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) 2022 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 enum counter_scope { SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE }; 57 enum counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC }; 58 enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT }; 59 60 struct msr_counter { 61 unsigned int msr_num; 62 char name[NAME_BYTES]; 63 char path[PATH_BYTES]; 64 unsigned int width; 65 enum counter_type type; 66 enum counter_format format; 67 struct msr_counter *next; 68 unsigned int flags; 69 #define FLAGS_HIDE (1 << 0) 70 #define FLAGS_SHOW (1 << 1) 71 #define SYSFS_PERCPU (1 << 1) 72 }; 73 74 struct msr_counter bic[] = { 75 { 0x0, "usec", "", 0, 0, 0, NULL, 0 }, 76 { 0x0, "Time_Of_Day_Seconds", "", 0, 0, 0, NULL, 0 }, 77 { 0x0, "Package", "", 0, 0, 0, NULL, 0 }, 78 { 0x0, "Node", "", 0, 0, 0, NULL, 0 }, 79 { 0x0, "Avg_MHz", "", 0, 0, 0, NULL, 0 }, 80 { 0x0, "Busy%", "", 0, 0, 0, NULL, 0 }, 81 { 0x0, "Bzy_MHz", "", 0, 0, 0, NULL, 0 }, 82 { 0x0, "TSC_MHz", "", 0, 0, 0, NULL, 0 }, 83 { 0x0, "IRQ", "", 0, 0, 0, NULL, 0 }, 84 { 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL, 0 }, 85 { 0x0, "sysfs", "", 0, 0, 0, NULL, 0 }, 86 { 0x0, "CPU%c1", "", 0, 0, 0, NULL, 0 }, 87 { 0x0, "CPU%c3", "", 0, 0, 0, NULL, 0 }, 88 { 0x0, "CPU%c6", "", 0, 0, 0, NULL, 0 }, 89 { 0x0, "CPU%c7", "", 0, 0, 0, NULL, 0 }, 90 { 0x0, "ThreadC", "", 0, 0, 0, NULL, 0 }, 91 { 0x0, "CoreTmp", "", 0, 0, 0, NULL, 0 }, 92 { 0x0, "CoreCnt", "", 0, 0, 0, NULL, 0 }, 93 { 0x0, "PkgTmp", "", 0, 0, 0, NULL, 0 }, 94 { 0x0, "GFX%rc6", "", 0, 0, 0, NULL, 0 }, 95 { 0x0, "GFXMHz", "", 0, 0, 0, NULL, 0 }, 96 { 0x0, "Pkg%pc2", "", 0, 0, 0, NULL, 0 }, 97 { 0x0, "Pkg%pc3", "", 0, 0, 0, NULL, 0 }, 98 { 0x0, "Pkg%pc6", "", 0, 0, 0, NULL, 0 }, 99 { 0x0, "Pkg%pc7", "", 0, 0, 0, NULL, 0 }, 100 { 0x0, "Pkg%pc8", "", 0, 0, 0, NULL, 0 }, 101 { 0x0, "Pkg%pc9", "", 0, 0, 0, NULL, 0 }, 102 { 0x0, "Pk%pc10", "", 0, 0, 0, NULL, 0 }, 103 { 0x0, "CPU%LPI", "", 0, 0, 0, NULL, 0 }, 104 { 0x0, "SYS%LPI", "", 0, 0, 0, NULL, 0 }, 105 { 0x0, "PkgWatt", "", 0, 0, 0, NULL, 0 }, 106 { 0x0, "CorWatt", "", 0, 0, 0, NULL, 0 }, 107 { 0x0, "GFXWatt", "", 0, 0, 0, NULL, 0 }, 108 { 0x0, "PkgCnt", "", 0, 0, 0, NULL, 0 }, 109 { 0x0, "RAMWatt", "", 0, 0, 0, NULL, 0 }, 110 { 0x0, "PKG_%", "", 0, 0, 0, NULL, 0 }, 111 { 0x0, "RAM_%", "", 0, 0, 0, NULL, 0 }, 112 { 0x0, "Pkg_J", "", 0, 0, 0, NULL, 0 }, 113 { 0x0, "Cor_J", "", 0, 0, 0, NULL, 0 }, 114 { 0x0, "GFX_J", "", 0, 0, 0, NULL, 0 }, 115 { 0x0, "RAM_J", "", 0, 0, 0, NULL, 0 }, 116 { 0x0, "Mod%c6", "", 0, 0, 0, NULL, 0 }, 117 { 0x0, "Totl%C0", "", 0, 0, 0, NULL, 0 }, 118 { 0x0, "Any%C0", "", 0, 0, 0, NULL, 0 }, 119 { 0x0, "GFX%C0", "", 0, 0, 0, NULL, 0 }, 120 { 0x0, "CPUGFX%", "", 0, 0, 0, NULL, 0 }, 121 { 0x0, "Core", "", 0, 0, 0, NULL, 0 }, 122 { 0x0, "CPU", "", 0, 0, 0, NULL, 0 }, 123 { 0x0, "APIC", "", 0, 0, 0, NULL, 0 }, 124 { 0x0, "X2APIC", "", 0, 0, 0, NULL, 0 }, 125 { 0x0, "Die", "", 0, 0, 0, NULL, 0 }, 126 { 0x0, "GFXAMHz", "", 0, 0, 0, NULL, 0 }, 127 { 0x0, "IPC", "", 0, 0, 0, NULL, 0 }, 128 { 0x0, "CoreThr", "", 0, 0, 0, NULL, 0 }, 129 { 0x0, "UncMHz", "", 0, 0, 0, NULL, 0 }, 130 }; 131 132 #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter)) 133 #define BIC_USEC (1ULL << 0) 134 #define BIC_TOD (1ULL << 1) 135 #define BIC_Package (1ULL << 2) 136 #define BIC_Node (1ULL << 3) 137 #define BIC_Avg_MHz (1ULL << 4) 138 #define BIC_Busy (1ULL << 5) 139 #define BIC_Bzy_MHz (1ULL << 6) 140 #define BIC_TSC_MHz (1ULL << 7) 141 #define BIC_IRQ (1ULL << 8) 142 #define BIC_SMI (1ULL << 9) 143 #define BIC_sysfs (1ULL << 10) 144 #define BIC_CPU_c1 (1ULL << 11) 145 #define BIC_CPU_c3 (1ULL << 12) 146 #define BIC_CPU_c6 (1ULL << 13) 147 #define BIC_CPU_c7 (1ULL << 14) 148 #define BIC_ThreadC (1ULL << 15) 149 #define BIC_CoreTmp (1ULL << 16) 150 #define BIC_CoreCnt (1ULL << 17) 151 #define BIC_PkgTmp (1ULL << 18) 152 #define BIC_GFX_rc6 (1ULL << 19) 153 #define BIC_GFXMHz (1ULL << 20) 154 #define BIC_Pkgpc2 (1ULL << 21) 155 #define BIC_Pkgpc3 (1ULL << 22) 156 #define BIC_Pkgpc6 (1ULL << 23) 157 #define BIC_Pkgpc7 (1ULL << 24) 158 #define BIC_Pkgpc8 (1ULL << 25) 159 #define BIC_Pkgpc9 (1ULL << 26) 160 #define BIC_Pkgpc10 (1ULL << 27) 161 #define BIC_CPU_LPI (1ULL << 28) 162 #define BIC_SYS_LPI (1ULL << 29) 163 #define BIC_PkgWatt (1ULL << 30) 164 #define BIC_CorWatt (1ULL << 31) 165 #define BIC_GFXWatt (1ULL << 32) 166 #define BIC_PkgCnt (1ULL << 33) 167 #define BIC_RAMWatt (1ULL << 34) 168 #define BIC_PKG__ (1ULL << 35) 169 #define BIC_RAM__ (1ULL << 36) 170 #define BIC_Pkg_J (1ULL << 37) 171 #define BIC_Cor_J (1ULL << 38) 172 #define BIC_GFX_J (1ULL << 39) 173 #define BIC_RAM_J (1ULL << 40) 174 #define BIC_Mod_c6 (1ULL << 41) 175 #define BIC_Totl_c0 (1ULL << 42) 176 #define BIC_Any_c0 (1ULL << 43) 177 #define BIC_GFX_c0 (1ULL << 44) 178 #define BIC_CPUGFX (1ULL << 45) 179 #define BIC_Core (1ULL << 46) 180 #define BIC_CPU (1ULL << 47) 181 #define BIC_APIC (1ULL << 48) 182 #define BIC_X2APIC (1ULL << 49) 183 #define BIC_Die (1ULL << 50) 184 #define BIC_GFXACTMHz (1ULL << 51) 185 #define BIC_IPC (1ULL << 52) 186 #define BIC_CORE_THROT_CNT (1ULL << 53) 187 #define BIC_UNCORE_MHZ (1ULL << 54) 188 189 #define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die ) 190 #define BIC_THERMAL_PWR ( BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__) 191 #define BIC_FREQUENCY ( BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_UNCORE_MHZ) 192 #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) 193 #define BIC_OTHER ( BIC_IRQ | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) 194 195 #define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) 196 197 unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT); 198 unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC; 199 200 #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME) 201 #define DO_BIC_READ(COUNTER_NAME) (bic_present & COUNTER_NAME) 202 #define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME) 203 #define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT) 204 #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT) 205 #define BIC_IS_ENABLED(COUNTER_BIT) (bic_enabled & COUNTER_BIT) 206 207 char *proc_stat = "/proc/stat"; 208 FILE *outf; 209 int *fd_percpu; 210 int *fd_instr_count_percpu; 211 struct timeval interval_tv = { 5, 0 }; 212 struct timespec interval_ts = { 5, 0 }; 213 214 /* Save original CPU model */ 215 unsigned int model_orig; 216 217 unsigned int num_iterations; 218 unsigned int header_iterations; 219 unsigned int debug; 220 unsigned int quiet; 221 unsigned int shown; 222 unsigned int sums_need_wide_columns; 223 unsigned int rapl_joules; 224 unsigned int summary_only; 225 unsigned int list_header_only; 226 unsigned int dump_only; 227 unsigned int do_snb_cstates; 228 unsigned int do_knl_cstates; 229 unsigned int do_slm_cstates; 230 unsigned int use_c1_residency_msr; 231 unsigned int has_aperf; 232 unsigned int has_epb; 233 unsigned int is_hybrid; 234 unsigned int do_irtl_snb; 235 unsigned int do_irtl_hsw; 236 unsigned int units = 1000000; /* MHz etc */ 237 unsigned int genuine_intel; 238 unsigned int authentic_amd; 239 unsigned int hygon_genuine; 240 unsigned int max_level, max_extended_level; 241 unsigned int has_invariant_tsc; 242 unsigned int do_nhm_platform_info; 243 unsigned int no_MSR_MISC_PWR_MGMT; 244 unsigned int aperf_mperf_multiplier = 1; 245 double bclk; 246 double base_hz; 247 unsigned int has_base_hz; 248 double tsc_tweak = 1.0; 249 unsigned int show_pkg_only; 250 unsigned int show_core_only; 251 char *output_buffer, *outp; 252 unsigned int do_rapl; 253 unsigned int do_dts; 254 unsigned int do_ptm; 255 unsigned int do_ipc; 256 unsigned long long gfx_cur_rc6_ms; 257 unsigned long long cpuidle_cur_cpu_lpi_us; 258 unsigned long long cpuidle_cur_sys_lpi_us; 259 unsigned int gfx_cur_mhz; 260 unsigned int gfx_act_mhz; 261 unsigned int tj_max; 262 unsigned int tj_max_override; 263 int tcc_offset_bits; 264 double rapl_power_units, rapl_time_units; 265 double rapl_dram_energy_units, rapl_energy_units; 266 double rapl_joule_counter_range; 267 unsigned int do_core_perf_limit_reasons; 268 unsigned int has_automatic_cstate_conversion; 269 unsigned int dis_cstate_prewake; 270 unsigned int do_gfx_perf_limit_reasons; 271 unsigned int do_ring_perf_limit_reasons; 272 unsigned int crystal_hz; 273 unsigned long long tsc_hz; 274 int base_cpu; 275 double discover_bclk(unsigned int family, unsigned int model); 276 unsigned int has_hwp; /* IA32_PM_ENABLE, IA32_HWP_CAPABILITIES */ 277 /* IA32_HWP_REQUEST, IA32_HWP_STATUS */ 278 unsigned int has_hwp_notify; /* IA32_HWP_INTERRUPT */ 279 unsigned int has_hwp_activity_window; /* IA32_HWP_REQUEST[bits 41:32] */ 280 unsigned int has_hwp_epp; /* IA32_HWP_REQUEST[bits 31:24] */ 281 unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */ 282 unsigned int has_misc_feature_control; 283 unsigned int first_counter_read = 1; 284 int ignore_stdin; 285 286 #define RAPL_PKG (1 << 0) 287 /* 0x610 MSR_PKG_POWER_LIMIT */ 288 /* 0x611 MSR_PKG_ENERGY_STATUS */ 289 #define RAPL_PKG_PERF_STATUS (1 << 1) 290 /* 0x613 MSR_PKG_PERF_STATUS */ 291 #define RAPL_PKG_POWER_INFO (1 << 2) 292 /* 0x614 MSR_PKG_POWER_INFO */ 293 294 #define RAPL_DRAM (1 << 3) 295 /* 0x618 MSR_DRAM_POWER_LIMIT */ 296 /* 0x619 MSR_DRAM_ENERGY_STATUS */ 297 #define RAPL_DRAM_PERF_STATUS (1 << 4) 298 /* 0x61b MSR_DRAM_PERF_STATUS */ 299 #define RAPL_DRAM_POWER_INFO (1 << 5) 300 /* 0x61c MSR_DRAM_POWER_INFO */ 301 302 #define RAPL_CORES_POWER_LIMIT (1 << 6) 303 /* 0x638 MSR_PP0_POWER_LIMIT */ 304 #define RAPL_CORE_POLICY (1 << 7) 305 /* 0x63a MSR_PP0_POLICY */ 306 307 #define RAPL_GFX (1 << 8) 308 /* 0x640 MSR_PP1_POWER_LIMIT */ 309 /* 0x641 MSR_PP1_ENERGY_STATUS */ 310 /* 0x642 MSR_PP1_POLICY */ 311 312 #define RAPL_CORES_ENERGY_STATUS (1 << 9) 313 /* 0x639 MSR_PP0_ENERGY_STATUS */ 314 #define RAPL_PER_CORE_ENERGY (1 << 10) 315 /* Indicates cores energy collection is per-core, 316 * not per-package. */ 317 #define RAPL_AMD_F17H (1 << 11) 318 /* 0xc0010299 MSR_RAPL_PWR_UNIT */ 319 /* 0xc001029a MSR_CORE_ENERGY_STAT */ 320 /* 0xc001029b MSR_PKG_ENERGY_STAT */ 321 #define RAPL_CORES (RAPL_CORES_ENERGY_STATUS | RAPL_CORES_POWER_LIMIT) 322 #define TJMAX_DEFAULT 100 323 324 /* MSRs that are not yet in the kernel-provided header. */ 325 #define MSR_RAPL_PWR_UNIT 0xc0010299 326 #define MSR_CORE_ENERGY_STAT 0xc001029a 327 #define MSR_PKG_ENERGY_STAT 0xc001029b 328 329 #define MAX(a, b) ((a) > (b) ? (a) : (b)) 330 331 int backwards_count; 332 char *progname; 333 334 #define CPU_SUBSET_MAXCPUS 1024 /* need to use before probe... */ 335 cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_subset; 336 size_t cpu_present_setsize, cpu_affinity_setsize, cpu_subset_size; 337 #define MAX_ADDED_COUNTERS 8 338 #define MAX_ADDED_THREAD_COUNTERS 24 339 #define BITMASK_SIZE 32 340 341 struct thread_data { 342 struct timeval tv_begin; 343 struct timeval tv_end; 344 struct timeval tv_delta; 345 unsigned long long tsc; 346 unsigned long long aperf; 347 unsigned long long mperf; 348 unsigned long long c1; 349 unsigned long long instr_count; 350 unsigned long long irq_count; 351 unsigned int smi_count; 352 unsigned int cpu_id; 353 unsigned int apic_id; 354 unsigned int x2apic_id; 355 unsigned int flags; 356 bool is_atom; 357 #define CPU_IS_FIRST_THREAD_IN_CORE 0x2 358 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 359 unsigned long long counter[MAX_ADDED_THREAD_COUNTERS]; 360 } *thread_even, *thread_odd; 361 362 struct core_data { 363 unsigned long long c3; 364 unsigned long long c6; 365 unsigned long long c7; 366 unsigned long long mc6_us; /* duplicate as per-core for now, even though per module */ 367 unsigned int core_temp_c; 368 unsigned int core_energy; /* MSR_CORE_ENERGY_STAT */ 369 unsigned int core_id; 370 unsigned long long core_throt_cnt; 371 unsigned long long counter[MAX_ADDED_COUNTERS]; 372 } *core_even, *core_odd; 373 374 struct pkg_data { 375 unsigned long long pc2; 376 unsigned long long pc3; 377 unsigned long long pc6; 378 unsigned long long pc7; 379 unsigned long long pc8; 380 unsigned long long pc9; 381 unsigned long long pc10; 382 unsigned long long cpu_lpi; 383 unsigned long long sys_lpi; 384 unsigned long long pkg_wtd_core_c0; 385 unsigned long long pkg_any_core_c0; 386 unsigned long long pkg_any_gfxe_c0; 387 unsigned long long pkg_both_core_gfxe_c0; 388 long long gfx_rc6_ms; 389 unsigned int gfx_mhz; 390 unsigned int gfx_act_mhz; 391 unsigned int package_id; 392 unsigned long long energy_pkg; /* MSR_PKG_ENERGY_STATUS */ 393 unsigned long long energy_dram; /* MSR_DRAM_ENERGY_STATUS */ 394 unsigned long long energy_cores; /* MSR_PP0_ENERGY_STATUS */ 395 unsigned long long energy_gfx; /* MSR_PP1_ENERGY_STATUS */ 396 unsigned long long rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ 397 unsigned long long rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ 398 unsigned int pkg_temp_c; 399 unsigned int uncore_mhz; 400 unsigned long long counter[MAX_ADDED_COUNTERS]; 401 } *package_even, *package_odd; 402 403 #define ODD_COUNTERS thread_odd, core_odd, package_odd 404 #define EVEN_COUNTERS thread_even, core_even, package_even 405 406 #define GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no) \ 407 ((thread_base) + \ 408 ((pkg_no) * \ 409 topo.nodes_per_pkg * topo.cores_per_node * topo.threads_per_core) + \ 410 ((node_no) * topo.cores_per_node * topo.threads_per_core) + \ 411 ((core_no) * topo.threads_per_core) + \ 412 (thread_no)) 413 414 #define GET_CORE(core_base, core_no, node_no, pkg_no) \ 415 ((core_base) + \ 416 ((pkg_no) * topo.nodes_per_pkg * topo.cores_per_node) + \ 417 ((node_no) * topo.cores_per_node) + \ 418 (core_no)) 419 420 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) 421 422 /* 423 * The accumulated sum of MSR is defined as a monotonic 424 * increasing MSR, it will be accumulated periodically, 425 * despite its register's bit width. 426 */ 427 enum { 428 IDX_PKG_ENERGY, 429 IDX_DRAM_ENERGY, 430 IDX_PP0_ENERGY, 431 IDX_PP1_ENERGY, 432 IDX_PKG_PERF, 433 IDX_DRAM_PERF, 434 IDX_COUNT, 435 }; 436 437 int get_msr_sum(int cpu, off_t offset, unsigned long long *msr); 438 439 struct msr_sum_array { 440 /* get_msr_sum() = sum + (get_msr() - last) */ 441 struct { 442 /*The accumulated MSR value is updated by the timer */ 443 unsigned long long sum; 444 /*The MSR footprint recorded in last timer */ 445 unsigned long long last; 446 } entries[IDX_COUNT]; 447 }; 448 449 /* The percpu MSR sum array.*/ 450 struct msr_sum_array *per_cpu_msr_sum; 451 452 off_t idx_to_offset(int idx) 453 { 454 off_t offset; 455 456 switch (idx) { 457 case IDX_PKG_ENERGY: 458 if (do_rapl & RAPL_AMD_F17H) 459 offset = MSR_PKG_ENERGY_STAT; 460 else 461 offset = MSR_PKG_ENERGY_STATUS; 462 break; 463 case IDX_DRAM_ENERGY: 464 offset = MSR_DRAM_ENERGY_STATUS; 465 break; 466 case IDX_PP0_ENERGY: 467 offset = MSR_PP0_ENERGY_STATUS; 468 break; 469 case IDX_PP1_ENERGY: 470 offset = MSR_PP1_ENERGY_STATUS; 471 break; 472 case IDX_PKG_PERF: 473 offset = MSR_PKG_PERF_STATUS; 474 break; 475 case IDX_DRAM_PERF: 476 offset = MSR_DRAM_PERF_STATUS; 477 break; 478 default: 479 offset = -1; 480 } 481 return offset; 482 } 483 484 int offset_to_idx(off_t offset) 485 { 486 int idx; 487 488 switch (offset) { 489 case MSR_PKG_ENERGY_STATUS: 490 case MSR_PKG_ENERGY_STAT: 491 idx = IDX_PKG_ENERGY; 492 break; 493 case MSR_DRAM_ENERGY_STATUS: 494 idx = IDX_DRAM_ENERGY; 495 break; 496 case MSR_PP0_ENERGY_STATUS: 497 idx = IDX_PP0_ENERGY; 498 break; 499 case MSR_PP1_ENERGY_STATUS: 500 idx = IDX_PP1_ENERGY; 501 break; 502 case MSR_PKG_PERF_STATUS: 503 idx = IDX_PKG_PERF; 504 break; 505 case MSR_DRAM_PERF_STATUS: 506 idx = IDX_DRAM_PERF; 507 break; 508 default: 509 idx = -1; 510 } 511 return idx; 512 } 513 514 int idx_valid(int idx) 515 { 516 switch (idx) { 517 case IDX_PKG_ENERGY: 518 return do_rapl & (RAPL_PKG | RAPL_AMD_F17H); 519 case IDX_DRAM_ENERGY: 520 return do_rapl & RAPL_DRAM; 521 case IDX_PP0_ENERGY: 522 return do_rapl & RAPL_CORES_ENERGY_STATUS; 523 case IDX_PP1_ENERGY: 524 return do_rapl & RAPL_GFX; 525 case IDX_PKG_PERF: 526 return do_rapl & RAPL_PKG_PERF_STATUS; 527 case IDX_DRAM_PERF: 528 return do_rapl & RAPL_DRAM_PERF_STATUS; 529 default: 530 return 0; 531 } 532 } 533 534 struct sys_counters { 535 unsigned int added_thread_counters; 536 unsigned int added_core_counters; 537 unsigned int added_package_counters; 538 struct msr_counter *tp; 539 struct msr_counter *cp; 540 struct msr_counter *pp; 541 } sys; 542 543 struct system_summary { 544 struct thread_data threads; 545 struct core_data cores; 546 struct pkg_data packages; 547 } average; 548 549 struct cpu_topology { 550 int physical_package_id; 551 int die_id; 552 int logical_cpu_id; 553 int physical_node_id; 554 int logical_node_id; /* 0-based count within the package */ 555 int physical_core_id; 556 int thread_id; 557 cpu_set_t *put_ids; /* Processing Unit/Thread IDs */ 558 } *cpus; 559 560 struct topo_params { 561 int num_packages; 562 int num_die; 563 int num_cpus; 564 int num_cores; 565 int max_cpu_num; 566 int max_node_num; 567 int nodes_per_pkg; 568 int cores_per_node; 569 int threads_per_core; 570 } topo; 571 572 struct timeval tv_even, tv_odd, tv_delta; 573 574 int *irq_column_2_cpu; /* /proc/interrupts column numbers */ 575 int *irqs_per_cpu; /* indexed by cpu_num */ 576 577 void setup_all_buffers(void); 578 579 char *sys_lpi_file; 580 char *sys_lpi_file_sysfs = "/sys/devices/system/cpu/cpuidle/low_power_idle_system_residency_us"; 581 char *sys_lpi_file_debugfs = "/sys/kernel/debug/pmc_core/slp_s0_residency_usec"; 582 583 int cpu_is_not_present(int cpu) 584 { 585 return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); 586 } 587 588 /* 589 * run func(thread, core, package) in topology order 590 * skip non-present cpus 591 */ 592 593 int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pkg_data *), 594 struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base) 595 { 596 int retval, pkg_no, core_no, thread_no, node_no; 597 598 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { 599 for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) { 600 for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { 601 for (thread_no = 0; thread_no < topo.threads_per_core; ++thread_no) { 602 struct thread_data *t; 603 struct core_data *c; 604 struct pkg_data *p; 605 606 t = GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no); 607 608 if (cpu_is_not_present(t->cpu_id)) 609 continue; 610 611 c = GET_CORE(core_base, core_no, node_no, pkg_no); 612 p = GET_PKG(pkg_base, pkg_no); 613 614 retval = func(t, c, p); 615 if (retval) 616 return retval; 617 } 618 } 619 } 620 } 621 return 0; 622 } 623 624 int cpu_migrate(int cpu) 625 { 626 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); 627 CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set); 628 if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) 629 return -1; 630 else 631 return 0; 632 } 633 634 int get_msr_fd(int cpu) 635 { 636 char pathname[32]; 637 int fd; 638 639 fd = fd_percpu[cpu]; 640 641 if (fd) 642 return fd; 643 644 sprintf(pathname, "/dev/cpu/%d/msr", cpu); 645 fd = open(pathname, O_RDONLY); 646 if (fd < 0) 647 err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname); 648 649 fd_percpu[cpu] = fd; 650 651 return fd; 652 } 653 654 static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags) 655 { 656 return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags); 657 } 658 659 static int perf_instr_count_open(int cpu_num) 660 { 661 struct perf_event_attr pea; 662 int fd; 663 664 memset(&pea, 0, sizeof(struct perf_event_attr)); 665 pea.type = PERF_TYPE_HARDWARE; 666 pea.size = sizeof(struct perf_event_attr); 667 pea.config = PERF_COUNT_HW_INSTRUCTIONS; 668 669 /* counter for cpu_num, including user + kernel and all processes */ 670 fd = perf_event_open(&pea, -1, cpu_num, -1, 0); 671 if (fd == -1) { 672 warn("cpu%d: perf instruction counter", cpu_num); 673 BIC_NOT_PRESENT(BIC_IPC); 674 } 675 676 return fd; 677 } 678 679 int get_instr_count_fd(int cpu) 680 { 681 if (fd_instr_count_percpu[cpu]) 682 return fd_instr_count_percpu[cpu]; 683 684 fd_instr_count_percpu[cpu] = perf_instr_count_open(cpu); 685 686 return fd_instr_count_percpu[cpu]; 687 } 688 689 int get_msr(int cpu, off_t offset, unsigned long long *msr) 690 { 691 ssize_t retval; 692 693 retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset); 694 695 if (retval != sizeof *msr) 696 err(-1, "cpu%d: msr offset 0x%llx read failed", cpu, (unsigned long long)offset); 697 698 return 0; 699 } 700 701 #define MAX_DEFERRED 16 702 char *deferred_add_names[MAX_DEFERRED]; 703 char *deferred_skip_names[MAX_DEFERRED]; 704 int deferred_add_index; 705 int deferred_skip_index; 706 707 /* 708 * HIDE_LIST - hide this list of counters, show the rest [default] 709 * SHOW_LIST - show this list of counters, hide the rest 710 */ 711 enum show_hide_mode { SHOW_LIST, HIDE_LIST } global_show_hide_mode = HIDE_LIST; 712 713 void help(void) 714 { 715 fprintf(outf, 716 "Usage: turbostat [OPTIONS][(--interval seconds) | COMMAND ...]\n" 717 "\n" 718 "Turbostat forks the specified COMMAND and prints statistics\n" 719 "when COMMAND completes.\n" 720 "If no COMMAND is specified, turbostat wakes every 5-seconds\n" 721 "to print statistics, until interrupted.\n" 722 " -a, --add add a counter\n" 723 " eg. --add msr0x10,u64,cpu,delta,MY_TSC\n" 724 " -c, --cpu cpu-set limit output to summary plus cpu-set:\n" 725 " {core | package | j,k,l..m,n-p }\n" 726 " -d, --debug displays usec, Time_Of_Day_Seconds and more debugging\n" 727 " -D, --Dump displays the raw counter values\n" 728 " -e, --enable [all | column]\n" 729 " shows all or the specified disabled column\n" 730 " -H, --hide [column|column,column,...]\n" 731 " hide the specified column(s)\n" 732 " -i, --interval sec.subsec\n" 733 " Override default 5-second measurement interval\n" 734 " -J, --Joules displays energy in Joules instead of Watts\n" 735 " -l, --list list column headers only\n" 736 " -n, --num_iterations num\n" 737 " number of the measurement iterations\n" 738 " -N, --header_iterations num\n" 739 " print header every num iterations\n" 740 " -o, --out file\n" 741 " create or truncate \"file\" for all output\n" 742 " -q, --quiet skip decoding system configuration header\n" 743 " -s, --show [column|column,column,...]\n" 744 " show only the specified column(s)\n" 745 " -S, --Summary\n" 746 " limits output to 1-line system summary per interval\n" 747 " -T, --TCC temperature\n" 748 " sets the Thermal Control Circuit temperature in\n" 749 " degrees Celsius\n" 750 " -h, --help print this help message\n" 751 " -v, --version print version information\n" "\n" "For more help, run \"man turbostat\"\n"); 752 } 753 754 /* 755 * bic_lookup 756 * for all the strings in comma separate name_list, 757 * set the approprate bit in return value. 