*/
#define _GNU_SOURCE
-#include <asm/msr-index.h>
-#include <stdarg.h>
-#include <stdio.h>
-#include <err.h>
-#include <unistd.h>
-#include <sys/types.h>
-#include <sys/wait.h>
-#include <sys/stat.h>
-#include <sys/resource.h>
-#include <fcntl.h>
-#include <signal.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <dirent.h>
-#include <string.h>
-#include <ctype.h>
-#include <sched.h>
-#include <cpuid.h>
-
#include "collectd.h"
#include "common.h"
#include "plugin.h"
+#include "utils_time.h"
+
+#include <asm/msr-index.h>
+#include <cpuid.h>
+#ifdef HAVE_SYS_CAPABILITY_H
+#include <sys/capability.h>
+#endif /* HAVE_SYS_CAPABILITY_H */
#define PLUGIN_NAME "turbostat"
-static const char *proc_stat = "/proc/stat";
-static unsigned int skip_c0;
-static unsigned int skip_c1;
+/*
+ * 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;
+
+/*
+ * 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 do_rapl;
-static unsigned int do_dts;
-static unsigned int do_ptm;
+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 tcc_activation_temp_override;
+
+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
-int aperf_mperf_unstable;
-int backwards_count;
-char *progname;
+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;
-cpu_set_t *cpu_present_set, *cpu_affinity_set;
-size_t cpu_present_setsize, cpu_affinity_setsize;
-
-struct thread_data {
+static struct thread_data {
unsigned long long tsc;
unsigned long long aperf;
unsigned long long mperf;
unsigned int flags;
#define CPU_IS_FIRST_THREAD_IN_CORE 0x2
#define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4
-} *thread_even, *thread_odd;
+} *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;
unsigned int core_temp_c;
unsigned int core_id;
-} *core_even, *core_odd;
+} *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;
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;
-} *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 = 1;
static _Bool initialized = 0;
#define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
- (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
- topo.num_threads_per_core + \
- (core_no) * topo.num_threads_per_core + (thread_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) * topo.num_cores_per_pkg + (core_no))
+ (core_base + \
+ (pkg_no) * topology.num_cores + \
+ (core_no))
#define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
-struct topo_params {
- int num_packages;
- int num_cpus;
- int num_cores;
- int max_cpu_num;
- int num_cores_per_pkg;
- int num_threads_per_core;
-} topo;
-
-struct timeval tv_even, tv_odd, tv_delta;
-
-enum return_values {
- OK = 0,
- ERR_CPU_MIGRATE,
- 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,
+struct cpu_topology {
+ unsigned int package_id;
+ unsigned int core_id;
+ _Bool first_core_in_package;
+ _Bool first_thread_in_core;
};
-static int setup_all_buffers(void);
+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 int
-cpu_is_not_present(int cpu)
-{
- return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
-}
-/*
- * run func(thread, core, package) in topology order
- * skip non-present cpus
- */
+static cdtime_t time_even, time_odd, time_delta;
-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)
+static const char *config_keys[] =
{
- int retval, pkg_no, core_no, thread_no;
-
- for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
- for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
- for (thread_no = 0; thread_no <
- topo.num_threads_per_core; ++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;
-}
+ "CoreCstates",
+ "PackageCstates",
+ "SystemManagementInterrupt",
+ "DigitalTemperatureSensor",
+ "PackageThermalManagement",
+ "TCCActivationTemp",
+ "RunningAveragePowerLimit",
+};
+static const int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
-static int __attribute__((warn_unused_result))
-cpu_migrate(int cpu)
-{
- 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)
- return -ERR_CPU_MIGRATE;
- else
- return 0;
-}
+/*****************************
+ * MSR Manipulation helpers *
+ *****************************/
+/*
+ * Open a MSR device for reading
+ * Can change the scheduling affinity of the current process if multiple_read is 1
+ */
static int __attribute__((warn_unused_result))
