2 * Copyright (c) 2012 - 2019 Intel Corporation. All rights reserved.
3 * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
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
35 #include <rdma/ib_smi.h>
36 #include <rdma/ib_verbs.h>
42 * qib_ud_loopback - handle send on loopback QPs
43 * @sqp: the sending QP
44 * @swqe: the send work request
46 * This is called from qib_make_ud_req() to forward a WQE addressed
48 * Note that the receive interrupt handler may be calling qib_ud_rcv()
49 * while this is being called.
51 static void qib_ud_loopback(struct rvt_qp *sqp, struct rvt_swqe *swqe)
53 struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
54 struct qib_pportdata *ppd = ppd_from_ibp(ibp);
55 struct qib_devdata *dd = ppd->dd;
56 struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
58 struct rdma_ah_attr *ah_attr;
60 struct rvt_sge_state ssge;
64 enum ib_qp_type sqptype, dqptype;
67 qp = rvt_lookup_qpn(rdi, &ibp->rvp, rvt_get_swqe_remote_qpn(swqe));
69 ibp->rvp.n_pkt_drops++;
73 sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
74 IB_QPT_UD : sqp->ibqp.qp_type;
75 dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
76 IB_QPT_UD : qp->ibqp.qp_type;
78 if (dqptype != sqptype ||
79 !(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
80 ibp->rvp.n_pkt_drops++;
84 ah_attr = rvt_get_swqe_ah_attr(swqe);
85 ppd = ppd_from_ibp(ibp);
87 if (qp->ibqp.qp_num > 1) {
92 pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
93 pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
94 if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
95 lid = ppd->lid | (rdma_ah_get_path_bits(ah_attr) &
96 ((1 << ppd->lmc) - 1));
97 qib_bad_pkey(ibp, pkey1,
98 rdma_ah_get_sl(ah_attr),
99 sqp->ibqp.qp_num, qp->ibqp.qp_num,
101 cpu_to_be16(rdma_ah_get_dlid(ah_attr)));
107 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
108 * Qkeys with the high order bit set mean use the
109 * qkey from the QP context instead of the WR (see 10.2.5).
111 if (qp->ibqp.qp_num) {
114 qkey = (int)rvt_get_swqe_remote_qkey(swqe) < 0 ?
115 sqp->qkey : rvt_get_swqe_remote_qkey(swqe);
116 if (unlikely(qkey != qp->qkey))
121 * A GRH is expected to precede the data even if not
122 * present on the wire.
124 length = swqe->length;
125 memset(&wc, 0, sizeof(wc));
126 wc.byte_len = length + sizeof(struct ib_grh);
128 if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
129 wc.wc_flags = IB_WC_WITH_IMM;
130 wc.ex.imm_data = swqe->wr.ex.imm_data;
133 spin_lock_irqsave(&qp->r_lock, flags);
136 * Get the next work request entry to find where to put the data.
