5421 	case 0x1F:		/* Core i7 and i5 Processor - Nehalem */
5422 	case INTEL_FAM6_WESTMERE:	/* Westmere Client - Clarkdale, Arrandale */
5423 	case INTEL_FAM6_WESTMERE_EP:	/* Westmere EP - Gulftown */
5424 		return INTEL_FAM6_NEHALEM;
5425 
5426 	case INTEL_FAM6_NEHALEM_EX:	/* Nehalem-EX Xeon - Beckton */
5427 	case INTEL_FAM6_WESTMERE_EX:	/* Westmere-EX Xeon - Eagleton */
5428 		return INTEL_FAM6_NEHALEM_EX;
5429 
5430 	case INTEL_FAM6_XEON_PHI_KNM:
5431 		return INTEL_FAM6_XEON_PHI_KNL;
5432 
5433 	case INTEL_FAM6_BROADWELL_X:
5434 	case INTEL_FAM6_BROADWELL_D:	/* BDX-DE */
5435 		return INTEL_FAM6_BROADWELL_X;
5436 
5437 	case INTEL_FAM6_SKYLAKE_L:
5438 	case INTEL_FAM6_SKYLAKE:
5439 	case INTEL_FAM6_KABYLAKE_L:
5440 	case INTEL_FAM6_KABYLAKE:
5441 	case INTEL_FAM6_COMETLAKE_L:
5442 	case INTEL_FAM6_COMETLAKE:
5443 		return INTEL_FAM6_SKYLAKE_L;
5444 
5445 	case INTEL_FAM6_ICELAKE_L:
5446 	case INTEL_FAM6_ICELAKE_NNPI:
5447 	case INTEL_FAM6_TIGERLAKE_L:
5448 	case INTEL_FAM6_TIGERLAKE:
5449 	case INTEL_FAM6_ROCKETLAKE:
5450 	case INTEL_FAM6_LAKEFIELD:
5451 	case INTEL_FAM6_ALDERLAKE:
5452 	case INTEL_FAM6_ALDERLAKE_L:
5453 	case INTEL_FAM6_ATOM_GRACEMONT:
5454 	case INTEL_FAM6_RAPTORLAKE:
5455 	case INTEL_FAM6_RAPTORLAKE_P:
5456 	case INTEL_FAM6_RAPTORLAKE_S:
5457 	case INTEL_FAM6_METEORLAKE:
5458 	case INTEL_FAM6_METEORLAKE_L:
5459 		return INTEL_FAM6_CANNONLAKE_L;
5460 
5461 	case INTEL_FAM6_ATOM_TREMONT_L:
5462 		return INTEL_FAM6_ATOM_TREMONT;
5463 
5464 	case INTEL_FAM6_ICELAKE_D:
5465 		return INTEL_FAM6_ICELAKE_X;
5466 
5467 	case INTEL_FAM6_EMERALDRAPIDS_X:
5468 		return INTEL_FAM6_SAPPHIRERAPIDS_X;
5469 	}
5470 	return model;
5471 }
5472 
5473 void print_dev_latency(void)
5474 {
5475 	char *path = "/dev/cpu_dma_latency";
5476 	int fd;
5477 	int value;
5478 	int retval;
5479 
5480 	fd = open(path, O_RDONLY);
5481 	if (fd < 0) {
5482 		if (debug)
5483 			warnx("Read %s failed", path);
5484 		return;
5485 	}
5486 
5487 	retval = read(fd, (void *)&value, sizeof(int));
5488 	if (retval != sizeof(int)) {
5489 		warn("read failed %s", path);
5490 		close(fd);
5491 		return;
5492 	}
5493 	fprintf(outf, "/dev/cpu_dma_latency: %d usec (%s)\n", value, value == 2000000000 ? "default" : "constrained");
5494 
5495 	close(fd);
5496 }
5497 
5498 /*
5499  * Linux-perf manages the HW instructions-retired counter
5500  * by enabling when requested, and hiding rollover
5501  */
5502 void linux_perf_init(void)
5503 {
5504 	if (!BIC_IS_ENABLED(BIC_IPC))
5505 		return;
5506 
5507 	if (access("/proc/sys/kernel/perf_event_paranoid", F_OK))
5508 		return;
5509 
5510 	fd_instr_count_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
5511 	if (fd_instr_count_percpu == NULL)
5512 		err(-1, "calloc fd_instr_count_percpu");
5513 
5514 	BIC_PRESENT(BIC_IPC);
5515 }
5516 
5517 void process_cpuid()
5518 {
5519 	unsigned int eax, ebx, ecx, edx;
5520 	unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
5521 	unsigned long long ucode_patch = 0;
5522 	bool ucode_patch_valid = false;
5523 
5524 	eax = ebx = ecx = edx = 0;
5525 
5526 	__cpuid(0, max_level, ebx, ecx, edx);
5527 
5528 	if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
5529 		genuine_intel = 1;
5530 	else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
5531 		authentic_amd = 1;
5532 	else if (ebx == 0x6f677948 && ecx == 0x656e6975 && edx == 0x6e65476e)
5533 		hygon_genuine = 1;
5534 
5535 	if (!quiet)
5536 		fprintf(outf, "CPUID(0): %.4s%.4s%.4s 0x%x CPUID levels\n",
5537 			(char *)&ebx, (char *)&edx, (char *)&ecx, max_level);
5538 
5539 	__cpuid(1, fms, ebx, ecx, edx);
5540 	family = (fms >> 8) & 0xf;
5541 	model = (fms >> 4) & 0xf;
5542 	stepping = fms & 0xf;
5543 	if (family == 0xf)
5544 		family += (fms >> 20) & 0xff;
5545 	if (family >= 6)
5546 		model += ((fms >> 16) & 0xf) << 4;
5547 	ecx_flags = ecx;
5548 	edx_flags = edx;
5549 
5550 	if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch))
5551 		warnx("get_msr(UCODE)");
5552 	else
5553 		ucode_patch_valid = true;
5554 
5555 	/*
5556 	 * check max extended function levels of CPUID.
5557 	 * This is needed to check for invariant TSC.
5558 	 * This check is valid for both Intel and AMD.