-open_msr(int cpu)
+open_msr(unsigned int cpu, _Bool multiple_read)
{
char pathname[32];
int fd;
- /* FIXME: Do we really need this, why? */
- if (cpu_migrate(cpu)) {
- ERROR("Could not migrate to CPU %d\n", cpu);
- return -ERR_CPU_MIGRATE;
+ /*
+ * 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;
+ }
}
- ssnprintf(pathname, STATIC_ARRAY_SIZE(pathname), "/dev/cpu/%d/msr", cpu);
+ 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;
}
+/*
+ * 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)
{
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;
}
-static int __attribute__((warn_unused_result))
-get_msr(int cpu, 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 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);
+ fd = open_msr(cpu, 0);
if (fd < 0)
return fd;
retval = read_msr(fd, offset, msr);
return retval;
}
-#define DELTA_WRAP32(new, old) \
- if (new > old) { \
- old = new - old; \
- } else { \
- old = 0x100000000 + new - old; \
- }
-
-static void
-delta_package(struct pkg_data *new, struct pkg_data *old)
-{
- old->pc2 = new->pc2 - old->pc2;
- old->pc3 = new->pc3 - old->pc3;
- old->pc6 = new->pc6 - old->pc6;
- old->pc7 = new->pc7 - old->pc7;
- old->pc8 = new->pc8 - old->pc8;
- old->pc9 = new->pc9 - old->pc9;
- old->pc10 = new->pc10 - old->pc10;
- old->pkg_temp_c = new->pkg_temp_c;
-
- DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
- DELTA_WRAP32(new->energy_cores, old->energy_cores);
- DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
- DELTA_WRAP32(new->energy_dram, old->energy_dram);
- DELTA_WRAP32(new->rapl_pkg_perf_status, old->rapl_pkg_perf_status);
- DELTA_WRAP32(new->rapl_dram_perf_status, old->rapl_dram_perf_status);
-}
-
-static void
-delta_core(struct core_data *new, struct core_data *old)
-{
- old->c3 = new->c3 - old->c3;
- old->c6 = new->c6 - old->c6;
- old->c7 = new->c7 - old->c7;
- old->core_temp_c = new->core_temp_c;
-}
-
-/*
- * old = new - old
- */
-static int __attribute__((warn_unused_result))
-delta_thread(struct thread_data *new, struct thread_data *old,
- struct core_data *core_delta)
-{
- old->tsc = new->tsc - old->tsc;
-
- /* check for TSC < 1 Mcycles over interval */
- if (old->tsc < (1000 * 1000)) {
- WARNING("Insanely slow TSC rate, TSC stops in idle?\n"
- "You can disable all c-states by booting with \"idle=poll\"\n"
- "or just the deep ones with \"processor.max_cstate=1\"");
- return -1;
- }
-
- old->c1 = new->c1 - old->c1;
-
- if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
- old->aperf = new->aperf - old->aperf;
- old->mperf = new->mperf - old->mperf;
- } else {
-
- if (!aperf_mperf_unstable) {
- WARNING("%s: APERF or MPERF went backwards *\n", progname);
- WARNING("* Frequency results do not cover entire interval *\n");
- WARNING("* fix this by running Linux-2.6.30 or later *\n");
-
- aperf_mperf_unstable = 1;
- }
- /*
- * mperf delta is likely a huge "positive" number
- * can not use it for calculating c0 time
- */
- skip_c0 = 1;
- skip_c1 = 1;
- }
-
-
- /*
- * 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 ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc)
- old->c1 = 0;
- else {
- /* normal case, derive c1 */
- old->c1 = old->tsc - old->mperf - core_delta->c3
- - core_delta->c6 - core_delta->c7;
- }
-
- if (old->mperf == 0) {
- WARNING("cpu%d MPERF 0!\n", old->cpu_id);
- old->mperf = 1; /* divide by 0 protection */
- }
-
- old->smi_count = new->smi_count - old->smi_count;
-
- return 0;
-}
-
-static int __attribute__((warn_unused_result))
-delta_cpu(struct thread_data *t, struct core_data *c,
- struct pkg_data *p, struct thread_data *t2,
- struct core_data *c2, struct pkg_data *p2)
-{
- int ret;
-
- /* calculate core delta only for 1st thread in core */
- if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE)
- delta_core(c, c2);
-
- /* always calculate thread delta */
- ret = delta_thread(t, t2, c2); /* c2 is core delta */
- if (ret != 0)
- return ret;
-
- /* calculate package delta only for 1st core in package */
- if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
- delta_package(p, p2);
- return 0;
-}
+/********************************
+ * Raw data acquisition (1 CPU) *
+ ********************************/
/*
- * get_counters(...)
