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 'turbostat' tool of the Linux kernel, found at
   6  * linux/tools/power/x86/turbostat/turbostat.c:
   7  * ----
   8  * Copyright (c) 2013 Intel Corporation.
   9  * Len Brown <len.brown@intel.com>
  10  *
  11  * This program is free software; you can redistribute it and/or modify it
  12  * under the terms and conditions of the GNU General Public License,
  13  * version 2, as published by the Free Software Foundation.
  14  *
  15  * This program is distributed in the hope it will be useful, but WITHOUT
  16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  17  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  18  * more details.
  19  *
  20  * You should have received a copy of the GNU General Public License along with
  21  * this program; if not, write to the Free Software Foundation, Inc.,
  22  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  23  * ----
  24  * Ported to collectd by Vincent Brillault <git@lerya.net>
  25  */
  26
  27 /*
  28  * _GNU_SOURCE is required because of the following functions:
  29  * - CPU_ISSET_S
  30  * - CPU_ZERO_S
  31  * - CPU_SET_S
  32  * - CPU_FREE
  33  * - CPU_ALLOC
  34  * - CPU_ALLOC_SIZE
  35  */
  36 #define _GNU_SOURCE
  37
  38 #include <asm/msr-index.h>
  39 #include <stdarg.h>
  40 #include <stdio.h>
  41 #include <err.h>
  42 #include <unistd.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/stat.h>
  46 #include <sys/resource.h>
  47 #include <fcntl.h>
  48 #include <signal.h>
  49 #include <sys/time.h>
  50 #include <stdlib.h>
  51 #include <dirent.h>
  52 #include <string.h>
  53 #include <ctype.h>
  54 #include <sched.h>
  55 #include <cpuid.h>
  56
  57 #include "collectd.h"
  58 #include "common.h"
  59 #include "plugin.h"
  60
  61 #define PLUGIN_NAME "turbostat"
  62
  63 static const char *proc_stat = "/proc/stat";
  64 static unsigned int skip_c0;
  65 static unsigned int skip_c1;
  66 static unsigned int do_nhm_cstates;
  67 static unsigned int do_snb_cstates;
  68 static unsigned int do_c8_c9_c10;
  69 static unsigned int do_slm_cstates;
  70 static unsigned int has_aperf;
  71 static unsigned int has_epb;
  72 static unsigned int units = 1000000000; /* Ghz etc */
  73 static unsigned int genuine_intel;
  74 static unsigned int has_invariant_tsc;
  75 static unsigned int do_nehalem_platform_info;
  76 static int do_smi;
  77 static unsigned int do_rapl;
  78 static unsigned int do_dts;
  79 static unsigned int do_ptm;
  80 static unsigned int tcc_activation_temp;
  81 static unsigned int tcc_activation_temp_override;
  82 static double rapl_power_units, rapl_energy_units, rapl_time_units;
  83 static double rapl_joule_counter_range;
  84
  85 #define RAPL_PKG                (1 << 0)
  86                                         /* 0x610 MSR_PKG_POWER_LIMIT */
  87                                         /* 0x611 MSR_PKG_ENERGY_STATUS */
  88 #define RAPL_PKG_PERF_STATUS    (1 << 1)
  89                                         /* 0x613 MSR_PKG_PERF_STATUS */
  90 #define RAPL_PKG_POWER_INFO     (1 << 2)
  91                                         /* 0x614 MSR_PKG_POWER_INFO */
  92
  93 #define RAPL_DRAM               (1 << 3)
  94                                         /* 0x618 MSR_DRAM_POWER_LIMIT */
  95                                         /* 0x619 MSR_DRAM_ENERGY_STATUS */
  96                                         /* 0x61c MSR_DRAM_POWER_INFO */
  97 #define RAPL_DRAM_PERF_STATUS   (1 << 4)
  98                                         /* 0x61b MSR_DRAM_PERF_STATUS */
  99
 100 #define RAPL_CORES              (1 << 5)
 101                                         /* 0x638 MSR_PP0_POWER_LIMIT */
 102                                         /* 0x639 MSR_PP0_ENERGY_STATUS */
 103 #define RAPL_CORE_POLICY        (1 << 6)
 104                                         /* 0x63a MSR_PP0_POLICY */
 105
 106
 107 #define RAPL_GFX                (1 << 7)
 108                                         /* 0x640 MSR_PP1_POWER_LIMIT */
 109                                         /* 0x641 MSR_PP1_ENERGY_STATUS */
 110                                         /* 0x642 MSR_PP1_POLICY */
 111 #define TJMAX_DEFAULT   100
 112
 113 int aperf_mperf_unstable;
 114 int backwards_count;
 115 char *progname;
 116
 117 cpu_set_t *cpu_present_set, *cpu_affinity_set;
 118 size_t cpu_present_setsize, cpu_affinity_setsize;
 119
 120 struct thread_data {
 121         unsigned long long tsc;
 122         unsigned long long aperf;
 123         unsigned long long mperf;
 124         unsigned long long c1;
 125         unsigned int smi_count;
 126         unsigned int cpu_id;
 127         unsigned int flags;
 128 #define CPU_IS_FIRST_THREAD_IN_CORE     0x2
 129 #define CPU_IS_FIRST_CORE_IN_PACKAGE    0x4
 130 } *thread_even, *thread_odd;
 131
 132 struct core_data {
 133         unsigned long long c3;
 134         unsigned long long c6;
 135         unsigned long long c7;
 136         unsigned int core_temp_c;
 137         unsigned int core_id;
 138 } *core_even, *core_odd;
 139
 140 struct pkg_data {
 141         unsigned long long pc2;
 142         unsigned long long pc3;
 143         unsigned long long pc6;
 144         unsigned long long pc7;
 145         unsigned long long pc8;
 146         unsigned long long pc9;
 147         unsigned long long pc10;
 148         unsigned int package_id;
 149         unsigned int energy_pkg;        /* MSR_PKG_ENERGY_STATUS */
 150         unsigned int energy_dram;       /* MSR_DRAM_ENERGY_STATUS */
 151         unsigned int energy_cores;      /* MSR_PP0_ENERGY_STATUS */
 152         unsigned int energy_gfx;        /* MSR_PP1_ENERGY_STATUS */
 153         unsigned int rapl_pkg_perf_status;      /* MSR_PKG_PERF_STATUS */
 154         unsigned int rapl_dram_perf_status;     /* MSR_DRAM_PERF_STATUS */
 155         unsigned int pkg_temp_c;
 156
 157 } *package_even, *package_odd;
 158
 159 #define ODD_COUNTERS thread_odd, core_odd, package_odd
 160 #define EVEN_COUNTERS thread_even, core_even, package_even