138 if (qp->r_flags & RVT_R_REUSE_SGE)
139 qp->r_flags &= ~RVT_R_REUSE_SGE;
143 ret = rvt_get_rwqe(qp, false);
145 rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
149 if (qp->ibqp.qp_num == 0)
150 ibp->rvp.n_vl15_dropped++;
154 /* Silently drop packets which are too big. */
155 if (unlikely(wc.byte_len > qp->r_len)) {
156 qp->r_flags |= RVT_R_REUSE_SGE;
157 ibp->rvp.n_pkt_drops++;
161 if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
163 const struct ib_global_route *grd = rdma_ah_read_grh(ah_attr);
165 qib_make_grh(ibp, &grh, grd, 0, 0);
166 rvt_copy_sge(qp, &qp->r_sge, &grh,
167 sizeof(grh), true, false);
168 wc.wc_flags |= IB_WC_GRH;
170 rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
171 ssge.sg_list = swqe->sg_list + 1;
172 ssge.sge = *swqe->sg_list;
173 ssge.num_sge = swqe->wr.num_sge;
176 u32 len = rvt_get_sge_length(sge, length);
178 rvt_copy_sge(qp, &qp->r_sge, sge->vaddr, len, true, false);
181 sge->sge_length -= len;
182 if (sge->sge_length == 0) {
184 *sge = *ssge.sg_list++;
185 } else if (sge->length == 0 && sge->mr->lkey) {
186 if (++sge->n >= RVT_SEGSZ) {
187 if (++sge->m >= sge->mr->mapsz)
192 sge->mr->map[sge->m]->segs[sge->n].vaddr;
194 sge->mr->map[sge->m]->segs[sge->n].length;
198 rvt_put_ss(&qp->r_sge);
199 if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
201 wc.wr_id = qp->r_wr_id;
202 wc.status = IB_WC_SUCCESS;
203 wc.opcode = IB_WC_RECV;
205 wc.src_qp = sqp->ibqp.qp_num;
206 wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
207 rvt_get_swqe_pkey_index(swqe) : 0;
208 wc.slid = ppd->lid | (rdma_ah_get_path_bits(ah_attr) &
209 ((1 << ppd->lmc) - 1));
210 wc.sl = rdma_ah_get_sl(ah_attr);
211 wc.dlid_path_bits = rdma_ah_get_dlid(ah_attr) & ((1 << ppd->lmc) - 1);
212 wc.port_num = qp->port_num;
213 /* Signal completion event if the solicited bit is set. */
214 rvt_recv_cq(qp, &wc, swqe->wr.send_flags & IB_SEND_SOLICITED);
215 ibp->rvp.n_loop_pkts++;
217 spin_unlock_irqrestore(&qp->r_lock, flags);
223 * qib_make_ud_req - construct a UD request packet
225 * @flags: flags to modify and pass back to caller
227 * Assumes the s_lock is held.
229 * Return 1 if constructed; otherwise, return 0.
231 int qib_make_ud_req(struct rvt_qp *qp, unsigned long *flags)
233 struct qib_qp_priv *priv = qp->priv;
234 struct ib_other_headers *ohdr;
235 struct rdma_ah_attr *ah_attr;
236 struct qib_pportdata *ppd;
237 struct qib_ibport *ibp;
238 struct rvt_swqe *wqe;
247 if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK)) {
248 if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
250 /* We are in the error state, flush the work request. */
251 if (qp->s_last == READ_ONCE(qp->s_head))
253 /* If DMAs are in progress, we can't flush immediately. */
254 if (atomic_read(&priv->s_dma_busy)) {
255 qp->s_flags |= RVT_S_WAIT_DMA;
258 wqe = rvt_get_swqe_ptr(qp, qp->s_last);
259 rvt_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
263 /* see post_one_send() */
264 if (qp->s_cur == READ_ONCE(qp->s_head))
267 wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
268 next_cur = qp->s_cur + 1;
269 if (next_cur >= qp->s_size)
272 /* Construct the header. */
273 ibp = to_iport(qp->ibqp.device, qp->port_num);
274 ppd = ppd_from_ibp(ibp);
275 ah_attr = rvt_get_swqe_ah_attr(wqe);
276 if (rdma_ah_get_dlid(ah_attr) >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
277 if (rdma_ah_get_dlid(ah_attr) !=
278 be16_to_cpu(IB_LID_PERMISSIVE))
279 this_cpu_inc(ibp->pmastats->n_multicast_xmit);
281 this_cpu_inc(ibp->pmastats->n_unicast_xmit);
283 this_cpu_inc(ibp->pmastats->n_unicast_xmit);
284 lid = rdma_ah_get_dlid(ah_attr) & ~((1 << ppd->lmc) - 1);
285 if (unlikely(lid == ppd->lid)) {
286 unsigned long tflags = *flags;
288 * If DMAs are in progress, we can't generate
289 * a completion for the loopback packet since
290 * it would be out of order.
291 * XXX Instead of waiting, we could queue a
292 * zero length descriptor so we get a callback.
