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 {
+struct cpu_topology {
+ int package_id;
+ int core_id;
+ _Bool first_core_in_package;
+ _Bool first_thread_in_core;
+};
+
+struct topology {
+ int max_cpu_id;
int num_packages;
- int num_cpus;
int num_cores;
- int max_cpu_num;
- int num_cores_per_pkg;
- int num_threads_per_core;
-} topo;
+ int num_threads;
+ struct cpu_topology *cpus;
+} topology;
struct timeval tv_even, tv_odd, tv_delta;
UNSUPPORTED_CPU,
};
-static int setup_all_buffers(void);
-
-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 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;
-
- 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;
-}
+/*****************************
+ * MSR Manipulation helpers *
+ *****************************/
/*
* Open a MSR device for reading
return retval;
}
+
+/********************************
+ * Raw data acquisition (1 CPU) *
+ ********************************/
+
+/*
+ * Read every data avalaible for a single CPU
+ *
+ * Core data is shared for all threads in one core: extracted only for the first thread
+ * Package data is shared for all core in one package: extracted only for the first thread of the first core
+ *
+ * Side effect: migrates to the targeted CPU
+ */
+static int __attribute__((warn_unused_result))
+get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
+{
+ int cpu = t->cpu_id;
+ unsigned long long msr;
+ int msr_fd;
+ int retval = 0;
+
+ msr_fd = open_msr(cpu, 1);
+ if (msr_fd < 0)
+ return msr_fd;
+
+#define READ_MSR(msr, dst) \
+do { \
+ if (read_msr(msr_fd, msr, dst)) { \
+ retval = -ERR_##msr; \
+ 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;
+
+ /* collect core counters only for 1st thread in core */
+ if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) {
+ retval = 0;
+ goto out;
+ }
+
+ if (do_core_cstate & (1 << 3))
+ READ_MSR(MSR_CORE_C3_RESIDENCY, &c->c3);
+ if (do_core_cstate & (1 << 6))
+ READ_MSR(MSR_CORE_C6_RESIDENCY, &c->c6);
+ if (do_core_cstate & (1 << 7))
+ READ_MSR(MSR_CORE_C7_RESIDENCY, &c->c7);
+
+ if (do_dts) {
+ READ_MSR(MSR_IA32_THERM_STATUS, &msr);
+ c->core_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
+ }
+
+ /* collect package counters only for 1st core in package */
+ if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
+ retval = 0;
+ goto out;
+ }
+
+ if (do_pkg_cstate & (1 << 2))
+ READ_MSR(MSR_PKG_C2_RESIDENCY, &p->pc2);
+ if (do_pkg_cstate & (1 << 3))
+ READ_MSR(MSR_PKG_C3_RESIDENCY, &p->pc3);
+ if (do_pkg_cstate & (1 << 6))
+ READ_MSR(MSR_PKG_C6_RESIDENCY, &p->pc6);
+ if (do_pkg_cstate & (1 << 7))
+ READ_MSR(MSR_PKG_C7_RESIDENCY, &p->pc7);
+ if (do_pkg_cstate & (1 << 8))
+ READ_MSR(MSR_PKG_C8_RESIDENCY, &p->pc8);
+ if (do_pkg_cstate & (1 << 9))
+ READ_MSR(MSR_PKG_C9_RESIDENCY, &p->pc9);
+ if (do_pkg_cstate & (1 << 10))
+ READ_MSR(MSR_PKG_C10_RESIDENCY, &p->pc10);
+
+ if (do_rapl & RAPL_PKG) {
+ READ_MSR(MSR_PKG_ENERGY_STATUS, &msr);
+ p->energy_pkg = msr & 0xFFFFFFFF;
+ }
+ if (do_rapl & RAPL_CORES) {
+ READ_MSR(MSR_PP0_ENERGY_STATUS, &msr);
+ p->energy_cores = msr & 0xFFFFFFFF;
+ }
+ if (do_rapl & RAPL_DRAM) {
+ READ_MSR(MSR_DRAM_ENERGY_STATUS, &msr);
+ p->energy_dram = msr & 0xFFFFFFFF;
+ }
