contrib/ofed/management/opensm/opensm/osm_trap_rcv.c

   1 /*
   2  * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
   3  * Copyright (c) 2002-2006 Mellanox Technologies LTD. All rights reserved.
   4  * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
   5  *
   6  * This software is available to you under a choice of one of two
   7  * licenses.  You may choose to be licensed under the terms of the GNU
   8  * General Public License (GPL) Version 2, available from the file
   9  * COPYING in the main directory of this source tree, or the
  10  * OpenIB.org BSD license below:
  11  *
  12  *     Redistribution and use in source and binary forms, with or
  13  *     without modification, are permitted provided that the following
  14  *     conditions are met:
  15  *
  16  *      - Redistributions of source code must retain the above
  17  *        copyright notice, this list of conditions and the following
  18  *        disclaimer.
  19  *
  20  *      - Redistributions in binary form must reproduce the above
  21  *        copyright notice, this list of conditions and the following
  22  *        disclaimer in the documentation and/or other materials
  23  *        provided with the distribution.
  24  *
  25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32  * SOFTWARE.
  33  *
  34  */
  35
  36 /*
  37  * Abstract:
  38  *    Implementation of osm_trap_rcv_t.
  39  * This object represents the Trap Receiver object.
  40  * This object is part of the opensm family of objects.
  41  */
  42
  43 #if HAVE_CONFIG_H
  44 #  include <config.h>
  45 #endif                          /* HAVE_CONFIG_H */
  46
  47 #include <string.h>
  48 #include <iba/ib_types.h>
  49 #include <complib/cl_qmap.h>
  50 #include <complib/cl_debug.h>
  51 #include <opensm/osm_madw.h>
  52 #include <opensm/osm_log.h>
  53 #include <opensm/osm_node.h>
  54 #include <opensm/osm_helper.h>
  55 #include <opensm/osm_subnet.h>
  56 #include <opensm/osm_inform.h>
  57 #include <opensm/osm_opensm.h>
  58
  59 extern void osm_req_get_node_desc(IN osm_sm_t * sm, osm_physp_t *p_physp);
  60
  61 /**********************************************************************
  62  *
  63  * TRAP HANDLING:
  64  *
  65  * Assuming traps can be caused by bad hardware we should provide
  66  * a mechanism for filtering their propagation into the actual logic
  67  * of OpenSM such that it is not overloaded by them.
  68  *
  69  * We will provide a trap filtering mechanism with "Aging" capability.
  70  * This mechanism will track incoming traps, clasify them by their
  71  * source and content and provide back their age.
  72  *
  73  * A timer running in the background will toggle a timer counter
  74  * that should be referenced by the aging algorithm.
  75  * To provide an efficient handling of aging. We also track all traps
  76  * in a sorted list by their aging.
  77  *
  78  * The generic Aging Tracker mechanism is implemented in the
  79  * cl_aging_tracker object.
  80  *
  81  **********************************************************************/
  82
  83 typedef struct osm_trap_agingracker_context {
  84         osm_log_t *p_log;
  85         osm_physp_t *p_physp;
  86 } osm_trap_aging_tracker_context_t;
  87
  88 /**********************************************************************
  89  **********************************************************************/
  90 static osm_physp_t *get_physp_by_lid_and_num(IN osm_sm_t * sm,
  91                                              IN uint16_t lid, IN uint8_t num)
  92 {
  93         cl_ptr_vector_t *p_vec = &(sm->p_subn->port_lid_tbl);
  94         osm_port_t *p_port;
  95
  96         if (lid > cl_ptr_vector_get_size(p_vec))
  97                 return NULL;
  98
  99         p_port = (osm_port_t *) cl_ptr_vector_get(p_vec, lid);
 100         if (!p_port)
 101                 return NULL;
 102
 103         if (osm_node_get_num_physp(p_port->p_node) < num)
 104                 return NULL;
 105
 106         return osm_node_get_physp_ptr(p_port->p_node, num);
 107 }
 108
 109 /**********************************************************************
 110  **********************************************************************/
 111 uint64_t
 112 osm_trap_rcv_aging_tracker_callback(IN uint64_t key,
 113                                     IN uint32_t num_regs, IN void *context)
 114 {
 115         osm_sm_t *sm = context;
 116         uint16_t lid;
 117         uint8_t port_num;
 118         osm_physp_t *p_physp;
 119