758 */ 759 unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode) 760 { 761 unsigned int i; 762 unsigned long long retval = 0; 763 764 while (name_list) { 765 char *comma; 766 767 comma = strchr(name_list, ','); 768 769 if (comma) 770 *comma = '\0'; 771 772 for (i = 0; i < MAX_BIC; ++i) { 773 if (!strcmp(name_list, bic[i].name)) { 774 retval |= (1ULL << i); 775 break; 776 } 777 if (!strcmp(name_list, "all")) { 778 retval |= ~0; 779 break; 780 } else if (!strcmp(name_list, "topology")) { 781 retval |= BIC_TOPOLOGY; 782 break; 783 } else if (!strcmp(name_list, "power")) { 784 retval |= BIC_THERMAL_PWR; 785 break; 786 } else if (!strcmp(name_list, "idle")) { 787 retval |= BIC_IDLE; 788 break; 789 } else if (!strcmp(name_list, "frequency")) { 790 retval |= BIC_FREQUENCY; 791 break; 792 } else if (!strcmp(name_list, "other")) { 793 retval |= BIC_OTHER; 794 break; 795 } 796 797 } 798 if (i == MAX_BIC) { 799 if (mode == SHOW_LIST) { 800 deferred_add_names[deferred_add_index++] = name_list; 801 if (deferred_add_index >= MAX_DEFERRED) { 802 fprintf(stderr, "More than max %d un-recognized --add options '%s'\n", 803 MAX_DEFERRED, name_list); 804 help(); 805 exit(1); 806 } 807 } else { 808 deferred_skip_names[deferred_skip_index++] = name_list; 809 if (debug) 810 fprintf(stderr, "deferred \"%s\"\n", name_list); 811 if (deferred_skip_index >= MAX_DEFERRED) { 812 fprintf(stderr, "More than max %d un-recognized --skip options '%s'\n", 813 MAX_DEFERRED, name_list); 814 help(); 815 exit(1); 816 } 817 } 818 } 819 820 name_list = comma; 821 if (name_list) 822 name_list++; 823 824 } 825 return retval; 826 } 827 828 void print_header(char *delim) 829 { 830 struct msr_counter *mp; 831 int printed = 0; 832 833 if (DO_BIC(BIC_USEC)) 834 outp += sprintf(outp, "%susec", (printed++ ? delim : "")); 835 if (DO_BIC(BIC_TOD)) 836 outp += sprintf(outp, "%sTime_Of_Day_Seconds", (printed++ ? delim : "")); 837 if (DO_BIC(BIC_Package)) 838 outp += sprintf(outp, "%sPackage", (printed++ ? delim : "")); 839 if (DO_BIC(BIC_Die)) 840 outp += sprintf(outp, "%sDie", (printed++ ? delim : "")); 841 if (DO_BIC(BIC_Node)) 842 outp += sprintf(outp, "%sNode", (printed++ ? delim : "")); 843 if (DO_BIC(BIC_Core)) 844 outp += sprintf(outp, "%sCore", (printed++ ? delim : "")); 845 if (DO_BIC(BIC_CPU)) 846 outp += sprintf(outp, "%sCPU", (printed++ ? delim : "")); 847 if (DO_BIC(BIC_APIC)) 848 outp += sprintf(outp, "%sAPIC", (printed++ ? delim : "")); 849 if (DO_BIC(BIC_X2APIC)) 850 outp += sprintf(outp, "%sX2APIC", (printed++ ? delim : "")); 851 if (DO_BIC(BIC_Avg_MHz)) 852 outp += sprintf(outp, "%sAvg_MHz", (printed++ ? delim : "")); 853 if (DO_BIC(BIC_Busy)) 854 outp += sprintf(outp, "%sBusy%%", (printed++ ? delim : "")); 855 if (DO_BIC(BIC_Bzy_MHz)) 856 outp += sprintf(outp, "%sBzy_MHz", (printed++ ? delim : "")); 857 if (DO_BIC(BIC_TSC_MHz)) 858 outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : "")); 859 860 if (DO_BIC(BIC_IPC)) 861 outp += sprintf(outp, "%sIPC", (printed++ ? delim : "")); 862 863 if (DO_BIC(BIC_IRQ)) { 864 if (sums_need_wide_columns) 865 outp += sprintf(outp, "%s IRQ", (printed++ ? delim : "")); 866 else 867 outp += sprintf(outp, "%sIRQ", (printed++ ? delim : "")); 868 } 869 870 if (DO_BIC(BIC_SMI)) 871 outp += sprintf(outp, "%sSMI", (printed++ ? delim : "")); 872 873 for (mp = sys.tp; mp; mp = mp->next) { 874 875 if (mp->format == FORMAT_RAW) { 876 if (mp->width == 64) 877 outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), mp->name); 878 else 879 outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), mp->name); 880 } else { 881 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 882 outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), mp->name); 883 else 884 outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), mp->name); 885 } 886 } 887 888 if (DO_BIC(BIC_CPU_c1)) 889 outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : "")); 890 if (DO_BIC(BIC_CPU_c3)) 891 outp += sprintf(outp, "%sCPU%%c3", (printed++ ? delim : "")); 892 if (DO_BIC(BIC_CPU_c6)) 893 outp += sprintf(outp, "%sCPU%%c6", (printed++ ? delim : "")); 894 if (DO_BIC(BIC_CPU_c7)) 895 outp += sprintf(outp, "%sCPU%%c7", (printed++ ? delim : "")); 896 897 if (DO_BIC(BIC_Mod_c6)) 898 outp += sprintf(outp, "%sMod%%c6", (printed++ ? delim : "")); 899 900 if (DO_BIC(BIC_CoreTmp)) 901 outp += sprintf(outp, "%sCoreTmp", (printed++ ? delim : "")); 902 903 if (DO_BIC(BIC_CORE_THROT_CNT)) 904 outp += sprintf(outp, "%sCoreThr", (printed++ ? delim : "")); 905 906 if (do_rapl && !rapl_joules) { 907 if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY)) 908 outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : "")); 909 } else if (do_rapl && rapl_joules) { 910 if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY)) 911 outp += sprintf(outp, "%sCor_J", (printed++ ? delim : "")); 912 } 913 914 for (mp = sys.cp; mp; mp = mp->next) { 915 if (mp->format == FORMAT_RAW) { 916 if (mp->width == 64) 917 outp += sprintf(outp, "%s%18.18s", delim, mp->name); 918 else 919 outp += sprintf(outp, "%s%10.10s", delim, mp->name); 920 } else { 921 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 922 outp += sprintf(outp, "%s%8s", delim, mp->name); 923 else 924 outp += sprintf(outp, "%s%s", delim, mp->name); 925 } 926 } 927 928 if (DO_BIC(BIC_PkgTmp)) 929 outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : "")); 930 931 if (DO_BIC(BIC_GFX_rc6)) 932 outp += sprintf(outp, "%sGFX%%rc6", (printed++ ? delim : "")); 933 934 if (DO_BIC(BIC_GFXMHz)) 935 outp += sprintf(outp, "%sGFXMHz", (printed++ ? delim : "")); 936 937 if (DO_BIC(BIC_GFXACTMHz)) 938 outp += sprintf(outp, "%sGFXAMHz", (printed++ ? delim : "")); 939 940 if (DO_BIC(BIC_Totl_c0)) 941 outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : "")); 942 if (DO_BIC(BIC_Any_c0)) 943 outp += sprintf(outp, "%sAny%%C0", (printed++ ? delim : "")); 944 if (DO_BIC(BIC_GFX_c0)) 945 outp += sprintf(outp, "%sGFX%%C0", (printed++ ? delim : "")); 946 if (DO_BIC(BIC_CPUGFX)) 947 outp += sprintf(outp, "%sCPUGFX%%", (printed++ ? delim : "")); 948 949 if (DO_BIC(BIC_Pkgpc2)) 950 outp += sprintf(outp, "%sPkg%%pc2", (printed++ ? delim : "")); 951 if (DO_BIC(BIC_Pkgpc3)) 952 outp += sprintf(outp, "%sPkg%%pc3", (printed++ ? delim : "")); 953 if (DO_BIC(BIC_Pkgpc6)) 954 outp += sprintf(outp, "%sPkg%%pc6", (printed++ ? delim : "")); 955 if (DO_BIC(BIC_Pkgpc7)) 956 outp += sprintf(outp, "%sPkg%%pc7", (printed++ ? delim : "")); 957 if (DO_BIC(BIC_Pkgpc8)) 958 outp += sprintf(outp, "%sPkg%%pc8", (printed++ ? delim : "")); 959 if (DO_BIC(BIC_Pkgpc9)) 960 outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : "")); 961 if (DO_BIC(BIC_Pkgpc10)) 962 outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : "")); 963 if (DO_BIC(BIC_CPU_LPI)) 964 outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : "")); 965 if (DO_BIC(BIC_SYS_LPI)) 966 outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : "")); 967 968 if (do_rapl && !rapl_joules) { 969 if (DO_BIC(BIC_PkgWatt)) 970 outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : "")); 971 if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY)) 972 outp += sprintf(outp, "%sCorWatt", (printed++ ? delim : "")); 973 if (DO_BIC(BIC_GFXWatt)) 974 outp += sprintf(outp, "%sGFXWatt", (printed++ ? delim : "")); 975 if (DO_BIC(BIC_RAMWatt)) 976 outp += sprintf(outp, "%sRAMWatt", (printed++ ? delim : "")); 977 if (DO_BIC(BIC_PKG__)) 978 outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : "")); 979 if (DO_BIC(BIC_RAM__)) 980 outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : "")); 981 } else if (do_rapl && rapl_joules) { 982 if (DO_BIC(BIC_Pkg_J)) 983 outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : "")); 984 if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY)) 985 outp += sprintf(outp, "%sCor_J", (printed++ ? delim : "")); 986 if (DO_BIC(BIC_GFX_J)) 987 outp += sprintf(outp, "%sGFX_J", (printed++ ? delim : "")); 988 if (DO_BIC(BIC_RAM_J)) 989 outp += sprintf(outp, "%sRAM_J", (printed++ ? delim : "")); 990 if (DO_BIC(BIC_PKG__)) 991 outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : "")); 992 if (DO_BIC(BIC_RAM__)) 993 outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : "")); 994 } 995 if (DO_BIC(BIC_UNCORE_MHZ)) 996 outp += sprintf(outp, "%sUncMHz", (printed++ ? delim : "")); 997 998 for (mp = sys.pp; mp; mp = mp->next) { 999 if (mp->format == FORMAT_RAW) { 1000 if (mp->width == 64) 1001 outp += sprintf(outp, "%s%18.18s", delim, mp->name); 1002 else 1003 outp += sprintf(outp, "%s%10.10s", delim, mp->name); 1004 } else { 1005 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 1006 outp += sprintf(outp, "%s%8s", delim, mp->name); 1007 else 1008 outp += sprintf(outp, "%s%s", delim, mp->name); 1009 } 1010 } 1011 1012 outp += sprintf(outp, "\n"); 1013 } 1014 1015 int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1016 { 1017 int i; 1018 struct msr_counter *mp; 1019 1020 outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p); 1021 1022 if (t) { 1023 outp += sprintf(outp, "CPU: %d flags 0x%x\n", t->cpu_id, t->flags); 1024 outp += sprintf(outp, "TSC: %016llX\n", t->tsc); 1025 outp += sprintf(outp, "aperf: %016llX\n", t->aperf); 1026 outp += sprintf(outp, "mperf: %016llX\n", t->mperf); 1027 outp += sprintf(outp, "c1: %016llX\n", t->c1); 1028 1029 if (DO_BIC(BIC_IPC)) 1030 outp += sprintf(outp, "IPC: %lld\n", t->instr_count); 1031 1032 if (DO_BIC(BIC_IRQ)) 1033 outp += sprintf(outp, "IRQ: %lld\n", t->irq_count); 1034 if (DO_BIC(BIC_SMI)) 1035 outp += sprintf(outp, "SMI: %d\n", t->smi_count); 1036 1037 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 1038 outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, t->counter[i]); 1039 } 1040 } 1041 1042 if (c) { 1043 outp += sprintf(outp, "core: %d\n", c->core_id); 1044 outp += sprintf(outp, "c3: %016llX\n", c->c3); 1045 outp += sprintf(outp, "c6: %016llX\n", c->c6); 1046 outp += sprintf(outp, "c7: %016llX\n", c->c7); 1047 outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c); 1048 outp += sprintf(outp, "cpu_throt_count: %016llX\n", c->core_throt_cnt); 1049 outp += sprintf(outp, "Joules: %0X\n", c->core_energy); 1050 1051 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 1052 outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, c->counter[i]); 1053 } 1054 outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us); 1055 } 1056 1057 if (p) { 1058 outp += sprintf(outp, "package: %d\n", p->package_id); 1059 1060 outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0); 1061 outp += sprintf(outp, "Any cores: %016llX\n", p->pkg_any_core_c0); 1062 outp += sprintf(outp, "Any GFX: %016llX\n", p->pkg_any_gfxe_c0); 1063 outp += sprintf(outp, "CPU + GFX: %016llX\n", p->pkg_both_core_gfxe_c0); 1064 1065 outp += sprintf(outp, "pc2: %016llX\n", p->pc2); 1066 if (DO_BIC(BIC_Pkgpc3)) 1067 outp += sprintf(outp, "pc3: %016llX\n", p->pc3); 1068 if (DO_BIC(BIC_Pkgpc6)) 1069 outp += sprintf(outp, "pc6: %016llX\n", p->pc6); 1070 if (DO_BIC(BIC_Pkgpc7)) 1071 outp += sprintf(outp, "pc7: %016llX\n", p->pc7); 1072 outp += sprintf(outp, "pc8: %016llX\n", p->pc8); 1073 outp += sprintf(outp, "pc9: %016llX\n", p->pc9); 1074 outp += sprintf(outp, "pc10: %016llX\n", p->pc10); 1075 outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi); 1076 outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi); 1077 outp += sprintf(outp, "Joules PKG: %0llX\n", p->energy_pkg); 1078 outp += sprintf(outp, "Joules COR: %0llX\n", p->energy_cores); 1079 outp += sprintf(outp, "Joules GFX: %0llX\n", p->energy_gfx); 1080 outp += sprintf(outp, "Joules RAM: %0llX\n", p->energy_dram); 1081 outp += sprintf(outp, "Throttle PKG: %0llX\n", p->rapl_pkg_perf_status); 1082 outp += sprintf(outp, "Throttle RAM: %0llX\n", p->rapl_dram_perf_status); 1083 outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); 1084 1085 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 1086 outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, p->counter[i]); 1087 } 1088 } 1089 1090 outp += sprintf(outp, "\n"); 1091 1092 return 0; 1093 } 1094 1095 /* 1096 * column formatting convention & formats 1097 */ 1098 int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1099 { 1100 double interval_float, tsc; 1101 char *fmt8; 1102 int i; 1103 struct msr_counter *mp; 1104 char *delim = "\t"; 1105 int printed = 0; 1106 1107 /* if showing only 1st thread in core and this isn't one, bail out */ 1108 if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 1109 return 0; 1110 1111 /* if showing only 1st thread in pkg and this isn't one, bail out */ 1112 if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 1113 return 0; 1114 1115 /*if not summary line and --cpu is used */ 1116 if ((t != &average.threads) && (cpu_subset && !CPU_ISSET_S(t->cpu_id, cpu_subset_size, cpu_subset))) 1117 return 0; 1118 1119 if (DO_BIC(BIC_USEC)) { 1120 /* on each row, print how many usec each timestamp took to gather */ 1121 struct timeval tv; 1122 1123 timersub(&t->tv_end, &t->tv_begin, &tv); 1124 outp += sprintf(outp, "%5ld\t", tv.tv_sec * 1000000 + tv.tv_usec); 1125 } 1126 1127 /* Time_Of_Day_Seconds: on each row, print sec.usec last timestamp taken */ 1128 if (DO_BIC(BIC_TOD)) 1129 outp += sprintf(outp, "%10ld.%06ld\t", t->tv_end.tv_sec, t->tv_end.tv_usec); 1130 1131 interval_float = t->tv_delta.tv_sec + t->tv_delta.tv_usec / 1000000.0; 1132 1133 tsc = t->tsc * tsc_tweak; 1134 1135 /* topo columns, print blanks on 1st (average) line */ 1136 if (t == &average.threads) { 1137 if (DO_BIC(BIC_Package)) 1138 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1139 if (DO_BIC(BIC_Die)) 1140 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1141 if (DO_BIC(BIC_Node)) 1142 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1143 if (DO_BIC(BIC_Core)) 1144 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1145 if (DO_BIC(BIC_CPU)) 1146 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1147 if (DO_BIC(BIC_APIC)) 1148 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1149 if (DO_BIC(BIC_X2APIC)) 1150 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1151 } else { 1152 if (DO_BIC(BIC_Package)) { 1153 if (p) 1154 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->package_id); 1155 else 1156 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1157 } 1158 if (DO_BIC(BIC_Die)) { 1159 if (c) 1160 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), cpus[t->cpu_id].die_id); 1161 else 1162 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1163 } 1164 if (DO_BIC(BIC_Node)) { 1165 if (t) 1166 outp += sprintf(outp, "%s%d", 1167 (printed++ ? delim : ""), cpus[t->cpu_id].physical_node_id); 1168 else 1169 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1170 } 1171 if (DO_BIC(BIC_Core)) { 1172 if (c) 1173 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_id); 1174 else 1175 outp += sprintf(outp, "%s-", (printed++ ? delim : "")); 1176 } 1177 if (DO_BIC(BIC_CPU)) 1178 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->cpu_id); 1179 if (DO_BIC(BIC_APIC)) 1180 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->apic_id); 1181 if (DO_BIC(BIC_X2APIC)) 1182 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->x2apic_id); 1183 } 1184 1185 if (DO_BIC(BIC_Avg_MHz)) 1186 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 / units * t->aperf / interval_float); 1187 1188 if (DO_BIC(BIC_Busy)) 1189 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->mperf / tsc); 1190 1191 if (DO_BIC(BIC_Bzy_MHz)) { 1192 if (has_base_hz) 1193 outp += 1194 sprintf(outp, "%s%.0f", (printed++ ? delim : ""), base_hz / units * t->aperf / t->mperf); 1195 else 1196 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1197 tsc / units * t->aperf / t->mperf / interval_float); 1198 } 1199 1200 if (DO_BIC(BIC_TSC_MHz)) 1201 outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc / units / interval_float); 1202 1203 if (DO_BIC(BIC_IPC)) 1204 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 1.0 * t->instr_count / t->aperf); 1205 1206 /* IRQ */ 1207 if (DO_BIC(BIC_IRQ)) { 1208 if (sums_need_wide_columns) 1209 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->irq_count); 1210 else 1211 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count); 1212 } 1213 1214 /* SMI */ 1215 if (DO_BIC(BIC_SMI)) 1216 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count); 1217 1218 /* Added counters */ 1219 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 1220 if (mp->format == FORMAT_RAW) { 1221 if (mp->width == 32) 1222 outp += 1223 sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)t->counter[i]); 1224 else 1225 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->counter[i]); 1226 } else if (mp->format == FORMAT_DELTA) { 1227 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 1228 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->counter[i]); 1229 else 1230 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->counter[i]); 1231 } else if (mp->format == FORMAT_PERCENT) { 1232 if (mp->type == COUNTER_USEC) 1233 outp += 1234 sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 1235 t->counter[i] / interval_float / 10000); 1236 else 1237 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->counter[i] / tsc); 1238 } 1239 } 1240 1241 /* C1 */ 1242 if (DO_BIC(BIC_CPU_c1)) 1243 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1 / tsc); 1244 1245 /* print per-core data only for 1st thread in core */ 1246 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 1247 goto done; 1248 1249 if (DO_BIC(BIC_CPU_c3)) 1250 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c3 / tsc); 1251 if (DO_BIC(BIC_CPU_c6)) 1252 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c6 / tsc); 1253 if (DO_BIC(BIC_CPU_c7)) 1254 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->c7 / tsc); 1255 1256 /* Mod%c6 */ 1257 if (DO_BIC(BIC_Mod_c6)) 1258 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->mc6_us / tsc); 1259 1260 if (DO_BIC(BIC_CoreTmp)) 1261 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), c->core_temp_c); 1262 1263 /* Core throttle count */ 1264 if (DO_BIC(BIC_CORE_THROT_CNT)) 1265 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->core_throt_cnt); 1266 1267 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 1268 if (mp->format == FORMAT_RAW) { 1269 if (mp->width == 32) 1270 outp += 1271 sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)c->counter[i]); 1272 else 1273 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->counter[i]); 1274 } else if (mp->format == FORMAT_DELTA) { 1275 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 1276 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->counter[i]); 1277 else 1278 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->counter[i]); 1279 } else if (mp->format == FORMAT_PERCENT) { 1280 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->counter[i] / tsc); 1281 } 1282 } 1283 1284 fmt8 = "%s%.2f"; 1285 1286 if (DO_BIC(BIC_CorWatt) && (do_rapl & RAPL_PER_CORE_ENERGY)) 1287 outp += 1288 sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float); 1289 if (DO_BIC(BIC_Cor_J) && (do_rapl & RAPL_PER_CORE_ENERGY)) 1290 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units); 1291 1292 /* print per-package data only for 1st core in package */ 1293 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 1294 goto done; 1295 1296 /* PkgTmp */ 1297 if (DO_BIC(BIC_PkgTmp)) 1298 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->pkg_temp_c); 1299 1300 /* GFXrc6 */ 1301 if (DO_BIC(BIC_GFX_rc6)) { 1302 if (p->gfx_rc6_ms == -1) { /* detect GFX counter reset */ 1303 outp += sprintf(outp, "%s**.**", (printed++ ? delim : "")); 1304 } else { 1305 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 1306 p->gfx_rc6_ms / 10.0 / interval_float); 1307 } 1308 } 1309 1310 /* GFXMHz */ 1311 if (DO_BIC(BIC_GFXMHz)) 1312 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_mhz); 1313 1314 /* GFXACTMHz */ 1315 if (DO_BIC(BIC_GFXACTMHz)) 1316 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_act_mhz); 1317 1318 /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */ 1319 if (DO_BIC(BIC_Totl_c0)) 1320 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0 / tsc); 1321 if (DO_BIC(BIC_Any_c0)) 1322 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_core_c0 / tsc); 1323 if (DO_BIC(BIC_GFX_c0)) 1324 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_any_gfxe_c0 / tsc); 1325 if (DO_BIC(BIC_CPUGFX)) 1326 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_both_core_gfxe_c0 / tsc); 1327 1328 if (DO_BIC(BIC_Pkgpc2)) 1329 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc2 / tsc); 1330 if (DO_BIC(BIC_Pkgpc3)) 1331 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc3 / tsc); 1332 if (DO_BIC(BIC_Pkgpc6)) 1333 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc6 / tsc); 1334 if (DO_BIC(BIC_Pkgpc7)) 1335 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc7 / tsc); 1336 if (DO_BIC(BIC_Pkgpc8)) 1337 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc8 / tsc); 1338 if (DO_BIC(BIC_Pkgpc9)) 1339 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc9 / tsc); 1340 if (DO_BIC(BIC_Pkgpc10)) 1341 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10 / tsc); 1342 1343 if (DO_BIC(BIC_CPU_LPI)) 1344 outp += 1345 sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float); 1346 if (DO_BIC(BIC_SYS_LPI)) 1347 outp += 1348 sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float); 1349 1350 if (DO_BIC(BIC_PkgWatt)) 1351 outp += 1352 sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float); 1353 1354 if (DO_BIC(BIC_CorWatt) && !(do_rapl & RAPL_PER_CORE_ENERGY)) 1355 outp += 1356 sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float); 1357 if (DO_BIC(BIC_GFXWatt)) 1358 outp += 1359 sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float); 1360 if (DO_BIC(BIC_RAMWatt)) 1361 outp += 1362 sprintf(outp, fmt8, (printed++ ? delim : ""), 1363 p->energy_dram * rapl_dram_energy_units / interval_float); 1364 if (DO_BIC(BIC_Pkg_J)) 1365 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units); 1366 if (DO_BIC(BIC_Cor_J) && !(do_rapl & RAPL_PER_CORE_ENERGY)) 1367 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units); 1368 if (DO_BIC(BIC_GFX_J)) 1369 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units); 1370 if (DO_BIC(BIC_RAM_J)) 1371 outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units); 1372 if (DO_BIC(BIC_PKG__)) 1373 outp += 1374 sprintf(outp, fmt8, (printed++ ? delim : ""), 1375 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); 1376 if (DO_BIC(BIC_RAM__)) 1377 outp += 1378 sprintf(outp, fmt8, (printed++ ? delim : ""), 1379 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); 1380 /* UncMHz */ 1381 if (DO_BIC(BIC_UNCORE_MHZ)) 1382 outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->uncore_mhz); 1383 1384 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 1385 if (mp->format == FORMAT_RAW) { 1386 if (mp->width == 32) 1387 outp += 1388 sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), (unsigned int)p->counter[i]); 1389 else 1390 outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->counter[i]); 1391 } else if (mp->format == FORMAT_DELTA) { 1392 if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) 1393 outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->counter[i]); 1394 else 1395 outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]); 1396 } else if (mp->format == FORMAT_PERCENT) { 1397 outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i] / tsc); 1398 } 1399 } 1400 1401 done: 1402 if (*(outp - 1) != '\n') 1403 outp += sprintf(outp, "\n"); 1404 1405 return 0; 1406 } 1407 1408 void flush_output_stdout(void) 1409 { 1410 FILE *filep; 1411 1412 if (outf == stderr) 1413 filep = stdout; 1414 else 1415 filep = outf; 1416 1417 fputs(output_buffer, filep); 1418 fflush(filep); 1419 1420 outp = output_buffer; 1421 } 1422 1423 void flush_output_stderr(void) 1424 { 1425 fputs(output_buffer, outf); 1426 fflush(outf); 1427 outp = output_buffer; 1428 } 1429 1430 void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1431 { 1432 static int count; 1433 1434 if ((!count || (header_iterations && !