5559 	 */
5560 	ebx = ecx = edx = 0;
5561 	__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
5562 
5563 	if (!quiet) {
5564 		fprintf(outf, "CPUID(1): family:model:stepping 0x%x:%x:%x (%d:%d:%d)",
5565 			family, model, stepping, family, model, stepping);
5566 		if (ucode_patch_valid)
5567 			fprintf(outf, " microcode 0x%x", (unsigned int)((ucode_patch >> 32) & 0xFFFFFFFF));
5568 		fputc('\n', outf);
5569 
5570 		fprintf(outf, "CPUID(0x80000000): max_extended_levels: 0x%x\n", max_extended_level);
5571 		fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n",
5572 			ecx_flags & (1 << 0) ? "SSE3" : "-",
5573 			ecx_flags & (1 << 3) ? "MONITOR" : "-",
5574 			ecx_flags & (1 << 6) ? "SMX" : "-",
5575 			ecx_flags & (1 << 7) ? "EIST" : "-",
5576 			ecx_flags & (1 << 8) ? "TM2" : "-",
5577 			edx_flags & (1 << 4) ? "TSC" : "-",
5578 			edx_flags & (1 << 5) ? "MSR" : "-",
5579 			edx_flags & (1 << 22) ? "ACPI-TM" : "-",
5580 			edx_flags & (1 << 28) ? "HT" : "-", edx_flags & (1 << 29) ? "TM" : "-");
5581 	}
5582 	if (genuine_intel) {
5583 		model_orig = model;
5584 		model = intel_model_duplicates(model);
5585 	}
5586 
5587 	if (!(edx_flags & (1 << 5)))
5588 		errx(1, "CPUID: no MSR");
5589 
5590 	if (max_extended_level >= 0x80000007) {
5591 
5592 		/*
5593 		 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
5594 		 * this check is valid for both Intel and AMD
5595 		 */
5596 		__cpuid(0x80000007, eax, ebx, ecx, edx);
5597 		has_invariant_tsc = edx & (1 << 8);
5598 	}
5599 
5600 	/*
5601 	 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
5602 	 * this check is valid for both Intel and AMD
5603 	 */
5604 
5605 	__cpuid(0x6, eax, ebx, ecx, edx);
5606 	has_aperf = ecx & (1 << 0);
5607 	if (has_aperf) {
5608 		BIC_PRESENT(BIC_Avg_MHz);
5609 		BIC_PRESENT(BIC_Busy);
5610 		BIC_PRESENT(BIC_Bzy_MHz);
5611 	}
5612 	do_dts = eax & (1 << 0);
5613 	if (do_dts)
5614 		BIC_PRESENT(BIC_CoreTmp);
5615 	has_turbo = eax & (1 << 1);
5616 	do_ptm = eax & (1 << 6);
5617 	if (do_ptm)
5618 		BIC_PRESENT(BIC_PkgTmp);
5619 	has_hwp = eax & (1 << 7);
5620 	has_hwp_notify = eax & (1 << 8);
5621 	has_hwp_activity_window = eax & (1 << 9);
5622 	has_hwp_epp = eax & (1 << 10);
5623 	has_hwp_pkg = eax & (1 << 11);
5624 	has_epb = ecx & (1 << 3);
5625 
5626 	if (!quiet)
5627 		fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, "
5628 			"%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n",
5629 			has_aperf ? "" : "No-",
5630 			has_turbo ? "" : "No-",
5631 			do_dts ? "" : "No-",
5632 			do_ptm ? "" : "No-",
5633 			has_hwp ? "" : "No-",
5634 			has_hwp_notify ? "" : "No-",
5635 			has_hwp_activity_window ? "" : "No-",
5636 			has_hwp_epp ? "" : "No-", has_hwp_pkg ? "" : "No-", has_epb ? "" : "No-");
5637 
5638 	if (!quiet)
5639 		decode_misc_enable_msr();
5640 
5641 	if (max_level >= 0x7 && !quiet) {
5642 		int has_sgx;
5643 
5644 		ecx = 0;
5645 
5646 		__cpuid_count(0x7, 0, eax, ebx, ecx, edx);
5647 
5648 		has_sgx = ebx & (1 << 2);
5649 
5650 		is_hybrid = edx & (1 << 15);
5651 
5652 		fprintf(outf, "CPUID(7): %sSGX %sHybrid\n", has_sgx ? "" : "No-", is_hybrid ? "" : "No-");
5653 
5654 		if (has_sgx)
5655 			decode_feature_control_msr();
5656 	}
5657 
5658 	if (max_level >= 0x15) {
5659 		unsigned int eax_crystal;
5660 		unsigned int ebx_tsc;
5661 
5662 		/*
5663 		 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz
5664 		 */
5665 		eax_crystal = ebx_tsc = crystal_hz = edx = 0;
5666 		__cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx);
5667 
5668 		if (ebx_tsc != 0) {
5669 
5670 			if (!quiet && (ebx != 0))
5671 				fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n",
5672 					eax_crystal, ebx_tsc, crystal_hz);
5673 
5674 			if (crystal_hz == 0)
5675 				switch (model) {
5676 				case INTEL_FAM6_SKYLAKE_L:	/* SKL */
5677 					crystal_hz = 24000000;	/* 24.0 MHz */
5678 					break;
5679 				case INTEL_FAM6_ATOM_GOLDMONT_D:	/* DNV */
5680 					crystal_hz = 25000000;	/* 25.0 MHz */
5681 					break;
5682 				case INTEL_FAM6_ATOM_GOLDMONT:	/* BXT */
5683 				case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
5684 					crystal_hz = 19200000;	/* 19.2 MHz */
5685 					break;
5686 				default:
5687 					crystal_hz = 0;
5688 				}
5689 
5690 			if (crystal_hz) {
5691 				tsc_hz = (unsigned long long)crystal_hz *ebx_tsc / eax_crystal;
5692 				if (!quiet)
5693 					fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n",
5694 						tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal);
5695 			}
5696 		}
5697 	}
5698 	if (max_level >= 0x16) {
5699 		unsigned int base_mhz, max_mhz, bus_mhz, edx;
5700 
5701 		/*
5702 		 * CPUID 16H Base MHz, Max MHz, Bus MHz
5703 		 */
5704 		base_mhz = max_mhz = bus_mhz = edx = 0;
5705 
5706 		__cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx);
5707 		if (!quiet)
