2 * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
4 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
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.
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
37 /* cut down ridiculously long IB macro names */
38 #define OP(x) IB_OPCODE_UC_##x
41 * qib_make_uc_req - construct a request packet (SEND, RDMA write)
42 * @qp: a pointer to the QP
44 * Assumes the s_lock is held.
46 * Return 1 if constructed; otherwise, return 0.
48 int qib_make_uc_req(struct rvt_qp *qp, unsigned long *flags)
50 struct qib_qp_priv *priv = qp->priv;
51 struct ib_other_headers *ohdr;
59 if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK)) {
60 if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
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))
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;
71 wqe = rvt_get_swqe_ptr(qp, qp->s_last);
72 qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
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;
80 /* header size in 32-bit words LRH+BTH = (8+12)/4. */
84 /* Get the next send request. */
85 wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
87 switch (qp->s_state) {
89 if (!(ib_rvt_state_ops[qp->state] &
90 RVT_PROCESS_NEXT_SEND_OK))
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))
97 * Start a new request.
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;
106 switch (wqe->wr.opcode) {
108 case IB_WR_SEND_WITH_IMM:
110 qp->s_state = OP(SEND_FIRST);
114 if (wqe->wr.opcode == IB_WR_SEND)
115 qp->s_state = OP(SEND_ONLY);
118 OP(SEND_ONLY_WITH_IMMEDIATE);
119 /* Immediate data comes after the BTH */
120 ohdr->u.imm_data = wqe->wr.ex.imm_data;
123 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
124 bth0 |= IB_BTH_SOLICITED;
126 if (++qp->s_cur >= qp->s_size)
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;
139 qp->s_state = OP(RDMA_WRITE_FIRST);
143 if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
144 qp->s_state = OP(RDMA_WRITE_ONLY);
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;
151 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
152 bth0 |= IB_BTH_SOLICITED;
155 if (++qp->s_cur >= qp->s_size)
165 qp->s_state = OP(SEND_MIDDLE);
167 case OP(SEND_MIDDLE):
173 if (wqe->wr.opcode == IB_WR_SEND)
174 qp->s_state = OP(SEND_LAST);
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;
181 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
182 bth0 |= IB_BTH_SOLICITED;
184 if (++qp->s_cur >= qp->s_size)
188 case OP(RDMA_WRITE_FIRST):
189 qp->s_state = OP(RDMA_WRITE_MIDDLE);
191 case OP(RDMA_WRITE_MIDDLE):
197 if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
198 qp->s_state = OP(RDMA_WRITE_LAST);
201 OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
202 /* Immediate data comes after the BTH */
203 ohdr->u.imm_data = wqe->wr.ex.imm_data;
205 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
206 bth0 |= IB_BTH_SOLICITED;
209 if (++qp->s_cur >= qp->s_size)
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);
222 qp->s_flags &= ~RVT_S_BUSY;
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.
235 * This is called from qib_qp_rcv() to process an incoming UC packet
237 * Called at interrupt level.
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)
242 struct ib_other_headers *ohdr;
249 struct ib_reth *reth;
255 hdrsize = 8 + 12; /* LRH + BTH */
257 ohdr = &hdr->u.l.oth;
258 hdrsize = 8 + 40 + 12; /* LRH + GRH + BTH */
261 opcode = be32_to_cpu(ohdr->bth[0]);
262 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
265 psn = be32_to_cpu(ohdr->bth[2]);
268 /* Compare the PSN verses the expected PSN. */
269 if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
271 * Handle a sequence error.
272 * Silently drop any current message.
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;
281 rvt_put_ss(&qp->r_sge);
282 qp->r_state = OP(SEND_LAST);
286 case OP(SEND_ONLY_WITH_IMMEDIATE):
289 case OP(RDMA_WRITE_FIRST):
290 case OP(RDMA_WRITE_ONLY):
291 case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
299 /* Check for opcode sequence errors. */
300 switch (qp->r_state) {
302 case OP(SEND_MIDDLE):
303 if (opcode == OP(SEND_MIDDLE) ||
304 opcode == OP(SEND_LAST) ||
305 opcode == OP(SEND_LAST_WITH_IMMEDIATE))
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))
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))
328 if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST))
331 /* OK, process the packet. */
335 case OP(SEND_ONLY_WITH_IMMEDIATE):
337 if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
338 qp->r_sge = qp->s_rdma_read_sge;
340 ret = qib_get_rwqe(qp, 0);
346 * qp->s_rdma_read_sge will be the owner
347 * of the mr references.
349 qp->s_rdma_read_sge = qp->r_sge;
352 if (opcode == OP(SEND_ONLY))
353 goto no_immediate_data;
354 else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
357 case OP(SEND_MIDDLE):
358 /* Check for invalid length PMTU or posted rwqe len. */
359 if (unlikely(tlen != (hdrsize + pmtu + 4)))
361 qp->r_rcv_len += pmtu;
362 if (unlikely(qp->r_rcv_len > qp->r_len))
364 qib_copy_sge(&qp->r_sge, data, pmtu, 0);
367 case OP(SEND_LAST_WITH_IMMEDIATE):
369 wc.ex.imm_data = ohdr->u.imm_data;
371 wc.wc_flags = IB_WC_WITH_IMM;
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)))
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))
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);
393 wc.wr_id = qp->r_wr_id;
394 wc.status = IB_WC_SUCCESS;
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 */
402 wc.dlid_path_bits = 0;
404 /* Signal completion event if the solicited bit is set. */
405 rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
407 cpu_to_be32(IB_BTH_SOLICITED)) != 0);
410 case OP(RDMA_WRITE_FIRST):
411 case OP(RDMA_WRITE_ONLY):
412 case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
414 if (unlikely(!(qp->qp_access_flags &
415 IB_ACCESS_REMOTE_WRITE))) {
418 reth = &ohdr->u.rc.reth;
419 hdrsize += sizeof(*reth);
420 qp->r_len = be32_to_cpu(reth->length);
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);
429 ok = rvt_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
430 vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
433 qp->r_sge.num_sge = 1;
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;
441 if (opcode == OP(RDMA_WRITE_ONLY))
443 else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE)) {
444 wc.ex.imm_data = ohdr->u.rc.imm_data;
448 case OP(RDMA_WRITE_MIDDLE):
449 /* Check for invalid length PMTU or posted rwqe len. */
450 if (unlikely(tlen != (hdrsize + pmtu + 4)))
452 qp->r_rcv_len += pmtu;
453 if (unlikely(qp->r_rcv_len > qp->r_len))
455 qib_copy_sge(&qp->r_sge, data, pmtu, 1);
458 case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
459 wc.ex.imm_data = ohdr->u.imm_data;
462 wc.wc_flags = IB_WC_WITH_IMM;
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)))
470 /* Don't count the CRC. */
471 tlen -= (hdrsize + pad + 4);
472 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
474 if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
475 rvt_put_ss(&qp->s_rdma_read_sge);
477 ret = qib_get_rwqe(qp, 1);
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);
489 case OP(RDMA_WRITE_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)))
497 /* Don't count the CRC. */
498 tlen -= (hdrsize + pad + 4);
499 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
501 qib_copy_sge(&qp->r_sge, data, tlen, 1);
502 rvt_put_ss(&qp->r_sge);
506 /* Drop packet for unknown opcodes. */
510 qp->r_state = opcode;
514 set_bit(RVT_R_REWIND_SGE, &qp->r_aflags);
515 qp->r_sge.num_sge = 0;
517 ibp->rvp.n_pkt_drops++;
521 rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);