Changes for power reporting on SKX and CLX.

[collectd.git] / src / turbostat.c

/*
 * turbostat -- Log CPU frequency and C-state residency
 * on modern Intel turbo-capable processors for collectd.
 *
 * Based on the 'turbostat' tool of the Linux kernel, found at
 * linux/tools/power/x86/turbostat/turbostat.c:
 * ----
 * Copyright (c) 2013 Intel Corporation.
 * Len Brown <len.brown@intel.com>
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
 * ----
 * Ported to collectd by Vincent Brillault <git@lerya.net>
 */

/*
 * _GNU_SOURCE is required because of the following functions:
 * - CPU_ISSET_S
 * - CPU_ZERO_S
 * - CPU_SET_S
 * - CPU_FREE
 * - CPU_ALLOC
 * - CPU_ALLOC_SIZE
 */
#define _GNU_SOURCE

#include "collectd.h"

#include "plugin.h"
#include "utils/common/common.h"
#include "utils_time.h"

#include "msr-index.h"
#include <cpuid.h>
#ifdef HAVE_SYS_CAPABILITY_H
#include <sys/capability.h>
#endif /* HAVE_SYS_CAPABILITY_H */

#define PLUGIN_NAME "turbostat"

typedef enum affinity_policy_enum {
  policy_restore_affinity, /* restore cpu affinity to whatever it was before */
  policy_allcpus_affinity  /* do not restore affinity, set to all cpus */
} affinity_policy_t;

/* the default is to set cpu affinity to all cpus */
static affinity_policy_t affinity_policy = policy_allcpus_affinity;

/*
 * This tool uses the Model-Specific Registers (MSRs) present on Intel
 * processors.
 * The general description each of these registers, depending on the
 * architecture,
 * can be found in the Intel® 64 and IA-32 Architectures Software Developer
 * Manual,
 * Volume 3 Chapter 35.
 */

/*
 * If set, aperf_mperf_unstable disables a/mperf based stats.
 * This includes: C0 & C1 states, frequency
 *
 * This value is automatically set if mperf or aperf go backward
 */
static bool aperf_mperf_unstable;

/*
 * If set, use kernel logical core numbering for all "per core" metrics.
 */
static bool config_lcn;

/*
 * Bitmask of the list of core C states supported by the processor.
 * Currently supported C-states (by this plugin): 3, 6, 7
 */
static unsigned int do_core_cstate;
static unsigned int config_core_cstate;
static bool apply_config_core_cstate;

/*
 * Bitmask of the list of pacages C states supported by the processor.
 * Currently supported C-states (by this plugin): 2, 3, 6, 7, 8, 9, 10
 */
static unsigned int do_pkg_cstate;
static unsigned int config_pkg_cstate;
static bool apply_config_pkg_cstate;

/*
 * Boolean indicating if the processor supports 'I/O System-Management Interrupt
 * counter'
 */
static bool do_smi;
static bool config_smi;
static bool apply_config_smi;

/*
 * Boolean indicating if the processor supports 'Digital temperature sensor'
 * This feature enables the monitoring of the temperature of each core
 *
 * This feature has two limitations:
 *  - if MSR_IA32_TEMPERATURE_TARGET is not supported, the absolute temperature
 * might be wrong
 *  - Temperatures above the tcc_activation_temp are not recorded
 */
static bool do_dts;
static bool config_dts;
static bool apply_config_dts;

/*
 * Boolean indicating if the processor supports 'Package thermal management'
 * This feature allows the monitoring of the temperature of each package
 *
 * This feature has two limitations:
 *  - if MSR_IA32_TEMPERATURE_TARGET is not supported, the absolute temperature
 * might be wrong
 *  - Temperatures above the tcc_activation_temp are not recorded
 */
static bool do_ptm;
static bool config_ptm;
static bool apply_config_ptm;

/*
 * Thermal Control Circuit Activation Temperature as configured by the user.
 * This override the automated detection via MSR_IA32_TEMPERATURE_TARGET
 * and should only be used if the automated detection fails.
 */
static unsigned int tcc_activation_temp;

static unsigned int do_power_fields;
#define UFS_PLATFORM (1 << 0)
#define TURBO_PLATFORM (1 << 1)
#define PSTATES_PLATFORM (1 << 2)

static unsigned int do_rapl;
static unsigned int config_rapl;
static bool apply_config_rapl;
static double rapl_energy_units;
static double rapl_power_units;

#define RAPL_PKG (1 << 0)
/* 0x610 MSR_PKG_POWER_LIMIT */
/* 0x611 MSR_PKG_ENERGY_STATUS */
/* 0x614 MSR_PKG_POWER_INFO */
#define RAPL_DRAM (1 << 1)
/* 0x618 MSR_DRAM_POWER_LIMIT */
/* 0x619 MSR_DRAM_ENERGY_STATUS */
/* 0x61c MSR_DRAM_POWER_INFO */
#define RAPL_CORES (1 << 2)
/* 0x638 MSR_PP0_POWER_LIMIT */
/* 0x639 MSR_PP0_ENERGY_STATUS */

#define RAPL_GFX (1 << 3)
/* 0x640 MSR_PP1_POWER_LIMIT */
/* 0x641 MSR_PP1_ENERGY_STATUS */
/* 0x642 MSR_PP1_POLICY */
#define TJMAX_DEFAULT 100

static cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_saved_affinity_set;
static size_t cpu_present_setsize, cpu_affinity_setsize,
    cpu_saved_affinity_setsize;

static struct thread_data {
  unsigned long long tsc;
  unsigned long long aperf;
  unsigned long long mperf;
  unsigned long long c1;
  unsigned int smi_count;
  unsigned int cpu_id;
  unsigned int flags;
#define CPU_IS_FIRST_THREAD_IN_CORE 0x2
#define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
} * thread_delta, *thread_even, *thread_odd;

static struct core_data {
  unsigned long long c3;
  unsigned long long c6;
  unsigned long long c7;
  unsigned int core_temp_c;
  unsigned int core_id;
} * core_delta, *core_even, *core_odd;

static struct pkg_data {
  unsigned long long pc2;
  unsigned long long pc3;
  unsigned long long pc6;
  unsigned long long pc7;
  unsigned long long pc8;
  unsigned long long pc9;
  unsigned long long pc10;
  unsigned int package_id;
  uint32_t energy_pkg;   /* MSR_PKG_ENERGY_STATUS */
  uint32_t energy_dram;  /* MSR_DRAM_ENERGY_STATUS */
  uint32_t energy_cores; /* MSR_PP0_ENERGY_STATUS */
  uint32_t energy_gfx;   /* MSR_PP1_ENERGY_STATUS */
  uint32_t tdp;
  uint8_t turbo_enabled;
  uint8_t pstates_enabled;
  uint32_t uncore;
  unsigned int tcc_activation_temp;
  unsigned int pkg_temp_c;
} * package_delta, *package_even, *package_odd;

