gpxe/src/net/infiniband/ib_cm.c

   1 /*
   2  * Copyright (C) 2009 Michael Brown <mbrown@fensystems.co.uk>.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License as
   6  * published by the Free Software Foundation; either version 2 of the
   7  * License, or any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  17  */
  18
  19 FILE_LICENCE ( GPL2_OR_LATER );
  20
  21 #include <stdint.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <byteswap.h>
  25 #include <errno.h>
  26 #include <assert.h>
  27 #include <gpxe/infiniband.h>
  28 #include <gpxe/ib_mi.h>
  29 #include <gpxe/ib_pathrec.h>
  30 #include <gpxe/ib_cm.h>
  31
  32 /**
  33  * @file
  34  *
  35  * Infiniband communication management
  36  *
  37  */
  38
  39 /** List of connections */
  40 static LIST_HEAD ( ib_cm_conns );
  41
  42 /**
  43  * Send "ready to use" response
  44  *
  45  * @v ibdev             Infiniband device
  46  * @v mi                Management interface
  47  * @v conn              Connection
  48  * @v av                Address vector
  49  * @ret rc              Return status code
  50  */
  51 static int ib_cm_send_rtu ( struct ib_device *ibdev,
  52                             struct ib_mad_interface *mi,
  53                             struct ib_connection *conn,
  54                             struct ib_address_vector *av ) {
  55         union ib_mad mad;
  56         struct ib_cm_ready_to_use *ready =
  57                 &mad.cm.cm_data.ready_to_use;
  58         int rc;
  59
  60         /* Construct "ready to use" response */
  61         memset ( &mad, 0, sizeof ( mad ) );
  62         mad.hdr.mgmt_class = IB_MGMT_CLASS_CM;
  63         mad.hdr.class_version = IB_CM_CLASS_VERSION;
  64         mad.hdr.method = IB_MGMT_METHOD_SEND;
  65         mad.hdr.attr_id = htons ( IB_CM_ATTR_READY_TO_USE );
  66         ready->local_id = htonl ( conn->local_id );
  67         ready->remote_id = htonl ( conn->remote_id );
  68         if ( ( rc = ib_mi_send ( ibdev, mi, &mad, av ) ) != 0 ){
  69                 DBGC ( conn, "CM %p could not send RTU: %s\n",
  70                        conn, strerror ( rc ) );
  71                 return rc;
  72         }
  73
  74         return 0;
  75 }
  76
  77 /**
  78  * Handle duplicate connection replies
  79  *
  80  * @v ibdev             Infiniband device
  81  * @v mi                Management interface
  82  * @v mad               Received MAD
  83  * @v av                Source address vector
  84  * @ret rc              Return status code
  85  *
  86  * If a "ready to use" MAD is lost, the peer may resend the connection
  87  * reply.  We have to respond to these with duplicate "ready to use"
  88  * MADs, otherwise the peer may time out and drop the connection.
  89  */
  90 static void ib_cm_connect_rep ( struct ib_device *ibdev,
  91                                 struct ib_mad_interface *mi,
  92                                 union ib_mad *mad,
  93                                 struct ib_address_vector *av ) {
  94         struct ib_cm_connect_reply *connect_rep =
  95                 &mad->cm.cm_data.connect_reply;
  96         struct ib_connection *conn;
  97         int rc;
  98
  99         /* Identify connection */
 100         list_for_each_entry ( conn, &ib_cm_conns, list ) {
 101                 if ( ntohl ( connect_rep->remote_id ) != conn->local_id )
 102                         continue;
 103                 /* Try to send "ready to use" reply */
 104                 if ( ( rc = ib_cm_send_rtu ( ibdev, mi, conn, av ) ) != 0 ) {
 105                         /* Ignore errors */
 106                         return;
 107                 }
 108                 return;
 109         }
 110
 111         DBG ( "CM unidentified connection %08x\n",
 112               ntohl ( connect_rep->remote_id ) );
 113 }
 114
 115 /** Communication management agents */
 116 struct ib_mad_agent ib_cm_agent[] __ib_mad_agent = {
 117         {
 118                 .mgmt_class = IB_MGMT_CLASS_CM,
 119                 .class_version = IB_CM_CLASS_VERSION,
 120                 .attr_id = htons ( IB_CM_ATTR_CONNECT_REPLY ),
 121                 .handle = ib_cm_connect_rep,
 122         },
 123 };
 124
 125 /**
 126  * Convert connection rejection reason to return status code
