/**
* collectd - src/intel_rdt.c
*
- * Copyright(c) 2016 Intel Corporation. All rights reserved.
+ * Copyright(c) 2016-2018 Intel Corporation. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
*
* Authors:
* Serhiy Pshyk <serhiyx.pshyk@intel.com>
+ * Starzyk, Mateusz <mateuszx.starzyk@intel.com>
**/
-#include "common.h"
#include "collectd.h"
-
-#include "utils_config_cores.h"
-
+#include "utils/common/common.h"
+#include "utils/config_cores/config_cores.h"
#include <pqos.h>
#define RDT_PLUGIN "intel_rdt"
#define RDT_MAX_SOCKETS 8
#define RDT_MAX_SOCKET_CORES 64
#define RDT_MAX_CORES (RDT_MAX_SOCKET_CORES * RDT_MAX_SOCKETS)
+/*
+ * Process name inside comm file is limited to 16 chars.
+ * More info here: http://man7.org/linux/man-pages/man5/proc.5.html
+ */
+#define RDT_MAX_PROC_COMM_LENGTH 16
typedef enum {
UNKNOWN = 0,
};
typedef struct rdt_ctx_s rdt_ctx_t;
-static rdt_ctx_t *g_rdt = NULL;
+static rdt_ctx_t *g_rdt;
static rdt_config_status g_state = UNKNOWN;
return;
DEBUG(RDT_PLUGIN ": Core Groups Dump");
- DEBUG(RDT_PLUGIN ": groups count: %zu", g_rdt->num_groups);
+ DEBUG(RDT_PLUGIN ": groups count: %" PRIsz, g_rdt->num_groups);
- for (int i = 0; i < g_rdt->num_groups; i++) {
+ for (size_t i = 0; i < g_rdt->num_groups; i++) {
core_group_t *cgroup = g_rdt->cores.cgroups + i;
memset(cores, 0, sizeof(cores));
- for (int j = 0; j < cgroup->num_cores; j++) {
+ for (size_t j = 0; j < cgroup->num_cores; j++) {
snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
cgroup->cores[j]);
}
- DEBUG(RDT_PLUGIN ": group[%d]:", i);
+ DEBUG(RDT_PLUGIN ": group[%zu]:", i);
DEBUG(RDT_PLUGIN ": description: %s", cgroup->desc);
DEBUG(RDT_PLUGIN ": cores: %s", cores);
DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->events[i]);
* MBR - remote memory bandwidth
*/
DEBUG(" CORE RMID LLC[KB] MBL[MB] MBR[MB]");
- for (int i = 0; i < g_rdt->num_groups; i++) {
-
+ for (size_t i = 0; i < g_rdt->num_groups; i++) {
const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
double llc = bytes_to_kb(pv->llc);
static int rdt_default_cgroups(void) {
unsigned num_cores = g_rdt->pqos_cpu->num_cores;
- g_rdt->cores.cgroups = calloc(num_cores, sizeof(*(g_rdt->cores.cgroups)));
+ g_rdt->cores.cgroups = calloc(num_cores, sizeof(*g_rdt->cores.cgroups));
if (g_rdt->cores.cgroups == NULL) {
ERROR(RDT_PLUGIN ": Error allocating core groups array");
return -ENOMEM;
char desc[DATA_MAX_NAME_LEN];
/* set core group info */
- cgroup->cores = calloc(1, sizeof(*(cgroup->cores)));
+ cgroup->cores = calloc(1, sizeof(*cgroup->cores));
if (cgroup->cores == NULL) {
ERROR(RDT_PLUGIN ": Error allocating cores array");
rdt_free_cgroups();
return num_cores;
}
-static int rdt_is_core_id_valid(int core_id) {
+static int rdt_is_core_id_valid(unsigned int core_id) {
- for (int i = 0; i < g_rdt->pqos_cpu->num_cores; i++)
+ for (unsigned int i = 0; i < g_rdt->pqos_cpu->num_cores; i++)
if (core_id == g_rdt->pqos_cpu->cores[i].lcore)
return 1;
for (size_t group_idx = 0; group_idx < n; group_idx++) {
core_group_t *cgroup = g_rdt->cores.cgroups + group_idx;
for (size_t core_idx = 0; core_idx < cgroup->num_cores; core_idx++) {
- if (!rdt_is_core_id_valid((int) cgroup->cores[core_idx])) {
- ERROR(RDT_PLUGIN ": Core group '%s' contains invalid core id '%d'",
- cgroup->desc, (int)cgroup->cores[core_idx]);
+ if (!rdt_is_core_id_valid(cgroup->cores[core_idx])) {
+ ERROR(RDT_PLUGIN ": Core group '%s' contains invalid core id '%u'",
+ cgroup->desc, cgroup->cores[core_idx]);
rdt_free_cgroups();
return -EINVAL;
}
ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
return ret;
}
- n = (size_t) ret;
+ n = (size_t)ret;
INFO(RDT_PLUGIN
": No core groups configured. Default core groups created.");
}
/* Get all available events on this platform */
- for (int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
+ for (unsigned int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
events |= g_rdt->cap_mon->u.mon->events[i].type;
events &= ~(PQOS_PERF_EVENT_LLC_MISS);
return 0;
}
+/* Helper typedef for process name array
+ * Extra 1 char is added for string null termination.
