| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | // Heavily inspired from proc_net_dev.c |
| 4 | #include "plugin_proc.h" |
| 5 | |
| 6 | #define PLUGIN_PROC_MODULE_INFINIBAND_NAME "/sys/class/infiniband" |
| 7 | #define CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND \ |
| 8 | "plugin:" PLUGIN_PROC_CONFIG_NAME ":" PLUGIN_PROC_MODULE_INFINIBAND_NAME |
| 9 | |
| 10 | // ib_device::name[IB_DEVICE_NAME_MAX(64)] + "-" + ib_device::phys_port_cnt[u8 = 3 chars] |
| 11 | #define IBNAME_MAX 68 |
| 12 | |
| 13 | // ---------------------------------------------------------------------------- |
| 14 | // infiniband & omnipath standard counters |
| 15 | |
| 16 | // I use macro as there's no single file acting as summary, but a lot of different files, so can't use helpers like |
| 17 | // procfile(). Also, omnipath generates other counters, that are not provided by infiniband |
| 18 | #define FOREACH_COUNTER(GEN, ...) \ |
| 19 | FOREACH_COUNTER_BYTES(GEN, __VA_ARGS__) \ |
| 20 | FOREACH_COUNTER_PACKETS(GEN, __VA_ARGS__) \ |
| 21 | FOREACH_COUNTER_ERRORS(GEN, __VA_ARGS__) |
| 22 | |
| 23 | #define FOREACH_COUNTER_BYTES(GEN, ...) \ |
| 24 | GEN(port_rcv_data, bytes, "Received", 1, __VA_ARGS__) \ |
| 25 | GEN(port_xmit_data, bytes, "Sent", -1, __VA_ARGS__) |
| 26 | |
| 27 | #define FOREACH_COUNTER_PACKETS(GEN, ...) \ |
| 28 | GEN(port_rcv_packets, packets, "Received", 1, __VA_ARGS__) \ |
| 29 | GEN(port_xmit_packets, packets, "Sent", -1, __VA_ARGS__) \ |
| 30 | GEN(multicast_rcv_packets, packets, "Mcast rcvd", 1, __VA_ARGS__) \ |
| 31 | GEN(multicast_xmit_packets, packets, "Mcast sent", -1, __VA_ARGS__) \ |
| 32 | GEN(unicast_rcv_packets, packets, "Ucast rcvd", 1, __VA_ARGS__) \ |
| 33 | GEN(unicast_xmit_packets, packets, "Ucast sent", -1, __VA_ARGS__) |
| 34 | |
| 35 | #define FOREACH_COUNTER_ERRORS(GEN, ...) \ |
| 36 | GEN(port_rcv_errors, errors, "Pkts malformated", 1, __VA_ARGS__) \ |
| 37 | GEN(port_rcv_constraint_errors, errors, "Pkts rcvd discarded ", 1, __VA_ARGS__) \ |
| 38 | GEN(port_xmit_discards, errors, "Pkts sent discarded", 1, __VA_ARGS__) \ |
| 39 | GEN(port_xmit_wait, errors, "Tick Wait to send", 1, __VA_ARGS__) \ |
| 40 | GEN(VL15_dropped, errors, "Pkts missed resource", 1, __VA_ARGS__) \ |
| 41 | GEN(excessive_buffer_overrun_errors, errors, "Buffer overrun", 1, __VA_ARGS__) \ |
| 42 | GEN(link_downed, errors, "Link Downed", 1, __VA_ARGS__) \ |
| 43 | GEN(link_error_recovery, errors, "Link recovered", 1, __VA_ARGS__) \ |
| 44 | GEN(local_link_integrity_errors, errors, "Link integrity err", 1, __VA_ARGS__) \ |
| 45 | GEN(symbol_error, errors, "Link minor errors", 1, __VA_ARGS__) \ |
| 46 | GEN(port_rcv_remote_physical_errors, errors, "Pkts rcvd with EBP", 1, __VA_ARGS__) \ |
| 47 | GEN(port_rcv_switch_relay_errors, errors, "Pkts rcvd discarded by switch", 1, __VA_ARGS__) \ |
| 48 | GEN(port_xmit_constraint_errors, errors, "Pkts sent discarded by switch", 1, __VA_ARGS__) |
| 49 | |
| 50 | // |
| 51 | // Hardware Counters |
| 52 | // |
| 53 | |
| 54 | // IMPORTANT: These are vendor-specific fields. |
| 55 | // If you want to add a new vendor, search this for for 'VENDORS:' keyword and |
| 56 | // add your definition as 'VENDOR-<key>:' where <key> if the string part that |
| 57 | // is shown in /sys/class/infiniband/<key>X_Y |
| 58 | // EG: for Mellanox, shown as mlx0_1, it's 'mlx' |
| 59 | // for Intel, shown as hfi1_1, it's 'hfi' |
| 60 | |
| 61 | // VENDORS: List of implemented hardware vendors |
| 62 | #define FOREACH_HWCOUNTER_NAME(GEN, ...) GEN(mlx, __VA_ARGS__) |
| 63 | |
| 64 | // VENDOR-MLX: HW Counters for Mellanox ConnectX Devices |
| 65 | #define FOREACH_HWCOUNTER_MLX(GEN, ...) \ |
| 66 | FOREACH_HWCOUNTER_MLX_PACKETS(GEN, __VA_ARGS__) \ |
| 67 | FOREACH_HWCOUNTER_MLX_ERRORS(GEN, __VA_ARGS__) |
| 68 | |