5708 			fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n",
5709 				base_mhz, max_mhz, bus_mhz);
5710 	}
5711 
5712 	if (has_aperf)
5713 		aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model);
5714 
5715 	BIC_PRESENT(BIC_IRQ);
5716 	BIC_PRESENT(BIC_TSC_MHz);
5717 
5718 	if (probe_nhm_msrs(family, model)) {
5719 		do_nhm_platform_info = 1;
5720 		BIC_PRESENT(BIC_CPU_c1);
5721 		BIC_PRESENT(BIC_CPU_c3);
5722 		BIC_PRESENT(BIC_CPU_c6);
5723 		BIC_PRESENT(BIC_SMI);
5724 	}
5725 	do_snb_cstates = has_snb_msrs(family, model);
5726 
5727 	if (do_snb_cstates)
5728 		BIC_PRESENT(BIC_CPU_c7);
5729 
5730 	do_irtl_snb = has_snb_msrs(family, model);
5731 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__2))
5732 		BIC_PRESENT(BIC_Pkgpc2);
5733 	if (pkg_cstate_limit >= PCL__3)
5734 		BIC_PRESENT(BIC_Pkgpc3);
5735 	if (pkg_cstate_limit >= PCL__6)
5736 		BIC_PRESENT(BIC_Pkgpc6);
5737 	if (do_snb_cstates && (pkg_cstate_limit >= PCL__7))
5738 		BIC_PRESENT(BIC_Pkgpc7);
5739 	if (has_slv_msrs(family, model)) {
5740 		BIC_NOT_PRESENT(BIC_Pkgpc2);
5741 		BIC_NOT_PRESENT(BIC_Pkgpc3);
5742 		BIC_PRESENT(BIC_Pkgpc6);
5743 		BIC_NOT_PRESENT(BIC_Pkgpc7);
5744 		BIC_PRESENT(BIC_Mod_c6);
5745 		use_c1_residency_msr = 1;
5746 	}
5747 	if (is_jvl(family, model)) {
5748 		BIC_NOT_PRESENT(BIC_CPU_c3);
5749 		BIC_NOT_PRESENT(BIC_CPU_c7);
5750 		BIC_NOT_PRESENT(BIC_Pkgpc2);
5751 		BIC_NOT_PRESENT(BIC_Pkgpc3);
5752 		BIC_NOT_PRESENT(BIC_Pkgpc6);
5753 		BIC_NOT_PRESENT(BIC_Pkgpc7);
5754 	}
5755 	if (is_dnv(family, model)) {
5756 		BIC_PRESENT(BIC_CPU_c1);
5757 		BIC_NOT_PRESENT(BIC_CPU_c3);
5758 		BIC_NOT_PRESENT(BIC_Pkgpc3);
5759 		BIC_NOT_PRESENT(BIC_CPU_c7);
5760 		BIC_NOT_PRESENT(BIC_Pkgpc7);
5761 		use_c1_residency_msr = 1;
5762 	}
5763 	if (is_skx(family, model) || is_icx(family, model) || is_spr(family, model)) {
5764 		BIC_NOT_PRESENT(BIC_CPU_c3);
5765 		BIC_NOT_PRESENT(BIC_Pkgpc3);
5766 		BIC_NOT_PRESENT(BIC_CPU_c7);
5767 		BIC_NOT_PRESENT(BIC_Pkgpc7);
5768 	}
5769 	if (is_bdx(family, model)) {
5770 		BIC_NOT_PRESENT(BIC_CPU_c7);
5771 		BIC_NOT_PRESENT(BIC_Pkgpc7);
5772 	}
5773 	if (has_c8910_msrs(family, model)) {
5774 		if (pkg_cstate_limit >= PCL__8)
5775 			BIC_PRESENT(BIC_Pkgpc8);
5776 		if (pkg_cstate_limit >= PCL__9)
5777 			BIC_PRESENT(BIC_Pkgpc9);
5778 		if (pkg_cstate_limit >= PCL_10)
5779 			BIC_PRESENT(BIC_Pkgpc10);
5780 	}
5781 	do_irtl_hsw = has_c8910_msrs(family, model);
5782 	if (has_skl_msrs(family, model)) {
5783 		BIC_PRESENT(BIC_Totl_c0);
5784 		BIC_PRESENT(BIC_Any_c0);
5785 		BIC_PRESENT(BIC_GFX_c0);
5786 		BIC_PRESENT(BIC_CPUGFX);
5787 	}
5788 	do_slm_cstates = is_slm(family, model);
5789 	do_knl_cstates = is_knl(family, model);
5790 
5791 	if (do_slm_cstates || do_knl_cstates || is_cnl(family, model) || is_ehl(family, model))
5792 		BIC_NOT_PRESENT(BIC_CPU_c3);
5793 
5794 	if (!quiet)
5795 		decode_misc_pwr_mgmt_msr();
5796 
5797 	if (!quiet && has_slv_msrs(family, model))
5798 		decode_c6_demotion_policy_msr();
5799 
5800 	rapl_probe(family, model);
5801 	perf_limit_reasons_probe(family, model);
5802 	automatic_cstate_conversion_probe(family, model);
5803 	prewake_cstate_probe(family, model);
5804 
5805 	check_tcc_offset(model_orig);
5806 
5807 	if (!quiet)
5808 		dump_cstate_pstate_config_info(family, model);
5809 	intel_uncore_frequency_probe();
5810 
5811 	if (!quiet)
5812 		print_dev_latency();
5813 	if (!quiet)
5814 		dump_sysfs_cstate_config();
5815 	if (!quiet)
5816 		dump_sysfs_pstate_config();
5817 
5818 	if (has_skl_msrs(family, model) || is_ehl(family, model))
5819 		calculate_tsc_tweak();
5820 
5821 	if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK))
5822 		BIC_PRESENT(BIC_GFX_rc6);
5823 
5824 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK))
5825 		BIC_PRESENT(BIC_GFXMHz);
5826 
5827 	if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", R_OK))
5828 		BIC_PRESENT(BIC_GFXACTMHz);
5829 
5830 	if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK))
5831 		BIC_PRESENT(BIC_CPU_LPI);
5832 	else
5833 		BIC_NOT_PRESENT(BIC_CPU_LPI);
5834 
5835 	if (!access("/sys/devices/system/cpu/cpu0/thermal_throttle/core_throttle_count", R_OK))
5836 		BIC_PRESENT(BIC_CORE_THROT_CNT);
5837 	else
5838 		BIC_NOT_PRESENT(BIC_CORE_THROT_CNT);
5839 
5840 	if (!access(sys_lpi_file_sysfs, R_OK)) {
5841 		sys_lpi_file = sys_lpi_file_sysfs;
5842 		BIC_PRESENT(BIC_SYS_LPI);
5843 	} else if (!access(sys_lpi_file_debugfs, R_OK)) {
5844 		sys_lpi_file = sys_lpi_file_debugfs;
5845 		BIC_PRESENT(BIC_SYS_LPI);
5846 	} else {
5847 		sys_lpi_file_sysfs = NULL;
5848 		BIC_NOT_PRESENT(BIC_SYS_LPI);
5849 	}
5850 
5851 	if (!quiet)
5852 		decode_misc_feature_control();
5853 
5854 	return;
5855 }
5856 
5857 /*
5858  * in /dev/cpu/ return success for names that are numbers
5859  * ie. filter out ".", "..", "microcode".