+ if (do_rapl & RAPL_GFX) {
+ READ_MSR(MSR_PP1_ENERGY_STATUS, &msr);
+ p->energy_gfx = msr & 0xFFFFFFFF;
+ }
+ if (do_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);
+ }
+
+out:
+ close(msr_fd);
+ return retval;
+}
+
+
+/**********************************
+ * Evaluating the changes (1 CPU) *
+ **********************************/
+
+/*
+ * Do delta = new - old on 32bits cyclique intergers
+ */
#define DELTA_WRAP32(delta, new, old) \
if (new > old) { \
delta = new - old; \
delta = 0x100000000 + new - old; \
}
-static void
+/*
+ * Extract the evolution old->new in delta at a package level
+ * (some are not new-delta, e.g. temperature)
+ */
+static inline void
delta_package(struct pkg_data *delta, const struct pkg_data *new, const struct pkg_data *old)
{
delta->pc2 = new->pc2 - old->pc2;
DELTA_WRAP32(delta->rapl_dram_perf_status, new->rapl_dram_perf_status, old->rapl_dram_perf_status);
}
-static void
+/*
+ * Extract the evolution old->new in delta at a core level
+ * (some are not new-delta, e.g. temperature)
+ */
+static inline void
delta_core(struct core_data *delta, const struct core_data *new, const struct core_data *old)
{
delta->c3 = new->c3 - old->c3;
delta->core_temp_c = new->core_temp_c;
}
-static int __attribute__((warn_unused_result))
+/*
+ * Extract the evolution old->new in delta at a package level
+ * core_delta is required for c1 estimation (tsc - c0 - all core cstates)
+ */
+static inline int __attribute__((warn_unused_result))
delta_thread(struct thread_data *delta, const struct thread_data *new, const struct thread_data *old,
const struct core_data *core_delta)
{
return 0;
}
-static int __attribute__((warn_unused_result))
-delta_cpu(struct thread_data *t_delta, struct core_data *c_delta, struct pkg_data *p_delta,
- const struct thread_data *t_new, const struct core_data *c_new, const struct pkg_data *p_new,
- const struct thread_data *t_old, const struct core_data *c_old, const struct pkg_data *p_old)
-{
- int ret;
+/**********************************
+ * Submitting the results (1 CPU) *
+ **********************************/
- /* calculate core delta only for 1st thread in core */
- if (t_new->flags & CPU_IS_FIRST_THREAD_IN_CORE)
- delta_core(c_delta, c_new, c_old);
-
- /* always calculate thread delta */
- ret = delta_thread(t_delta, t_new, t_old, c_delta);
- if (ret != 0)
- return ret;
+/*
+ * Submit one gauge value
+ */
+static void
+turbostat_submit (const char *plugin_instance,
+ const char *type, const char *type_instance,
+ gauge_t value)
+{
+ value_list_t vl = VALUE_LIST_INIT;
+ value_t v;
- /* calculate package delta only for 1st core in package */
- if (t_new->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)
- delta_package(p_delta, p_new, p_old);
+ v.gauge = value;
+ vl.values = &v;
+ vl.values_len = 1;
+ sstrncpy (vl.host, hostname_g, sizeof (vl.host));
+ sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
+ if (plugin_instance != NULL)
+ sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
+ sstrncpy (vl.type, type, sizeof (vl.type));
+ if (type_instance != NULL)
+ sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
- return 0;
+ plugin_dispatch_values (&vl);
}
-
/*
- * get_counters(...)