(count % header_iterations))) || !summary_only) 1435 print_header("\t"); 1436 1437 format_counters(&average.threads, &average.cores, &average.packages); 1438 1439 count++; 1440 1441 if (summary_only) 1442 return; 1443 1444 for_all_cpus(format_counters, t, c, p); 1445 } 1446 1447 #define DELTA_WRAP32(new, old) \ 1448 old = ((((unsigned long long)new << 32) - ((unsigned long long)old << 32)) >> 32); 1449 1450 int delta_package(struct pkg_data *new, struct pkg_data *old) 1451 { 1452 int i; 1453 struct msr_counter *mp; 1454 1455 if (DO_BIC(BIC_Totl_c0)) 1456 old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0; 1457 if (DO_BIC(BIC_Any_c0)) 1458 old->pkg_any_core_c0 = new->pkg_any_core_c0 - old->pkg_any_core_c0; 1459 if (DO_BIC(BIC_GFX_c0)) 1460 old->pkg_any_gfxe_c0 = new->pkg_any_gfxe_c0 - old->pkg_any_gfxe_c0; 1461 if (DO_BIC(BIC_CPUGFX)) 1462 old->pkg_both_core_gfxe_c0 = new->pkg_both_core_gfxe_c0 - old->pkg_both_core_gfxe_c0; 1463 1464 old->pc2 = new->pc2 - old->pc2; 1465 if (DO_BIC(BIC_Pkgpc3)) 1466 old->pc3 = new->pc3 - old->pc3; 1467 if (DO_BIC(BIC_Pkgpc6)) 1468 old->pc6 = new->pc6 - old->pc6; 1469 if (DO_BIC(BIC_Pkgpc7)) 1470 old->pc7 = new->pc7 - old->pc7; 1471 old->pc8 = new->pc8 - old->pc8; 1472 old->pc9 = new->pc9 - old->pc9; 1473 old->pc10 = new->pc10 - old->pc10; 1474 old->cpu_lpi = new->cpu_lpi - old->cpu_lpi; 1475 old->sys_lpi = new->sys_lpi - old->sys_lpi; 1476 old->pkg_temp_c = new->pkg_temp_c; 1477 1478 /* flag an error when rc6 counter resets/wraps */ 1479 if (old->gfx_rc6_ms > new->gfx_rc6_ms) 1480 old->gfx_rc6_ms = -1; 1481 else 1482 old->gfx_rc6_ms = new->gfx_rc6_ms - old->gfx_rc6_ms; 1483 1484 old->uncore_mhz = new->uncore_mhz; 1485 old->gfx_mhz = new->gfx_mhz; 1486 old->gfx_act_mhz = new->gfx_act_mhz; 1487 1488 old->energy_pkg = new->energy_pkg - old->energy_pkg; 1489 old->energy_cores = new->energy_cores - old->energy_cores; 1490 old->energy_gfx = new->energy_gfx - old->energy_gfx; 1491 old->energy_dram = new->energy_dram - old->energy_dram; 1492 old->rapl_pkg_perf_status = new->rapl_pkg_perf_status - old->rapl_pkg_perf_status; 1493 old->rapl_dram_perf_status = new->rapl_dram_perf_status - old->rapl_dram_perf_status; 1494 1495 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 1496 if (mp->format == FORMAT_RAW) 1497 old->counter[i] = new->counter[i]; 1498 else 1499 old->counter[i] = new->counter[i] - old->counter[i]; 1500 } 1501 1502 return 0; 1503 } 1504 1505 void delta_core(struct core_data *new, struct core_data *old) 1506 { 1507 int i; 1508 struct msr_counter *mp; 1509 1510 old->c3 = new->c3 - old->c3; 1511 old->c6 = new->c6 - old->c6; 1512 old->c7 = new->c7 - old->c7; 1513 old->core_temp_c = new->core_temp_c; 1514 old->core_throt_cnt = new->core_throt_cnt; 1515 old->mc6_us = new->mc6_us - old->mc6_us; 1516 1517 DELTA_WRAP32(new->core_energy, old->core_energy); 1518 1519 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 1520 if (mp->format == FORMAT_RAW) 1521 old->counter[i] = new->counter[i]; 1522 else 1523 old->counter[i] = new->counter[i] - old->counter[i]; 1524 } 1525 } 1526 1527 int soft_c1_residency_display(int bic) 1528 { 1529 if (!DO_BIC(BIC_CPU_c1) || use_c1_residency_msr) 1530 return 0; 1531 1532 return DO_BIC_READ(bic); 1533 } 1534 1535 /* 1536 * old = new - old 1537 */ 1538 int delta_thread(struct thread_data *new, struct thread_data *old, struct core_data *core_delta) 1539 { 1540 int i; 1541 struct msr_counter *mp; 1542 1543 /* we run cpuid just the 1st time, copy the results */ 1544 if (DO_BIC(BIC_APIC)) 1545 new->apic_id = old->apic_id; 1546 if (DO_BIC(BIC_X2APIC)) 1547 new->x2apic_id = old->x2apic_id; 1548 1549 /* 1550 * the timestamps from start of measurement interval are in "old" 1551 * the timestamp from end of measurement interval are in "new" 1552 * over-write old w/ new so we can print end of interval values 1553 */ 1554 1555 timersub(&new->tv_begin, &old->tv_begin, &old->tv_delta); 1556 old->tv_begin = new->tv_begin; 1557 old->tv_end = new->tv_end; 1558 1559 old->tsc = new->tsc - old->tsc; 1560 1561 /* check for TSC < 1 Mcycles over interval */ 1562 if (old->tsc < (1000 * 1000)) 1563 errx(-3, "Insanely slow TSC rate, TSC stops in idle?\n" 1564 "You can disable all c-states by booting with \"idle=poll\"\n" 1565 "or just the deep ones with \"processor.max_cstate=1\""); 1566 1567 old->c1 = new->c1 - old->c1; 1568 1569 if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) || soft_c1_residency_display(BIC_Avg_MHz)) { 1570 if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) { 1571 old->aperf = new->aperf - old->aperf; 1572 old->mperf = new->mperf - old->mperf; 1573 } else { 1574 return -1; 1575 } 1576 } 1577 1578 if (use_c1_residency_msr) { 1579 /* 1580 * Some models have a dedicated C1 residency MSR, 1581 * which should be more accurate than the derivation below. 1582 */ 1583 } else { 1584 /* 1585 * As counter collection is not atomic, 1586 * it is possible for mperf's non-halted cycles + idle states 1587 * to exceed TSC's all cycles: show c1 = 0% in that case. 1588 */ 1589 if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > (old->tsc * tsc_tweak)) 1590 old->c1 = 0; 1591 else { 1592 /* normal case, derive c1 */ 1593 old->c1 = (old->tsc * tsc_tweak) - old->mperf - core_delta->c3 1594 - core_delta->c6 - core_delta->c7; 1595 } 1596 } 1597 1598 if (old->mperf == 0) { 1599 if (debug > 1) 1600 fprintf(outf, "cpu%d MPERF 0!\n", old->cpu_id); 1601 old->mperf = 1; /* divide by 0 protection */ 1602 } 1603 1604 if (DO_BIC(BIC_IPC)) 1605 old->instr_count = new->instr_count - old->instr_count; 1606 1607 if (DO_BIC(BIC_IRQ)) 1608 old->irq_count = new->irq_count - old->irq_count; 1609 1610 if (DO_BIC(BIC_SMI)) 1611 old->smi_count = new->smi_count - old->smi_count; 1612 1613 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 1614 if (mp->format == FORMAT_RAW) 1615 old->counter[i] = new->counter[i]; 1616 else 1617 old->counter[i] = new->counter[i] - old->counter[i]; 1618 } 1619 return 0; 1620 } 1621 1622 int delta_cpu(struct thread_data *t, struct core_data *c, 1623 struct pkg_data *p, struct thread_data *t2, struct core_data *c2, struct pkg_data *p2) 1624 { 1625 int retval = 0; 1626 1627 /* calculate core delta only for 1st thread in core */ 1628 if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE) 1629 delta_core(c, c2); 1630 1631 /* always calculate thread delta */ 1632 retval = delta_thread(t, t2, c2); /* c2 is core delta */ 1633 if (retval) 1634 return retval; 1635 1636 /* calculate package delta only for 1st core in package */ 1637 if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) 1638 retval = delta_package(p, p2); 1639 1640 return retval; 1641 } 1642 1643 void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1644 { 1645 int i; 1646 struct msr_counter *mp; 1647 1648 t->tv_begin.tv_sec = 0; 1649 t->tv_begin.tv_usec = 0; 1650 t->tv_end.tv_sec = 0; 1651 t->tv_end.tv_usec = 0; 1652 t->tv_delta.tv_sec = 0; 1653 t->tv_delta.tv_usec = 0; 1654 1655 t->tsc = 0; 1656 t->aperf = 0; 1657 t->mperf = 0; 1658 t->c1 = 0; 1659 1660 t->instr_count = 0; 1661 1662 t->irq_count = 0; 1663 t->smi_count = 0; 1664 1665 /* tells format_counters to dump all fields from this set */ 1666 t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE; 1667 1668 c->c3 = 0; 1669 c->c6 = 0; 1670 c->c7 = 0; 1671 c->mc6_us = 0; 1672 c->core_temp_c = 0; 1673 c->core_energy = 0; 1674 c->core_throt_cnt = 0; 1675 1676 p->pkg_wtd_core_c0 = 0; 1677 p->pkg_any_core_c0 = 0; 1678 p->pkg_any_gfxe_c0 = 0; 1679 p->pkg_both_core_gfxe_c0 = 0; 1680 1681 p->pc2 = 0; 1682 if (DO_BIC(BIC_Pkgpc3)) 1683 p->pc3 = 0; 1684 if (DO_BIC(BIC_Pkgpc6)) 1685 p->pc6 = 0; 1686 if (DO_BIC(BIC_Pkgpc7)) 1687 p->pc7 = 0; 1688 p->pc8 = 0; 1689 p->pc9 = 0; 1690 p->pc10 = 0; 1691 p->cpu_lpi = 0; 1692 p->sys_lpi = 0; 1693 1694 p->energy_pkg = 0; 1695 p->energy_dram = 0; 1696 p->energy_cores = 0; 1697 p->energy_gfx = 0; 1698 p->rapl_pkg_perf_status = 0; 1699 p->rapl_dram_perf_status = 0; 1700 p->pkg_temp_c = 0; 1701 1702 p->gfx_rc6_ms = 0; 1703 p->uncore_mhz = 0; 1704 p->gfx_mhz = 0; 1705 p->gfx_act_mhz = 0; 1706 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) 1707 t->counter[i] = 0; 1708 1709 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) 1710 c->counter[i] = 0; 1711 1712 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) 1713 p->counter[i] = 0; 1714 } 1715 1716 int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1717 { 1718 int i; 1719 struct msr_counter *mp; 1720 1721 /* copy un-changing apic_id's */ 1722 if (DO_BIC(BIC_APIC)) 1723 average.threads.apic_id = t->apic_id; 1724 if (DO_BIC(BIC_X2APIC)) 1725 average.threads.x2apic_id = t->x2apic_id; 1726 1727 /* remember first tv_begin */ 1728 if (average.threads.tv_begin.tv_sec == 0) 1729 average.threads.tv_begin = t->tv_begin; 1730 1731 /* remember last tv_end */ 1732 average.threads.tv_end = t->tv_end; 1733 1734 average.threads.tsc += t->tsc; 1735 average.threads.aperf += t->aperf; 1736 average.threads.mperf += t->mperf; 1737 average.threads.c1 += t->c1; 1738 1739 average.threads.instr_count += t->instr_count; 1740 1741 average.threads.irq_count += t->irq_count; 1742 average.threads.smi_count += t->smi_count; 1743 1744 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 1745 if (mp->format == FORMAT_RAW) 1746 continue; 1747 average.threads.counter[i] += t->counter[i]; 1748 } 1749 1750 /* sum per-core values only for 1st thread in core */ 1751 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 1752 return 0; 1753 1754 average.cores.c3 += c->c3; 1755 average.cores.c6 += c->c6; 1756 average.cores.c7 += c->c7; 1757 average.cores.mc6_us += c->mc6_us; 1758 1759 average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c); 1760 average.cores.core_throt_cnt = MAX(average.cores.core_throt_cnt, c->core_throt_cnt); 1761 1762 average.cores.core_energy += c->core_energy; 1763 1764 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 1765 if (mp->format == FORMAT_RAW) 1766 continue; 1767 average.cores.counter[i] += c->counter[i]; 1768 } 1769 1770 /* sum per-pkg values only for 1st core in pkg */ 1771 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 1772 return 0; 1773 1774 if (DO_BIC(BIC_Totl_c0)) 1775 average.packages.pkg_wtd_core_c0 += p->pkg_wtd_core_c0; 1776 if (DO_BIC(BIC_Any_c0)) 1777 average.packages.pkg_any_core_c0 += p->pkg_any_core_c0; 1778 if (DO_BIC(BIC_GFX_c0)) 1779 average.packages.pkg_any_gfxe_c0 += p->pkg_any_gfxe_c0; 1780 if (DO_BIC(BIC_CPUGFX)) 1781 average.packages.pkg_both_core_gfxe_c0 += p->pkg_both_core_gfxe_c0; 1782 1783 average.packages.pc2 += p->pc2; 1784 if (DO_BIC(BIC_Pkgpc3)) 1785 average.packages.pc3 += p->pc3; 1786 if (DO_BIC(BIC_Pkgpc6)) 1787 average.packages.pc6 += p->pc6; 1788 if (DO_BIC(BIC_Pkgpc7)) 1789 average.packages.pc7 += p->pc7; 1790 average.packages.pc8 += p->pc8; 1791 average.packages.pc9 += p->pc9; 1792 average.packages.pc10 += p->pc10; 1793 1794 average.packages.cpu_lpi = p->cpu_lpi; 1795 average.packages.sys_lpi = p->sys_lpi; 1796 1797 average.packages.energy_pkg += p->energy_pkg; 1798 average.packages.energy_dram += p->energy_dram; 1799 average.packages.energy_cores += p->energy_cores; 1800 average.packages.energy_gfx += p->energy_gfx; 1801 1802 average.packages.gfx_rc6_ms = p->gfx_rc6_ms; 1803 average.packages.uncore_mhz = p->uncore_mhz; 1804 average.packages.gfx_mhz = p->gfx_mhz; 1805 average.packages.gfx_act_mhz = p->gfx_act_mhz; 1806 1807 average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); 1808 1809 average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status; 1810 average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; 1811 1812 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 1813 if (mp->format == FORMAT_RAW) 1814 continue; 1815 average.packages.counter[i] += p->counter[i]; 1816 } 1817 return 0; 1818 } 1819 1820 /* 1821 * sum the counters for all cpus in the system 1822 * compute the weighted average 1823 */ 1824 void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data *p) 1825 { 1826 int i; 1827 struct msr_counter *mp; 1828 1829 clear_counters(&average.threads, &average.cores, &average.packages); 1830 1831 for_all_cpus(sum_counters, t, c, p); 1832 1833 /* Use the global time delta for the average. */ 1834 average.threads.tv_delta = tv_delta; 1835 1836 average.threads.tsc /= topo.num_cpus; 1837 average.threads.aperf /= topo.num_cpus; 1838 average.threads.mperf /= topo.num_cpus; 1839 average.threads.instr_count /= topo.num_cpus; 1840 average.threads.c1 /= topo.num_cpus; 1841 1842 if (average.threads.irq_count > 9999999) 1843 sums_need_wide_columns = 1; 1844 1845 average.cores.c3 /= topo.num_cores; 1846 average.cores.c6 /= topo.num_cores; 1847 average.cores.c7 /= topo.num_cores; 1848 average.cores.mc6_us /= topo.num_cores; 1849 1850 if (DO_BIC(BIC_Totl_c0)) 1851 average.packages.pkg_wtd_core_c0 /= topo.num_packages; 1852 if (DO_BIC(BIC_Any_c0)) 1853 average.packages.pkg_any_core_c0 /= topo.num_packages; 1854 if (DO_BIC(BIC_GFX_c0)) 1855 average.packages.pkg_any_gfxe_c0 /= topo.num_packages; 1856 if (DO_BIC(BIC_CPUGFX)) 1857 average.packages.pkg_both_core_gfxe_c0 /= topo.num_packages; 1858 1859 average.packages.pc2 /= topo.num_packages; 1860 if (DO_BIC(BIC_Pkgpc3)) 1861 average.packages.pc3 /= topo.num_packages; 1862 if (DO_BIC(BIC_Pkgpc6)) 1863 average.packages.pc6 /= topo.num_packages; 1864 if (DO_BIC(BIC_Pkgpc7)) 1865 average.packages.pc7 /= topo.num_packages; 1866 1867 average.packages.pc8 /= topo.num_packages; 1868 average.packages.pc9 /= topo.num_packages; 1869 average.packages.pc10 /= topo.num_packages; 1870 1871 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 1872 if (mp->format == FORMAT_RAW) 1873 continue; 1874 if (mp->type == COUNTER_ITEMS) { 1875 if (average.threads.counter[i] > 9999999) 1876 sums_need_wide_columns = 1; 1877 continue; 1878 } 1879 average.threads.counter[i] /= topo.num_cpus; 1880 } 1881 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 1882 if (mp->format == FORMAT_RAW) 1883 continue; 1884 if (mp->type == COUNTER_ITEMS) { 1885 if (average.cores.counter[i] > 9999999) 1886 sums_need_wide_columns = 1; 1887 } 1888 average.cores.counter[i] /= topo.num_cores; 1889 } 1890 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 1891 if (mp->format == FORMAT_RAW) 1892 continue; 1893 if (mp->type == COUNTER_ITEMS) { 1894 if (average.packages.counter[i] > 9999999) 1895 sums_need_wide_columns = 1; 1896 } 1897 average.packages.counter[i] /= topo.num_packages; 1898 } 1899 } 1900 1901 static unsigned long long rdtsc(void) 1902 { 1903 unsigned int low, high; 1904 1905 asm volatile ("rdtsc":"=a" (low), "=d"(high)); 1906 1907 return low | ((unsigned long long)high) << 32; 1908 } 1909 1910 /* 1911 * Open a file, and exit on failure 1912 */ 1913 FILE *fopen_or_die(const char *path, const char *mode) 1914 { 1915 FILE *filep = fopen(path, mode); 1916 1917 if (!filep) 1918 err(1, "%s: open failed", path); 1919 return filep; 1920 } 1921 1922 /* 1923 * snapshot_sysfs_counter() 1924 * 1925 * return snapshot of given counter 1926 */ 1927 unsigned long long snapshot_sysfs_counter(char *path) 1928 { 1929 FILE *fp; 1930 int retval; 1931 unsigned long long counter; 1932 1933 fp = fopen_or_die(path, "r"); 1934 1935 retval = fscanf(fp, "%lld", &counter); 1936 if (retval != 1) 1937 err(1, "snapshot_sysfs_counter(%s)", path); 1938 1939 fclose(fp); 1940 1941 return counter; 1942 } 1943 1944 int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp) 1945 { 1946 if (mp->msr_num != 0) { 1947 if (get_msr(cpu, mp->msr_num, counterp)) 1948 return -1; 1949 } else { 1950 char path[128 + PATH_BYTES]; 1951 1952 if (mp->flags & SYSFS_PERCPU) { 1953 sprintf(path, "/sys/devices/system/cpu/cpu%d/%s", cpu, mp->path); 1954 1955 *counterp = snapshot_sysfs_counter(path); 1956 } else { 1957 *counterp = snapshot_sysfs_counter(mp->path); 1958 } 1959 } 1960 1961 return 0; 1962 } 1963 1964 unsigned long long get_uncore_mhz(int package, int die) 1965 { 1966 char path[128]; 1967 1968 sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/current_freq_khz", package, 1969 die); 1970 1971 return (snapshot_sysfs_counter(path) / 1000); 1972 } 1973 1974 int get_epb(int cpu) 1975 { 1976 char path[128 + PATH_BYTES]; 1977 unsigned long long msr; 1978 int ret, epb = -1; 1979 FILE *fp; 1980 1981 sprintf(path, "/sys/devices/system/cpu/cpu%d/power/energy_perf_bias", cpu); 1982 1983 fp = fopen(path, "r"); 1984 if (!fp) 1985 goto msr_fallback; 1986 1987 ret = fscanf(fp, "%d", &epb); 1988 if (ret != 1) 1989 err(1, "%s(%s)", __func__, path); 1990 1991 fclose(fp); 1992 1993 return epb; 1994 1995 msr_fallback: 1996 get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr); 1997 1998 return msr & 0xf; 1999 } 2000 2001 void get_apic_id(struct thread_data *t) 2002 { 2003 unsigned int eax, ebx, ecx, edx; 2004 2005 if (DO_BIC(BIC_APIC)) { 2006 eax = ebx = ecx = edx = 0; 2007 __cpuid(1, eax, ebx, ecx, edx); 2008 2009 t->apic_id = (ebx >> 24) & 0xff; 2010 } 2011 2012 if (!DO_BIC(BIC_X2APIC)) 2013 return; 2014 2015 if (authentic_amd || hygon_genuine) { 2016 unsigned int topology_extensions; 2017 2018 if (max_extended_level < 0x8000001e) 2019 return; 2020 2021 eax = ebx = ecx = edx = 0; 2022 __cpuid(0x80000001, eax, ebx, ecx, edx); 2023 topology_extensions = ecx & (1 << 22); 2024 2025 if (topology_extensions == 0) 2026 return; 2027 2028 eax = ebx = ecx = edx = 0; 2029 __cpuid(0x8000001e, eax, ebx, ecx, edx); 2030 2031 t->x2apic_id = eax; 2032 return; 2033 } 2034 2035 if (!genuine_intel) 2036 return; 2037 2038 if (max_level < 0xb) 2039 return; 2040 2041 ecx = 0; 2042 __cpuid(0xb, eax, ebx, ecx, edx); 2043 t->x2apic_id = edx; 2044 2045 if (debug && (t->apic_id != (t->x2apic_id & 0xff))) 2046 fprintf(outf, "cpu%d: BIOS BUG: apic 0x%x x2apic 0x%x\n", t->cpu_id, t->apic_id, t->x2apic_id); 2047 } 2048 2049 int get_core_throt_cnt(int cpu, unsigned long long *cnt) 2050 { 2051 char path[128 + PATH_BYTES]; 2052 unsigned long long tmp; 2053 FILE *fp; 2054 int ret; 2055 2056 sprintf(path, "/sys/devices/system/cpu/cpu%d/thermal_throttle/core_throttle_count", cpu); 2057 fp = fopen(path, "r"); 2058 if (!fp) 2059 return -1; 2060 ret = fscanf(fp, "%lld", &tmp); 2061 fclose(fp); 2062 if (ret != 1) 2063 return -1; 2064 *cnt = tmp; 2065 2066 return 0; 2067 } 2068 2069 /* 2070 * get_counters(...) 2071 * migrate to cpu 2072 * acquire and record local counters for that cpu 2073 */ 2074 int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) 2075 { 2076 int cpu = t->cpu_id; 2077 unsigned long long msr; 2078 int aperf_mperf_retry_count = 0; 2079 struct msr_counter *mp; 2080 int i; 2081 2082 if (cpu_migrate(cpu)) { 2083 fprintf(outf, "get_counters: Could not migrate to CPU %d\n", cpu); 2084 return -1; 2085 } 2086 2087 gettimeofday(&t->tv_begin, (struct timezone *)NULL); 2088 2089 if (first_counter_read) 2090 get_apic_id(t); 2091 retry: 2092 t->tsc = rdtsc(); /* we are running on local CPU of interest */ 2093 2094 if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) || soft_c1_residency_display(BIC_Avg_MHz)) { 2095 unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time; 2096 2097 /* 2098 * The TSC, APERF and MPERF must be read together for 2099 * APERF/MPERF and MPERF/TSC to give accurate results. 2100 * 2101 * Unfortunately, APERF and MPERF are read by 2102 * individual system call, so delays may occur 2103 * between them. If the time to read them 2104 * varies by a large amount, we re-read them. 2105 */ 2106 2107 /* 2108 * This initial dummy APERF read has been seen to 2109 * reduce jitter in the subsequent reads. 2110 */ 2111 2112 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) 2113 return -3; 2114 2115 t->tsc = rdtsc(); /* re-read close to APERF */ 2116 2117 tsc_before = t->tsc; 2118 2119 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) 2120 return -3; 2121 2122 tsc_between = rdtsc(); 2123 2124 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) 2125 return -4; 2126 2127 tsc_after = rdtsc(); 2128 2129 aperf_time = tsc_between - tsc_before; 2130 mperf_time = tsc_after - tsc_between; 2131 2132 /* 2133 * If the system call latency to read APERF and MPERF 2134 * differ by more than 2x, then try again. 2135 */ 2136 if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) { 2137 aperf_mperf_retry_count++; 2138 if (aperf_mperf_retry_count < 5) 2139 goto retry; 2140 else 2141 warnx("cpu%d jitter %lld %lld", cpu, aperf_time, mperf_time); 2142 } 2143 aperf_mperf_retry_count = 0; 2144 2145 t->aperf = t->aperf * aperf_mperf_multiplier; 2146 t->mperf = t->mperf * aperf_mperf_multiplier; 2147 } 2148 2149 if (DO_BIC(BIC_IPC)) 2150 if (read(get_instr_count_fd(cpu), &t->instr_count, sizeof(long long)) != sizeof(long long)) 2151 return -4; 2152 2153 if (DO_BIC(BIC_IRQ)) 2154 t->irq_count = irqs_per_cpu[cpu]; 2155 if (DO_BIC(BIC_SMI)) { 2156 if (get_msr(cpu, MSR_SMI_COUNT, &msr)) 2157 return -5; 2158 t->smi_count = msr & 0xFFFFFFFF; 2159 } 2160 if (DO_BIC(BIC_CPU_c1) && use_c1_residency_msr) { 2161 if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1)) 2162 return -6; 2163 } 2164 2165 for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { 2166 if (get_mp(cpu, mp, &t->counter[i])) 2167 return -10; 2168 } 2169 2170 /* collect core counters only for 1st thread in core */ 2171 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 2172 goto done; 2173 2174 if (DO_BIC(BIC_CPU_c3) || soft_c1_residency_display(BIC_CPU_c3)) { 2175 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) 2176 return -6; 2177 } 2178 2179 if ((DO_BIC(BIC_CPU_c6) || soft_c1_residency_display(BIC_CPU_c6)) && !do_knl_cstates) { 2180 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) 2181 return -7; 2182 } else if (do_knl_cstates || soft_c1_residency_display(BIC_CPU_c6)) { 2183 if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6)) 2184 return -7; 2185 } 2186 2187 if (DO_BIC(BIC_CPU_c7) || soft_c1_residency_display(BIC_CPU_c7)) { 2188 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) 2189 return -8; 2190 else if (t->is_atom) { 2191 /* 2192 * For Atom CPUs that has core cstate deeper than c6, 2193 * MSR_CORE_C6_RESIDENCY returns residency of cc6 and deeper. 2194 * Minus CC7 (and deeper cstates) residency to get 2195 * accturate cc6 residency. 2196 */ 2197 c->c6 -= c->c7; 2198 } 2199 } 2200 2201 if (DO_BIC(BIC_Mod_c6)) 2202 if (get_msr(cpu, MSR_MODULE_C6_RES_MS, &c->mc6_us)) 2203 return -8; 2204 2205 if (DO_BIC(BIC_CoreTmp)) { 2206 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) 2207 return -9; 2208 c->core_temp_c = tj_max - ((msr >> 16) & 0x7F); 2209 } 2210 2211 if (DO_BIC(BIC_CORE_THROT_CNT)) 2212 get_core_throt_cnt(cpu, &c->core_throt_cnt); 2213 2214 if (do_rapl & RAPL_AMD_F17H) { 2215 if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr)) 2216 return -14; 2217 c->core_energy = msr & 0xFFFFFFFF; 2218 } 2219 2220 for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { 2221 if (get_mp(cpu, mp, &c->counter[i])) 2222 return -10; 2223 } 2224 2225 /* collect package counters only for 1st core in package */ 2226 if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 2227 goto done; 2228 2229 if (DO_BIC(BIC_Totl_c0)) { 2230 if (get_msr(cpu, MSR_PKG_WEIGHTED_CORE_C0_RES, &p->pkg_wtd_core_c0)) 2231 return -10; 2232 } 2233 if (DO_BIC(BIC_Any_c0)) { 2234 if (get_msr(cpu, MSR_PKG_ANY_CORE_C0_RES, &p->pkg_any_core_c0)) 2235 return -11; 2236 } 2237 if (DO_BIC(BIC_GFX_c0)) { 2238 if (get_msr(cpu, MSR_PKG_ANY_GFXE_C0_RES, &p->pkg_any_gfxe_c0)) 2239 return -12; 2240 } 2241 if (DO_BIC(BIC_CPUGFX)) { 2242 if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0)) 2243 return -13; 2244 } 2245 if (DO_BIC(BIC_Pkgpc3)) 2246 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) 2247 return -9; 2248 if (DO_BIC(BIC_Pkgpc6)) { 2249 if (do_slm_cstates) { 2250 if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6)) 2251 return -10; 2252 } else { 2253 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) 2254 return -10; 2255 } 2256 } 2257 2258 if (DO_BIC(BIC_Pkgpc2)) 2259 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) 2260 return -11; 2261 if (DO_BIC(BIC_Pkgpc7)) 2262 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) 2263 return -12; 2264 if (DO_BIC(BIC_Pkgpc8)) 2265 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8)) 2266 return -13; 2267 if (DO_BIC(BIC_Pkgpc9)) 2268 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9)) 2269 return -13; 2270 if (DO_BIC(BIC_Pkgpc10)) 2271 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10)) 2272 return -13; 2273 2274 if (DO_BIC(BIC_CPU_LPI)) 2275 p->cpu_lpi = cpuidle_cur_cpu_lpi_us; 2276 if (DO_BIC(BIC_SYS_LPI)) 2277 p->sys_lpi = cpuidle_cur_sys_lpi_us; 2278 2279 if (do_rapl & RAPL_PKG) { 2280 if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr)) 2281 return -13; 2282 p->energy_pkg = msr; 2283 } 2284 if (do_rapl & RAPL_CORES_ENERGY_STATUS) { 2285 if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr)) 2286 return -14; 2287 p->energy_cores = msr; 2288 } 2289 if (do_rapl & RAPL_DRAM) { 2290 if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) 2291 return -15; 2292 p->energy_dram = msr; 2293 } 2294 if (do_rapl & RAPL_GFX) { 2295 if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr)) 2296 return -16; 2297 p->energy_gfx = msr; 2298 } 2299 if (do_rapl & RAPL_PKG_PERF_STATUS) { 2300 if (get_msr_sum(cpu, MSR_PKG_PERF_STATUS, &msr)) 2301 return -16; 2302 p->rapl_pkg_perf_status = msr; 2303 } 2304 if (do_rapl & RAPL_DRAM_PERF_STATUS) { 2305 if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr)) 2306 return -16; 2307 p->rapl_dram_perf_status = msr; 2308 } 2309 if (do_rapl & RAPL_AMD_F17H) { 2310 if (get_msr_sum(cpu, MSR_PKG_ENERGY_STAT, &msr)) 2311 return -13; 2312 p->energy_pkg = msr; 2313 } 2314 if (DO_BIC(BIC_PkgTmp)) { 2315 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) 2316 return -17; 2317 p->pkg_temp_c = tj_max - ((msr >> 16) & 0x7F); 2318 } 2319 2320 if (DO_BIC(BIC_GFX_rc6)) 2321 p->gfx_rc6_ms = gfx_cur_rc6_ms; 2322 2323 /* n.b. assume die0 uncore frequency applies to whole package */ 2324 if (DO_BIC(BIC_UNCORE_MHZ)) 2325 p->uncore_mhz = get_uncore_mhz(p->package_id, 0); 2326 2327 if (DO_BIC(BIC_GFXMHz)) 2328 p->gfx_mhz = gfx_cur_mhz; 2329 2330 if (DO_BIC(BIC_GFXACTMHz)) 2331 p->gfx_act_mhz = gfx_act_mhz; 2332 2333 for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { 2334 if (get_mp(cpu, mp, &p->counter[i])) 2335 return -10; 2336 } 2337 done: 2338 gettimeofday(&t->tv_end, (struct timezone *)NULL); 2339 2340 return 0; 2341 } 2342 2343 /* 2344 * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: 2345 * If you change the values, note they are used both in comparisons 2346 * (>= PCL__7) and to index pkg_cstate_limit_strings[]. 