5860  */
5861 int dir_filter(const struct dirent *dirp)
5862 {
5863 	if (isdigit(dirp->d_name[0]))
5864 		return 1;
5865 	else
5866 		return 0;
5867 }
5868 
5869 void topology_probe()
5870 {
5871 	int i;
5872 	int max_core_id = 0;
5873 	int max_package_id = 0;
5874 	int max_die_id = 0;
5875 	int max_siblings = 0;
5876 
5877 	/* Initialize num_cpus, max_cpu_num */
5878 	set_max_cpu_num();
5879 	topo.num_cpus = 0;
5880 	for_all_proc_cpus(count_cpus);
5881 	if (!summary_only && topo.num_cpus > 1)
5882 		BIC_PRESENT(BIC_CPU);
5883 
5884 	if (debug > 1)
5885 		fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
5886 
5887 	cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
5888 	if (cpus == NULL)
5889 		err(1, "calloc cpus");
5890 
5891 	/*
5892 	 * Allocate and initialize cpu_present_set
5893 	 */
5894 	cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
5895 	if (cpu_present_set == NULL)
5896 		err(3, "CPU_ALLOC");
5897 	cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
5898 	CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
5899 	for_all_proc_cpus(mark_cpu_present);
5900 
5901 	/*
5902 	 * Validate that all cpus in cpu_subset are also in cpu_present_set
5903 	 */
5904 	for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) {
5905 		if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset))
5906 			if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set))
5907 				err(1, "cpu%d not present", i);
5908 	}
5909 
5910 	/*
5911 	 * Allocate and initialize cpu_affinity_set
5912 	 */
5913 	cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
5914 	if (cpu_affinity_set == NULL)
5915 		err(3, "CPU_ALLOC");
5916 	cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
5917 	CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
5918 
5919 	for_all_proc_cpus(init_thread_id);
5920 
5921 	/*
5922 	 * For online cpus
5923 	 * find max_core_id, max_package_id
5924 	 */
5925 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5926 		int siblings;
5927 
5928 		if (cpu_is_not_present(i)) {
5929 			if (debug > 1)
5930 				fprintf(outf, "cpu%d NOT PRESENT\n", i);
5931 			continue;
5932 		}
5933 
5934 		cpus[i].logical_cpu_id = i;
5935 
5936 		/* get package information */
5937 		cpus[i].physical_package_id = get_physical_package_id(i);
5938 		if (cpus[i].physical_package_id > max_package_id)
5939 			max_package_id = cpus[i].physical_package_id;
5940 
5941 		/* get die information */
5942 		cpus[i].die_id = get_die_id(i);
5943 		if (cpus[i].die_id > max_die_id)
5944 			max_die_id = cpus[i].die_id;
5945 
5946 		/* get numa node information */
5947 		cpus[i].physical_node_id = get_physical_node_id(&cpus[i]);
5948 		if (cpus[i].physical_node_id > topo.max_node_num)
5949 			topo.max_node_num = cpus[i].physical_node_id;
5950 
5951 		/* get core information */
5952 		cpus[i].physical_core_id = get_core_id(i);
5953 		if (cpus[i].physical_core_id > max_core_id)
5954 			max_core_id = cpus[i].physical_core_id;
5955 
5956 		/* get thread information */
5957 		siblings = get_thread_siblings(&cpus[i]);
5958 		if (siblings > max_siblings)
5959 			max_siblings = siblings;
5960 		if (cpus[i].thread_id == 0)
5961 			topo.num_cores++;
5962 	}
5963 
5964 	topo.cores_per_node = max_core_id + 1;
5965 	if (debug > 1)
5966 		fprintf(outf, "max_core_id %d, sizing for %d cores per package\n", max_core_id, topo.cores_per_node);
5967 	if (!summary_only && topo.cores_per_node > 1)
5968 		BIC_PRESENT(BIC_Core);
5969 
5970 	topo.num_die = max_die_id + 1;
5971 	if (debug > 1)
5972 		fprintf(outf, "max_die_id %d, sizing for %d die\n", max_die_id, topo.num_die);
5973 	if (!summary_only && topo.num_die > 1)
5974 		BIC_PRESENT(BIC_Die);
5975 
5976 	topo.num_packages = max_package_id + 1;
5977 	if (debug > 1)
5978 		fprintf(outf, "max_package_id %d, sizing for %d packages\n", max_package_id, topo.num_packages);
5979 	if (!summary_only && topo.num_packages > 1)
5980 		BIC_PRESENT(BIC_Package);
5981 
5982 	set_node_data();
5983 	if (debug > 1)
5984 		fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg);
5985 	if (!summary_only && topo.nodes_per_pkg > 1)
5986 		BIC_PRESENT(BIC_Node);
5987 
5988 	topo.threads_per_core = max_siblings;
5989 	if (debug > 1)
5990 		fprintf(outf, "max_siblings %d\n", max_siblings);
5991 
5992 	if (debug < 1)
5993 		return;
5994 
5995 	for (i = 0; i <= topo.max_cpu_num; ++i) {
5996 		if (cpu_is_not_present(i))
5997 			continue;
5998 		fprintf(outf,
5999 			"cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n",
6000 			i, cpus[i].physical_package_id, cpus[i].die_id,
6001 			cpus[i].physical_node_id, cpus[i].logical_node_id, cpus[i].physical_core_id, cpus[i].thread_id);
6002 	}
6003 
6004 }
6005 
6006 void allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
6007 {
6008 	int i;
6009 	int num_cores = topo.cores_per_node * topo.nodes_per_pkg * topo.num_packages;
6010 	int num_threads = topo.threads_per_core * num_cores;
6011 