 120         OSM_LOG_ENTER(sm->p_log);
 121
 122         if (osm_exit_flag)
 123                 /* We got an exit flag - do nothing */
 124                 return 0;
 125
 126         lid = cl_ntoh16((uint16_t) ((key & 0x0000FFFF00000000ULL) >> 32));
 127         port_num = (uint8_t) ((key & 0x00FF000000000000ULL) >> 48);
 128
 129         p_physp = get_physp_by_lid_and_num(sm, lid, port_num);
 130         if (!p_physp)
 131                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 132                         "Cannot find port num:%u with lid:%u\n",
 133                         port_num, lid);
 134         /* make sure the physp is still valid */
 135         /* If the health port was false - set it to true */
 136         else if (!osm_physp_is_healthy(p_physp)) {
 137                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 138                         "Clearing health bit of port num:%u with lid:%u\n",
 139                         port_num, lid);
 140
 141                 /* Clear its health bit */
 142                 osm_physp_set_health(p_physp, TRUE);
 143         }
 144
 145         OSM_LOG_EXIT(sm->p_log);
 146
 147         /* We want to remove the event from the tracker - so
 148            need to return zero. */
 149         return 0;
 150 }
 151
 152 /**********************************************************************
 153  * CRC calculation for notice identification
 154  **********************************************************************/
 155
 156 #define CRC32_POLYNOMIAL   0xEDB88320L
 157
 158 /* calculate the crc for a given buffer */
 159 static uint32_t __osm_trap_calc_crc32(void *buffer, uint32_t count)
 160 {
 161         uint32_t temp1, temp2;
 162         uint32_t crc = -1L;
 163         unsigned char *p = (unsigned char *)buffer;
 164         /* pre - calculated table for faster crc calculation */
 165         static uint32_t crc_table[256];
 166         static boolean_t first = TRUE;
 167         int i, j;
 168
 169         /* if we need to initialize the lookup table */
 170         if (first) {
 171                 /* calc the CRC table */
 172                 for (i = 0; i <= 255; i++) {
 173                         crc = i;
 174                         for (j = 8; j > 0; j--)
 175                                 if (crc & 1)
 176                                         crc = (crc >> 1) ^ CRC32_POLYNOMIAL;
 177                                 else
 178                                         crc >>= 1;
 179                         crc_table[i] = crc;
 180                 }
 181                 first = FALSE;
 182         }
 183
 184         crc = -1L;
 185         /* do the calculation */
 186         while (count-- != 0) {
 187                 temp1 = (crc >> 8) & 0x00FFFFFFL;
 188                 temp2 = crc_table[((int)crc ^ *p++) & 0xFF];
 189                 crc = temp1 ^ temp2;
 190         }
 191         return crc;
 192 }
 193
 194 /********************************************************************
 195  ********************************************************************/
 196
 197 /* The key is created in the following manner:
 198    port_num  lid   crc
 199    \______/ \___/ \___/
 200      16b     16b   32b
 201 */
 202 static void
 203 __osm_trap_get_key(IN uint16_t lid,
 204                    IN uint8_t port_num,
 205                    IN ib_mad_notice_attr_t * p_ntci, OUT uint64_t * trap_key)
 206 {
 207         uint32_t crc = 0;
 208
 209         CL_ASSERT(trap_key);
 210
 211         crc = __osm_trap_calc_crc32(p_ntci, sizeof(ib_mad_notice_attr_t));
 212         *trap_key = ((uint64_t) port_num << 48) | ((uint64_t) lid << 32) | crc;
 213 }
 214
 215 /**********************************************************************
 216  **********************************************************************/
 217 static int __print_num_received(IN uint32_t num_received)
 218 {
 219         uint32_t i;
 220
 221         /* Series is 10, 20, 50, 100, 200, 500, ... */
 222         i = num_received;
 223         while (i >= 10) {
 224                 if (i % 10)
 225                         break;
 226                 i = i / 10;
 227         }
 228
 229         if (i == 1 || i == 2 || i == 5)
 230                 return 1;
 231         else
 232                 return 0;
 233 }
 234
 235 static int disable_port(osm_sm_t *sm, osm_physp_t *p)
 236 {
 237         uint8_t payload[IB_SMP_DATA_SIZE];
 238         osm_madw_context_t context;
 239         ib_port_info_t *pi = (ib_port_info_t *)payload;
