src/turbostat.c

   1 /*
   2  * turbostat -- Log CPU frequency and C-state residency
   3  * on modern Intel turbo-capable processors for collectd.
   4  *
   5  * Based on the kernel tools by:
   6  * Copyright (c) 2013 Intel Corporation.
   7  * Len Brown <len.brown@intel.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms and conditions of the GNU General Public License,
  11  * version 2, as published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope it will be useful, but WITHOUT
  14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  16  * more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along with
  19  * this program; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  21  *
  22  * Ported to collectd by Vincent Brillault <git@lerya.net>
  23  */
  24
  25 #define _GNU_SOURCE
  26 #include "msr-index.h"
  27 #include <stdarg.h>
  28 #include <stdio.h>
  29 #include <stdbool.h>
  30 #include <err.h>
  31 #include <unistd.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <sys/stat.h>
  35 #include <sys/resource.h>
  36 #include <fcntl.h>
  37 #include <signal.h>
  38 #include <sys/time.h>
  39 #include <stdlib.h>
  40 #include <dirent.h>
  41 #include <string.h>
  42 #include <ctype.h>
  43 #include <sched.h>
  44 #include <cpuid.h>
  45
  46 #include "collectd.h"
  47 #include "common.h"
  48 #include "plugin.h"
  49
  50 #define PLUGIN_NAME "turbostat"
  51
  52 static const char *proc_stat = "/proc/stat";
  53 static unsigned int interval_sec = 5;   /* set with -i interval_sec */
  54 static unsigned int skip_c0;
  55 static unsigned int skip_c1;
  56 static unsigned int do_nhm_cstates;
  57 static unsigned int do_snb_cstates;
  58 static unsigned int do_c8_c9_c10;
  59 static unsigned int do_slm_cstates;
  60 static unsigned int has_aperf;
  61 static unsigned int has_epb;
  62 static unsigned int units = 1000000000; /* Ghz etc */
  63 static unsigned int genuine_intel;
  64 static unsigned int has_invariant_tsc;
  65 static unsigned int do_nehalem_platform_info;
  66 static int do_smi;
  67 static unsigned int show_pkg;
  68 static unsigned int show_core;
  69 static unsigned int show_cpu;
  70 static unsigned int do_rapl;
  71 static unsigned int do_dts;
  72 static unsigned int do_ptm;
  73 static unsigned int tcc_activation_temp;
  74 static unsigned int tcc_activation_temp_override;
  75 static double rapl_power_units, rapl_energy_units, rapl_time_units;
  76 static double rapl_joule_counter_range;
  77
  78 #define RAPL_PKG                (1 << 0)
  79                                         /* 0x610 MSR_PKG_POWER_LIMIT */
  80                                         /* 0x611 MSR_PKG_ENERGY_STATUS */
  81 #define RAPL_PKG_PERF_STATUS    (1 << 1)
  82                                         /* 0x613 MSR_PKG_PERF_STATUS */
  83 #define RAPL_PKG_POWER_INFO     (1 << 2)
  84                                         /* 0x614 MSR_PKG_POWER_INFO */
  85
  86 #define RAPL_DRAM               (1 << 3)
  87                                         /* 0x618 MSR_DRAM_POWER_LIMIT */
  88                                         /* 0x619 MSR_DRAM_ENERGY_STATUS */
  89                                         /* 0x61c MSR_DRAM_POWER_INFO */
  90 #define RAPL_DRAM_PERF_STATUS   (1 << 4)
  91                                         /* 0x61b MSR_DRAM_PERF_STATUS */
  92
  93 #define RAPL_CORES              (1 << 5)
  94                                         /* 0x638 MSR_PP0_POWER_LIMIT */
  95                                         /* 0x639 MSR_PP0_ENERGY_STATUS */
  96 #define RAPL_CORE_POLICY        (1 << 6)
  97                                         /* 0x63a MSR_PP0_POLICY */
  98
  99
 100 #define RAPL_GFX                (1 << 7)
 101                                         /* 0x640 MSR_PP1_POWER_LIMIT */
 102                                         /* 0x641 MSR_PP1_ENERGY_STATUS */
 103                                         /* 0x642 MSR_PP1_POLICY */
 104 #define TJMAX_DEFAULT   100
 105
 106 int aperf_mperf_unstable;
 107 int backwards_count;
 108 char *progname;
 109
 110 cpu_set_t *cpu_present_set, *cpu_affinity_set;
 111 size_t cpu_present_setsize, cpu_affinity_setsize;
 112
 113 struct thread_data {
 114         unsigned long long tsc;
 115         unsigned long long aperf;
 116         unsigned long long mperf;
 117         unsigned long long c1;
 118         unsigned int smi_count;
 119         unsigned int cpu_id;
 120         unsigned int flags;
 121 #define CPU_IS_FIRST_THREAD_IN_CORE     0x2
 122 #define CPU_IS_FIRST_CORE_IN_PACKAGE    0x4
 123 } *thread_even, *thread_odd;
 124
 125 struct core_data {
 126         unsigned long long c3;
 127         unsigned long long c6;
 128         unsigned long long c7;
 129         unsigned int core_temp_c;
 130         unsigned int core_id;
 131 } *core_even, *core_odd;
 132
 133 struct pkg_data {
 134         unsigned long long pc2;
 135         unsigned long long pc3;
 136         unsigned long long pc6;
 137         unsigned long long pc7;
 138         unsigned long long pc8;
 139         unsigned long long pc9;
 140         unsigned long long pc10;
 141         unsigned int package_id;
 142         unsigned int energy_pkg;        /* MSR_PKG_ENERGY_STATUS */
 143         unsigned int energy_dram;       /* MSR_DRAM_ENERGY_STATUS */
 144         unsigned int energy_cores;      /* MSR_PP0_ENERGY_STATUS */
 145         unsigned int energy_gfx;        /* MSR_PP1_ENERGY_STATUS */
 146         unsigned int rapl_pkg_perf_status;      /* MSR_PKG_PERF_STATUS */
 147         unsigned int rapl_dram_perf_status;     /* MSR_DRAM_PERF_STATUS */
 148         unsigned int pkg_temp_c;
 149
 150 } *package_even, *package_odd;
 151
 152 #define ODD_COUNTERS thread_odd, core_odd, package_odd
 153 #define EVEN_COUNTERS thread_even, core_even, package_even
 154 static bool is_even = true;
 155
 156 static bool allocated = false;
 157 static bool initialized = false;
 158