 161 static _Bool is_even = 1;
 162
 163 static _Bool allocated = 0;
 164 static _Bool initialized = 0;
 165
 166 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
 167         (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
 168                 topo.num_threads_per_core + \
 169                 (core_no) * topo.num_threads_per_core + (thread_no))
 170 #define GET_CORE(core_base, core_no, pkg_no) \
 171         (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
 172 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
 173
 174 struct topo_params {
 175         int num_packages;
 176         int num_cpus;
 177         int num_cores;
 178         int max_cpu_num;
 179         int num_cores_per_pkg;
 180         int num_threads_per_core;
 181 } topo;
 182
 183 struct timeval tv_even, tv_odd, tv_delta;
 184
 185 enum return_values {
 186         OK = 0,
 187         ERR_CPU_MIGRATE,
 188         ERR_MSR_IA32_APERF,
 189         ERR_MSR_IA32_MPERF,
 190         ERR_MSR_SMI_COUNT,
 191         ERR_MSR_CORE_C3_RESIDENCY,
 192         ERR_MSR_CORE_C6_RESIDENCY,
 193         ERR_MSR_CORE_C7_RESIDENCY,
 194         ERR_MSR_IA32_THERM_STATUS,
 195         ERR_MSR_PKG_C3_RESIDENCY,
 196         ERR_MSR_PKG_C6_RESIDENCY,
 197         ERR_MSR_PKG_C2_RESIDENCY,
 198         ERR_MSR_PKG_C7_RESIDENCY,
 199         ERR_MSR_PKG_C8_RESIDENCY,
 200         ERR_MSR_PKG_C9_RESIDENCY,
 201         ERR_MSR_PKG_C10_RESIDENCY,
 202         ERR_MSR_PKG_ENERGY_STATUS,
 203         ERR_MSR_PP0_ENERGY_STATUS,
 204         ERR_MSR_DRAM_ENERGY_STATUS,
 205         ERR_MSR_PP1_ENERGY_STATUS,
 206         ERR_MSR_PKG_PERF_STATUS,
 207         ERR_MSR_DRAM_PERF_STATUS,
 208         ERR_MSR_IA32_PACKAGE_THERM_STATUS,
 209         ERR_CPU_NOT_PRESENT,
 210         ERR_NO_MSR,
 211         ERR_CANT_OPEN_FILE,
 212         ERR_CANT_READ_NUMBER,
 213         ERR_CANT_READ_PROC_STAT,
 214         ERR_NO_INVARIANT_TSC,
 215         ERR_NO_APERF,
 216         ERR_CALLOC,
 217         ERR_CPU_ALLOC,
 218         ERR_NOT_ROOT,
 219 };
 220
 221 static int setup_all_buffers(void);
 222
 223 static int
 224 cpu_is_not_present(int cpu)
 225 {
 226         return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
 227 }
 228 /*
 229  * run func(thread, core, package) in topology order
 230  * skip non-present cpus
 231  */
 232
 233 static int __attribute__((warn_unused_result))
 234 for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
 235         struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
 236 {
 237         int retval, pkg_no, core_no, thread_no;
 238
 239         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
 240                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
 241                         for (thread_no = 0; thread_no <
 242                                 topo.num_threads_per_core; ++thread_no) {
 243                                 struct thread_data *t;
 244                                 struct core_data *c;
 245                                 struct pkg_data *p;
 246
 247                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
 248
 249                                 if (cpu_is_not_present(t->cpu_id))
 250                                         continue;
 251
 252                                 c = GET_CORE(core_base, core_no, pkg_no);
 253                                 p = GET_PKG(pkg_base, pkg_no);
 254
 255                                 retval = func(t, c, p);
 256                                 if (retval)
 257                                         return retval;
 258                         }
 259                 }
 260         }
 261         return 0;
 262 }
 263
 264 static int __attribute__((warn_unused_result))
 265 cpu_migrate(int cpu)
 266 {
 267         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
 268         CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
 269         if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1)
 270                 return -ERR_CPU_MIGRATE;
 271         else
 272                 return 0;
 273 }
 274
 275 static int __attribute__((warn_unused_result))
 276 get_msr(int cpu, off_t offset, unsigned long long *msr)
 277 {
 278         ssize_t retval;
 279         char pathname[32];
 280         int fd;
 281
 282         ssnprintf(pathname, 32, "/dev/cpu/%d/msr", cpu);
 283         fd = open(pathname, O_RDONLY);
 284         if (fd < 0)
 285                 return -1;
 286
 287         retval = pread(fd, msr, sizeof *msr, offset);
 288         close(fd);
 289
 290         if (retval != sizeof *msr) {
 291                 ERROR ("%s offset 0x%llx read failed\n", pathname, (unsigned long long)offset);
 292                 return -1;
 293         }
 294
 295         return 0;
 296 }
 297
 298 #define DELTA_WRAP32(new, old)                  \
 299         if (new > old) {                        \
 300                 old = new - old;                \
 301         } else {                                \
 302                 old = 0x100000000 + new - old;  \
 303         }
 304
 305 static void
 306 delta_package(struct pkg_data *new, struct pkg_data *old)
 307 {
 308         old->pc2 = new->pc2 - old->pc2;
 309         old->pc3 = new->pc3 - old->pc3;
 310         old->pc6 = new->pc6 - old->pc6;
 311         old->pc7 = new->pc7 - old->pc7;
 312         old->pc8 = new->pc8 - old->pc8;
 313         old->pc9 = new->pc9 - old->pc9;
 314         old->pc10 = new->pc10 - old->pc10;
 315         old->pkg_temp_c = new->pkg_temp_c;
 316
 317         DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
 318         DELTA_WRAP32(new->energy_cores, old->energy_cores);
 319         DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
 320         DELTA_WRAP32(new->energy_dram, old->energy_dram);
 321         DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
 322         DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
 323 }
 324
 325 static void
 326 delta_core(struct core_data *new, struct core_data *old)
 327 {
 328         old->c3 = new->c3 - old->c3;