 240         int ret;
 241
 242         /* select the nearest port to master opensm */
 243         if (p->p_remote_physp &&
 244             p->dr_path.hop_count > p->p_remote_physp->dr_path.hop_count)
 245                 p = p->p_remote_physp;
 246
 247         /* If trap 131, might want to disable peer port if available */
 248         /* but peer port has been observed not to respond to SM requests */
 249
 250         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3810: "
 251                 "Disabling physical port 0x%016" PRIx64 " num:%u\n",
 252                 cl_ntoh64(osm_physp_get_port_guid(p)), p->port_num);
 253
 254         memcpy(payload, &p->port_info, sizeof(ib_port_info_t));
 255
 256         /* Set port to disabled/down */
 257         ib_port_info_set_port_state(pi, IB_LINK_DOWN);
 258         ib_port_info_set_port_phys_state(IB_PORT_PHYS_STATE_DISABLED, pi);
 259
 260         /* Issue set of PortInfo */
 261         context.pi_context.node_guid = osm_node_get_node_guid(p->p_node);
 262         context.pi_context.port_guid = osm_physp_get_port_guid(p);
 263         context.pi_context.set_method = TRUE;
 264         context.pi_context.light_sweep = FALSE;
 265         context.pi_context.active_transition = FALSE;
 266
 267         ret = osm_req_set(sm, osm_physp_get_dr_path_ptr(p),
 268                           payload, sizeof(payload), IB_MAD_ATTR_PORT_INFO,
 269                           cl_hton32(osm_physp_get_port_num(p)),
 270                           CL_DISP_MSGID_NONE, &context);
 271         if (ret)
 272                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3811: "
 273                         "Request to set PortInfo failed\n");
 274
 275         return ret;
 276 }
 277
 278 /**********************************************************************
 279  **********************************************************************/
 280 static void
 281 __osm_trap_rcv_process_request(IN osm_sm_t * sm,
 282                                IN const osm_madw_t * const p_madw)
 283 {
 284         uint8_t payload[sizeof(ib_mad_notice_attr_t)];
 285         ib_smp_t *p_smp;
 286         ib_mad_notice_attr_t *p_ntci = (ib_mad_notice_attr_t *) payload;
 287         ib_api_status_t status;
 288         osm_madw_t tmp_madw;    /* we need a copy to last after repress */
 289         uint64_t trap_key;
 290         uint32_t num_received;
 291         osm_physp_t *p_physp;
 292         cl_ptr_vector_t *p_tbl;
 293         osm_port_t *p_port;
 294         ib_net16_t source_lid = 0;
 295         boolean_t is_gsi = TRUE;
 296         uint8_t port_num = 0;
 297         boolean_t physp_change_trap = FALSE;
 298         uint64_t event_wheel_timeout = OSM_DEFAULT_TRAP_SUPRESSION_TIMEOUT;
 299         boolean_t run_heavy_sweep = FALSE;
 300
 301         OSM_LOG_ENTER(sm->p_log);
 302
 303         CL_ASSERT(p_madw);
 304
 305         if (osm_exit_flag)
 306                 /*
 307                    We got an exit flag - do nothing
 308                    Otherwise we start a sweep on the trap 144 caused by
 309                    cleaning up SM Cap bit...
 310                  */
 311                 goto Exit;
 312
 313         /* update the is_gsi flag according to the mgmt_class field */
 314         if (p_madw->p_mad->mgmt_class == IB_MCLASS_SUBN_LID ||
 315             p_madw->p_mad->mgmt_class == IB_MCLASS_SUBN_DIR)
 316                 is_gsi = FALSE;
 317
 318         /* No real need to grab the lock for this function. */
 319         memset(payload, 0, sizeof(payload));
 320         memset(&tmp_madw, 0, sizeof(tmp_madw));
 321
 322         p_smp = osm_madw_get_smp_ptr(p_madw);
 323
 324         if (p_smp->method != IB_MAD_METHOD_TRAP) {
 325                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3801: "
 326                         "Unsupported method 0x%X\n", p_smp->method);
 327                 goto Exit;
 328         }
 329
 330         /*
 331          * The NOTICE Attribute is part of the SMP CLASS attributes
 332          * As such the actual attribute data resides inside the SMP
 333          * payload.
 334          */
 335
 336         memcpy(payload, &(p_smp->data), IB_SMP_DATA_SIZE);
 337         memcpy(&tmp_madw, p_madw, sizeof(tmp_madw));
 338
 339         if (is_gsi == FALSE) {
 340                 /* We are in smi flow */
 341                 /*
 342                  * When we received a TRAP with dlid = 0 - it means it
 343                  * came from our own node. So we need to fix it.
