drivers/infiniband/hw/qib/qib_uc.c

   1 /*
   2  * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
   3  * All rights reserved.
   4  * Copyright (c) 2005, 2006 PathScale, Inc. 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 #include "qib.h"
  36
  37 /* cut down ridiculously long IB macro names */
  38 #define OP(x) IB_OPCODE_UC_##x
  39
  40 /**
  41  * qib_make_uc_req - construct a request packet (SEND, RDMA write)
  42  * @qp: a pointer to the QP
  43  *
  44  * Assumes the s_lock is held.
  45  *
  46  * Return 1 if constructed; otherwise, return 0.
  47  */
  48 int qib_make_uc_req(struct rvt_qp *qp, unsigned long *flags)
  49 {
  50         struct qib_qp_priv *priv = qp->priv;
  51         struct ib_other_headers *ohdr;
  52         struct rvt_swqe *wqe;
  53         u32 hwords;
  54         u32 bth0;
  55         u32 len;
  56         u32 pmtu = qp->pmtu;
  57         int ret = 0;
  58
  59         if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK)) {
  60                 if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
  61                         goto bail;
  62                 /* We are in the error state, flush the work request. */
  63                 smp_read_barrier_depends(); /* see post_one_send() */
  64                 if (qp->s_last == ACCESS_ONCE(qp->s_head))
  65                         goto bail;
  66                 /* If DMAs are in progress, we can't flush immediately. */
  67                 if (atomic_read(&priv->s_dma_busy)) {
  68                         qp->s_flags |= RVT_S_WAIT_DMA;
  69                         goto bail;
  70                 }
  71                 wqe = rvt_get_swqe_ptr(qp, qp->s_last);
  72                 qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
  73                 goto done;
  74         }
  75
  76         ohdr = &priv->s_hdr->u.oth;
  77         if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
  78                 ohdr = &priv->s_hdr->u.l.oth;
  79
  80         /* header size in 32-bit words LRH+BTH = (8+12)/4. */
  81         hwords = 5;
  82         bth0 = 0;
  83
  84         /* Get the next send request. */
  85         wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
  86         qp->s_wqe = NULL;
  87         switch (qp->s_state) {
  88         default:
  89                 if (!(ib_rvt_state_ops[qp->state] &
  90                     RVT_PROCESS_NEXT_SEND_OK))
  91                         goto bail;
  92                 /* Check if send work queue is empty. */
  93                 smp_read_barrier_depends(); /* see post_one_send() */
  94                 if (qp->s_cur == ACCESS_ONCE(qp->s_head))
  95                         goto bail;
  96                 /*
  97                  * Start a new request.
  98                  */
  99                 qp->s_psn = wqe->psn;
 100                 qp->s_sge.sge = wqe->sg_list[0];
 101                 qp->s_sge.sg_list = wqe->sg_list + 1;
 102                 qp->s_sge.num_sge = wqe->wr.num_sge;
 103                 qp->s_sge.total_len = wqe->length;
 104                 len = wqe->length;
 105                 qp->s_len = len;
 106                 switch (wqe->wr.opcode) {
 107                 case IB_WR_SEND:
 108                 case IB_WR_SEND_WITH_IMM:
 109                         if (len > pmtu) {
 110                                 qp->s_state = OP(SEND_FIRST);
 111                                 len = pmtu;
 112                                 break;
 113                         }
 114                         if (wqe->wr.opcode == IB_WR_SEND)
 115                                 qp->s_state = OP(SEND_ONLY);
 116                         else {
 117                                 qp->s_state =
 118                                         OP(SEND_ONLY_WITH_IMMEDIATE);
 119                                 /* Immediate data comes after the BTH */
 120                                 ohdr->u.imm_data = wqe->wr.ex.imm_data;
 121                                 hwords += 1;
 122                         }
 123                         if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 124                                 bth0 |= IB_BTH_SOLICITED;
 125                         qp->s_wqe = wqe;
 126                         if (++qp->s_cur >= qp->s_size)
 127                                 qp->s_cur = 0;
 128                         break;
 129
 130                 case IB_WR_RDMA_WRITE:
 131                 case IB_WR_RDMA_WRITE_WITH_IMM:
 132                         ohdr->u.rc.reth.vaddr =