- * migrate to cpu
- * acquire and record local counters for that cpu
+ * Submit every data for a single CPU
+ *
+ * Core data is shared for all threads in one core: submitted only for the first thread
+ * Package data is shared for all core in one package: submitted only for the first thread of the first core
*/
-static int __attribute__((warn_unused_result))
-get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
+static int
+submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
{
- int cpu = t->cpu_id;
- unsigned long long msr;
- int msr_fd;
- int retval = 0;
+ char name[12];
+ double interval_float;
- msr_fd = open_msr(cpu, 1);
- if (msr_fd < 0)
- return msr_fd;
+ interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
-#define READ_MSR(msr, dst) \
-do { \
- if (read_msr(msr_fd, msr, dst)) { \
- retval = -ERR_##msr; \
- goto out; \
- } \
-} while (0)
+ ssnprintf(name, sizeof(name), "cpu%02d", t->cpu_id);
- READ_MSR(MSR_IA32_TSC, &t->tsc);
+ if (!aperf_mperf_unstable)
+ turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
+ if (!aperf_mperf_unstable)
+ turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
- READ_MSR(MSR_IA32_APERF, &t->aperf);
- READ_MSR(MSR_IA32_MPERF, &t->mperf);
+ /* GHz */
+ 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);
- READ_MSR(MSR_SMI_COUNT, &msr);
- t->smi_count = msr & 0xFFFFFFFF;
+ /* SMI */
+ turbostat_submit(NULL, "current", name, t->smi_count);
- /* collect core counters only for 1st thread in core */
- if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) {
- retval = 0;
- goto out;
- }
+ /* submit per-core data only for 1st thread in core */
+ if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
+ goto done;
+
+ ssnprintf(name, sizeof(name), "core%02d", c->core_id);
if (do_core_cstate & (1 << 3))
- READ_MSR(MSR_CORE_C3_RESIDENCY, &c->c3);
+ turbostat_submit(name, "percent", "c3", 100.0 * c->c3/t->tsc);
if (do_core_cstate & (1 << 6))
- READ_MSR(MSR_CORE_C6_RESIDENCY, &c->c6);
+ turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
if (do_core_cstate & (1 << 7))
- READ_MSR(MSR_CORE_C7_RESIDENCY, &c->c7);
+ turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
- if (do_dts) {
- READ_MSR(MSR_IA32_THERM_STATUS, &msr);
- c->core_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
- }
+ if (do_dts)
+ turbostat_submit(NULL, "temperature", name, c->core_temp_c);
- /* collect package counters only for 1st core in package */
- if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) {
- retval = 0;
- goto out;
- }
+ /* submit per-package data only for 1st core in package */
+ if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
+ goto done;
- 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);
+ ssnprintf(name, sizeof(name), "pkg%02d", p->package_id);
- if (do_rapl & RAPL_PKG) {
- READ_MSR(MSR_PKG_ENERGY_STATUS, &msr);
- p->energy_pkg = msr & 0xFFFFFFFF;
- }
- if (do_rapl & RAPL_CORES) {
- READ_MSR(MSR_PP0_ENERGY_STATUS, &msr);
- p->energy_cores = msr & 0xFFFFFFFF;
- }
- if (do_rapl & RAPL_DRAM) {
- READ_MSR(MSR_DRAM_ENERGY_STATUS, &msr);
- p->energy_dram = msr & 0xFFFFFFFF;
- }
- if (do_rapl & RAPL_GFX) {
- READ_MSR(MSR_PP1_ENERGY_STATUS, &msr);
- p->energy_gfx = msr & 0xFFFFFFFF;
- }
- if (do_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)
+ turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
+
+ if (do_pkg_cstate & (1 << 2))
+ turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
+ if (do_pkg_cstate & (1 << 3))
+ turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
+ if (do_pkg_cstate & (1 << 6))
+ turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
+ if (do_pkg_cstate & (1 << 7))
+ turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
+ if (do_pkg_cstate & (1 << 8))
+ turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
+ if (do_pkg_cstate & (1 << 9))
+ turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
+ if (do_pkg_cstate & (1 << 10))
+ turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
+
+ if (do_rapl) {
+ if (do_rapl & RAPL_PKG)