294 if (atomic_read(&priv->s_dma_busy)) {
295 qp->s_flags |= RVT_S_WAIT_DMA;
298 qp->s_cur = next_cur;
299 spin_unlock_irqrestore(&qp->s_lock, tflags);
300 qib_ud_loopback(qp, wqe);
301 spin_lock_irqsave(&qp->s_lock, tflags);
303 rvt_send_complete(qp, wqe, IB_WC_SUCCESS);
308 qp->s_cur = next_cur;
309 extra_bytes = -wqe->length & 3;
310 nwords = (wqe->length + extra_bytes) >> 2;
312 /* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
314 qp->s_cur_size = wqe->length;
315 qp->s_cur_sge = &qp->s_sge;
316 qp->s_srate = rdma_ah_get_static_rate(ah_attr);
318 qp->s_sge.sge = wqe->sg_list[0];
319 qp->s_sge.sg_list = wqe->sg_list + 1;
320 qp->s_sge.num_sge = wqe->wr.num_sge;
321 qp->s_sge.total_len = wqe->length;
323 if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
324 /* Header size in 32-bit words. */
325 qp->s_hdrwords += qib_make_grh(ibp, &priv->s_hdr->u.l.grh,
326 rdma_ah_read_grh(ah_attr),
327 qp->s_hdrwords, nwords);
329 ohdr = &priv->s_hdr->u.l.oth;
331 * Don't worry about sending to locally attached multicast
332 * QPs. It is unspecified by the spec. what happens.
335 /* Header size in 32-bit words. */
337 ohdr = &priv->s_hdr->u.oth;
339 if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
341 ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
342 bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
344 bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
345 lrh0 |= rdma_ah_get_sl(ah_attr) << 4;
346 if (qp->ibqp.qp_type == IB_QPT_SMI)
347 lrh0 |= 0xF000; /* Set VL (see ch. 13.5.3.1) */
349 lrh0 |= ibp->sl_to_vl[rdma_ah_get_sl(ah_attr)] << 12;
350 priv->s_hdr->lrh[0] = cpu_to_be16(lrh0);
351 priv->s_hdr->lrh[1] =
352 cpu_to_be16(rdma_ah_get_dlid(ah_attr)); /* DEST LID */
353 priv->s_hdr->lrh[2] =
354 cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
357 lid |= rdma_ah_get_path_bits(ah_attr) &
358 ((1 << ppd->lmc) - 1);
359 priv->s_hdr->lrh[3] = cpu_to_be16(lid);
361 priv->s_hdr->lrh[3] = IB_LID_PERMISSIVE;
362 if (wqe->wr.send_flags & IB_SEND_SOLICITED)
363 bth0 |= IB_BTH_SOLICITED;
364 bth0 |= extra_bytes << 20;
365 bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
366 qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
367 rvt_get_swqe_pkey_index(wqe) : qp->s_pkey_index);
368 ohdr->bth[0] = cpu_to_be32(bth0);
370 * Use the multicast QP if the destination LID is a multicast LID.
372 ohdr->bth[1] = rdma_ah_get_dlid(ah_attr) >=
373 be16_to_cpu(IB_MULTICAST_LID_BASE) &&
374 rdma_ah_get_dlid(ah_attr) != be16_to_cpu(IB_LID_PERMISSIVE) ?
375 cpu_to_be32(QIB_MULTICAST_QPN) :
376 cpu_to_be32(rvt_get_swqe_remote_qpn(wqe));
377 ohdr->bth[2] = cpu_to_be32(wqe->psn & QIB_PSN_MASK);
379 * Qkeys with the high order bit set mean use the
380 * qkey from the QP context instead of the WR (see 10.2.5).
383 cpu_to_be32((int)rvt_get_swqe_remote_qkey(wqe) < 0 ? qp->qkey :
384 rvt_get_swqe_remote_qkey(wqe));
385 ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
390 qp->s_flags &= ~RVT_S_BUSY;
394 static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
396 struct qib_pportdata *ppd = ppd_from_ibp(ibp);
397 struct qib_devdata *dd = ppd->dd;
398 unsigned ctxt = ppd->hw_pidx;
401 pkey &= 0x7fff; /* remove limited/full membership bit */
403 for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
404 if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
408 * Should not get here, this means hardware failed to validate pkeys.
