X-Git-Url: https://git.octo.it/?a=blobdiff_plain;f=src%2Fturbostat.c;h=80650dff38539c781bf106a2cdcc462ceaaf3300;hb=363bd0300297144a29b58cc813b23d9353525178;hp=e35f3d20dbd09048791b36c66b762d1656049b7b;hpb=6f00bf172139b5561acf8a2622af85b44dea6af9;p=collectd.git diff --git a/src/turbostat.c b/src/turbostat.c index e35f3d20..80650dff 100644 --- a/src/turbostat.c +++ b/src/turbostat.c @@ -35,28 +35,16 @@ */ #define _GNU_SOURCE -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - #include "collectd.h" #include "common.h" #include "plugin.h" +#include "utils_time.h" + +#include +#include +#ifdef HAVE_SYS_CAPABILITY_H +#include +#endif /* HAVE_SYS_CAPABILITY_H */ #define PLUGIN_NAME "turbostat" @@ -80,12 +68,23 @@ static _Bool aperf_mperf_unstable; * 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' @@ -96,6 +95,8 @@ static unsigned int do_pkg_cstate; * - 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' @@ -106,6 +107,8 @@ static _Bool do_dts; * - 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. @@ -115,40 +118,31 @@ static _Bool do_ptm; static unsigned int tcc_activation_temp; static unsigned int do_rapl; +static unsigned int config_rapl; +static _Bool apply_config_rapl; static double rapl_energy_units; #define RAPL_PKG (1 << 0) /* 0x610 MSR_PKG_POWER_LIMIT */ /* 0x611 MSR_PKG_ENERGY_STATUS */ -#define RAPL_PKG_PERF_STATUS (1 << 1) - /* 0x613 MSR_PKG_PERF_STATUS */ -#define RAPL_PKG_POWER_INFO (1 << 2) - /* 0x614 MSR_PKG_POWER_INFO */ - -#define RAPL_DRAM (1 << 3) +#define RAPL_DRAM (1 << 1) /* 0x618 MSR_DRAM_POWER_LIMIT */ /* 0x619 MSR_DRAM_ENERGY_STATUS */ /* 0x61c MSR_DRAM_POWER_INFO */ -#define RAPL_DRAM_PERF_STATUS (1 << 4) - /* 0x61b MSR_DRAM_PERF_STATUS */ - -#define RAPL_CORES (1 << 5) +#define RAPL_CORES (1 << 2) /* 0x638 MSR_PP0_POWER_LIMIT */ /* 0x639 MSR_PP0_ENERGY_STATUS */ -#define RAPL_CORE_POLICY (1 << 6) - /* 0x63a MSR_PP0_POLICY */ - -#define RAPL_GFX (1 << 7) +#define RAPL_GFX (1 << 3) /* 0x640 MSR_PP1_POWER_LIMIT */ /* 0x641 MSR_PP1_ENERGY_STATUS */ /* 0x642 MSR_PP1_POLICY */ #define TJMAX_DEFAULT 100 -cpu_set_t *cpu_present_set, *cpu_affinity_set, *cpu_saved_affinity_set; -size_t cpu_present_setsize, cpu_affinity_setsize, cpu_saved_affinity_setsize; +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; -struct thread_data { +static struct thread_data { unsigned long long tsc; unsigned long long aperf; unsigned long long mperf; @@ -160,7 +154,7 @@ struct thread_data { #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 } *thread_delta, *thread_even, *thread_odd; -struct core_data { +static struct core_data { unsigned long long c3; unsigned long long c6; unsigned long long c7; @@ -168,7 +162,7 @@ struct core_data { unsigned int core_id; } *core_delta, *core_even, *core_odd; -struct pkg_data { +static struct pkg_data { unsigned long long pc2; unsigned long long pc3; unsigned long long pc6; @@ -177,12 +171,10 @@ struct pkg_data { unsigned long long pc9; unsigned long long pc10; unsigned int package_id; - unsigned int energy_pkg; /* MSR_PKG_ENERGY_STATUS */ - unsigned int energy_dram; /* MSR_DRAM_ENERGY_STATUS */ - unsigned int energy_cores; /* MSR_PP0_ENERGY_STATUS */ - unsigned int energy_gfx; /* MSR_PP1_ENERGY_STATUS */ - unsigned int rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ - unsigned int rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ + 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 */ unsigned int tcc_activation_temp; unsigned int pkg_temp_c; } *package_delta, *package_even, *package_odd; @@ -207,63 +199,33 @@ static _Bool initialized = 0; #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) struct cpu_topology { - int package_id; - int core_id; + unsigned int package_id; + unsigned int core_id; _Bool first_core_in_package; _Bool first_thread_in_core; }; -struct topology { - int max_cpu_id; - int num_packages; - int num_cores; - int num_threads; +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; -struct