6012 	*t = calloc(num_threads, sizeof(struct thread_data));
6013 	if (*t == NULL)
6014 		goto error;
6015 
6016 	for (i = 0; i < num_threads; i++)
6017 		(*t)[i].cpu_id = -1;
6018 
6019 	*c = calloc(num_cores, sizeof(struct core_data));
6020 	if (*c == NULL)
6021 		goto error;
6022 
6023 	for (i = 0; i < num_cores; i++)
6024 		(*c)[i].core_id = -1;
6025 
6026 	*p = calloc(topo.num_packages, sizeof(struct pkg_data));
6027 	if (*p == NULL)
6028 		goto error;
6029 
6030 	for (i = 0; i < topo.num_packages; i++)
6031 		(*p)[i].package_id = i;
6032 
6033 	return;
6034 error:
6035 	err(1, "calloc counters");
6036 }
6037 
6038 /*
6039  * init_counter()
6040  *
6041  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
6042  */
6043 void init_counter(struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base, int cpu_id)
6044 {
6045 	int pkg_id = cpus[cpu_id].physical_package_id;
6046 	int node_id = cpus[cpu_id].logical_node_id;
6047 	int core_id = cpus[cpu_id].physical_core_id;
6048 	int thread_id = cpus[cpu_id].thread_id;
6049 	struct thread_data *t;
6050 	struct core_data *c;
6051 	struct pkg_data *p;
6052 
6053 	/* Workaround for systems where physical_node_id==-1
6054 	 * and logical_node_id==(-1 - topo.num_cpus)
6055 	 */
6056 	if (node_id < 0)
6057 		node_id = 0;
6058 
6059 	t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id);
6060 	c = GET_CORE(core_base, core_id, node_id, pkg_id);
6061 	p = GET_PKG(pkg_base, pkg_id);
6062 
6063 	t->cpu_id = cpu_id;
6064 	if (thread_id == 0) {
6065 		t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
6066 		if (cpu_is_first_core_in_package(cpu_id))
6067 			t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
6068 	}
6069 
6070 	c->core_id = core_id;
6071 	p->package_id = pkg_id;
6072 }
6073 
6074 int initialize_counters(int cpu_id)
6075 {
6076 	init_counter(EVEN_COUNTERS, cpu_id);
6077 	init_counter(ODD_COUNTERS, cpu_id);
6078 	return 0;
6079 }
6080 
6081 void allocate_output_buffer()
6082 {
6083 	output_buffer = calloc(1, (1 + topo.num_cpus) * 2048);
6084 	outp = output_buffer;
6085 	if (outp == NULL)
6086 		err(-1, "calloc output buffer");
6087 }
6088 
6089 void allocate_fd_percpu(void)
6090 {
6091 	fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int));
6092 	if (fd_percpu == NULL)
6093 		err(-1, "calloc fd_percpu");
6094 }
6095 
6096 void allocate_irq_buffers(void)
6097 {
6098 	irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int));
6099 	if (irq_column_2_cpu == NULL)
6100 		err(-1, "calloc %d", topo.num_cpus);
6101 
6102 	irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int));
6103 	if (irqs_per_cpu == NULL)
6104 		err(-1, "calloc %d", topo.max_cpu_num + 1);
6105 }
6106 
6107 void setup_all_buffers(void)
6108 {
6109 	topology_probe();
6110 	allocate_irq_buffers();
6111 	allocate_fd_percpu();
6112 	allocate_counters(&thread_even, &core_even, &package_even);
6113 	allocate_counters(&thread_odd, &core_odd, &package_odd);
6114 	allocate_output_buffer();
6115 	for_all_proc_cpus(initialize_counters);
6116 }
6117 
6118 void set_base_cpu(void)
6119 {
6120 	base_cpu = sched_getcpu();
6121 	if (base_cpu < 0)
6122 		err(-ENODEV, "No valid cpus found");
6123 
6124 	if (debug > 1)
6125 		fprintf(outf, "base_cpu = %d\n", base_cpu);
6126 }
6127 
6128 void turbostat_init()
6129 {
6130 	setup_all_buffers();
6131 	set_base_cpu();
6132 	check_dev_msr();
6133 	check_permissions();
6134 	process_cpuid();
6135 	linux_perf_init();
6136 
6137 	if (!quiet)
6138 		for_all_cpus(print_hwp, ODD_COUNTERS);
6139 
6140 	if (!quiet)
6141 		for_all_cpus(print_epb, ODD_COUNTERS);
6142 
6143 	if (!quiet)
6144 		for_all_cpus(print_perf_limit, ODD_COUNTERS);
6145 
6146 	if (!quiet)
6147 		for_all_cpus(print_rapl, ODD_COUNTERS);
6148 
6149 	for_all_cpus(set_temperature_target, ODD_COUNTERS);
6150 
6151 	for_all_cpus(get_cpu_type, ODD_COUNTERS);
6152 	for_all_cpus(get_cpu_type, EVEN_COUNTERS);
6153 
6154 	if (!quiet)
6155 		for_all_cpus(print_thermal, ODD_COUNTERS);
6156 
6157 	if (!quiet && do_irtl_snb)
6158 		print_irtl();
6159 
6160 	if (DO_BIC(BIC_IPC))
6161 		(void)get_instr_count_fd(base_cpu);
6162 }
6163 
6164 int fork_it(char **argv)
6165 {
6166 	pid_t child_pid;
6167 	int status;
6168 
6169 	snapshot_proc_sysfs_files();
6170 	status = for_all_cpus(get_counters, EVEN_COUNTERS);
6171 	first_counter_read = 0;
6172 	if (status)
6173 		exit(status);
6174 	/* clear affinity side-effect of get_counters() */
6175 	sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
6176 	gettimeofday(&tv_even, (struct timezone *)NULL);
6177 
6178 	child_pid = fork();
6179 	if (!child_pid) {
6180 		/* child */
6181 		execvp(argv[0], argv);
6182 		err(errno, "exec %s", argv[0]);
6183 	} else {
6184 
6185 		/* parent */
6186 		if (child_pid == -1)
6187 			err(1, "fork");
6188 
6189 		signal(SIGINT, SIG_IGN);
6190 		signal(SIGQUIT, SIG_IGN);
6191 		if (waitpid(child_pid, &status, 0) == -1)
6192 			err(status, "waitpid");
6193 
6194 		if (WIFEXITED(status))