 159 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
 160         (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
 161                 topo.num_threads_per_core + \
 162                 (core_no) * topo.num_threads_per_core + (thread_no))
 163 #define GET_CORE(core_base, core_no, pkg_no) \
 164         (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
 165 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
 166
 167 struct topo_params {
 168         int num_packages;
 169         int num_cpus;
 170         int num_cores;
 171         int max_cpu_num;
 172         int num_cores_per_pkg;
 173         int num_threads_per_core;
 174 } topo;
 175
 176 struct timeval tv_even, tv_odd, tv_delta;
 177
 178 enum return_values {
 179         OK = 0,
 180         ERR_CPU_MIGRATE,
 181         ERR_MSR_IA32_APERF,
 182         ERR_MSR_IA32_MPERF,
 183         ERR_MSR_SMI_COUNT,
 184         ERR_MSR_CORE_C3_RESIDENCY,
 185         ERR_MSR_CORE_C6_RESIDENCY,
 186         ERR_MSR_CORE_C7_RESIDENCY,
 187         ERR_MSR_IA32_THERM_STATUS,
 188         ERR_MSR_PKG_C3_RESIDENCY,
 189         ERR_MSR_PKG_C6_RESIDENCY,
 190         ERR_MSR_PKG_C2_RESIDENCY,
 191         ERR_MSR_PKG_C7_RESIDENCY,
 192         ERR_MSR_PKG_C8_RESIDENCY,
 193         ERR_MSR_PKG_C9_RESIDENCY,
 194         ERR_MSR_PKG_C10_RESIDENCY,
 195         ERR_MSR_PKG_ENERGY_STATUS,
 196         ERR_MSR_PP0_ENERGY_STATUS,
 197         ERR_MSR_DRAM_ENERGY_STATUS,
 198         ERR_MSR_PP1_ENERGY_STATUS,
 199         ERR_MSR_PKG_PERF_STATUS,
 200         ERR_MSR_DRAM_PERF_STATUS,
 201         ERR_MSR_IA32_PACKAGE_THERM_STATUS,
 202         ERR_CPU_NOT_PRESENT,
 203         ERR_NO_MSR,
 204         ERR_CANT_OPEN_FILE,
 205         ERR_CANT_READ_NUMBER,
 206         ERR_CANT_READ_PROC_STAT,
 207         ERR_NO_INVARIANT_TSC,
 208         ERR_NO_APERF,
 209         ERR_CALLOC,
 210         ERR_CPU_ALLOC,
 211         ERR_NOT_ROOT,
 212 };
 213
 214 #define __must_check __attribute__((warn_unused_result))
 215
 216 static int setup_all_buffers(void);
 217
 218 static int
 219 cpu_is_not_present(int cpu)
 220 {
 221         return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
 222 }
 223 /*
 224  * run func(thread, core, package) in topology order
 225  * skip non-present cpus
 226  */
 227
 228 static int __must_check
 229 for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
 230         struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
 231 {
 232         int retval, pkg_no, core_no, thread_no;
 233
 234         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
 235                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
 236                         for (thread_no = 0; thread_no <
 237                                 topo.num_threads_per_core; ++thread_no) {
 238                                 struct thread_data *t;
 239                                 struct core_data *c;
 240                                 struct pkg_data *p;
 241
 242                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
 243
 244                                 if (cpu_is_not_present(t->cpu_id))
 245                                         continue;
 246
 247                                 c = GET_CORE(core_base, core_no, pkg_no);
 248                                 p = GET_PKG(pkg_base, pkg_no);
 249
 250                                 retval = func(t, c, p);
 251                                 if (retval)
 252                                         return retval;
 253                         }
 254                 }
 255         }
 256         return 0;
 257 }
 258
 259 static int __must_check
 260 cpu_migrate(int cpu)
 261 {
 262         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
 263         CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
 264         if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
 265                 return -ERR_CPU_MIGRATE;
 266         else
 267                 return 0;
 268 }
 269
 270 static int __must_check
 271 get_msr(int cpu, off_t offset, unsigned long long *msr)
 272 {
 273         ssize_t retval;
 274         char pathname[32];
 275         int fd;
 276
 277         sprintf(pathname, "/dev/cpu/%d/msr", cpu);
 278         fd = open(pathname, O_RDONLY);
 279         if (fd < 0)
 280                 return -1;
 281
 282         retval = pread(fd, msr, sizeof *msr, offset);
 283         close(fd);
 284
 285         if (retval != sizeof *msr) {
 286                 ERROR ("%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
 287                 return -1;
 288         }
 289
 290         return 0;
 291 }
 292
 293 #define DELTA_WRAP32(new, old)                  \
 294         if (new > old) {                        \
 295                 old = new - old;                \
 296         } else {                                \
 297                 old = 0x100000000 + new - old;  \
 298         }
 299
 300 static void
 301 delta_package(struct pkg_data *new, struct pkg_data *old)
 302 {
 303         old->pc2 = new->pc2 - old->pc2;
 304         old->pc3 = new->pc3 - old->pc3;
 305         old->pc6 = new->pc6 - old->pc6;
 306         old->pc7 = new->pc7 - old->pc7;
 307         old->pc8 = new->pc8 - old->pc8;
 308         old->pc9 = new->pc9 - old->pc9;
 309         old->pc10 = new->pc10 - old->pc10;
 310         old->pkg_temp_c = new->pkg_temp_c;
 311
 312         DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
 313         DELTA_WRAP32(new->energy_cores, old->energy_cores);
 314         DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
 315         DELTA_WRAP32(new->energy_dram, old->energy_dram);
 316         DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
 317         DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
 318 }
 319
 320 static void
 321 delta_core(struct core_data *new, struct core_data *old)
 322 {
 323         old->c3 = new->c3 - old->c3;
 324         old->c6 = new->c6 - old->c6;