| 69 | #define FOREACH_HWCOUNTER_MLX_PACKETS(GEN, ...) \ |
| 70 | GEN(np_cnp_sent, hwpackets, "RoCEv2 Congestion sent", 1, __VA_ARGS__) \ |
| 71 | GEN(np_ecn_marked_roce_packets, hwpackets, "RoCEv2 Congestion rcvd", -1, __VA_ARGS__) \ |
| 72 | GEN(rp_cnp_handled, hwpackets, "IB Congestion handled", 1, __VA_ARGS__) \ |
| 73 | GEN(rx_atomic_requests, hwpackets, "ATOMIC req. rcvd", 1, __VA_ARGS__) \ |
| 74 | GEN(rx_dct_connect, hwpackets, "Connection req. rcvd", 1, __VA_ARGS__) \ |
| 75 | GEN(rx_read_requests, hwpackets, "Read req. rcvd", 1, __VA_ARGS__) \ |
| 76 | GEN(rx_write_requests, hwpackets, "Write req. rcvd", 1, __VA_ARGS__) \ |
| 77 | GEN(roce_adp_retrans, hwpackets, "RoCE retrans adaptive", 1, __VA_ARGS__) \ |
| 78 | GEN(roce_adp_retrans_to, hwpackets, "RoCE retrans timeout", 1, __VA_ARGS__) \ |
| 79 | GEN(roce_slow_restart, hwpackets, "RoCE slow restart", 1, __VA_ARGS__) \ |
| 80 | GEN(roce_slow_restart_cnps, hwpackets, "RoCE slow restart congestion", 1, __VA_ARGS__) \ |
| 81 | GEN(roce_slow_restart_trans, hwpackets, "RoCE slow restart count", 1, __VA_ARGS__) |
| 82 | |
| 83 | #define FOREACH_HWCOUNTER_MLX_ERRORS(GEN, ...) \ |
| 84 | GEN(duplicate_request, hwerrors, "Duplicated packets", -1, __VA_ARGS__) \ |
| 85 | GEN(implied_nak_seq_err, hwerrors, "Pkt Seq Num gap", 1, __VA_ARGS__) \ |
| 86 | GEN(local_ack_timeout_err, hwerrors, "Ack timer expired", 1, __VA_ARGS__) \ |
| 87 | GEN(out_of_buffer, hwerrors, "Drop missing buffer", 1, __VA_ARGS__) \ |
| 88 | GEN(out_of_sequence, hwerrors, "Drop out of sequence", 1, __VA_ARGS__) \ |
| 89 | GEN(packet_seq_err, hwerrors, "NAK sequence rcvd", 1, __VA_ARGS__) \ |
| 90 | GEN(req_cqe_error, hwerrors, "CQE err Req", 1, __VA_ARGS__) \ |
| 91 | GEN(resp_cqe_error, hwerrors, "CQE err Resp", 1, __VA_ARGS__) \ |
| 92 | GEN(req_cqe_flush_error, hwerrors, "CQE Flushed err Req", 1, __VA_ARGS__) \ |
| 93 | GEN(resp_cqe_flush_error, hwerrors, "CQE Flushed err Resp", 1, __VA_ARGS__) \ |
| 94 | GEN(req_remote_access_errors, hwerrors, "Remote access err Req", 1, __VA_ARGS__) \ |
| 95 | GEN(resp_remote_access_errors, hwerrors, "Remote access err Resp", 1, __VA_ARGS__) \ |
| 96 | GEN(req_remote_invalid_request, hwerrors, "Remote invalid req", 1, __VA_ARGS__) \ |
| 97 | GEN(resp_local_length_error, hwerrors, "Local length err Resp", 1, __VA_ARGS__) \ |
| 98 | GEN(rnr_nak_retry_err, hwerrors, "RNR NAK Packets", 1, __VA_ARGS__) \ |
| 99 | GEN(rp_cnp_ignored, hwerrors, "CNP Pkts ignored", 1, __VA_ARGS__) \ |
| 100 | GEN(rx_icrc_encapsulated, hwerrors, "RoCE ICRC Errors", 1, __VA_ARGS__) |
| 101 | |
| 102 | // Common definitions used more than once |
| 103 | #define GEN_RRD_DIM_ADD(NAME, GRP, DESC, DIR, PORT) \ |
| 104 | GEN_RRD_DIM_ADD_CUSTOM(NAME, GRP, DESC, DIR, PORT, 1, 1, RRD_ALGORITHM_INCREMENTAL) |
| 105 | |
| 106 | #define GEN_RRD_DIM_ADD_CUSTOM(NAME, GRP, DESC, DIR, PORT, MULT, DIV, ALGO) \ |
| 107 | PORT->rd_##NAME = rrddim_add(PORT->st_##GRP, DESC, NULL, DIR * MULT, DIV, ALGO); |
| 108 | |
| 109 | #define GEN_RRD_DIM_ADD_HW(NAME, GRP, DESC, DIR, PORT, HW) \ |
| 110 | HW->rd_##NAME = rrddim_add(PORT->st_##GRP, DESC, NULL, DIR, 1, RRD_ALGORITHM_INCREMENTAL); |
| 111 | |
| 112 | #define GEN_RRD_DIM_SETP(NAME, GRP, DESC, DIR, PORT) \ |
| 113 | rrddim_set_by_pointer(PORT->st_##GRP, PORT->rd_##NAME, (collected_number)PORT->NAME); |
| 114 | |
| 115 | #define GEN_RRD_DIM_SETP_HW(NAME, GRP, DESC, DIR, PORT, HW) \ |
| 116 | rrddim_set_by_pointer(PORT->st_##GRP, HW->rd_##NAME, (collected_number)HW->NAME); |
| 117 | |
| 118 | // https://community.mellanox.com/s/article/understanding-mlx5-linux-counters-and-status-parameters |
| 119 | // https://community.mellanox.com/s/article/infiniband-port-counters |
| 120 | static struct ibport { |
| 121 | char *name; |
| 122 | char *counters_path; |
| 123 | char *hwcounters_path; |
| 124 | int len; |