+ turbostat_submit(name, "power", "Pkg_W", 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);
+ if (do_rapl & RAPL_GFX)
+ turbostat_submit(name, "power", "GFX_W", 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);
}
- if (do_ptm) {
- READ_MSR(MSR_IA32_PACKAGE_THERM_STATUS, &msr);
- p->pkg_temp_c = p->tcc_activation_temp - ((msr >> 16) & 0x7F);
+done:
+ return 0;
+}
+
+
+/**********************************
+ * Looping function over all CPUs *
+ **********************************/
+
+/*
+ * Check if a given cpu id is in our compiled list of existing CPUs
+ */
+static int
+cpu_is_not_present(int cpu)
+{
+ return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set);
+}
+
+/*
+ * Loop on all CPUs in topological order
+ *
+ * Skip non-present cpus
+ * Return the error code at the first error or 0
+ */
+static int __attribute__((warn_unused_result))
+for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *),
+ struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base)
+{
+ int retval, 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;
+ 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;
+}
-out:
- close(msr_fd);
- return retval;
+/*
+ * Dedicated loop: Extract every data evolution for all CPU
+ *
+ * Skip non-present cpus
+ * Return the error code at the first error or 0
+ *
+ * Core data is shared for all threads in one core: extracted only for the first thread
+ * Package data is shared for all core in one package: extracted only for the first thread of the first core
+ */
+static int __attribute__((warn_unused_result))
+for_all_cpus_delta(const struct thread_data *thread_new_base, const struct core_data *core_new_base, const struct pkg_data *pkg_new_base,
+ const struct thread_data *thread_old_base, const struct core_data *core_old_base, const struct pkg_data *pkg_old_base)
+{
+ int retval, 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;
}
+
static void
free_all_buffers(void)
{
package_delta = NULL;
}
+
+/****************
+ * File helpers *
+ ****************/
+
/*
- * Parse a file containing a single int.
+ * Read a single int from a file.
*/
static int __attribute__ ((format(printf,1,2)))
parse_int_file(const char *fmt, ...)
return value;
}
-/*
- * cpu_is_first_sibling_in_core(cpu)
- * return 1 if given CPU is 1st HT sibling in the core
- */
-static int
-cpu_is_first_sibling_in_core(int cpu)
-{
- return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
-}
-
-/*
- * cpu_is_first_core_in_package(cpu)
- * return 1 if given CPU is 1st core in package
- */
-static int
-cpu_is_first_core_in_package(int cpu)
-{
- return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
-}
-
-static int
-get_physical_package_id(int cpu)
-{
- return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
-}
-
-static int
-get_core_id(int cpu)
-{
- return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
-}
-
static int
-get_num_ht_siblings(int cpu)
+get_threads_on_core(int cpu)
{
char path[80];
FILE *filep;
return 1;
}
-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;
-
- 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_delta;
- const struct thread_data *t_old, *t_new;
- struct core_data *c_delta;
- const struct core_data *c_old, *c_new;
- struct pkg_data *p_delta;
- const struct pkg_data *p_old, *p_new;
-
- 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);
- if (cpu_is_not_present(t_delta->cpu_id))
- continue;
-
- c_delta = GET_CORE(core_delta, core_no, pkg_no);
- c_new = GET_CORE(core_new_base, core_no, pkg_no);
- c_old = GET_CORE(core_old_base, core_no, pkg_no);
-
- 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);
-
- retval = delta_cpu(t_delta, c_delta, p_delta,
- t_new, c_new, p_new,
- t_old, c_old, p_old);
- if (retval)
- return retval;
- }
- }
- }
- return 0;
-}
-
/*
* run func(cpu) on every cpu in /proc/stat
* return max_cpu number
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;
}
}
/*
- * count_cpus()
- * remember the last one seen, it will be the max
+ * Update the stored topology.max_cpu_id
*/
static int
-count_cpus(int cpu)