 325         old->c7 = new->c7 - old->c7;
 326         old->core_temp_c = new->core_temp_c;
 327 }
 328
 329 /*
 330  * old = new - old
 331  */
 332 static int __must_check
 333 delta_thread(struct thread_data *new, struct thread_data *old,
 334         struct core_data *core_delta)
 335 {
 336         old->tsc = new->tsc - old->tsc;
 337
 338         /* check for TSC < 1 Mcycles over interval */
 339         if (old->tsc < (1000 * 1000)) {
 340                 WARNING("Insanely slow TSC rate, TSC stops in idle?\n"
 341                         "You can disable all c-states by booting with \"idle=poll\"\n"
 342                         "or just the deep ones with \"processor.max_cstate=1\"");
 343                 return -1;
 344         }
 345
 346         old->c1 = new->c1 - old->c1;
 347
 348         if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
 349                 old->aperf = new->aperf - old->aperf;
 350                 old->mperf = new->mperf - old->mperf;
 351         } else {
 352
 353                 if (!aperf_mperf_unstable) {
 354                         WARNING("%s: APERF or MPERF went backwards *\n", progname);
 355                         WARNING("* Frequency results do not cover entire interval *\n");
 356                         WARNING("* fix this by running Linux-2.6.30 or later *\n");
 357
 358                         aperf_mperf_unstable = 1;
 359                 }
 360                 /*
 361                  * mperf delta is likely a huge "positive" number
 362                  * can not use it for calculating c0 time
 363                  */
 364                 skip_c0 = 1;
 365                 skip_c1 = 1;
 366         }
 367
 368
 369         /*
 370          * As counter collection is not atomic,
 371          * it is possible for mperf's non-halted cycles + idle states
 372          * to exceed TSC's all cycles: show c1 = 0% in that case.
 373          */
 374         if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
 375                 old->c1 = 0;
 376         else {
 377                 /* normal case, derive c1 */
 378                 old->c1 = old->tsc - old->mperf - core_delta->c3
 379                         - core_delta->c6 - core_delta->c7;
 380         }
 381
 382         if (old->mperf == 0) {
 383                 WARNING("cpu%d MPERF 0!\n", old->cpu_id);
 384                 old->mperf = 1; /* divide by 0 protection */
 385         }
 386
 387         if (do_smi)
 388                 old->smi_count = new->smi_count - old->smi_count;
 389
 390         return 0;
 391 }
 392
 393 static int __must_check
 394 delta_cpu(struct thread_data *t, struct core_data *c,
 395         struct pkg_data *p, struct thread_data *t2,
 396         struct core_data *c2, struct pkg_data *p2)
 397 {
 398         int ret;
 399
 400         /* calculate core delta only for 1st thread in core */
 401         if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
 402                 delta_core(c, c2);
 403
 404         /* always calculate thread delta */
 405         ret = delta_thread(t, t2, c2);  /* c2 is core delta */
 406         if (ret != 0)
 407                 return ret;
 408
 409         /* calculate package delta only for 1st core in package */
 410         if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
 411                 delta_package(p, p2);
 412
 413         return 0;
 414 }
 415
 416 static unsigned long long
 417 rdtsc(void)
 418 {
 419         unsigned int low, high;
 420
 421         asm volatile("rdtsc" : "=a" (low), "=d" (high));
 422
 423         return low | ((unsigned long long)high) << 32;
 424 }
 425
 426
 427 /*
 428  * get_counters(...)
 429  * migrate to cpu
 430  * acquire and record local counters for that cpu
 431  */
 432 static int __must_check
 433 get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 434 {
 435         int cpu = t->cpu_id;
 436         unsigned long long msr;
 437
 438         if (cpu_migrate(cpu)) {
 439                 WARNING("Could not migrate to CPU %d\n", cpu);
 440                 return -ERR_CPU_MIGRATE;
 441         }
 442
 443         t->tsc = rdtsc();       /* we are running on local CPU of interest */
 444
 445         if (has_aperf) {
 446                 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
 447                         return -ERR_MSR_IA32_APERF;
 448                 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
 449                         return -ERR_MSR_IA32_MPERF;
 450         }
 451
 452         if (do_smi) {
 453                 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
 454                         return -ERR_MSR_SMI_COUNT;
 455                 t->smi_count = msr & 0xFFFFFFFF;
 456         }
 457
 458         /* collect core counters only for 1st thread in core */
 459         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
 460                 return 0;
 461
 462         if (do_nhm_cstates && !do_slm_cstates) {
 463                 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
 464                         return -ERR_MSR_CORE_C3_RESIDENCY;
 465         }
 466
 467         if (do_nhm_cstates) {
 468                 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
 469                         return -ERR_MSR_CORE_C6_RESIDENCY;
 470         }
 471
 472         if (do_snb_cstates)
 473                 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
 474                         return -ERR_MSR_CORE_C7_RESIDENCY;
 475
 476         if (do_dts) {
 477                 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
 478                         return -ERR_MSR_IA32_THERM_STATUS;
 479                 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
 480         }
 481
 482
 483         /* collect package counters only for 1st core in package */
 484         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
 485                 return 0;
 486