| 125 | |
| 126 | // flags |
| 127 | int configured; |
| 128 | int enabled; |
| 129 | int discovered; |
| 130 | |
| 131 | int do_bytes; |
| 132 | int do_packets; |
| 133 | int do_errors; |
| 134 | int do_hwpackets; |
| 135 | int do_hwerrors; |
| 136 | |
| 137 | const char *chart_type_bytes; |
| 138 | const char *chart_type_packets; |
| 139 | const char *chart_type_errors; |
| 140 | const char *chart_type_hwpackets; |
| 141 | const char *chart_type_hwerrors; |
| 142 | |
| 143 | const char *chart_id_bytes; |
| 144 | const char *chart_id_packets; |
| 145 | const char *chart_id_errors; |
| 146 | const char *chart_id_hwpackets; |
| 147 | const char *chart_id_hwerrors; |
| 148 | |
| 149 | const char *chart_family; |
| 150 | |
| 151 | unsigned long priority; |
| 152 | |
| 153 | // Port details using drivers/infiniband/core/sysfs.c :: rate_show() |
| 154 | RRDDIM *rd_speed; |
| 155 | uint64_t speed; |
| 156 | uint64_t width; |
| 157 | |
| 158 | // Stats from /$device/ports/$portid/counters |
| 159 | // as drivers/infiniband/hw/qib/qib_verbs.h |
| 160 | // All uint64 except vl15_dropped, local_link_integrity_errors, excessive_buffer_overrun_errors uint32 |
| 161 | // Will generate 2 elements for each counter: |
| 162 | // - uint64_t to store the value |
| 163 | // - char* to store the filename path |
| 164 | // - RRDDIM* to store the RRD Dimension |
| 165 | #define GEN_DEF_COUNTER(NAME, ...) \ |
| 166 | uint64_t NAME; \ |
| 167 | char *file_##NAME; \ |
| 168 | RRDDIM *rd_##NAME; |
| 169 | FOREACH_COUNTER(GEN_DEF_COUNTER) |
| 170 | |
| 171 | // Vendor specific hwcounters from /$device/ports/$portid/hw_counters |
| 172 | // We will generate one struct pointer per vendor to avoid future casting |
| 173 | #define GEN_DEF_HWCOUNTER_PTR(VENDOR, ...) struct ibporthw_##VENDOR *hwcounters_##VENDOR; |
| 174 | FOREACH_HWCOUNTER_NAME(GEN_DEF_HWCOUNTER_PTR) |
| 175 | |
| 176 | // Function pointer to the "infiniband_hwcounters_parse_<vendor>" function |
| 177 | void (*hwcounters_parse)(struct ibport *); |
| 178 | void (*hwcounters_dorrd)(struct ibport *); |
| 179 | |
| 180 | // charts and dim |
| 181 | RRDSET *st_bytes; |
| 182 | RRDSET *st_packets; |
| 183 | RRDSET *st_errors; |
| 184 | RRDSET *st_hwpackets; |
| 185 | RRDSET *st_hwerrors; |
| 186 | |
| 187 | const RRDVAR_ACQUIRED *stv_speed; |
| 188 | |
| 189 | |
| 190 | usec_t speed_last_collected_usec; |
| 191 | |
| 192 | struct ibport *next; |
| 193 | } *ibport_root = NULL, *ibport_last_used = NULL; |
| 194 | |
| 195 | // VENDORS: reading / calculation functions |
| 196 | #define GEN_DEF_HWCOUNTER(NAME, ...) \ |
| 197 | uint64_t NAME; \ |
| 198 | char *file_##NAME; \ |
| 199 | RRDDIM *rd_##NAME; |
| 200 | |
| 201 | #define GEN_DO_HWCOUNTER_READ(NAME, GRP, DESC, DIR, PORT, HW, ...) \ |
| 202 | if (HW->file_##NAME) { \ |
| 203 | if (read_single_number_file(HW->file_##NAME, (unsigned long long *)&HW->NAME)) { \ |
| 204 | collector_error("cannot read iface '%s' hwcounter '" #HW "'", PORT->name); \ |
| 205 | HW->file_##NAME = NULL; \ |
| 206 | } \ |
| 207 | } |
| 208 | |
| 209 | // VENDOR-MLX: Mellanox |
| 210 | struct ibporthw_mlx { |
| 211 | FOREACH_HWCOUNTER_MLX(GEN_DEF_HWCOUNTER) |
| 212 | }; |
| 213 | void infiniband_hwcounters_parse_mlx(struct ibport *port) |
| 214 | { |
| 215 | if (port->do_hwerrors != CONFIG_BOOLEAN_NO) |
| 216 | FOREACH_HWCOUNTER_MLX_ERRORS(GEN_DO_HWCOUNTER_READ, port, port->hwcounters_mlx) |
| 217 | if (port->do_hwpackets != CONFIG_BOOLEAN_NO) |
| 218 | FOREACH_HWCOUNTER_MLX_PACKETS(GEN_DO_HWCOUNTER_READ, port, port->hwcounters_mlx) |
| 219 | } |
| 220 | void sys_class_infiniband_plugin_cleanup(void) { |
| 221 | // Cleanup any acquired RRDVARs |
| 222 | struct ibport *port; |
| 223 | for (port = ibport_root; port; port = port->next) { |
| 224 | if (port->stv_speed) { |