 133                                 cpu_to_be64(wqe->rdma_wr.remote_addr);
 134                         ohdr->u.rc.reth.rkey =
 135                                 cpu_to_be32(wqe->rdma_wr.rkey);
 136                         ohdr->u.rc.reth.length = cpu_to_be32(len);
 137                         hwords += sizeof(struct ib_reth) / 4;
 138                         if (len > pmtu) {
 139                                 qp->s_state = OP(RDMA_WRITE_FIRST);
 140                                 len = pmtu;
 141                                 break;
 142                         }
 143                         if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
 144                                 qp->s_state = OP(RDMA_WRITE_ONLY);
 145                         else {
 146                                 qp->s_state =
 147                                         OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
 148                                 /* Immediate data comes after the RETH */
 149                                 ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
 150                                 hwords += 1;
 151                                 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 152                                         bth0 |= IB_BTH_SOLICITED;
 153                         }
 154                         qp->s_wqe = wqe;
 155                         if (++qp->s_cur >= qp->s_size)
 156                                 qp->s_cur = 0;
 157                         break;
 158
 159                 default:
 160                         goto bail;
 161                 }
 162                 break;
 163
 164         case OP(SEND_FIRST):
 165                 qp->s_state = OP(SEND_MIDDLE);
 166                 /* FALLTHROUGH */
 167         case OP(SEND_MIDDLE):
 168                 len = qp->s_len;
 169                 if (len > pmtu) {
 170                         len = pmtu;
 171                         break;
 172                 }
 173                 if (wqe->wr.opcode == IB_WR_SEND)
 174                         qp->s_state = OP(SEND_LAST);
 175                 else {
 176                         qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
 177                         /* Immediate data comes after the BTH */
 178                         ohdr->u.imm_data = wqe->wr.ex.imm_data;
 179                         hwords += 1;
 180                 }
 181                 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 182                         bth0 |= IB_BTH_SOLICITED;
 183                 qp->s_wqe = wqe;
 184                 if (++qp->s_cur >= qp->s_size)
 185                         qp->s_cur = 0;
 186                 break;
 187
 188         case OP(RDMA_WRITE_FIRST):
 189                 qp->s_state = OP(RDMA_WRITE_MIDDLE);
 190                 /* FALLTHROUGH */
 191         case OP(RDMA_WRITE_MIDDLE):
 192                 len = qp->s_len;
 193                 if (len > pmtu) {
 194                         len = pmtu;
 195                         break;
 196                 }
 197                 if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
 198                         qp->s_state = OP(RDMA_WRITE_LAST);
 199                 else {
 200                         qp->s_state =
 201                                 OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
 202                         /* Immediate data comes after the BTH */
 203                         ohdr->u.imm_data = wqe->wr.ex.imm_data;
 204                         hwords += 1;
 205                         if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 206                                 bth0 |= IB_BTH_SOLICITED;
 207                 }
 208                 qp->s_wqe = wqe;
 209                 if (++qp->s_cur >= qp->s_size)
 210                         qp->s_cur = 0;
 211                 break;
 212         }
 213         qp->s_len -= len;
 214         qp->s_hdrwords = hwords;
 215         qp->s_cur_sge = &qp->s_sge;
 216         qp->s_cur_size = len;
 217         qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
 218                             qp->s_psn++ & QIB_PSN_MASK);
 219 done:
 220         return 1;
 221 bail:
 222         qp->s_flags &= ~RVT_S_BUSY;
 223         return ret;
 224 }
 225
 226 /**
 227  * qib_uc_rcv - handle an incoming UC packet
 228  * @ibp: the port the packet came in on
 229  * @hdr: the header of the packet
 230  * @has_grh: true if the packet has a GRH
 231  * @data: the packet data
 232  * @tlen: the length of the packet
 233  * @qp: the QP for this packet.