 344                  */
 345
 346                 if (p_madw->mad_addr.addr_type.smi.source_lid == 0) {
 347                         /* Check if the sm_base_lid is 0. If yes - this means
 348                            that the local lid wasn't configured yet. Don't send
 349                            a response to the trap. */
 350                         if (sm->p_subn->sm_base_lid == 0) {
 351                                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 352                                         "Received SLID=0 Trap with local LID=0. Ignoring MAD\n");
 353                                 goto Exit;
 354                         }
 355                         OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 356                                 "Received SLID=0 Trap. Using local LID:%u instead\n",
 357                                 cl_ntoh16(sm->p_subn->sm_base_lid));
 358                         tmp_madw.mad_addr.addr_type.smi.source_lid =
 359                             sm->p_subn->sm_base_lid;
 360                 }
 361
 362                 source_lid = tmp_madw.mad_addr.addr_type.smi.source_lid;
 363
 364                 /* Print some info about the incoming Trap */
 365                 if (ib_notice_is_generic(p_ntci)) {
 366                         if ((p_ntci->g_or_v.generic.trap_num == CL_HTON16(129))
 367                             || (p_ntci->g_or_v.generic.trap_num ==
 368                                 CL_HTON16(130))
 369                             || (p_ntci->g_or_v.generic.trap_num ==
 370                                 CL_HTON16(131)))
 371                                 OSM_LOG(sm->p_log, OSM_LOG_ERROR,
 372                                         "Received Generic Notice type:%u "
 373                                         "num:%u (%s) Producer:%u (%s) "
 374                                         "from LID:%u Port %d TID:0x%016"
 375                                         PRIx64 "\n", ib_notice_get_type(p_ntci),
 376                                         cl_ntoh16(p_ntci->g_or_v.generic.
 377                                                   trap_num),
 378                                         ib_get_trap_str(p_ntci->g_or_v.generic.
 379                                                         trap_num),
 380                                         cl_ntoh32(ib_notice_get_prod_type
 381                                                   (p_ntci)),
 382                                         ib_get_producer_type_str
 383                                         (ib_notice_get_prod_type(p_ntci)),
 384                                         cl_hton16(source_lid),
 385                                         p_ntci->data_details.ntc_129_131.
 386                                         port_num, cl_ntoh64(p_smp->trans_id));
 387                         else
 388                                 OSM_LOG(sm->p_log, OSM_LOG_ERROR,
 389                                         "Received Generic Notice type:%u "
 390                                         "num:%u (%s) Producer:%u (%s) "
 391                                         "from LID:%u TID:0x%016" PRIx64
 392                                         "\n", ib_notice_get_type(p_ntci),
 393                                         cl_ntoh16(p_ntci->g_or_v.generic.
 394                                                   trap_num),
 395                                         ib_get_trap_str(p_ntci->g_or_v.generic.
 396                                                         trap_num),
 397                                         cl_ntoh32(ib_notice_get_prod_type
 398                                                   (p_ntci)),
 399                                         ib_get_producer_type_str
 400                                         (ib_notice_get_prod_type(p_ntci)),
 401                                         cl_hton16(source_lid),
 402                                         cl_ntoh64(p_smp->trans_id));
 403                 } else
 404                         OSM_LOG(sm->p_log, OSM_LOG_ERROR,
 405                                 "Received Vendor Notice type:%u vend:0x%06X "
 406                                 "dev:%u from LID:%u TID:0x%016" PRIx64 "\n",
 407                                 ib_notice_get_type(p_ntci),
 408                                 cl_ntoh32(ib_notice_get_vend_id(p_ntci)),
 409                                 cl_ntoh16(p_ntci->g_or_v.vend.dev_id),
 410                                 cl_ntoh16(source_lid),
 411                                 cl_ntoh64(p_smp->trans_id));
 412         }
 413
 414         osm_dump_notice(sm->p_log, p_ntci, OSM_LOG_VERBOSE);
 415
 416         p_physp = osm_get_physp_by_mad_addr(sm->p_log,
 417                                             sm->p_subn, &tmp_madw.mad_addr);
 418         if (p_physp)
 419                 p_smp->m_key = p_physp->port_info.m_key;
 420         else
 421                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3809: "
 422                         "Failed to find source physical port for trap\n");
 423
 424         status = osm_resp_send(sm, &tmp_madw, 0, payload);
 425         if (status != IB_SUCCESS) {
 426                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3802: "
 427                         "Error sending response (%s)\n",
 428                         ib_get_err_str(status));
 429                 goto Exit;
 430         }
 431
 432         /*
 433          * We would like to filter out recurring Traps so we track them by
 434          * their source lid and content. If the same trap was already
 435          * received within the aging time window more than 10 times,
 436          * we simply ignore it. This is done only if we are in smi mode
 437          */
 438
 439         if (is_gsi == FALSE) {
 440                 if (ib_notice_is_generic(p_ntci) &&
 441                     ((p_ntci->g_or_v.generic.trap_num == CL_HTON16(129)) ||
 442                      (p_ntci->g_or_v.generic.trap_num == CL_HTON16(130)) ||
 443                      (p_ntci->g_or_v.generic.trap_num == CL_HTON16(131)))) {
 444                         /* If this is a trap 129, 130, or 131 - then this is a
 445                          * trap signaling a change on a physical port.