| 225 | rrdvar_chart_variable_release(port->st_bytes, port->stv_speed); |
| 226 | port->stv_speed = NULL; |
| 227 | } |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | void infiniband_hwcounters_dorrd_mlx(struct ibport *port) |
| 232 | { |
| 233 | if (port->do_hwerrors != CONFIG_BOOLEAN_NO) { |
| 234 | FOREACH_HWCOUNTER_MLX_ERRORS(GEN_RRD_DIM_SETP_HW, port, port->hwcounters_mlx) |
| 235 | rrdset_done(port->st_hwerrors); |
| 236 | } |
| 237 | if (port->do_hwpackets != CONFIG_BOOLEAN_NO) { |
| 238 | FOREACH_HWCOUNTER_MLX_PACKETS(GEN_RRD_DIM_SETP_HW, port, port->hwcounters_mlx) |
| 239 | rrdset_done(port->st_hwpackets); |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | // ---------------------------------------------------------------------------- |
| 244 | |
| 245 | static struct ibport *get_ibport(const char *dev, const char *port) |
| 246 | { |
| 247 | struct ibport *p; |
| 248 | |
| 249 | char name[IBNAME_MAX + 1]; |
| 250 | snprintfz(name, IBNAME_MAX, "%s-%s", dev, port); |
| 251 | |
| 252 | // search it, resuming from the last position in sequence |
| 253 | for (p = ibport_last_used; p; p = p->next) { |
| 254 | if (unlikely(!strcmp(name, p->name))) { |
| 255 | ibport_last_used = p->next; |
| 256 | return p; |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | // new round, from the beginning to the last position used this time |
| 261 | for (p = ibport_root; p != ibport_last_used; p = p->next) { |
| 262 | if (unlikely(!strcmp(name, p->name))) { |
| 263 | ibport_last_used = p->next; |
| 264 | return p; |
| 265 | } |
| 266 | } |
| 267 | |
| 268 | // create a new one |
| 269 | p = callocz(1, sizeof(struct ibport)); |
| 270 | p->name = strdupz(name); |
| 271 | p->len = strlen(p->name); |
| 272 | |
| 273 | p->chart_type_bytes = strdupz("infiniband_cnt_bytes"); |
| 274 | p->chart_type_packets = strdupz("infiniband_cnt_packets"); |
| 275 | p->chart_type_errors = strdupz("infiniband_cnt_errors"); |
| 276 | p->chart_type_hwpackets = strdupz("infiniband_hwc_packets"); |
| 277 | p->chart_type_hwerrors = strdupz("infiniband_hwc_errors"); |
| 278 | |
| 279 | char buffer[RRD_ID_LENGTH_MAX + 1]; |
| 280 | snprintfz(buffer, RRD_ID_LENGTH_MAX, "ib_cntbytes_%s", p->name); |
| 281 | p->chart_id_bytes = strdupz(buffer); |
| 282 | |
| 283 | snprintfz(buffer, RRD_ID_LENGTH_MAX, "ib_cntpackets_%s", p->name); |
| 284 | p->chart_id_packets = strdupz(buffer); |
| 285 | |
| 286 | snprintfz(buffer, RRD_ID_LENGTH_MAX, "ib_cnterrors_%s", p->name); |
| 287 | p->chart_id_errors = strdupz(buffer); |
| 288 | |
| 289 | snprintfz(buffer, RRD_ID_LENGTH_MAX, "ib_hwcntpackets_%s", p->name); |
| 290 | p->chart_id_hwpackets = strdupz(buffer); |
| 291 | |
| 292 | snprintfz(buffer, RRD_ID_LENGTH_MAX, "ib_hwcnterrors_%s", p->name); |
| 293 | p->chart_id_hwerrors = strdupz(buffer); |
| 294 | |
| 295 | p->chart_family = strdupz(p->name); |
| 296 | p->priority = NETDATA_CHART_PRIO_INFINIBAND; |
| 297 | |
| 298 | // Link current ibport to last one in the list |
| 299 | if (ibport_root) { |
| 300 | struct ibport *t; |
| 301 | for (t = ibport_root; t->next; t = t->next) |
| 302 | ; |
| 303 | t->next = p; |
| 304 | } else |
| 305 | ibport_root = p; |
| 306 | |
| 307 | return p; |
| 308 | } |
| 309 | |
| 310 | int do_sys_class_infiniband(int update_every, usec_t dt) |
| 311 | { |
| 312 | (void)dt; |
| 313 | static SIMPLE_PATTERN *disabled_list = NULL; |
| 314 | static int initialized = 0; |
| 315 | static int enable_new_ports = -1, enable_only_active = CONFIG_BOOLEAN_YES; |
| 316 | static int do_bytes = -1, do_packets = -1, do_errors = -1, do_hwpackets = -1, do_hwerrors = -1; |
| 317 | static const char *sys_class_infiniband_dirname = NULL; |
| 318 | |
| 319 | static long long int dt_to_refresh_ports = 0, last_refresh_ports_usec = 0; |
| 320 | |
| 321 | if (unlikely(enable_new_ports == -1)) { |