#define DELTA_COUNTERS thread_delta, core_delta, package_delta
#define ODD_COUNTERS thread_odd, core_odd, package_odd
#define EVEN_COUNTERS thread_even, core_even, package_even
static bool is_even = true;

static bool allocated;
static bool initialized;

#define GET_THREAD(thread_base, thread_no, core_no, pkg_no)                    \
  (thread_base + (pkg_no)*topology.num_cores * topology.num_threads +          \
   (core_no)*topology.num_threads + (thread_no))
#define GET_CORE(core_base, core_no, pkg_no)                                   \
  (core_base + (pkg_no)*topology.num_cores + (core_no))
#define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)

struct cpu_topology {
  unsigned int package_id;
  unsigned int core_id;
  bool first_core_in_package;
  bool first_thread_in_core;
};

static struct topology {
  unsigned int max_cpu_id;
  unsigned int num_packages;
  unsigned int num_cores;
  unsigned int num_threads;
  struct cpu_topology *cpus;
} topology;

static cdtime_t time_even, time_odd, time_delta;

static const char *config_keys[] = {
    "CoreCstates",
    "PackageCstates",
    "SystemManagementInterrupt",
    "DigitalTemperatureSensor",
    "PackageThermalManagement",
    "TCCActivationTemp",
    "RunningAveragePowerLimit",
    "LogicalCoreNames",
    "RestoreAffinityPolicy",
};
static const int config_keys_num = STATIC_ARRAY_SIZE(config_keys);

/*****************************
 *  MSR Manipulation helpers *
 *****************************/

/*
 * Open a MSR device for reading
 * Can change the scheduling affinity of the current process if multiple_read is
 * true
 */
static int __attribute__((warn_unused_result))
open_msr(unsigned int cpu, bool multiple_read) {
  char pathname[32];
  int fd;

  /*
   * If we need to do multiple read, let's migrate to the CPU
   * Otherwise, we would lose time calling functions on another CPU
   *
   * If we are not yet initialized (cpu_affinity_setsize = 0),
   * we need to skip this optimisation.
   */
  if (multiple_read && cpu_affinity_setsize) {
    CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
    CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
    if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) {
      ERROR("turbostat plugin: Could not migrate to CPU %d", cpu);
      return -1;
    }
  }

  snprintf(pathname, sizeof(pathname), "/dev/cpu/%d/msr", cpu);
  fd = open(pathname, O_RDONLY);
  if (fd < 0) {
    ERROR("turbostat plugin: failed to open %s", pathname);
    return -1;
  }
  return fd;
}

/*
 * Read a single MSR from an open file descriptor
 */
static int __attribute__((warn_unused_result))
read_msr(int fd, off_t offset, unsigned long long *msr) {
  ssize_t retval;

  retval = pread(fd, msr, sizeof *msr, offset);

  if (retval != sizeof *msr) {
    ERROR("turbostat plugin: MSR offset 0x%llx read failed",
          (unsigned long long)offset);
    return -1;
  }
  return 0;
}

/*
 * Open a MSR device for reading, read the value asked for and close it.
 * This call will not affect the scheduling affinity of this thread.
 */
static ssize_t __attribute__((warn_unused_result))
get_msr(unsigned int cpu, off_t offset, unsigned long long *msr) {
  ssize_t retval;
  int fd;

  fd = open_msr(cpu, 0);
  if (fd < 0)
    return fd;
  retval = read_msr(fd, offset, msr);
  close(fd);
  return retval;
}

/********************************
 * Raw data acquisition (1 CPU) *
 ********************************/

/*
 * Read every data avalaible for a single CPU
 *
 * Core data is shared for all threads in one core: extracted only for the first
 * thread
 * Package data is shared for all core in one package: extracted only for the
 * first thread of the first core
 *
 * Side effect: migrates to the targeted CPU
 */
static int __attribute__((warn_unused_result))
get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) {
  unsigned int cpu = t->cpu_id;
  unsigned long long msr;
  int msr_fd;
  int retval = 0;

  msr_fd = open_msr(cpu, 1);
  if (msr_fd < 0)
    return msr_fd;

#define READ_MSR(msr, dst)                                                     \
  do {                                                                         \
    if (read_msr(msr_fd, msr, dst)) {                                          \
      ERROR("turbostat plugin: Unable to read " #msr);                         \
      retval = -1;                                                             \
      goto out;                                                                \
    }                                                                          \
  } while (0)

  READ_MSR(MSR_IA32_TSC, &t->tsc);

  READ_MSR(MSR_IA32_APERF, &t->aperf);
  READ_MSR(MSR_IA32_MPERF, &t->mperf);

  if (do_smi) {
    READ_MSR(MSR_SMI_COUNT, &msr);
    t->smi_count = msr & 0xFFFFFFFF;
  }

  /* collect core counters only for 1st thread in core */
  if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) {
    retval = 0;
    goto out;
  }

  if (do_core_cstate & (1 << 3))
    READ_MSR(MSR_CORE_C3_RESIDENCY, &c->c3);
  if (do_core_cstate & (1 << 6))
    READ_MSR(MSR_CORE_C6_RESIDENCY, &c->c6);
  if (do_core_cstate & (1 << 7))
    READ_MSR(MSR_CORE_C7_RESIDENCY, &c->c7);

  if (do_dts) {
    READ_MSR(MSR_IA32_THERM_STATUS, &msr);
    c->core_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
  }

  /* collect package counters only for 1st core in package */
  if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
    retval = 0;
    goto out;
  }

  if (do_pkg_cstate & (1 << 2))
    READ_MSR(MSR_PKG_C2_RESIDENCY, &p->pc2);
  if (do_pkg_cstate & (1 << 3))
    READ_MSR(MSR_PKG_C3_RESIDENCY, &p->pc3);
  if (do_pkg_cstate & (1 << 6))
    READ_MSR(MSR_PKG_C6_RESIDENCY, &p->pc6);
  if (do_pkg_cstate & (1 << 7))
    READ_MSR(MSR_PKG_C7_RESIDENCY, &p->pc7);
  if (do_pkg_cstate & (1 << 8))
    READ_MSR(MSR_PKG_C8_RESIDENCY, &p->pc8);
  if (do_pkg_cstate & (1 << 9))
    READ_MSR(MSR_PKG_C9_RESIDENCY, &p->pc9);
  if (do_pkg_cstate & (1 << 10))
    READ_MSR(MSR_PKG_C10_RESIDENCY, &p->pc10);