 487         if (do_nhm_cstates && !do_slm_cstates) {
 488                 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
 489                         return -ERR_MSR_PKG_C3_RESIDENCY;
 490                 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
 491                         return -ERR_MSR_PKG_C6_RESIDENCY;
 492         }
 493         if (do_snb_cstates) {
 494                 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
 495                         return -ERR_MSR_PKG_C2_RESIDENCY;
 496                 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
 497                         return -ERR_MSR_PKG_C7_RESIDENCY;
 498         }
 499         if (do_c8_c9_c10) {
 500                 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
 501                         return -ERR_MSR_PKG_C8_RESIDENCY;
 502                 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
 503                         return -ERR_MSR_PKG_C9_RESIDENCY;
 504                 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
 505                         return -ERR_MSR_PKG_C10_RESIDENCY;
 506         }
 507         if (do_rapl & RAPL_PKG) {
 508                 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
 509                         return -ERR_MSR_PKG_ENERGY_STATUS;
 510                 p->energy_pkg = msr & 0xFFFFFFFF;
 511         }
 512         if (do_rapl & RAPL_CORES) {
 513                 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
 514                         return MSR_PP0_ENERGY_STATUS;
 515                 p->energy_cores = msr & 0xFFFFFFFF;
 516         }
 517         if (do_rapl & RAPL_DRAM) {
 518                 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
 519                         return -ERR_MSR_DRAM_ENERGY_STATUS;
 520                 p->energy_dram = msr & 0xFFFFFFFF;
 521         }
 522         if (do_rapl & RAPL_GFX) {
 523                 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
 524                         return -ERR_MSR_PP1_ENERGY_STATUS;
 525                 p->energy_gfx = msr & 0xFFFFFFFF;
 526         }
 527         if (do_rapl & RAPL_PKG_PERF_STATUS) {
 528                 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
 529                         return -ERR_MSR_PKG_PERF_STATUS;
 530                 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
 531         }
 532         if (do_rapl & RAPL_DRAM_PERF_STATUS) {
 533                 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
 534                         return -ERR_MSR_DRAM_PERF_STATUS;
 535                 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
 536         }
 537         if (do_ptm) {
 538                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
 539                         return -ERR_MSR_IA32_PACKAGE_THERM_STATUS;
 540                 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
 541         }
 542         return 0;
 543 }
 544
 545 static void
 546 free_all_buffers(void)
 547 {
 548         allocated = false;
 549         initialized = false;
 550
 551         CPU_FREE(cpu_present_set);
 552         cpu_present_set = NULL;
 553         cpu_present_set = 0;
 554
 555         CPU_FREE(cpu_affinity_set);
 556         cpu_affinity_set = NULL;
 557         cpu_affinity_setsize = 0;
 558
 559         free(thread_even);
 560         free(core_even);
 561         free(package_even);
 562
 563         thread_even = NULL;
 564         core_even = NULL;
 565         package_even = NULL;
 566
 567         free(thread_odd);
 568         free(core_odd);
 569         free(package_odd);
 570
 571         thread_odd = NULL;
 572         core_odd = NULL;
 573         package_odd = NULL;
 574 }
 575
 576 /*
 577  * Parse a file containing a single int.
 578  */
 579 static int
 580 parse_int_file(const char *fmt, ...)
 581 {
 582         va_list args;
 583         char path[PATH_MAX];
 584         FILE *filep;
 585         int value;
 586
 587         va_start(args, fmt);
 588         vsnprintf(path, sizeof(path), fmt, args);
 589         va_end(args);
 590         filep = fopen(path, "r");
 591         if (!filep) {
 592                 ERROR("%s: open failed", path);
 593                 return -ERR_CANT_OPEN_FILE;
 594         }
 595         if (fscanf(filep, "%d", &value) != 1) {
 596                 ERROR("%s: failed to parse number from file", path);
 597                 return -ERR_CANT_READ_NUMBER;
 598         }
 599         fclose(filep);
 600         return value;
 601 }
 602
 603 /*
 604  * cpu_is_first_sibling_in_core(cpu)
 605  * return 1 if given CPU is 1st HT sibling in the core
 606  */
 607 static int
 608 cpu_is_first_sibling_in_core(int cpu)
 609 {
 610         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
 611 }
 612
 613 /*
 614  * cpu_is_first_core_in_package(cpu)
 615  * return 1 if given CPU is 1st core in package
 616  */
 617 static int
 618 cpu_is_first_core_in_package(int cpu)
 619 {
 620         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
 621 }
 622
 623 static int
 624 get_physical_package_id(int cpu)
 625 {
 626         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
 627 }
 628
 629 static int
 630 get_core_id(int cpu)
 631 {
 632         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
 633 }
 634
 635 static int
 636 get_num_ht_siblings(int cpu)
 637 {
 638         char path[80];
 639         FILE *filep;
 640         int sib1, sib2;
 641         int matches;
 642         char character;
 643
 644         sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
 645         filep = fopen(path, "r");
 646         if (!filep) {
 647                 ERROR("%s: open failed", path);
 648                 return -ERR_CANT_OPEN_FILE;
 649         }
 650         /*
 651          * file format:
 652          * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
 653          * otherwinse 1 sibling (self).
