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 * Return 1 if constructed; otherwise, return 0.
46 int qib_make_uc_req(struct qib_qp *qp)
48 struct qib_other_headers *ohdr;
54 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
57 spin_lock_irqsave(&qp->s_lock, flags);
59 if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
60 if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
62 /* We are in the error state, flush the work request. */
63 if (qp->s_last == qp->s_head)
65 /* If DMAs are in progress, we can't flush immediately. */
66 if (atomic_read(&qp->s_dma_busy)) {
67 qp->s_flags |= QIB_S_WAIT_DMA;
70 wqe = get_swqe_ptr(qp, qp->s_last);
71 qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
75 ohdr = &qp->s_hdr.u.oth;
76 if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
77 ohdr = &qp->s_hdr.u.l.oth;
79 /* header size in 32-bit words LRH+BTH = (8+12)/4. */
83 /* Get the next send request. */
84 wqe = get_swqe_ptr(qp, qp->s_cur);
86 switch (qp->s_state) {
88 if (!(ib_qib_state_ops[qp->state] &
89 QIB_PROCESS_NEXT_SEND_OK))
91 /* Check if send work queue is empty. */
92 if (qp->s_cur == qp->s_head)
95 * Start a new request.
97 wqe->psn = qp->s_next_psn;
98 qp->s_psn = qp->s_next_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;
105 switch (wqe->wr.opcode) {
107 case IB_WR_SEND_WITH_IMM:
109 qp->s_state = OP(SEND_FIRST);
113 if (wqe->wr.opcode == IB_WR_SEND)
114 qp->s_state = OP(SEND_ONLY);
117 OP(SEND_ONLY_WITH_IMMEDIATE);
118 /* Immediate data comes after the BTH */
119 ohdr->u.imm_data = wqe->wr.ex.imm_data;
122 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
123 bth0 |= IB_BTH_SOLICITED;
125 if (++qp->s_cur >= qp->s_size)
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->wr.wr.rdma.remote_addr);
133 ohdr->u.rc.reth.rkey =
134 cpu_to_be32(wqe->wr.wr.rdma.rkey);
135 ohdr->u.rc.reth.length = cpu_to_be32(len);
136 hwords += sizeof(struct ib_reth) / 4;
138 qp->s_state = OP(RDMA_WRITE_FIRST);
142 if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
143 qp->s_state = OP(RDMA_WRITE_ONLY);
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;
150 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
151 bth0 |= IB_BTH_SOLICITED;
154 if (++qp->s_cur >= qp->s_size)
164 qp->s_state = OP(SEND_MIDDLE);
166 case OP(SEND_MIDDLE):
172 if (wqe->wr.opcode == IB_WR_SEND)
173 qp->s_state = OP(SEND_LAST);
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;
180 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
181 bth0 |= IB_BTH_SOLICITED;
183 if (++qp->s_cur >= qp->s_size)
187 case OP(RDMA_WRITE_FIRST):
188 qp->s_state = OP(RDMA_WRITE_MIDDLE);
190 case OP(RDMA_WRITE_MIDDLE):
196 if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
197 qp->s_state = OP(RDMA_WRITE_LAST);
200 OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
201 /* Immediate data comes after the BTH */
202 ohdr->u.imm_data = wqe->wr.ex.imm_data;
204 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
205 bth0 |= IB_BTH_SOLICITED;
208 if (++qp->s_cur >= qp->s_size)
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_next_psn++ & QIB_PSN_MASK);
223 qp->s_flags &= ~QIB_S_BUSY;
225 spin_unlock_irqrestore(&qp->s_lock, flags);
230 * qib_uc_rcv - handle an incoming UC packet
231 * @ibp: the port the packet came in on
232 * @hdr: the header of the packet
233 * @has_grh: true if the packet has a GRH
234 * @data: the packet data
235 * @tlen: the length of the packet
236 * @qp: the QP for this packet.
238 * This is called from qib_qp_rcv() to process an incoming UC packet
240 * Called at interrupt level.