 446                          * Mark the physp_change_trap flag as TRUE.
 447                          */
 448                         physp_change_trap = TRUE;
 449                         /* The source_lid should be based on the source_lid from the trap */
 450                         source_lid = p_ntci->data_details.ntc_129_131.lid;
 451                 }
 452
 453                 /* If physp_change_trap is TRUE - the key will include the port number.
 454                    If not - the port_number in the key will be zero. */
 455                 if (physp_change_trap == TRUE) {
 456                         port_num = p_ntci->data_details.ntc_129_131.port_num;
 457                         __osm_trap_get_key(source_lid, port_num, p_ntci,
 458                                            &trap_key);
 459                 } else
 460                         __osm_trap_get_key(source_lid, 0, p_ntci, &trap_key);
 461
 462                 /* try to find it in the aging tracker */
 463                 num_received =
 464                     cl_event_wheel_num_regs(&sm->trap_aging_tracker,
 465                                             trap_key);
 466
 467                 /* Now we know how many times it provided this trap */
 468                 if (num_received > 10) {
 469                         if (__print_num_received(num_received))
 470                                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3804: "
 471                                         "Received trap %u times consecutively\n",
 472                                         num_received);
 473                         /*
 474                          * If the trap provides info about a bad port
 475                          * we mark it as unhealthy.
 476                          */
 477                         if (physp_change_trap == TRUE) {
 478                                 /* get the port */
 479                                 p_physp = get_physp_by_lid_and_num(sm,
 480                                                                    cl_ntoh16
 481                                                                    (p_ntci->
 482                                                                     data_details.
 483                                                                     ntc_129_131.
 484                                                                     lid),
 485                                                                    port_num);
 486
 487                                 if (!p_physp)
 488                                         OSM_LOG(sm->p_log, OSM_LOG_ERROR,
 489                                                 "ERR 3805: "
 490                                                 "Failed to find physical port by lid:%u num:%u\n",
 491                                                 cl_ntoh16(p_ntci->data_details.
 492                                                           ntc_129_131.lid),
 493                                                 p_ntci->data_details.
 494                                                 ntc_129_131.port_num);
 495                                 else {
 496                                         /* When babbling port policy option is enabled and
 497                                            Threshold for disabling a "babbling" port is exceeded */
 498                                         if (sm->p_subn->opt.
 499                                             babbling_port_policy
 500                                             && num_received >= 250
 501                                             && disable_port(sm, p_physp) == 0)
 502                                                 goto Exit;
 503
 504                                         OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 505                                                 "Marking unhealthy physical port by lid:%u num:%u\n",
 506                                                 cl_ntoh16(p_ntci->data_details.
 507                                                           ntc_129_131.lid),
 508                                                 p_ntci->data_details.
 509                                                 ntc_129_131.port_num);
 510                                         /* check if the current state of the p_physp is healthy. If
 511                                            it is - then this is a first change of state. Run a heavy sweep.