 654          */
 655         matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
 656
 657         fclose(filep);
 658
 659         if (matches == 3)
 660                 return 2;
 661         else
 662                 return 1;
 663 }
 664
 665 /*
 666  * run func(thread, core, package) in topology order
 667  * skip non-present cpus
 668  */
 669
 670
 671 static int __must_check
 672 for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
 673         struct pkg_data *, struct thread_data *, struct core_data *,
 674         struct pkg_data *), struct thread_data *thread_base,
 675         struct core_data *core_base, struct pkg_data *pkg_base,
 676         struct thread_data *thread_base2, struct core_data *core_base2,
 677         struct pkg_data *pkg_base2)
 678 {
 679         int retval, pkg_no, core_no, thread_no;
 680
 681         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
 682                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
 683                         for (thread_no = 0; thread_no <
 684                                 topo.num_threads_per_core; ++thread_no) {
 685                                 struct thread_data *t, *t2;
 686                                 struct core_data *c, *c2;
 687                                 struct pkg_data *p, *p2;
 688
 689                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
 690
 691                                 if (cpu_is_not_present(t->cpu_id))
 692                                         continue;
 693
 694                                 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
 695
 696                                 c = GET_CORE(core_base, core_no, pkg_no);
 697                                 c2 = GET_CORE(core_base2, core_no, pkg_no);
 698
 699                                 p = GET_PKG(pkg_base, pkg_no);
 700                                 p2 = GET_PKG(pkg_base2, pkg_no);
 701
 702                                 retval = func(t, c, p, t2, c2, p2);
 703                                 if (retval)
 704                                         return retval;
 705                         }
 706                 }
 707         }
 708         return 0;
 709 }
 710
 711 /*
 712  * run func(cpu) on every cpu in /proc/stat
 713  * return max_cpu number
 714  */
 715 static int __must_check
 716 for_all_proc_cpus(int (func)(int))
 717 {
 718         FILE *fp;
 719         int cpu_num;
 720         int retval;
 721
 722         fp = fopen(proc_stat, "r");
 723         if (!fp) {
 724                 ERROR("%s: open failed", proc_stat);
 725                 return -ERR_CANT_OPEN_FILE;
 726         }
 727
 728         retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
 729         if (retval != 0) {
 730                 ERROR("%s: failed to parse format", proc_stat);
 731                 return -ERR_CANT_READ_PROC_STAT;
 732         }
 733
 734         while (1) {
 735                 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
 736                 if (retval != 1)
 737                         break;
 738
 739                 retval = func(cpu_num);
 740                 if (retval) {
 741                         fclose(fp);
 742                         return(retval);
 743                 }
 744         }
 745         fclose(fp);
 746         return 0;
 747 }
 748
 749 /*
 750  * count_cpus()
 751  * remember the last one seen, it will be the max
 752  */
 753 static int
 754 count_cpus(int cpu)
 755 {
 756         if (topo.max_cpu_num < cpu)
 757                 topo.max_cpu_num = cpu;
 758
 759         topo.num_cpus += 1;
 760         return 0;
 761 }
 762 static int
 763 mark_cpu_present(int cpu)
 764 {
 765         CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
 766         return 0;
 767 }
 768
 769
 770 static void
 771 turbostat_submit (const char *plugin_instance,
 772         const char *type, const char *type_instance,
 773         gauge_t value)
 774 {
 775         value_list_t vl = VALUE_LIST_INIT;
 776         value_t v;
 777
 778         v.gauge = value;
 779         vl.values = &v;
 780         vl.values_len = 1;
 781         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 782         sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
 783         if (plugin_instance != NULL)
 784                 sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 785         sstrncpy (vl.type, type, sizeof (vl.type));
 786         if (type_instance != NULL)
 787                 sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
 788
 789         plugin_dispatch_values (&vl);
 790 }
 791
 792 /*
 793  * column formatting convention & formats
 794  * package: "pk" 2 columns %2d
 795  * core: "cor" 3 columns %3d
 796  * CPU: "CPU" 3 columns %3d
 797  * Pkg_W: %6.2
 798  * Cor_W: %6.2
 799  * GFX_W: %5.2
 800  * RAM_W: %5.2
 801  * GHz: "GHz" 3 columns %3.2
 802  * TSC: "TSC" 3 columns %3.2
 803  * SMI: "SMI" 4 columns %4d
 804  * percentage " %pc3" %6.2
 805  * Perf Status percentage: %5.2
 806  * "CTMP" 4 columns %4d
 807  */
 808 #define NAME_LEN 12
 809 static int
 810 submit_counters(struct thread_data *t, struct core_data *c,
 811         struct pkg_data *p)
 812 {
 813         char name[NAME_LEN];
 814         double interval_float;
 815
 816         interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
 817
 818         snprintf(name, NAME_LEN, "cpu%02d", t->cpu_id);
 819
 820         if (do_nhm_cstates) {
 821                 if (!skip_c0)
 822                         turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
 823                 if (!skip_c1)
 824                         turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
 825         }
 826
 827         /* GHz */
 828         if (has_aperf && ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc))))
 829                 turbostat_submit(NULL, "frequency", name, 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
 830
 831         /* SMI */
 832         if (do_smi)
 833                 turbostat_submit(NULL, "current", name, t->smi_count);
 834
 835         /* print per-core data only for 1st thread in core */
 836         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
 837                 goto done;
 838
 839         snprintf(name, NAME_LEN, "core%02d", c->core_id);
 840
 841         if (do_nhm_cstates && !do_slm_cstates)
 842                 turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
 843         if (do_nhm_cstates)
 844                 turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
 845         if (do_snb_cstates)
 846                 turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
 847
 848         if (do_dts)
 849                 turbostat_submit(NULL, "temperature", name, c->core_temp_c);
 850
 851         /* print per-package data only for 1st core in package */
 852         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
 853                 goto done;
 854
 855         snprintf(name, NAME_LEN, "pkg%02d", p->package_id);
 856
 857         if (do_ptm)
 858                 turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
 859
 860         if (do_snb_cstates)
 861                 turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
 862         if (do_nhm_cstates && !do_slm_cstates)
 863                 turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
 864         if (do_nhm_cstates && !do_slm_cstates)
 865                 turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
 866         if (do_snb_cstates)
 867                 turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
 868         if (do_c8_c9_c10) {
 869                 turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
 870                 turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