 329         old->c6 = new->c6 - old->c6;
 330         old->c7 = new->c7 - old->c7;
 331         old->core_temp_c = new->core_temp_c;
 332 }
 333
 334 /*
 335  * old = new - old
 336  */
 337 static int __attribute__((warn_unused_result))
 338 delta_thread(struct thread_data *new, struct thread_data *old,
 339         struct core_data *core_delta)
 340 {
 341         old->tsc = new->tsc - old->tsc;
 342
 343         /* check for TSC < 1 Mcycles over interval */
 344         if (old->tsc < (1000 * 1000)) {
 345                 WARNING("Insanely slow TSC rate, TSC stops in idle?\n"
 346                         "You can disable all c-states by booting with \"idle=poll\"\n"
 347                         "or just the deep ones with \"processor.max_cstate=1\"");
 348                 return -1;
 349         }
 350
 351         old->c1 = new->c1 - old->c1;
 352
 353         if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
 354                 old->aperf = new->aperf - old->aperf;
 355                 old->mperf = new->mperf - old->mperf;
 356         } else {
 357
 358                 if (!aperf_mperf_unstable) {
 359                         WARNING("%s: APERF or MPERF went backwards *\n", progname);
 360                         WARNING("* Frequency results do not cover entire interval *\n");
 361                         WARNING("* fix this by running Linux-2.6.30 or later *\n");
 362
 363                         aperf_mperf_unstable = 1;
 364                 }
 365                 /*
 366                  * mperf delta is likely a huge "positive" number
 367                  * can not use it for calculating c0 time
 368                  */
 369                 skip_c0 = 1;
 370                 skip_c1 = 1;
 371         }
 372
 373
 374         /*
 375          * As counter collection is not atomic,
 376          * it is possible for mperf's non-halted cycles + idle states
 377          * to exceed TSC's all cycles: show c1 = 0% in that case.
 378          */
 379         if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
 380                 old->c1 = 0;
 381         else {
 382                 /* normal case, derive c1 */
 383                 old->c1 = old->tsc - old->mperf - core_delta->c3
 384                         - core_delta->c6 - core_delta->c7;
 385         }
 386
 387         if (old->mperf == 0) {
 388                 WARNING("cpu%d MPERF 0!\n", old->cpu_id);
 389                 old->mperf = 1; /* divide by 0 protection */
 390         }
 391
 392         if (do_smi)
 393                 old->smi_count = new->smi_count - old->smi_count;
 394
 395         return 0;
 396 }
 397
 398 static int __attribute__((warn_unused_result))
 399 delta_cpu(struct thread_data *t, struct core_data *c,
 400         struct pkg_data *p, struct thread_data *t2,
 401         struct core_data *c2, struct pkg_data *p2)
 402 {
 403         int ret;
 404
 405         /* calculate core delta only for 1st thread in core */
 406         if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
 407                 delta_core(c, c2);
 408
 409         /* always calculate thread delta */
 410         ret = delta_thread(t, t2, c2);  /* c2 is core delta */
 411         if (ret != 0)
 412                 return ret;
 413
 414         /* calculate package delta only for 1st core in package */
 415         if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
 416                 delta_package(p, p2);
 417
 418         return 0;
 419 }
 420
 421 static unsigned long long
 422 rdtsc(void)
 423 {
 424         unsigned int low, high;
 425
 426         asm volatile("rdtsc" : "=a" (low), "=d" (high));
 427
 428         return low | ((unsigned long long)high) << 32;
 429 }
 430
 431
 432 /*
 433  * get_counters(...)
 434  * migrate to cpu
 435  * acquire and record local counters for that cpu
 436  */
 437 static int __attribute__((warn_unused_result))
 438 get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 439 {
 440         int cpu = t->cpu_id;
 441         unsigned long long msr;
 442
 443         if (cpu_migrate(cpu)) {
 444                 WARNING("Could not migrate to CPU %d\n", cpu);
 445                 return -ERR_CPU_MIGRATE;
 446         }
 447
 448         t->tsc = rdtsc();       /* we are running on local CPU of interest */
 449
 450         if (has_aperf) {
 451                 if (get_msr(cpu, MSR_IA32_APERF, &t->aperf))
 452                         return -ERR_MSR_IA32_APERF;
 453                 if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf))
 454                         return -ERR_MSR_IA32_MPERF;
 455         }
 456
 457         if (do_smi) {
 458                 if (get_msr(cpu, MSR_SMI_COUNT, &msr))
 459                         return -ERR_MSR_SMI_COUNT;
 460                 t->smi_count = msr & 0xFFFFFFFF;
 461         }
 462
 463         /* collect core counters only for 1st thread in core */
 464         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
 465                 return 0;
 466
 467         if (do_nhm_cstates && !do_slm_cstates) {
 468                 if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3))
 469                         return -ERR_MSR_CORE_C3_RESIDENCY;
 470         }
 471
 472         if (do_nhm_cstates) {
 473                 if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6))
 474                         return -ERR_MSR_CORE_C6_RESIDENCY;
 475         }
 476
 477         if (do_snb_cstates)
 478                 if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7))
 479                         return -ERR_MSR_CORE_C7_RESIDENCY;
 480
 481         if (do_dts) {
 482                 if (get_msr(cpu, MSR_IA32_THERM_STATUS, &msr))
 483                         return -ERR_MSR_IA32_THERM_STATUS;
 484                 c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
 485         }
 486
 487
 488         /* collect package counters only for 1st core in package */
 489         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
 490                 return 0;
 491