| 322 | char dirname[FILENAME_MAX + 1]; |
| 323 | |
| 324 | snprintfz(dirname, FILENAME_MAX, "%s%s", netdata_configured_host_prefix, "/sys/class/infiniband"); |
| 325 | sys_class_infiniband_dirname = |
| 326 | inicfg_get(&netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "dirname to monitor", dirname); |
| 327 | |
| 328 | do_bytes = inicfg_get_boolean_ondemand( |
| 329 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "bandwidth counters", CONFIG_BOOLEAN_YES); |
| 330 | do_packets = inicfg_get_boolean_ondemand( |
| 331 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "packets counters", CONFIG_BOOLEAN_YES); |
| 332 | do_errors = inicfg_get_boolean_ondemand( |
| 333 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "errors counters", CONFIG_BOOLEAN_YES); |
| 334 | do_hwpackets = inicfg_get_boolean_ondemand( |
| 335 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "hardware packets counters", CONFIG_BOOLEAN_AUTO); |
| 336 | do_hwerrors = inicfg_get_boolean_ondemand( |
| 337 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "hardware errors counters", CONFIG_BOOLEAN_AUTO); |
| 338 | |
| 339 | enable_only_active = inicfg_get_boolean_ondemand( |
| 340 | &netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "monitor only active ports", CONFIG_BOOLEAN_AUTO); |
| 341 | disabled_list = simple_pattern_create( |
| 342 | inicfg_get(&netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "disable by default interfaces matching", ""), |
| 343 | NULL, |
| 344 | SIMPLE_PATTERN_EXACT, true); |
| 345 | |
| 346 | dt_to_refresh_ports = |
| 347 | inicfg_get_duration_seconds(&netdata_config, CONFIG_SECTION_PLUGIN_SYS_CLASS_INFINIBAND, "refresh ports state every", 30) * |
| 348 | USEC_PER_SEC; |
| 349 | if (dt_to_refresh_ports < 0) |
| 350 | dt_to_refresh_ports = 0; |
| 351 | } |
| 352 | |
| 353 | // init listing of /sys/class/infiniband/ (or rediscovery) |
| 354 | if (unlikely(!initialized) || unlikely(last_refresh_ports_usec >= dt_to_refresh_ports)) { |
| 355 | // If folder does not exists, return 1 to disable |
| 356 | DIR *devices_dir = opendir(sys_class_infiniband_dirname); |
| 357 | if (unlikely(!devices_dir)) |
| 358 | return 1; |
| 359 | |
| 360 | // Work on all device available |
| 361 | struct dirent *dev_dent; |
| 362 | while ((dev_dent = readdir(devices_dir))) { |
| 363 | // Skip special folders |
| 364 | if (!strcmp(dev_dent->d_name, "..") || !strcmp(dev_dent->d_name, ".")) |
| 365 | continue; |
| 366 | |
| 367 | // /sys/class/infiniband/<dev>/ports |
| 368 | char ports_dirname[FILENAME_MAX + 1]; |
| 369 | snprintfz(ports_dirname, FILENAME_MAX, "%s/%s/%s", sys_class_infiniband_dirname, dev_dent->d_name, "ports"); |
| 370 | |
| 371 | DIR *ports_dir = opendir(ports_dirname); |
| 372 | if (unlikely(!ports_dir)) |
| 373 | continue; |
| 374 | |
| 375 | struct dirent *port_dent; |
| 376 | while ((port_dent = readdir(ports_dir))) { |
| 377 | // Skip special folders |
| 378 | if (!strcmp(port_dent->d_name, "..") || !strcmp(port_dent->d_name, ".")) |
| 379 | continue; |
| 380 | |
| 381 | char buffer[FILENAME_MAX + 1]; |
| 382 | |
| 383 | // Check if counters are available (mandatory) |
| 384 | // /sys/class/infiniband/<device>/ports/<port>/counters |
| 385 | char counters_dirname[FILENAME_MAX + 1]; |
| 386 | snprintfz(counters_dirname, FILENAME_MAX, "%s/%s/%s", ports_dirname, port_dent->d_name, "counters"); |
| 387 | DIR *counters_dir = opendir(counters_dirname); |
| 388 | // Standard counters are mandatory |
| 389 | if (!counters_dir) |
| 390 | continue; |
| 391 | closedir(counters_dir); |
| 392 | |
| 393 | // Hardware Counters are optional, used later |
| 394 | char hwcounters_dirname[FILENAME_MAX + 1]; |
| 395 | snprintfz( |
| 396 | hwcounters_dirname, FILENAME_MAX, "%s/%s/%s", ports_dirname, port_dent->d_name, "hw_counters"); |
| 397 | |