 127  *
 128  * @v reason            Rejection reason (in network byte order)
 129  * @ret rc              Return status code
 130  */
 131 static int ib_cm_rejection_reason_to_rc ( uint16_t reason ) {
 132         switch ( reason ) {
 133         case htons ( IB_CM_REJECT_BAD_SERVICE_ID ) :
 134                 return -ENODEV;
 135         case htons ( IB_CM_REJECT_STALE_CONN ) :
 136                 return -EALREADY;
 137         case htons ( IB_CM_REJECT_CONSUMER ) :
 138                 return -ENOTTY;
 139         default:
 140                 return -EPERM;
 141         }
 142 }
 143
 144 /**
 145  * Handle connection request transaction completion
 146  *
 147  * @v ibdev             Infiniband device
 148  * @v mi                Management interface
 149  * @v madx              Management transaction
 150  * @v rc                Status code
 151  * @v mad               Received MAD (or NULL on error)
 152  * @v av                Source address vector (or NULL on error)
 153  */
 154 static void ib_cm_req_complete ( struct ib_device *ibdev,
 155                                  struct ib_mad_interface *mi,
 156                                  struct ib_mad_transaction *madx,
 157                                  int rc, union ib_mad *mad,
 158                                  struct ib_address_vector *av ) {
 159         struct ib_connection *conn = ib_madx_get_ownerdata ( madx );
 160         struct ib_queue_pair *qp = conn->qp;
 161         struct ib_cm_common *common = &mad->cm.cm_data.common;
 162         struct ib_cm_connect_reply *connect_rep =
 163                 &mad->cm.cm_data.connect_reply;
 164         struct ib_cm_connect_reject *connect_rej =
 165                 &mad->cm.cm_data.connect_reject;
 166         void *private_data = NULL;
 167         size_t private_data_len = 0;
 168
 169         /* Report failures */
 170         if ( ( rc == 0 ) && ( mad->hdr.status != htons ( IB_MGMT_STATUS_OK ) ))
 171                 rc = -EIO;
 172         if ( rc != 0 ) {
 173                 DBGC ( conn, "CM %p connection request failed: %s\n",
 174                        conn, strerror ( rc ) );
 175                 goto out;
 176         }
 177
 178         /* Record remote communication ID */
 179         conn->remote_id = ntohl ( common->local_id );
 180
 181         /* Handle response */
 182         switch ( mad->hdr.attr_id ) {
 183
 184         case htons ( IB_CM_ATTR_CONNECT_REPLY ) :
 185                 /* Extract fields */
 186                 qp->av.qpn = ( ntohl ( connect_rep->local_qpn ) >> 8 );
 187                 qp->send.psn = ( ntohl ( connect_rep->starting_psn ) >> 8 );
 188                 private_data = &connect_rep->private_data;
 189                 private_data_len = sizeof ( connect_rep->private_data );
 190                 DBGC ( conn, "CM %p connected to QPN %lx PSN %x\n",
 191                        conn, qp->av.qpn, qp->send.psn );
 192
 193                 /* Modify queue pair */
 194                 if ( ( rc = ib_modify_qp ( ibdev, qp ) ) != 0 ) {
 195                         DBGC ( conn, "CM %p could not modify queue pair: %s\n",
 196                                conn, strerror ( rc ) );
 197                         goto out;
 198                 }
 199
 200                 /* Send "ready to use" reply */
 201                 if ( ( rc = ib_cm_send_rtu ( ibdev, mi, conn, av ) ) != 0 ) {
 202                         /* Treat as non-fatal */
 203                         rc = 0;
 204                 }
 205                 break;
 206
 207         case htons ( IB_CM_ATTR_CONNECT_REJECT ) :
 208                 /* Extract fields */
 209                 DBGC ( conn, "CM %p connection rejected (reason %d)\n",
 210                        conn, ntohs ( connect_rej->reason ) );
 211                 /* Private data is valid only for a Consumer Reject */
 212                 if ( connect_rej->reason == htons ( IB_CM_REJECT_CONSUMER ) ) {
 213                         private_data = &connect_rej->private_data;
 214                         private_data_len = sizeof (connect_rej->private_data);
 215                 }
 216                 rc = ib_cm_rejection_reason_to_rc ( connect_rej->reason );
 217                 break;
 218
 219         default:
 220                 DBGC ( conn, "CM %p unexpected response (attribute %04x)\n",
 221                        conn, ntohs ( mad->hdr.attr_id ) );
 222                 rc = -ENOTSUP;
 223                 break;
 224         }
 225
 226  out:
 227         /* Destroy the completed transaction */
 228         ib_destroy_madx ( ibdev, ibdev->gsi, madx );
 229         conn->madx = NULL;
 230
 231         /* Hand off to the upper completion handler */
 232         conn->op->changed ( ibdev, qp, conn, rc, private_data,
 233                             private_data_len );
 234 }
 235
 236 /** Connection request operations */
 237 static struct ib_mad_transaction_operations ib_cm_req_op = {
 238         .complete = ib_cm_req_complete,
 239 };
 240
 241 /**
 242  * Handle connection path transaction completion
 243  *
 244  * @v ibdev             Infiniband device
 245  * @v path              Path
 246  * @v rc                Status code
 247  * @v av                Address vector, or NULL on error
 248  */
 249 static void ib_cm_path_complete ( struct ib_device *ibdev,
 250                                   struct ib_path *path, int rc,
 251                                   struct ib_address_vector *av ) {
 252         struct ib_connection *conn = ib_path_get_ownerdata ( path );
 253         struct ib_queue_pair *qp = conn->qp;
 254         union ib_mad mad;
 255         struct ib_cm_connect_request *connect_req =
 256                 &mad.cm.cm_data.connect_request;
 257         size_t private_data_len;
 258
 259         /* Report failures */
 260         if ( rc != 0 ) {
 261                 DBGC ( conn, "CM %p path lookup failed: %s\n",
 262                        conn, strerror ( rc ) );
 263                 conn->op->changed ( ibdev, qp, conn, rc, NULL, 0 );
 264                 goto out;
 265         }
 266
 267         /* Update queue pair peer path */
 268         memcpy ( &qp->av, av, sizeof ( qp->av ) );
 269
 270         /* Construct connection request */
 271         memset ( &mad, 0, sizeof ( mad ) );
 272         mad.hdr.mgmt_class = IB_MGMT_CLASS_CM;
 273         mad.hdr.class_version = IB_CM_CLASS_VERSION;
 274         mad.hdr.method = IB_MGMT_METHOD_SEND;
 275         mad.hdr.attr_id = htons ( IB_CM_ATTR_CONNECT_REQUEST );
 276         connect_req->local_id = htonl ( conn->local_id );
 277         memcpy ( &connect_req->service_id, &conn->service_id,
 278                  sizeof ( connect_req->service_id ) );
 279         ib_get_hca_info ( ibdev, &connect_req->local_ca );
 280         connect_req->local_qpn__responder_resources =
 281                 htonl ( ( qp->qpn << 8 ) | 1 );
 282         connect_req->local_eecn__initiator_depth = htonl ( ( 0 << 8 ) | 1 );
 283         connect_req->remote_eecn__remote_timeout__service_type__ee_flow_ctrl =
 284                 htonl ( ( 0x14 << 3 ) | ( IB_CM_TRANSPORT_RC << 1 ) |
 285                         ( 0 << 0 ) );
 286         connect_req->starting_psn__local_timeout__retry_count =
 287                 htonl ( ( qp->recv.psn << 8 ) | ( 0x14 << 3 ) |
 288                         ( 0x07 << 0 ) );
 289         connect_req->pkey = htons ( ibdev->pkey );
 290         connect_req->payload_mtu__rdc_exists__rnr_retry =
 291                 ( ( IB_MTU_2048 << 4 ) | ( 1 << 3 ) | ( 0x07 << 0 ) );
 292         connect_req->max_cm_retries__srq =
 293                 ( ( 0x0f << 4 ) | ( 0 << 3 ) );
 294         connect_req->primary.local_lid = htons ( ibdev->lid );
 295         connect_req->primary.remote_lid = htons ( conn->qp->av.lid );
 296         memcpy ( &connect_req->primary.local_gid, &ibdev->gid,
 297                  sizeof ( connect_req->primary.local_gid ) );
 298         memcpy ( &connect_req->primary.remote_gid, &conn->qp->av.gid,
 299                  sizeof ( connect_req->primary.remote_gid ) );
 300         connect_req->primary.flow_label__rate =
 301                 htonl ( ( 0 << 12 ) | ( conn->qp->av.rate << 0 ) );
 302         connect_req->primary.hop_limit = 0;
 303         connect_req->primary.sl__subnet_local =
 304                 ( ( conn->qp->av.sl << 4 ) | ( 1 << 3 ) );
 305         connect_req->primary.local_ack_timeout = ( 0x13 << 3 );
 306         private_data_len = conn->private_data_len;
 307         if ( private_data_len > sizeof ( connect_req->private_data ) )