6195 			status = WEXITSTATUS(status);
6196 	}
6197 	/*
6198 	 * n.b. fork_it() does not check for errors from for_all_cpus()
6199 	 * because re-starting is problematic when forking
6200 	 */
6201 	snapshot_proc_sysfs_files();
6202 	for_all_cpus(get_counters, ODD_COUNTERS);
6203 	gettimeofday(&tv_odd, (struct timezone *)NULL);
6204 	timersub(&tv_odd, &tv_even, &tv_delta);
6205 	if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS))
6206 		fprintf(outf, "%s: Counter reset detected\n", progname);
6207 	else {
6208 		compute_average(EVEN_COUNTERS);
6209 		format_all_counters(EVEN_COUNTERS);
6210 	}
6211 
6212 	fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec / 1000000.0);
6213 
6214 	flush_output_stderr();
6215 
6216 	return status;
6217 }
6218 
6219 int get_and_dump_counters(void)
6220 {
6221 	int status;
6222 
6223 	snapshot_proc_sysfs_files();
6224 	status = for_all_cpus(get_counters, ODD_COUNTERS);
6225 	if (status)
6226 		return status;
6227 
6228 	status = for_all_cpus(dump_counters, ODD_COUNTERS);
6229 	if (status)
6230 		return status;
6231 
6232 	flush_output_stdout();
6233 
6234 	return status;
6235 }
6236 
6237 void print_version()
6238 {
6239 	fprintf(outf, "turbostat version 2023.03.17 - Len Brown <lenb@kernel.org>\n");
6240 }
6241 
6242 #define COMMAND_LINE_SIZE 2048
6243 
6244 void print_bootcmd(void)
6245 {
6246 	char bootcmd[COMMAND_LINE_SIZE];
6247 	FILE *fp;
6248 	int ret;
6249 
6250 	memset(bootcmd, 0, COMMAND_LINE_SIZE);
6251 	fp = fopen("/proc/cmdline", "r");
6252 	if (!fp)
6253 		return;
6254 
6255 	ret = fread(bootcmd, sizeof(char), COMMAND_LINE_SIZE - 1, fp);
6256 	if (ret) {
6257 		bootcmd[ret] = '\0';
6258 		/* the last character is already '\n' */
6259 		fprintf(outf, "Kernel command line: %s", bootcmd);
6260 	}
6261 
6262 	fclose(fp);
6263 }
6264 
6265 int add_counter(unsigned int msr_num, char *path, char *name,
6266 		unsigned int width, enum counter_scope scope,
6267 		enum counter_type type, enum counter_format format, int flags)
6268 {
6269 	struct msr_counter *msrp;
6270 
6271 	msrp = calloc(1, sizeof(struct msr_counter));
6272 	if (msrp == NULL) {
6273 		perror("calloc");
6274 		exit(1);
6275 	}
6276 
6277 	msrp->msr_num = msr_num;
6278 	strncpy(msrp->name, name, NAME_BYTES - 1);
6279 	if (path)
6280 		strncpy(msrp->path, path, PATH_BYTES - 1);
6281 	msrp->width = width;
6282 	msrp->type = type;
6283 	msrp->format = format;
6284 	msrp->flags = flags;
6285 
6286 	switch (scope) {
6287 
6288 	case SCOPE_CPU:
6289 		msrp->next = sys.tp;
6290 		sys.tp = msrp;
6291 		sys.added_thread_counters++;
6292 		if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) {
6293 			fprintf(stderr, "exceeded max %d added thread counters\n", MAX_ADDED_COUNTERS);
6294 			exit(-1);
6295 		}
6296 		break;
6297 
6298 	case SCOPE_CORE:
6299 		msrp->next = sys.cp;
6300 		sys.cp = msrp;
6301 		sys.added_core_counters++;
6302 		if (sys.added_core_counters > MAX_ADDED_COUNTERS) {
6303 			fprintf(stderr, "exceeded max %d added core counters\n", MAX_ADDED_COUNTERS);
6304 			exit(-1);
6305 		}
6306 		break;
6307 
6308 	case SCOPE_PACKAGE:
6309 		msrp->next = sys.pp;
6310 		sys.pp = msrp;
6311 		sys.added_package_counters++;
6312 		if (sys.added_package_counters > MAX_ADDED_COUNTERS) {
6313 			fprintf(stderr, "exceeded max %d added package counters\n", MAX_ADDED_COUNTERS);
6314 			exit(-1);
6315 		}
6316 		break;
6317 	}
6318 
6319 	return 0;
6320 }
6321 
6322 void parse_add_command(char *add_command)
6323 {
6324 	int msr_num = 0;
6325 	char *path = NULL;
6326 	char name_buffer[NAME_BYTES] = "";
6327 	int width = 64;
6328 	int fail = 0;
6329 	enum counter_scope scope = SCOPE_CPU;
6330 	enum counter_type type = COUNTER_CYCLES;
6331 	enum counter_format format = FORMAT_DELTA;
6332 
6333 	while (add_command) {
6334 
6335 		if (sscanf(add_command, "msr0x%x", &msr_num) == 1)
6336 			goto next;
6337 
6338 		if (sscanf(add_command, "msr%d", &msr_num) == 1)
6339 			goto next;
6340 
6341 		if (*add_command == '/') {
6342 			path = add_command;
6343 			goto next;
6344 		}
6345 
6346 		if (sscanf(add_command, "u%d", &width) == 1) {
6347 			if ((width == 32) || (width == 64))
6348 				goto next;
6349 			width = 64;
6350 		}
6351 		if (!strncmp(add_command, "cpu", strlen("cpu"))) {
6352 			scope = SCOPE_CPU;
6353 			goto next;
6354 		}
6355 		if (!strncmp(add_command, "core", strlen("core"))) {
6356 			scope = SCOPE_CORE;
6357 			goto next;
6358 		}
6359 		if (!strncmp(add_command, "package", strlen("package"))) {
6360 			scope = SCOPE_PACKAGE;
6361 			goto next;
6362 		}
6363 		if (!strncmp(add_command, "cycles", strlen("cycles"))) {
6364 			type = COUNTER_CYCLES;
6365 			goto next;
6366 		}
6367 		if (!strncmp(add_command, "seconds", strlen("seconds"))) {
6368 			type = COUNTER_SECONDS;
6369 			goto next;
6370 		}
6371 		if (!strncmp(add_command, "usec", strlen("usec"))) {
6372 			type = COUNTER_USEC;
6373 			goto next;
6374 		}