| 398 | // Get new or existing ibport |
| 399 | struct ibport *p = get_ibport(dev_dent->d_name, port_dent->d_name); |
| 400 | if (!p) |
| 401 | continue; |
| 402 | |
| 403 | // Prepare configuration |
| 404 | if (!p->configured) { |
| 405 | p->configured = 1; |
| 406 | |
| 407 | // Enable by default, will be filtered out later |
| 408 | p->enabled = 1; |
| 409 | |
| 410 | p->counters_path = strdupz(counters_dirname); |
| 411 | p->hwcounters_path = strdupz(hwcounters_dirname); |
| 412 | |
| 413 | snprintfz(buffer, FILENAME_MAX, "plugin:proc:/sys/class/infiniband:%s", p->name); |
| 414 | |
| 415 | // Standard counters |
| 416 | p->do_bytes = inicfg_get_boolean_ondemand(&netdata_config, buffer, "bytes", do_bytes); |
| 417 | p->do_packets = inicfg_get_boolean_ondemand(&netdata_config, buffer, "packets", do_packets); |
| 418 | p->do_errors = inicfg_get_boolean_ondemand(&netdata_config, buffer, "errors", do_errors); |
| 419 | |
| 420 | // Gen filename allocation and concatenation |
| 421 | #define GEN_DO_COUNTER_NAME(NAME, GRP, DESC, DIR, PORT, ...) \ |
| 422 | PORT->file_##NAME = callocz(1, strlen(PORT->counters_path) + sizeof(#NAME) + 3); \ |
| 423 | strcat(PORT->file_##NAME, PORT->counters_path); \ |
| 424 | strcat(PORT->file_##NAME, "/" #NAME); |
| 425 | FOREACH_COUNTER(GEN_DO_COUNTER_NAME, p) |
| 426 | |
| 427 | // Check HW Counters vendor dependent |
| 428 | DIR *hwcounters_dir = opendir(hwcounters_dirname); |
| 429 | if (hwcounters_dir) { |
| 430 | // By default set standard |
| 431 | p->do_hwpackets = inicfg_get_boolean_ondemand(&netdata_config, buffer, "hwpackets", do_hwpackets); |
| 432 | p->do_hwerrors = inicfg_get_boolean_ondemand(&netdata_config, buffer, "hwerrors", do_hwerrors); |
| 433 | |
| 434 | // VENDORS: Set your own |
| 435 | |
| 436 | // Allocate the chars to the filenames |
| 437 | #define GEN_DO_HWCOUNTER_NAME(NAME, GRP, DESC, DIR, PORT, HW, ...) \ |
| 438 | HW->file_##NAME = callocz(1, strlen(PORT->hwcounters_path) + sizeof(#NAME) + 3); \ |
| 439 | strcat(HW->file_##NAME, PORT->hwcounters_path); \ |
| 440 | strcat(HW->file_##NAME, "/" #NAME); |
| 441 | |
| 442 | // VENDOR-MLX: Mellanox |
| 443 | if (strncmp(dev_dent->d_name, "mlx", 3) == 0) { |
| 444 | // Allocate the vendor specific struct |
| 445 | p->hwcounters_mlx = callocz(1, sizeof(struct ibporthw_mlx)); |
| 446 | |
| 447 | FOREACH_HWCOUNTER_MLX(GEN_DO_HWCOUNTER_NAME, p, p->hwcounters_mlx) |
| 448 | |
| 449 | // Set the function pointer for hwcounter parsing |
| 450 | p->hwcounters_parse = &infiniband_hwcounters_parse_mlx; |
| 451 | p->hwcounters_dorrd = &infiniband_hwcounters_dorrd_mlx; |
| 452 | } |
| 453 | |
| 454 | // VENDOR: Unknown |
| 455 | else { |
| 456 | p->do_hwpackets = CONFIG_BOOLEAN_NO; |
| 457 | p->do_hwerrors = CONFIG_BOOLEAN_NO; |
| 458 | } |
| 459 | closedir(hwcounters_dir); |
| 460 | } |
| 461 | } |
| 462 | |
| 463 | // Check port state to keep activation |
| 464 | if (enable_only_active) { |
| 465 | snprintfz(buffer, FILENAME_MAX, "%s/%s/%s", ports_dirname, port_dent->d_name, "state"); |
| 466 | unsigned long long active; |
| 467 | // File is "1: DOWN" or "4: ACTIVE", but str2ull will stop on first non-decimal char |
| 468 | read_single_number_file(buffer, &active); |
| 469 | |
| 470 | // Want "IB_PORT_ACTIVE" == "4", as defined by drivers/infiniband/core/sysfs.c::state_show() |
| 471 | if (active == 4) |
| 472 | p->enabled = 1; |
| 473 | else |
| 474 | p->enabled = 0; |
| 475 | } |
| 476 | |
| 477 | if (p->enabled) |
| 478 | p->enabled = !simple_pattern_matches(disabled_list, p->name); |
| 479 | |
| 480 | // Check / Update the link speed & width frm "rate" file |
| 481 | // Sample output: "100 Gb/sec (4X EDR)" |
| 482 | snprintfz(buffer, FILENAME_MAX, "%s/%s/%s", ports_dirname, port_dent->d_name, "rate"); |
| 483 | char buffer_rate[65]; |
| 484 | p->width = 4; |