- * migrate to cpu
- * acquire and record local counters for that 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)
{
- int cpu = t->cpu_id;
+ unsigned int cpu = t->cpu_id;
unsigned long long msr;
int msr_fd;
int retval = 0;
- msr_fd = open_msr(cpu);
+ 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)) { \
- 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);
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)) {
if (do_dts) {
READ_MSR(MSR_IA32_THERM_STATUS, &msr);
- c->core_temp_c = tcc_activation_temp - ((msr >> 16) & 0x7F);
+ c->core_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
}
/* collect package counters only for 1st core in package */
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);
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 = tcc_activation_temp - ((msr >> 16) & 0x7F);
+ p->pkg_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
}
out:
return retval;
}
-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;
-
- 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;
-}
+/**********************************
+ * Evaluating the changes (1 CPU) *
+ **********************************/
/*
- * Parse a file containing a single int.
+ * Extract the evolution old->new in delta at a package level
+ * (some are not new-delta, e.g. temperature)
*/
-static int __attribute__ ((format(printf,1,2)))
-parse_int_file(const char *fmt, ...)
+static inline void
+delta_package(struct pkg_data *delta, const struct pkg_data *new, const struct pkg_data *old)
{
- 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;
+ 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;
}
/*
- * cpu_is_first_sibling_in_core(cpu)
- * return 1 if given CPU is 1st HT sibling in the core
+ * Extract the evolution old->new in delta at a core level
+ * (some are not new-delta, e.g. temperature)
*/
-static int
-cpu_is_first_sibling_in_core(int cpu)
+static inline void
+delta_core(struct core_data *delta, const struct core_data *new, const struct core_data *old)
{
- return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
+ 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;
}
/*
- * cpu_is_first_core_in_package(cpu)
- * return 1 if given CPU is 1st core in package
+ * Extract the evolution old->new in delta at a package level
+ * core_delta is required for c1 estimation (tsc - c0 - all core cstates)
*/
-static int
-cpu_is_first_core_in_package(int cpu)
+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 *core_delta)
{
- return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
-}
+ delta->tsc = new->tsc - old->tsc;
-static int
-get_physical_package_id(int cpu)
-{
- return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
-}
+ /* 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;
+ }
-static int
-get_core_id(int cpu)
-{
- return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
-}
+ delta->c1 = new->c1 - old->c1;
-static int
-get_num_ht_siblings(int cpu)
-{
- char path[80];
- FILE *filep;
- int sib1, sib2;
- int matches;
- char character;
+ 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 = 1;
+ }
+ }
- ssnprintf(path, 80, "/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).
+ * 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.
*/
- matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
+ if ((delta->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > delta->tsc)
+ delta->c1 = 0;
+ else {
+ /* normal case, derive c1 */
+ delta->c1 = delta->tsc - delta->mperf - core_delta->c3
+ - core_delta->c6 - core_delta->c7;
+ }
- fclose(filep);
+ if (delta->mperf == 0) {
+ WARNING("turbostat plugin: cpu%d MPERF 0!", old->cpu_id);
+ delta->mperf = 1; /* divide by 0 protection */
+ }
- if (matches == 3)
- return 2;
- else
- return 1;
+ if (do_smi)
+ delta->smi_count = new->smi_count - old->smi_count;
+
+ return 0;
}
+/**********************************
+ * Submitting the results (1 CPU) *
+ **********************************/
+
/*
- * run func(thread, core, package) in topology order
- * skip non-present cpus
+ * Submit one gauge value
*/
-
-
-static int __attribute__((warn_unused_result))
-for_all_cpus_2(int (func)(struct thread_data *, struct core_data *,
- struct pkg_data *, struct thread_data *, struct core_data *,
- struct pkg_data *), struct thread_data *thread_base,
- struct core_data *core_base, struct pkg_data *pkg_base,
- struct thread_data *thread_base2, struct core_data *core_base2,
- struct pkg_data *pkg_base2)
-{
- int retval, pkg_no, core_no, thread_no;
-
- for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
- for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
- for (thread_no = 0; thread_no <
- topo.num_threads_per_core; ++thread_no) {
- struct thread_data *t, *t2;
- struct core_data *c, *c2;
- struct pkg_data *p, *p2;
-
- t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
-
- if (cpu_is_not_present(t->cpu_id))
- continue;
-
- t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no);
-
- c = GET_CORE(core_base, core_no, pkg_no);
- c2 = GET_CORE(core_base2, core_no, pkg_no);
-
- p = GET_PKG(pkg_base, pkg_no);