timeval tv_even, tv_odd, tv_delta; - -enum return_values { - OK = 0, - ERR_CPU_MIGRATE, - ERR_CPU_SAVE_SCHED_AFFINITY, - ERR_MSR_IA32_APERF, - ERR_MSR_IA32_MPERF, - ERR_MSR_SMI_COUNT, - ERR_MSR_CORE_C3_RESIDENCY, - ERR_MSR_CORE_C6_RESIDENCY, - ERR_MSR_CORE_C7_RESIDENCY, - ERR_MSR_IA32_THERM_STATUS, - ERR_MSR_PKG_C3_RESIDENCY, - ERR_MSR_PKG_C6_RESIDENCY, - ERR_MSR_PKG_C2_RESIDENCY, - ERR_MSR_PKG_C7_RESIDENCY, - ERR_MSR_PKG_C8_RESIDENCY, - ERR_MSR_PKG_C9_RESIDENCY, - ERR_MSR_PKG_C10_RESIDENCY, - ERR_MSR_PKG_ENERGY_STATUS, - ERR_MSR_PKG_POWER_INFO, - ERR_MSR_PP0_ENERGY_STATUS, - ERR_MSR_DRAM_ENERGY_STATUS, - ERR_MSR_PP1_ENERGY_STATUS, - ERR_MSR_PKG_PERF_STATUS, - ERR_MSR_DRAM_PERF_STATUS, - ERR_MSR_IA32_PACKAGE_THERM_STATUS, - ERR_MSR_IA32_TSC, - ERR_CPU_NOT_PRESENT, - ERR_NO_MSR, - ERR_CANT_OPEN_MSR, - ERR_CANT_OPEN_FILE, - ERR_CANT_READ_NUMBER, - ERR_CANT_READ_PROC_STAT, - ERR_NO_INVARIANT_TSC, - ERR_NO_APERF, - ERR_CALLOC, - ERR_CPU_ALLOC, - ERR_NOT_ROOT, - UNSUPPORTED_CPU, -}; +static cdtime_t time_even, time_odd, time_delta; +static const char *config_keys[] = +{ + "CoreCstates", + "PackageCstates", + "SystemManagementInterrupt", + "DigitalTemperatureSensor", + "PackageThermalManagement", + "TCCActivationTemp", + "RunningAveragePowerLimit", +}; +static const int config_keys_num = STATIC_ARRAY_SIZE (config_keys); /***************************** * MSR Manipulation helpers * @@ -274,7 +236,7 @@ enum return_values { * Can change the scheduling affinity of the current process if multiple_read is 1 */ static int __attribute__((warn_unused_result)) -open_msr(int cpu, _Bool multiple_read) +open_msr(unsigned int cpu, _Bool multiple_read) { char pathname[32]; int fd; @@ -290,15 +252,17 @@ open_msr(int cpu, _Bool multiple_read) 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("Could not migrate to CPU %d", cpu); - return -ERR_CPU_MIGRATE; + ERROR("turbostat plugin: Could not migrate to CPU %d", cpu); + return -1; } } ssnprintf(pathname, sizeof(pathname), "/dev/cpu/%d/msr", cpu); fd = open(pathname, O_RDONLY); - if (fd < 0) - return -ERR_CANT_OPEN_MSR; + if (fd < 0) { + ERROR("turbostat plugin: failed to open %s", pathname); + return -1; + } return fd; } @@ -313,7 +277,8 @@ read_msr(int fd, off_t offset, unsigned long long *msr) retval = pread(fd, msr, sizeof *msr, offset); if (retval != sizeof *msr) { - ERROR("MSR offset 0x%llx read failed", (unsigned long long)offset); + ERROR("turbostat plugin: MSR offset 0x%llx read failed", + (unsigned long long)offset); return -1; } return 0; @@ -323,8 +288,8 @@ read_msr(int fd, off_t offset, unsigned long long *msr) * 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 int __attribute__((warn_unused_result)) -get_msr(int cpu, off_t offset, unsigned long long *msr) +static ssize_t __attribute__((warn_unused_result)) +get_msr(unsigned int cpu, off_t offset, unsigned long long *msr) { ssize_t retval; int fd; @@ -353,7 +318,7 @@ get_msr(int cpu, off_t offset, unsigned long long *msr) static int __attribute__((warn_unused_result)) get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { - int cpu = t->cpu_id; + unsigned int cpu = t->cpu_id; unsigned long long msr; int msr_fd; int retval = 0; @@ -362,12 +327,13 @@ get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (msr_fd < 0) return msr_fd; -#define READ_MSR(msr, dst) \ -do { \ - if (read_msr(msr_fd, msr, dst)) { \ - retval = -ERR_##msr; \ - goto out; \ - } \ +#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); @@ -375,8 +341,10 @@ do { \ READ_MSR(MSR_IA32_APERF, &t->aperf); READ_MSR(MSR_IA32_MPERF, &t->mperf); - READ_MSR(MSR_SMI_COUNT, &msr); - t->smi_count = msr & 0xFFFFFFFF; + 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)) { @@ -402,20 +370,20 @@ do { \ 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_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); @@ -433,14 +401,6 @@ do { \ READ_MSR(MSR_PP1_ENERGY_STATUS, &msr); p->energy_gfx = msr & 0xFFFFFFFF; } - if (do_rapl & RAPL_PKG_PERF_STATUS) { - READ_MSR(MSR_PKG_PERF_STATUS, &msr); - p->rapl_pkg_perf_status = msr & 0xFFFFFFFF; - } - if (do_rapl & RAPL_DRAM_PERF_STATUS) { - READ_MSR(MSR_DRAM_PERF_STATUS, &msr); - p->rapl_dram_perf_status = msr & 0xFFFFFFFF; - } if (do_ptm) { READ_MSR(MSR_IA32_PACKAGE_THERM_STATUS, &msr); p->pkg_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F); @@ -457,16 +417,6 @@ out: **********************************/ /* - * Do delta = new - old on 32bits cyclique intergers - */ -#define DELTA_WRAP32(delta, new, old) \ - if (new > old) { \ - delta = new - old; \ - } else { \ - delta = 0x100000000 + new - old; \ - } - -/* * Extract the evolution old->new in delta at a package level * (some are not new-delta, e.g. temperature) */ @@ -482,12 +432,10 @@ delta_package(struct pkg_data *delta, const struct pkg_data *new, const struct p delta->pc10 = new->pc10 - old->pc10; delta->pkg_temp_c = new->pkg_temp_c; - DELTA_WRAP32(delta->energy_pkg, new->energy_pkg, old->energy_pkg); - DELTA_WRAP32(delta->energy_cores, new->energy_cores, old->energy_cores); - DELTA_WRAP32(delta->energy_gfx, new->energy_gfx, old->energy_gfx); - DELTA_WRAP32(delta->energy_dram, new->energy_dram, old->energy_dram); - DELTA_WRAP32(delta->rapl_pkg_perf_status, new->rapl_pkg_perf_status, old->rapl_pkg_perf_status); - DELTA_WRAP32(delta->rapl_dram_perf_status, new->rapl_dram_perf_status, old->rapl_dram_perf_status); + 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; } /* @@ -515,9 +463,10 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str /* check for TSC < 1 Mcycles over interval */ if (delta->tsc < (1000 * 1000)) { - WARNING("Insanely slow TSC rate, TSC stops in idle? "); - WARNING("You can disable all c-states by booting with \"idle=poll\" "); - WARNING("or just the deep ones with \"processor.max_cstate=1\""); + 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; } @@ -528,9 +477,10 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str delta->mperf = new->mperf - old->mperf; } else { if (!aperf_mperf_unstable) { - WARNING(" APERF or MPERF went backwards * "); - WARNING("* Frequency results do not cover entire interval *"); - WARNING("* fix this by running Linux-2.6.30 or later *"); + 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 = 1; } @@ -550,11 +500,12 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str } if (delta->mperf == 0) { - WARNING("cpu%d MPERF 0!", old->cpu_id); + WARNING("turbostat plugin: cpu%d MPERF 0!", old->cpu_id); delta->mperf = 1; /* divide by 0 protection */ } - delta->smi_count = new->smi_count - old->smi_count; + if (do_smi) + delta->smi_count = new->smi_count - old->smi_count; return 0; } @@ -597,10 +548,10 @@ turbostat_submit (const char *plugin_instance, static int submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { - char name[12]; + char name[DATA_MAX_NAME_LEN]; double interval_float; - interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; + interval_float = CDTIME_T_TO_DOUBLE(time_delta); ssnprintf(name, sizeof(name), "cpu%02d", t->cpu_id); @@ -609,12 +560,17 @@ submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (!aperf_mperf_unstable) turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc); - /* GHz */ + 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(NULL, "frequency", name, 1.0 * t->tsc / 1000000000 * t->aperf / t->mperf / interval_float); + 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 */ - turbostat_submit(NULL, "current", name, t->smi_count); + 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)) @@ -630,7 +586,7 @@ submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc); if (do_dts) - turbostat_submit(NULL, "temperature", name, c->core_temp_c); + 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)) @@ -639,7 +595,7 @@ submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) ssnprintf(name, sizeof(name), "pkg%02d", p->package_id); if (do_ptm) - turbostat_submit(NULL, "temperature", name, p->pkg_temp_c); + 