  if (do_rapl & RAPL_PKG) {
    READ_MSR(MSR_PKG_ENERGY_STATUS, &msr);
    p->energy_pkg = msr & 0xFFFFFFFF;
    READ_MSR(MSR_PKG_POWER_INFO, &msr);
    p->tdp = msr & 0x7FFF;
  }
  if (do_rapl & RAPL_CORES) {
    READ_MSR(MSR_PP0_ENERGY_STATUS, &msr);
    p->energy_cores = msr & 0xFFFFFFFF;
  }
  if (do_rapl & RAPL_DRAM) {
    READ_MSR(MSR_DRAM_ENERGY_STATUS, &msr);
    p->energy_dram = msr & 0xFFFFFFFF;
  }
  if (do_rapl & RAPL_GFX) {
    READ_MSR(MSR_PP1_ENERGY_STATUS, &msr);
    p->energy_gfx = msr & 0xFFFFFFFF;
  }
  if (do_ptm) {
    READ_MSR(MSR_IA32_PACKAGE_THERM_STATUS, &msr);
    p->pkg_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
  }
  if (do_power_fields & TURBO_PLATFORM) {
    READ_MSR(MSR_IA32_MISC_ENABLE, &msr);
    p->turbo_enabled = !((msr >> 38) & 0x1);
  }
  if (do_power_fields & PSTATES_PLATFORM) {
    READ_MSR(MSR_IA32_MISC_ENABLE, &msr);
    p->pstates_enabled = (msr >> 16) & 0x1;
  }
  if (do_power_fields & UFS_PLATFORM) {
    READ_MSR(MSR_UNCORE_FREQ_SCALING, &msr);
    p->uncore = msr & 0x1F;
  }

out:
  close(msr_fd);
  return retval;
}

/**********************************
 * Evaluating the changes (1 CPU) *
 **********************************/

/*
 * Extract the evolution old->new in delta at a package level
 * (some are not new-delta, e.g. temperature)
 */
static inline void delta_package(struct pkg_data *delta,
                                 const struct pkg_data *new,
                                 const struct pkg_data *old) {
  delta->pc2 = new->pc2 - old->pc2;
  delta->pc3 = new->pc3 - old->pc3;
  delta->pc6 = new->pc6 - old->pc6;
  delta->pc7 = new->pc7 - old->pc7;
  delta->pc8 = new->pc8 - old->pc8;
  delta->pc9 = new->pc9 - old->pc9;
  delta->pc10 = new->pc10 - old->pc10;
  delta->pkg_temp_c = new->pkg_temp_c;

  delta->energy_pkg = new->energy_pkg - old->energy_pkg;
  delta->energy_cores = new->energy_cores - old->energy_cores;
  delta->energy_gfx = new->energy_gfx - old->energy_gfx;
  delta->energy_dram = new->energy_dram - old->energy_dram;
  delta->tdp = new->tdp;
  delta->turbo_enabled = new->turbo_enabled;
  delta->pstates_enabled = new->pstates_enabled;
  delta->tcc_activation_temp = new->tcc_activation_temp;
  delta->uncore = new->uncore;
}

/*
 * Extract the evolution old->new in delta at a core level
 * (some are not new-delta, e.g. temperature)
 */
static inline void delta_core(struct core_data *delta,
                              const struct core_data *new,
                              const struct core_data *old) {
  delta->c3 = new->c3 - old->c3;
  delta->c6 = new->c6 - old->c6;
  delta->c7 = new->c7 - old->c7;
  delta->core_temp_c = new->core_temp_c;
}

/*
 * Extract the evolution old->new in delta at a package level
 * core_delta is required for c1 estimation (tsc - c0 - all core cstates)
 */
static inline int __attribute__((warn_unused_result))
delta_thread(struct thread_data *delta, const struct thread_data *new,
             const struct thread_data *old, const struct core_data *cdelta) {
  delta->tsc = new->tsc - old->tsc;

  /* check for TSC < 1 Mcycles over interval */
  if (delta->tsc < (1000 * 1000)) {
    WARNING("turbostat plugin: Insanely slow TSC rate, TSC stops "
            "in idle? You can disable all c-states by booting with"
            " 'idle=poll' or just the deep ones with"
            " 'processor.max_cstate=1'");
    return -1;
  }

  delta->c1 = new->c1 - old->c1;

  if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
    delta->aperf = new->aperf - old->aperf;
    delta->mperf = new->mperf - old->mperf;
  } else {
    if (!aperf_mperf_unstable) {
      WARNING("turbostat plugin: APERF or MPERF went "
              "backwards. Frequency results do not cover "
              "the entire interval. Fix this by running "
              "Linux-2.6.30 or later.");

      aperf_mperf_unstable = true;
    }
  }

  /*
   * As counter collection is not atomic,
   * it is possible for mperf's non-halted cycles + idle states
   * to exceed TSC's all cycles: show c1 = 0% in that case.
   */
  if ((delta->mperf + cdelta->c3 + cdelta->c6 + cdelta->c7) > delta->tsc)
    delta->c1 = 0;
  else {
    /* normal case, derive c1 */
    delta->c1 =
        delta->tsc - delta->mperf - cdelta->c3 - cdelta->c6 - cdelta->c7;
  }

  if (delta->mperf == 0) {
    WARNING("turbostat plugin: cpu%d MPERF 0!", old->cpu_id);
    delta->mperf = 1; /* divide by 0 protection */
  }

  if (do_smi)
    delta->smi_count = new->smi_count - old->smi_count;

  return 0;
}

/**********************************
 * Submitting the results (1 CPU) *
 **********************************/

/*
 * Submit one gauge value
 */
static void turbostat_submit(const char *plugin_instance, const char *type,
                             const char *type_instance, gauge_t value) {
  value_list_t vl = VALUE_LIST_INIT;

  vl.values = &(value_t){.gauge = value};
  vl.values_len = 1;
  sstrncpy(vl.plugin, PLUGIN_NAME, sizeof(vl.plugin));
  if (plugin_instance != NULL)
    sstrncpy(vl.plugin_instance, plugin_instance, sizeof(vl.plugin_instance));
  sstrncpy(vl.type, type, sizeof(vl.type));
  if (type_instance != NULL)
    sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));

  plugin_dispatch_values(&vl);
}

/*
 * Submit every data for a single CPU
 *
 * Core data is shared for all threads in one core: submitted only for the first
 * thread
 * Package data is shared for all core in one package: submitted only for the
 * first thread of the first core
 */
static int submit_counters(struct thread_data *t, struct core_data *c,
                           struct pkg_data *p) {
  char name[DATA_MAX_NAME_LEN];
  double interval_float;

  interval_float = CDTIME_T_TO_DOUBLE(time_delta);