 234  *
 235  * This is called from qib_qp_rcv() to process an incoming UC packet
 236  * for the given QP.
 237  * Called at interrupt level.
 238  */
 239 void qib_uc_rcv(struct qib_ibport *ibp, struct ib_header *hdr,
 240                 int has_grh, void *data, u32 tlen, struct rvt_qp *qp)
 241 {
 242         struct ib_other_headers *ohdr;
 243         u32 opcode;
 244         u32 hdrsize;
 245         u32 psn;
 246         u32 pad;
 247         struct ib_wc wc;
 248         u32 pmtu = qp->pmtu;
 249         struct ib_reth *reth;
 250         int ret;
 251
 252         /* Check for GRH */
 253         if (!has_grh) {
 254                 ohdr = &hdr->u.oth;
 255                 hdrsize = 8 + 12;       /* LRH + BTH */
 256         } else {
 257                 ohdr = &hdr->u.l.oth;
 258                 hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
 259         }
 260
 261         opcode = be32_to_cpu(ohdr->bth[0]);
 262         if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
 263                 return;
 264
 265         psn = be32_to_cpu(ohdr->bth[2]);
 266         opcode >>= 24;
 267
 268         /* Compare the PSN verses the expected PSN. */
 269         if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
 270                 /*
 271                  * Handle a sequence error.
 272                  * Silently drop any current message.
 273                  */
 274                 qp->r_psn = psn;
 275 inv:
 276                 if (qp->r_state == OP(SEND_FIRST) ||
 277                     qp->r_state == OP(SEND_MIDDLE)) {
 278                         set_bit(RVT_R_REWIND_SGE, &qp->r_aflags);
 279                         qp->r_sge.num_sge = 0;
 280                 } else
 281                         rvt_put_ss(&qp->r_sge);
 282                 qp->r_state = OP(SEND_LAST);
 283                 switch (opcode) {
 284                 case OP(SEND_FIRST):
 285                 case OP(SEND_ONLY):
 286                 case OP(SEND_ONLY_WITH_IMMEDIATE):
 287                         goto send_first;
 288
 289                 case OP(RDMA_WRITE_FIRST):
 290                 case OP(RDMA_WRITE_ONLY):
 291                 case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
 292                         goto rdma_first;
 293
 294                 default:
 295                         goto drop;
 296                 }
 297         }
 298
 299         /* Check for opcode sequence errors. */
 300         switch (qp->r_state) {
 301         case OP(SEND_FIRST):
 302         case OP(SEND_MIDDLE):
 303                 if (opcode == OP(SEND_MIDDLE) ||
 304                     opcode == OP(SEND_LAST) ||
 305                     opcode == OP(SEND_LAST_WITH_IMMEDIATE))
 306                         break;
 307                 goto inv;
 308
 309         case OP(RDMA_WRITE_FIRST):
 310         case OP(RDMA_WRITE_MIDDLE):
 311                 if (opcode == OP(RDMA_WRITE_MIDDLE) ||
 312                     opcode == OP(RDMA_WRITE_LAST) ||
 313                     opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
 314                         break;
 315                 goto inv;
 316
 317         default:
 318                 if (opcode == OP(SEND_FIRST) ||
 319                     opcode == OP(SEND_ONLY) ||
 320                     opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
 321                     opcode == OP(RDMA_WRITE_FIRST) ||
 322                     opcode == OP(RDMA_WRITE_ONLY) ||
 323                     opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
 324                         break;
 325                 goto inv;
 326         }
 327
 328         if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST))
 329                 rvt_comm_est(qp);
 330
 331         /* OK, process the packet. */
 332         switch (opcode) {
 333         case OP(SEND_FIRST):
 334         case OP(SEND_ONLY):
 335         case OP(SEND_ONLY_WITH_IMMEDIATE):
 336 send_first:
 337                 if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
 338                         qp->r_sge = qp->s_rdma_read_sge;
 339                 else {
 340                         ret = qib_get_rwqe(qp, 0);
 341                         if (ret < 0)
 342                                 goto op_err;
 343                         if (!ret)
 344                                 goto drop;
 345                         /*
 346                          * qp->s_rdma_read_sge will be the owner
 347                          * of the mr references.