 871                 turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
 872         }
 873
 874         if (do_rapl) {
 875                 if (do_rapl & RAPL_PKG)
 876                         turbostat_submit(name, "power", "Pkg_W", p->energy_pkg * rapl_energy_units / interval_float);
 877                 if (do_rapl & RAPL_CORES)
 878                         turbostat_submit(name, "power", "Cor_W", p->energy_cores * rapl_energy_units / interval_float);
 879                 if (do_rapl & RAPL_GFX)
 880                         turbostat_submit(name, "power", "GFX_W", p->energy_gfx * rapl_energy_units / interval_float);
 881                 if (do_rapl & RAPL_DRAM)
 882                         turbostat_submit(name, "power", "RAM_W", p->energy_dram * rapl_energy_units / interval_float);
 883         }
 884 done:
 885         return 0;
 886 }
 887
 888 static int
 889 turbostat_read(user_data_t * not_used)
 890 {
 891         int ret;
 892
 893         if (!allocated) {
 894                 if ((ret = setup_all_buffers()) < 0)
 895                         return ret;
 896         }
 897
 898         if (for_all_proc_cpus(cpu_is_not_present)) {
 899                 free_all_buffers();
 900                 if ((ret = setup_all_buffers()) < 0)
 901                         return ret;
 902                 if (for_all_proc_cpus(cpu_is_not_present))
 903                         return -ERR_CPU_NOT_PRESENT;
 904         }
 905
 906         if (!initialized) {
 907                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
 908                         return ret;
 909                 gettimeofday(&tv_even, (struct timezone *)NULL);
 910                 is_even = true;
 911                 initialized = true;
 912                 return 0;
 913         }
 914
 915         if (is_even) {
 916                 if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
 917                         return ret;
 918                 gettimeofday(&tv_odd, (struct timezone *)NULL);
 919                 is_even = false;
 920                 timersub(&tv_odd, &tv_even, &tv_delta);
 921                 if ((ret = for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) < 0)
 922                         return ret;
 923                 if ((ret = for_all_cpus(submit_counters, EVEN_COUNTERS)) < 0)
 924                         return ret;
 925         } else {
 926                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
 927                         return ret;
 928                 gettimeofday(&tv_even, (struct timezone *)NULL);
 929                 is_even = true;
 930                 timersub(&tv_even, &tv_odd, &tv_delta);
 931                 if ((ret = for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) < 0)
 932                         return ret;
 933                 if ((ret = for_all_cpus(submit_counters, ODD_COUNTERS)) < 0)
 934                         return ret;
 935         }
 936         return 0;
 937 }
 938
 939 static int __must_check
 940 check_dev_msr()
 941 {
 942         struct stat sb;
 943
 944         if (stat("/dev/cpu/0/msr", &sb)) {
 945                 ERROR("no /dev/cpu/0/msr\n"
 946                         "Try \"# modprobe msr\"");
 947                 return -ERR_NO_MSR;
 948         }
 949         return 0;
 950 }
 951
 952 static int __must_check
 953 check_super_user()
 954 {
 955         if (getuid() != 0) {
 956                 ERROR("must be root");
 957                 return -ERR_NOT_ROOT;
 958         }
 959         return 0;
 960 }
 961
 962
 963 #define RAPL_POWER_GRANULARITY  0x7FFF  /* 15 bit power granularity */
 964 #define RAPL_TIME_GRANULARITY   0x3F /* 6 bit time granularity */
 965
 966 static double
 967 get_tdp(unsigned int model)
 968 {
 969         unsigned long long msr;
 970
 971         if (do_rapl & RAPL_PKG_POWER_INFO)
 972                 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
 973                         return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
 974
 975         switch (model) {
 976         case 0x37:
 977         case 0x4D:
 978                 return 30.0;
 979         default:
 980                 return 135.0;
 981         }
 982 }
 983
 984
 985 /*
 986  * rapl_probe()
 987  *
 988  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
 989  */
 990 static void
 991 rapl_probe(unsigned int family, unsigned int model)
 992 {
 993         unsigned long long msr;
 994         unsigned int time_unit;
 995         double tdp;
 996
 997         if (!genuine_intel)
 998                 return;
 999
1000         if (family != 6)
1001                 return;
1002
1003         switch (model) {
1004         case 0x2A:
1005         case 0x3A:
1006         case 0x3C:      /* HSW */
1007         case 0x45:      /* HSW */
1008         case 0x46:      /* HSW */
1009                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1010                 break;
1011         case 0x3F:      /* HSX */
1012                 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1013                 break;
1014         case 0x2D:
1015         case 0x3E:
1016                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1017                 break;
1018         case 0x37:      /* BYT */
1019         case 0x4D:      /* AVN */
1020                 do_rapl = RAPL_PKG | RAPL_CORES ;
1021                 break;
1022         default:
1023                 return;
1024         }
1025
1026         /* units on package 0, verify later other packages match */
1027         if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1028                 return;
1029
1030         rapl_power_units = 1.0 / (1 << (msr & 0xF));
1031         if (model == 0x37)
1032                 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1033         else
1034                 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1035
1036         time_unit = msr >> 16 & 0xF;
1037         if (time_unit == 0)
1038                 time_unit = 0xA;
1039
1040         rapl_time_units = 1.0 / (1 << (time_unit));
1041
1042         tdp = get_tdp(model);
1043
1044         rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1045 //      if (verbose)
1046 //              fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1047
1048         return;
1049 }
1050
1051 static int
1052 is_snb(unsigned int family, unsigned int model)
1053 {
1054         if (!genuine_intel)
1055                 return 0;
1056
1057         switch (model) {
1058         case 0x2A:
1059         case 0x2D:
1060         case 0x3A:      /* IVB */
1061         case 0x3E:      /* IVB Xeon */
1062         case 0x3C:      /* HSW */
1063         case 0x3F:      /* HSW */
1064         case 0x45:      /* HSW */
1065         case 0x46:      /* HSW */
1066                 return 1;
1067         }
1068         return 0;
1069 }
1070
1071 static int
1072 has_c8_c9_c10(unsigned int family, unsigned int model)
1073 {
1074         if (!genuine_intel)
1075                 return 0;
1076
1077         switch (model) {
1078         case 0x45:
1079                 return 1;
1080         }
1081         return 0;
1082 }
1083
1084
1085 static int
1086 is_slm(unsigned int family, unsigned int model)
1087 {
1088         if (!genuine_intel)
1089                 return 0;
1090         switch (model) {
1091         case 0x37:      /* BYT */
1092         case 0x4D:      /* AVN */
1093                 return 1;
1094         }
1095         return 0;
1096 }
1097
1098 /*
1099  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1100  * the Thermal Control Circuit (TCC) activates.
1101  * This is usually equal to tjMax.
1102  *
1103  * Older processors do not have this MSR, so there we guess,
1104  * but also allow cmdline over-ride with -T.