2347 */ 2348 2349 #define PCLUKN 0 /* Unknown */ 2350 #define PCLRSV 1 /* Reserved */ 2351 #define PCL__0 2 /* PC0 */ 2352 #define PCL__1 3 /* PC1 */ 2353 #define PCL__2 4 /* PC2 */ 2354 #define PCL__3 5 /* PC3 */ 2355 #define PCL__4 6 /* PC4 */ 2356 #define PCL__6 7 /* PC6 */ 2357 #define PCL_6N 8 /* PC6 No Retention */ 2358 #define PCL_6R 9 /* PC6 Retention */ 2359 #define PCL__7 10 /* PC7 */ 2360 #define PCL_7S 11 /* PC7 Shrink */ 2361 #define PCL__8 12 /* PC8 */ 2362 #define PCL__9 13 /* PC9 */ 2363 #define PCL_10 14 /* PC10 */ 2364 #define PCLUNL 15 /* Unlimited */ 2365 2366 int pkg_cstate_limit = PCLUKN; 2367 char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2", 2368 "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited" 2369 }; 2370 2371 int nhm_pkg_cstate_limits[16] = 2372 { PCL__0, PCL__1, PCL__3, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2373 PCLRSV, PCLRSV 2374 }; 2375 2376 int snb_pkg_cstate_limits[16] = 2377 { PCL__0, PCL__2, PCL_6N, PCL_6R, PCL__7, PCL_7S, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2378 PCLRSV, PCLRSV 2379 }; 2380 2381 int hsw_pkg_cstate_limits[16] = 2382 { PCL__0, PCL__2, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2383 PCLRSV, PCLRSV 2384 }; 2385 2386 int slv_pkg_cstate_limits[16] = 2387 { PCL__0, PCL__1, PCLRSV, PCLRSV, PCL__4, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2388 PCL__6, PCL__7 2389 }; 2390 2391 int amt_pkg_cstate_limits[16] = 2392 { PCLUNL, PCL__1, PCL__2, PCLRSV, PCLRSV, PCLRSV, PCL__6, PCL__7, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2393 PCLRSV, PCLRSV 2394 }; 2395 2396 int phi_pkg_cstate_limits[16] = 2397 { PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2398 PCLRSV, PCLRSV 2399 }; 2400 2401 int glm_pkg_cstate_limits[16] = 2402 { PCLUNL, PCL__1, PCL__3, PCL__6, PCL__7, PCL_7S, PCL__8, PCL__9, PCL_10, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2403 PCLRSV, PCLRSV 2404 }; 2405 2406 int skx_pkg_cstate_limits[16] = 2407 { PCL__0, PCL__2, PCL_6N, PCL_6R, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2408 PCLRSV, PCLRSV 2409 }; 2410 2411 int icx_pkg_cstate_limits[16] = 2412 { PCL__0, PCL__2, PCL__6, PCL__6, PCLRSV, PCLRSV, PCLRSV, PCLUNL, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, PCLRSV, 2413 PCLRSV, PCLRSV 2414 }; 2415 2416 static void calculate_tsc_tweak() 2417 { 2418 tsc_tweak = base_hz / tsc_hz; 2419 } 2420 2421 void prewake_cstate_probe(unsigned int family, unsigned int model); 2422 2423 static void dump_nhm_platform_info(void) 2424 { 2425 unsigned long long msr; 2426 unsigned int ratio; 2427 2428 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); 2429 2430 fprintf(outf, "cpu%d: MSR_PLATFORM_INFO: 0x%08llx\n", base_cpu, msr); 2431 2432 ratio = (msr >> 40) & 0xFF; 2433 fprintf(outf, "%d * %.1f = %.1f MHz max efficiency frequency\n", ratio, bclk, ratio * bclk); 2434 2435 ratio = (msr >> 8) & 0xFF; 2436 fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n", ratio, bclk, ratio * bclk); 2437 2438 get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr); 2439 fprintf(outf, "cpu%d: MSR_IA32_POWER_CTL: 0x%08llx (C1E auto-promotion: %sabled)\n", 2440 base_cpu, msr, msr & 0x2 ? "EN" : "DIS"); 2441 2442 /* C-state Pre-wake Disable (CSTATE_PREWAKE_DISABLE) */ 2443 if (dis_cstate_prewake) 2444 fprintf(outf, "C-state Pre-wake: %sabled\n", msr & 0x40000000 ? "DIS" : "EN"); 2445 2446 return; 2447 } 2448 2449 static void dump_hsw_turbo_ratio_limits(void) 2450 { 2451 unsigned long long msr; 2452 unsigned int ratio; 2453 2454 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT2, &msr); 2455 2456 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT2: 0x%08llx\n", base_cpu, msr); 2457 2458 ratio = (msr >> 8) & 0xFF; 2459 if (ratio) 2460 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 18 active cores\n", ratio, bclk, ratio * bclk); 2461 2462 ratio = (msr >> 0) & 0xFF; 2463 if (ratio) 2464 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 17 active cores\n", ratio, bclk, ratio * bclk); 2465 return; 2466 } 2467 2468 static void dump_ivt_turbo_ratio_limits(void) 2469 { 2470 unsigned long long msr; 2471 unsigned int ratio; 2472 2473 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &msr); 2474 2475 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, msr); 2476 2477 ratio = (msr >> 56) & 0xFF; 2478 if (ratio) 2479 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 16 active cores\n", ratio, bclk, ratio * bclk); 2480 2481 ratio = (msr >> 48) & 0xFF; 2482 if (ratio) 2483 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 15 active cores\n", ratio, bclk, ratio * bclk); 2484 2485 ratio = (msr >> 40) & 0xFF; 2486 if (ratio) 2487 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 14 active cores\n", ratio, bclk, ratio * bclk); 2488 2489 ratio = (msr >> 32) & 0xFF; 2490 if (ratio) 2491 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 13 active cores\n", ratio, bclk, ratio * bclk); 2492 2493 ratio = (msr >> 24) & 0xFF; 2494 if (ratio) 2495 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 12 active cores\n", ratio, bclk, ratio * bclk); 2496 2497 ratio = (msr >> 16) & 0xFF; 2498 if (ratio) 2499 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 11 active cores\n", ratio, bclk, ratio * bclk); 2500 2501 ratio = (msr >> 8) & 0xFF; 2502 if (ratio) 2503 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 10 active cores\n", ratio, bclk, ratio * bclk); 2504 2505 ratio = (msr >> 0) & 0xFF; 2506 if (ratio) 2507 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 9 active cores\n", ratio, bclk, ratio * bclk); 2508 return; 2509 } 2510 2511 int has_turbo_ratio_group_limits(int family, int model) 2512 { 2513 2514 if (!genuine_intel) 2515 return 0; 2516 2517 if (family != 6) 2518 return 0; 2519 2520 switch (model) { 2521 case INTEL_FAM6_ATOM_GOLDMONT: 2522 case INTEL_FAM6_SKYLAKE_X: 2523 case INTEL_FAM6_ICELAKE_X: 2524 case INTEL_FAM6_ATOM_GOLDMONT_D: 2525 case INTEL_FAM6_ATOM_TREMONT_D: 2526 return 1; 2527 default: 2528 return 0; 2529 } 2530 } 2531 2532 static void dump_turbo_ratio_limits(int trl_msr_offset, int family, int model) 2533 { 2534 unsigned long long msr, core_counts; 2535 int shift; 2536 2537 get_msr(base_cpu, trl_msr_offset, &msr); 2538 fprintf(outf, "cpu%d: MSR_%sTURBO_RATIO_LIMIT: 0x%08llx\n", 2539 base_cpu, trl_msr_offset == MSR_SECONDARY_TURBO_RATIO_LIMIT ? "SECONDARY" : "", msr); 2540 2541 if (has_turbo_ratio_group_limits(family, model)) { 2542 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT1, &core_counts); 2543 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT1: 0x%08llx\n", base_cpu, core_counts); 2544 } else { 2545 core_counts = 0x0807060504030201; 2546 } 2547 2548 for (shift = 56; shift >= 0; shift -= 8) { 2549 unsigned int ratio, group_size; 2550 2551 ratio = (msr >> shift) & 0xFF; 2552 group_size = (core_counts >> shift) & 0xFF; 2553 if (ratio) 2554 fprintf(outf, "%d * %.1f = %.1f MHz max turbo %d active cores\n", 2555 ratio, bclk, ratio * bclk, group_size); 2556 } 2557 2558 return; 2559 } 2560 2561 static void dump_atom_turbo_ratio_limits(void) 2562 { 2563 unsigned long long msr; 2564 unsigned int ratio; 2565 2566 get_msr(base_cpu, MSR_ATOM_CORE_RATIOS, &msr); 2567 fprintf(outf, "cpu%d: MSR_ATOM_CORE_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF); 2568 2569 ratio = (msr >> 0) & 0x3F; 2570 if (ratio) 2571 fprintf(outf, "%d * %.1f = %.1f MHz minimum operating frequency\n", ratio, bclk, ratio * bclk); 2572 2573 ratio = (msr >> 8) & 0x3F; 2574 if (ratio) 2575 fprintf(outf, "%d * %.1f = %.1f MHz low frequency mode (LFM)\n", ratio, bclk, ratio * bclk); 2576 2577 ratio = (msr >> 16) & 0x3F; 2578 if (ratio) 2579 fprintf(outf, "%d * %.1f = %.1f MHz base frequency\n", ratio, bclk, ratio * bclk); 2580 2581 get_msr(base_cpu, MSR_ATOM_CORE_TURBO_RATIOS, &msr); 2582 fprintf(outf, "cpu%d: MSR_ATOM_CORE_TURBO_RATIOS: 0x%08llx\n", base_cpu, msr & 0xFFFFFFFF); 2583 2584 ratio = (msr >> 24) & 0x3F; 2585 if (ratio) 2586 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 4 active cores\n", ratio, bclk, ratio * bclk); 2587 2588 ratio = (msr >> 16) & 0x3F; 2589 if (ratio) 2590 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 3 active cores\n", ratio, bclk, ratio * bclk); 2591 2592 ratio = (msr >> 8) & 0x3F; 2593 if (ratio) 2594 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 2 active cores\n", ratio, bclk, ratio * bclk); 2595 2596 ratio = (msr >> 0) & 0x3F; 2597 if (ratio) 2598 fprintf(outf, "%d * %.1f = %.1f MHz max turbo 1 active core\n", ratio, bclk, ratio * bclk); 2599 } 2600 2601 static void dump_knl_turbo_ratio_limits(void) 2602 { 2603 const unsigned int buckets_no = 7; 2604 2605 unsigned long long msr; 2606 int delta_cores, delta_ratio; 2607 int i, b_nr; 2608 unsigned int cores[buckets_no]; 2609 unsigned int ratio[buckets_no]; 2610 2611 get_msr(base_cpu, MSR_TURBO_RATIO_LIMIT, &msr); 2612 2613 fprintf(outf, "cpu%d: MSR_TURBO_RATIO_LIMIT: 0x%08llx\n", base_cpu, msr); 2614 2615 /* 2616 * Turbo encoding in KNL is as follows: 2617 * [0] -- Reserved 2618 * [7:1] -- Base value of number of active cores of bucket 1. 2619 * [15:8] -- Base value of freq ratio of bucket 1. 2620 * [20:16] -- +ve delta of number of active cores of bucket 2. 2621 * i.e. active cores of bucket 2 = 2622 * active cores of bucket 1 + delta 2623 * [23:21] -- Negative delta of freq ratio of bucket 2. 2624 * i.e. freq ratio of bucket 2 = 2625 * freq ratio of bucket 1 - delta 2626 * [28:24]-- +ve delta of number of active cores of bucket 3. 2627 * [31:29]-- -ve delta of freq ratio of bucket 3. 2628 * [36:32]-- +ve delta of number of active cores of bucket 4. 2629 * [39:37]-- -ve delta of freq ratio of bucket 4. 2630 * [44:40]-- +ve delta of number of active cores of bucket 5. 2631 * [47:45]-- -ve delta of freq ratio of bucket 5. 2632 * [52:48]-- +ve delta of number of active cores of bucket 6. 2633 * [55:53]-- -ve delta of freq ratio of bucket 6. 2634 * [60:56]-- +ve delta of number of active cores of bucket 7. 2635 * [63:61]-- -ve delta of freq ratio of bucket 7. 2636 */ 2637 2638 b_nr = 0; 2639 cores[b_nr] = (msr & 0xFF) >> 1; 2640 ratio[b_nr] = (msr >> 8) & 0xFF; 2641 2642 for (i = 16; i < 64; i += 8) { 2643 delta_cores = (msr >> i) & 0x1F; 2644 delta_ratio = (msr >> (i + 5)) & 0x7; 2645 2646 cores[b_nr + 1] = cores[b_nr] + delta_cores; 2647 ratio[b_nr + 1] = ratio[b_nr] - delta_ratio; 2648 b_nr++; 2649 } 2650 2651 for (i = buckets_no - 1; i >= 0; i--) 2652 if (i > 0 ? ratio[i] != ratio[i - 1] : 1) 2653 fprintf(outf, 2654 "%d * %.1f = %.1f MHz max turbo %d active cores\n", 2655 ratio[i], bclk, ratio[i] * bclk, cores[i]); 2656 } 2657 2658 static void dump_nhm_cst_cfg(void) 2659 { 2660 unsigned long long msr; 2661 2662 get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr); 2663 2664 fprintf(outf, "cpu%d: MSR_PKG_CST_CONFIG_CONTROL: 0x%08llx", base_cpu, msr); 2665 2666 fprintf(outf, " (%s%s%s%s%slocked, pkg-cstate-limit=%d (%s)", 2667 (msr & SNB_C3_AUTO_UNDEMOTE) ? "UNdemote-C3, " : "", 2668 (msr & SNB_C1_AUTO_UNDEMOTE) ? "UNdemote-C1, " : "", 2669 (msr & NHM_C3_AUTO_DEMOTE) ? "demote-C3, " : "", 2670 (msr & NHM_C1_AUTO_DEMOTE) ? "demote-C1, " : "", 2671 (msr & (1 << 15)) ? "" : "UN", (unsigned int)msr & 0xF, pkg_cstate_limit_strings[pkg_cstate_limit]); 2672 2673 #define AUTOMATIC_CSTATE_CONVERSION (1UL << 16) 2674 if (has_automatic_cstate_conversion) { 2675 fprintf(outf, ", automatic c-state conversion=%s", (msr & AUTOMATIC_CSTATE_CONVERSION) ? "on" : "off"); 2676 } 2677 2678 fprintf(outf, ")\n"); 2679 2680 return; 2681 } 2682 2683 static void dump_config_tdp(void) 2684 { 2685 unsigned long long msr; 2686 2687 get_msr(base_cpu, MSR_CONFIG_TDP_NOMINAL, &msr); 2688 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_NOMINAL: 0x%08llx", base_cpu, msr); 2689 fprintf(outf, " (base_ratio=%d)\n", (unsigned int)msr & 0xFF); 2690 2691 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_1, &msr); 2692 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_1: 0x%08llx (", base_cpu, msr); 2693 if (msr) { 2694 fprintf(outf, "PKG_MIN_PWR_LVL1=%d ", (unsigned int)(msr >> 48) & 0x7FFF); 2695 fprintf(outf, "PKG_MAX_PWR_LVL1=%d ", (unsigned int)(msr >> 32) & 0x7FFF); 2696 fprintf(outf, "LVL1_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF); 2697 fprintf(outf, "PKG_TDP_LVL1=%d", (unsigned int)(msr) & 0x7FFF); 2698 } 2699 fprintf(outf, ")\n"); 2700 2701 get_msr(base_cpu, MSR_CONFIG_TDP_LEVEL_2, &msr); 2702 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_LEVEL_2: 0x%08llx (", base_cpu, msr); 2703 if (msr) { 2704 fprintf(outf, "PKG_MIN_PWR_LVL2=%d ", (unsigned int)(msr >> 48) & 0x7FFF); 2705 fprintf(outf, "PKG_MAX_PWR_LVL2=%d ", (unsigned int)(msr >> 32) & 0x7FFF); 2706 fprintf(outf, "LVL2_RATIO=%d ", (unsigned int)(msr >> 16) & 0xFF); 2707 fprintf(outf, "PKG_TDP_LVL2=%d", (unsigned int)(msr) & 0x7FFF); 2708 } 2709 fprintf(outf, ")\n"); 2710 2711 get_msr(base_cpu, MSR_CONFIG_TDP_CONTROL, &msr); 2712 fprintf(outf, "cpu%d: MSR_CONFIG_TDP_CONTROL: 0x%08llx (", base_cpu, msr); 2713 if ((msr) & 0x3) 2714 fprintf(outf, "TDP_LEVEL=%d ", (unsigned int)(msr) & 0x3); 2715 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1); 2716 fprintf(outf, ")\n"); 2717 2718 get_msr(base_cpu, MSR_TURBO_ACTIVATION_RATIO, &msr); 2719 fprintf(outf, "cpu%d: MSR_TURBO_ACTIVATION_RATIO: 0x%08llx (", base_cpu, msr); 2720 fprintf(outf, "MAX_NON_TURBO_RATIO=%d", (unsigned int)(msr) & 0xFF); 2721 fprintf(outf, " lock=%d", (unsigned int)(msr >> 31) & 1); 2722 fprintf(outf, ")\n"); 2723 } 2724 2725 unsigned int irtl_time_units[] = { 1, 32, 1024, 32768, 1048576, 33554432, 0, 0 }; 2726 2727 void print_irtl(void) 2728 { 2729 unsigned long long msr; 2730 2731 get_msr(base_cpu, MSR_PKGC3_IRTL, &msr); 2732 fprintf(outf, "cpu%d: MSR_PKGC3_IRTL: 0x%08llx (", base_cpu, msr); 2733 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT", 2734 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]); 2735 2736 get_msr(base_cpu, MSR_PKGC6_IRTL, &msr); 2737 fprintf(outf, "cpu%d: MSR_PKGC6_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_PKGC7_IRTL, &msr); 2742 fprintf(outf, "cpu%d: MSR_PKGC7_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 if (!do_irtl_hsw) 2747 return; 2748 2749 get_msr(base_cpu, MSR_PKGC8_IRTL, &msr); 2750 fprintf(outf, "cpu%d: MSR_PKGC8_IRTL: 0x%08llx (", base_cpu, msr); 2751 fprintf(outf, "%svalid, %lld ns)\n", msr & (1 << 15) ? "" : "NOT", 2752 (msr & 0x3FF) * irtl_time_units[(msr >> 10) & 0x3]); 2753 2754 get_msr(base_cpu, MSR_PKGC9_IRTL, &msr); 2755 fprintf(outf, "cpu%d: MSR_PKGC9_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_PKGC10_IRTL, &msr); 2760 fprintf(outf, "cpu%d: MSR_PKGC10_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 } 2765 2766 void free_fd_percpu(void) 2767 { 2768 int i; 2769 2770 for (i = 0; i < topo.max_cpu_num + 1; ++i) { 2771 if (fd_percpu[i] != 0) 2772 close(fd_percpu[i]); 2773 } 2774 2775 free(fd_percpu); 2776 } 2777 2778 void free_all_buffers(void) 2779 { 2780 int i; 2781 2782 CPU_FREE(cpu_present_set); 2783 cpu_present_set = NULL; 2784 cpu_present_setsize = 0; 2785 2786 CPU_FREE(cpu_affinity_set); 2787 cpu_affinity_set = NULL; 2788 cpu_affinity_setsize = 0; 2789 2790 free(thread_even); 2791 free(core_even); 2792 free(package_even); 2793 2794 thread_even = NULL; 2795 core_even = NULL; 2796 package_even = NULL; 2797 2798 free(thread_odd); 2799 free(core_odd); 2800 free(package_odd); 2801 2802 thread_odd = NULL; 2803 core_odd = NULL; 2804 package_odd = NULL; 2805 2806 free(output_buffer); 2807 output_buffer = NULL; 2808 outp = NULL; 2809 2810 free_fd_percpu(); 2811 2812 free(irq_column_2_cpu); 2813 free(irqs_per_cpu); 2814 2815 for (i = 0; i <= topo.max_cpu_num; ++i) { 2816 if (cpus[i].put_ids) 2817 CPU_FREE(cpus[i].put_ids); 2818 } 2819 free(cpus); 2820 } 2821 2822 /* 2823 * Parse a file containing a single int. 2824 * Return 0 if file can not be opened 2825 * Exit if file can be opened, but can not be parsed 2826 */ 2827 int parse_int_file(const char *fmt, ...) 2828 { 2829 va_list args; 2830 char path[PATH_MAX]; 2831 FILE *filep; 2832 int value; 2833 2834 va_start(args, fmt); 2835 vsnprintf(path, sizeof(path), fmt, args); 2836 va_end(args); 2837 filep = fopen(path, "r"); 2838 if (!filep) 2839 return 0; 2840 if (fscanf(filep, "%d", &value) != 1) 2841 err(1, "%s: failed to parse number from file", path); 2842 fclose(filep); 2843 return value; 2844 } 2845 2846 /* 2847 * cpu_is_first_core_in_package(cpu) 2848 * return 1 if given CPU is 1st core in package 2849 */ 2850 int cpu_is_first_core_in_package(int cpu) 2851 { 2852 return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu); 2853 } 2854 2855 int get_physical_package_id(int cpu) 2856 { 2857 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu); 2858 } 2859 2860 int get_die_id(int cpu) 2861 { 2862 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/die_id", cpu); 2863 } 2864 2865 int get_core_id(int cpu) 2866 { 2867 return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu); 2868 } 2869 2870 void set_node_data(void) 2871 { 2872 int pkg, node, lnode, cpu, cpux; 2873 int cpu_count; 2874 2875 /* initialize logical_node_id */ 2876 for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) 2877 cpus[cpu].logical_node_id = -1; 2878 2879 cpu_count = 0; 2880 for (pkg = 0; pkg < topo.num_packages; pkg++) { 2881 lnode = 0; 2882 for (cpu = 0; cpu <= topo.max_cpu_num; ++cpu) { 2883 if (cpus[cpu].physical_package_id != pkg) 2884 continue; 2885 /* find a cpu with an unset logical_node_id */ 2886 if (cpus[cpu].logical_node_id != -1) 2887 continue; 2888 cpus[cpu].logical_node_id = lnode; 2889 node = cpus[cpu].physical_node_id; 2890 cpu_count++; 2891 /* 2892 * find all matching cpus on this pkg and set 2893 * the logical_node_id 2894 */ 2895 for (cpux = cpu; cpux <= topo.max_cpu_num; cpux++) { 2896 if ((cpus[cpux].physical_package_id == pkg) && (cpus[cpux].physical_node_id == node)) { 2897 cpus[cpux].logical_node_id = lnode; 2898 cpu_count++; 2899 } 2900 } 2901 lnode++; 2902 if (lnode > topo.nodes_per_pkg) 2903 topo.nodes_per_pkg = lnode; 2904 } 2905 if (cpu_count >= topo.max_cpu_num) 2906 break; 2907 } 2908 } 2909 2910 int get_physical_node_id(struct cpu_topology *thiscpu) 2911 { 2912 char path[80]; 2913 FILE *filep; 2914 int i; 2915 int cpu = thiscpu->logical_cpu_id; 2916 2917 for (i = 0; i <= topo.max_cpu_num; i++) { 2918 sprintf(path, "/sys/devices/system/cpu/cpu%d/node%i/cpulist", cpu, i); 2919 filep = fopen(path, "r"); 2920 if (!filep) 2921 continue; 2922 fclose(filep); 2923 return i; 2924 } 2925 return -1; 2926 } 2927 2928 int get_thread_siblings(struct cpu_topology *thiscpu) 2929 { 2930 char path[80], character; 2931 FILE *filep; 2932 unsigned long map; 2933 int so, shift, sib_core; 2934 int cpu = thiscpu->logical_cpu_id; 2935 int offset = topo.max_cpu_num + 1; 2936 size_t size; 2937 int thread_id = 0; 2938 2939 thiscpu->put_ids = CPU_ALLOC((topo.max_cpu_num + 1)); 2940 if (thiscpu->thread_id < 0) 2941 thiscpu->thread_id = thread_id++; 2942 if (!thiscpu->put_ids) 2943 return -1; 2944 2945 size = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); 2946 CPU_ZERO_S(size, thiscpu->put_ids); 2947 2948 sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings", cpu); 2949 filep = fopen(path, "r"); 2950 2951 if (!filep) { 2952 warnx("%s: open failed", path); 2953 return -1; 2954 } 2955 do { 2956 offset -= BITMASK_SIZE; 2957 if (fscanf(filep, "%lx%c", &map, &character) != 2) 2958 err(1, "%s: failed to parse file", path); 2959 for (shift = 0; shift < BITMASK_SIZE; shift++) { 2960 if ((map >> shift) & 0x1) { 2961 so = shift + offset; 2962 sib_core = get_core_id(so); 2963 if (sib_core == thiscpu->physical_core_id) { 2964 CPU_SET_S(so, size, thiscpu->put_ids); 2965 if ((so != cpu) && (cpus[so].thread_id < 0)) 2966 cpus[so].thread_id = thread_id++; 2967 } 2968 } 2969 } 2970 } while (character == ','); 2971 fclose(filep); 2972 2973 return CPU_COUNT_S(size, thiscpu->put_ids); 2974 } 2975 2976 /* 2977 * run func(thread, core, package) in topology order 2978 * skip non-present cpus 2979 */ 2980 2981 int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *, 2982 struct pkg_data *, struct thread_data *, struct core_data *, 2983 struct pkg_data *), struct thread_data *thread_base, 2984 struct core_data *core_base, struct pkg_data *pkg_base, 2985 struct thread_data *thread_base2, struct core_data *core_base2, struct pkg_data *pkg_base2) 2986 { 2987 int retval, pkg_no, node_no, core_no, thread_no; 2988 2989 for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { 2990 for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) { 2991 for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { 2992 for (thread_no = 0; thread_no < topo.threads_per_core; ++thread_no) { 2993 struct thread_data *t, *t2; 2994 struct core_data *c, *c2; 2995 struct pkg_data *p, *p2; 2996 2997 t = GET_THREAD(thread_base, thread_no, core_no, node_no, pkg_no); 2998 2999 if (cpu_is_not_present(t->cpu_id)) 3000 continue; 3001 3002 t2 = GET_THREAD(thread_base2, thread_no, core_no, node_no, pkg_no); 3003 3004 c = GET_CORE(core_base, core_no, node_no, pkg_no); 3005 c2 = GET_CORE(core_base2, core_no, node_no, pkg_no); 3006 3007 p = GET_PKG(pkg_base, pkg_no); 3008 p2 = GET_PKG(pkg_base2, pkg_no); 3009 3010 retval = func(t, c, p, t2, c2, p2); 3011 if (retval) 3012 return retval; 3013 } 3014 } 3015 } 3016 } 3017 return 0; 3018 } 3019 3020 /* 3021 * run func(cpu) on every cpu in /proc/stat 3022 * return max_cpu number 3023 */ 3024 int for_all_proc_cpus(int (func) (int)) 3025 { 3026 FILE *fp; 3027 int cpu_num; 3028 int retval; 3029 3030 fp = fopen_or_die(proc_stat, "r"); 3031 3032 retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); 3033 if (retval != 0) 3034 err(1, "%s: failed to parse format", proc_stat); 3035 3036 while (1) { 3037 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num); 3038 if (retval != 1) 3039 break; 3040 3041 retval = func(cpu_num); 3042 if (retval) { 3043 fclose(fp); 3044 return (retval); 3045 } 3046 } 3047 fclose(fp); 3048 return 0; 3049 } 3050 3051 void re_initialize(void) 3052 { 3053 free_all_buffers(); 3054 setup_all_buffers(); 3055 fprintf(outf, "turbostat: re-initialized with num_cpus %d\n", topo.num_cpus); 3056 } 3057 3058 void set_max_cpu_num(void) 3059 { 3060 FILE *filep; 3061 int base_cpu; 3062 unsigned long dummy; 3063 char pathname[64]; 3064 3065 base_cpu = sched_getcpu(); 3066 if (base_cpu < 0) 3067 err(1, "cannot find calling cpu ID"); 3068 sprintf(pathname, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings", base_cpu); 3069 3070 filep = fopen_or_die(pathname, "r"); 3071 topo.max_cpu_num = 0; 3072 while (fscanf(filep, "%lx,", &dummy) == 1) 3073 topo.max_cpu_num += BITMASK_SIZE; 3074 fclose(filep); 3075 topo.max_cpu_num--; /* 0 based */ 3076 } 3077 3078 /* 3079 * count_cpus() 3080 * remember the last one seen, it will be the max 3081 */ 3082 int count_cpus(int cpu) 3083 { 3084 UNUSED(cpu); 3085 3086 topo.num_cpus++; 3087 return 0; 3088 } 3089 3090 int mark_cpu_present(int cpu) 3091 { 3092 CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); 3093 return 0; 3094 } 3095 3096 int init_thread_id(int cpu) 3097 { 3098 cpus[cpu].thread_id = -1; 3099 return 0; 3100 } 3101 3102 /* 3103 * snapshot_proc_interrupts() 3104 * 3105 * read and record summary of /proc/interrupts 3106 * 3107 * return 1 if config change requires a restart, else return 0 3108 */ 3109 int snapshot_proc_interrupts(void) 3110 { 3111 static FILE *fp; 3112 int column, retval; 3113 3114 if (fp == NULL) 3115 fp = fopen_or_die("/proc/interrupts", "r"); 3116 else 3117 rewind(fp); 3118 3119 /* read 1st line of /proc/interrupts to get cpu* name for each column */ 3120 for (column = 0; column < topo.num_cpus; ++column) { 3121 int cpu_number; 3122 3123 retval = fscanf(fp, " CPU%d", &cpu_number); 3124 if (retval != 1) 3125 break; 3126 3127 if (cpu_number > topo.max_cpu_num) { 3128 warn("/proc/interrupts: cpu%d: > %d", cpu_number, topo.max_cpu_num); 3129 return 1; 3130 } 3131 3132 irq_column_2_cpu[column] = cpu_number; 3133 irqs_per_cpu[cpu_number] = 0; 3134 } 3135 3136 /* read /proc/interrupt count lines and sum up irqs per cpu */ 3137 while (1) { 3138 int column; 3139 char buf[64]; 3140 3141 retval = fscanf(fp, " %s:", buf); /* flush irq# "N:" */ 3142 if (retval != 1) 3143 break; 3144 3145 /* read the count per cpu */ 3146 for (column = 0; column < topo.num_cpus; ++column) { 3147 3148 int cpu_number, irq_count; 3149 3150 retval = fscanf(fp, " %d", &irq_count); 3151 if (retval != 1) 3152 break; 3153 3154 cpu_number = irq_column_2_cpu[column]; 3155 irqs_per_cpu[cpu_number] += irq_count; 3156 3157 } 3158 3159 while (getc(fp) != '\n') ; /* flush interrupt description */ 3160 3161 } 3162 return 0; 3163 } 3164 3165 /* 3166 * snapshot_gfx_rc6_ms() 3167 * 3168 * record snapshot of 3169 * /sys/class/drm/card0/power/rc6_residency_ms 3170 * 3171 * return 1 if config change requires a restart, else return 0 3172 */ 3173 int snapshot_gfx_rc6_ms(void) 3174 { 3175 FILE *fp; 3176 int retval; 3177 3178 fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r"); 3179 3180 retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms); 3181 if (retval != 1) 3182 err(1, "GFX rc6"); 3183 3184 fclose(fp); 3185 3186 return 0; 3187 } 3188 3189 /* 3190 * snapshot_gfx_mhz() 3191 * 3192 * record snapshot of 3193 * /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz 3194 * 3195 * return 1 if config change requires a restart, else return 0 3196 */ 3197 int snapshot_gfx_mhz(void) 3198 { 3199 static FILE *fp; 3200 int retval; 3201 3202 if (fp == NULL) 3203 fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r"); 3204 else { 3205 rewind(fp); 3206 fflush(fp); 3207 } 3208 3209 retval = fscanf(fp, "%d", &gfx_cur_mhz); 3210 if (retval != 1) 3211 err(1, "GFX MHz"); 3212 3213 return 0; 3214 } 3215 3216 /* 3217 * snapshot_gfx_cur_mhz() 3218 * 3219 * record snapshot of 3220 * /sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz 3221 * 3222 * return 1 if config change requires a restart, else return 0 3223 */ 3224 int snapshot_gfx_act_mhz(void) 3225 { 3226 static FILE *fp; 3227 int retval; 3228 3229 if (fp == NULL) 3230 fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", "r"); 3231 else { 3232 rewind(fp); 3233 fflush(fp); 3234 } 3235 3236 retval = fscanf(fp, "%d", &gfx_act_mhz); 3237 if (retval != 1) 3238 err(1, "GFX ACT MHz"); 3239 3240 return 0; 3241 } 3242 3243 /* 3244 * snapshot_cpu_lpi() 3245 * 3246 * record snapshot of 3247 * /sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us 3248 */ 3249 int snapshot_cpu_lpi_us(void) 3250 { 3251 FILE *fp; 3252 int retval; 3253 3254 fp = fopen_or_die("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", "r"); 3255 3256 retval = fscanf(fp, "%lld", &cpuidle_cur_cpu_lpi_us); 3257 if (retval != 1) { 3258 fprintf(stderr, "Disabling Low Power Idle CPU output\n"); 3259 BIC_NOT_PRESENT(BIC_CPU_LPI); 3260 fclose(fp); 3261 return -1; 3262 } 3263 3264 fclose(fp); 3265 3266 return 0; 3267 } 3268 3269 /* 3270 * snapshot_sys_lpi() 3271 * 3272 * record snapshot of sys_lpi_file 3273 */ 3274 int snapshot_sys_lpi_us(void) 3275 { 3276 FILE *fp; 3277 int retval; 3278 3279 fp = fopen_or_die(sys_lpi_file, "r"); 3280 3281 retval = fscanf(fp, "%lld", &cpuidle_cur_sys_lpi_us); 3282 if (retval != 1) { 3283 fprintf(stderr, "Disabling Low Power Idle System output\n"); 3284 BIC_NOT_PRESENT(BIC_SYS_LPI); 3285 fclose(fp); 3286 return -1; 3287 } 3288 fclose(fp); 3289 3290 return 0; 3291 } 3292 3293 /* 3294 * snapshot /proc and /sys files 3295 * 3296 * return 1 if configuration restart needed, else return 0 3297 */ 3298 int snapshot_proc_sysfs_files(void) 3299 { 3300 if (DO_BIC(BIC_IRQ)) 3301 if (snapshot_proc_interrupts()) 3302 return 1; 3303 3304 if (DO_BIC(BIC_GFX_rc6)) 3305 snapshot_gfx_rc6_ms(); 3306 3307 if (DO_BIC(BIC_GFXMHz)) 3308 snapshot_gfx_mhz(); 3309 3310 if (DO_BIC(BIC_GFXACTMHz)) 3311 snapshot_gfx_act_mhz(); 3312 3313 if (DO_BIC(BIC_CPU_LPI)) 3314 snapshot_cpu_lpi_us(); 3315 3316 if (DO_BIC(BIC_SYS_LPI)) 3317 snapshot_sys_lpi_us(); 3318 3319 return 0; 3320 } 3321 3322 int exit_requested; 3323 3324 static void signal_handler(int signal) 3325 { 3326 switch (signal) { 3327 case SIGINT: 3328 exit_requested = 1; 3329 if (debug) 3330 fprintf(stderr, " SIGINT\n"); 3331 break; 3332 case SIGUSR1: 3333 if (debug > 1) 3334 fprintf(stderr, "SIGUSR1\n"); 3335 break; 3336 } 3337 } 3338 3339 void setup_signal_handler(void) 3340 { 3341 struct sigaction sa; 3342 3343 memset(&sa, 0, sizeof(sa)); 3344 3345 sa.sa_handler = &signal_handler; 3346 3347 if (sigaction(SIGINT, &sa, NULL) < 0) 3348 err(1, "sigaction SIGINT"); 3349 if (sigaction(SIGUSR1, &sa, NULL) < 0) 3350 err(1, "sigaction SIGUSR1"); 3351 } 3352 3353 void do_sleep(void) 3354 { 3355 struct timeval tout; 3356 struct timespec rest; 3357 fd_set readfds; 3358 int retval; 3359 3360 FD_ZERO(&readfds); 3361 FD_SET(0, &readfds); 3362 3363 if (ignore_stdin) { 3364 nanosleep(&interval_ts, NULL); 3365 return; 3366 } 3367 3368 tout = interval_tv; 3369 retval = select(1, &readfds, NULL, NULL, &tout); 3370 3371 if (retval == 1) { 3372 switch (getc(stdin)) { 3373 case 'q': 3374 exit_requested = 1; 3375 break; 3376 case EOF: 3377 /* 3378 * 'stdin' is a pipe closed on the other end. There 3379 * won't be any further input. 3380 */ 3381 ignore_stdin = 1; 3382 /* Sleep the rest of the time */ 3383 rest.tv_sec = (tout.tv_sec + tout.tv_usec / 1000000); 3384 rest.tv_nsec = (tout.tv_usec % 1000000) * 1000; 3385 nanosleep(&rest, NULL); 3386 } 3387 } 3388 } 3389 3390 int get_msr_sum(int cpu, off_t offset, unsigned long long *msr) 3391 { 3392 int ret, idx; 3393 unsigned long long msr_cur, msr_last; 3394 3395 if (!per_cpu_msr_sum) 3396 return 1; 3397 3398 idx = offset_to_idx(offset); 3399 if (idx < 0) 3400 return idx; 3401 /* get_msr_sum() = sum + (get_msr() - last) */ 3402 ret = get_msr(cpu, offset, &msr_cur); 3403 if (ret) 3404 return ret; 3405 msr_last = per_cpu_msr_sum[cpu].entries[idx].last; 3406 DELTA_WRAP32(msr_cur, msr_last); 3407 *msr = msr_last + per_cpu_msr_sum[cpu].entries[idx].sum; 3408 3409 return 0; 3410 } 3411 3412 timer_t timerid; 3413 3414 /* Timer callback, update the sum of MSRs periodically. */ 3415 static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg_data *p) 3416 { 3417 int i, ret; 3418 int cpu = t->cpu_id; 3419 3420 UNUSED(c); 3421 UNUSED(p); 3422 3423 for (i = IDX_PKG_ENERGY; i < IDX_COUNT; i++) { 3424 unsigned long long msr_cur, msr_last; 3425 off_t offset; 3426 3427 if (!idx_valid(i)) 3428 continue; 3429 offset = idx_to_offset(i); 3430 if (offset < 0) 3431 continue; 3432 ret = get_msr(cpu, offset, &msr_cur); 3433 if (ret) { 3434 fprintf(outf, "Can not update msr(0x%llx)\n", (unsigned long long)offset); 3435 continue; 3436 } 3437 3438 msr_last = per_cpu_msr_sum[cpu].entries[i].last; 3439 per_cpu_msr_sum[cpu].entries[i].last = msr_cur & 0xffffffff; 3440 3441 DELTA_WRAP32(msr_cur, msr_last); 3442 per_cpu_msr_sum[cpu].entries[i].sum += msr_last; 3443 } 3444 return 0; 3445 } 3446 3447 static void msr_record_handler(union sigval v) 3448 { 3449 UNUSED(v); 3450 3451 for_all_cpus(update_msr_sum, EVEN_COUNTERS); 3452 } 3453 3454 void msr_sum_record(void) 3455 { 3456 struct itimerspec its; 3457 struct sigevent sev; 3458 3459 per_cpu_msr_sum = calloc(topo.max_cpu_num + 1, sizeof(struct msr_sum_array)); 3460 if (!per_cpu_msr_sum) { 3461 fprintf(outf, "Can not allocate memory for long time MSR.\n"); 3462 return; 3463 } 3464 /* 3465 * Signal handler might be restricted, so use thread notifier instead. 3466 */ 3467 memset(&sev, 0, sizeof(struct sigevent)); 3468 sev.sigev_notify = SIGEV_THREAD; 3469 sev.sigev_notify_function = msr_record_handler; 3470 3471 sev.sigev_value.sival_ptr = &timerid; 3472 if (timer_create(CLOCK_REALTIME, &sev, &timerid) == -1) { 3473 fprintf(outf, "Can not create timer.\n"); 3474 goto release_msr; 3475 } 3476 3477 its.it_value.tv_sec = 0; 3478 its.it_value.tv_nsec = 1; 3479 /* 3480 * A wraparound time has been calculated early. 3481 * Some sources state that the peak power for a 3482 * microprocessor is usually 1.5 times the TDP rating, 3483 * use 2 * TDP for safety. 3484 */ 3485 its.it_interval.tv_sec = rapl_joule_counter_range / 2; 3486 its.it_interval.tv_nsec = 0; 3487 3488 if (timer_settime(timerid, 0, &its, NULL) == -1) { 3489 fprintf(outf, "Can not set timer.\n"); 3490 goto release_timer; 3491 } 3492 return; 3493 3494 release_timer: 3495 timer_delete(timerid); 3496 release_msr: 3497 free(per_cpu_msr_sum); 3498 } 3499 3500 /* 3501 * set_my_sched_priority(pri) 3502 * return previous 3503 * 3504 * if non-root, do this: 3505 * # /sbin/setcap cap_sys_rawio,cap_sys_nice=+ep /usr/bin/turbostat 3506 */ 3507 int set_my_sched_priority(int priority) 3508 { 3509 int retval; 3510 int original_priority; 3511 3512 errno = 0; 3513 original_priority = getpriority(PRIO_PROCESS, 0); 3514 if (errno && (original_priority == -1)) 3515 err(errno, "getpriority"); 3516 3517 retval = setpriority(PRIO_PROCESS, 0, priority); 3518 if (retval) 3519 err(retval, "setpriority(%d)", priority); 3520 3521 errno = 0; 3522 retval = getpriority(PRIO_PROCESS, 0); 3523 if (retval != priority) 3524 err(retval, "getpriority(%d) != setpriority(%d)", retval, priority); 3525 3526 return original_priority; 3527 } 3528 3529 void turbostat_loop() 3530 { 3531 int retval; 3532 int restarted = 0; 3533 unsigned int done_iters = 0; 3534 3535 setup_signal_handler(); 3536 3537 /* 3538 * elevate own priority for interval mode 3539 */ 3540 set_my_sched_priority(-20); 3541 3542 restart: 3543 restarted++; 3544 3545 snapshot_proc_sysfs_files(); 3546 retval = for_all_cpus(get_counters, EVEN_COUNTERS); 3547 first_counter_read = 0; 3548 if (retval < -1) { 3549 exit(retval); 3550 } else if (retval == -1) { 3551 if (restarted > 10) { 3552 exit(retval); 3553 } 3554 re_initialize(); 3555 goto restart; 3556 } 3557 restarted = 0; 3558 done_iters = 0; 3559 gettimeofday(&tv_even, (struct timezone *)NULL); 3560 3561 while (1) { 3562 if (for_all_proc_cpus(cpu_is_not_present)) { 3563 re_initialize(); 3564 goto restart; 3565 } 3566 do_sleep(); 3567 if (snapshot_proc_sysfs_files()) 3568 goto restart; 3569 retval = for_all_cpus(get_counters, ODD_COUNTERS); 3570 if (retval < -1) { 3571 exit(retval); 3572 } else if (retval == -1) { 3573 re_initialize(); 3574 goto restart; 3575 } 3576 gettimeofday(&tv_odd, (struct timezone *)NULL); 3577 timersub(&tv_odd, &tv_even, &tv_delta); 3578 if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) { 3579 re_initialize(); 3580 goto restart; 3581 } 3582 compute_average(EVEN_COUNTERS); 3583 format_all_counters(EVEN_COUNTERS); 3584 flush_output_stdout(); 3585 if (exit_requested) 3586 break; 3587 if (num_iterations && ++done_iters >= num_iterations) 3588 break; 3589 do_sleep(); 3590 if (snapshot_proc_sysfs_files()) 3591 goto restart; 3592 retval = for_all_cpus(get_counters, EVEN_COUNTERS); 3593 if (retval < -1) { 3594 exit(retval); 3595 } else if (retval == -1) { 3596 re_initialize(); 3597 goto restart; 3598 } 3599 gettimeofday(&tv_even, (struct timezone *)NULL); 3600 timersub(&tv_even, &tv_odd, &tv_delta); 3601 if (for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) { 3602 re_initialize(); 3603 goto restart; 3604 } 3605 compute_average(ODD_COUNTERS); 3606 format_all_counters(ODD_COUNTERS); 3607 flush_output_stdout(); 3608 if (exit_requested) 3609 break; 3610 if (num_iterations && ++done_iters >= num_iterations) 3611 break; 3612 } 3613 } 3614 3615 void check_dev_msr() 3616 { 3617 struct stat sb; 3618 char pathname[32]; 3619 3620 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); 3621 if (stat(pathname, &sb)) 3622 if (system("/sbin/modprobe msr > /dev/null 2>&1")) 3623 err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" "); 3624 } 3625 3626 /* 3627 * check for CAP_SYS_RAWIO 3628 * return 0 on success 3629 * return 1 on fail 3630 */ 3631 int check_for_cap_sys_rawio(void) 3632 { 3633 cap_t caps; 3634 cap_flag_value_t cap_flag_value; 3635 3636 caps = cap_get_proc(); 3637 if (caps == NULL) 3638 err(-6, "cap_get_proc\n"); 3639 3640 if (cap_get_flag(caps, CAP_SYS_RAWIO, CAP_EFFECTIVE, &cap_flag_value)) 3641 err(-6, "cap_get\n"); 3642 3643 if (cap_flag_value != CAP_SET) { 3644 warnx("capget(CAP_SYS_RAWIO) failed," " try \"# setcap cap_sys_rawio=ep %s\"", progname); 3645 return 1; 3646 } 3647 3648 if (cap_free(caps) == -1) 3649 err(-6, "cap_free\n"); 3650 3651 return 0; 3652 } 3653 3654 void check_permissions(void) 3655 { 3656 int do_exit = 0; 3657 char pathname[32]; 3658 3659 /* check for CAP_SYS_RAWIO */ 3660 do_exit += check_for_cap_sys_rawio(); 3661 3662 /* test file permissions */ 3663 sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); 3664 if (euidaccess(pathname, R_OK)) { 3665 do_exit++; 3666 warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr"); 3667 } 3668 3669 /* if all else fails, thell them to be root */ 3670 if (do_exit) 3671 if (getuid() != 0) 3672 warnx("... or simply run as root"); 3673 3674 if (do_exit) 3675 exit(-6); 3676 } 3677 3678 /* 3679 * NHM adds support for additional MSRs: 3680 * 3681 * MSR_SMI_COUNT 0x00000034 3682 * 3683 * MSR_PLATFORM_INFO 0x000000ce 3684 * MSR_PKG_CST_CONFIG_CONTROL 0x000000e2 3685 * 3686 * MSR_MISC_PWR_MGMT 0x000001aa 3687 * 3688 * MSR_PKG_C3_RESIDENCY 0x000003f8 3689 * MSR_PKG_C6_RESIDENCY 0x000003f9 3690 * MSR_CORE_C3_RESIDENCY 0x000003fc 3691 * MSR_CORE_C6_RESIDENCY 0x000003fd 3692 * 3693 * Side effect: 3694 * sets global pkg_cstate_limit to decode MSR_PKG_CST_CONFIG_CONTROL 3695 * sets has_misc_feature_control 3696 */ 3697 int probe_nhm_msrs(unsigned int family, unsigned int model) 3698 { 3699 unsigned long long msr; 3700 unsigned int base_ratio; 3701 int *pkg_cstate_limits; 3702 3703 if (!genuine_intel) 3704 return 0; 3705 3706 if (family != 6) 3707 return 0; 3708 3709 bclk = discover_bclk(family, model); 3710 3711 switch (model) { 3712 case INTEL_FAM6_NEHALEM: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */ 3713 case INTEL_FAM6_NEHALEM_EX: /* Nehalem-EX Xeon - Beckton */ 3714 pkg_cstate_limits = nhm_pkg_cstate_limits; 3715 break; 3716 case INTEL_FAM6_SANDYBRIDGE: /* SNB */ 3717 case INTEL_FAM6_SANDYBRIDGE_X: /* SNB Xeon */ 3718 case INTEL_FAM6_IVYBRIDGE: /* IVB */ 3719 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */ 3720 pkg_cstate_limits = snb_pkg_cstate_limits; 3721 has_misc_feature_control = 1; 3722 break; 3723 case INTEL_FAM6_HASWELL: /* HSW */ 3724 case INTEL_FAM6_HASWELL_G: /* HSW */ 3725 case INTEL_FAM6_HASWELL_X: /* HSX */ 3726 case INTEL_FAM6_HASWELL_L: /* HSW */ 3727 case INTEL_FAM6_BROADWELL: /* BDW */ 3728 case INTEL_FAM6_BROADWELL_G: /* BDW */ 3729 case INTEL_FAM6_BROADWELL_X: /* BDX */ 3730 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 3731 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 3732 pkg_cstate_limits = hsw_pkg_cstate_limits; 3733 has_misc_feature_control = 1; 3734 break; 3735 case INTEL_FAM6_SKYLAKE_X: /* SKX */ 3736 pkg_cstate_limits = skx_pkg_cstate_limits; 3737 has_misc_feature_control = 1; 3738 break; 3739 case INTEL_FAM6_ICELAKE_X: /* ICX */ 3740 pkg_cstate_limits = icx_pkg_cstate_limits; 3741 has_misc_feature_control = 1; 3742 break; 3743 case INTEL_FAM6_ATOM_SILVERMONT: /* BYT */ 3744 no_MSR_MISC_PWR_MGMT = 1; 3745 /* FALLTHRU */ 3746 case INTEL_FAM6_ATOM_SILVERMONT_D: /* AVN */ 3747 pkg_cstate_limits = slv_pkg_cstate_limits; 3748 break; 3749 case INTEL_FAM6_ATOM_AIRMONT: /* AMT */ 3750 pkg_cstate_limits = amt_pkg_cstate_limits; 3751 no_MSR_MISC_PWR_MGMT = 1; 3752 break; 3753 case INTEL_FAM6_XEON_PHI_KNL: /* PHI */ 3754 pkg_cstate_limits = phi_pkg_cstate_limits; 3755 break; 3756 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */ 3757 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 3758 case INTEL_FAM6_ATOM_GOLDMONT_D: /* DNV */ 3759 case INTEL_FAM6_ATOM_TREMONT: /* EHL */ 3760 case INTEL_FAM6_ATOM_TREMONT_D: /* JVL */ 3761 pkg_cstate_limits = glm_pkg_cstate_limits; 3762 break; 3763 default: 3764 return 0; 3765 } 3766 get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr); 3767 pkg_cstate_limit = pkg_cstate_limits[msr & 0xF]; 3768 3769 get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); 3770 base_ratio = (msr >> 8) & 0xFF; 3771 3772 base_hz = base_ratio * bclk * 1000000; 3773 has_base_hz = 1; 3774 return 1; 3775 } 3776 3777 /* 3778 * SLV client has support for unique MSRs: 3779 * 3780 * MSR_CC6_DEMOTION_POLICY_CONFIG 3781 * MSR_MC6_DEMOTION_POLICY_CONFIG 3782 */ 3783 3784 int has_slv_msrs(unsigned int family, unsigned int model) 3785 { 3786 if (!genuine_intel) 3787 return 0; 3788 3789 if (family != 6) 3790 return 0; 3791 3792 switch (model) { 3793 case INTEL_FAM6_ATOM_SILVERMONT: 3794 case INTEL_FAM6_ATOM_SILVERMONT_MID: 3795 case INTEL_FAM6_ATOM_AIRMONT_MID: 3796 return 1; 3797 } 3798 return 0; 3799 } 3800 3801 int is_dnv(unsigned int family, unsigned int model) 3802 { 3803 3804 if (!genuine_intel) 3805 return 0; 3806 3807 if (family != 6) 3808 return 0; 3809 3810 switch (model) { 3811 case INTEL_FAM6_ATOM_GOLDMONT_D: 3812 return 1; 3813 } 3814 return 0; 3815 } 3816 3817 int is_bdx(unsigned int family, unsigned int model) 3818 { 3819 3820 if (!genuine_intel) 3821 return 0; 3822 3823 if (family != 6) 3824 return 0; 3825 3826 switch (model) { 3827 case INTEL_FAM6_BROADWELL_X: 3828 return 1; 3829 } 3830 return 0; 3831 } 3832 3833 int is_skx(unsigned int family, unsigned int model) 3834 { 3835 3836 if (!genuine_intel) 3837 return 0; 3838 3839 if (family != 6) 3840 return 0; 3841 3842 switch (model) { 3843 case INTEL_FAM6_SKYLAKE_X: 3844 return 1; 3845 } 3846 return 0; 3847 } 3848 3849 int is_icx(unsigned int family, unsigned int model) 3850 { 3851 3852 if (!genuine_intel) 3853 return 0; 3854 3855 if (family != 6) 3856 return 0; 3857 3858 switch (model) { 3859 case INTEL_FAM6_ICELAKE_X: 3860 return 1; 3861 } 3862 return 0; 3863 } 3864 3865 int is_ehl(unsigned int family, unsigned int model) 3866 { 3867 if (!genuine_intel) 3868 return 0; 3869 3870 if (family != 6) 3871 return 0; 3872 3873 switch (model) { 3874 case INTEL_FAM6_ATOM_TREMONT: 3875 return 1; 3876 } 3877 return 0; 3878 } 3879 3880 int is_jvl(unsigned int family, unsigned int model) 3881 { 3882 if (!genuine_intel) 3883 return 0; 3884 3885 if (family != 6) 3886 return 0; 3887 3888 switch (model) { 3889 case INTEL_FAM6_ATOM_TREMONT_D: 3890 return 1; 3891 } 3892 return 0; 3893 } 3894 3895 int has_turbo_ratio_limit(unsigned int family, unsigned int model) 3896 { 3897 if (has_slv_msrs(family, model)) 3898 return 0; 3899 3900 if (family != 6) 3901 return 0; 3902 3903 switch (model) { 3904 /* Nehalem compatible, but do not include turbo-ratio limit support */ 3905 case INTEL_FAM6_NEHALEM_EX: /* Nehalem-EX Xeon - Beckton */ 3906 case INTEL_FAM6_XEON_PHI_KNL: /* PHI - Knights Landing (different MSR definition) */ 3907 return 0; 3908 default: 3909 return 1; 3910 } 3911 } 3912 3913 int has_atom_turbo_ratio_limit(unsigned int family, unsigned int model) 3914 { 3915 if (has_slv_msrs(family, model)) 3916 return 1; 3917 3918 return 0; 3919 } 3920 3921 int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model) 3922 { 3923 if (!genuine_intel) 3924 return 0; 3925 3926 if (family != 6) 3927 return 0; 3928 3929 switch (model) { 3930 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */ 3931 case INTEL_FAM6_HASWELL_X: /* HSW Xeon */ 3932 return 1; 3933 default: 3934 return 0; 3935 } 3936 } 3937 3938 int has_hsw_turbo_ratio_limit(unsigned int family, unsigned int model) 3939 { 3940 if (!genuine_intel) 3941 return 0; 3942 3943 if (family != 6) 3944 return 0; 3945 3946 switch (model) { 3947 case INTEL_FAM6_HASWELL_X: /* HSW Xeon */ 3948 return 1; 3949 default: 3950 return 0; 3951 } 3952 } 3953 3954 int has_knl_turbo_ratio_limit(unsigned int family, unsigned int model) 3955 { 3956 if (!genuine_intel) 3957 return 0; 3958 3959 if (family != 6) 3960 return 0; 3961 3962 switch (model) { 3963 case INTEL_FAM6_XEON_PHI_KNL: /* Knights Landing */ 3964 return 1; 3965 default: 3966 return 0; 3967 } 3968 } 3969 3970 int has_glm_turbo_ratio_limit(unsigned int family, unsigned int model) 3971 { 3972 if (!genuine_intel) 3973 return 0; 3974 3975 if (family != 6) 3976 return 0; 3977 3978 switch (model) { 3979 case INTEL_FAM6_ATOM_GOLDMONT: 3980 case INTEL_FAM6_SKYLAKE_X: 3981 case INTEL_FAM6_ICELAKE_X: 3982 return 1; 3983 default: 3984 return 0; 3985 } 3986 } 3987 3988 int has_config_tdp(unsigned int family, unsigned int model) 3989 { 3990 if (!genuine_intel) 3991 return 0; 3992 3993 if (family != 6) 3994 return 0; 3995 3996 switch (model) { 3997 case INTEL_FAM6_IVYBRIDGE: /* IVB */ 3998 case INTEL_FAM6_HASWELL: /* HSW */ 3999 case INTEL_FAM6_HASWELL_X: /* HSX */ 4000 case INTEL_FAM6_HASWELL_L: /* HSW */ 4001 case INTEL_FAM6_HASWELL_G: /* HSW */ 4002 case INTEL_FAM6_BROADWELL: /* BDW */ 4003 case INTEL_FAM6_BROADWELL_G: /* BDW */ 4004 case INTEL_FAM6_BROADWELL_X: /* BDX */ 4005 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 4006 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 4007 case INTEL_FAM6_SKYLAKE_X: /* SKX */ 4008 case INTEL_FAM6_ICELAKE_X: /* ICX */ 4009 4010 case INTEL_FAM6_XEON_PHI_KNL: /* Knights Landing */ 4011 return 1; 4012 default: 4013 return 0; 4014 } 4015 } 4016 4017 /* 4018 * tcc_offset_bits: 4019 * 0: Tcc Offset not supported (Default) 4020 * 6: Bit 29:24 of MSR_PLATFORM_INFO 4021 * 4: Bit 27:24 of MSR_PLATFORM_INFO 4022 */ 4023 void check_tcc_offset(int model) 4024 { 4025 unsigned long long msr; 4026 4027 if (!genuine_intel) 4028 return; 4029 4030 switch (model) { 4031 case INTEL_FAM6_SKYLAKE_L: 4032 case INTEL_FAM6_SKYLAKE: 4033 case INTEL_FAM6_KABYLAKE_L: 4034 case INTEL_FAM6_KABYLAKE: 4035 case INTEL_FAM6_ICELAKE_L: 4036 case INTEL_FAM6_ICELAKE: 4037 case INTEL_FAM6_TIGERLAKE_L: 4038 case INTEL_FAM6_TIGERLAKE: 4039 case INTEL_FAM6_COMETLAKE: 4040 if (!get_msr(base_cpu, MSR_PLATFORM_INFO, &msr)) { 4041 msr = (msr >> 30) & 1; 4042 if (msr) 4043 tcc_offset_bits = 6; 4044 } 4045 return; 4046 default: 4047 return; 4048 } 4049 } 4050 4051 static void remove_underbar(char *s) 4052 { 4053 char *to = s; 4054 4055 while (*s) { 4056 if (*s != '_') 4057 *to++ = *s; 4058 s++; 4059 } 4060 4061 *to = 0; 4062 } 4063 4064 static void dump_cstate_pstate_config_info(unsigned int family, unsigned int model) 4065 { 4066 if (!do_nhm_platform_info) 4067 return; 4068 4069 dump_nhm_platform_info(); 4070 4071 if (has_hsw_turbo_ratio_limit(family, model)) 4072 dump_hsw_turbo_ratio_limits(); 4073 4074 if (has_ivt_turbo_ratio_limit(family, model)) 4075 dump_ivt_turbo_ratio_limits(); 4076 4077 if (has_turbo_ratio_limit(family, model)) { 4078 dump_turbo_ratio_limits(MSR_TURBO_RATIO_LIMIT, family, model); 4079 4080 if (is_hybrid) 4081 dump_turbo_ratio_limits(MSR_SECONDARY_TURBO_RATIO_LIMIT, family, model); 4082 } 4083 4084 if (has_atom_turbo_ratio_limit(family, model)) 4085 dump_atom_turbo_ratio_limits(); 4086 4087 if (has_knl_turbo_ratio_limit(family, model)) 4088 dump_knl_turbo_ratio_limits(); 4089 4090 if (has_config_tdp(family, model)) 4091 dump_config_tdp(); 4092 4093 dump_nhm_cst_cfg(); 4094 } 4095 4096 static int read_sysfs_int(char *path) 4097 { 4098 FILE *input; 4099 int retval = -1; 4100 4101 input = fopen(path, "r"); 4102 if (input == NULL) { 4103 if (debug) 4104 fprintf(outf, "NSFOD %s\n", path); 4105 return (-1); 4106 } 4107 if (fscanf(input, "%d", &retval) != 1) 4108 err(1, "%s: failed to read int from file", path); 4109 fclose(input); 4110 4111 return (retval); 4112 } 4113 4114 static void dump_sysfs_file(char *path) 4115 { 4116 FILE *input; 4117 char cpuidle_buf[64]; 4118 4119 input = fopen(path, "r"); 4120 if (input == NULL) { 4121 if (debug) 4122 fprintf(outf, "NSFOD %s\n", path); 4123 return; 4124 } 4125 if (!fgets(cpuidle_buf, sizeof(cpuidle_buf), input)) 4126 err(1, "%s: failed to read file", path); 4127 fclose(input); 4128 4129 fprintf(outf, "%s: %s", strrchr(path, '/') + 1, cpuidle_buf); 4130 } 4131 4132 static void intel_uncore_frequency_probe(void) 4133 { 4134 int i, j; 4135 char path[128]; 4136 4137 if (!genuine_intel) 4138 return; 4139 4140 if (access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00", R_OK)) 4141 return; 4142 4143 if (!access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00/current_freq_khz", R_OK)) 4144 BIC_PRESENT(BIC_UNCORE_MHZ); 4145 4146 if (quiet) 4147 return; 4148 4149 for (i = 0; i < topo.num_packages; ++i) { 