6375 		if (!strncmp(add_command, "raw", strlen("raw"))) {
6376 			format = FORMAT_RAW;
6377 			goto next;
6378 		}
6379 		if (!strncmp(add_command, "delta", strlen("delta"))) {
6380 			format = FORMAT_DELTA;
6381 			goto next;
6382 		}
6383 		if (!strncmp(add_command, "percent", strlen("percent"))) {
6384 			format = FORMAT_PERCENT;
6385 			goto next;
6386 		}
6387 
6388 		if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) {	/* 18 < NAME_BYTES */
6389 			char *eos;
6390 
6391 			eos = strchr(name_buffer, ',');
6392 			if (eos)
6393 				*eos = '\0';
6394 			goto next;
6395 		}
6396 
6397 next:
6398 		add_command = strchr(add_command, ',');
6399 		if (add_command) {
6400 			*add_command = '\0';
6401 			add_command++;
6402 		}
6403 
6404 	}
6405 	if ((msr_num == 0) && (path == NULL)) {
6406 		fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n");
6407 		fail++;
6408 	}
6409 
6410 	/* generate default column header */
6411 	if (*name_buffer == '\0') {
6412 		if (width == 32)
6413 			sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
6414 		else
6415 			sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : "");
6416 	}
6417 
6418 	if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0))
6419 		fail++;
6420 
6421 	if (fail) {
6422 		help();
6423 		exit(1);
6424 	}
6425 }
6426 
6427 int is_deferred_add(char *name)
6428 {
6429 	int i;
6430 
6431 	for (i = 0; i < deferred_add_index; ++i)
6432 		if (!strcmp(name, deferred_add_names[i]))
6433 			return 1;
6434 	return 0;
6435 }
6436 
6437 int is_deferred_skip(char *name)
6438 {
6439 	int i;
6440 
6441 	for (i = 0; i < deferred_skip_index; ++i)
6442 		if (!strcmp(name, deferred_skip_names[i]))
6443 			return 1;
6444 	return 0;
6445 }
6446 
6447 void probe_sysfs(void)
6448 {
6449 	char path[64];
6450 	char name_buf[16];
6451 	FILE *input;
6452 	int state;
6453 	char *sp;
6454 
6455 	for (state = 10; state >= 0; --state) {
6456 
6457 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state);
6458 		input = fopen(path, "r");
6459 		if (input == NULL)
6460 			continue;
6461 		if (!fgets(name_buf, sizeof(name_buf), input))
6462 			err(1, "%s: failed to read file", path);
6463 
6464 		/* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
6465 		sp = strchr(name_buf, '-');
6466 		if (!sp)
6467 			sp = strchrnul(name_buf, '\n');
6468 		*sp = '%';
6469 		*(sp + 1) = '\0';
6470 
6471 		remove_underbar(name_buf);
6472 
6473 		fclose(input);
6474 
6475 		sprintf(path, "cpuidle/state%d/time", state);
6476 
6477 		if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf))
6478 			continue;
6479 
6480 		if (is_deferred_skip(name_buf))
6481 			continue;
6482 
6483 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC, FORMAT_PERCENT, SYSFS_PERCPU);
6484 	}
6485 
6486 	for (state = 10; state >= 0; --state) {
6487 
6488 		sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state);
6489 		input = fopen(path, "r");
6490 		if (input == NULL)
6491 			continue;
6492 		if (!fgets(name_buf, sizeof(name_buf), input))
6493 			err(1, "%s: failed to read file", path);
6494 		/* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */
6495 		sp = strchr(name_buf, '-');
6496 		if (!sp)
6497 			sp = strchrnul(name_buf, '\n');
6498 		*sp = '\0';
6499 		fclose(input);
6500 
6501 		remove_underbar(name_buf);
6502 
6503 		sprintf(path, "cpuidle/state%d/usage", state);
6504 
6505 		if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf))
6506 			continue;
6507 
6508 		if (is_deferred_skip(name_buf))
6509 			continue;
6510 
6511 		add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU);
6512 	}
6513 
6514 }
6515 
6516 /*
6517  * parse cpuset with following syntax
6518  * 1,2,4..6,8-10 and set bits in cpu_subset
6519  */
6520 void parse_cpu_command(char *optarg)
6521 {
6522 	unsigned int start, end;
6523 	char *next;
6524 
6525 	if (!strcmp(optarg, "core")) {
6526 		if (cpu_subset)
6527 			goto error;
6528 		show_core_only++;
6529 		return;
6530 	}
6531 	if (!strcmp(optarg, "package")) {
6532 		if (cpu_subset)
6533 			goto error;
6534 		show_pkg_only++;
6535 		return;
6536 	}
6537 	if (show_core_only || show_pkg_only)
6538 		goto error;
6539 
6540 	cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS);
6541 	if (cpu_subset == NULL)
6542 		err(3, "CPU_ALLOC");
6543 	cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS);
6544 
6545 	CPU_ZERO_S(cpu_subset_size, cpu_subset);
6546 
6547 	next = optarg;
6548 
6549 	while (next && *next) {
6550 
6551 		if (*next == '-')	/* no negative cpu numbers */
6552 			goto error;
6553 
6554 		start = strtoul(next, &next, 10);
6555 
6556 		if (start >= CPU_SUBSET_MAXCPUS)
6557 			goto error;
6558 		CPU_SET_S(start, cpu_subset_size, cpu_subset);
6559 
6560 		if (*next == '\0')
6561 			break;
6562 
6563 		if (*next == ',') {
6564 			next += 1;
6565 			continue;
6566 		}
6567 
6568 		if (*next == '-') {
6569 			next += 1;	/* start range */
6570 		} else if (*next == '.') {
6571 			next += 1;
6572 			if (*next == '.')