- p2 = GET_PKG(pkg_base2, pkg_no);
-
- retval = func(t, c, p, t2, c2, p2);
- if (retval)
- return retval;
- }
- }
- }
- return 0;
-}
-
-/*
- * 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("%s: open failed", 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("%s: failed to parse format", proc_stat);
- 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;
-}
-
-/*
- * count_cpus()
- * remember the last one seen, it will be the max
- */
-static int
-count_cpus(int cpu)
-{
- if (topo.max_cpu_num < cpu)
- topo.max_cpu_num = cpu;
-
- topo.num_cpus += 1;
- return 0;
-}
-static int
-mark_cpu_present(int cpu)
-{
- CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
- return 0;
-}
-
-
static void
turbostat_submit (const char *plugin_instance,
const char *type, const char *type_instance,
}
/*
- * column formatting convention & formats
- * package: "pk" 2 columns %2d
- * core: "cor" 3 columns %3d
- * CPU: "CPU" 3 columns %3d
- * Pkg_W: %6.2
- * Cor_W: %6.2
- * GFX_W: %5.2
- * RAM_W: %5.2
- * GHz: "GHz" 3 columns %3.2
- * TSC: "TSC" 3 columns %3.2
- * SMI: "SMI" 4 columns %4d
- * percentage " %pc3" %6.2
- * Perf Status percentage: %5.2
- * "CTMP" 4 columns %4d
+ * 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
*/
-#define NAME_LEN 12
static int
-submit_counters(struct thread_data *t, struct core_data *c,
- struct pkg_data *p)
+submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
{
- char name[NAME_LEN];
+ 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);
- snprintf(name, NAME_LEN, "cpu%02d", t->cpu_id);
+ ssnprintf(name, sizeof(name), "cpu%02d", t->cpu_id);
- if (!skip_c0)
+ if (!aperf_mperf_unstable)
turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
- if (!skip_c1)
+ 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);
- /* print per-core data only for 1st thread in core */
+ /* submit per-core data only for 1st thread in core */
if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
goto done;
- snprintf(name, NAME_LEN, "core%02d", c->core_id);
+ ssnprintf(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);
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);
- /* print per-package data only for 1st core in package */
+ /* submit per-package data only for 1st core in package */
if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
goto done;
- snprintf(name, NAME_LEN, "pkg%02d", p->package_id);
+ 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);
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;
}
+
+/**********************************
+ * Looping function over all CPUs *
+ **********************************/
+
+/*
+ * Check if a given cpu id is in our compiled list of existing CPUs
+ */
static int
-turbostat_read(user_data_t * not_used)
+cpu_is_not_present(unsigned int cpu)
{
- int ret;
+ return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
+}
- if (!allocated) {
- if ((ret = setup_all_buffers()) < 0)
- return ret;
- }
+/*
+ * 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;
+ unsigned int pkg_no, core_no, thread_no;
- 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;
- }
+ for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
+ for (core_no = 0; core_no < topology.num_cores; ++core_no) {
+ for (thread_no = 0; thread_no < topology.num_threads; ++thread_no) {
+ struct thread_data *t;
+ struct core_data *c;
+ struct pkg_data *p;
- if (!initialized) {
- if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
- return ret;
- gettimeofday(&tv_even, (struct timezone *)NULL);
- is_even = 1;
- initialized = 1;
- return 0;
- }
+ t = GET_THREAD(thread_base, thread_no, core_no, pkg_no);
- if (is_even) {
- if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
- return ret;
- gettimeofday(&tv_odd, (struct timezone *)NULL);
- is_even = 0;
- timersub(&tv_odd, &tv_even, &tv_delta);
- if ((ret = for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) < 0)
- return ret;
- if ((ret = for_all_cpus(submit_counters, EVEN_COUNTERS)) < 0)
- return ret;
- } else {
- if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
- return ret;
- gettimeofday(&tv_even, (struct timezone *)NULL);
- is_even = 1;
- timersub(&tv_even, &tv_odd, &tv_delta);
- if ((ret = for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS)) < 0)
- return ret;
- if ((ret = for_all_cpus(submit_counters, ODD_COUNTERS)) < 0)
- return ret;
- }
- return 0;
-}
+ if (cpu_is_not_present(t->cpu_id))
+ continue;
-static int __attribute__((warn_unused_result))
-check_dev_msr()
-{
- struct stat sb;
+ c = GET_CORE(core_base, core_no, pkg_no);
+ p = GET_PKG(pkg_base, pkg_no);
- if (stat("/dev/cpu/0/msr", &sb)) {
- ERROR("no /dev/cpu/0/msr\n"
- "Try \"# modprobe msr\"");
- return -ERR_NO_MSR;
+ 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))
-check_super_user()
+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)
{
- if (getuid() != 0) {
- ERROR("must be root");
- return -ERR_NOT_ROOT;
+ 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) {
+ for (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 cmdline over-ride with -T.