4150 for (j = 0; j < topo.num_die; ++j) { 4151 int k, l; 4152 4153 sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/min_freq_khz", 4154 i, j); 4155 k = read_sysfs_int(path); 4156 sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/max_freq_khz", 4157 i, j); 4158 l = read_sysfs_int(path); 4159 fprintf(outf, "Uncore Frequency pkg%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000); 4160 4161 sprintf(path, 4162 "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_min_freq_khz", 4163 i, j); 4164 k = read_sysfs_int(path); 4165 sprintf(path, 4166 "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_max_freq_khz", 4167 i, j); 4168 l = read_sysfs_int(path); 4169 fprintf(outf, "(%d - %d MHz)\n", k / 1000, l / 1000); 4170 } 4171 } 4172 } 4173 4174 static void dump_sysfs_cstate_config(void) 4175 { 4176 char path[64]; 4177 char name_buf[16]; 4178 char desc[64]; 4179 FILE *input; 4180 int state; 4181 char *sp; 4182 4183 if (access("/sys/devices/system/cpu/cpuidle", R_OK)) { 4184 fprintf(outf, "cpuidle not loaded\n"); 4185 return; 4186 } 4187 4188 dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_driver"); 4189 dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_governor"); 4190 dump_sysfs_file("/sys/devices/system/cpu/cpuidle/current_governor_ro"); 4191 4192 for (state = 0; state < 10; ++state) { 4193 4194 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state); 4195 input = fopen(path, "r"); 4196 if (input == NULL) 4197 continue; 4198 if (!fgets(name_buf, sizeof(name_buf), input)) 4199 err(1, "%s: failed to read file", path); 4200 4201 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */ 4202 sp = strchr(name_buf, '-'); 4203 if (!sp) 4204 sp = strchrnul(name_buf, '\n'); 4205 *sp = '\0'; 4206 fclose(input); 4207 4208 remove_underbar(name_buf); 4209 4210 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/desc", base_cpu, state); 4211 input = fopen(path, "r"); 4212 if (input == NULL) 4213 continue; 4214 if (!fgets(desc, sizeof(desc), input)) 4215 err(1, "%s: failed to read file", path); 4216 4217 fprintf(outf, "cpu%d: %s: %s", base_cpu, name_buf, desc); 4218 fclose(input); 4219 } 4220 } 4221 4222 static void dump_sysfs_pstate_config(void) 4223 { 4224 char path[64]; 4225 char driver_buf[64]; 4226 char governor_buf[64]; 4227 FILE *input; 4228 int turbo; 4229 4230 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_driver", base_cpu); 4231 input = fopen(path, "r"); 4232 if (input == NULL) { 4233 fprintf(outf, "NSFOD %s\n", path); 4234 return; 4235 } 4236 if (!fgets(driver_buf, sizeof(driver_buf), input)) 4237 err(1, "%s: failed to read file", path); 4238 fclose(input); 4239 4240 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor", base_cpu); 4241 input = fopen(path, "r"); 4242 if (input == NULL) { 4243 fprintf(outf, "NSFOD %s\n", path); 4244 return; 4245 } 4246 if (!fgets(governor_buf, sizeof(governor_buf), input)) 4247 err(1, "%s: failed to read file", path); 4248 fclose(input); 4249 4250 fprintf(outf, "cpu%d: cpufreq driver: %s", base_cpu, driver_buf); 4251 fprintf(outf, "cpu%d: cpufreq governor: %s", base_cpu, governor_buf); 4252 4253 sprintf(path, "/sys/devices/system/cpu/cpufreq/boost"); 4254 input = fopen(path, "r"); 4255 if (input != NULL) { 4256 if (fscanf(input, "%d", &turbo) != 1) 4257 err(1, "%s: failed to parse number from file", path); 4258 fprintf(outf, "cpufreq boost: %d\n", turbo); 4259 fclose(input); 4260 } 4261 4262 sprintf(path, "/sys/devices/system/cpu/intel_pstate/no_turbo"); 4263 input = fopen(path, "r"); 4264 if (input != NULL) { 4265 if (fscanf(input, "%d", &turbo) != 1) 4266 err(1, "%s: failed to parse number from file", path); 4267 fprintf(outf, "cpufreq intel_pstate no_turbo: %d\n", turbo); 4268 fclose(input); 4269 } 4270 } 4271 4272 /* 4273 * print_epb() 4274 * Decode the ENERGY_PERF_BIAS MSR 4275 */ 4276 int print_epb(struct thread_data *t, struct core_data *c, struct pkg_data *p) 4277 { 4278 char *epb_string; 4279 int cpu, epb; 4280 4281 UNUSED(c); 4282 UNUSED(p); 4283 4284 if (!has_epb) 4285 return 0; 4286 4287 cpu = t->cpu_id; 4288 4289 /* EPB is per-package */ 4290 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 4291 return 0; 4292 4293 if (cpu_migrate(cpu)) { 4294 fprintf(outf, "print_epb: Could not migrate to CPU %d\n", cpu); 4295 return -1; 4296 } 4297 4298 epb = get_epb(cpu); 4299 if (epb < 0) 4300 return 0; 4301 4302 switch (epb) { 4303 case ENERGY_PERF_BIAS_PERFORMANCE: 4304 epb_string = "performance"; 4305 break; 4306 case ENERGY_PERF_BIAS_NORMAL: 4307 epb_string = "balanced"; 4308 break; 4309 case ENERGY_PERF_BIAS_POWERSAVE: 4310 epb_string = "powersave"; 4311 break; 4312 default: 4313 epb_string = "custom"; 4314 break; 4315 } 4316 fprintf(outf, "cpu%d: EPB: %d (%s)\n", cpu, epb, epb_string); 4317 4318 return 0; 4319 } 4320 4321 /* 4322 * print_hwp() 4323 * Decode the MSR_HWP_CAPABILITIES 4324 */ 4325 int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p) 4326 { 4327 unsigned long long msr; 4328 int cpu; 4329 4330 UNUSED(c); 4331 UNUSED(p); 4332 4333 if (!has_hwp) 4334 return 0; 4335 4336 cpu = t->cpu_id; 4337 4338 /* MSR_HWP_CAPABILITIES is per-package */ 4339 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 4340 return 0; 4341 4342 if (cpu_migrate(cpu)) { 4343 fprintf(outf, "print_hwp: Could not migrate to CPU %d\n", cpu); 4344 return -1; 4345 } 4346 4347 if (get_msr(cpu, MSR_PM_ENABLE, &msr)) 4348 return 0; 4349 4350 fprintf(outf, "cpu%d: MSR_PM_ENABLE: 0x%08llx (%sHWP)\n", cpu, msr, (msr & (1 << 0)) ? "" : "No-"); 4351 4352 /* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */ 4353 if ((msr & (1 << 0)) == 0) 4354 return 0; 4355 4356 if (get_msr(cpu, MSR_HWP_CAPABILITIES, &msr)) 4357 return 0; 4358 4359 fprintf(outf, "cpu%d: MSR_HWP_CAPABILITIES: 0x%08llx " 4360 "(high %d guar %d eff %d low %d)\n", 4361 cpu, msr, 4362 (unsigned int)HWP_HIGHEST_PERF(msr), 4363 (unsigned int)HWP_GUARANTEED_PERF(msr), 4364 (unsigned int)HWP_MOSTEFFICIENT_PERF(msr), (unsigned int)HWP_LOWEST_PERF(msr)); 4365 4366 if (get_msr(cpu, MSR_HWP_REQUEST, &msr)) 4367 return 0; 4368 4369 fprintf(outf, "cpu%d: MSR_HWP_REQUEST: 0x%08llx " 4370 "(min %d max %d des %d epp 0x%x window 0x%x pkg 0x%x)\n", 4371 cpu, msr, 4372 (unsigned int)(((msr) >> 0) & 0xff), 4373 (unsigned int)(((msr) >> 8) & 0xff), 4374 (unsigned int)(((msr) >> 16) & 0xff), 4375 (unsigned int)(((msr) >> 24) & 0xff), 4376 (unsigned int)(((msr) >> 32) & 0xff3), (unsigned int)(((msr) >> 42) & 0x1)); 4377 4378 if (has_hwp_pkg) { 4379 if (get_msr(cpu, MSR_HWP_REQUEST_PKG, &msr)) 4380 return 0; 4381 4382 fprintf(outf, "cpu%d: MSR_HWP_REQUEST_PKG: 0x%08llx " 4383 "(min %d max %d des %d epp 0x%x window 0x%x)\n", 4384 cpu, msr, 4385 (unsigned int)(((msr) >> 0) & 0xff), 4386 (unsigned int)(((msr) >> 8) & 0xff), 4387 (unsigned int)(((msr) >> 16) & 0xff), 4388 (unsigned int)(((msr) >> 24) & 0xff), (unsigned int)(((msr) >> 32) & 0xff3)); 4389 } 4390 if (has_hwp_notify) { 4391 if (get_msr(cpu, MSR_HWP_INTERRUPT, &msr)) 4392 return 0; 4393 4394 fprintf(outf, "cpu%d: MSR_HWP_INTERRUPT: 0x%08llx " 4395 "(%s_Guaranteed_Perf_Change, %s_Excursion_Min)\n", 4396 cpu, msr, ((msr) & 0x1) ? "EN" : "Dis", ((msr) & 0x2) ? "EN" : "Dis"); 4397 } 4398 if (get_msr(cpu, MSR_HWP_STATUS, &msr)) 4399 return 0; 4400 4401 fprintf(outf, "cpu%d: MSR_HWP_STATUS: 0x%08llx " 4402 "(%sGuaranteed_Perf_Change, %sExcursion_Min)\n", 4403 cpu, msr, ((msr) & 0x1) ? "" : "No-", ((msr) & 0x2) ? "" : "No-"); 4404 4405 return 0; 4406 } 4407 4408 /* 4409 * print_perf_limit() 4410 */ 4411 int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data *p) 4412 { 4413 unsigned long long msr; 4414 int cpu; 4415 4416 UNUSED(c); 4417 UNUSED(p); 4418 4419 cpu = t->cpu_id; 4420 4421 /* per-package */ 4422 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 4423 return 0; 4424 4425 if (cpu_migrate(cpu)) { 4426 fprintf(outf, "print_perf_limit: Could not migrate to CPU %d\n", cpu); 4427 return -1; 4428 } 4429 4430 if (do_core_perf_limit_reasons) { 4431 get_msr(cpu, MSR_CORE_PERF_LIMIT_REASONS, &msr); 4432 fprintf(outf, "cpu%d: MSR_CORE_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); 4433 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)", 4434 (msr & 1 << 15) ? "bit15, " : "", 4435 (msr & 1 << 14) ? "bit14, " : "", 4436 (msr & 1 << 13) ? "Transitions, " : "", 4437 (msr & 1 << 12) ? "MultiCoreTurbo, " : "", 4438 (msr & 1 << 11) ? "PkgPwrL2, " : "", 4439 (msr & 1 << 10) ? "PkgPwrL1, " : "", 4440 (msr & 1 << 9) ? "CorePwr, " : "", 4441 (msr & 1 << 8) ? "Amps, " : "", 4442 (msr & 1 << 6) ? "VR-Therm, " : "", 4443 (msr & 1 << 5) ? "Auto-HWP, " : "", 4444 (msr & 1 << 4) ? "Graphics, " : "", 4445 (msr & 1 << 2) ? "bit2, " : "", 4446 (msr & 1 << 1) ? "ThermStatus, " : "", (msr & 1 << 0) ? "PROCHOT, " : ""); 4447 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n", 4448 (msr & 1 << 31) ? "bit31, " : "", 4449 (msr & 1 << 30) ? "bit30, " : "", 4450 (msr & 1 << 29) ? "Transitions, " : "", 4451 (msr & 1 << 28) ? "MultiCoreTurbo, " : "", 4452 (msr & 1 << 27) ? "PkgPwrL2, " : "", 4453 (msr & 1 << 26) ? "PkgPwrL1, " : "", 4454 (msr & 1 << 25) ? "CorePwr, " : "", 4455 (msr & 1 << 24) ? "Amps, " : "", 4456 (msr & 1 << 22) ? "VR-Therm, " : "", 4457 (msr & 1 << 21) ? "Auto-HWP, " : "", 4458 (msr & 1 << 20) ? "Graphics, " : "", 4459 (msr & 1 << 18) ? "bit18, " : "", 4460 (msr & 1 << 17) ? "ThermStatus, " : "", (msr & 1 << 16) ? "PROCHOT, " : ""); 4461 4462 } 4463 if (do_gfx_perf_limit_reasons) { 4464 get_msr(cpu, MSR_GFX_PERF_LIMIT_REASONS, &msr); 4465 fprintf(outf, "cpu%d: MSR_GFX_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); 4466 fprintf(outf, " (Active: %s%s%s%s%s%s%s%s)", 4467 (msr & 1 << 0) ? "PROCHOT, " : "", 4468 (msr & 1 << 1) ? "ThermStatus, " : "", 4469 (msr & 1 << 4) ? "Graphics, " : "", 4470 (msr & 1 << 6) ? "VR-Therm, " : "", 4471 (msr & 1 << 8) ? "Amps, " : "", 4472 (msr & 1 << 9) ? "GFXPwr, " : "", 4473 (msr & 1 << 10) ? "PkgPwrL1, " : "", (msr & 1 << 11) ? "PkgPwrL2, " : ""); 4474 fprintf(outf, " (Logged: %s%s%s%s%s%s%s%s)\n", 4475 (msr & 1 << 16) ? "PROCHOT, " : "", 4476 (msr & 1 << 17) ? "ThermStatus, " : "", 4477 (msr & 1 << 20) ? "Graphics, " : "", 4478 (msr & 1 << 22) ? "VR-Therm, " : "", 4479 (msr & 1 << 24) ? "Amps, " : "", 4480 (msr & 1 << 25) ? "GFXPwr, " : "", 4481 (msr & 1 << 26) ? "PkgPwrL1, " : "", (msr & 1 << 27) ? "PkgPwrL2, " : ""); 4482 } 4483 if (do_ring_perf_limit_reasons) { 4484 get_msr(cpu, MSR_RING_PERF_LIMIT_REASONS, &msr); 4485 fprintf(outf, "cpu%d: MSR_RING_PERF_LIMIT_REASONS, 0x%08llx", cpu, msr); 4486 fprintf(outf, " (Active: %s%s%s%s%s%s)", 4487 (msr & 1 << 0) ? "PROCHOT, " : "", 4488 (msr & 1 << 1) ? "ThermStatus, " : "", 4489 (msr & 1 << 6) ? "VR-Therm, " : "", 4490 (msr & 1 << 8) ? "Amps, " : "", 4491 (msr & 1 << 10) ? "PkgPwrL1, " : "", (msr & 1 << 11) ? "PkgPwrL2, " : ""); 4492 fprintf(outf, " (Logged: %s%s%s%s%s%s)\n", 4493 (msr & 1 << 16) ? "PROCHOT, " : "", 4494 (msr & 1 << 17) ? "ThermStatus, " : "", 4495 (msr & 1 << 22) ? "VR-Therm, " : "", 4496 (msr & 1 << 24) ? "Amps, " : "", 4497 (msr & 1 << 26) ? "PkgPwrL1, " : "", (msr & 1 << 27) ? "PkgPwrL2, " : ""); 4498 } 4499 return 0; 4500 } 4501 4502 #define RAPL_POWER_GRANULARITY 0x7FFF /* 15 bit power granularity */ 4503 #define RAPL_TIME_GRANULARITY 0x3F /* 6 bit time granularity */ 4504 4505 double get_tdp_intel(unsigned int model) 4506 { 4507 unsigned long long msr; 4508 4509 if (do_rapl & RAPL_PKG_POWER_INFO) 4510 if (!get_msr(base_cpu, MSR_PKG_POWER_INFO, &msr)) 4511 return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units; 4512 4513 switch (model) { 4514 case INTEL_FAM6_ATOM_SILVERMONT: 4515 case INTEL_FAM6_ATOM_SILVERMONT_D: 4516 return 30.0; 4517 default: 4518 return 135.0; 4519 } 4520 } 4521 4522 double get_tdp_amd(unsigned int family) 4523 { 4524 UNUSED(family); 4525 4526 /* This is the max stock TDP of HEDT/Server Fam17h+ chips */ 4527 return 280.0; 4528 } 4529 4530 /* 4531 * rapl_dram_energy_units_probe() 4532 * Energy units are either hard-coded, or come from RAPL Energy Unit MSR. 4533 */ 4534 static double rapl_dram_energy_units_probe(int model, double rapl_energy_units) 4535 { 4536 /* only called for genuine_intel, family 6 */ 4537 4538 switch (model) { 4539 case INTEL_FAM6_HASWELL_X: /* HSX */ 4540 case INTEL_FAM6_BROADWELL_X: /* BDX */ 4541 case INTEL_FAM6_SKYLAKE_X: /* SKX */ 4542 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */ 4543 case INTEL_FAM6_ICELAKE_X: /* ICX */ 4544 return (rapl_dram_energy_units = 15.3 / 1000000); 4545 default: 4546 return (rapl_energy_units); 4547 } 4548 } 4549 4550 void rapl_probe_intel(unsigned int family, unsigned int model) 4551 { 4552 unsigned long long msr; 4553 unsigned int time_unit; 4554 double tdp; 4555 4556 if (family != 6) 4557 return; 4558 4559 switch (model) { 4560 case INTEL_FAM6_SANDYBRIDGE: 4561 case INTEL_FAM6_IVYBRIDGE: 4562 case INTEL_FAM6_HASWELL: /* HSW */ 4563 case INTEL_FAM6_HASWELL_L: /* HSW */ 4564 case INTEL_FAM6_HASWELL_G: /* HSW */ 4565 case INTEL_FAM6_BROADWELL: /* BDW */ 4566 case INTEL_FAM6_BROADWELL_G: /* BDW */ 4567 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO; 4568 if (rapl_joules) { 4569 BIC_PRESENT(BIC_Pkg_J); 4570 BIC_PRESENT(BIC_Cor_J); 4571 BIC_PRESENT(BIC_GFX_J); 4572 } else { 4573 BIC_PRESENT(BIC_PkgWatt); 4574 BIC_PRESENT(BIC_CorWatt); 4575 BIC_PRESENT(BIC_GFXWatt); 4576 } 4577 break; 4578 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */ 4579 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 4580 do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO; 4581 if (rapl_joules) 4582 BIC_PRESENT(BIC_Pkg_J); 4583 else 4584 BIC_PRESENT(BIC_PkgWatt); 4585 break; 4586 case INTEL_FAM6_ATOM_TREMONT: /* EHL */ 4587 do_rapl = 4588 RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS 4589 | RAPL_GFX | RAPL_PKG_POWER_INFO; 4590 if (rapl_joules) { 4591 BIC_PRESENT(BIC_Pkg_J); 4592 BIC_PRESENT(BIC_Cor_J); 4593 BIC_PRESENT(BIC_RAM_J); 4594 BIC_PRESENT(BIC_GFX_J); 4595 } else { 4596 BIC_PRESENT(BIC_PkgWatt); 4597 BIC_PRESENT(BIC_CorWatt); 4598 BIC_PRESENT(BIC_RAMWatt); 4599 BIC_PRESENT(BIC_GFXWatt); 4600 } 4601 break; 4602 case INTEL_FAM6_ATOM_TREMONT_D: /* JVL */ 4603 do_rapl = RAPL_PKG | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO; 4604 BIC_PRESENT(BIC_PKG__); 4605 if (rapl_joules) 4606 BIC_PRESENT(BIC_Pkg_J); 4607 else 4608 BIC_PRESENT(BIC_PkgWatt); 4609 break; 4610 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 4611 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 4612 do_rapl = 4613 RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS 4614 | RAPL_GFX | RAPL_PKG_POWER_INFO; 4615 BIC_PRESENT(BIC_PKG__); 4616 BIC_PRESENT(BIC_RAM__); 4617 if (rapl_joules) { 4618 BIC_PRESENT(BIC_Pkg_J); 4619 BIC_PRESENT(BIC_Cor_J); 4620 BIC_PRESENT(BIC_RAM_J); 4621 BIC_PRESENT(BIC_GFX_J); 4622 } else { 4623 BIC_PRESENT(BIC_PkgWatt); 4624 BIC_PRESENT(BIC_CorWatt); 4625 BIC_PRESENT(BIC_RAMWatt); 4626 BIC_PRESENT(BIC_GFXWatt); 4627 } 4628 break; 4629 case INTEL_FAM6_HASWELL_X: /* HSX */ 4630 case INTEL_FAM6_BROADWELL_X: /* BDX */ 4631 case INTEL_FAM6_SKYLAKE_X: /* SKX */ 4632 case INTEL_FAM6_ICELAKE_X: /* ICX */ 4633 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */ 4634 do_rapl = 4635 RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | 4636 RAPL_PKG_POWER_INFO; 4637 BIC_PRESENT(BIC_PKG__); 4638 BIC_PRESENT(BIC_RAM__); 4639 if (rapl_joules) { 4640 BIC_PRESENT(BIC_Pkg_J); 4641 BIC_PRESENT(BIC_RAM_J); 4642 } else { 4643 BIC_PRESENT(BIC_PkgWatt); 4644 BIC_PRESENT(BIC_RAMWatt); 4645 } 4646 break; 4647 case INTEL_FAM6_SANDYBRIDGE_X: 4648 case INTEL_FAM6_IVYBRIDGE_X: 4649 do_rapl = 4650 RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_PKG_PERF_STATUS | 4651 RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO; 4652 BIC_PRESENT(BIC_PKG__); 4653 BIC_PRESENT(BIC_RAM__); 4654 if (rapl_joules) { 4655 BIC_PRESENT(BIC_Pkg_J); 4656 BIC_PRESENT(BIC_Cor_J); 4657 BIC_PRESENT(BIC_RAM_J); 4658 } else { 4659 BIC_PRESENT(BIC_PkgWatt); 4660 BIC_PRESENT(BIC_CorWatt); 4661 BIC_PRESENT(BIC_RAMWatt); 4662 } 4663 break; 4664 case INTEL_FAM6_ATOM_SILVERMONT: /* BYT */ 4665 case INTEL_FAM6_ATOM_SILVERMONT_D: /* AVN */ 4666 do_rapl = RAPL_PKG | RAPL_CORES; 4667 if (rapl_joules) { 4668 BIC_PRESENT(BIC_Pkg_J); 4669 BIC_PRESENT(BIC_Cor_J); 4670 } else { 4671 BIC_PRESENT(BIC_PkgWatt); 4672 BIC_PRESENT(BIC_CorWatt); 4673 } 4674 break; 4675 case INTEL_FAM6_ATOM_GOLDMONT_D: /* DNV */ 4676 do_rapl = 4677 RAPL_PKG | RAPL_DRAM | RAPL_DRAM_POWER_INFO | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | 4678 RAPL_PKG_POWER_INFO | RAPL_CORES_ENERGY_STATUS; 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 default: 4692 return; 4693 } 4694 4695 /* units on package 0, verify later other packages match */ 4696 if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr)) 4697 return; 4698 4699 rapl_power_units = 1.0 / (1 << (msr & 0xF)); 4700 if (model == INTEL_FAM6_ATOM_SILVERMONT) 4701 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000; 4702 else 4703 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F)); 4704 4705 rapl_dram_energy_units = rapl_dram_energy_units_probe(model, rapl_energy_units); 4706 4707 time_unit = msr >> 16 & 0xF; 4708 if (time_unit == 0) 4709 time_unit = 0xA; 4710 4711 rapl_time_units = 1.0 / (1 << (time_unit)); 4712 4713 tdp = get_tdp_intel(model); 4714 4715 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp; 4716 if (!quiet) 4717 fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp); 4718 } 4719 4720 void rapl_probe_amd(unsigned int family, unsigned int model) 4721 { 4722 unsigned long long msr; 4723 unsigned int eax, ebx, ecx, edx; 4724 unsigned int has_rapl = 0; 4725 double tdp; 4726 4727 UNUSED(model); 4728 4729 if (max_extended_level >= 0x80000007) { 4730 __cpuid(0x80000007, eax, ebx, ecx, edx); 4731 /* RAPL (Fam 17h+) */ 4732 has_rapl = edx & (1 << 14); 4733 } 4734 4735 if (!has_rapl || family < 0x17) 4736 return; 4737 4738 do_rapl = RAPL_AMD_F17H | RAPL_PER_CORE_ENERGY; 4739 if (rapl_joules) { 4740 BIC_PRESENT(BIC_Pkg_J); 4741 BIC_PRESENT(BIC_Cor_J); 4742 } else { 4743 BIC_PRESENT(BIC_PkgWatt); 4744 BIC_PRESENT(BIC_CorWatt); 4745 } 4746 4747 if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr)) 4748 return; 4749 4750 rapl_time_units = ldexp(1.0, -(msr >> 16 & 0xf)); 4751 rapl_energy_units = ldexp(1.0, -(msr >> 8 & 0x1f)); 4752 rapl_power_units = ldexp(1.0, -(msr & 0xf)); 4753 4754 tdp = get_tdp_amd(family); 4755 4756 rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp; 4757 if (!quiet) 4758 fprintf(outf, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp); 4759 } 4760 4761 /* 4762 * rapl_probe() 4763 * 4764 * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units 4765 */ 4766 void rapl_probe(unsigned int family, unsigned int model) 4767 { 4768 if (genuine_intel) 4769 rapl_probe_intel(family, model); 4770 if (authentic_amd || hygon_genuine) 4771 rapl_probe_amd(family, model); 4772 } 4773 4774 void perf_limit_reasons_probe(unsigned int family, unsigned int model) 4775 { 4776 if (!genuine_intel) 4777 return; 4778 4779 if (family != 6) 4780 return; 4781 4782 switch (model) { 4783 case INTEL_FAM6_HASWELL: /* HSW */ 4784 case INTEL_FAM6_HASWELL_L: /* HSW */ 4785 case INTEL_FAM6_HASWELL_G: /* HSW */ 4786 do_gfx_perf_limit_reasons = 1; 4787 /* FALLTHRU */ 4788 case INTEL_FAM6_HASWELL_X: /* HSX */ 4789 do_core_perf_limit_reasons = 1; 4790 do_ring_perf_limit_reasons = 1; 4791 default: 4792 return; 4793 } 4794 } 4795 4796 void automatic_cstate_conversion_probe(unsigned int family, unsigned int model) 4797 { 4798 if (is_skx(family, model) || is_bdx(family, model) || is_icx(family, model)) 4799 has_automatic_cstate_conversion = 1; 4800 } 4801 4802 void prewake_cstate_probe(unsigned int family, unsigned int model) 4803 { 4804 if (is_icx(family, model)) 4805 dis_cstate_prewake = 1; 4806 } 4807 4808 int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p) 4809 { 4810 unsigned long long msr; 4811 unsigned int dts, dts2; 4812 int cpu; 4813 4814 UNUSED(c); 4815 UNUSED(p); 4816 4817 if (!(do_dts || do_ptm)) 4818 return 0; 4819 4820 cpu = t->cpu_id; 4821 4822 /* DTS is per-core, no need to print for each thread */ 4823 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) 4824 return 0; 4825 4826 if (cpu_migrate(cpu)) { 4827 fprintf(outf, "print_thermal: Could not migrate to CPU %d\n", cpu); 4828 return -1; 4829 } 4830 4831 if (do_ptm && (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) { 4832 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) 4833 return 0; 4834 4835 dts = (msr >> 16) & 0x7F; 4836 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_STATUS: 0x%08llx (%d C)\n", cpu, msr, tj_max - dts); 4837 4838 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &msr)) 4839 return 0; 4840 4841 dts = (msr >> 16) & 0x7F; 4842 dts2 = (msr >> 8) & 0x7F; 4843 fprintf(outf, "cpu%d: MSR_IA32_PACKAGE_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", 4844 cpu, msr, tj_max - dts, tj_max - dts2); 4845 } 4846 4847 if (do_dts && debug) { 4848 unsigned int resolution; 4849 4850 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr)) 4851 return 0; 4852 4853 dts = (msr >> 16) & 0x7F; 4854 resolution = (msr >> 27) & 0xF; 4855 fprintf(outf, "cpu%d: MSR_IA32_THERM_STATUS: 0x%08llx (%d C +/- %d)\n", 4856 cpu, msr, tj_max - dts, resolution); 4857 4858 if (get_msr(cpu, MSR_IA32_THERM_INTERRUPT, &msr)) 4859 return 0; 4860 4861 dts = (msr >> 16) & 0x7F; 4862 dts2 = (msr >> 8) & 0x7F; 4863 fprintf(outf, "cpu%d: MSR_IA32_THERM_INTERRUPT: 0x%08llx (%d C, %d C)\n", 4864 cpu, msr, tj_max - dts, tj_max - dts2); 4865 } 4866 4867 return 0; 4868 } 4869 4870 void print_power_limit_msr(int cpu, unsigned long long msr, char *label) 4871 { 4872 fprintf(outf, "cpu%d: %s: %sabled (%0.3f Watts, %f sec, clamp %sabled)\n", 4873 cpu, label, 4874 ((msr >> 15) & 1) ? "EN" : "DIS", 4875 ((msr >> 0) & 0x7FFF) * rapl_power_units, 4876 (1.0 + (((msr >> 22) & 0x3) / 4.0)) * (1 << ((msr >> 17) & 0x1F)) * rapl_time_units, 4877 (((msr >> 16) & 1) ? "EN" : "DIS")); 4878 4879 return; 4880 } 4881 4882 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p) 4883 { 4884 unsigned long long msr; 4885 const char *msr_name; 4886 int cpu; 4887 4888 UNUSED(c); 4889 UNUSED(p); 4890 4891 if (!do_rapl) 4892 return 0; 4893 4894 /* RAPL counters are per package, so print only for 1st thread/package */ 4895 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 4896 return 0; 4897 4898 cpu = t->cpu_id; 4899 if (cpu_migrate(cpu)) { 4900 fprintf(outf, "print_rapl: Could not migrate to CPU %d\n", cpu); 4901 return -1; 4902 } 4903 4904 if (do_rapl & RAPL_AMD_F17H) { 4905 msr_name = "MSR_RAPL_PWR_UNIT"; 4906 if (get_msr(cpu, MSR_RAPL_PWR_UNIT, &msr)) 4907 return -1; 4908 } else { 4909 msr_name = "MSR_RAPL_POWER_UNIT"; 4910 if (get_msr(cpu, MSR_RAPL_POWER_UNIT, &msr)) 4911 return -1; 4912 } 4913 4914 fprintf(outf, "cpu%d: %s: 0x%08llx (%f Watts, %f Joules, %f sec.)\n", cpu, msr_name, msr, 4915 rapl_power_units, rapl_energy_units, rapl_time_units); 4916 4917 if (do_rapl & RAPL_PKG_POWER_INFO) { 4918 4919 if (get_msr(cpu, MSR_PKG_POWER_INFO, &msr)) 4920 return -5; 4921 4922 fprintf(outf, "cpu%d: MSR_PKG_POWER_INFO: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", 4923 cpu, msr, 4924 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4925 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4926 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4927 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); 4928 4929 } 4930 if (do_rapl & RAPL_PKG) { 4931 4932 if (get_msr(cpu, MSR_PKG_POWER_LIMIT, &msr)) 4933 return -9; 4934 4935 fprintf(outf, "cpu%d: MSR_PKG_POWER_LIMIT: 0x%08llx (%slocked)\n", 4936 cpu, msr, (msr >> 63) & 1 ? "" : "UN"); 4937 4938 print_power_limit_msr(cpu, msr, "PKG Limit #1"); 4939 fprintf(outf, "cpu%d: PKG Limit #2: %sabled (%0.3f Watts, %f* sec, clamp %sabled)\n", 4940 cpu, 4941 ((msr >> 47) & 1) ? "EN" : "DIS", 4942 ((msr >> 32) & 0x7FFF) * rapl_power_units, 4943 (1.0 + (((msr >> 54) & 0x3) / 4.0)) * (1 << ((msr >> 49) & 0x1F)) * rapl_time_units, 4944 ((msr >> 48) & 1) ? "EN" : "DIS"); 4945 4946 if (get_msr(cpu, MSR_VR_CURRENT_CONFIG, &msr)) 4947 return -9; 4948 4949 fprintf(outf, "cpu%d: MSR_VR_CURRENT_CONFIG: 0x%08llx\n", cpu, msr); 4950 fprintf(outf, "cpu%d: PKG Limit #4: %f Watts (%slocked)\n", 4951 cpu, ((msr >> 0) & 0x1FFF) * rapl_power_units, (msr >> 31) & 1 ? "" : "UN"); 4952 } 4953 4954 if (do_rapl & RAPL_DRAM_POWER_INFO) { 4955 if (get_msr(cpu, MSR_DRAM_POWER_INFO, &msr)) 4956 return -6; 4957 4958 fprintf(outf, "cpu%d: MSR_DRAM_POWER_INFO,: 0x%08llx (%.0f W TDP, RAPL %.0f - %.0f W, %f sec.)