+ */
+typedef char proc_comm_t[RDT_MAX_PROC_COMM_LENGTH + 1];
+
+/* Linked one-way list of pids. */
+typedef struct pids_list_s {
+ pid_t pid;
+ struct pids_list_s *next;
+} pids_list_t;
+
+/* Holds process name and list of pids assigned to that name */
+typedef struct proc_pids_s {
+ proc_comm_t proccess_name;
+ pids_list_t *pids;
+} proc_pids_t;
+
+/*
+ * NAME
+ * pids_list_add_pid
+ *
+ * DESCRIPTION
+ * Adds pid at the end of the pids list.
+ * Allocates memory for new pid element, it is up to user to free it.
+ *
+ * PARAMETERS
+ * `list' Head of target pids_list.
+ * `pid' Pid to be added.
+ *
+ * RETURN VALUE
+ * On success, returns 0.
+ * -1 on memory allocation error.
+ */
+static int pids_list_add_pid(pids_list_t **list, const pid_t pid) {
+ pids_list_t *new_element = calloc(1, sizeof(*new_element));
+
+ if (new_element == NULL) {
+ ERROR(RDT_PLUGIN ": Alloc error\n");
+ return -1;
+ }
+ new_element->pid = pid;
+ new_element->next = NULL;
+
+ pids_list_t **current = list;
+ while (*current != NULL) {
+ current = &((*current)->next);
+ }
+ *current = new_element;
+ return 0;
+}
+
+/*
+ * NAME
+ * read_proc_name
+ *
+ * DESCRIPTION
+ * Reads process name from given pid directory.
+ * Strips new-line character (\n).
+ *
+ * PARAMETERS
+ * `procfs_path` Path to systems proc directory (e.g. /proc)
+ * `pid_entry' Dirent for PID directory
+ * `name' Output buffer for process name, recommended proc_comm.
+ * `out_size' Output buffer size, recommended sizeof(proc_comm)
+ *
+ * RETURN VALUE
+ * On success, the number of read bytes (includes stripped \n).
+ * -1 on file open error
+ */
+static int read_proc_name(const char *procfs_path,
+ const struct dirent *pid_entry, char *name,
+ const size_t out_size) {
+ assert(procfs_path);
+ assert(pid_entry);
+ assert(name);
+ assert(out_size);
+ memset(name, 0, out_size);
+
+ const char *comm_file_name = "comm";
+
+ char *path = ssnprintf_alloc("%s/%s/%s", procfs_path, pid_entry->d_name,
+ comm_file_name);
+
+ FILE *f = fopen(path, "r");
+ if (f == NULL) {
+ ERROR(RDT_PLUGIN ": Failed to open comm file, error: %d\n", errno);
+ sfree(path);
+ return -1;
+ }
+ size_t read_length = fread(name, sizeof(char), out_size, f);
+ fclose(f);
+ sfree(path);
+ /* strip new line ending */
+ char *newline = strchr(name, '\n');
+ if (newline) {
+ *newline = '\0';
+ }
+
+ return read_length;
+}
+
+/*
+ * NAME
+ * get_pid_number
+ *
+ * DESCRIPTION
+ * Gets pid number for given /proc/pid directory entry or
+ * returns error if input directory does not hold PID information.
+ *
+ * PARAMETERS
+ * `entry' Dirent for PID directory
+ * `pid' PID number to be filled
+ *
+ * RETURN VALUE
+ * 0 on success. Negative number on error:
+ * -1: given entry is not a directory
+ * -2: PID conversion error
+ */
+static int get_pid_number(struct dirent *entry, pid_t *pid) {
+ char *tmp_end; /* used for strtoul error check*/
+
+ if (pid == NULL || entry == NULL)
+ return -1;
+
+ if (entry->d_type != DT_DIR)
+ return -1;
+
+ /* trying to get pid number from directory name*/
+ *pid = strtoul(entry->d_name, &tmp_end, 10);
+ if (*tmp_end != '\0') {
+ return -2; /* conversion failed, not proc-pid */
+ }
+ /* all checks passed, marking as success */
+ return 0;
+}
+
+/*
+ * NAME
+ * fetch_pids_for_procs
+ *
+ * DESCRIPTION
+ * Finds PIDs matching given process's names.
