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