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); @@ -658,13 +614,13 @@ submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (do_rapl) { if (do_rapl & RAPL_PKG) - turbostat_submit(name, "power", "Pkg_W", p->energy_pkg * rapl_energy_units / interval_float); + turbostat_submit(name, "power", "pkg", p->energy_pkg * rapl_energy_units / interval_float); if (do_rapl & RAPL_CORES) - turbostat_submit(name, "power", "Cor_W", p->energy_cores * rapl_energy_units / interval_float); + turbostat_submit(name, "power", "cores", p->energy_cores * rapl_energy_units / interval_float); if (do_rapl & RAPL_GFX) - turbostat_submit(name, "power", "GFX_W", p->energy_gfx * rapl_energy_units / interval_float); + turbostat_submit(name, "power", "GFX", p->energy_gfx * rapl_energy_units / interval_float); if (do_rapl & RAPL_DRAM) - turbostat_submit(name, "power", "RAM_W", p->energy_dram * rapl_energy_units / interval_float); + turbostat_submit(name, "power", "DRAM", p->energy_dram * rapl_energy_units / interval_float); } done: return 0; @@ -679,7 +635,7 @@ done: * Check if a given cpu id is in our compiled list of existing CPUs */ static int -cpu_is_not_present(int cpu) +cpu_is_not_present(unsigned int cpu) { return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); } @@ -694,7 +650,8 @@ 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, pkg_no, core_no, thread_no; + int retval; + unsigned int pkg_no, core_no, thread_no; for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) { for (core_no = 0; core_no < topology.num_cores; ++core_no) { @@ -733,7 +690,8 @@ 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, pkg_no, core_no, thread_no; + int retval; + unsigned int pkg_no, core_no, thread_no; for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) { for (core_no = 0; core_no < topology.num_cores; ++core_no) { @@ -787,255 +745,9 @@ for_all_cpus_delta(const struct thread_data *thread_new_base, const struct core_ } -static void -free_all_buffers(void) -{ - allocated = 0; - initialized = 0; - - CPU_FREE(cpu_present_set); - cpu_present_set = NULL; - cpu_present_set = 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; -} - - -/**************** - * File helpers * - ****************/ - -/* - * 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]; - FILE *filep; - int value; - - va_start(args, fmt); - vsnprintf(path, sizeof(path), fmt, args); - va_end(args); - filep = fopen(path, "r"); - if (!filep) { - ERROR("%s: open failed", path); - return -ERR_CANT_OPEN_FILE; - } - if (fscanf(filep, "%d", &value) != 1) { - ERROR("%s: failed to parse number from file", path); - return -ERR_CANT_READ_NUMBER; - } - fclose(filep); - return value; -} - -static int -get_threads_on_core(int cpu) -{ - char path[80]; - FILE *filep; - int sib1, sib2; - int matches; - char character; - - ssnprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); - filep = fopen(path, "r"); - if (!filep) { - ERROR("%s: open failed", path); - return -ERR_CANT_OPEN_FILE; - } - /* - * 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)(int)) -{ - FILE *fp; - int cpu_num; - int retval; - - fp = fopen("/proc/stat", "r"); - if (!fp) { - ERROR("Failed to open /proc/stat"); - return -ERR_CANT_OPEN_FILE; - } - - retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); - if (retval != 0) { - ERROR("Failed to parse /proc/stat"); - fclose(fp); - return -ERR_CANT_READ_PROC_STAT; - } - - 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(int cpu) -{ - if (topology.max_cpu_id < cpu) - topology.max_cpu_id = cpu; - return 0; -} - -static int -mark_cpu_present(int cpu) -{ - CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); - return 0; -} - - -static int setup_all_buffers(void); - -static int -turbostat_read(user_data_t * not_used) -{ - 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)) - return -ERR_CPU_NOT_PRESENT; - } - - /* Saving 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 -ERR_CPU_SAVE_SCHED_AFFINITY; - - if (!initialized) { - if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0) - goto out; - gettimeofday(&tv_even, (struct timezone *)NULL); - is_even = 1; - initialized = 1; - ret = 0; - goto out; - } - - if (is_even) { - if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0) - goto out; - gettimeofday(&tv_odd, (struct timezone *)NULL); - is_even = 0; - timersub(&tv_odd, &tv_even, &tv_delta); - 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; - gettimeofday(&tv_even, (struct timezone *)NULL); - is_even = 1; - timersub(&tv_even, &tv_odd, &tv_delta); - 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: - /* - * Let's restore the affinity - * This might fail if the number of CPU changed, but we can't do anything in that case.. - */ - (void)sched_setaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set); - return ret; -} - -static int __attribute__((warn_unused_result)) -check_dev_msr() -{ - struct stat sb; - - if (stat("/dev/cpu/0/msr", &sb)) { - ERROR("no /dev/cpu/0/msr, try \"# modprobe msr\""); - return -ERR_NO_MSR; - } - return 0; -} - -static int __attribute__((warn_unused_result)) -check_super_user() -{ - if (getuid() != 0) { - ERROR("must be root"); - return -ERR_NOT_ROOT; - } - return 0; -} +/*************** + * CPU Probing * + ***************/ /* * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where @@ -1082,7 +794,8 @@ set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_da guess: p->tcc_activation_temp = TJMAX_DEFAULT; - WARNING("cpu%d: Guessing tjMax %d C, Please use TCCActivationTemp to specify", + WARNING("turbostat plugin: cpu%d: Guessing tjMax %d C," + " Please use TCCActivationTemp to specify it.", t->cpu_id, p->tcc_activation_temp); return 0; @@ -1101,25 +814,25 @@ probe_cpu() * - EAX: Maximum Input Value for Basic CPUID Information * - EBX: "Genu" (0x756e6547) * - EDX: "ineI" (0x49656e69) - * - ECX: "ntel" (0x6c65746e) - */ + * - 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("Unsupported CPU"); - return -UNSUPPORTED_CPU; + 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 + * + 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; @@ -1129,37 +842,8 @@ probe_cpu() if (family == 6 || family == 0xf) model += ((fms >> 16) & 0xf) << 4; if (!(edx & (1 << 5))) { - ERROR("CPUID: no MSR"); - return -ERR_NO_MSR; - } - - /* - * CPUID(0x80000000): - * - EAX: Maximum Input Value for Extended Function CPUID Information - * - * This allows us to verify if the CPUID(0x80000007) can be called - * - * This check is valid for both Intel and AMD. - */ - max_level = ebx = ecx = edx = 0; - __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx); - if (max_level < 0x80000007) { - ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level); - return -ERR_NO_INVARIANT_TSC; - } - - /* - * CPUID(0x80000007): - * - EDX: - * + 8: Invariant TSC available if set - * - * This check is valid for both Intel and AMD - */ - eax = ebx = ecx = edx = 0; - __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx); - if (!(edx & (1 << 8))) { - ERROR("No invariant TSC"); - return -ERR_NO_INVARIANT_TSC; + ERROR("turbostat plugin: Unsupported CPU (no MSR support)"); + return -1; } /* @@ -1179,8 +863,8 @@ probe_cpu() do_dts = eax & (1 << 0); do_ptm = eax & (1 << 6); if (!(ecx & (1 << 0))) { - ERROR("No APERF"); - return -ERR_NO_APERF; + ERROR("turbostat plugin: Unsupported CPU (No APERF)"); + return -1; } /* @@ -1190,15 +874,14 @@ probe_cpu() switch (model) { /* Atom (partial) */ case 0x27: + do_smi = 0; do_core_cstate = 0; do_pkg_cstate = (1 << 2) | (1 << 4) | (1 << 6); break; /* Silvermont */ case 0x37: /* BYT */ - case 0x4A: case 0x4D: /* AVN */ - case 0x5A: - case 0x5D: + do_smi = 1; do_core_cstate = (1 << 1) | (1 << 6); do_pkg_cstate = (1 << 6); break; @@ -1207,6 +890,7 @@ probe_cpu() 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 = 1; do_core_cstate = (1 << 3) | (1 << 6); do_pkg_cstate = (1 << 3) | (1 << 6) | (1 << 7); break; @@ -1214,18 +898,21 @@ probe_cpu() case 0x25: /* Westmere Client - Clarkdale, Arrandale */ case 0x2C: /* Westmere EP - Gulftown */ case 0x2F: /* Westmere-EX Xeon - Eagleton */ + do_smi = 1; 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 = 1; 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 */ + do_smi = 1; do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7); do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7); break; @@ -1233,53 +920,78 @@ probe_cpu() case 0x3C: /* HSW */ case 0x3F: /* HSW */ case 0x46: /* HSW */ + do_smi = 1; 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 = 1; 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; /* Broadwel */ case 0x4F: /* BDW */ case 0x56: /* BDX-DE */ + do_smi = 1; 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 = 1; 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: - ERROR("Unsupported CPU"); + do_smi = 0; + do_core_cstate = 0; + do_pkg_cstate = 0; + break; } switch (model) { - case 0x2A: - case 0x3A: - case 0x3C: - case 0x45: - case 0x46: - do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_PKG_POWER_INFO | RAPL_GFX; + case 0x2A: /* SNB */ + case 0x3A: /* IVB */ + case 0x3C: /* HSW */ + case 0x45: /* HSW */ + case 0x46: /* HSW */ + case 0x3D: /* BDW */ + do_rapl = RAPL_PKG | RAPL_CORES | RAPL_GFX; break; - case 0x3F: - do_rapl = RAPL_PKG | RAPL_PKG_POWER_INFO | RAPL_PKG_PERF_STATUS | RAPL_DRAM | RAPL_DRAM_PERF_STATUS; + case 0x3F: /* HSX */ + case 0x4F: /* BDX */ + case 0x56: /* BDX-DE */ + do_rapl = RAPL_PKG | RAPL_DRAM ; break; - case 0x2D: - case 0x3E: - do_rapl = RAPL_PKG | RAPL_CORES | RAPL_CORE_POLICY | RAPL_PKG_POWER_INFO | RAPL_PKG_PERF_STATUS | RAPL_DRAM | RAPL_DRAM_PERF_STATUS; + case 0x2D: /* SNB Xeon */ + case 0x3E: /* IVB Xeon */ + do_rapl = RAPL_PKG | RAPL_CORES | RAPL_DRAM; break; - case 0x37: - case 0x4D: + case 0x37: /* BYT */ + case 0x4D: /* AVN */ do_rapl = RAPL_PKG | RAPL_CORES; break; default: do_rapl = 0; } } else { - ERROR("Unsupported CPU"); - return -UNSUPPORTED_CPU; + 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)) @@ -1294,15 +1006,139 @@ probe_cpu() 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]; + FILE *filep; + int len, value; + + 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; + } + + filep = fopen(path, "r"); + if (!filep) { + ERROR("turbostat plugin: Failed to open '%s'", path); + return -1; + } + if (fscanf(filep, "%d", &value) != 1) { + ERROR("turbostat plugin: Failed to parse number from '%s'", path); + return -1; + } + fclose(filep); + return value; +} + +static int +get_threads_on_core(unsigned int cpu) +{ + char path[80]; + FILE *filep; + int sib1, sib2; + int matches; + char character; + + ssnprintf(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("Unable to allocate CPU state"); - return -ERR_CPU_ALLOC; +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); + CPU_ZERO_S(*size, *set); return 0; } @@ -1312,47 +1148,49 @@ allocate_cpu_set(cpu_set_t * set, size_t * size) { static int __attribute__((warn_unused_result)) topology_probe() { - int i; + unsigned int i; int ret; - int max_package_id, max_core_id, max_thread_id; - max_package_id = max_core_id = max_thread_id = 0; + 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)); - /* Can't fail (update_max_cpu_id always returns 0) */ - assert(for_all_proc_cpus(update_max_cpu_id)); + 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("Unable to allocate memory for cpu topology"); - return -ERR_CALLOC; + ERROR("turbostat plugin: Unable to allocate memory for CPU topology"); + return -1; } - ret = allocate_cpu_set(cpu_present_set, &cpu_present_setsize); + 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); + 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); + ret = allocate_cpu_set(&cpu_saved_affinity_set, &cpu_saved_affinity_setsize); if (ret != 0) goto err; - /* Can't fail (mark_cpu_present always returns 0) */ - assert(for_all_proc_cpus(mark_cpu_present)); + ret = for_all_proc_cpus(mark_cpu_present); + if (ret != 0) + goto err; /* * For online cpus * find max_core_id, max_package_id */ for (i = 0; i <= topology.max_cpu_id; ++i) { - int num_threads; + unsigned int num_threads; struct cpu_topology *cpu = &topology.cpus[i]; if (cpu_is_not_present(i)) { - WARNING("cpu%d NOT PRESENT", i); + WARNING("turbostat plugin: cpu%d NOT PRESENT", i); continue; } @@ -1360,7 +1198,7 @@ topology_probe() if (ret < 0) goto err; else - cpu->package_id = ret; + cpu->package_id = (unsigned int) ret; if (cpu->package_id > max_package_id) max_package_id = cpu->package_id; @@ -1368,36 +1206,35 @@ topology_probe() if (ret < 0) goto err; else - cpu->core_id = ret; + 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 (ret == cpu->core_id) + else if ((unsigned int) ret == i) cpu->first_core_in_package = 1; ret = get_threads_on_core(i); if (ret < 0) goto err; else - num_threads = ret; - if (num_threads > max_thread_id) - max_thread_id = num_threads; - if (num_threads > 1) { - ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", i); - if (ret < 0) - goto err; - else if (ret == num_threads) - cpu->first_thread_in_core = 1; - } - DEBUG("cpu %d pkg %d core %d\n", + 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 = 1; + + 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_thread_id + 1; + topology.num_threads = max_threads; return 0; err: @@ -1405,11 +1242,16 @@ err: return ret; } + +/************************ + * Main alloc/init/free * + ************************/ + static int allocate_counters(struct thread_data **threads, struct core_data **cores, struct pkg_data **packages) { - int i; - int total_threads, total_cores; + unsigned int i; + unsigned int total_threads, total_cores; total_threads = topology.num_threads * topology.num_cores * topology.num_packages; *threads = calloc(total_threads, sizeof(struct thread_data)); @@ -1417,23 +1259,17 @@ allocate_counters(struct thread_data **threads, struct core_data **cores, struct goto err; for (i = 0; i < total_threads; ++i) - (*threads)[i].cpu_id = -1; + (*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) goto err_clean_threads; - for (i = 0; i < total_cores; ++i) - (*cores)[i].core_id = -1; - *packages = calloc(topology.num_packages, sizeof(struct pkg_data)); if (*packages == NULL) goto err_clean_cores; - for (i = 0; i < topology.num_packages; i++) - (*packages)[i].package_id = i; - return 0; err_clean_cores: @@ -1441,21 +1277,13 @@ err_clean_cores: err_clean_threads: free(*threads); err: - ERROR("calloc counters"); - return -ERR_CALLOC; + ERROR("turbostat plugin: Failled to allocate memory for counters"); + return -1; } -/* - * init_counter() - * - * set cpu_id, core_id, package_id - * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE - * - * increment topo.num_cores when 1st core in pkg seen - */ -static int +static void init_counter(struct thread_data *thread_base, struct core_data *core_base, - struct pkg_data *pkg_base, int cpu_id) + struct pkg_data *pkg_base, unsigned int cpu_id) { struct thread_data *t; struct core_data *c; @@ -1474,40 +1302,77 @@ init_counter(struct thread_data *thread_base, struct core_data *core_base, c->core_id = cpu->core_id; p->package_id = cpu->package_id; - - return 0; } - -static int +static void initialize_counters(void) { - int ret; - int cpu_id; + unsigned int cpu_id; for (cpu_id = 0; cpu_id <= topology.max_cpu_id; ++cpu_id) { - if (cpu_is_not_present(cpu_id)) { + if (cpu_is_not_present(cpu_id)) continue; - } - - ret = init_counter(EVEN_COUNTERS, cpu_id); - if (ret < 0) - return ret; - ret = init_counter(ODD_COUNTERS, cpu_id); - if (ret < 0) - return ret; - ret = init_counter(DELTA_COUNTERS, cpu_id); - if (ret < 0) - return ret; + init_counter(EVEN_COUNTERS, cpu_id); + init_counter(ODD_COUNTERS, cpu_id); + init_counter(DELTA_COUNTERS, cpu_id); } - return 0; } + + +static void +free_all_buffers(void) +{ + allocated = 0; + initialized = 0; + + CPU_FREE(cpu_present_set); + cpu_present_set = NULL; + cpu_present_set = 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; \ +do { \ + ret = (something); \ + if (ret < 0) \ + goto err; \ } while (0) static int setup_all_buffers(void) @@ -1518,7 +1383,7 @@ static int setup_all_buffers(void) 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)); - DO_OR_GOTO_ERR(initialize_counters()); + 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)); @@ -1530,16 +1395,150 @@ err: } 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; + } + } + + /* Saving the scheduling affinity, as it will be modified by get_counters */ + if (sched_getaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set) != 0) { + ERROR("turbostat plugin: Unable to save the CPU affinity"); + return -1; + } + + if (!initialized) { + if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0) + goto out; + time_even = cdtime(); + is_even = 1; + initialized = 1; + ret = 0; + goto out; + } + + if (is_even) { + if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0) + goto out; + time_odd = cdtime(); + is_even = 0; + 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 = 1; + 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: + /* + * Let's restore the affinity + * This might fail if the number of CPU changed, but we can't do anything in that case.. + */ + (void)sched_setaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set); + return ret; +} + +static int +check_permissions(void) +{ +#ifdef HAVE_SYS_CAPABILITY_H + struct __user_cap_header_struct cap_header_data; + cap_user_header_t cap_header = &cap_header_data; + struct __user_cap_data_struct cap_data_data; + cap_user_data_t cap_data = &cap_data_data; + int ret = 0; +#endif /* HAVE_SYS_CAPABILITY_H */ + + if (getuid() == 0) { + /* We have everything we need */ + return 0; +#ifndef HAVE_SYS_CAPABILITY_H + } else { + ERROR("turbostat plugin: Initialization failed: this plugin " + "requires collectd to run as root"); + return -1; + } +#else /* HAVE_SYS_CAPABILITY_H */ + } + + /* check for CAP_SYS_RAWIO */ + cap_header->pid = getpid(); + cap_header->version = _LINUX_CAPABILITY_VERSION; + if (capget(cap_header, cap_data) < 0) { + ERROR("turbostat plugin: capget failed"); + return -1; + } + + if ((cap_data->effective & (1 << 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 */ +} + +static int turbostat_init(void) { + struct stat sb; int ret; - DO_OR_GOTO_ERR(check_super_user()); + 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(check_dev_msr()); + DO_OR_GOTO_ERR(setup_all_buffers()); - plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, NULL, NULL); + plugin_register_read(PLUGIN_NAME, turbostat_read); return 0; err: @@ -1547,30 +1546,64 @@ err: return ret; } -static const char *config_keys[] = -{ - "TCCActivationTemp", -}; -static const int config_keys_num = STATIC_ARRAY_SIZE (config_keys); - static int turbostat_config(const char *key, const char *value) { long unsigned int tmp_val; char *end; - if (strcasecmp("TCCActivationTemp", key) == 0) { + if (strcasecmp("CoreCstates", key) == 0) { tmp_val = strtoul(value, &end, 0); - if (*end != '\0' || tmp_val > UINT_MAX) + 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 = 1; + } 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 = 1; + } else if (strcasecmp("SystemManagementInterrupt", key) == 0) { + config_smi = IS_TRUE(value); + apply_config_smi = 1; + } else if (strcasecmp("DigitalTemperatureSensor", key) == 0) { + config_dts = IS_TRUE(value); + apply_config_dts = 1; + } else if (strcasecmp("PackageThermalManagement", key) == 0) { + config_ptm = IS_TRUE(value); + apply_config_ptm = 1; + } 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 = 1; + } 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 { + ERROR("turbostat plugin: Invalid configuration option '%s'", + key); return -1; } return 0; } -void module_register(void); void module_register(void) { plugin_register_init(PLUGIN_NAME, turbostat_init);