contrib/ofed/management/opensm/opensm/osm_sw_info_rcv.c

   1 /*
   2  * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
   3  * Copyright (c) 2002-2005 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_si_rcv_t.
  39  * This object represents the SwitchInfo 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_passivelock.h>
  51 #include <complib/cl_debug.h>
  52 #include <opensm/osm_log.h>
  53 #include <opensm/osm_switch.h>
  54 #include <opensm/osm_subnet.h>
  55 #include <opensm/osm_helper.h>
  56 #include <opensm/osm_opensm.h>
  57
  58 /**********************************************************************
  59  The plock must be held before calling this function.
  60 **********************************************************************/
  61 static void
  62 __osm_si_rcv_get_port_info(IN osm_sm_t * sm, IN osm_switch_t * const p_sw)
  63 {
  64         osm_madw_context_t context;
  65         uint8_t port_num;
  66         osm_physp_t *p_physp;
  67         osm_node_t *p_node;
  68         uint8_t num_ports;
  69         ib_api_status_t status = IB_SUCCESS;
  70
  71         OSM_LOG_ENTER(sm->p_log);
  72
  73         CL_ASSERT(p_sw);
  74
  75         p_node = p_sw->p_node;
  76
  77         CL_ASSERT(osm_node_get_type(p_node) == IB_NODE_TYPE_SWITCH);
  78
  79         /*
  80            Request PortInfo attribute for each port on the switch.
  81          */
  82         p_physp = osm_node_get_physp_ptr(p_node, 0);
  83
  84         context.pi_context.node_guid = osm_node_get_node_guid(p_node);
  85         context.pi_context.port_guid = osm_physp_get_port_guid(p_physp);
  86         context.pi_context.set_method = FALSE;
  87         context.pi_context.light_sweep = FALSE;
  88         context.pi_context.active_transition = FALSE;
  89
  90         num_ports = osm_node_get_num_physp(p_node);
  91
  92         for (port_num = 0; port_num < num_ports; port_num++) {
  93                 status = osm_req_get(sm, osm_physp_get_dr_path_ptr(p_physp),
  94                                      IB_MAD_ATTR_PORT_INFO, cl_hton32(port_num),
  95                                      CL_DISP_MSGID_NONE, &context);
  96                 if (status != IB_SUCCESS)
  97                         /* continue the loop despite the error */
  98                         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3602: "
  99                                 "Failure initiating PortInfo request (%s)\n",
 100                                 ib_get_err_str(status));
 101         }
 102
 103         OSM_LOG_EXIT(sm->p_log);
 104 }
 105
 106 #if 0
 107 /**********************************************************************
 108  The plock must be held before calling this function.
 109 **********************************************************************/
 110 static void
 111 __osm_si_rcv_get_fwd_tbl(IN osm_sm_t * sm, IN osm_switch_t * const p_sw)
 112 {
 113         osm_madw_context_t context;
 114         osm_dr_path_t *p_dr_path;
 115         osm_physp_t *p_physp;
 116         osm_node_t *p_node;
 117         uint32_t block_id_ho;
 118         uint32_t max_block_id_ho;
 119         ib_api_status_t status = IB_SUCCESS;
 120
 121         OSM_LOG_ENTER(sm->p_log);
 122
 123         CL_ASSERT(p_sw);
 124
 125         p_node = p_sw->p_node;
 126
 127         CL_ASSERT(osm_node_get_type(p_node) == IB_NODE_TYPE_SWITCH);
 128
 129         context.lft_context.node_guid = osm_node_get_node_guid(p_node);
 130         context.lft_context.set_method = FALSE;
 131
 132         max_block_id_ho = osm_switch_get_max_block_id_in_use(p_sw);
 133
 134         p_physp = osm_node_get_physp_ptr(p_node, 0);
 135         p_dr_path = osm_physp_get_dr_path_ptr(p_physp);
 136
 137         for (block_id_ho = 0; block_id_ho <= max_block_id_ho; block_id_ho++) {
 138                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 139                         "Retrieving FT block %u\n", block_id_ho);
 140
 141                 status = osm_req_get(sm, p_dr_path, IB_MAD_ATTR_LIN_FWD_TBL,
 142                                      cl_hton32(block_id_ho),
 143                                      CL_DISP_MSGID_NONE, &context);
 144                 if (status != IB_SUCCESS)
 145                         /* continue the loop despite the error */
 146                         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3603: "
 147                                 "Failure initiating PortInfo request (%s)\n",
 148                                 ib_get_err_str(status));