+ * 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),
if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
return 0;
- if (tcc_activation_temp_override != 0) {
- tcc_activation_temp = tcc_activation_temp_override;
- ERROR("cpu%d: Using cmdline TCC Target (%d C)\n",
- cpu, tcc_activation_temp);
+ 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) & 0x7F;
+ target_c_local = (msr >> 16) & 0xFF;
- if (target_c_local < 85 || target_c_local > 127)
+ if (!target_c_local)
goto guess;
- tcc_activation_temp = target_c_local;
+ p->tcc_activation_temp = target_c_local;
return 0;
guess:
- tcc_activation_temp = TJMAX_DEFAULT;
- WARNING("cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
- t->cpu_id, tcc_activation_temp);
+ 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;
}
* - 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;
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
- */
- __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;
}
/*
*
* 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("No APERF");
- return -ERR_NO_APERF;
+ ERROR("turbostat plugin: Unsupported CPU (No APERF)");
+ return -1;
}
/*
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;
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;
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 << 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 = 1;
do_core_cstate = (1 << 3) | (1 << 6) | (1 << 7);
- do_pkg_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 << 3) | (1 << 6) | (1 << 7) | (1 << 8) | (1 << 9) | (1 << 10);
+ 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 << 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 << 3) | (1 << 6) | (1 << 7) | (1 << 8) | (1 << 9) | (1 << 10);
+ 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 msr;
+ unsigned long long msr;
if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
return 0;
}
+/********************
+ * Topology Probing *
+ ********************/
-static int __attribute__((warn_unused_result))
-topology_probe()
+/*
+ * Read a single int from a file.
+ */
+static int __attribute__ ((format(printf,1,2)))
+parse_int_file(const char *fmt, ...)
{
- int i;
- int ret;
- int max_core_id = 0;
- int max_package_id = 0;
- int max_siblings = 0;
- struct cpu_topology {
- int core_id;
- int physical_package_id;
- } *cpus;
-
- /* Initialize num_cpus, max_cpu_num */
- topo.num_cpus = 0;
- topo.max_cpu_num = 0;
- ret = for_all_proc_cpus(count_cpus);
- if (ret < 0)
- return ret;
-
- DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
-
- cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology));
- if (cpus == NULL) {
- ERROR("calloc cpus");
- return -ERR_CALLOC;
+ 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;
}
- /*
- * Allocate and initialize cpu_present_set
- */
- cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
- if (cpu_present_set == NULL) {
- free(cpus);
- ERROR("CPU_ALLOC");
- return -ERR_CPU_ALLOC;
+ filep = fopen(path, "r");
+ if (!filep) {
+ ERROR("turbostat plugin: Failed to open '%s'", path);
+ return -1;
}
- cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
- CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
- ret = for_all_proc_cpus(mark_cpu_present);
- if (ret < 0) {
- free(cpus);
- return ret;
+ 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;
+ }
/*
- * Allocate and initialize cpu_affinity_set
+ * file format:
+ * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
+ * otherwinse 1 sibling (self).
*/
- cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
- if (cpu_affinity_set == NULL) {
- free(cpus);
- ERROR("CPU_ALLOC");
- return -ERR_CPU_ALLOC;
+ 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;
}
- cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
- CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
+ 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()
+{
+ unsigned int i;
+ 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 (i = 0; i <= topo.max_cpu_num; ++i) {
- int siblings;
+ for (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)) {
- //if (verbose > 1)
- fprintf(stderr, "cpu%d NOT PRESENT\n", i);
+ WARNING("turbostat plugin: cpu%d NOT PRESENT", i);
continue;
}
- cpus[i].core_id = get_core_id(i);
- if (cpus[i].core_id < 0)
- return cpus[i].core_id;
- if (cpus[i].core_id > max_core_id)
- max_core_id = cpus[i].core_id;
-
- cpus[i].physical_package_id = get_physical_package_id(i);
- if (cpus[i].physical_package_id < 0)
- return cpus[i].physical_package_id;
- if (cpus[i].physical_package_id > max_package_id)
- max_package_id = cpus[i].physical_package_id;
-
- siblings = get_num_ht_siblings(i);
- if (siblings < 0)
- return siblings;
- if (siblings > max_siblings)
- max_siblings = siblings;
- DEBUG("cpu %d pkg %d core %d\n",
- i, cpus[i].physical_package_id, cpus[i].core_id);
- }
- topo.num_cores_per_pkg = max_core_id + 1;
- DEBUG("max_core_id %d, sizing for %d cores per package\n",