 348                          */
 349                         qp->s_rdma_read_sge = qp->r_sge;
 350                 }
 351                 qp->r_rcv_len = 0;
 352                 if (opcode == OP(SEND_ONLY))
 353                         goto no_immediate_data;
 354                 else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
 355                         goto send_last_imm;
 356                 /* FALLTHROUGH */
 357         case OP(SEND_MIDDLE):
 358                 /* Check for invalid length PMTU or posted rwqe len. */
 359                 if (unlikely(tlen != (hdrsize + pmtu + 4)))
 360                         goto rewind;
 361                 qp->r_rcv_len += pmtu;
 362                 if (unlikely(qp->r_rcv_len > qp->r_len))
 363                         goto rewind;
 364                 qib_copy_sge(&qp->r_sge, data, pmtu, 0);
 365                 break;
 366
 367         case OP(SEND_LAST_WITH_IMMEDIATE):
 368 send_last_imm:
 369                 wc.ex.imm_data = ohdr->u.imm_data;
 370                 hdrsize += 4;
 371                 wc.wc_flags = IB_WC_WITH_IMM;
 372                 goto send_last;
 373         case OP(SEND_LAST):
 374 no_immediate_data:
 375                 wc.ex.imm_data = 0;
 376                 wc.wc_flags = 0;
 377 send_last:
 378                 /* Get the number of bytes the message was padded by. */
 379                 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
 380                 /* Check for invalid length. */
 381                 /* XXX LAST len should be >= 1 */
 382                 if (unlikely(tlen < (hdrsize + pad + 4)))
 383                         goto rewind;
 384                 /* Don't count the CRC. */
 385                 tlen -= (hdrsize + pad + 4);
 386                 wc.byte_len = tlen + qp->r_rcv_len;
 387                 if (unlikely(wc.byte_len > qp->r_len))
 388                         goto rewind;
 389                 wc.opcode = IB_WC_RECV;
 390                 qib_copy_sge(&qp->r_sge, data, tlen, 0);
 391                 rvt_put_ss(&qp->s_rdma_read_sge);
 392 last_imm:
 393                 wc.wr_id = qp->r_wr_id;
 394                 wc.status = IB_WC_SUCCESS;
 395                 wc.qp = &qp->ibqp;
 396                 wc.src_qp = qp->remote_qpn;
 397                 wc.slid = rdma_ah_get_dlid(&qp->remote_ah_attr);
 398                 wc.sl = rdma_ah_get_sl(&qp->remote_ah_attr);
 399                 /* zero fields that are N/A */
 400                 wc.vendor_err = 0;
 401                 wc.pkey_index = 0;
 402                 wc.dlid_path_bits = 0;
 403                 wc.port_num = 0;
 404                 /* Signal completion event if the solicited bit is set. */
 405                 rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
 406                              (ohdr->bth[0] &
 407                                 cpu_to_be32(IB_BTH_SOLICITED)) != 0);
 408                 break;
 409
 410         case OP(RDMA_WRITE_FIRST):
 411         case OP(RDMA_WRITE_ONLY):
 412         case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
 413 rdma_first:
 414                 if (unlikely(!(qp->qp_access_flags &
 415                                IB_ACCESS_REMOTE_WRITE))) {
 416                         goto drop;
 417                 }
 418                 reth = &ohdr->u.rc.reth;
 419                 hdrsize += sizeof(*reth);
 420                 qp->r_len = be32_to_cpu(reth->length);
 421                 qp->r_rcv_len = 0;
 422                 qp->r_sge.sg_list = NULL;
 423                 if (qp->r_len != 0) {
 424                         u32 rkey = be32_to_cpu(reth->rkey);
 425                         u64 vaddr = be64_to_cpu(reth->vaddr);
 426                         int ok;
 427
 428                         /* Check rkey */
 429                         ok = rvt_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
 430                                          vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
 431                         if (unlikely(!ok))