 492         if (do_nhm_cstates && !do_slm_cstates) {
 493                 if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3))
 494                         return -ERR_MSR_PKG_C3_RESIDENCY;
 495                 if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6))
 496                         return -ERR_MSR_PKG_C6_RESIDENCY;
 497         }
 498         if (do_snb_cstates) {
 499                 if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2))
 500                         return -ERR_MSR_PKG_C2_RESIDENCY;
 501                 if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7))
 502                         return -ERR_MSR_PKG_C7_RESIDENCY;
 503         }
 504         if (do_c8_c9_c10) {
 505                 if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8))
 506                         return -ERR_MSR_PKG_C8_RESIDENCY;
 507                 if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9))
 508                         return -ERR_MSR_PKG_C9_RESIDENCY;
 509                 if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10))
 510                         return -ERR_MSR_PKG_C10_RESIDENCY;
 511         }
 512         if (do_rapl & RAPL_PKG) {
 513                 if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
 514                         return -ERR_MSR_PKG_ENERGY_STATUS;
 515                 p->energy_pkg = msr & 0xFFFFFFFF;
 516         }
 517         if (do_rapl & RAPL_CORES) {
 518                 if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
 519                         return MSR_PP0_ENERGY_STATUS;
 520                 p->energy_cores = msr & 0xFFFFFFFF;
 521         }
 522         if (do_rapl & RAPL_DRAM) {
 523                 if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
 524                         return -ERR_MSR_DRAM_ENERGY_STATUS;
 525                 p->energy_dram = msr & 0xFFFFFFFF;
 526         }
 527         if (do_rapl & RAPL_GFX) {
 528                 if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
 529                         return -ERR_MSR_PP1_ENERGY_STATUS;
 530                 p->energy_gfx = msr & 0xFFFFFFFF;
 531         }
 532         if (do_rapl & RAPL_PKG_PERF_STATUS) {
 533                 if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
 534                         return -ERR_MSR_PKG_PERF_STATUS;
 535                 p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
 536         }
 537         if (do_rapl & RAPL_DRAM_PERF_STATUS) {
 538                 if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
 539                         return -ERR_MSR_DRAM_PERF_STATUS;
 540                 p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
 541         }
 542         if (do_ptm) {
 543                 if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr))
 544                         return -ERR_MSR_IA32_PACKAGE_THERM_STATUS;
 545                 p->pkg_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
 546         }
 547         return 0;
 548 }
 549
 550 static void
 551 free_all_buffers(void)
 552 {
 553         allocated = 0;
 554         initialized = 0;
 555
 556         CPU_FREE(cpu_present_set);
 557         cpu_present_set = NULL;
 558         cpu_present_set = 0;
 559
 560         CPU_FREE(cpu_affinity_set);
 561         cpu_affinity_set = NULL;
 562         cpu_affinity_setsize = 0;
 563
 564         free(thread_even);
 565         free(core_even);
 566         free(package_even);
 567
 568         thread_even = NULL;
 569         core_even = NULL;
 570         package_even = NULL;
 571
 572         free(thread_odd);
 573         free(core_odd);
 574         free(package_odd);
 575
 576         thread_odd = NULL;
 577         core_odd = NULL;
 578         package_odd = NULL;
 579 }
 580
 581 /*
 582  * Parse a file containing a single int.
 583  */
 584 static int __attribute__ ((format(printf,1,2)))
 585 parse_int_file(const char *fmt, ...)
 586 {
 587         va_list args;
 588         char path[PATH_MAX];
 589         FILE *filep;
 590         int value;
 591
 592         va_start(args, fmt);
 593         vsnprintf(path, sizeof(path), fmt, args);
 594         va_end(args);
 595         filep = fopen(path, "r");
 596         if (!filep) {
 597                 ERROR("%s: open failed", path);
 598                 return -ERR_CANT_OPEN_FILE;
 599         }
 600         if (fscanf(filep, "%d", &value) != 1) {
 601                 ERROR("%s: failed to parse number from file", path);
 602                 return -ERR_CANT_READ_NUMBER;
 603         }
 604         fclose(filep);
 605         return value;
 606 }
 607
 608 /*
 609  * cpu_is_first_sibling_in_core(cpu)
 610  * return 1 if given CPU is 1st HT sibling in the core
 611  */
 612 static int
 613 cpu_is_first_sibling_in_core(int cpu)
 614 {
 615         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
 616 }
 617
 618 /*
 619  * cpu_is_first_core_in_package(cpu)
 620  * return 1 if given CPU is 1st core in package
 621  */
 622 static int
 623 cpu_is_first_core_in_package(int cpu)
 624 {
 625         return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
 626 }
 627
 628 static int
 629 get_physical_package_id(int cpu)
 630 {
 631         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
 632 }
 633
 634 static int
 635 get_core_id(int cpu)
 636 {
 637         return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
 638 }
 639
 640 static int
 641 get_num_ht_siblings(int cpu)
 642 {
 643         char path[80];
 644         FILE *filep;
 645         int sib1, sib2;
 646         int matches;
 647         char character;
 648
 649         ssnprintf(path, 80, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
 650         filep = fopen(path, "r");
 651         if (!filep) {
 652                 ERROR("%s: open failed", path);
 653                 return -ERR_CANT_OPEN_FILE;
 654         }
 655         /*
 656          * file format:
 657          * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
 658          * otherwinse 1 sibling (self).