+update_max_cpu_id(int cpu)
{
- if (topo.max_cpu_num < cpu)
- topo.max_cpu_num = cpu;
-
- topo.num_cpus += 1;
+ if (topology.max_cpu_id < cpu)
+ topology.max_cpu_id = cpu;
return 0;
}
+
static int
mark_cpu_present(int cpu)
{
}
-static void
-turbostat_submit (const char *plugin_instance,
- const char *type, const char *type_instance,
- gauge_t value)
-{
- value_list_t vl = VALUE_LIST_INIT;
- value_t v;
-
- v.gauge = value;
- vl.values = &v;
- vl.values_len = 1;
- sstrncpy (vl.host, hostname_g, sizeof (vl.host));
- sstrncpy (vl.plugin, PLUGIN_NAME, sizeof (vl.plugin));
- if (plugin_instance != NULL)
- sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
- sstrncpy (vl.type, type, sizeof (vl.type));
- if (type_instance != NULL)
- sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
-
- plugin_dispatch_values (&vl);
-}
-
-/*
- * 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
- */
-static int
-submit_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
-{
- char name[12];
- double interval_float;
-
- interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
-
- ssnprintf(name, sizeof(name), "cpu%02d", t->cpu_id);
-
- if (!aperf_mperf_unstable)
- turbostat_submit(name, "percent", "c0", 100.0 * t->mperf/t->tsc);
- if (!aperf_mperf_unstable)
- turbostat_submit(name, "percent", "c1", 100.0 * t->c1/t->tsc);
-
- /* GHz */
- 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);
-
- /* SMI */
- turbostat_submit(NULL, "current", name, t->smi_count);
-
- /* print per-core data only for 1st thread in core */
- if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE))
- goto done;
-
- 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);
- if (do_core_cstate & (1 << 6))
- turbostat_submit(name, "percent", "c6", 100.0 * c->c6/t->tsc);
- if (do_core_cstate & (1 << 7))
- turbostat_submit(name, "percent", "c7", 100.0 * c->c7/t->tsc);
-
- if (do_dts)
- turbostat_submit(NULL, "temperature", name, c->core_temp_c);
-
- /* print per-package data only for 1st core in package */
- if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
- goto done;
-
- ssnprintf(name, sizeof(name), "pkg%02d", p->package_id);
-
- if (do_ptm)
- turbostat_submit(NULL, "temperature", name, p->pkg_temp_c);
-
- if (do_pkg_cstate & (1 << 2))
- turbostat_submit(name, "percent", "pc2", 100.0 * p->pc2/t->tsc);
- if (do_pkg_cstate & (1 << 3))
- turbostat_submit(name, "percent", "pc3", 100.0 * p->pc3/t->tsc);
- if (do_pkg_cstate & (1 << 6))
- turbostat_submit(name, "percent", "pc6", 100.0 * p->pc6/t->tsc);
- if (do_pkg_cstate & (1 << 7))
- turbostat_submit(name, "percent", "pc7", 100.0 * p->pc7/t->tsc);
- if (do_pkg_cstate & (1 << 8))
- turbostat_submit(name, "percent", "pc8", 100.0 * p->pc8/t->tsc);
- if (do_pkg_cstate & (1 << 9))
- turbostat_submit(name, "percent", "pc9", 100.0 * p->pc9/t->tsc);
- if (do_pkg_cstate & (1 << 10))
- turbostat_submit(name, "percent", "pc10", 100.0 * p->pc10/t->tsc);
-
- if (do_rapl) {
- if (do_rapl & RAPL_PKG)
- turbostat_submit(name, "power", "Pkg_W", 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);
- if (do_rapl & RAPL_GFX)
- turbostat_submit(name, "power", "GFX_W", 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);
- }
-done:
- return 0;
-}
+static int setup_all_buffers(void);
static int
turbostat_read(user_data_t * not_used)
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;
p->tcc_activation_temp = target_c_local;
*
* 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");
*
* 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);
case 0x3A: /* IVB */
case 0x3E: /* IVB Xeon */
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_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_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_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_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");
}
if (do_rapl) {
- unsigned long msr;
+ unsigned long long msr;
if (get_msr(0, MSR_RAPL_POWER_UNIT, &msr))
return 0;
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;
+ }