 149         }
 150
 151         OSM_LOG_EXIT(sm->p_log);
 152 }
 153
 154 /**********************************************************************
 155  The plock must be held before calling this function.
 156 **********************************************************************/
 157 static void
 158 __osm_si_rcv_get_mcast_fwd_tbl(IN osm_sm_t * sm, IN osm_switch_t * const p_sw)
 159 {
 160         osm_madw_context_t context;
 161         osm_dr_path_t *p_dr_path;
 162         osm_physp_t *p_physp;
 163         osm_node_t *p_node;
 164         osm_mcast_tbl_t *p_tbl;
 165         uint32_t block_id_ho;
 166         uint32_t max_block_id_ho;
 167         uint32_t position;
 168         uint32_t max_position;
 169         uint32_t attr_mod_ho;
 170         ib_api_status_t status = IB_SUCCESS;
 171
 172         OSM_LOG_ENTER(sm->p_log);
 173
 174         CL_ASSERT(p_sw);
 175
 176         p_node = p_sw->p_node;
 177
 178         CL_ASSERT(osm_node_get_type(p_node) == IB_NODE_TYPE_SWITCH);
 179
 180         if (osm_switch_get_mcast_fwd_tbl_size(p_sw) == 0) {
 181                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 182                         "Multicast not supported by switch 0x%016" PRIx64 "\n",
 183                         cl_ntoh64(osm_node_get_node_guid(p_node)));
 184                 goto Exit;
 185         }
 186
 187         context.mft_context.node_guid = osm_node_get_node_guid(p_node);
 188         context.mft_context.set_method = FALSE;
 189
 190         p_tbl = osm_switch_get_mcast_tbl_ptr(p_sw);
 191         max_block_id_ho = osm_mcast_tbl_get_max_block(p_tbl);
 192
 193         if (max_block_id_ho > IB_MCAST_MAX_BLOCK_ID) {
 194                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3609: "
 195                         "Out-of-range mcast block size = %u on switch 0x%016"
 196                         PRIx64 "\n", max_block_id_ho,
 197                         cl_ntoh64(osm_node_get_node_guid(p_node)));
 198                 goto Exit;
 199         }
 200
 201         max_position = osm_mcast_tbl_get_max_position(p_tbl);
 202
 203         CL_ASSERT(max_position <= IB_MCAST_POSITION_MAX);
 204
 205         OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 206                 "Max MFT block = %u, Max position = %u\n", max_block_id_ho,
 207                 max_position);
 208
 209         p_physp = osm_node_get_physp_ptr(p_node, 0);
 210         p_dr_path = osm_physp_get_dr_path_ptr(p_physp);
 211
 212         for (block_id_ho = 0; block_id_ho <= max_block_id_ho; block_id_ho++) {
 213                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 214                         "Retrieving MFT block %u\n", block_id_ho);
 215
 216                 for (position = 0; position <= max_position; position++) {
 217                         OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 218                                 "Retrieving MFT position %u\n", position);
 219
 220                         attr_mod_ho =
 221                             block_id_ho | position << IB_MCAST_POSITION_SHIFT;
 222                         status =
 223                             osm_req_get(sm, p_dr_path,
 224                                         IB_MAD_ATTR_MCAST_FWD_TBL,
 225                                         cl_hton32(attr_mod_ho),
 226                                         CL_DISP_MSGID_NONE, &context);
 227                         if (status != IB_SUCCESS)
 228                                 /* continue the loop despite the error */
 229                                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3607: "
 230                                         "Failure initiating PortInfo request (%s)\n",
 231                                         ib_get_err_str(status));
 232                 }
 233         }
 234
 235 Exit:
 236         OSM_LOG_EXIT(sm->p_log);
 237 }
 238 #endif
 239
 240 /**********************************************************************
 241    Lock must be held on entry to this function.
 242 **********************************************************************/
 243 static void
 244 __osm_si_rcv_process_new(IN osm_sm_t * sm,
 245                          IN osm_node_t * const p_node,
 246                          IN const osm_madw_t * const p_madw)
 247 {
 248         osm_switch_t *p_sw;
 249         osm_switch_t *p_check;
 250         ib_switch_info_t *p_si;
 251         ib_smp_t *p_smp;
 252         cl_qmap_t *p_sw_guid_tbl;
 253
 254         CL_ASSERT(sm);
 255
 256         OSM_LOG_ENTER(sm->p_log);
 257
 258         CL_ASSERT(p_madw);
 259
 260         p_sw_guid_tbl = &sm->p_subn->sw_guid_tbl;
 261
 262         p_smp = osm_madw_get_smp_ptr(p_madw);
 263         p_si = (ib_switch_info_t *) ib_smp_get_payload_ptr(p_smp);
 264
 265         osm_dump_switch_info(sm->p_log, p_si, OSM_LOG_DEBUG);
 266
 267         /*
 268            Allocate a new switch object for this switch,
 269            and place it in the switch table.