  DEBUG("turbostat plugin: submit stats for cpu: %d, core: %d, pkg: %d",
        t->cpu_id, c->core_id, p->package_id);

  snprintf(name, sizeof(name), "cpu%02d", t->cpu_id);

  if (!aperf_mperf_unstable)
    turbostat_submit(name, "percent", "c0", 100.0 * t->mperf / t->tsc);
  if (!aperf_mperf_unstable)
    turbostat_submit(name, "percent", "c1", 100.0 * t->c1 / t->tsc);

  turbostat_submit(name, "frequency", "average",
                   1.0 / 1000000 * t->aperf / interval_float);

  if ((!aperf_mperf_unstable) || (!(t->aperf > t->tsc || t->mperf > t->tsc)))
    turbostat_submit(name, "frequency", "busy",
                     1.0 * t->tsc / 1000000 * t->aperf / t->mperf /
                         interval_float);

  /* Sanity check (should stay stable) */
  turbostat_submit(name, "gauge", "TSC",
                   1.0 * t->tsc / 1000000 / interval_float);

  /* SMI */
  if (do_smi)
    turbostat_submit(name, "count", NULL, t->smi_count);

  /* submit per-core data only for 1st thread in core */
  if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
    goto done;

  /* If not using logical core numbering, set core id */
  if (!config_lcn) {
    if (topology.num_packages > 1)
      snprintf(name, sizeof(name), "pkg%02d-core%02d", p->package_id,
               c->core_id);
    else
      snprintf(name, sizeof(name), "core%02d", c->core_id);
  }

  if (do_core_cstate & (1 << 3))
    turbostat_submit(name, "percent", "c3", 100.0 * c->c3 / t->tsc);
  if (do_core_cstate & (1 << 6))
    turbostat_submit(name, "percent", "c6", 100.0 * c->c6 / t->tsc);
  if (do_core_cstate & (1 << 7))
    turbostat_submit(name, "percent", "c7", 100.0 * c->c7 / t->tsc);

  if (do_dts)
    turbostat_submit(name, "temperature", NULL, c->core_temp_c);

  /* submit per-package data only for 1st core in package */
  if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
    goto done;

  snprintf(name, sizeof(name), "pkg%02d", p->package_id);

  if (do_ptm)
    turbostat_submit(name, "temperature", NULL, p->pkg_temp_c);

  if (do_pkg_cstate & (1 << 2))
    turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2 / t->tsc);
  if (do_pkg_cstate & (1 << 3))
    turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3 / t->tsc);
  if (do_pkg_cstate & (1 << 6))
    turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6 / t->tsc);
  if (do_pkg_cstate & (1 << 7))
    turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7 / t->tsc);
  if (do_pkg_cstate & (1 << 8))
    turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8 / t->tsc);
  if (do_pkg_cstate & (1 << 9))
    turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9 / t->tsc);
  if (do_pkg_cstate & (1 << 10))
    turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10 / t->tsc);

  if (do_rapl) {
    if (do_rapl & RAPL_PKG) {
      turbostat_submit(name, "power", "pkg",
                       p->energy_pkg * rapl_energy_units / interval_float);
      turbostat_submit(name, "tdp", "pkg", p->tdp * rapl_power_units);
    }
    if (do_rapl & RAPL_CORES)
      turbostat_submit(name, "power", "cores",
                       p->energy_cores * rapl_energy_units / interval_float);
    if (do_rapl & RAPL_GFX)
      turbostat_submit(name, "power", "GFX",
                       p->energy_gfx * rapl_energy_units / interval_float);
    if (do_rapl & RAPL_DRAM)
      turbostat_submit(name, "power", "DRAM",
                       p->energy_dram * rapl_energy_units / interval_float);
  }

  if (do_power_fields & TURBO_PLATFORM) {
    turbostat_submit(name, "turbo_enabled", NULL, p->turbo_enabled);
  }
  if (do_power_fields & PSTATES_PLATFORM) {
    turbostat_submit(name, "pstates_enabled", NULL, p->pstates_enabled);
  }
  if (do_power_fields & UFS_PLATFORM) {
    turbostat_submit(name, "uncore_ratio", NULL, p->uncore);
  }
  turbostat_submit(name, "temperature", "tcc_activation",
                   p->tcc_activation_temp);
done:
  return 0;
}

/**********************************
 * Looping function over all CPUs *
 **********************************/

/*
 * Check if a given cpu id is in our compiled list of existing CPUs
 */
static int cpu_is_not_present(unsigned int cpu) {
  return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
}

/*
 * Loop on all CPUs in topological order
 *
 * Skip non-present cpus
 * Return the error code at the first error or 0
 */
static int __attribute__((warn_unused_result))
for_all_cpus(int(func)(struct thread_data *, struct core_data *,
                       struct pkg_data *),
             struct thread_data *thread_base, struct core_data *core_base,
             struct pkg_data *pkg_base) {
  int retval;

  for (unsigned int pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
    for (unsigned int core_no = 0; core_no < topology.num_cores; ++core_no) {
      for (unsigned int thread_no = 0; thread_no < topology.num_threads;
           ++thread_no) {
        struct thread_data *t;
        struct core_data *c;
        struct pkg_data *p;

        t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);

        if (cpu_is_not_present(t->cpu_id))
          continue;

        c = GET_CORE(core_base, core_no, pkg_no);
        p = GET_PKG(pkg_base, pkg_no);

        retval = func(t, c, p);
        if (retval)
          return retval;
      }
    }
  }
  return 0;
}

/*
 * Dedicated loop: Extract every data evolution for all CPU
 *
 * Skip non-present cpus
 * Return the error code at the first error or 0
 *
 * Core data is shared for all threads in one core: extracted only for the first
 * thread
 * Package data is shared for all core in one package: extracted only for the
 * first thread of the first core
 */
static int __attribute__((warn_unused_result))
for_all_cpus_delta(const struct thread_data *thread_new_base,
                   const struct core_data *core_new_base,
                   const struct pkg_data *pkg_new_base,
                   const struct thread_data *thread_old_base,
                   const struct core_data *core_old_base,
                   const struct pkg_data *pkg_old_base) {
  int retval;

  for (unsigned int pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
    for (unsigned int core_no = 0; core_no < topology.num_cores; ++core_no) {
      for (unsigned int thread_no = 0; thread_no < topology.num_threads;
           ++thread_no) {
        struct thread_data *t_delta;
        const struct thread_data *t_old, *t_new;
        struct core_data *c_delta;

        /* Get correct pointers for threads */
        t_delta = GET_THREAD(thread_delta, thread_no, core_no, pkg_no);
        t_new = GET_THREAD(thread_new_base, thread_no, core_no, pkg_no);
        t_old = GET_THREAD(thread_old_base, thread_no, core_no, pkg_no);

        /* Skip threads that disappeared */
        if (cpu_is_not_present(t_delta->cpu_id))
          continue;