 659          */
 660         matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
 661
 662         fclose(filep);
 663
 664         if (matches == 3)
 665                 return 2;
 666         else
 667                 return 1;
 668 }
 669
 670 /*
 671  * run func(thread, core, package) in topology order
 672  * skip non-present cpus
 673  */
 674
 675
 676 static int __attribute__((warn_unused_result))
 677 for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
 678         struct pkg_data *, struct thread_data *, struct core_data *,
 679         struct pkg_data *), struct thread_data *thread_base,
 680         struct core_data *core_base, struct pkg_data *pkg_base,
 681         struct thread_data *thread_base2, struct core_data *core_base2,
 682         struct pkg_data *pkg_base2)
 683 {
 684         int retval, pkg_no, core_no, thread_no;
 685
 686         for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
 687                 for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
 688                         for (thread_no = 0; thread_no <
 689                                 topo.num_threads_per_core; ++thread_no) {
 690                                 struct thread_data *t, *t2;
 691                                 struct core_data *c, *c2;
 692                                 struct pkg_data *p, *p2;
 693
 694                                 t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
 695
 696                                 if (cpu_is_not_present(t->cpu_id))
 697                                         continue;
 698
 699                                 t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
 700
 701                                 c = GET_CORE(core_base, core_no, pkg_no);
 702                                 c2 = GET_CORE(core_base2, core_no, pkg_no);
 703
 704                                 p = GET_PKG(pkg_base, pkg_no);
 705                                 p2 = GET_PKG(pkg_base2, pkg_no);
 706
 707                                 retval = func(t, c, p, t2, c2, p2);
 708                                 if (retval)
 709                                         return retval;
 710                         }
 711                 }
 712         }
 713         return 0;
 714 }
 715
 716 /*
 717  * run func(cpu) on every cpu in /proc/stat
 718  * return max_cpu number
 719  */
 720 static int __attribute__((warn_unused_result))
 721 for_all_proc_cpus(int (func)(int))
 722 {
 723         FILE *fp;
 724         int cpu_num;
 725         int retval;
 726
 727         fp = fopen(proc_stat, "r");
 728         if (!fp) {
 729                 ERROR("%s: open failed", proc_stat);
 730                 return -ERR_CANT_OPEN_FILE;
 731         }
 732
 733         retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
 734         if (retval != 0) {
 735                 ERROR("%s: failed to parse format", proc_stat);
 736                 return -ERR_CANT_READ_PROC_STAT;
 737         }
 738
 739         while (1) {
 740                 retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
 741                 if (retval != 1)
 742                         break;
 743
 744                 retval = func(cpu_num);
 745                 if (retval) {
 746                         fclose(fp);
 747                         return(retval);
 748                 }
 749         }
 750         fclose(fp);
 751         return 0;
 752 }
 753
 754 /*
 755  * count_cpus()
 756  * remember the last one seen, it will be the max
 757  */
 758 static int
 759 count_cpus(int cpu)
 760 {
 761         if (topo.max_cpu_num < cpu)
 762                 topo.max_cpu_num = cpu;
 763
 764         topo.num_cpus += 1;
 765         return 0;
 766 }
 767 static int
 768 mark_cpu_present(int cpu)
 769 {
 770         CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
 771         return 0;
 772 }
 773
 774
 775 static void
 776 turbostat_submit (const char *plugin_instance,
 777         const char *type, const char *type_instance,
 778         gauge_t value)
 779 {
 780         value_list_t vl = VALUE_LIST_INIT;
 781         value_t v;
 782
 783         v.gauge = value;
 784         vl.values = &v;
 785         vl.values_len = 1;
 786         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 787         sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
 788         if (plugin_instance != NULL)
 789                 sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 790         sstrncpy (vl.type, type, sizeof (vl.type));
 791         if (type_instance != NULL)
 792                 sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
 793
 794         plugin_dispatch_values (&vl);
 795 }
 796
 797 /*
 798  * column formatting convention & formats
 799  * package: "pk" 2 columns %2d
 800  * core: "cor" 3 columns %3d
 801  * CPU: "CPU" 3 columns %3d
 802  * Pkg_W: %6.2
 803  * Cor_W: %6.2
 804  * GFX_W: %5.2
 805  * RAM_W: %5.2
 806  * GHz: "GHz" 3 columns %3.2
 807  * TSC: "TSC" 3 columns %3.2
 808  * SMI: "SMI" 4 columns %4d
 809  * percentage " %pc3" %6.2
 810  * Perf Status percentage: %5.2
 811  * "CTMP" 4 columns %4d
 812  */
 813 #define NAME_LEN 12
 814 static int
 815 submit_counters(struct thread_data *t, struct core_data *c,
 816         struct pkg_data *p)
 817 {
 818         char name[NAME_LEN];
 819         double interval_float;
 820
 821         interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
 822
 823         snprintf(name, NAME_LEN, "cpu%02d", t->cpu_id);
 824
 825         if (do_nhm_cstates) {
 826                 if (!skip_c0)
 827                         turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
 828                 if (!skip_c1)
 829                         turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
 830         }
 831
 832         /* GHz */
 833         if (has_aperf && ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc))))
 834                 turbostat_submit(NULL, "frequency", name, 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float);
 835
 836         /* SMI */
 837         if (do_smi)
 838                 turbostat_submit(NULL, "current", name, t->smi_count);
 839
 840         /* print per-core data only for 1st thread in core */
 841         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
 842                 goto done;
 843
 844         snprintf(name, NAME_LEN, "core%02d", c->core_id);
 845
 846         if (do_nhm_cstates && !do_slm_cstates)
 847                 turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
 848         if (do_nhm_cstates)
 849                 turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
 850         if (do_snb_cstates)
 851                 turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
 852
 853         if (do_dts)
 854                 turbostat_submit(NULL, "temperature", name, c->core_temp_c);
 855
 856         /* print per-package data only for 1st core in package */
 857         if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
 858                 goto done;
 859
 860         snprintf(name, NAME_LEN, "pkg%02d", p->package_id);
 861
 862         if (do_ptm)
 863                 turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
 864
 865         if (do_snb_cstates)
 866                 turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
 867         if (do_nhm_cstates && !do_slm_cstates)
 868                 turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
 869         if (do_nhm_cstates && !do_slm_cstates)
 870                 turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
 871         if (do_snb_cstates)
 872                 turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
 873         if (do_c8_c9_c10) {
 874                 turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
 875                 turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