 270          */
 271         p_sw = osm_switch_new(p_node, p_madw);
 272         if (p_sw == NULL) {
 273                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3608: "
 274                         "Unable to allocate new switch object\n");
 275                 goto Exit;
 276         }
 277
 278         /* set subnet max mlid to the minimum MulticastFDBCap of all switches */
 279         if (p_sw->mcast_tbl.max_mlid_ho < sm->p_subn->max_mcast_lid_ho) {
 280                 sm->p_subn->max_mcast_lid_ho = p_sw->mcast_tbl.max_mlid_ho;
 281                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 282                         "Subnet max multicast lid is 0x%X\n",
 283                         sm->p_subn->max_mcast_lid_ho);
 284         }
 285
 286         /* set subnet max unicast lid to the minimum LinearFDBCap of all switches */
 287         if (cl_ntoh16(p_si->lin_cap) < sm->p_subn->max_ucast_lid_ho) {
 288                 sm->p_subn->max_ucast_lid_ho = cl_ntoh16(p_si->lin_cap);
 289                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 290                         "Subnet max unicast lid is 0x%X\n",
 291                         sm->p_subn->max_ucast_lid_ho);
 292         }
 293
 294         p_check = (osm_switch_t *) cl_qmap_insert(p_sw_guid_tbl,
 295                                                   osm_node_get_node_guid
 296                                                   (p_node), &p_sw->map_item);
 297
 298         if (p_check != p_sw) {
 299                 /*
 300                    This shouldn't happen since we hold the lock!
 301                  */
 302                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3605: "
 303                         "Unable to add new switch object to database\n");
 304                 osm_switch_delete(&p_sw);
 305                 goto Exit;
 306         }
 307
 308         p_node->sw = p_sw;
 309
 310         /*
 311            Update the switch info according to the
 312            info we just received.
 313          */
 314         osm_switch_set_switch_info(p_sw, p_si);
 315         p_sw->discovery_count++;
 316
 317         /*
 318            Get the PortInfo attribute for every port.
 319          */
 320         __osm_si_rcv_get_port_info(sm, p_sw);
 321
 322         /*
 323            Don't bother retrieving the current unicast and multicast tables
 324            from the switches.  The current version of SM does
 325            not support silent take-over of an existing multicast
 326            configuration.
 327
 328            Gathering the multicast tables can also generate large amounts
 329            of extra subnet-init traffic.
 330
 331            The code to retrieve the tables was fully debugged.
 332          */
 333 #if 0
 334         __osm_si_rcv_get_fwd_tbl(sm, p_sw);
 335         if (!sm->p_subn->opt.disable_multicast)
 336                 __osm_si_rcv_get_mcast_fwd_tbl(sm, p_sw);
 337 #endif
 338
 339 Exit:
 340         OSM_LOG_EXIT(sm->p_log);
 341 }
 342
 343 /**********************************************************************
 344    Lock must be held on entry to this function.
 345    Return 1 if the caller is expected to send a change_detected event.
 346    this can not be done internally as the event needs the lock...
 347 **********************************************************************/
 348 static boolean_t
 349 __osm_si_rcv_process_existing(IN osm_sm_t * sm,
 350                               IN osm_node_t * const p_node,
 351                               IN const osm_madw_t * const p_madw)
 352 {
 353         osm_switch_t *p_sw = p_node->sw;
 354         ib_switch_info_t *p_si;
 355         osm_si_context_t *p_si_context;
 356         ib_smp_t *p_smp;
 357         boolean_t is_change_detected = FALSE;
 358
 359         OSM_LOG_ENTER(sm->p_log);
 360
 361         CL_ASSERT(p_madw);
 362
 363         p_smp = osm_madw_get_smp_ptr(p_madw);
 364         p_si = (ib_switch_info_t *) ib_smp_get_payload_ptr(p_smp);
 365         p_si_context = osm_madw_get_si_context_ptr(p_madw);
 366
 367         if (p_si_context->set_method) {
 368                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 369                         "Received logical SetResp()\n");
 370
 371                 osm_switch_set_switch_info(p_sw, p_si);
 372         } else {
 373                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 374                         "Received logical GetResp()\n");
 375
 376                 osm_switch_set_switch_info(p_sw, p_si);
 377
 378                 /*
 379                    Check the port state change bit.  If true, then this switch
 380                    has seen a port state transition, so continue probing.