        /* c_delta is always required for delta_thread */
        c_delta = GET_CORE(core_delta, core_no, pkg_no);

        /* calculate core delta only for 1st thread in core */
        if (t_new->flags & CPU_IS_FIRST_THREAD_IN_CORE) {
          const struct core_data *c_old, *c_new;

          c_new = GET_CORE(core_new_base, core_no, pkg_no);
          c_old = GET_CORE(core_old_base, core_no, pkg_no);

          delta_core(c_delta, c_new, c_old);
        }

        /* Always calculate thread delta */
        retval = delta_thread(t_delta, t_new, t_old, c_delta);
        if (retval)
          return retval;

        /* calculate package delta only for 1st core in package */
        if (t_new->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) {
          struct pkg_data *p_delta;
          const struct pkg_data *p_old, *p_new;

          p_delta = GET_PKG(package_delta, pkg_no);
          p_new = GET_PKG(pkg_new_base, pkg_no);
          p_old = GET_PKG(pkg_old_base, pkg_no);

          delta_package(p_delta, p_new, p_old);
        }
      }
    }
  }
  return 0;
}

/***************
 * CPU Probing *
 ***************/

/*
 * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
 * the Thermal Control Circuit (TCC) activates.
 * This is usually equal to tjMax.
 *
 * Older processors do not have this MSR, so there we guess,
 * but also allow conficuration over-ride with "TCCActivationTemp".
 *
 * Several MSR temperature values are in units of degrees-C
 * below this value, including the Digital Thermal Sensor (DTS),
 * Package Thermal Management Sensor (PTM), and thermal event thresholds.
 */
static int __attribute__((warn_unused_result))
set_temperature_target(struct thread_data *t, struct core_data *c,
                       struct pkg_data *p) {
  unsigned long long msr;
  unsigned int target_c_local;

  /* tcc_activation_temp is used only for dts or ptm */
  if (!(do_dts || do_ptm))
    return 0;

  /* this is a per-package concept */
  if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) ||
      !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
    return 0;

  if (tcc_activation_temp != 0) {
    p->tcc_activation_temp = tcc_activation_temp;
    return 0;
  }

  if (get_msr(t->cpu_id, MSR_IA32_TEMPERATURE_TARGET, &msr))
    goto guess;

  target_c_local = (msr >> 16) & 0xFF;

  if (!target_c_local)
    goto guess;

  p->tcc_activation_temp = target_c_local;

  return 0;

guess:
  p->tcc_activation_temp = TJMAX_DEFAULT;
  WARNING("turbostat plugin: cpu%d: Guessing tjMax %d C,"
          " Please use TCCActivationTemp to specify it.",
          t->cpu_id, p->tcc_activation_temp);

  return 0;
}

/*
 * Identify the functionality of the CPU
 */
static int __attribute__((warn_unused_result)) probe_cpu(void) {
  unsigned int eax, ebx, ecx, edx, max_level;
  unsigned int fms, family, model;

  /* CPUID(0):
   * - EAX: Maximum Input Value for Basic CPUID Information
   * - EBX: "Genu" (0x756e6547)
   * - EDX: "ineI" (0x49656e69)
   * - ECX: "ntel" (0x6c65746e)
   */
  max_level = ebx = ecx = edx = 0;
  __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
  if (ebx != 0x756e6547 && edx != 0x49656e69 && ecx != 0x6c65746e) {
    ERROR("turbostat plugin: Unsupported CPU (not Intel)");
    return -1;
  }

  /* CPUID(1):
   * - EAX: Version Information: Type, Family, Model, and Stepping ID
   *  + 4-7:   Model ID
   *  + 8-11:  Family ID
   *  + 12-13: Processor type
   *  + 16-19: Extended Model ID
   *  + 20-27: Extended Family ID
   * - EDX: Feature Information:
   *  + 5: Support for MSR read/write operations
   */
  fms = ebx = ecx = edx = 0;
  __get_cpuid(1, &fms, &ebx, &ecx, &edx);
  family = (fms >> 8) & 0xf;
  model = (fms >> 4) & 0xf;
  if (family == 0xf)
    family += (fms >> 20) & 0xf;
  if (family == 6 || family == 0xf)
    model += ((fms >> 16) & 0xf) << 4;
  if (!(edx & (1 << 5))) {
    ERROR("turbostat plugin: Unsupported CPU (no MSR support)");
    return -1;
  }

  /*
   * CPUID(6):
   * - EAX:
   *  + 0: Digital temperature sensor is supported if set
   *  + 6: Package thermal management is supported if set
   * - ECX:
   *  + 0: Hardware Coordination Feedback Capability (Presence of IA32_MPERF and
   * IA32_APERF).
   *  + 3: The processor supports performance-energy bias preference if set.
   *       It also implies the presence of a new architectural MSR called
   * IA32_ENERGY_PERF_BIAS
   *
   * This check is valid for both Intel and AMD
   */
  eax = ebx = ecx = edx = 0;
  __get_cpuid(0x6, &eax, &ebx, &ecx, &edx);
  do_dts = eax & (1 << 0);
  do_ptm = eax & (1 << 6);
  if (!(ecx & (1 << 0))) {
    ERROR("turbostat plugin: Unsupported CPU (No APERF)");
    return -1;
  }