242 void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
243 int has_grh, void *data, u32 tlen, struct qib_qp *qp)
245 struct qib_other_headers *ohdr;
252 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
253 struct ib_reth *reth;
259 hdrsize = 8 + 12; /* LRH + BTH */
261 ohdr = &hdr->u.l.oth;
262 hdrsize = 8 + 40 + 12; /* LRH + GRH + BTH */
265 opcode = be32_to_cpu(ohdr->bth[0]);
266 spin_lock_irqsave(&qp->s_lock, flags);
267 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
269 spin_unlock_irqrestore(&qp->s_lock, flags);
271 psn = be32_to_cpu(ohdr->bth[2]);
273 memset(&wc, 0, sizeof wc);
275 /* Compare the PSN verses the expected PSN. */
276 if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
278 * Handle a sequence error.
279 * Silently drop any current message.
283 if (qp->r_state == OP(SEND_FIRST) ||
284 qp->r_state == OP(SEND_MIDDLE)) {
285 set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
286 qp->r_sge.num_sge = 0;
288 while (qp->r_sge.num_sge) {
289 atomic_dec(&qp->r_sge.sge.mr->refcount);
290 if (--qp->r_sge.num_sge)
291 qp->r_sge.sge = *qp->r_sge.sg_list++;
293 qp->r_state = OP(SEND_LAST);
297 case OP(SEND_ONLY_WITH_IMMEDIATE):
300 case OP(RDMA_WRITE_FIRST):
301 case OP(RDMA_WRITE_ONLY):
302 case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
310 /* Check for opcode sequence errors. */
311 switch (qp->r_state) {
313 case OP(SEND_MIDDLE):
314 if (opcode == OP(SEND_MIDDLE) ||
315 opcode == OP(SEND_LAST) ||
316 opcode == OP(SEND_LAST_WITH_IMMEDIATE))
320 case OP(RDMA_WRITE_FIRST):
321 case OP(RDMA_WRITE_MIDDLE):
322 if (opcode == OP(RDMA_WRITE_MIDDLE) ||
323 opcode == OP(RDMA_WRITE_LAST) ||
324 opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
329 if (opcode == OP(SEND_FIRST) ||
330 opcode == OP(SEND_ONLY) ||
331 opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
332 opcode == OP(RDMA_WRITE_FIRST) ||
333 opcode == OP(RDMA_WRITE_ONLY) ||
334 opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
339 if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
340 qp->r_flags |= QIB_R_COMM_EST;
341 if (qp->ibqp.event_handler) {
344 ev.device = qp->ibqp.device;
345 ev.element.qp = &qp->ibqp;
346 ev.event = IB_EVENT_COMM_EST;
347 qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
351 /* OK, process the packet. */
355 case OP(SEND_ONLY_WITH_IMMEDIATE):
357 if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
358 qp->r_sge = qp->s_rdma_read_sge;
360 ret = qib_get_rwqe(qp, 0);
366 * qp->s_rdma_read_sge will be the owner
367 * of the mr references.