 308                 private_data_len = sizeof ( connect_req->private_data );
 309         memcpy ( &connect_req->private_data, &conn->private_data,
 310                  private_data_len );
 311
 312         /* Create connection request */
 313         av->qpn = IB_QPN_GSI;
 314         av->qkey = IB_QKEY_GSI;
 315         conn->madx = ib_create_madx ( ibdev, ibdev->gsi, &mad, av,
 316                                       &ib_cm_req_op );
 317         if ( ! conn->madx ) {
 318                 DBGC ( conn, "CM %p could not create connection request\n",
 319                        conn );
 320                 conn->op->changed ( ibdev, qp, conn, rc, NULL, 0 );
 321                 goto out;
 322         }
 323         ib_madx_set_ownerdata ( conn->madx, conn );
 324
 325  out:
 326         /* Destroy the completed transaction */
 327         ib_destroy_path ( ibdev, path );
 328         conn->path = NULL;
 329 }
 330
 331 /** Connection path operations */
 332 static struct ib_path_operations ib_cm_path_op = {
 333         .complete = ib_cm_path_complete,
 334 };
 335
 336 /**
 337  * Create connection to remote QP
 338  *
 339  * @v ibdev             Infiniband device
 340  * @v qp                Queue pair
 341  * @v dgid              Target GID
 342  * @v service_id        Target service ID
 343  * @v private_data      Connection request private data
 344  * @v private_data_len  Length of connection request private data
 345  * @v op                Connection operations
 346  * @ret conn            Connection
 347  */
 348 struct ib_connection *
 349 ib_create_conn ( struct ib_device *ibdev, struct ib_queue_pair *qp,
 350                  struct ib_gid *dgid, struct ib_gid_half *service_id,
 351                  void *private_data, size_t private_data_len,
 352                  struct ib_connection_operations *op ) {
 353         struct ib_connection *conn;
 354
 355         /* Allocate and initialise request */
 356         conn = zalloc ( sizeof ( *conn ) + private_data_len );
 357         if ( ! conn )
 358                 goto err_alloc_conn;
 359         conn->ibdev = ibdev;
 360         conn->qp = qp;
 361         memset ( &qp->av, 0, sizeof ( qp->av ) );
 362         qp->av.gid_present = 1;
 363         memcpy ( &qp->av.gid, dgid, sizeof ( qp->av.gid ) );
 364         conn->local_id = random();
 365         memcpy ( &conn->service_id, service_id, sizeof ( conn->service_id ) );
 366         conn->op = op;
 367         conn->private_data_len = private_data_len;
 368         memcpy ( &conn->private_data, private_data, private_data_len );
 369
 370         /* Create path */
 371         conn->path = ib_create_path ( ibdev, &qp->av, &ib_cm_path_op );
 372         if ( ! conn->path )
 373                 goto err_create_path;
 374         ib_path_set_ownerdata ( conn->path, conn );
 375
 376         /* Add to list of connections */
 377         list_add ( &conn->list, &ib_cm_conns );
 378
 379         DBGC ( conn, "CM %p created for IBDEV %p QPN %lx\n",
 380                conn, ibdev, qp->qpn );
 381         DBGC ( conn, "CM %p connecting to %08x:%08x:%08x:%08x %08x:%08x\n",
 382                conn, ntohl ( dgid->u.dwords[0] ), ntohl ( dgid->u.dwords[1] ),
 383                ntohl ( dgid->u.dwords[2] ), ntohl ( dgid->u.dwords[3] ),
 384                ntohl ( service_id->u.dwords[0] ),
 385                ntohl ( service_id->u.dwords[1] ) );
 386
 387         return conn;
 388
 389         ib_destroy_path ( ibdev, conn->path );
 390  err_create_path:
 391         free ( conn );
 392  err_alloc_conn:
 393         return NULL;
 394 }
 395
 396 /**
 397  * Destroy connection to remote QP
 398  *
 399  * @v ibdev             Infiniband device
 400  * @v qp                Queue pair
 401  * @v conn              Connection
 402  */
 403 void ib_destroy_conn ( struct ib_device *ibdev,
 404                        struct ib_queue_pair *qp __unused,
 405                        struct ib_connection *conn ) {
 406
 407         list_del ( &conn->list );
 408         if ( conn->madx )
 409                 ib_destroy_madx ( ibdev, ibdev->gsi, conn->madx );
 410         if ( conn->path )
 411                 ib_destroy_path ( ibdev, conn->path );
 412         free ( conn );
 413 }