 876                 turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
 877         }
 878
 879         if (do_rapl) {
 880                 if (do_rapl & RAPL_PKG)
 881                         turbostat_submit(name, "power", "Pkg_W", p->energy_pkg * rapl_energy_units / interval_float);
 882                 if (do_rapl & RAPL_CORES)
 883                         turbostat_submit(name, "power", "Cor_W", p->energy_cores * rapl_energy_units / interval_float);
 884                 if (do_rapl & RAPL_GFX)
 885                         turbostat_submit(name, "power", "GFX_W", p->energy_gfx * rapl_energy_units / interval_float);
 886                 if (do_rapl & RAPL_DRAM)
 887                         turbostat_submit(name, "power", "RAM_W", p->energy_dram * rapl_energy_units / interval_float);
 888         }
 889 done:
 890         return 0;
 891 }
 892
 893 static int
 894 turbostat_read(user_data_t * not_used)
 895 {
 896         int ret;
 897
 898         if (!allocated) {
 899                 if ((ret = setup_all_buffers()) < 0)
 900                         return ret;
 901         }
 902
 903         if (for_all_proc_cpus(cpu_is_not_present)) {
 904                 free_all_buffers();
 905                 if ((ret = setup_all_buffers()) < 0)
 906                         return ret;
 907                 if (for_all_proc_cpus(cpu_is_not_present))
 908                         return -ERR_CPU_NOT_PRESENT;
 909         }
 910
 911         if (!initialized) {
 912                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
 913                         return ret;
 914                 gettimeofday(&tv_even, (struct timezone *)NULL);
 915                 is_even = 1;
 916                 initialized = 1;
 917                 return 0;
 918         }
 919
 920         if (is_even) {
 921                 if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
 922                         return ret;
 923                 gettimeofday(&tv_odd, (struct timezone *)NULL);
 924                 is_even = 0;
 925                 timersub(&tv_odd, &tv_even, &tv_delta);
 926                 if ((ret = for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) < 0)
 927                         return ret;
 928                 if ((ret = for_all_cpus(submit_counters, EVEN_COUNTERS)) < 0)
 929                         return ret;
 930         } else {
 931                 if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
 932                         return ret;
 933                 gettimeofday(&tv_even, (struct timezone *)NULL);
 934                 is_even = 1;
 935                 timersub(&tv_even, &tv_odd, &tv_delta);
 936                 if ((ret = for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) < 0)
 937                         return ret;
 938                 if ((ret = for_all_cpus(submit_counters, ODD_COUNTERS)) < 0)
 939                         return ret;
 940         }
 941         return 0;
 942 }
 943
 944 static int __attribute__((warn_unused_result))
 945 check_dev_msr()
 946 {
 947         struct stat sb;
 948
 949         if (stat("/dev/cpu/0/msr", &sb)) {
 950                 ERROR("no /dev/cpu/0/msr\n"
 951                         "Try \"# modprobe msr\"");
 952                 return -ERR_NO_MSR;
 953         }
 954         return 0;
 955 }
 956
 957 static int __attribute__((warn_unused_result))
 958 check_super_user()
 959 {
 960         if (getuid() != 0) {
 961                 ERROR("must be root");
 962                 return -ERR_NOT_ROOT;
 963         }
 964         return 0;
 965 }
 966
 967
 968 #define RAPL_POWER_GRANULARITY  0x7FFF  /* 15 bit power granularity */
 969 #define RAPL_TIME_GRANULARITY   0x3F /* 6 bit time granularity */
 970
 971 static double
 972 get_tdp(unsigned int model)
 973 {
 974         unsigned long long msr;
 975
 976         if (do_rapl & RAPL_PKG_POWER_INFO)
 977                 if (!get_msr(0, MSR_PKG_POWER_INFO, &msr))
 978                         return ((msr >> 0) & RAPL_POWER_GRANULARITY) * rapl_power_units;
 979
 980         switch (model) {
 981         case 0x37:
 982         case 0x4D:
 983                 return 30.0;
 984         default:
 985                 return 135.0;
 986         }
 987 }
 988
 989
 990 /*
 991  * rapl_probe()
 992  *
 993  * sets do_rapl, rapl_power_units, rapl_energy_units, rapl_time_units
 994  */
 995 static void
 996 rapl_probe(unsigned int family, unsigned int model)
 997 {
 998         unsigned long long msr;
 999         unsigned int time_unit;
1000         double tdp;
1001
1002         if (!genuine_intel)
1003                 return;
1004
1005         if (family != 6)
1006                 return;
1007
1008         switch (model) {
1009         case 0x2A:
1010         case 0x3A:
1011         case 0x3C:      /* HSW */
1012         case 0x45:      /* HSW */
1013         case 0x46:      /* HSW */
1014                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_GFX | RAPL_PKG_POWER_INFO;
1015                 break;
1016         case 0x3F:      /* HSX */
1017                 do_rapl = RAPL_PKG | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO;
1018                 break;
1019         case 0x2D:
1020         case 0x3E:
1021                 do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_DRAM | RAPL_PKG_PERF_STATUS | RAPL_DRAM_PERF_STATUS | RAPL_PKG_POWER_INFO;
1022                 break;
1023         case 0x37:      /* BYT */
1024         case 0x4D:      /* AVN */
1025                 do_rapl = RAPL_PKG | RAPL_CORES ;
1026                 break;
1027         default:
1028                 return;
1029         }
1030
1031         /* units on package 0, verify later other packages match */
1032         if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
1033                 return;
1034
1035         rapl_power_units = 1.0 / (1 << (msr & 0xF));
1036         if (model == 0x37)
1037                 rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
1038         else
1039                 rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
1040
1041         time_unit = msr >> 16 & 0xF;
1042         if (time_unit == 0)
1043                 time_unit = 0xA;
1044
1045         rapl_time_units = 1.0 / (1 << (time_unit));
1046
1047         tdp = get_tdp(model);
1048
1049         rapl_joule_counter_range = 0xFFFFFFFF * rapl_energy_units / tdp;
1050 //      if (verbose)
1051 //              fprintf(stderr, "RAPL: %.0f sec. Joule Counter Range, at %.0f Watts\n", rapl_joule_counter_range, tdp);
1052
1053         return;
1054 }
1055
1056 static int
1057 is_snb(unsigned int family, unsigned int model)
1058 {
1059         if (!genuine_intel)
1060                 return 0;
1061
1062         switch (model) {
1063         case 0x2A:
1064         case 0x2D:
1065         case 0x3A:      /* IVB */
1066         case 0x3E:      /* IVB Xeon */
1067         case 0x3C:      /* HSW */
1068         case 0x3F:      /* HSW */
1069         case 0x45:      /* HSW */
1070         case 0x46:      /* HSW */
1071                 return 1;
1072         }
1073         return 0;
1074 }
1075
1076 static int
1077 has_c8_c9_c10(unsigned int family, unsigned int model)
1078 {
1079         if (!genuine_intel)
1080                 return 0;
1081
1082         switch (model) {
1083         case 0x45:
1084                 return 1;
1085         }
1086         return 0;
1087 }
1088
1089
1090 static int
1091 is_slm(unsigned int family, unsigned int model)
1092 {
1093         if (!genuine_intel)
1094                 return 0;
1095         switch (model) {
1096         case 0x37:      /* BYT */
1097         case 0x4D:      /* AVN */
1098                 return 1;
1099         }
1100         return 0;
1101 }
1102
1103 /*
1104  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
1105  * the Thermal Control Circuit (TCC) activates.