\n", 4959 cpu, msr, 4960 ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4961 ((msr >> 16) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4962 ((msr >> 32) & RAPL_POWER_GRANULARITY) * rapl_power_units, 4963 ((msr >> 48) & RAPL_TIME_GRANULARITY) * rapl_time_units); 4964 } 4965 if (do_rapl & RAPL_DRAM) { 4966 if (get_msr(cpu, MSR_DRAM_POWER_LIMIT, &msr)) 4967 return -9; 4968 fprintf(outf, "cpu%d: MSR_DRAM_POWER_LIMIT: 0x%08llx (%slocked)\n", 4969 cpu, msr, (msr >> 31) & 1 ? "" : "UN"); 4970 4971 print_power_limit_msr(cpu, msr, "DRAM Limit"); 4972 } 4973 if (do_rapl & RAPL_CORE_POLICY) { 4974 if (get_msr(cpu, MSR_PP0_POLICY, &msr)) 4975 return -7; 4976 4977 fprintf(outf, "cpu%d: MSR_PP0_POLICY: %lld\n", cpu, msr & 0xF); 4978 } 4979 if (do_rapl & RAPL_CORES_POWER_LIMIT) { 4980 if (get_msr(cpu, MSR_PP0_POWER_LIMIT, &msr)) 4981 return -9; 4982 fprintf(outf, "cpu%d: MSR_PP0_POWER_LIMIT: 0x%08llx (%slocked)\n", 4983 cpu, msr, (msr >> 31) & 1 ? "" : "UN"); 4984 print_power_limit_msr(cpu, msr, "Cores Limit"); 4985 } 4986 if (do_rapl & RAPL_GFX) { 4987 if (get_msr(cpu, MSR_PP1_POLICY, &msr)) 4988 return -8; 4989 4990 fprintf(outf, "cpu%d: MSR_PP1_POLICY: %lld\n", cpu, msr & 0xF); 4991 4992 if (get_msr(cpu, MSR_PP1_POWER_LIMIT, &msr)) 4993 return -9; 4994 fprintf(outf, "cpu%d: MSR_PP1_POWER_LIMIT: 0x%08llx (%slocked)\n", 4995 cpu, msr, (msr >> 31) & 1 ? "" : "UN"); 4996 print_power_limit_msr(cpu, msr, "GFX Limit"); 4997 } 4998 return 0; 4999 } 5000 5001 /* 5002 * SNB adds support for additional MSRs: 5003 * 5004 * MSR_PKG_C7_RESIDENCY 0x000003fa 5005 * MSR_CORE_C7_RESIDENCY 0x000003fe 5006 * MSR_PKG_C2_RESIDENCY 0x0000060d 5007 */ 5008 5009 int has_snb_msrs(unsigned int family, unsigned int model) 5010 { 5011 if (!genuine_intel) 5012 return 0; 5013 5014 if (family != 6) 5015 return 0; 5016 5017 switch (model) { 5018 case INTEL_FAM6_SANDYBRIDGE: 5019 case INTEL_FAM6_SANDYBRIDGE_X: 5020 case INTEL_FAM6_IVYBRIDGE: /* IVB */ 5021 case INTEL_FAM6_IVYBRIDGE_X: /* IVB Xeon */ 5022 case INTEL_FAM6_HASWELL: /* HSW */ 5023 case INTEL_FAM6_HASWELL_X: /* HSW */ 5024 case INTEL_FAM6_HASWELL_L: /* HSW */ 5025 case INTEL_FAM6_HASWELL_G: /* HSW */ 5026 case INTEL_FAM6_BROADWELL: /* BDW */ 5027 case INTEL_FAM6_BROADWELL_G: /* BDW */ 5028 case INTEL_FAM6_BROADWELL_X: /* BDX */ 5029 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 5030 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 5031 case INTEL_FAM6_SKYLAKE_X: /* SKX */ 5032 case INTEL_FAM6_ICELAKE_X: /* ICX */ 5033 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */ 5034 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 5035 case INTEL_FAM6_ATOM_GOLDMONT_D: /* DNV */ 5036 case INTEL_FAM6_ATOM_TREMONT: /* EHL */ 5037 case INTEL_FAM6_ATOM_TREMONT_D: /* JVL */ 5038 return 1; 5039 } 5040 return 0; 5041 } 5042 5043 /* 5044 * HSW ULT added support for C8/C9/C10 MSRs: 5045 * 5046 * MSR_PKG_C8_RESIDENCY 0x00000630 5047 * MSR_PKG_C9_RESIDENCY 0x00000631 5048 * MSR_PKG_C10_RESIDENCY 0x00000632 5049 * 5050 * MSR_PKGC8_IRTL 0x00000633 5051 * MSR_PKGC9_IRTL 0x00000634 5052 * MSR_PKGC10_IRTL 0x00000635 5053 * 5054 */ 5055 int has_c8910_msrs(unsigned int family, unsigned int model) 5056 { 5057 if (!genuine_intel) 5058 return 0; 5059 5060 if (family != 6) 5061 return 0; 5062 5063 switch (model) { 5064 case INTEL_FAM6_HASWELL_L: /* HSW */ 5065 case INTEL_FAM6_BROADWELL: /* BDW */ 5066 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 5067 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 5068 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */ 5069 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 5070 case INTEL_FAM6_ATOM_TREMONT: /* EHL */ 5071 return 1; 5072 } 5073 return 0; 5074 } 5075 5076 /* 5077 * SKL adds support for additional MSRS: 5078 * 5079 * MSR_PKG_WEIGHTED_CORE_C0_RES 0x00000658 5080 * MSR_PKG_ANY_CORE_C0_RES 0x00000659 5081 * MSR_PKG_ANY_GFXE_C0_RES 0x0000065A 5082 * MSR_PKG_BOTH_CORE_GFXE_C0_RES 0x0000065B 5083 */ 5084 int has_skl_msrs(unsigned int family, unsigned int model) 5085 { 5086 if (!genuine_intel) 5087 return 0; 5088 5089 if (family != 6) 5090 return 0; 5091 5092 switch (model) { 5093 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 5094 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 5095 return 1; 5096 } 5097 return 0; 5098 } 5099 5100 int is_slm(unsigned int family, unsigned int model) 5101 { 5102 if (!genuine_intel) 5103 return 0; 5104 5105 if (family != 6) 5106 return 0; 5107 5108 switch (model) { 5109 case INTEL_FAM6_ATOM_SILVERMONT: /* BYT */ 5110 case INTEL_FAM6_ATOM_SILVERMONT_D: /* AVN */ 5111 return 1; 5112 } 5113 return 0; 5114 } 5115 5116 int is_knl(unsigned int family, unsigned int model) 5117 { 5118 if (!genuine_intel) 5119 return 0; 5120 5121 if (family != 6) 5122 return 0; 5123 5124 switch (model) { 5125 case INTEL_FAM6_XEON_PHI_KNL: /* KNL */ 5126 return 1; 5127 } 5128 return 0; 5129 } 5130 5131 int is_cnl(unsigned int family, unsigned int model) 5132 { 5133 if (!genuine_intel) 5134 return 0; 5135 5136 if (family != 6) 5137 return 0; 5138 5139 switch (model) { 5140 case INTEL_FAM6_CANNONLAKE_L: /* CNL */ 5141 return 1; 5142 } 5143 5144 return 0; 5145 } 5146 5147 unsigned int get_aperf_mperf_multiplier(unsigned int family, unsigned int model) 5148 { 5149 if (is_knl(family, model)) 5150 return 1024; 5151 return 1; 5152 } 5153 5154 #define SLM_BCLK_FREQS 5 5155 double slm_freq_table[SLM_BCLK_FREQS] = { 83.3, 100.0, 133.3, 116.7, 80.0 }; 5156 5157 double slm_bclk(void) 5158 { 5159 unsigned long long msr = 3; 5160 unsigned int i; 5161 double freq; 5162 5163 if (get_msr(base_cpu, MSR_FSB_FREQ, &msr)) 5164 fprintf(outf, "SLM BCLK: unknown\n"); 5165 5166 i = msr & 0xf; 5167 if (i >= SLM_BCLK_FREQS) { 5168 fprintf(outf, "SLM BCLK[%d] invalid\n", i); 5169 i = 3; 5170 } 5171 freq = slm_freq_table[i]; 5172 5173 if (!quiet) 5174 fprintf(outf, "SLM BCLK: %.1f Mhz\n", freq); 5175 5176 return freq; 5177 } 5178 5179 double discover_bclk(unsigned int family, unsigned int model) 5180 { 5181 if (has_snb_msrs(family, model) || is_knl(family, model)) 5182 return 100.00; 5183 else if (is_slm(family, model)) 5184 return slm_bclk(); 5185 else 5186 return 133.33; 5187 } 5188 5189 int get_cpu_type(struct thread_data *t, struct core_data *c, struct pkg_data *p) 5190 { 5191 unsigned int eax, ebx, ecx, edx; 5192 5193 UNUSED(c); 5194 UNUSED(p); 5195 5196 if (!genuine_intel) 5197 return 0; 5198 5199 if (cpu_migrate(t->cpu_id)) { 5200 fprintf(outf, "Could not migrate to CPU %d\n", t->cpu_id); 5201 return -1; 5202 } 5203 5204 if (max_level < 0x1a) 5205 return 0; 5206 5207 __cpuid(0x1a, eax, ebx, ecx, edx); 5208 eax = (eax >> 24) & 0xFF; 5209 if (eax == 0x20) 5210 t->is_atom = true; 5211 return 0; 5212 } 5213 5214 /* 5215 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where 5216 * the Thermal Control Circuit (TCC) activates. 5217 * This is usually equal to tjMax. 5218 * 5219 * Older processors do not have this MSR, so there we guess, 5220 * but also allow cmdline over-ride with -T. 5221 * 5222 * Several MSR temperature values are in units of degrees-C 5223 * below this value, including the Digital Thermal Sensor (DTS), 5224 * Package Thermal Management Sensor (PTM), and thermal event thresholds. 5225 */ 5226 int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p) 5227 { 5228 unsigned long long msr; 5229 unsigned int tcc_default, tcc_offset; 5230 int cpu; 5231 5232 UNUSED(c); 5233 UNUSED(p); 5234 5235 /* tj_max is used only for dts or ptm */ 5236 if (!(do_dts || do_ptm)) 5237 return 0; 5238 5239 /* this is a per-package concept */ 5240 if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) 5241 return 0; 5242 5243 cpu = t->cpu_id; 5244 if (cpu_migrate(cpu)) { 5245 fprintf(outf, "Could not migrate to CPU %d\n", cpu); 5246 return -1; 5247 } 5248 5249 if (tj_max_override != 0) { 5250 tj_max = tj_max_override; 5251 fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n", cpu, tj_max); 5252 return 0; 5253 } 5254 5255 /* Temperature Target MSR is Nehalem and newer only */ 5256 if (!do_nhm_platform_info) 5257 goto guess; 5258 5259 if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr)) 5260 goto guess; 5261 5262 tcc_default = (msr >> 16) & 0xFF; 5263 5264 if (!quiet) { 5265 switch (tcc_offset_bits) { 5266 case 4: 5267 tcc_offset = (msr >> 24) & 0xF; 5268 fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C) (%d default - %d offset)\n", 5269 cpu, msr, tcc_default - tcc_offset, tcc_default, tcc_offset); 5270 break; 5271 case 6: 5272 tcc_offset = (msr >> 24) & 0x3F; 5273 fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C) (%d default - %d offset)\n", 5274 cpu, msr, tcc_default - tcc_offset, tcc_default, tcc_offset); 5275 break; 5276 default: 5277 fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n", cpu, msr, tcc_default); 5278 break; 5279 } 5280 } 5281 5282 if (!tcc_default) 5283 goto guess; 5284 5285 tj_max = tcc_default; 5286 5287 return 0; 5288 5289 guess: 5290 tj_max = TJMAX_DEFAULT; 5291 fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n", cpu, tj_max); 5292 5293 return 0; 5294 } 5295 5296 void decode_feature_control_msr(void) 5297 { 5298 unsigned long long msr; 5299 5300 if (!get_msr(base_cpu, MSR_IA32_FEAT_CTL, &msr)) 5301 fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n", 5302 base_cpu, msr, msr & FEAT_CTL_LOCKED ? "" : "UN-", msr & (1 << 18) ? "SGX" : ""); 5303 } 5304 5305 void decode_misc_enable_msr(void) 5306 { 5307 unsigned long long msr; 5308 5309 if (!genuine_intel) 5310 return; 5311 5312 if (!get_msr(base_cpu, MSR_IA32_MISC_ENABLE, &msr)) 5313 fprintf(outf, "cpu%d: MSR_IA32_MISC_ENABLE: 0x%08llx (%sTCC %sEIST %sMWAIT %sPREFETCH %sTURBO)\n", 5314 base_cpu, msr, 5315 msr & MSR_IA32_MISC_ENABLE_TM1 ? "" : "No-", 5316 msr & MSR_IA32_MISC_ENABLE_ENHANCED_SPEEDSTEP ? "" : "No-", 5317 msr & MSR_IA32_MISC_ENABLE_MWAIT ? "" : "No-", 5318 msr & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE ? "No-" : "", 5319 msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ? "No-" : ""); 5320 } 5321 5322 void decode_misc_feature_control(void) 5323 { 5324 unsigned long long msr; 5325 5326 if (!has_misc_feature_control) 5327 return; 5328 5329 if (!get_msr(base_cpu, MSR_MISC_FEATURE_CONTROL, &msr)) 5330 fprintf(outf, 5331 "cpu%d: MSR_MISC_FEATURE_CONTROL: 0x%08llx (%sL2-Prefetch %sL2-Prefetch-pair %sL1-Prefetch %sL1-IP-Prefetch)\n", 5332 base_cpu, msr, msr & (0 << 0) ? "No-" : "", msr & (1 << 0) ? "No-" : "", 5333 msr & (2 << 0) ? "No-" : "", msr & (3 << 0) ? "No-" : ""); 5334 } 5335 5336 /* 5337 * Decode MSR_MISC_PWR_MGMT 5338 * 5339 * Decode the bits according to the Nehalem documentation 5340 * bit[0] seems to continue to have same meaning going forward 5341 * bit[1] less so... 5342 */ 5343 void decode_misc_pwr_mgmt_msr(void) 5344 { 5345 unsigned long long msr; 5346 5347 if (!do_nhm_platform_info) 5348 return; 5349 5350 if (no_MSR_MISC_PWR_MGMT) 5351 return; 5352 5353 if (!get_msr(base_cpu, MSR_MISC_PWR_MGMT, &msr)) 5354 fprintf(outf, "cpu%d: MSR_MISC_PWR_MGMT: 0x%08llx (%sable-EIST_Coordination %sable-EPB %sable-OOB)\n", 5355 base_cpu, msr, 5356 msr & (1 << 0) ? "DIS" : "EN", msr & (1 << 1) ? "EN" : "DIS", msr & (1 << 8) ? "EN" : "DIS"); 5357 } 5358 5359 /* 5360 * Decode MSR_CC6_DEMOTION_POLICY_CONFIG, MSR_MC6_DEMOTION_POLICY_CONFIG 5361 * 5362 * This MSRs are present on Silvermont processors, 5363 * Intel Atom processor E3000 series (Baytrail), and friends. 5364 */ 5365 void decode_c6_demotion_policy_msr(void) 5366 { 5367 unsigned long long msr; 5368 5369 if (!get_msr(base_cpu, MSR_CC6_DEMOTION_POLICY_CONFIG, &msr)) 5370 fprintf(outf, "cpu%d: MSR_CC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-CC6-Demotion)\n", 5371 base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS"); 5372 5373 if (!get_msr(base_cpu, MSR_MC6_DEMOTION_POLICY_CONFIG, &msr)) 5374 fprintf(outf, "cpu%d: MSR_MC6_DEMOTION_POLICY_CONFIG: 0x%08llx (%sable-MC6-Demotion)\n", 5375 base_cpu, msr, msr & (1 << 0) ? "EN" : "DIS"); 5376 } 5377 5378 /* 5379 * When models are the same, for the purpose of turbostat, reuse 5380 */ 5381 unsigned int intel_model_duplicates(unsigned int model) 5382 { 5383 5384 switch (model) { 5385 case INTEL_FAM6_NEHALEM_EP: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */ 5386 case INTEL_FAM6_NEHALEM: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */ 5387 case 0x1F: /* Core i7 and i5 Processor - Nehalem */ 5388 case INTEL_FAM6_WESTMERE: /* Westmere Client - Clarkdale, Arrandale */ 5389 case INTEL_FAM6_WESTMERE_EP: /* Westmere EP - Gulftown */ 5390 return INTEL_FAM6_NEHALEM; 5391 5392 case INTEL_FAM6_NEHALEM_EX: /* Nehalem-EX Xeon - Beckton */ 5393 case INTEL_FAM6_WESTMERE_EX: /* Westmere-EX Xeon - Eagleton */ 5394 return INTEL_FAM6_NEHALEM_EX; 5395 5396 case INTEL_FAM6_XEON_PHI_KNM: 5397 return INTEL_FAM6_XEON_PHI_KNL; 5398 5399 case INTEL_FAM6_BROADWELL_X: 5400 case INTEL_FAM6_BROADWELL_D: /* BDX-DE */ 5401 return INTEL_FAM6_BROADWELL_X; 5402 5403 case INTEL_FAM6_SKYLAKE_L: 5404 case INTEL_FAM6_SKYLAKE: 5405 case INTEL_FAM6_KABYLAKE_L: 5406 case INTEL_FAM6_KABYLAKE: 5407 case INTEL_FAM6_COMETLAKE_L: 5408 case INTEL_FAM6_COMETLAKE: 5409 return INTEL_FAM6_SKYLAKE_L; 5410 5411 case INTEL_FAM6_ICELAKE_L: 5412 case INTEL_FAM6_ICELAKE_NNPI: 5413 case INTEL_FAM6_TIGERLAKE_L: 5414 case INTEL_FAM6_TIGERLAKE: 5415 case INTEL_FAM6_ROCKETLAKE: 5416 case INTEL_FAM6_LAKEFIELD: 5417 case INTEL_FAM6_ALDERLAKE: 5418 case INTEL_FAM6_ALDERLAKE_L: 5419 case INTEL_FAM6_ALDERLAKE_N: 5420 case INTEL_FAM6_RAPTORLAKE: 5421 case INTEL_FAM6_RAPTORLAKE_P: 5422 return INTEL_FAM6_CANNONLAKE_L; 5423 5424 case INTEL_FAM6_ATOM_TREMONT_L: 5425 return INTEL_FAM6_ATOM_TREMONT; 5426 5427 case INTEL_FAM6_ICELAKE_D: 5428 case INTEL_FAM6_SAPPHIRERAPIDS_X: 5429 return INTEL_FAM6_ICELAKE_X; 5430 } 5431 return model; 5432 } 5433 5434 void print_dev_latency(void) 5435 { 5436 char *path = "/dev/cpu_dma_latency"; 5437 int fd; 5438 int value; 5439 int retval; 5440 5441 fd = open(path, O_RDONLY); 5442 if (fd < 0) { 5443 warn("fopen %s\n", path); 5444 return; 5445 } 5446 5447 retval = read(fd, (void *)&value, sizeof(int)); 5448 if (retval != sizeof(int)) { 5449 warn("read %s\n", path); 5450 close(fd); 5451 return; 5452 } 5453 fprintf(outf, "/dev/cpu_dma_latency: %d usec (%s)\n", value, value == 2000000000 ? "default" : "constrained"); 5454 5455 close(fd); 5456 } 5457 5458 /* 5459 * Linux-perf manages the the HW instructions-retired counter 5460 * by enabling when requested, and hiding rollover 5461 */ 5462 void linux_perf_init(void) 5463 { 5464 if (!BIC_IS_ENABLED(BIC_IPC)) 5465 return; 5466 5467 if (access("/proc/sys/kernel/perf_event_paranoid", F_OK)) 5468 return; 5469 5470 fd_instr_count_percpu = calloc(topo.max_cpu_num + 1, sizeof(int)); 5471 if (fd_instr_count_percpu == NULL) 5472 err(-1, "calloc fd_instr_count_percpu"); 5473 5474 BIC_PRESENT(BIC_IPC); 5475 } 5476 5477 void process_cpuid() 5478 { 5479 unsigned int eax, ebx, ecx, edx; 5480 unsigned int fms, family, model, stepping, ecx_flags, edx_flags; 5481 unsigned int has_turbo; 5482 unsigned long long ucode_patch = 0; 5483 5484 eax = ebx = ecx = edx = 0; 5485 5486 __cpuid(0, max_level, ebx, ecx, edx); 5487 5488 if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69) 5489 genuine_intel = 1; 5490 else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65) 5491 authentic_amd = 1; 5492 else if (ebx == 0x6f677948 && ecx == 0x656e6975 && edx == 0x6e65476e) 5493 hygon_genuine = 1; 5494 5495 if (!quiet) 5496 fprintf(outf, "CPUID(0): %.4s%.4s%.4s 0x%x CPUID levels\n", 5497 (char *)&ebx, (char *)&edx, (char *)&ecx, max_level); 5498 5499 __cpuid(1, fms, ebx, ecx, edx); 5500 family = (fms >> 8) & 0xf; 5501 model = (fms >> 4) & 0xf; 5502 stepping = fms & 0xf; 5503 if (family == 0xf) 5504 family += (fms >> 20) & 0xff; 5505 if (family >= 6) 5506 model += ((fms >> 16) & 0xf) << 4; 5507 ecx_flags = ecx; 5508 edx_flags = edx; 5509 5510 if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch)) 5511 warnx("get_msr(UCODE)\n"); 5512 5513 /* 5514 * check max extended function levels of CPUID. 5515 * This is needed to check for invariant TSC. 5516 * This check is valid for both Intel and AMD. 5517 */ 5518 ebx = ecx = edx = 0; 5519 __cpuid(0x80000000, max_extended_level, ebx, ecx, edx); 5520 5521 if (!quiet) { 5522 fprintf(outf, "CPUID(1): family:model:stepping 0x%x:%x:%x (%d:%d:%d) microcode 0x%x\n", 5523 family, model, stepping, family, model, stepping, 5524 (unsigned int)((ucode_patch >> 32) & 0xFFFFFFFF)); 5525 fprintf(outf, "CPUID(0x80000000): max_extended_levels: 0x%x\n", max_extended_level); 5526 fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n", 5527 ecx_flags & (1 << 0) ? "SSE3" : "-", 5528 ecx_flags & (1 << 3) ? "MONITOR" : "-", 5529 ecx_flags & (1 << 6) ? "SMX" : "-", 5530 ecx_flags & (1 << 7) ? "EIST" : "-", 5531 ecx_flags & (1 << 8) ? "TM2" : "-", 5532 edx_flags & (1 << 4) ? "TSC" : "-", 5533 edx_flags & (1 << 5) ? "MSR" : "-", 5534 edx_flags & (1 << 22) ? "ACPI-TM" : "-", 5535 edx_flags & (1 << 28) ? "HT" : "-", edx_flags & (1 << 29) ? "TM" : "-"); 5536 } 5537 if (genuine_intel) { 5538 model_orig = model; 5539 model = intel_model_duplicates(model); 5540 } 5541 5542 if (!(edx_flags & (1 << 5))) 5543 errx(1, "CPUID: no MSR"); 5544 5545 if (max_extended_level >= 0x80000007) { 5546 5547 /* 5548 * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8 5549 * this check is valid for both Intel and AMD 5550 */ 5551 __cpuid(0x80000007, eax, ebx, ecx, edx); 5552 has_invariant_tsc = edx & (1 << 8); 5553 } 5554 5555 /* 5556 * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0 5557 * this check is valid for both Intel and AMD 5558 */ 5559 5560 __cpuid(0x6, eax, ebx, ecx, edx); 5561 has_aperf = ecx & (1 << 0); 5562 if (has_aperf) { 5563 BIC_PRESENT(BIC_Avg_MHz); 5564 BIC_PRESENT(BIC_Busy); 5565 BIC_PRESENT(BIC_Bzy_MHz); 5566 } 5567 do_dts = eax & (1 << 0); 5568 if (do_dts) 5569 BIC_PRESENT(BIC_CoreTmp); 5570 has_turbo = eax & (1 << 1); 5571 do_ptm = eax & (1 << 6); 5572 if (do_ptm) 5573 BIC_PRESENT(BIC_PkgTmp); 5574 has_hwp = eax & (1 << 7); 5575 has_hwp_notify = eax & (1 << 8); 5576 has_hwp_activity_window = eax & (1 << 9); 5577 has_hwp_epp = eax & (1 << 10); 5578 has_hwp_pkg = eax & (1 << 11); 5579 has_epb = ecx & (1 << 3); 5580 5581 if (!quiet) 5582 fprintf(outf, "CPUID(6): %sAPERF, %sTURBO, %sDTS, %sPTM, %sHWP, " 5583 "%sHWPnotify, %sHWPwindow, %sHWPepp, %sHWPpkg, %sEPB\n", 5584 has_aperf ? "" : "No-", 5585 has_turbo ? "" : "No-", 5586 do_dts ? "" : "No-", 5587 do_ptm ? "" : "No-", 5588 has_hwp ? "" : "No-", 5589 has_hwp_notify ? "" : "No-", 5590 has_hwp_activity_window ? "" : "No-", 5591 has_hwp_epp ? "" : "No-", has_hwp_pkg ? "" : "No-", has_epb ? "" : "No-"); 5592 5593 if (!quiet) 5594 decode_misc_enable_msr(); 5595 5596 if (max_level >= 0x7 && !quiet) { 5597 int has_sgx; 5598 5599 ecx = 0; 5600 5601 __cpuid_count(0x7, 0, eax, ebx, ecx, edx); 5602 5603 has_sgx = ebx & (1 << 2); 5604 5605 is_hybrid = edx & (1 << 15); 5606 5607 fprintf(outf, "CPUID(7): %sSGX %sHybrid\n", has_sgx ? "" : "No-", is_hybrid ? "" : "No-"); 5608 5609 if (has_sgx) 5610 decode_feature_control_msr(); 5611 } 5612 5613 if (max_level >= 0x15) { 5614 unsigned int eax_crystal; 5615 unsigned int ebx_tsc; 5616 5617 /* 5618 * CPUID 15H TSC/Crystal ratio, possibly Crystal Hz 5619 */ 5620 eax_crystal = ebx_tsc = crystal_hz = edx = 0; 5621 __cpuid(0x15, eax_crystal, ebx_tsc, crystal_hz, edx); 5622 5623 if (ebx_tsc != 0) { 5624 5625 if (!quiet && (ebx != 0)) 5626 fprintf(outf, "CPUID(0x15): eax_crystal: %d ebx_tsc: %d ecx_crystal_hz: %d\n", 5627 eax_crystal, ebx_tsc, crystal_hz); 5628 5629 if (crystal_hz == 0) 5630 switch (model) { 5631 case INTEL_FAM6_SKYLAKE_L: /* SKL */ 5632 crystal_hz = 24000000; /* 24.0 MHz */ 5633 break; 5634 case INTEL_FAM6_ATOM_GOLDMONT_D: /* DNV */ 5635 crystal_hz = 25000000; /* 25.0 MHz */ 5636 break; 5637 case INTEL_FAM6_ATOM_GOLDMONT: /* BXT */ 5638 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 5639 crystal_hz = 19200000; /* 19.2 MHz */ 5640 break; 5641 default: 5642 crystal_hz = 0; 5643 } 5644 5645 if (crystal_hz) { 5646 tsc_hz = (unsigned long long)crystal_hz *ebx_tsc / eax_crystal; 5647 if (!quiet) 5648 fprintf(outf, "TSC: %lld MHz (%d Hz * %d / %d / 1000000)\n", 5649 tsc_hz / 1000000, crystal_hz, ebx_tsc, eax_crystal); 5650 } 5651 } 5652 } 5653 if (max_level >= 0x16) { 5654 unsigned int base_mhz, max_mhz, bus_mhz, edx; 5655 5656 /* 5657 * CPUID 16H Base MHz, Max MHz, Bus MHz 5658 */ 5659 base_mhz = max_mhz = bus_mhz = edx = 0; 5660 5661 __cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx); 5662 if (!quiet) 5663 fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n", 5664 base_mhz, max_mhz, bus_mhz); 5665 } 5666 5667 if (has_aperf) 5668 aperf_mperf_multiplier = get_aperf_mperf_multiplier(family, model); 5669 5670 BIC_PRESENT(BIC_IRQ); 5671 BIC_PRESENT(BIC_TSC_MHz); 5672 5673 if (probe_nhm_msrs(family, model)) { 5674 do_nhm_platform_info = 1; 5675 BIC_PRESENT(BIC_CPU_c1); 5676 BIC_PRESENT(BIC_CPU_c3); 5677 BIC_PRESENT(BIC_CPU_c6); 5678 BIC_PRESENT(BIC_SMI); 5679 } 5680 do_snb_cstates = has_snb_msrs(family, model); 5681 5682 if (do_snb_cstates) 5683 BIC_PRESENT(BIC_CPU_c7); 5684 5685 do_irtl_snb = has_snb_msrs(family, model); 5686 if (do_snb_cstates && (pkg_cstate_limit >= PCL__2)) 5687 BIC_PRESENT(BIC_Pkgpc2); 5688 if (pkg_cstate_limit >= PCL__3) 5689 BIC_PRESENT(BIC_Pkgpc3); 5690 if (pkg_cstate_limit >= PCL__6) 5691 BIC_PRESENT(BIC_Pkgpc6); 5692 if (do_snb_cstates && (pkg_cstate_limit >= PCL__7)) 5693 BIC_PRESENT(BIC_Pkgpc7); 5694 if (has_slv_msrs(family, model)) { 5695 BIC_NOT_PRESENT(BIC_Pkgpc2); 5696 BIC_NOT_PRESENT(BIC_Pkgpc3); 5697 BIC_PRESENT(BIC_Pkgpc6); 5698 BIC_NOT_PRESENT(BIC_Pkgpc7); 5699 BIC_PRESENT(BIC_Mod_c6); 5700 use_c1_residency_msr = 1; 5701 } 5702 if (is_jvl(family, model)) { 5703 BIC_NOT_PRESENT(BIC_CPU_c3); 5704 BIC_NOT_PRESENT(BIC_CPU_c7); 5705 BIC_NOT_PRESENT(BIC_Pkgpc2); 5706 BIC_NOT_PRESENT(BIC_Pkgpc3); 5707 BIC_NOT_PRESENT(BIC_Pkgpc6); 5708 BIC_NOT_PRESENT(BIC_Pkgpc7); 5709 } 5710 if (is_dnv(family, model)) { 5711 BIC_PRESENT(BIC_CPU_c1); 5712 BIC_NOT_PRESENT(BIC_CPU_c3); 5713 BIC_NOT_PRESENT(BIC_Pkgpc3); 5714 BIC_NOT_PRESENT(BIC_CPU_c7); 5715 BIC_NOT_PRESENT(BIC_Pkgpc7); 5716 use_c1_residency_msr = 1; 5717 } 5718 if (is_skx(family, model) || is_icx(family, model)) { 5719 BIC_NOT_PRESENT(BIC_CPU_c3); 5720 BIC_NOT_PRESENT(BIC_Pkgpc3); 5721 BIC_NOT_PRESENT(BIC_CPU_c7); 5722 BIC_NOT_PRESENT(BIC_Pkgpc7); 5723 } 5724 if (is_bdx(family, model)) { 5725 BIC_NOT_PRESENT(BIC_CPU_c7); 5726 BIC_NOT_PRESENT(BIC_Pkgpc7); 5727 } 5728 if (has_c8910_msrs(family, model)) { 5729 if (pkg_cstate_limit >= PCL__8) 5730 BIC_PRESENT(BIC_Pkgpc8); 5731 if (pkg_cstate_limit >= PCL__9) 5732 BIC_PRESENT(BIC_Pkgpc9); 5733 if (pkg_cstate_limit >= PCL_10) 5734 BIC_PRESENT(BIC_Pkgpc10); 5735 } 5736 do_irtl_hsw = has_c8910_msrs(family, model); 5737 if (has_skl_msrs(family, model)) { 5738 BIC_PRESENT(BIC_Totl_c0); 5739 BIC_PRESENT(BIC_Any_c0); 5740 BIC_PRESENT(BIC_GFX_c0); 5741 BIC_PRESENT(BIC_CPUGFX); 5742 } 5743 do_slm_cstates = is_slm(family, model); 5744 do_knl_cstates = is_knl(family, model); 5745 5746 if (do_slm_cstates || do_knl_cstates || is_cnl(family, model) || is_ehl(family, model)) 5747 BIC_NOT_PRESENT(BIC_CPU_c3); 5748 5749 if (!quiet) 5750 decode_misc_pwr_mgmt_msr(); 5751 5752 if (!quiet && has_slv_msrs(family, model)) 5753 decode_c6_demotion_policy_msr(); 5754 5755 rapl_probe(family, model); 5756 perf_limit_reasons_probe(family, model); 5757 automatic_cstate_conversion_probe(family, model); 5758 5759 check_tcc_offset(model_orig); 5760 5761 if (!quiet) 5762 dump_cstate_pstate_config_info(family, model); 5763 intel_uncore_frequency_probe(); 5764 5765 if (!quiet) 5766 print_dev_latency(); 5767 if (!quiet) 5768 dump_sysfs_cstate_config(); 5769 if (!quiet) 5770 dump_sysfs_pstate_config(); 5771 5772 if (has_skl_msrs(family, model) || is_ehl(family, model)) 5773 calculate_tsc_tweak(); 5774 5775 if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK)) 5776 BIC_PRESENT(BIC_GFX_rc6); 5777 5778 if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK)) 5779 BIC_PRESENT(BIC_GFXMHz); 5780 5781 if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", R_OK)) 5782 BIC_PRESENT(BIC_GFXACTMHz); 5783 5784 if (!access("/sys/devices/system/cpu/cpuidle/low_power_idle_cpu_residency_us", R_OK)) 5785 BIC_PRESENT(BIC_CPU_LPI); 5786 else 5787 BIC_NOT_PRESENT(BIC_CPU_LPI); 5788 5789 if (!access("/sys/devices/system/cpu/cpu0/thermal_throttle/core_throttle_count", R_OK)) 5790 BIC_PRESENT(BIC_CORE_THROT_CNT); 5791 else 5792 BIC_NOT_PRESENT(BIC_CORE_THROT_CNT); 5793 5794 if (!access(sys_lpi_file_sysfs, R_OK)) { 5795 sys_lpi_file = sys_lpi_file_sysfs; 5796 BIC_PRESENT(BIC_SYS_LPI); 5797 } else if (!access(sys_lpi_file_debugfs, R_OK)) { 5798 sys_lpi_file = sys_lpi_file_debugfs; 5799 BIC_PRESENT(BIC_SYS_LPI); 5800 } else { 5801 sys_lpi_file_sysfs = NULL; 5802 BIC_NOT_PRESENT(BIC_SYS_LPI); 5803 } 5804 5805 if (!quiet) 5806 decode_misc_feature_control(); 5807 5808 return; 5809 } 5810 5811 /* 5812 * in /dev/cpu/ return success for names that are numbers 5813 * ie. filter out ".", "..", "microcode". 