 381                  */
 382                 if (p_si_context->light_sweep == TRUE) {
 383                         /* This is a light sweep */
 384                         /* If the mad was returned with an error -
 385                            signal a change to the state manager. */
 386                         if (ib_smp_get_status(p_smp) != 0) {
 387                                 OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 388                                         "GetResp() received with error in light sweep. "
 389                                         "Commencing heavy sweep\n");
 390                                 is_change_detected = TRUE;
 391                         } else {
 392                                 /*
 393                                    If something changed, then just signal the
 394                                    state manager.  Don't attempt to probe
 395                                    further during a light sweep.
 396                                  */
 397                                 if (ib_switch_info_get_state_change(p_si)) {
 398                                         osm_dump_switch_info(sm->p_log, p_si,
 399                                                              OSM_LOG_DEBUG);
 400                                         is_change_detected = TRUE;
 401                                 }
 402                         }
 403                 } else {
 404                         /*
 405                            This is a heavy sweep.  Get information regardless
 406                            of the state change bit.
 407                          */
 408                         p_sw->discovery_count++;
 409                         OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
 410                                 "discovery_count is:%u\n",
 411                                 p_sw->discovery_count);
 412
 413                         /* If this is the first discovery - then get the port_info */
 414                         if (p_sw->discovery_count == 1)
 415                                 __osm_si_rcv_get_port_info(sm, p_sw);
 416                         else
 417                                 OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 418                                         "Not discovering again through switch:0x%"
 419                                         PRIx64 "\n",
 420                                         osm_node_get_node_guid(p_sw->p_node));
 421                 }
 422         }
 423
 424         OSM_LOG_EXIT(sm->p_log);
 425         return is_change_detected;
 426 }
 427
 428 /**********************************************************************
 429  **********************************************************************/
 430 void osm_si_rcv_process(IN void *context, IN void *data)
 431 {
 432         osm_sm_t *sm = context;
 433         osm_madw_t *p_madw = data;
 434         ib_switch_info_t *p_si;
 435         ib_smp_t *p_smp;
 436         osm_node_t *p_node;
 437         ib_net64_t node_guid;
 438         osm_si_context_t *p_context;
 439
 440         CL_ASSERT(sm);
 441
 442         OSM_LOG_ENTER(sm->p_log);
 443
 444         CL_ASSERT(p_madw);
 445
 446         p_smp = osm_madw_get_smp_ptr(p_madw);
 447         p_si = (ib_switch_info_t *) ib_smp_get_payload_ptr(p_smp);
 448
 449         /*
 450            Acquire the switch object and add the switch info.
 451          */
 452         p_context = osm_madw_get_si_context_ptr(p_madw);
 453
 454         node_guid = p_context->node_guid;
 455
 456         OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
 457                 "Switch GUID 0x%016" PRIx64 ", TID 0x%" PRIx64 "\n",
 458                 cl_ntoh64(node_guid), cl_ntoh64(p_smp->trans_id));
 459
 460         CL_PLOCK_EXCL_ACQUIRE(sm->p_lock);
 461
 462         p_node = osm_get_node_by_guid(sm->p_subn, node_guid);
 463         if (!p_node)
 464                 OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3606: "
 465                         "SwitchInfo received for nonexistent node "
 466                         "with GUID 0x%" PRIx64 "\n", cl_ntoh64(node_guid));
 467         else {
 468
 469                 /*
 470                    Hack for bad value in Mellanox switch
 471                  */
 472                 if (cl_ntoh16(p_si->lin_top) > IB_LID_UCAST_END_HO) {
 473                         OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3610: "
 474                                 "\n\t\t\t\tBad LinearFDBTop value = 0x%X "
 475                                 "on switch 0x%" PRIx64
 476                                 "\n\t\t\t\tForcing internal correction to 0x%X\n",
 477                                 cl_ntoh16(p_si->lin_top),
 478                                 cl_ntoh64(osm_node_get_node_guid(p_node)), 0);
 479
 480                         p_si->lin_top = 0;
 481                 }
 482
 483                 /*
 484                    Acquire the switch object for this switch.
 485                  */
 486                 if (!p_node->sw) {
 487                         __osm_si_rcv_process_new(sm, p_node, p_madw);
 488                         /*
 489                            A new switch was found during the sweep so we need
 490                            to ignore the current LFT settings.
 491                          */
 492                         sm->p_subn->ignore_existing_lfts = TRUE;
 493                 } else {
 494                         /* we might get back a request for signaling change was detected */
 495                         if (__osm_si_rcv_process_existing(sm, p_node, p_madw)) {
 496                                 CL_PLOCK_RELEASE(sm->p_lock);
 497                                 sm->p_subn->force_heavy_sweep = TRUE;
 498                                 goto Exit;
 499                         }
 500                 }
 501         }
 502
 503         CL_PLOCK_RELEASE(sm->p_lock);
 504 Exit:
 505         OSM_LOG_EXIT(sm->p_log);
 506 }