  /*
   * Enable or disable C states depending on the model and family
   */
  if (family == 6) {
    switch (model) {
    /* Atom (partial) */
    case 0x27:
      do_smi = false;
      do_core_cstate = 0;
      do_pkg_cstate = (1 << 2) | (1 << 4) | (1 << 6);
      break;
    /* Silvermont */
    case 0x37: /* BYT */
    case 0x4D: /* AVN */
      do_smi = true;
      do_core_cstate = (1 << 1) | (1 << 6);
      do_pkg_cstate = (1 << 6);
      break;
    /* Nehalem */
    case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
    case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper
                  Forest */
    case 0x1F: /* Core i7 and i5 Processor - Nehalem */
    case 0x2E: /* Nehalem-EX Xeon - Beckton */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6);
      do_pkg_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      break;
    /* Westmere */
    case 0x25: /* Westmere Client - Clarkdale, Arrandale */
    case 0x2C: /* Westmere EP - Gulftown */
    case 0x2F: /* Westmere-EX Xeon - Eagleton */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6);
      do_pkg_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      break;
    /* Sandy Bridge */
    case 0x2A: /* SNB */
    case 0x2D: /* SNB Xeon */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
      break;
    /* Ivy Bridge */
    case 0x3A: /* IVB */
    case 0x3E: /* IVB Xeon */
    case 0x55: /* SKX,CLX Xeon */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
      break;
    /* Haswell Bridge */
    case 0x3C: /* HSW */
    case 0x3F: /* HSW */
    case 0x46: /* HSW */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
      break;
    case 0x45: /* HSW */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7) | (1 << 8) |
                      (1 << 9) | (1 << 10);
      break;
    /* Broadwell */
    case 0x4F: /* BDW */
    case 0x56: /* BDX-DE */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7);
      break;
    case 0x3D: /* BDW */
      do_smi = true;
      do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
      do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7) | (1 << 8) |
                      (1 << 9) | (1 << 10);
      break;
    default:
      do_smi = false;
      do_core_cstate = 0;
      do_pkg_cstate = 0;
      break;
    }
    switch (model) {
    case 0x2A: /* SNB */
    case 0x3A: /* IVB */
    case 0x3C: /* HSW */
    case 0x45: /* HSW */
    case 0x46: /* HSW */
    case 0x3D: /* BDW */
    case 0x5E: /* SKL */
      do_rapl = RAPL_PKG | RAPL_CORES | RAPL_GFX;
      break;
    case 0x3F: /* HSX */
    case 0x4F: /* BDX */
    case 0x56: /* BDX-DE */
      do_rapl = RAPL_PKG | RAPL_DRAM;
      do_power_fields = TURBO_PLATFORM | UFS_PLATFORM | PSTATES_PLATFORM;
      break;
    case 0x2D: /* SNB Xeon */
    case 0x3E: /* IVB Xeon */
    case 0x55: /* SKX,CLX Xeon */
      do_rapl = RAPL_PKG | RAPL_CORES | RAPL_DRAM;
      do_power_fields = TURBO_PLATFORM | PSTATES_PLATFORM;
      break;
    case 0x37: /* BYT */
    case 0x4D: /* AVN */
      do_rapl = RAPL_PKG | RAPL_CORES;
      break;
    default:
      do_rapl = 0;
    }
  } else {
    ERROR("turbostat plugin: Unsupported CPU (family: %#x, "
          "model: %#x)",
          family, model);
    return -1;
  }

  /* Override detected values with configuration */
  if (apply_config_core_cstate)
    do_core_cstate = config_core_cstate;
  if (apply_config_pkg_cstate)
    do_pkg_cstate = config_pkg_cstate;
  if (apply_config_smi)
    do_smi = config_smi;
  if (apply_config_dts)
    do_dts = config_dts;
  if (apply_config_ptm)
    do_ptm = config_ptm;
  if (apply_config_rapl)
    do_rapl = config_rapl;

  if (do_rapl) {
    unsigned long long msr;
    if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
      return 0;

    rapl_power_units = 1.0 / (1 << (msr & 0xF));
    if (model == 0x37)
      rapl_energy_units = 1.0 * (1 << (msr >> 8 & 0x1F)) / 1000000;
    else
      rapl_energy_units = 1.0 / (1 << (msr >> 8 & 0x1F));
  }

  return 0;
}

/********************
 * Topology Probing *
 ********************/

/*
 * Read a single int from a file.
 */
static int __attribute__((format(printf, 1, 2)))
parse_int_file(const char *fmt, ...) {
  va_list args;
  char path[PATH_MAX];
  int len;

  va_start(args, fmt);
  len = vsnprintf(path, sizeof(path), fmt, args);
  va_end(args);
  if (len < 0 || len >= PATH_MAX) {
    ERROR("turbostat plugin: path truncated: '%s'", path);
    return -1;
  }

  value_t v;
  if (parse_value_file(path, &v, DS_TYPE_DERIVE) != 0) {
    ERROR("turbostat plugin: Parsing \"%s\" failed.", path);
    return -1;
  }

  return (int)v.derive;
}

static int get_threads_on_core(unsigned int cpu) {
  char path[80];
  FILE *filep;
  int sib1, sib2;
  int matches;
  char character;

  snprintf(path, sizeof(path),
           "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
  filep = fopen(path, "r");
  if (!filep) {
    ERROR("turbostat plugin: Failed to open '%s'", path);
    return -1;
  }
  /*
   * file format:
   * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
   * otherwinse 1 sibling (self).
   */
  matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);

  fclose(filep);

  if (matches == 3)
    return 2;
  else
    return 1;
}

/*
 * run func(cpu) on every cpu in /proc/stat
 * return max_cpu number
 */
static int __attribute__((warn_unused_result))
for_all_proc_cpus(int(func)(unsigned int)) {
  FILE *fp;
  unsigned int cpu_num;
  int retval;

  fp = fopen("/proc/stat", "r");
  if (!fp) {
    ERROR("turbostat plugin: Failed to open /proc/stat");
    return -1;
  }

  retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
  if (retval != 0) {
    ERROR("turbostat plugin: Failed to parse /proc/stat");
    fclose(fp);
    return -1;
  }

  while (1) {
    retval =
        fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
    if (retval != 1)
      break;

    retval = func(cpu_num);
    if (retval) {
      fclose(fp);
      return retval;
    }
  }
  fclose(fp);
  return 0;
}

/*
 * Update the stored topology.max_cpu_id
 */
static int update_max_cpu_id(unsigned int cpu) {
  if (topology.max_cpu_id < cpu)
    topology.max_cpu_id = cpu;
  return 0;
}

static int mark_cpu_present(unsigned int cpu) {
  CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
  return 0;
}

static int __attribute__((warn_unused_result))
allocate_cpu_set(cpu_set_t **set, size_t *size) {
  *set = CPU_ALLOC(topology.max_cpu_id + 1);
  if (*set == NULL) {
    ERROR("turbostat plugin: Unable to allocate CPU state");
    return -1;
  }
  *size = CPU_ALLOC_SIZE(topology.max_cpu_id + 1);
  CPU_ZERO_S(*size, *set);
  return 0;
}

/*
 * Build a local representation of the cpu distribution
 */
static int __attribute__((warn_unused_result)) topology_probe(void) {
  int ret;
  unsigned int max_package_id, max_core_id, max_threads;
  max_package_id = max_core_id = max_threads = 0;

  /* Clean topology */
  free(topology.cpus);
  memset(&topology, 0, sizeof(topology));

  ret = for_all_proc_cpus(update_max_cpu_id);
  if (ret != 0)
    goto err;

  topology.cpus =
      calloc(1, (topology.max_cpu_id + 1) * sizeof(struct cpu_topology));
  if (topology.cpus == NULL) {
    ERROR("turbostat plugin: Unable to allocate memory for CPU topology");
    return -1;
  }

  ret = allocate_cpu_set(&cpu_present_set, &cpu_present_setsize);
  if (ret != 0)
    goto err;
  ret = allocate_cpu_set(&cpu_affinity_set, &cpu_affinity_setsize);
  if (ret != 0)
    goto err;
  ret = allocate_cpu_set(&cpu_saved_affinity_set, &cpu_saved_affinity_setsize);
  if (ret != 0)
    goto err;

  ret = for_all_proc_cpus(mark_cpu_present);
  if (ret != 0)
    goto err;