| 485 | if (read_txt_file(buffer, buffer_rate, sizeof(buffer_rate))) { |
| 486 | collector_error("Unable to read '%s'", buffer); |
| 487 | } else { |
| 488 | char *buffer_width = strstr(buffer_rate, "("); |
| 489 | if (buffer_width) { |
| 490 | buffer_width++; |
| 491 | // str2ull will stop on first non-decimal value |
| 492 | p->speed = str2ull(buffer_rate, NULL); |
| 493 | p->width = str2ull(buffer_width, NULL); |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | if (!p->discovered) |
| 498 | collector_info("Infiniband card %s port %s at speed %" PRIu64 " width %" PRIu64 "", |
| 499 | dev_dent->d_name, |
| 500 | port_dent->d_name, |
| 501 | p->speed, |
| 502 | p->width); |
| 503 | |
| 504 | p->discovered = 1; |
| 505 | } |
| 506 | closedir(ports_dir); |
| 507 | } |
| 508 | closedir(devices_dir); |
| 509 | |
| 510 | initialized = 1; |
| 511 | last_refresh_ports_usec = 0; |
| 512 | } |
| 513 | last_refresh_ports_usec += dt; |
| 514 | |
| 515 | // Update all ports values |
| 516 | struct ibport *port; |
| 517 | for (port = ibport_root; port; port = port->next) { |
| 518 | if (!port->enabled) |
| 519 | continue; |
| 520 | // |
| 521 | // Read values from system to struct |
| 522 | // |
| 523 | |
| 524 | // counter from file and place it in ibport struct |
| 525 | #define GEN_DO_COUNTER_READ(NAME, GRP, DESC, DIR, PORT, ...) \ |
| 526 | if (PORT->file_##NAME) { \ |
| 527 | if (read_single_number_file(PORT->file_##NAME, (unsigned long long *)&PORT->NAME)) { \ |
| 528 | collector_error("cannot read iface '%s' counter '" #NAME "'", PORT->name); \ |
| 529 | PORT->file_##NAME = NULL; \ |
| 530 | } \ |
| 531 | } |
| 532 | |
| 533 | // Update charts |
| 534 | if (port->do_bytes != CONFIG_BOOLEAN_NO) { |
| 535 | // Read values from sysfs |
| 536 | FOREACH_COUNTER_BYTES(GEN_DO_COUNTER_READ, port) |
| 537 | |
| 538 | // First creation of RRD Set (charts) |
| 539 | if (unlikely(!port->st_bytes)) { |
| 540 | port->st_bytes = rrdset_create_localhost( |
| 541 | "Infiniband", |
| 542 | port->chart_id_bytes, |
| 543 | NULL, |
| 544 | port->chart_family, |
| 545 | "ib.bytes", |
| 546 | "Bandwidth usage", |
| 547 | "kilobits/s", |
| 548 | PLUGIN_PROC_NAME, |
| 549 | PLUGIN_PROC_MODULE_INFINIBAND_NAME, |
| 550 | port->priority + 1, |
| 551 | update_every, |
| 552 | RRDSET_TYPE_AREA); |
| 553 | |
| 554 | // On this chart, we want to have a KB/s so the dashboard will autoscale it |
| 555 | // The reported values are also per-lane, so we must multiply it by the width |
| 556 | // x4 lanes multiplier as per Documentation/ABI/stable/sysfs-class-infiniband |
| 557 | FOREACH_COUNTER_BYTES(GEN_RRD_DIM_ADD_CUSTOM, port, port->width * 8, 1000, RRD_ALGORITHM_INCREMENTAL) |
| 558 | |
| 559 | port->stv_speed = rrdvar_chart_variable_add_and_acquire(port->st_bytes, "link_speed"); |
| 560 | } |
| 561 | |
| 562 | // Link read values to dimensions |
| 563 | FOREACH_COUNTER_BYTES(GEN_RRD_DIM_SETP, port) |
| 564 | |
| 565 | // For link speed set only variable |
| 566 | rrdvar_chart_variable_set(port->st_bytes, port->stv_speed, port->speed); |
| 567 | |
| 568 | rrdset_done(port->st_bytes); |
| 569 | } |
| 570 | |
| 571 | if (port->do_packets != CONFIG_BOOLEAN_NO) { |
| 572 | // Read values from sysfs |
| 573 | FOREACH_COUNTER_PACKETS(GEN_DO_COUNTER_READ, port) |
| 574 | |
| 575 | // First creation of RRD Set (charts) |
| 576 | if (unlikely(!port->st_packets)) { |
| 577 | port->st_packets = rrdset_create_localhost( |
| 578 | "Infiniband", |
| 579 | port->chart_id_packets, |
| 580 | NULL, |
| 581 | port->chart_family, |
| 582 | "ib.packets", |
| 583 | "Packets Statistics", |
| 584 | "packets/s", |
| 585 | PLUGIN_PROC_NAME, |
| 586 | PLUGIN_PROC_MODULE_INFINIBAND_NAME, |
| 587 | port->priority + 2, |
| 588 | update_every, |
| 589 | RRDSET_TYPE_AREA); |
| 590 | |
| 591 | FOREACH_COUNTER_PACKETS(GEN_RRD_DIM_ADD, port) |