369 qp->s_rdma_read_sge = qp->r_sge;
372 if (opcode == OP(SEND_ONLY))
374 else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
377 case OP(SEND_MIDDLE):
378 /* Check for invalid length PMTU or posted rwqe len. */
379 if (unlikely(tlen != (hdrsize + pmtu + 4)))
381 qp->r_rcv_len += pmtu;
382 if (unlikely(qp->r_rcv_len > qp->r_len))
384 qib_copy_sge(&qp->r_sge, data, pmtu, 0);
387 case OP(SEND_LAST_WITH_IMMEDIATE):
389 wc.ex.imm_data = ohdr->u.imm_data;
391 wc.wc_flags = IB_WC_WITH_IMM;
395 /* Get the number of bytes the message was padded by. */
396 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
397 /* Check for invalid length. */
398 /* XXX LAST len should be >= 1 */
399 if (unlikely(tlen < (hdrsize + pad + 4)))
401 /* Don't count the CRC. */
402 tlen -= (hdrsize + pad + 4);
403 wc.byte_len = tlen + qp->r_rcv_len;
404 if (unlikely(wc.byte_len > qp->r_len))
406 wc.opcode = IB_WC_RECV;
408 qib_copy_sge(&qp->r_sge, data, tlen, 0);
409 while (qp->s_rdma_read_sge.num_sge) {
410 atomic_dec(&qp->s_rdma_read_sge.sge.mr->refcount);
411 if (--qp->s_rdma_read_sge.num_sge)
412 qp->s_rdma_read_sge.sge =
413 *qp->s_rdma_read_sge.sg_list++;
415 wc.wr_id = qp->r_wr_id;
416 wc.status = IB_WC_SUCCESS;
418 wc.src_qp = qp->remote_qpn;
419 wc.slid = qp->remote_ah_attr.dlid;
420 wc.sl = qp->remote_ah_attr.sl;
421 /* Signal completion event if the solicited bit is set. */
422 qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
424 cpu_to_be32(IB_BTH_SOLICITED)) != 0);
427 case OP(RDMA_WRITE_FIRST):
428 case OP(RDMA_WRITE_ONLY):
429 case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
431 if (unlikely(!(qp->qp_access_flags &
432 IB_ACCESS_REMOTE_WRITE))) {
435 reth = &ohdr->u.rc.reth;
436 hdrsize += sizeof(*reth);
437 qp->r_len = be32_to_cpu(reth->length);
439 qp->r_sge.sg_list = NULL;
440 if (qp->r_len != 0) {
441 u32 rkey = be32_to_cpu(reth->rkey);
442 u64 vaddr = be64_to_cpu(reth->vaddr);
446 ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
447 vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
450 qp->r_sge.num_sge = 1;
452 qp->r_sge.num_sge = 0;
453 qp->r_sge.sge.mr = NULL;
454 qp->r_sge.sge.vaddr = NULL;
455 qp->r_sge.sge.length = 0;
456 qp->r_sge.sge.sge_length = 0;
458 if (opcode == OP(RDMA_WRITE_ONLY))
460 else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
463 case OP(RDMA_WRITE_MIDDLE):
464 /* Check for invalid length PMTU or posted rwqe len. */
465 if (unlikely(tlen != (hdrsize + pmtu + 4)))
467 qp->r_rcv_len += pmtu;
468 if (unlikely(qp->r_rcv_len > qp->r_len))
470 qib_copy_sge(&qp->r_sge, data, pmtu, 1);
473 case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
475 wc.ex.imm_data = ohdr->u.imm_data;
477 wc.wc_flags = IB_WC_WITH_IMM;
479 /* Get the number of bytes the message was padded by. */
480 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
481 /* Check for invalid length. */
482 /* XXX LAST len should be >= 1 */
483 if (unlikely(tlen < (hdrsize + pad + 4)))
485 /* Don't count the CRC. */
486 tlen -= (hdrsize + pad + 4);
487 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
489 if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
490 while (qp->s_rdma_read_sge.num_sge) {
491 atomic_dec(&qp->s_rdma_read_sge.sge.mr->
493 if (--qp->s_rdma_read_sge.num_sge)
494 qp->s_rdma_read_sge.sge =
495 *qp->s_rdma_read_sge.sg_list++;
498 ret = qib_get_rwqe(qp, 1);
504 wc.byte_len = qp->r_len;
505 wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
508 case OP(RDMA_WRITE_LAST):
510 /* Get the number of bytes the message was padded by. */
511 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
512 /* Check for invalid length. */
513 /* XXX LAST len should be >= 1 */
514 if (unlikely(tlen < (hdrsize + pad + 4)))
516 /* Don't count the CRC. */
517 tlen -= (hdrsize + pad + 4);
518 if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
520 qib_copy_sge(&qp->r_sge, data, tlen, 1);
521 while (qp->r_sge.num_sge) {
522 atomic_dec(&qp->r_sge.sge.mr->refcount);
523 if (--qp->r_sge.num_sge)
524 qp->r_sge.sge = *qp->r_sge.sg_list++;
529 /* Drop packet for unknown opcodes. */
533 qp->r_state = opcode;
537 set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
538 qp->r_sge.num_sge = 0;
544 qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
548 spin_unlock_irqrestore(&qp->s_lock, flags);