+ *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()
{
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;
- }
-
- /*
- * 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;
- }
- 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;
- }
+ int max_package_id, max_core_id, max_thread_id;
+ max_package_id = max_core_id = max_thread_id = 0;
- /*
- * Allocate and initialize cpu_affinity_set
- */
- cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
- if (cpu_affinity_set == NULL) {
- free(cpus);
- ERROR("CPU_ALLOC");
- return -ERR_CPU_ALLOC;
- }
- cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
- CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
+ /* 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));
- /*
- * Allocate and initialize cpu_saved_affinity_set
- */
- cpu_saved_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
- if (cpu_saved_affinity_set == NULL) {
- free(cpus);
- ERROR("CPU_ALLOC");
- return -ERR_CPU_ALLOC;
+ 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;
}
- cpu_saved_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
- CPU_ZERO_S(cpu_saved_affinity_setsize, cpu_saved_affinity_set);
+ 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;
+
+ /* Can't fail (mark_cpu_present always returns 0) */
+ assert(for_all_proc_cpus(mark_cpu_present));
/*
* 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) {
+ int num_threads;
+ struct cpu_topology *cpu = &topology.cpus[i];
if (cpu_is_not_present(i)) {
WARNING("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 = 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 = 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)
+ 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",
+ 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;
- 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)
+allocate_counters(struct thread_data **threads, struct core_data **cores, struct pkg_data **packages)
{
int i;
+ int total_threads, total_cores;
- *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
- topo.num_packages, sizeof(struct thread_data));
- if (*t == 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_threads_per_core *
- topo.num_cores_per_pkg * topo.num_packages; i++)
- (*t)[i].cpu_id = -1;
+ for (i = 0; i < total_threads; ++i)
+ (*threads)[i].cpu_id = -1;
- *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
- sizeof(struct core_data));
- if (*c == 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_cores_per_pkg * topo.num_packages; i++)
- (*c)[i].core_id = -1;
+ for (i = 0; i < total_cores; ++i)
+ (*cores)[i].core_id = -1;
- *p = calloc(topo.num_packages, sizeof(struct pkg_data));
- if (*p == NULL)
- goto error;
+ *packages = calloc(topology.num_packages, sizeof(struct pkg_data));
+ if (*packages == NULL)
+ goto err_clean_cores;
- for (i = 0; i < topo.num_packages; i++)
- (*p)[i].package_id = i;
+ for (i = 0; i < topology.num_packages; i++)
+ (*packages)[i].package_id = i;
return 0;
-error:
+
+err_clean_cores:
+ free(*cores);
+err_clean_threads:
+ free(*threads);
+err:
ERROR("calloc counters");
return -ERR_CALLOC;
}
+
/*
* init_counter()
*
- * set cpu_id, core_num, pkg_num
+ * 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
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, 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 int
-initialize_counters(int cpu_id)
+initialize_counters(void)
{
- int my_thread_id, my_core_id, my_package_id;
int ret;
+ int cpu_id;
- 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;
- }
+ for (cpu_id = 0; cpu_id <= topology.max_cpu_id; ++cpu_id) {
+ if (cpu_is_not_present(cpu_id)) {
+ continue;
+ }
- 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;
- ret = init_counter(DELTA_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
- if (ret < 0)
- return ret;
+ 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;
+ }
return 0;
}
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(for_all_proc_cpus(initialize_counters));
+ DO_OR_GOTO_ERR(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));
projects / collectd.git / blobdiff