| 592 | } |
| 593 | |
| 594 | // Link read values to dimensions |
| 595 | FOREACH_COUNTER_PACKETS(GEN_RRD_DIM_SETP, port) |
| 596 | rrdset_done(port->st_packets); |
| 597 | } |
| 598 | |
| 599 | if (port->do_errors != CONFIG_BOOLEAN_NO) { |
| 600 | // Read values from sysfs |
| 601 | FOREACH_COUNTER_ERRORS(GEN_DO_COUNTER_READ, port) |
| 602 | |
| 603 | // First creation of RRD Set (charts) |
| 604 | if (unlikely(!port->st_errors)) { |
| 605 | port->st_errors = rrdset_create_localhost( |
| 606 | "Infiniband", |
| 607 | port->chart_id_errors, |
| 608 | NULL, |
| 609 | port->chart_family, |
| 610 | "ib.errors", |
| 611 | "Error Counters", |
| 612 | "errors/s", |
| 613 | PLUGIN_PROC_NAME, |
| 614 | PLUGIN_PROC_MODULE_INFINIBAND_NAME, |
| 615 | port->priority + 3, |
| 616 | update_every, |
| 617 | RRDSET_TYPE_LINE); |
| 618 | |
| 619 | FOREACH_COUNTER_ERRORS(GEN_RRD_DIM_ADD, port) |
| 620 | } |
| 621 | |
| 622 | // Link read values to dimensions |
| 623 | FOREACH_COUNTER_ERRORS(GEN_RRD_DIM_SETP, port) |
| 624 | rrdset_done(port->st_errors); |
| 625 | } |
| 626 | |
| 627 | // |
| 628 | // HW Counters |
| 629 | // |
| 630 | |
| 631 | // Call the function for parsing and setting hwcounters |
| 632 | if (port->hwcounters_parse && port->hwcounters_dorrd) { |
| 633 | // Read all values (done by vendor-specific function) |
| 634 | (*port->hwcounters_parse)(port); |
| 635 | |
| 636 | if (port->do_hwerrors != CONFIG_BOOLEAN_NO) { |
| 637 | // First creation of RRD Set (charts) |
| 638 | if (unlikely(!port->st_hwerrors)) { |
| 639 | port->st_hwerrors = rrdset_create_localhost( |
| 640 | "Infiniband", |
| 641 | port->chart_id_hwerrors, |
| 642 | NULL, |
| 643 | port->chart_family, |
| 644 | "ib.hwerrors", |
| 645 | "Hardware Errors", |
| 646 | "errors/s", |
| 647 | PLUGIN_PROC_NAME, |
| 648 | PLUGIN_PROC_MODULE_INFINIBAND_NAME, |
| 649 | port->priority + 4, |
| 650 | update_every, |
| 651 | RRDSET_TYPE_LINE); |
| 652 | |
| 653 | // VENDORS: Set your selection |
| 654 | |
| 655 | // VENDOR: Mellanox |
| 656 | if (strncmp(port->name, "mlx", 3) == 0) { |
| 657 | FOREACH_HWCOUNTER_MLX_ERRORS(GEN_RRD_DIM_ADD_HW, port, port->hwcounters_mlx) |
| 658 | } |
| 659 | |
| 660 | // Unknown vendor, should not happen |
| 661 | else { |
| 662 | collector_error( |
| 663 | "Unmanaged vendor for '%s', do_hwerrors should have been set to no. Please report this bug", |
| 664 | port->name); |
| 665 | port->do_hwerrors = CONFIG_BOOLEAN_NO; |
| 666 | } |
| 667 | } |
| 668 | } |
| 669 | |
| 670 | if (port->do_hwpackets != CONFIG_BOOLEAN_NO) { |
| 671 | // First creation of RRD Set (charts) |
| 672 | if (unlikely(!port->st_hwpackets)) { |
| 673 | port->st_hwpackets = rrdset_create_localhost( |
| 674 | "Infiniband", |
| 675 | port->chart_id_hwpackets, |
| 676 | NULL, |
| 677 | port->chart_family, |
| 678 | "ib.hwpackets", |
| 679 | "Hardware Packets Statistics", |
| 680 | "packets/s", |
| 681 | PLUGIN_PROC_NAME, |
| 682 | PLUGIN_PROC_MODULE_INFINIBAND_NAME, |
| 683 | port->priority + 5, |
| 684 | update_every, |
| 685 | RRDSET_TYPE_LINE); |
| 686 | |
| 687 | // VENDORS: Set your selection |
| 688 | |
| 689 | // VENDOR: Mellanox |
| 690 | if (strncmp(port->name, "mlx", 3) == 0) { |
| 691 | FOREACH_HWCOUNTER_MLX_PACKETS(GEN_RRD_DIM_ADD_HW, port, port->hwcounters_mlx) |
| 692 | } |
| 693 | |
| 694 | // Unknown vendor, should not happen |
| 695 | else { |
| 696 | collector_error( |
| 697 | "Unmanaged vendor for '%s', do_hwpackets should have been set to no. Please report this bug", |
| 698 | port->name); |
| 699 | port->do_hwpackets = CONFIG_BOOLEAN_NO; |
| 700 | } |
| 701 | } |
| 702 | } |
| 703 | |
| 704 | // Update values to rrd (done by vendor-specific function) |
| 705 | (*port->hwcounters_dorrd)(port); |
| 706 | } |
| 707 | } |
| 708 | |
| 709 | return 0; |
| 710 | } |