1106  * This is usually equal to tjMax.
1107  *
1108  * Older processors do not have this MSR, so there we guess,
1109  * but also allow cmdline over-ride with -T.
1110  *
1111  * Several MSR temperature values are in units of degrees-C
1112  * below this value, including the Digital Thermal Sensor (DTS),
1113  * Package Thermal Management Sensor (PTM), and thermal event thresholds.
1114  */
1115 static int __attribute__((warn_unused_result))
1116 set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
1117 {
1118         unsigned long long msr;
1119         unsigned int target_c_local;
1120         int cpu;
1121
1122         /* tcc_activation_temp is used only for dts or ptm */
1123         if (!(do_dts || do_ptm))
1124                 return 0;
1125
1126         /* this is a per-package concept */
1127         if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
1128                 return 0;
1129
1130         cpu = t->cpu_id;
1131         if (cpu_migrate(cpu)) {
1132                 ERROR("Could not migrate to CPU %d\n", cpu);
1133                 return -ERR_CPU_MIGRATE;
1134         }
1135
1136         if (tcc_activation_temp_override != 0) {
1137                 tcc_activation_temp = tcc_activation_temp_override;
1138                 ERROR("cpu%d: Using cmdline TCC Target (%d C)\n",
1139                         cpu, tcc_activation_temp);
1140                 return 0;
1141         }
1142
1143         /* Temperature Target MSR is Nehalem and newer only */
1144         if (!do_nehalem_platform_info)
1145                 goto guess;
1146
1147         if (get_msr(0, MSR_IA32_TEMPERATURE_TARGET, &msr))
1148                 goto guess;
1149
1150         target_c_local = (msr >> 16) & 0x7F;
1151
1152         if (target_c_local < 85 || target_c_local > 127)
1153                 goto guess;
1154
1155         tcc_activation_temp = target_c_local;
1156
1157         return 0;
1158
1159 guess:
1160         tcc_activation_temp = TJMAX_DEFAULT;
1161         WARNING("cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
1162                 cpu, tcc_activation_temp);
1163
1164         return 0;
1165 }
1166
1167 static int __attribute__((warn_unused_result))
1168 check_cpuid()
1169 {
1170         unsigned int eax, ebx, ecx, edx, max_level;
1171         unsigned int fms, family, model;
1172
1173         eax = ebx = ecx = edx = 0;
1174
1175         __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
1176
1177         if (ebx == 0x756e6547 && edx == 0x49656e69 && ecx == 0x6c65746e)
1178                 genuine_intel = 1;
1179
1180         fms = 0;
1181         __get_cpuid(1, &fms, &ebx, &ecx, &edx);
1182         family = (fms >> 8) & 0xf;
1183         model = (fms >> 4) & 0xf;
1184         if (family == 6 || family == 0xf)
1185                 model += ((fms >> 16) & 0xf) << 4;
1186
1187         if (!(edx & (1 << 5))) {
1188                 ERROR("CPUID: no MSR");
1189                 return -ERR_NO_MSR;
1190         }
1191
1192         /*
1193          * check max extended function levels of CPUID.
1194          * This is needed to check for invariant TSC.
1195          * This check is valid for both Intel and AMD.
1196          */
1197         ebx = ecx = edx = 0;
1198         __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
1199
1200         if (max_level < 0x80000007) {
1201                 ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level);
1202                 return -ERR_NO_INVARIANT_TSC;
1203         }
1204
1205         /*
1206          * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
1207          * this check is valid for both Intel and AMD
1208          */
1209         __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1210         has_invariant_tsc = edx & (1 << 8);
1211
1212         if (!has_invariant_tsc) {
1213                 ERROR("No invariant TSC");
1214                 return -ERR_NO_INVARIANT_TSC;
1215         }
1216
1217         /*
1218          * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
1219          * this check is valid for both Intel and AMD
1220          */
1221
1222         __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
1223         has_aperf = ecx & (1 << 0);
1224         do_dts = eax & (1 << 0);
1225         do_ptm = eax & (1 << 6);
1226         has_epb = ecx & (1 << 3);
1227
1228         if (!has_aperf) {
1229                 ERROR("No APERF");
1230                 return -ERR_NO_APERF;
1231         }
1232
1233         do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
1234         do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
1235         do_smi = do_nhm_cstates;
1236         do_snb_cstates = is_snb(family, model);
1237         do_c8_c9_c10 = has_c8_c9_c10(family, model);
1238         do_slm_cstates = is_slm(family, model);
1239
1240         rapl_probe(family, model);
1241
1242         return 0;
1243 }
1244
1245
1246
1247 static int __attribute__((warn_unused_result))
1248 topology_probe()
1249 {
1250         int i;
1251         int ret;
1252         int max_core_id = 0;
1253         int max_package_id = 0;
1254         int max_siblings = 0;
1255         struct cpu_topology {
1256                 int core_id;
1257                 int physical_package_id;
1258         } *cpus;
1259
1260         /* Initialize num_cpus, max_cpu_num */
1261         topo.num_cpus = 0;
1262         topo.max_cpu_num = 0;
1263         ret = for_all_proc_cpus(count_cpus);
1264         if (ret < 0)
1265                 return ret;
1266
1267         DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
1268
1269         cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
1270         if (cpus == NULL) {
1271                 ERROR("calloc cpus");
1272                 return -ERR_CALLOC;
1273         }
1274
1275         /*
1276          * Allocate and initialize cpu_present_set
1277          */
1278         cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
1279         if (cpu_present_set == NULL) {
1280                 free(cpus);
1281                 ERROR("CPU_ALLOC");
1282                 return -ERR_CPU_ALLOC;
1283         }
1284         cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1285         CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
1286         ret = for_all_proc_cpus(mark_cpu_present);
1287         if (ret < 0) {
1288                 free(cpus);
1289                 return ret;
1290         }
1291
1292         /*
1293          * Allocate and initialize cpu_affinity_set
1294          */
1295         cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
1296         if (cpu_affinity_set == NULL) {
1297                 free(cpus);
1298                 ERROR("CPU_ALLOC");
1299                 return -ERR_CPU_ALLOC;
1300         }
1301         cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