1105  *
1106  * Several MSR temperature values are in units of degrees-C
1107  * below this value, including the Digital Thermal Sensor (DTS),
1108  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1109  */
1110 static int __must_check
1111 set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1112 {
1113         unsigned long long msr;
1114         unsigned int target_c_local;
1115         int cpu;
1116
1117         /* tcc_activation_temp is used only for dts or ptm */
1118         if (!(do_dts || do_ptm))
1119                 return 0;
1120
1121         /* this is a per-package concept */
1122         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1123                 return 0;
1124
1125         cpu = t->cpu_id;
1126         if (cpu_migrate(cpu)) {
1127                 ERROR("Could not migrate to CPU %d\n", cpu);
1128                 return -ERR_CPU_MIGRATE;
1129         }
1130
1131         if (tcc_activation_temp_override != 0) {
1132                 tcc_activation_temp = tcc_activation_temp_override;
1133                 ERROR("cpu%d: Using cmdline TCC Target (%d C)\n",
1134                         cpu, tcc_activation_temp);
1135                 return 0;
1136         }
1137
1138         /* Temperature Target MSR is Nehalem and newer only */
1139         if (!do_nehalem_platform_info)
1140                 goto guess;
1141
1142         if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1143                 goto guess;
1144
1145         target_c_local = (msr >> 16) & 0x7F;
1146
1147         if (target_c_local < 85 || target_c_local > 127)
1148                 goto guess;
1149
1150         tcc_activation_temp = target_c_local;
1151
1152         return 0;
1153
1154 guess:
1155         tcc_activation_temp = TJMAX_DEFAULT;
1156         WARNING("cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
1157                 cpu, tcc_activation_temp);
1158
1159         return 0;
1160 }
1161
1162 static int __must_check
1163 check_cpuid()
1164 {
1165         unsigned int eax, ebx, ecx, edx, max_level;
1166         unsigned int fms, family, model;
1167
1168         eax = ebx = ecx = edx = 0;
1169
1170         __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
1171
1172         if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
1173                 genuine_intel = 1;
1174
1175         fms = 0;
1176         __get_cpuid(1, &fms, &ebx, &ecx, &edx);
1177         family = (fms >> 8) & 0xf;
1178         model = (fms >> 4) & 0xf;
1179         if (family == 6 || family == 0xf)
1180                 model += ((fms >> 16) & 0xf) << 4;
1181
1182         if (!(edx & (1 << 5))) {
1183                 ERROR("CPUID: no MSR");
1184                 return -ERR_NO_MSR;
1185         }
1186
1187         /*
1188          * check max extended function levels of CPUID.
1189          * This is needed to check for invariant TSC.
1190          * This check is valid for both Intel and AMD.
1191          */
1192         ebx = ecx = edx = 0;
1193         __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
1194
1195         if (max_level < 0x80000007) {
1196                 ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level);
1197                 return -ERR_NO_INVARIANT_TSC;
1198         }
1199
1200         /*
1201          * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
1202          * this check is valid for both Intel and AMD
1203          */
1204         __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1205         has_invariant_tsc = edx & (1 << 8);
1206
1207         if (!has_invariant_tsc) {
1208                 ERROR("No invariant TSC");
1209                 return -ERR_NO_INVARIANT_TSC;
1210         }
1211
1212         /*
1213          * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
1214          * this check is valid for both Intel and AMD
1215          */
1216
1217         __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
1218         has_aperf = ecx & (1 << 0);
1219         do_dts = eax & (1 << 0);
1220         do_ptm = eax & (1 << 6);
1221         has_epb = ecx & (1 << 3);
1222
1223         if (!has_aperf) {
1224                 ERROR("No APERF");
1225                 return -ERR_NO_APERF;
1226         }
1227
1228         do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
1229         do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
1230         do_smi = do_nhm_cstates;
1231         do_snb_cstates = is_snb(family, model);
1232         do_c8_c9_c10 = has_c8_c9_c10(family, model);
1233         do_slm_cstates = is_slm(family, model);
1234
1235         rapl_probe(family, model);
1236
1237         return 0;
1238 }
1239
1240
1241
1242 static int __must_check
1243 topology_probe()
1244 {
1245         int i;
1246         int ret;
1247         int max_core_id = 0;
1248         int max_package_id = 0;
1249         int max_siblings = 0;
1250         struct cpu_topology {
1251                 int core_id;
1252                 int physical_package_id;
1253         } *cpus;
1254
1255         /* Initialize num_cpus, max_cpu_num */
1256         topo.num_cpus = 0;
1257         topo.max_cpu_num = 0;
1258         ret = for_all_proc_cpus(count_cpus);
1259         if (ret < 0)
1260                 return ret;
1261         if (topo.num_cpus > 1)
1262                 show_cpu = 1;
1263
1264         DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
1265
1266         cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
1267         if (cpus == NULL) {
1268                 ERROR("calloc cpus");
1269                 return -ERR_CALLOC;
1270         }
1271
1272         /*
1273          * Allocate and initialize cpu_present_set
1274          */
1275         cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
1276         if (cpu_present_set == NULL) {
1277                 free(cpus);
1278                 ERROR("CPU_ALLOC");
1279                 return -ERR_CPU_ALLOC;
1280         }
1281         cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1282         CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
1283         ret = for_all_proc_cpus(mark_cpu_present);
1284         if (ret < 0) {
1285                 free(cpus);
1286                 return ret;
1287         }
1288
1289         /*
1290          * Allocate and initialize cpu_affinity_set
1291          */
1292         cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
1293         if (cpu_affinity_set == NULL) {
1294                 free(cpus);
1295                 ERROR("CPU_ALLOC");
1296                 return -ERR_CPU_ALLOC;
1297         }
1298         cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1299         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
1300
1301
1302         /*
1303          * For online cpus
1304          * find max_core_id, max_package_id
1305          */
1306         for (i = 0; i <= topo.max_cpu_num; ++i) {
1307                 int siblings;
1308
1309                 if (cpu_is_not_present(i)) {
1310                         //if (verbose > 1)
1311                                 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
1312                         continue;