 512                                            if it is not - no need to mark it again - just restart the timer. */
 513                                         if (osm_physp_is_healthy(p_physp)) {
 514                                                 osm_physp_set_health(p_physp,
 515                                                                      FALSE);
 516                                                 /* Make sure we sweep again - force a heavy sweep. */
 517                                                 /* The sweep should be done only after the re-registration, or
 518                                                    else we'll be losing track of the timer. */
 519                                                 run_heavy_sweep = TRUE;
 520                                         }
 521                                         /* If we are marking the port as unhealthy - we want to
 522                                            keep this for a longer period of time than the
 523                                            OSM_DEFAULT_TRAP_SUPRESSION_TIMEOUT. Use the
 524                                            OSM_DEFAULT_UNHEALTHY_TIMEOUT */
 525                                         event_wheel_timeout =
 526                                             OSM_DEFAULT_UNHEALTHY_TIMEOUT;
 527                                 }
 528                         }
 529                 }
 530
 531                 /* restart the aging anyway */
 532                 /* If physp_change_trap is TRUE - then use a callback to unset the
 533                    healthy bit. If not - no need to use a callback. */
 534                 if (physp_change_trap == TRUE)
 535                         cl_event_wheel_reg(&sm->trap_aging_tracker, trap_key, cl_get_time_stamp() + event_wheel_timeout, osm_trap_rcv_aging_tracker_callback,   /* no callback */
 536                                            sm   /* no context */ );
 537                 else
 538                         cl_event_wheel_reg(&sm->trap_aging_tracker, trap_key, cl_get_time_stamp() + event_wheel_timeout, NULL,  /* no callback */
 539                                            NULL /* no context */ );
 540
 541                 /* If was already registered do nothing more */
 542                 if (num_received > 10 && run_heavy_sweep == FALSE) {
 543                         if (__print_num_received(num_received))
 544                                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 545                                         "Continuously received this trap %u times. Ignoring\n",
 546                                         num_received);
 547                         goto Exit;
 548                 }
 549         }
 550
 551         /* Check for node description update. IB Spec v1.2.1 pg 823 */
 552         if ((p_ntci->data_details.ntc_144.local_changes & TRAP_144_MASK_OTHER_LOCAL_CHANGES) &&
 553                 (p_ntci->data_details.ntc_144.change_flgs & TRAP_144_MASK_NODE_DESCRIPTION_CHANGE)
 554                 ) {
 555                 OSM_LOG(sm->p_log, OSM_LOG_INFO, "Trap 144 Node description update\n");
 556
 557                 if (p_physp) {
 558                         CL_PLOCK_ACQUIRE(sm->p_lock);
 559                         osm_req_get_node_desc(sm, p_physp);
 560                         CL_PLOCK_RELEASE(sm->p_lock);
 561                 } else {
 562                         OSM_LOG(sm->p_log, OSM_LOG_ERROR,
 563                                 "ERR 3812: No physical port found for "
 564                                 "trap 144: \"node description update\"\n");
 565                 }
 566         }
 567
 568         /* do a sweep if we received a trap */
 569         if (sm->p_subn->opt.sweep_on_trap) {
 570                 /* if this is trap number 128 or run_heavy_sweep is TRUE - update the
 571                    force_single_heavy_sweep flag of the subnet.
 572                    Sweep also on traps 144/145 - these traps signal a change of a certain
 573                    port capability/system image guid.
 574                    TODO: In the future we can change this to just getting PortInfo on
 575                    this port instead of sweeping the entire subnet. */
 576                 if (ib_notice_is_generic(p_ntci) &&
 577                     ((cl_ntoh16(p_ntci->g_or_v.generic.trap_num) == 128) ||
 578                      (cl_ntoh16(p_ntci->g_or_v.generic.trap_num) == 144) ||
 579                      (cl_ntoh16(p_ntci->g_or_v.generic.trap_num) == 145) ||
 580                      run_heavy_sweep)) {
 581                         OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 582                                 "Forcing heavy sweep. Received trap:%u\n",
 583                                 cl_ntoh16(p_ntci->g_or_v.generic.trap_num));
 584
 585                         sm->p_subn->force_heavy_sweep = TRUE;
 586                 }
 587                 osm_sm_signal(sm, OSM_SIGNAL_SWEEP);
 588         }
 589
 590         /* If we reached here due to trap 129/130/131 - do not need to do
 591            the notice report. Just goto exit. We know this is the case