  /*
   * For online cpus
   * find max_core_id, max_package_id
   */
  for (unsigned int i = 0; i <= topology.max_cpu_id; ++i) {
    unsigned int num_threads;
    struct cpu_topology *cpu = &topology.cpus[i];

    if (cpu_is_not_present(i)) {
      WARNING("turbostat plugin: cpu%d NOT PRESENT", i);
      continue;
    }

    ret = parse_int_file(
        "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", i);
    if (ret < 0)
      goto err;
    else
      cpu->package_id = (unsigned int)ret;
    if (cpu->package_id > max_package_id)
      max_package_id = cpu->package_id;

    ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", i);
    if (ret < 0)
      goto err;
    else
      cpu->core_id = (unsigned int)ret;
    if (cpu->core_id > max_core_id)
      max_core_id = cpu->core_id;
    ret = parse_int_file(
        "/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", i);
    if (ret < 0)
      goto err;
    else if ((unsigned int)ret == i)
      cpu->first_core_in_package = true;

    ret = get_threads_on_core(i);
    if (ret < 0)
      goto err;
    else
      num_threads = (unsigned int)ret;
    if (num_threads > max_threads)
      max_threads = num_threads;
    ret = parse_int_file(
        "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", i);
    if (ret < 0)
      goto err;
    else if ((unsigned int)ret == i)
      cpu->first_thread_in_core = true;

    DEBUG("turbostat plugin: cpu %d pkg %d core %d\n", i, cpu->package_id,
          cpu->core_id);
  }
  /* Num is max + 1 (need to count 0) */
  topology.num_packages = max_package_id + 1;
  topology.num_cores = max_core_id + 1;
  topology.num_threads = max_threads;

  return 0;
err:
  free(topology.cpus);
  return ret;
}

/************************
 * Main alloc/init/free *
 ************************/

static int allocate_counters(struct thread_data **threads,
                             struct core_data **cores,
                             struct pkg_data **packages) {
  unsigned int total_threads, total_cores;

  if ((topology.num_threads == 0) || (topology.num_cores == 0) ||
      (topology.num_packages == 0)) {
    ERROR(
        "turbostat plugin: Invalid topology: %u threads, %u cores, %u packages",
        topology.num_threads, topology.num_cores, topology.num_packages);
    return -1;
  }

  total_threads =
      topology.num_threads * topology.num_cores * topology.num_packages;
  *threads = calloc(total_threads, sizeof(struct thread_data));
  if (*threads == NULL) {
    ERROR("turbostat plugin: calloc failed");
    return -1;
  }

  for (unsigned int i = 0; i < total_threads; ++i)
    (*threads)[i].cpu_id = topology.max_cpu_id + 1;

  total_cores = topology.num_cores * topology.num_packages;
  *cores = calloc(total_cores, sizeof(struct core_data));
  if (*cores == NULL) {
    ERROR("turbostat plugin: calloc failed");
    sfree(*threads);
    return -1;
  }

  *packages = calloc(topology.num_packages, sizeof(struct pkg_data));
  if (*packages == NULL) {
    ERROR("turbostat plugin: calloc failed");
    sfree(*cores);
    sfree(*threads);
    return -1;
  }

  return 0;
}

static void init_counter(struct thread_data *thread_base,
                         struct core_data *core_base, struct pkg_data *pkg_base,
                         unsigned int cpu_id) {
  struct thread_data *t;
  struct core_data *c;
  struct pkg_data *p;
  struct cpu_topology *cpu = &topology.cpus[cpu_id];

  t = GET_THREAD(thread_base, !(cpu->first_thread_in_core), cpu->core_id,
                 cpu->package_id);
  c = GET_CORE(core_base, cpu->core_id, cpu->package_id);
  p = GET_PKG(pkg_base, cpu->package_id);

  t->cpu_id = cpu_id;
  if (cpu->first_thread_in_core)
    t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
  if (cpu->first_core_in_package)
    t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;

  c->core_id = cpu->core_id;
  p->package_id = cpu->package_id;
}

static void initialize_counters(void) {
  for (unsigned int cpu_id = 0; cpu_id <= topology.max_cpu_id; ++cpu_id) {
    if (cpu_is_not_present(cpu_id))
      continue;
    init_counter(EVEN_COUNTERS, cpu_id);
    init_counter(ODD_COUNTERS, cpu_id);
    init_counter(DELTA_COUNTERS, cpu_id);
  }
}

static void free_all_buffers(void) {
  allocated = false;
  initialized = false;

  CPU_FREE(cpu_present_set);
  cpu_present_set = NULL;
  cpu_present_setsize = 0;

  CPU_FREE(cpu_affinity_set);
  cpu_affinity_set = NULL;
  cpu_affinity_setsize = 0;

  CPU_FREE(cpu_saved_affinity_set);
  cpu_saved_affinity_set = NULL;
  cpu_saved_affinity_setsize = 0;

  free(thread_even);
  free(core_even);
  free(package_even);

  thread_even = NULL;
  core_even = NULL;
  package_even = NULL;

  free(thread_odd);
  free(core_odd);
  free(package_odd);

  thread_odd = NULL;
  core_odd = NULL;
  package_odd = NULL;

  free(thread_delta);
  free(core_delta);
  free(package_delta);

  thread_delta = NULL;
  core_delta = NULL;
  package_delta = NULL;
}

  /**********************
   * Collectd functions *
   **********************/

#define DO_OR_GOTO_ERR(something)                                              \
  do {                                                                         \
    ret = (something);                                                         \
    if (ret < 0)                                                               \
      goto err;                                                                \
  } while (0)

static int setup_all_buffers(void) {
  int ret;

  DO_OR_GOTO_ERR(topology_probe());
  DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
  DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
  DO_OR_GOTO_ERR(allocate_counters(&thread_delta, &core_delta, &package_delta));
  initialize_counters();
  DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
  DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, ODD_COUNTERS));

  allocated = true;
  return 0;
err:
  free_all_buffers();
  return ret;
}

static int save_affinity(void) {
  if (affinity_policy == policy_restore_affinity) {
    /* Try to save the scheduling affinity, as it will be modified by
     * get_counters().
     */
    if (sched_getaffinity(0, cpu_saved_affinity_setsize,
                          cpu_saved_affinity_set) != 0)
      return -1;
  }

  return 0;
}

static void restore_affinity(void) {
  /* Let's restore the affinity to the value saved in save_affinity */
  if (affinity_policy == policy_restore_affinity)
    (void)sched_setaffinity(0, cpu_saved_affinity_setsize,
                            cpu_saved_affinity_set);
  else {
    /* reset the affinity to all present cpus */
    (void)sched_setaffinity(0, cpu_present_setsize, cpu_present_set);
  }
}