+ * Searches all PID directories in /proc fs and
+ * allocates memory for proc_pids structs, it is up to user to free it.
+ * Output array will have same element count as input array.
+ *
+ * PARAMETERS
+ * `procfs_path' Path to systems proc directory (e.g. /proc)
+ * `procs' Array of null-terminated strings with
+ * process' names to search for
+ * `procs_size' procs array element count
+ * `proc_pids_array' Address of pointer, under which new
+ * array of proc_pids will be allocated. Must be NULL.
+ *
+ * RETURN VALUE
+ * 0 on success. Negative number on error:
+ * -1: could not open /proc dir
+ */
+__attribute__((unused)) /* TODO: remove this attribute when PID monitoring is
+ implemented */
+static int
+fetch_pids_for_procs(const char *procfs_path, const char **procs_names_array,
+ const size_t procs_names_array_size,
+ proc_pids_t **proc_pids_array) {
+ assert(procfs_path);
+ assert(procs_names_array);
+ assert(procs_names_array_size);
+ assert(proc_pids_array);
+ assert(NULL == *proc_pids_array);
+
+ DIR *proc_dir = opendir(procfs_path);
+ if (proc_dir == NULL) {
+ ERROR(RDT_PLUGIN ": Could not open %s directory, error: %d", procfs_path,
+ errno);
+ return -1;
+ }
+
+ /* Copy procs names to output array. Initialize pids list with NULL value. */
+ (*proc_pids_array) =
+ calloc(procs_names_array_size, sizeof(**proc_pids_array));
+ for (size_t i = 0; i < procs_names_array_size; ++i) {
+ sstrncpy((*proc_pids_array)[i].proccess_name, procs_names_array[i],
+ STATIC_ARRAY_SIZE((*proc_pids_array)[i].proccess_name));
+ (*proc_pids_array)[i].pids = NULL;
+ }
+
+ /* Go through procfs and find PIDS and their comms */
+ struct dirent *entry;
+ while ((entry = readdir(proc_dir)) != NULL) {
+
+ pid_t pid;
+ int pid_conversion = get_pid_number(entry, &pid);
+ if (pid_conversion < 0)
+ continue;
+
+ proc_comm_t comm;
+ int read_result =
+ read_proc_name(procfs_path, entry, comm, sizeof(proc_comm_t));
+ if (read_result <= 0) {
+ ERROR(RDT_PLUGIN ": Comm file skipped. Read result: %d", read_result);
+ continue;
+ }
+
+ /* Try to find comm in input procs array (proc_pids_array has same names) */
+ for (size_t i = 0; i < procs_names_array_size; ++i) {
+ if (0 == strncmp(comm, (*proc_pids_array)[i].proccess_name,
+ STATIC_ARRAY_SIZE(comm)))
+ pids_list_add_pid(&((*proc_pids_array)[i].pids), pid);
+ }
+ }
+
+ int close_result = closedir(proc_dir);
+ if (0 != close_result) {
+ ERROR(RDT_PLUGIN ": failed to close %s directory, error: %d", procfs_path,
+ errno);
+ return -1;
+ }
+ return 0;
+}
+
static void rdt_pqos_log(void *context, const size_t size, const char *msg) {
DEBUG(RDT_PLUGIN ": %s", msg);
}
return 0;
}
-static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
- derive_t value) {
+static void rdt_submit_derive(const char *cgroup, const char *type,
+ const char *type_instance, derive_t value) {
value_list_t vl = VALUE_LIST_INIT;
vl.values = &(value_t){.derive = value};
plugin_dispatch_values(&vl);
}
-static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
- gauge_t value) {
+static void rdt_submit_gauge(const char *cgroup, const char *type,
+ const char *type_instance, gauge_t value) {
value_list_t vl = VALUE_LIST_INIT;
vl.values = &(value_t){.gauge = value};
rdt_dump_data();
#endif /* COLLECT_DEBUG */
- for (int i = 0; i < g_rdt->num_groups; i++) {
+ for (size_t i = 0; i < g_rdt->num_groups; i++) {
core_group_t *cgroup = g_rdt->cores.cgroups + i;
enum pqos_mon_event mbm_events =
return ret;
/* Start monitoring */
- for (int i = 0; i < g_rdt->num_groups; i++) {
+ for (size_t i = 0; i < g_rdt->num_groups; i++) {
core_group_t *cg = g_rdt->cores.cgroups + i;
ret = pqos_mon_start(cg->num_cores, cg->cores, g_rdt->events[i],
return 0;
/* Stop monitoring */
- for (int i = 0; i < g_rdt->num_groups; i++) {
+ for (size_t i = 0; i < g_rdt->num_groups; i++) {
pqos_mon_stop(g_rdt->pgroups[i]);
}