409 * Punt and return index 0.
415 * qib_ud_rcv - receive an incoming UD packet
416 * @ibp: the port the packet came in on
417 * @hdr: the packet header
418 * @has_grh: true if the packet has a GRH
419 * @data: the packet data
420 * @tlen: the packet length
421 * @qp: the QP the packet came on
423 * This is called from qib_qp_rcv() to process an incoming UD packet
425 * Called at interrupt level.
427 void qib_ud_rcv(struct qib_ibport *ibp, struct ib_header *hdr,
428 int has_grh, void *data, u32 tlen, struct rvt_qp *qp)
430 struct ib_other_headers *ohdr;
442 hdrsize = 8 + 12 + 8; /* LRH + BTH + DETH */
444 ohdr = &hdr->u.l.oth;
445 hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
447 qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
448 src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK;
451 * Get the number of bytes the message was padded by
452 * and drop incomplete packets.
454 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
455 if (unlikely(tlen < (hdrsize + pad + 4)))
458 tlen -= hdrsize + pad + 4;
461 * Check that the permissive LID is only used on QP0
462 * and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
464 if (qp->ibqp.qp_num) {
465 if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
466 hdr->lrh[3] == IB_LID_PERMISSIVE))
468 if (qp->ibqp.qp_num > 1) {
471 pkey1 = be32_to_cpu(ohdr->bth[0]);
472 pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
473 if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
476 (be16_to_cpu(hdr->lrh[0]) >> 4) &
478 src_qp, qp->ibqp.qp_num,
479 hdr->lrh[3], hdr->lrh[1]);
483 if (unlikely(qkey != qp->qkey))
486 /* Drop invalid MAD packets (see 13.5.3.1). */
487 if (unlikely(qp->ibqp.qp_num == 1 &&
489 (be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
494 /* Drop invalid MAD packets (see 13.5.3.1). */
495 if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
497 smp = (struct ib_smp *) data;
498 if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
499 hdr->lrh[3] == IB_LID_PERMISSIVE) &&
500 smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
505 * The opcode is in the low byte when its in network order
506 * (top byte when in host order).
508 opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
509 if (qp->ibqp.qp_num > 1 &&
510 opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
511 wc.ex.imm_data = ohdr->u.ud.imm_data;
512 wc.wc_flags = IB_WC_WITH_IMM;
513 } else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
520 * A GRH is expected to precede the data even if not
521 * present on the wire.
523 wc.byte_len = tlen + sizeof(struct ib_grh);
526 * Get the next work request entry to find where to put the data.
528 if (qp->r_flags & RVT_R_REUSE_SGE)
529 qp->r_flags &= ~RVT_R_REUSE_SGE;
533 ret = rvt_get_rwqe(qp, false);
535 rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
539 if (qp->ibqp.qp_num == 0)
540 ibp->rvp.n_vl15_dropped++;
544 /* Silently drop packets which are too big. */
545 if (unlikely(wc.byte_len > qp->r_len)) {
546 qp->r_flags |= RVT_R_REUSE_SGE;
550 rvt_copy_sge(qp, &qp->r_sge, &hdr->u.l.grh,
551 sizeof(struct ib_grh), true, false);
552 wc.wc_flags |= IB_WC_GRH;
554 rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
555 rvt_copy_sge(qp, &qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh),
557 rvt_put_ss(&qp->r_sge);
558 if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
560 wc.wr_id = qp->r_wr_id;
561 wc.status = IB_WC_SUCCESS;
562 wc.opcode = IB_WC_RECV;
566 wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
567 qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
568 wc.slid = be16_to_cpu(hdr->lrh[3]);
569 wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
570 dlid = be16_to_cpu(hdr->lrh[1]);
572 * Save the LMC lower bits if the destination LID is a unicast LID.
574 wc.dlid_path_bits = dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE) ? 0 :
575 dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
576 wc.port_num = qp->port_num;
577 /* Signal completion event if the solicited bit is set. */
578 rvt_recv_cq(qp, &wc, ib_bth_is_solicited(ohdr));
582 ibp->rvp.n_pkt_drops++;