1302         CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
1303
1304
1305         /*
1306          * For online cpus
1307          * find max_core_id, max_package_id
1308          */
1309         for (i = 0; i <= topo.max_cpu_num; ++i) {
1310                 int siblings;
1311
1312                 if (cpu_is_not_present(i)) {
1313                         //if (verbose > 1)
1314                                 fprintf(stderr, "cpu%d NOT PRESENT\n", i);
1315                         continue;
1316                 }
1317                 cpus[i].core_id = get_core_id(i);
1318                 if (cpus[i].core_id < 0)
1319                         return cpus[i].core_id;
1320                 if (cpus[i].core_id > max_core_id)
1321                         max_core_id = cpus[i].core_id;
1322
1323                 cpus[i].physical_package_id = get_physical_package_id(i);
1324                 if (cpus[i].physical_package_id < 0)
1325                         return cpus[i].physical_package_id;
1326                 if (cpus[i].physical_package_id > max_package_id)
1327                         max_package_id = cpus[i].physical_package_id;
1328
1329                 siblings = get_num_ht_siblings(i);
1330                 if (siblings < 0)
1331                         return siblings;
1332                 if (siblings > max_siblings)
1333                         max_siblings = siblings;
1334                 DEBUG("cpu %d pkg %d core %d\n",
1335                         i, cpus[i].physical_package_id, cpus[i].core_id);
1336         }
1337         topo.num_cores_per_pkg = max_core_id + 1;
1338         DEBUG("max_core_id %d, sizing for %d cores per package\n",
1339                 max_core_id, topo.num_cores_per_pkg);
1340
1341         topo.num_packages = max_package_id + 1;
1342         DEBUG("max_package_id %d, sizing for %d packages\n",
1343                 max_package_id, topo.num_packages);
1344
1345         topo.num_threads_per_core = max_siblings;
1346         DEBUG("max_siblings %d\n", max_siblings);
1347
1348         free(cpus);
1349         return 0;
1350 }
1351
1352 static int
1353 allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
1354 {
1355         int i;
1356
1357         *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
1358                 topo.num_packages, sizeof(struct thread_data));
1359         if (*t == NULL)
1360                 goto error;
1361
1362         for (i = 0; i < topo.num_threads_per_core *
1363                 topo.num_cores_per_pkg * topo.num_packages; i++)
1364                 (*t)[i].cpu_id = -1;
1365
1366         *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
1367                 sizeof(struct core_data));
1368         if (*c == NULL)
1369                 goto error;
1370
1371         for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
1372                 (*c)[i].core_id = -1;
1373
1374         *p = calloc(topo.num_packages, sizeof(struct pkg_data));
1375         if (*p == NULL)
1376                 goto error;
1377
1378         for (i = 0; i < topo.num_packages; i++)
1379                 (*p)[i].package_id = i;
1380
1381         return 0;
1382 error:
1383         ERROR("calloc counters");
1384         return -ERR_CALLOC;
1385 }
1386 /*
1387  * init_counter()
1388  *
1389  * set cpu_id, core_num, pkg_num
1390  * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
1391  *
1392  * increment topo.num_cores when 1st core in pkg seen
1393  */
1394 static int
1395 init_counter(struct thread_data *thread_base, struct core_data *core_base,
1396         struct pkg_data *pkg_base, int thread_num, int core_num,
1397         int pkg_num, int cpu_id)
1398 {
1399         int ret;
1400         struct thread_data *t;
1401         struct core_data *c;
1402         struct pkg_data *p;
1403
1404         t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
1405         c = GET_CORE(core_base, core_num, pkg_num);
1406         p = GET_PKG(pkg_base, pkg_num);
1407
1408         t->cpu_id = cpu_id;
1409         if (thread_num == 0) {
1410                 t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
1411                 if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
1412                         return ret;
1413                 } else if (ret != 0) {
1414                         t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
1415                 }
1416         }
1417
1418         c->core_id = core_num;
1419         p->package_id = pkg_num;
1420
1421         return 0;
1422 }
1423
1424
1425 static int
1426 initialize_counters(int cpu_id)
1427 {
1428         int my_thread_id, my_core_id, my_package_id;
1429         int ret;
1430
1431         my_package_id = get_physical_package_id(cpu_id);
1432         if (my_package_id < 0)
1433                 return my_package_id;
1434         my_core_id = get_core_id(cpu_id);
1435         if (my_core_id < 0)
1436                 return my_core_id;
1437
1438         if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
1439                 return ret;
1440         } else if (ret != 0) {
1441                 my_thread_id = 0;
1442                 topo.num_cores++;
1443         } else {
1444                 my_thread_id = 1;
1445         }
1446
1447         ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1448         if (ret < 0)
1449                 return ret;
1450         ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
1451         if (ret < 0)
1452                 return ret;
1453         return 0;
1454 }
1455
1456 #define DO_OR_GOTO_ERR(something) \
1457 do {                         \
1458         ret = (something);     \
1459         if (ret < 0)         \
1460                 goto err;    \
1461 } while (0)
1462
1463 static int setup_all_buffers(void)
1464 {
1465         int ret;
1466
1467         DO_OR_GOTO_ERR(topology_probe());
1468         DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
1469         DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
1470         DO_OR_GOTO_ERR(for_all_proc_cpus(initialize_counters));
1471
1472         allocated = 1;
1473         return 0;
1474 err:
1475         free_all_buffers();
1476         return ret;
1477 }
1478
1479 static int
1480 turbostat_init(void)
1481 {
1482         int ret;
1483
1484         DO_OR_GOTO_ERR(check_cpuid());
1485         DO_OR_GOTO_ERR(check_dev_msr());
1486         DO_OR_GOTO_ERR(check_super_user());
1487         DO_OR_GOTO_ERR(setup_all_buffers());
1488         DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
1489
1490         plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, NULL, NULL);
1491
1492         return 0;
1493 err:
1494         free_all_buffers();
1495         return ret;
1496 }
1497
1498 void module_register(void);
1499 void module_register(void)
1500 {
1501         plugin_register_init(PLUGIN_NAME, turbostat_init);
1502 }