 592            if physp_change_trap is TRUE. */
 593         if (physp_change_trap == TRUE)
 594                 goto Exit;
 595
 596         /* Add a call to osm_report_notice */
 597         /* We are going to report the notice - so need to fix the IssuerGID
 598            accordingly. See IBA 1.2 p.739 or IBA 1.1 p.653 for details. */
 599         if (is_gsi) {
 600                 if (!tmp_madw.mad_addr.addr_type.gsi.global_route) {
 601                         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3806: "
 602                                 "Received gsi trap with global_route FALSE. "
 603                                 "Cannot update issuer_gid!\n");
 604                         goto Exit;
 605                 }
 606                 memcpy(&(p_ntci->issuer_gid),
 607                        &(tmp_madw.mad_addr.addr_type.gsi.grh_info.src_gid),
 608                        sizeof(ib_gid_t));
 609         } else {
 610                 /* Need to use the IssuerLID */
 611                 p_tbl = &sm->p_subn->port_lid_tbl;
 612
 613                 CL_ASSERT(cl_ptr_vector_get_size(p_tbl) < 0x10000);
 614
 615                 if ((uint16_t) cl_ptr_vector_get_size(p_tbl) <=
 616                     cl_ntoh16(source_lid)) {
 617                         /*  the source lid is out of range */
 618                         OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 619                                 "source lid is out of range:%u\n",
 620                                 cl_ntoh16(source_lid));
 621
 622                         goto Exit;
 623                 }
 624                 p_port = cl_ptr_vector_get(p_tbl, cl_ntoh16(source_lid));
 625                 if (p_port == 0) {
 626                         /* We have the lid - but no corresponding port */
 627                         OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 628                                 "Cannot find port corresponding to lid:%u\n",
 629                                 cl_ntoh16(source_lid));
 630
 631                         goto Exit;
 632                 }
 633
 634                 p_ntci->issuer_gid.unicast.prefix =
 635                     sm->p_subn->opt.subnet_prefix;
 636                 p_ntci->issuer_gid.unicast.interface_id = p_port->guid;
 637         }
 638
 639         /* we need a lock here as the InformInfo DB must be stable */
 640         CL_PLOCK_ACQUIRE(sm->p_lock);
 641         status = osm_report_notice(sm->p_log, sm->p_subn, p_ntci);
 642         CL_PLOCK_RELEASE(sm->p_lock);
 643         if (status != IB_SUCCESS) {
 644                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3803: "
 645                         "Error sending trap reports (%s)\n",
 646                         ib_get_err_str(status));
 647                 goto Exit;
 648         }
 649
 650 Exit:
 651         OSM_LOG_EXIT(sm->p_log);
 652 }
 653
 654 #if 0
 655 /**********************************************************************
 656  CURRENTLY WE ARE NOT CREATING TRAPS - SO THIS CALL IS AN ERROR
 657 **********************************************************************/
 658 static void
 659 __osm_trap_rcv_process_sm(IN osm_sm_t * sm,
 660                           IN const osm_remote_sm_t * const p_sm)
 661 {
 662         /* const ib_sm_info_t*        p_smi; */
 663
 664         OSM_LOG_ENTER(sm->p_log);
 665
 666         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3807: "
 667                 "This function is not supported yet\n");
 668
 669         OSM_LOG_EXIT(sm->p_log);
 670 }
 671 #endif
 672
 673 /**********************************************************************
 674  CURRENTLY WE ARE NOT CREATING TRAPS - SO THIS CALL IN AN ERROR
 675 **********************************************************************/
 676 static void
 677 __osm_trap_rcv_process_response(IN osm_sm_t * sm,
 678                                 IN const osm_madw_t * const p_madw)
 679 {
 680
 681         OSM_LOG_ENTER(sm->p_log);
 682
 683         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3808: "
 684                 "This function is not supported yet\n");
 685
 686         OSM_LOG_EXIT(sm->p_log);
 687 }
 688
 689 /**********************************************************************
 690  **********************************************************************/
 691 void osm_trap_rcv_process(IN void *context, IN void *data)
 692 {
 693         osm_sm_t *sm = context;
 694         osm_madw_t *p_madw = data;
 695         ib_smp_t *p_smp;
 696
 697         OSM_LOG_ENTER(sm->p_log);
 698
 699         CL_ASSERT(p_madw);
 700
 701         p_smp = osm_madw_get_smp_ptr(p_madw);
 702
 703         /*
 704            Determine if this is a request for our own Trap
 705            or if this is a response to our request for another
 706            SM's Trap.
 707          */
 708         if (ib_smp_is_response(p_smp))
 709                 __osm_trap_rcv_process_response(sm, p_madw);
 710         else
 711                 __osm_trap_rcv_process_request(sm, p_madw);
 712
 713         OSM_LOG_EXIT(sm->p_log);
 714 }