1313                 }
1314                 cpus[i].core_id = get_core_id(i);
1315                 if (cpus[i].core_id < 0)
1316                         return cpus[i].core_id;
1317                 if (cpus[i].core_id > max_core_id)
1318                         max_core_id = cpus[i].core_id;
1319
1320                 cpus[i].physical_package_id = get_physical_package_id(i);
1321                 if (cpus[i].physical_package_id < 0)
1322                         return cpus[i].physical_package_id;
1323                 if (cpus[i].physical_package_id > max_package_id)
1324                         max_package_id = cpus[i].physical_package_id;
1325
1326                 siblings = get_num_ht_siblings(i);
1327                 if (siblings < 0)
1328                         return siblings;
1329                 if (siblings > max_siblings)
1330                         max_siblings = siblings;
1331                 DEBUG("cpu %d pkg %d core %d\n",
1332                         i, cpus[i].physical_package_id, cpus[i].core_id);
1333         }
1334         topo.num_cores_per_pkg = max_core_id + 1;
1335         DEBUG("max_core_id %d, sizing for %d cores per package\n",
1336                 max_core_id, topo.num_cores_per_pkg);
1337         if (topo.num_cores_per_pkg > 1)
1338                 show_core = 1;
1339
1340         topo.num_packages = max_package_id + 1;
1341         DEBUG("max_package_id %d, sizing for %d packages\n",
1342                 max_package_id, topo.num_packages);
1343         if (topo.num_packages > 1)
1344                 show_pkg = 1;
1345
1346         topo.num_threads_per_core = max_siblings;
1347         DEBUG("max_siblings %d\n", max_siblings);
1348
1349         free(cpus);
1350         return 0;
1351 }
1352
1353 static int
1354 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
1355 {
1356         int i;
1357
1358         *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
1359                 topo.num_packages, sizeof(struct thread_data));
1360         if (*t == NULL)
1361                 goto error;
1362
1363         for (i = 0; i < topo.num_threads_per_core *
1364                 topo.num_cores_per_pkg * topo.num_packages; i++)
1365                 (*t)[i].cpu_id = -1;
1366
1367         *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
1368                 sizeof(struct core_data));
1369         if (*c == NULL)
1370                 goto error;
1371
1372         for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
1373                 (*c)[i].core_id = -1;
1374
1375         *p = calloc(topo.num_packages, sizeof(struct pkg_data));
1376         if (*p == NULL)
1377                 goto error;
1378
1379         for (i = 0; i < topo.num_packages; i++)
1380                 (*p)[i].package_id = i;
1381
1382         return 0;
1383 error:
1384         ERROR("calloc counters");
1385         return -ERR_CALLOC;
1386 }
1387 /*
1388  * init_counter()
1389  *
1390  * set cpu_id, core_num, pkg_num
1391  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
1392  *
1393  * increment topo.num_cores when 1st core in pkg seen
1394  */
1395 static int
1396 init_counter(struct thread_data *thread_base, struct core_data *core_base,
1397         struct pkg_data *pkg_base, int thread_num, int core_num,
1398         int pkg_num, int cpu_id)
1399 {
1400         int ret;
1401         struct thread_data *t;
1402         struct core_data *c;
1403         struct pkg_data *p;
1404
1405         t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
1406         c = GET_CORE(core_base, core_num, pkg_num);
1407         p = GET_PKG(pkg_base, pkg_num);
1408
1409         t->cpu_id = cpu_id;
1410         if (thread_num == 0) {
1411                 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1412                 if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
1413                         return ret;
1414                 } else if (ret != 0) {
1415                         t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1416                 }
1417         }
1418
1419         c->core_id = core_num;
1420         p->package_id = pkg_num;
1421
1422         return 0;
1423 }
1424
1425
1426 static int
1427 initialize_counters(int cpu_id)
1428 {
1429         int my_thread_id, my_core_id, my_package_id;
1430         int ret;
1431
1432         my_package_id = get_physical_package_id(cpu_id);
1433         if (my_package_id < 0)
1434                 return my_package_id;
1435         my_core_id = get_core_id(cpu_id);
1436         if (my_core_id < 0)
1437                 return my_core_id;
1438
1439         if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
1440                 return ret;
1441         } else if (ret != 0) {
1442                 my_thread_id = 0;
1443                 topo.num_cores++;
1444         } else {
1445                 my_thread_id = 1;
1446         }
1447
1448         ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1449         if (ret < 0)
1450                 return ret;
1451         ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1452         if (ret < 0)
1453                 return ret;
1454         return 0;
1455 }
1456
1457 #define DO_OR_GOTO_ERR(something) \
1458 do {                         \
1459         ret = (something);     \
1460         if (ret < 0)         \
1461                 goto err;    \
1462 } while (0)
1463
1464 static int setup_all_buffers(void)
1465 {
1466         int ret;
1467
1468         DO_OR_GOTO_ERR(topology_probe());
1469         DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
1470         DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
1471         DO_OR_GOTO_ERR(for_all_proc_cpus(initialize_counters));
1472
1473         allocated = true;
1474         return 0;
1475 err:
1476         free_all_buffers();
1477         return ret;
1478 }
1479
1480 static int
1481 turbostat_init(void)
1482 {
1483         int ret;
1484         struct timespec ts;
1485
1486         DO_OR_GOTO_ERR(check_cpuid());
1487         DO_OR_GOTO_ERR(check_dev_msr());
1488         DO_OR_GOTO_ERR(check_super_user());
1489         DO_OR_GOTO_ERR(setup_all_buffers());
1490         DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
1491
1492         ts.tv_sec = interval_sec;
1493         ts.tv_nsec = 0;
1494
1495         plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, &ts, NULL);
1496
1497         return 0;
1498 err:
1499         free_all_buffers();
1500         return ret;
1501 }
1502
1503 static const char *config_keys[] =
1504 {
1505         "Interval",
1506 };
1507 static int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
1508
1509 static int
1510 turbostat_config(const char *key, const char *value)
1511 {
1512         if (strcasecmp("Interval", key) == 0)
1513                 interval_sec = atoi(value);
1514         else
1515                 return -1;
1516         return 0;
1517 }
1518
1519 void module_register(void);
1520 void module_register(void)
1521 {
1522         plugin_register_init(PLUGIN_NAME, turbostat_init);
1523         plugin_register_config(PLUGIN_NAME, turbostat_config, config_keys, config_keys_num);
1524 }