static int turbostat_read(void) {
  int ret;

  if (!allocated) {
    if ((ret = setup_all_buffers()) < 0)
      return ret;
  }

  if (for_all_proc_cpus(cpu_is_not_present)) {
    free_all_buffers();
    if ((ret = setup_all_buffers()) < 0)
      return ret;
    if (for_all_proc_cpus(cpu_is_not_present)) {
      ERROR("turbostat plugin: CPU appeared just after "
            "initialization");
      return -1;
    }
  }

  if (save_affinity() != 0) {
    ERROR("turbostat plugin: Unable to save the CPU affinity. Please read the "
          "docs about RestoreAffinityPolicy option.");
    return -1;
  }

  if (!initialized) {
    if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
      goto out;
    time_even = cdtime();
    is_even = true;
    initialized = true;
    ret = 0;
    goto out;
  }

  if (is_even) {
    if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
      goto out;
    time_odd = cdtime();
    is_even = false;
    time_delta = time_odd - time_even;
    if ((ret = for_all_cpus_delta(ODD_COUNTERS, EVEN_COUNTERS)) < 0)
      goto out;
    if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
      goto out;
  } else {
    if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
      goto out;
    time_even = cdtime();
    is_even = true;
    time_delta = time_even - time_odd;
    if ((ret = for_all_cpus_delta(EVEN_COUNTERS, ODD_COUNTERS)) < 0)
      goto out;
    if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
      goto out;
  }
  ret = 0;
out:
  restore_affinity();

  return ret;
}

static int check_permissions(void) {

  if (getuid() == 0) {
    /* We have everything we need */
    return 0;
#if !defined(HAVE_SYS_CAPABILITY_H) && !defined(CAP_SYS_RAWIO)
  } else {
    ERROR("turbostat plugin: Initialization failed: this plugin "
          "requires collectd to run as root");
    return -1;
  }
#else  /* HAVE_SYS_CAPABILITY_H && CAP_SYS_RAWIO */
  }

  int ret = 0;

  if (check_capability(CAP_SYS_RAWIO) != 0) {
    WARNING("turbostat plugin: Collectd doesn't have the "
            "CAP_SYS_RAWIO capability. If you don't want to run "
            "collectd as root, try running \"setcap "
            "cap_sys_rawio=ep\" on collectd binary");
    ret = -1;
  }

  if (euidaccess("/dev/cpu/0/msr", R_OK)) {
    WARNING("turbostat plugin: Collectd cannot open "
            "/dev/cpu/0/msr. If you don't want to run collectd as "
            "root, you need to change the ownership (chown) and "
            "permissions on /dev/cpu/*/msr to allow such access");
    ret = -1;
  }

  if (ret != 0)
    ERROR("turbostat plugin: Initialization failed: this plugin "
          "requires collectd to either to run as root or give "
          "collectd a special capability (CAP_SYS_RAWIO) and read "
          "access to /dev/cpu/*/msr (see previous warnings)");
  return ret;
#endif /* HAVE_SYS_CAPABILITY_H && CAP_SYS_RAWIO */
}

static int turbostat_init(void) {
  struct stat sb;
  int ret;

  if (stat("/dev/cpu/0/msr", &sb)) {
    ERROR("turbostat plugin: Initialization failed: /dev/cpu/0/msr "
          "does not exist while the CPU supports MSR. You may be "
          "missing the corresponding kernel module, please try '# "
          "modprobe msr'");
    return -1;
  }

  DO_OR_GOTO_ERR(check_permissions());

  DO_OR_GOTO_ERR(probe_cpu());

  DO_OR_GOTO_ERR(setup_all_buffers());

  plugin_register_read(PLUGIN_NAME, turbostat_read);

  return 0;
err:
  free_all_buffers();
  return ret;
}

static int turbostat_config(const char *key, const char *value) {
  long unsigned int tmp_val;
  char *end;

  if (strcasecmp("CoreCstates", key) == 0) {
    tmp_val = strtoul(value, &end, 0);
    if (*end != '\0' || tmp_val > UINT_MAX) {
      ERROR("turbostat plugin: Invalid CoreCstates '%s'", value);
      return -1;
    }
    config_core_cstate = (unsigned int)tmp_val;
    apply_config_core_cstate = true;
  } else if (strcasecmp("PackageCstates", key) == 0) {
    tmp_val = strtoul(value, &end, 0);
    if (*end != '\0' || tmp_val > UINT_MAX) {
      ERROR("turbostat plugin: Invalid PackageCstates '%s'", value);
      return -1;
    }
    config_pkg_cstate = (unsigned int)tmp_val;
    apply_config_pkg_cstate = true;
  } else if (strcasecmp("SystemManagementInterrupt", key) == 0) {
    config_smi = IS_TRUE(value);
    apply_config_smi = true;
  } else if (strcasecmp("DigitalTemperatureSensor", key) == 0) {
    config_dts = IS_TRUE(value);
    apply_config_dts = true;
  } else if (strcasecmp("PackageThermalManagement", key) == 0) {
    config_ptm = IS_TRUE(value);
    apply_config_ptm = true;
  } else if (strcasecmp("LogicalCoreNames", key) == 0) {
    config_lcn = IS_TRUE(value);
  } else if (strcasecmp("RunningAveragePowerLimit", key) == 0) {
    tmp_val = strtoul(value, &end, 0);
    if (*end != '\0' || tmp_val > UINT_MAX) {
      ERROR("turbostat plugin: Invalid RunningAveragePowerLimit '%s'", value);
      return -1;
    }
    config_rapl = (unsigned int)tmp_val;
    apply_config_rapl = true;
  } else if (strcasecmp("TCCActivationTemp", key) == 0) {
    tmp_val = strtoul(value, &end, 0);
    if (*end != '\0' || tmp_val > UINT_MAX) {
      ERROR("turbostat plugin: Invalid TCCActivationTemp '%s'", value);
      return -1;
    }
    tcc_activation_temp = (unsigned int)tmp_val;
  } else if (strcasecmp("RestoreAffinityPolicy", key) == 0) {
    if (strcasecmp("Restore", value) == 0)
      affinity_policy = policy_restore_affinity;
    else if (strcasecmp("AllCPUs", value) == 0)
      affinity_policy = policy_allcpus_affinity;
    else {
      ERROR("turbostat plugin: Invalid RestoreAffinityPolicy '%s'", value);
      return -1;
    }
  } else {
    ERROR("turbostat plugin: Invalid configuration option '%s'", key);
    return -1;
  }
  return 0;
}

void module_register(void) {
  plugin_register_init(PLUGIN_NAME, turbostat_init);
  plugin_register_config(PLUGIN_NAME, turbostat_config, config_keys,
                         config_keys_num);
}