6573 				next += 1;	/* start range */
6574 			else
6575 				goto error;
6576 		}
6577 
6578 		end = strtoul(next, &next, 10);
6579 		if (end <= start)
6580 			goto error;
6581 
6582 		while (++start <= end) {
6583 			if (start >= CPU_SUBSET_MAXCPUS)
6584 				goto error;
6585 			CPU_SET_S(start, cpu_subset_size, cpu_subset);
6586 		}
6587 
6588 		if (*next == ',')
6589 			next += 1;
6590 		else if (*next != '\0')
6591 			goto error;
6592 	}
6593 
6594 	return;
6595 
6596 error:
6597 	fprintf(stderr, "\"--cpu %s\" malformed\n", optarg);
6598 	help();
6599 	exit(-1);
6600 }
6601 
6602 void cmdline(int argc, char **argv)
6603 {
6604 	int opt;
6605 	int option_index = 0;
6606 	static struct option long_options[] = {
6607 		{ "add", required_argument, 0, 'a' },
6608 		{ "cpu", required_argument, 0, 'c' },
6609 		{ "Dump", no_argument, 0, 'D' },
6610 		{ "debug", no_argument, 0, 'd' },	/* internal, not documented */
6611 		{ "enable", required_argument, 0, 'e' },
6612 		{ "interval", required_argument, 0, 'i' },
6613 		{ "IPC", no_argument, 0, 'I' },
6614 		{ "num_iterations", required_argument, 0, 'n' },
6615 		{ "header_iterations", required_argument, 0, 'N' },
6616 		{ "help", no_argument, 0, 'h' },
6617 		{ "hide", required_argument, 0, 'H' },	// meh, -h taken by --help
6618 		{ "Joules", no_argument, 0, 'J' },
6619 		{ "list", no_argument, 0, 'l' },
6620 		{ "out", required_argument, 0, 'o' },
6621 		{ "quiet", no_argument, 0, 'q' },
6622 		{ "show", required_argument, 0, 's' },
6623 		{ "Summary", no_argument, 0, 'S' },
6624 		{ "TCC", required_argument, 0, 'T' },
6625 		{ "version", no_argument, 0, 'v' },
6626 		{ 0, 0, 0, 0 }
6627 	};
6628 
6629 	progname = argv[0];
6630 
6631 	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v", long_options, &option_index)) != -1) {
6632 		switch (opt) {
6633 		case 'a':
6634 			parse_add_command(optarg);
6635 			break;
6636 		case 'c':
6637 			parse_cpu_command(optarg);
6638 			break;
6639 		case 'D':
6640 			dump_only++;
6641 			break;
6642 		case 'e':
6643 			/* --enable specified counter */
6644 			bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST);
6645 			break;
6646 		case 'd':
6647 			debug++;
6648 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
6649 			break;
6650 		case 'H':
6651 			/*
6652 			 * --hide: do not show those specified
6653 			 *  multiple invocations simply clear more bits in enabled mask
6654 			 */
6655 			bic_enabled &= ~bic_lookup(optarg, HIDE_LIST);
6656 			break;
6657 		case 'h':
6658 		default:
6659 			help();
6660 			exit(1);
6661 		case 'i':
6662 			{
6663 				double interval = strtod(optarg, NULL);
6664 
6665 				if (interval < 0.001) {
6666 					fprintf(outf, "interval %f seconds is too small\n", interval);
6667 					exit(2);
6668 				}
6669 
6670 				interval_tv.tv_sec = interval_ts.tv_sec = interval;
6671 				interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000;
6672 				interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000;
6673 			}
6674 			break;
6675 		case 'J':
6676 			rapl_joules++;
6677 			break;
6678 		case 'l':
6679 			ENABLE_BIC(BIC_DISABLED_BY_DEFAULT);
6680 			list_header_only++;
6681 			quiet++;
6682 			break;
6683 		case 'o':
6684 			outf = fopen_or_die(optarg, "w");
6685 			break;
6686 		case 'q':
6687 			quiet = 1;
6688 			break;
6689 		case 'n':
6690 			num_iterations = strtod(optarg, NULL);
6691 
6692 			if (num_iterations <= 0) {
6693 				fprintf(outf, "iterations %d should be positive number\n", num_iterations);
6694 				exit(2);
6695 			}
6696 			break;
6697 		case 'N':
6698 			header_iterations = strtod(optarg, NULL);
6699 
6700 			if (header_iterations <= 0) {
6701 				fprintf(outf, "iterations %d should be positive number\n", header_iterations);
6702 				exit(2);
6703 			}
6704 			break;
6705 		case 's':
6706 			/*
6707 			 * --show: show only those specified
6708 			 *  The 1st invocation will clear and replace the enabled mask
6709 			 *  subsequent invocations can add to it.
6710 			 */
6711 			if (shown == 0)
6712 				bic_enabled = bic_lookup(optarg, SHOW_LIST);
6713 			else
6714 				bic_enabled |= bic_lookup(optarg, SHOW_LIST);
6715 			shown = 1;
6716 			break;
6717 		case 'S':
6718 			summary_only++;
6719 			break;
6720 		case 'T':
6721 			tj_max_override = atoi(optarg);
6722 			break;
6723 		case 'v':
6724 			print_version();
6725 			exit(0);
6726 			break;
6727 		}
6728 	}
6729 }
6730 
6731 void set_rlimit(void)
6732 {
6733 	struct rlimit limit;
6734 
6735 	if (getrlimit(RLIMIT_NOFILE, &limit) < 0)
6736 		err(1, "Failed to get rlimit");
6737 
6738 	if (limit.rlim_max < MAX_NOFILE)
6739 		limit.rlim_max = MAX_NOFILE;
6740 	if (limit.rlim_cur < MAX_NOFILE)
6741 		limit.rlim_cur = MAX_NOFILE;
6742 
6743 	if (setrlimit(RLIMIT_NOFILE, &limit) < 0)
6744 		err(1, "Failed to set rlimit");
6745 }
6746 
6747 int main(int argc, char **argv)
6748 {
6749 	outf = stderr;
6750 	cmdline(argc, argv);
6751 
6752 	if (!quiet) {
6753 		print_version();
6754 		print_bootcmd();
6755 	}
6756 
6757 	probe_sysfs();
6758 
6759 	if (!getuid())
6760 		set_rlimit();
6761 
6762 	turbostat_init();
6763 
6764 	msr_sum_record();
6765 
6766 	/* dump counters and exit */
6767 	if (dump_only)
6768 		return get_and_dump_counters();
6769 
6770 	/* list header and exit */
6771 	if (list_header_only) {
6772 		print_header(",");
6773 		flush_output_stdout();
6774 		return 0;
6775 	}
6776 
6777 	/*
6778 	 * if any params left, it must be a command to fork
6779 	 */
6780 	if (argc - optind)
6781 		return fork_it(argv + optind);
6782 	else
6783 		turbostat_loop();
6784 
6785 	return 0;
6786 }
6787