 432                                 goto drop;
 433                         qp->r_sge.num_sge = 1;
 434                 } else {
 435                         qp->r_sge.num_sge = 0;
 436                         qp->r_sge.sge.mr = NULL;
 437                         qp->r_sge.sge.vaddr = NULL;
 438                         qp->r_sge.sge.length = 0;
 439                         qp->r_sge.sge.sge_length = 0;
 440                 }
 441                 if (opcode == OP(RDMA_WRITE_ONLY))
 442                         goto rdma_last;
 443                 else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE)) {
 444                         wc.ex.imm_data = ohdr->u.rc.imm_data;
 445                         goto rdma_last_imm;
 446                 }
 447                 /* FALLTHROUGH */
 448         case OP(RDMA_WRITE_MIDDLE):
 449                 /* Check for invalid length PMTU or posted rwqe len. */
 450                 if (unlikely(tlen != (hdrsize + pmtu + 4)))
 451                         goto drop;
 452                 qp->r_rcv_len += pmtu;
 453                 if (unlikely(qp->r_rcv_len > qp->r_len))
 454                         goto drop;
 455                 qib_copy_sge(&qp->r_sge, data, pmtu, 1);
 456                 break;
 457
 458         case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
 459                 wc.ex.imm_data = ohdr->u.imm_data;
 460 rdma_last_imm:
 461                 hdrsize += 4;
 462                 wc.wc_flags = IB_WC_WITH_IMM;
 463
 464                 /* Get the number of bytes the message was padded by. */
 465                 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
 466                 /* Check for invalid length. */
 467                 /* XXX LAST len should be >= 1 */
 468                 if (unlikely(tlen < (hdrsize + pad + 4)))
 469                         goto drop;
 470                 /* Don't count the CRC. */
 471                 tlen -= (hdrsize + pad + 4);
 472                 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
 473                         goto drop;
 474                 if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
 475                         rvt_put_ss(&qp->s_rdma_read_sge);
 476                 else {
 477                         ret = qib_get_rwqe(qp, 1);
 478                         if (ret < 0)
 479                                 goto op_err;
 480                         if (!ret)
 481                                 goto drop;
 482                 }
 483                 wc.byte_len = qp->r_len;
 484                 wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
 485                 qib_copy_sge(&qp->r_sge, data, tlen, 1);
 486                 rvt_put_ss(&qp->r_sge);
 487                 goto last_imm;
 488
 489         case OP(RDMA_WRITE_LAST):
 490 rdma_last:
 491                 /* Get the number of bytes the message was padded by. */
 492                 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
 493                 /* Check for invalid length. */
 494                 /* XXX LAST len should be >= 1 */
 495                 if (unlikely(tlen < (hdrsize + pad + 4)))
 496                         goto drop;
 497                 /* Don't count the CRC. */
 498                 tlen -= (hdrsize + pad + 4);
 499                 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
 500                         goto drop;
 501                 qib_copy_sge(&qp->r_sge, data, tlen, 1);
 502                 rvt_put_ss(&qp->r_sge);
 503                 break;
 504
 505         default:
 506                 /* Drop packet for unknown opcodes. */
 507                 goto drop;
 508         }
 509         qp->r_psn++;
 510         qp->r_state = opcode;
 511         return;
 512
 513 rewind:
 514         set_bit(RVT_R_REWIND_SGE, &qp->r_aflags);
 515         qp->r_sge.num_sge = 0;
 516 drop:
 517         ibp->rvp.n_pkt_drops++;
 518         return;
 519
 520 op_err:
 521         rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
 522         return;
 523
 524 }