5814 */ 5815 int dir_filter(const struct dirent *dirp) 5816 { 5817 if (isdigit(dirp->d_name[0])) 5818 return 1; 5819 else 5820 return 0; 5821 } 5822 5823 void topology_probe() 5824 { 5825 int i; 5826 int max_core_id = 0; 5827 int max_package_id = 0; 5828 int max_die_id = 0; 5829 int max_siblings = 0; 5830 5831 /* Initialize num_cpus, max_cpu_num */ 5832 set_max_cpu_num(); 5833 topo.num_cpus = 0; 5834 for_all_proc_cpus(count_cpus); 5835 if (!summary_only && topo.num_cpus > 1) 5836 BIC_PRESENT(BIC_CPU); 5837 5838 if (debug > 1) 5839 fprintf(outf, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num); 5840 5841 cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology)); 5842 if (cpus == NULL) 5843 err(1, "calloc cpus"); 5844 5845 /* 5846 * Allocate and initialize cpu_present_set 5847 */ 5848 cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1)); 5849 if (cpu_present_set == NULL) 5850 err(3, "CPU_ALLOC"); 5851 cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); 5852 CPU_ZERO_S(cpu_present_setsize, cpu_present_set); 5853 for_all_proc_cpus(mark_cpu_present); 5854 5855 /* 5856 * Validate that all cpus in cpu_subset are also in cpu_present_set 5857 */ 5858 for (i = 0; i < CPU_SUBSET_MAXCPUS; ++i) { 5859 if (CPU_ISSET_S(i, cpu_subset_size, cpu_subset)) 5860 if (!CPU_ISSET_S(i, cpu_present_setsize, cpu_present_set)) 5861 err(1, "cpu%d not present", i); 5862 } 5863 5864 /* 5865 * Allocate and initialize cpu_affinity_set 5866 */ 5867 cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1)); 5868 if (cpu_affinity_set == NULL) 5869 err(3, "CPU_ALLOC"); 5870 cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); 5871 CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); 5872 5873 for_all_proc_cpus(init_thread_id); 5874 5875 /* 5876 * For online cpus 5877 * find max_core_id, max_package_id 5878 */ 5879 for (i = 0; i <= topo.max_cpu_num; ++i) { 5880 int siblings; 5881 5882 if (cpu_is_not_present(i)) { 5883 if (debug > 1) 5884 fprintf(outf, "cpu%d NOT PRESENT\n", i); 5885 continue; 5886 } 5887 5888 cpus[i].logical_cpu_id = i; 5889 5890 /* get package information */ 5891 cpus[i].physical_package_id = get_physical_package_id(i); 5892 if (cpus[i].physical_package_id > max_package_id) 5893 max_package_id = cpus[i].physical_package_id; 5894 5895 /* get die information */ 5896 cpus[i].die_id = get_die_id(i); 5897 if (cpus[i].die_id > max_die_id) 5898 max_die_id = cpus[i].die_id; 5899 5900 /* get numa node information */ 5901 cpus[i].physical_node_id = get_physical_node_id(&cpus[i]); 5902 if (cpus[i].physical_node_id > topo.max_node_num) 5903 topo.max_node_num = cpus[i].physical_node_id; 5904 5905 /* get core information */ 5906 cpus[i].physical_core_id = get_core_id(i); 5907 if (cpus[i].physical_core_id > max_core_id) 5908 max_core_id = cpus[i].physical_core_id; 5909 5910 /* get thread information */ 5911 siblings = get_thread_siblings(&cpus[i]); 5912 if (siblings > max_siblings) 5913 max_siblings = siblings; 5914 if (cpus[i].thread_id == 0) 5915 topo.num_cores++; 5916 } 5917 5918 topo.cores_per_node = max_core_id + 1; 5919 if (debug > 1) 5920 fprintf(outf, "max_core_id %d, sizing for %d cores per package\n", max_core_id, topo.cores_per_node); 5921 if (!summary_only && topo.cores_per_node > 1) 5922 BIC_PRESENT(BIC_Core); 5923 5924 topo.num_die = max_die_id + 1; 5925 if (debug > 1) 5926 fprintf(outf, "max_die_id %d, sizing for %d die\n", max_die_id, topo.num_die); 5927 if (!summary_only && topo.num_die > 1) 5928 BIC_PRESENT(BIC_Die); 5929 5930 topo.num_packages = max_package_id + 1; 5931 if (debug > 1) 5932 fprintf(outf, "max_package_id %d, sizing for %d packages\n", max_package_id, topo.num_packages); 5933 if (!summary_only && topo.num_packages > 1) 5934 BIC_PRESENT(BIC_Package); 5935 5936 set_node_data(); 5937 if (debug > 1) 5938 fprintf(outf, "nodes_per_pkg %d\n", topo.nodes_per_pkg); 5939 if (!summary_only && topo.nodes_per_pkg > 1) 5940 BIC_PRESENT(BIC_Node); 5941 5942 topo.threads_per_core = max_siblings; 5943 if (debug > 1) 5944 fprintf(outf, "max_siblings %d\n", max_siblings); 5945 5946 if (debug < 1) 5947 return; 5948 5949 for (i = 0; i <= topo.max_cpu_num; ++i) { 5950 if (cpu_is_not_present(i)) 5951 continue; 5952 fprintf(outf, 5953 "cpu %d pkg %d die %d node %d lnode %d core %d thread %d\n", 5954 i, cpus[i].physical_package_id, cpus[i].die_id, 5955 cpus[i].physical_node_id, cpus[i].logical_node_id, cpus[i].physical_core_id, cpus[i].thread_id); 5956 } 5957 5958 } 5959 5960 void allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p) 5961 { 5962 int i; 5963 int num_cores = topo.cores_per_node * topo.nodes_per_pkg * topo.num_packages; 5964 int num_threads = topo.threads_per_core * num_cores; 5965 5966 *t = calloc(num_threads, sizeof(struct thread_data)); 5967 if (*t == NULL) 5968 goto error; 5969 5970 for (i = 0; i < num_threads; i++) 5971 (*t)[i].cpu_id = -1; 5972 5973 *c = calloc(num_cores, sizeof(struct core_data)); 5974 if (*c == NULL) 5975 goto error; 5976 5977 for (i = 0; i < num_cores; i++) 5978 (*c)[i].core_id = -1; 5979 5980 *p = calloc(topo.num_packages, sizeof(struct pkg_data)); 5981 if (*p == NULL) 5982 goto error; 5983 5984 for (i = 0; i < topo.num_packages; i++) 5985 (*p)[i].package_id = i; 5986 5987 return; 5988 error: 5989 err(1, "calloc counters"); 5990 } 5991 5992 /* 5993 * init_counter() 5994 * 5995 * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE 5996 */ 5997 void init_counter(struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base, int cpu_id) 5998 { 5999 int pkg_id = cpus[cpu_id].physical_package_id; 6000 int node_id = cpus[cpu_id].logical_node_id; 6001 int core_id = cpus[cpu_id].physical_core_id; 6002 int thread_id = cpus[cpu_id].thread_id; 6003 struct thread_data *t; 6004 struct core_data *c; 6005 struct pkg_data *p; 6006 6007 /* Workaround for systems where physical_node_id==-1 6008 * and logical_node_id==(-1 - topo.num_cpus) 6009 */ 6010 if (node_id < 0) 6011 node_id = 0; 6012 6013 t = GET_THREAD(thread_base, thread_id, core_id, node_id, pkg_id); 6014 c = GET_CORE(core_base, core_id, node_id, pkg_id); 6015 p = GET_PKG(pkg_base, pkg_id); 6016 6017 t->cpu_id = cpu_id; 6018 if (thread_id == 0) { 6019 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE; 6020 if (cpu_is_first_core_in_package(cpu_id)) 6021 t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE; 6022 } 6023 6024 c->core_id = core_id; 6025 p->package_id = pkg_id; 6026 } 6027 6028 int initialize_counters(int cpu_id) 6029 { 6030 init_counter(EVEN_COUNTERS, cpu_id); 6031 init_counter(ODD_COUNTERS, cpu_id); 6032 return 0; 6033 } 6034 6035 void allocate_output_buffer() 6036 { 6037 output_buffer = calloc(1, (1 + topo.num_cpus) * 2048); 6038 outp = output_buffer; 6039 if (outp == NULL) 6040 err(-1, "calloc output buffer"); 6041 } 6042 6043 void allocate_fd_percpu(void) 6044 { 6045 fd_percpu = calloc(topo.max_cpu_num + 1, sizeof(int)); 6046 if (fd_percpu == NULL) 6047 err(-1, "calloc fd_percpu"); 6048 } 6049 6050 void allocate_irq_buffers(void) 6051 { 6052 irq_column_2_cpu = calloc(topo.num_cpus, sizeof(int)); 6053 if (irq_column_2_cpu == NULL) 6054 err(-1, "calloc %d", topo.num_cpus); 6055 6056 irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int)); 6057 if (irqs_per_cpu == NULL) 6058 err(-1, "calloc %d", topo.max_cpu_num + 1); 6059 } 6060 6061 void setup_all_buffers(void) 6062 { 6063 topology_probe(); 6064 allocate_irq_buffers(); 6065 allocate_fd_percpu(); 6066 allocate_counters(&thread_even, &core_even, &package_even); 6067 allocate_counters(&thread_odd, &core_odd, &package_odd); 6068 allocate_output_buffer(); 6069 for_all_proc_cpus(initialize_counters); 6070 } 6071 6072 void set_base_cpu(void) 6073 { 6074 base_cpu = sched_getcpu(); 6075 if (base_cpu < 0) 6076 err(-ENODEV, "No valid cpus found"); 6077 6078 if (debug > 1) 6079 fprintf(outf, "base_cpu = %d\n", base_cpu); 6080 } 6081 6082 void turbostat_init() 6083 { 6084 setup_all_buffers(); 6085 set_base_cpu(); 6086 check_dev_msr(); 6087 check_permissions(); 6088 process_cpuid(); 6089 linux_perf_init(); 6090 6091 if (!quiet) 6092 for_all_cpus(print_hwp, ODD_COUNTERS); 6093 6094 if (!quiet) 6095 for_all_cpus(print_epb, ODD_COUNTERS); 6096 6097 if (!quiet) 6098 for_all_cpus(print_perf_limit, ODD_COUNTERS); 6099 6100 if (!quiet) 6101 for_all_cpus(print_rapl, ODD_COUNTERS); 6102 6103 for_all_cpus(set_temperature_target, ODD_COUNTERS); 6104 6105 for_all_cpus(get_cpu_type, ODD_COUNTERS); 6106 for_all_cpus(get_cpu_type, EVEN_COUNTERS); 6107 6108 if (!quiet) 6109 for_all_cpus(print_thermal, ODD_COUNTERS); 6110 6111 if (!quiet && do_irtl_snb) 6112 print_irtl(); 6113 6114 if (DO_BIC(BIC_IPC)) 6115 (void)get_instr_count_fd(base_cpu); 6116 } 6117 6118 int fork_it(char **argv) 6119 { 6120 pid_t child_pid; 6121 int status; 6122 6123 snapshot_proc_sysfs_files(); 6124 status = for_all_cpus(get_counters, EVEN_COUNTERS); 6125 first_counter_read = 0; 6126 if (status) 6127 exit(status); 6128 /* clear affinity side-effect of get_counters() */ 6129 sched_setaffinity(0, cpu_present_setsize, cpu_present_set); 6130 gettimeofday(&tv_even, (struct timezone *)NULL); 6131 6132 child_pid = fork(); 6133 if (!child_pid) { 6134 /* child */ 6135 execvp(argv[0], argv); 6136 err(errno, "exec %s", argv[0]); 6137 } else { 6138 6139 /* parent */ 6140 if (child_pid == -1) 6141 err(1, "fork"); 6142 6143 signal(SIGINT, SIG_IGN); 6144 signal(SIGQUIT, SIG_IGN); 6145 if (waitpid(child_pid, &status, 0) == -1) 6146 err(status, "waitpid"); 6147 6148 if (WIFEXITED(status)) 6149 status = WEXITSTATUS(status); 6150 } 6151 /* 6152 * n.b. fork_it() does not check for errors from for_all_cpus() 6153 * because re-starting is problematic when forking 6154 */ 6155 snapshot_proc_sysfs_files(); 6156 for_all_cpus(get_counters, ODD_COUNTERS); 6157 gettimeofday(&tv_odd, (struct timezone *)NULL); 6158 timersub(&tv_odd, &tv_even, &tv_delta); 6159 if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) 6160 fprintf(outf, "%s: Counter reset detected\n", progname); 6161 else { 6162 compute_average(EVEN_COUNTERS); 6163 format_all_counters(EVEN_COUNTERS); 6164 } 6165 6166 fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec / 1000000.0); 6167 6168 flush_output_stderr(); 6169 6170 return status; 6171 } 6172 6173 int get_and_dump_counters(void) 6174 { 6175 int status; 6176 6177 snapshot_proc_sysfs_files(); 6178 status = for_all_cpus(get_counters, ODD_COUNTERS); 6179 if (status) 6180 return status; 6181 6182 status = for_all_cpus(dump_counters, ODD_COUNTERS); 6183 if (status) 6184 return status; 6185 6186 flush_output_stdout(); 6187 6188 return status; 6189 } 6190 6191 void print_version() 6192 { 6193 fprintf(outf, "turbostat version 2022.04.16 - Len Brown <lenb@kernel.org>\n"); 6194 } 6195 6196 #define COMMAND_LINE_SIZE 2048 6197 6198 void print_bootcmd(void) 6199 { 6200 char bootcmd[COMMAND_LINE_SIZE]; 6201 FILE *fp; 6202 int ret; 6203 6204 memset(bootcmd, 0, COMMAND_LINE_SIZE); 6205 fp = fopen("/proc/cmdline", "r"); 6206 if (!fp) 6207 return; 6208 6209 ret = fread(bootcmd, sizeof(char), COMMAND_LINE_SIZE - 1, fp); 6210 if (ret) { 6211 bootcmd[ret] = '\0'; 6212 /* the last character is already '\n' */ 6213 fprintf(outf, "Kernel command line: %s", bootcmd); 6214 } 6215 6216 fclose(fp); 6217 } 6218 6219 int add_counter(unsigned int msr_num, char *path, char *name, 6220 unsigned int width, enum counter_scope scope, 6221 enum counter_type type, enum counter_format format, int flags) 6222 { 6223 struct msr_counter *msrp; 6224 6225 msrp = calloc(1, sizeof(struct msr_counter)); 6226 if (msrp == NULL) { 6227 perror("calloc"); 6228 exit(1); 6229 } 6230 6231 msrp->msr_num = msr_num; 6232 strncpy(msrp->name, name, NAME_BYTES - 1); 6233 if (path) 6234 strncpy(msrp->path, path, PATH_BYTES - 1); 6235 msrp->width = width; 6236 msrp->type = type; 6237 msrp->format = format; 6238 msrp->flags = flags; 6239 6240 switch (scope) { 6241 6242 case SCOPE_CPU: 6243 msrp->next = sys.tp; 6244 sys.tp = msrp; 6245 sys.added_thread_counters++; 6246 if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) { 6247 fprintf(stderr, "exceeded max %d added thread counters\n", MAX_ADDED_COUNTERS); 6248 exit(-1); 6249 } 6250 break; 6251 6252 case SCOPE_CORE: 6253 msrp->next = sys.cp; 6254 sys.cp = msrp; 6255 sys.added_core_counters++; 6256 if (sys.added_core_counters > MAX_ADDED_COUNTERS) { 6257 fprintf(stderr, "exceeded max %d added core counters\n", MAX_ADDED_COUNTERS); 6258 exit(-1); 6259 } 6260 break; 6261 6262 case SCOPE_PACKAGE: 6263 msrp->next = sys.pp; 6264 sys.pp = msrp; 6265 sys.added_package_counters++; 6266 if (sys.added_package_counters > MAX_ADDED_COUNTERS) { 6267 fprintf(stderr, "exceeded max %d added package counters\n", MAX_ADDED_COUNTERS); 6268 exit(-1); 6269 } 6270 break; 6271 } 6272 6273 return 0; 6274 } 6275 6276 void parse_add_command(char *add_command) 6277 { 6278 int msr_num = 0; 6279 char *path = NULL; 6280 char name_buffer[NAME_BYTES] = ""; 6281 int width = 64; 6282 int fail = 0; 6283 enum counter_scope scope = SCOPE_CPU; 6284 enum counter_type type = COUNTER_CYCLES; 6285 enum counter_format format = FORMAT_DELTA; 6286 6287 while (add_command) { 6288 6289 if (sscanf(add_command, "msr0x%x", &msr_num) == 1) 6290 goto next; 6291 6292 if (sscanf(add_command, "msr%d", &msr_num) == 1) 6293 goto next; 6294 6295 if (*add_command == '/') { 6296 path = add_command; 6297 goto next; 6298 } 6299 6300 if (sscanf(add_command, "u%d", &width) == 1) { 6301 if ((width == 32) || (width == 64)) 6302 goto next; 6303 width = 64; 6304 } 6305 if (!strncmp(add_command, "cpu", strlen("cpu"))) { 6306 scope = SCOPE_CPU; 6307 goto next; 6308 } 6309 if (!strncmp(add_command, "core", strlen("core"))) { 6310 scope = SCOPE_CORE; 6311 goto next; 6312 } 6313 if (!strncmp(add_command, "package", strlen("package"))) { 6314 scope = SCOPE_PACKAGE; 6315 goto next; 6316 } 6317 if (!strncmp(add_command, "cycles", strlen("cycles"))) { 6318 type = COUNTER_CYCLES; 6319 goto next; 6320 } 6321 if (!strncmp(add_command, "seconds", strlen("seconds"))) { 6322 type = COUNTER_SECONDS; 6323 goto next; 6324 } 6325 if (!strncmp(add_command, "usec", strlen("usec"))) { 6326 type = COUNTER_USEC; 6327 goto next; 6328 } 6329 if (!strncmp(add_command, "raw", strlen("raw"))) { 6330 format = FORMAT_RAW; 6331 goto next; 6332 } 6333 if (!strncmp(add_command, "delta", strlen("delta"))) { 6334 format = FORMAT_DELTA; 6335 goto next; 6336 } 6337 if (!strncmp(add_command, "percent", strlen("percent"))) { 6338 format = FORMAT_PERCENT; 6339 goto next; 6340 } 6341 6342 if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) { /* 18 < NAME_BYTES */ 6343 char *eos; 6344 6345 eos = strchr(name_buffer, ','); 6346 if (eos) 6347 *eos = '\0'; 6348 goto next; 6349 } 6350 6351 next: 6352 add_command = strchr(add_command, ','); 6353 if (add_command) { 6354 *add_command = '\0'; 6355 add_command++; 6356 } 6357 6358 } 6359 if ((msr_num == 0) && (path == NULL)) { 6360 fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n"); 6361 fail++; 6362 } 6363 6364 /* generate default column header */ 6365 if (*name_buffer == '\0') { 6366 if (width == 32) 6367 sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); 6368 else 6369 sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); 6370 } 6371 6372 if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0)) 6373 fail++; 6374 6375 if (fail) { 6376 help(); 6377 exit(1); 6378 } 6379 } 6380 6381 int is_deferred_add(char *name) 6382 { 6383 int i; 6384 6385 for (i = 0; i < deferred_add_index; ++i) 6386 if (!strcmp(name, deferred_add_names[i])) 6387 return 1; 6388 return 0; 6389 } 6390 6391 int is_deferred_skip(char *name) 6392 { 6393 int i; 6394 6395 for (i = 0; i < deferred_skip_index; ++i) 6396 if (!strcmp(name, deferred_skip_names[i])) 6397 return 1; 6398 return 0; 6399 } 6400 6401 void probe_sysfs(void) 6402 { 6403 char path[64]; 6404 char name_buf[16]; 6405 FILE *input; 6406 int state; 6407 char *sp; 6408 6409 for (state = 10; state >= 0; --state) { 6410 6411 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state); 6412 input = fopen(path, "r"); 6413 if (input == NULL) 6414 continue; 6415 if (!fgets(name_buf, sizeof(name_buf), input)) 6416 err(1, "%s: failed to read file", path); 6417 6418 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */ 6419 sp = strchr(name_buf, '-'); 6420 if (!sp) 6421 sp = strchrnul(name_buf, '\n'); 6422 *sp = '%'; 6423 *(sp + 1) = '\0'; 6424 6425 remove_underbar(name_buf); 6426 6427 fclose(input); 6428 6429 sprintf(path, "cpuidle/state%d/time", state); 6430 6431 if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf)) 6432 continue; 6433 6434 if (is_deferred_skip(name_buf)) 6435 continue; 6436 6437 add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC, FORMAT_PERCENT, SYSFS_PERCPU); 6438 } 6439 6440 for (state = 10; state >= 0; --state) { 6441 6442 sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state); 6443 input = fopen(path, "r"); 6444 if (input == NULL) 6445 continue; 6446 if (!fgets(name_buf, sizeof(name_buf), input)) 6447 err(1, "%s: failed to read file", path); 6448 /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */ 6449 sp = strchr(name_buf, '-'); 6450 if (!sp) 6451 sp = strchrnul(name_buf, '\n'); 6452 *sp = '\0'; 6453 fclose(input); 6454 6455 remove_underbar(name_buf); 6456 6457 sprintf(path, "cpuidle/state%d/usage", state); 6458 6459 if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf)) 6460 continue; 6461 6462 if (is_deferred_skip(name_buf)) 6463 continue; 6464 6465 add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU); 6466 } 6467 6468 } 6469 6470 /* 6471 * parse cpuset with following syntax 6472 * 1,2,4..6,8-10 and set bits in cpu_subset 6473 */ 6474 void parse_cpu_command(char *optarg) 6475 { 6476 unsigned int start, end; 6477 char *next; 6478 6479 if (!strcmp(optarg, "core")) { 6480 if (cpu_subset) 6481 goto error; 6482 show_core_only++; 6483 return; 6484 } 6485 if (!strcmp(optarg, "package")) { 6486 if (cpu_subset) 6487 goto error; 6488 show_pkg_only++; 6489 return; 6490 } 6491 if (show_core_only || show_pkg_only) 6492 goto error; 6493 6494 cpu_subset = CPU_ALLOC(CPU_SUBSET_MAXCPUS); 6495 if (cpu_subset == NULL) 6496 err(3, "CPU_ALLOC"); 6497 cpu_subset_size = CPU_ALLOC_SIZE(CPU_SUBSET_MAXCPUS); 6498 6499 CPU_ZERO_S(cpu_subset_size, cpu_subset); 6500 6501 next = optarg; 6502 6503 while (next && *next) { 6504 6505 if (*next == '-') /* no negative cpu numbers */ 6506 goto error; 6507 6508 start = strtoul(next, &next, 10); 6509 6510 if (start >= CPU_SUBSET_MAXCPUS) 6511 goto error; 6512 CPU_SET_S(start, cpu_subset_size, cpu_subset); 6513 6514 if (*next == '\0') 6515 break; 6516 6517 if (*next == ',') { 6518 next += 1; 6519 continue; 6520 } 6521 6522 if (*next == '-') { 6523 next += 1; /* start range */ 6524 } else if (*next == '.') { 6525 next += 1; 6526 if (*next == '.') 6527 next += 1; /* start range */ 6528 else 6529 goto error; 6530 } 6531 6532 end = strtoul(next, &next, 10); 6533 if (end <= start) 6534 goto error; 6535 6536 while (++start <= end) { 6537 if (start >= CPU_SUBSET_MAXCPUS) 6538 goto error; 6539 CPU_SET_S(start, cpu_subset_size, cpu_subset); 6540 } 6541 6542 if (*next == ',') 6543 next += 1; 6544 else if (*next != '\0') 6545 goto error; 6546 } 6547 6548 return; 6549 6550 error: 6551 fprintf(stderr, "\"--cpu %s\" malformed\n", optarg); 6552 help(); 6553 exit(-1); 6554 } 6555 6556 void cmdline(int argc, char **argv) 6557 { 6558 int opt; 6559 int option_index = 0; 6560 static struct option long_options[] = { 6561 { "add", required_argument, 0, 'a' }, 6562 { "cpu", required_argument, 0, 'c' }, 6563 { "Dump", no_argument, 0, 'D' }, 6564 { "debug", no_argument, 0, 'd' }, /* internal, not documented */ 6565 { "enable", required_argument, 0, 'e' }, 6566 { "interval", required_argument, 0, 'i' }, 6567 { "IPC", no_argument, 0, 'I' }, 6568 { "num_iterations", required_argument, 0, 'n' }, 6569 { "header_iterations", required_argument, 0, 'N' }, 6570 { "help", no_argument, 0, 'h' }, 6571 { "hide", required_argument, 0, 'H' }, // meh, -h taken by --help 6572 { "Joules", no_argument, 0, 'J' }, 6573 { "list", no_argument, 0, 'l' }, 6574 { "out", required_argument, 0, 'o' }, 6575 { "quiet", no_argument, 0, 'q' }, 6576 { "show", required_argument, 0, 's' }, 6577 { "Summary", no_argument, 0, 'S' }, 6578 { "TCC", required_argument, 0, 'T' }, 6579 { "version", no_argument, 0, 'v' }, 6580 { 0, 0, 0, 0 } 6581 }; 6582 6583 progname = argv[0]; 6584 6585 while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v", long_options, &option_index)) != -1) { 6586 switch (opt) { 6587 case 'a': 6588 parse_add_command(optarg); 6589 break; 6590 case 'c': 6591 parse_cpu_command(optarg); 6592 break; 6593 case 'D': 6594 dump_only++; 6595 break; 6596 case 'e': 6597 /* --enable specified counter */ 6598 bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST); 6599 break; 6600 case 'd': 6601 debug++; 6602 ENABLE_BIC(BIC_DISABLED_BY_DEFAULT); 6603 break; 6604 case 'H': 6605 /* 6606 * --hide: do not show those specified 6607 * multiple invocations simply clear more bits in enabled mask 6608 */ 6609 bic_enabled &= ~bic_lookup(optarg, HIDE_LIST); 6610 break; 6611 case 'h': 6612 default: 6613 help(); 6614 exit(1); 6615 case 'i': 6616 { 6617 double interval = strtod(optarg, NULL); 6618 6619 if (interval < 0.001) { 6620 fprintf(outf, "interval %f seconds is too small\n", interval); 6621 exit(2); 6622 } 6623 6624 interval_tv.tv_sec = interval_ts.tv_sec = interval; 6625 interval_tv.tv_usec = (interval - interval_tv.tv_sec) * 1000000; 6626 interval_ts.tv_nsec = (interval - interval_ts.tv_sec) * 1000000000; 6627 } 6628 break; 6629 case 'J': 6630 rapl_joules++; 6631 break; 6632 case 'l': 6633 ENABLE_BIC(BIC_DISABLED_BY_DEFAULT); 6634 list_header_only++; 6635 quiet++; 6636 break; 6637 case 'o': 6638 outf = fopen_or_die(optarg, "w"); 6639 break; 6640 case 'q': 6641 quiet = 1; 6642 break; 6643 case 'n': 6644 num_iterations = strtod(optarg, NULL); 6645 6646 if (num_iterations <= 0) { 6647 fprintf(outf, "iterations %d should be positive number\n", num_iterations); 6648 exit(2); 6649 } 6650 break; 6651 case 'N': 6652 header_iterations = strtod(optarg, NULL); 6653 6654 if (header_iterations <= 0) { 6655 fprintf(outf, "iterations %d should be positive number\n", header_iterations); 6656 exit(2); 6657 } 6658 break; 6659 case 's': 6660 /* 6661 * --show: show only those specified 6662 * The 1st invocation will clear and replace the enabled mask 6663 * subsequent invocations can add to it. 6664 */ 6665 if (shown == 0) 6666 bic_enabled = bic_lookup(optarg, SHOW_LIST); 6667 else 6668 bic_enabled |= bic_lookup(optarg, SHOW_LIST); 6669 shown = 1; 6670 break; 6671 case 'S': 6672 summary_only++; 6673 break; 6674 case 'T': 6675 tj_max_override = atoi(optarg); 6676 break; 6677 case 'v': 6678 print_version(); 6679 exit(0); 6680 break; 6681 } 6682 } 6683 } 6684 6685 int main(int argc, char **argv) 6686 { 6687 outf = stderr; 6688 cmdline(argc, argv); 6689 6690 if (!quiet) { 6691 print_version(); 6692 print_bootcmd(); 6693 } 6694 6695 probe_sysfs(); 6696 6697 turbostat_init(); 6698 6699 msr_sum_record(); 6700 6701 /* dump counters and exit */ 6702 if (dump_only) 6703 return get_and_dump_counters(); 6704 6705 /* list header and exit */ 6706 if (list_header_only) { 6707 print_header(","); 6708 flush_output_stdout(); 6709 return 0; 6710 } 6711 6712 /* 6713 * if any params left, it must be a command to fork 6714 */ 6715 if (argc - optind) 6716 return fork_it(argv + optind); 6717 else 6718 turbostat_loop(); 6719 6720 return 0; 6721 } 6722