- max_core_id, topo.num_cores_per_pkg);
- topo.num_packages = max_package_id + 1;
- DEBUG("max_package_id %d, sizing for %d packages\n",
- max_package_id, topo.num_packages);
+ 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;
- topo.num_threads_per_core = max_siblings;
- DEBUG("max_siblings %d\n", max_siblings);
+ 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 = 1;
+
+ 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 = 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_threads;
- free(cpus);
return 0;
+err:
+ free(topology.cpus);
+ return ret;
}
-static int
-allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
-{
- int i;
- *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
- topo.num_packages, sizeof(struct thread_data));
- if (*t == NULL)
- goto error;
+/************************
+ * Main alloc/init/free *
+ ************************/
- for (i = 0; i < topo.num_threads_per_core *
- topo.num_cores_per_pkg * topo.num_packages; i++)
- (*t)[i].cpu_id = -1;
+static int
+allocate_counters(struct thread_data **threads, struct core_data **cores, struct pkg_data **packages)
+{
+ unsigned int i;
+ unsigned int total_threads, total_cores;
- *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
- sizeof(struct core_data));
- if (*c == NULL)
- goto error;
+ total_threads = topology.num_threads * topology.num_cores * topology.num_packages;
+ *threads = calloc(total_threads, sizeof(struct thread_data));
+ if (*threads == NULL)
+ goto err;
- for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
- (*c)[i].core_id = -1;
+ for (i = 0; i < total_threads; ++i)
+ (*threads)[i].cpu_id = topology.max_cpu_id + 1;
- *p = calloc(topo.num_packages, sizeof(struct pkg_data));
- if (*p == NULL)
- goto error;
+ 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 < topo.num_packages; i++)
- (*p)[i].package_id = i;
+ *packages = calloc(topology.num_packages, sizeof(struct pkg_data));
+ if (*packages == NULL)
+ goto err_clean_cores;
return 0;
-error:
- ERROR("calloc counters");
- return -ERR_CALLOC;
+
+err_clean_cores:
+ free(*cores);
+err_clean_threads:
+ free(*threads);
+err:
+ ERROR("turbostat plugin: Failled to allocate memory for counters");
+ return -1;
}
-/*
- * init_counter()
- *
- * set cpu_id, core_num, pkg_num
- * 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 thread_num, int core_num,
- int pkg_num, int cpu_id)
+ struct pkg_data *pkg_base, unsigned int cpu_id)
{
- int ret;
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, thread_num, core_num, pkg_num);
- c = GET_CORE(core_base, core_num, pkg_num);
- p = GET_PKG(pkg_base, pkg_num);
+ 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 (thread_num == 0) {
+ if (cpu->first_thread_in_core)
t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
- if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
- return ret;
- } else if (ret != 0) {
- t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
- }
- }
+ if (cpu->first_core_in_package)
+ t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
- c->core_id = core_num;
- p->package_id = pkg_num;
+ c->core_id = cpu->core_id;
+ p->package_id = cpu->package_id;
+}
- return 0;
+static void
+initialize_counters(void)
+{
+ unsigned int cpu_id;
+
+ for (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 int
-initialize_counters(int cpu_id)
+
+static void
+free_all_buffers(void)
{
- int my_thread_id, my_core_id, my_package_id;
- int ret;
+ allocated = 0;
+ initialized = 0;
- my_package_id = get_physical_package_id(cpu_id);
- if (my_package_id < 0)
- return my_package_id;
- my_core_id = get_core_id(cpu_id);
- if (my_core_id < 0)
- return my_core_id;
-
- if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
- return ret;
- } else if (ret != 0) {
- my_thread_id = 0;
- topo.num_cores++;
- } else {
- my_thread_id = 1;
- }
+ CPU_FREE(cpu_present_set);
+ cpu_present_set = NULL;
+ cpu_present_set = 0;
- ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
- if (ret < 0)
- return ret;
- ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
- if (ret < 0)
- return ret;
- return 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)
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(for_all_proc_cpus(initialize_counters));
+ 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 = 1;
return 0;
}
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());
- DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
- plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, NULL, NULL);
+ plugin_register_read(PLUGIN_NAME, turbostat_read);
return 0;
err:
return ret;
}
-void module_register(void);
+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 = 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)
{
plugin_register_init(PLUGIN_NAME, turbostat_init);
+ plugin_register_config(PLUGIN_NAME, turbostat_config, config_keys, config_keys_num);
}
projects / collectd.git / blobdiff