[linux.git] / drivers / infiniband / hw / qib / qib_ud.c

/*
 * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <rdma/ib_smi.h>

#include "qib.h"
#include "qib_mad.h"

/**
 * qib_ud_loopback - handle send on loopback QPs
 * @sqp: the sending QP
 * @swqe: the send work request
 *
 * This is called from qib_make_ud_req() to forward a WQE addressed
 * to the same HCA.
 * Note that the receive interrupt handler may be calling qib_ud_rcv()
 * while this is being called.
 */
static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
{
	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
	struct qib_pportdata *ppd;
	struct qib_qp *qp;
	struct ib_ah_attr *ah_attr;
	unsigned long flags;
	struct qib_sge_state ssge;
	struct qib_sge *sge;
	struct ib_wc wc;
	u32 length;

	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
	if (!qp) {
		ibp->n_pkt_drops++;
		return;
	}
	if (qp->ibqp.qp_type != sqp->ibqp.qp_type ||
	    !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
		ibp->n_pkt_drops++;
		goto drop;
	}

	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
	ppd = ppd_from_ibp(ibp);

	if (qp->ibqp.qp_num > 1) {
		u16 pkey1;
		u16 pkey2;
		u16 lid;

		pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
		pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
		if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
			lid = ppd->lid | (ah_attr->src_path_bits &
					  ((1 << ppd->lmc) - 1));
			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
				      ah_attr->sl,
				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
				      cpu_to_be16(lid),
				      cpu_to_be16(ah_attr->dlid));
			goto drop;
		}
	}

	/*
	 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
	 * Qkeys with the high order bit set mean use the
	 * qkey from the QP context instead of the WR (see 10.2.5).
	 */
	if (qp->ibqp.qp_num) {
		u32 qkey;

		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
		if (unlikely(qkey != qp->qkey)) {
			u16 lid;

			lid = ppd->lid | (ah_attr->src_path_bits &
					  ((1 << ppd->lmc) - 1));
			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
				      ah_attr->sl,
				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
				      cpu_to_be16(lid),
				      cpu_to_be16(ah_attr->dlid));
			goto drop;
		}
	}

	/*
	 * A GRH is expected to precede the data even if not
	 * present on the wire.
	 */
	length = swqe->length;
	memset(&wc, 0, sizeof wc);
	wc.byte_len = length + sizeof(struct ib_grh);

	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
		wc.wc_flags = IB_WC_WITH_IMM;
		wc.ex.imm_data = swqe->wr.ex.imm_data;
	}

	spin_lock_irqsave(&qp->r_lock, flags);

	/*
	 * Get the next work request entry to find where to put the data.
	 */
	if (qp->r_flags & QIB_R_REUSE_SGE)
		qp->r_flags &= ~QIB_R_REUSE_SGE;
	else {
		int ret;

		ret = qib_get_rwqe(qp, 0);
		if (ret < 0) {
			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
			goto bail_unlock;
		}
		if (!ret) {
			if (qp->ibqp.qp_num == 0)
				ibp->n_vl15_dropped++;
			goto bail_unlock;
		}
	}
	/* Silently drop packets which are too big. */
	if (unlikely(wc.byte_len > qp->r_len)) {
		qp->r_flags |= QIB_R_REUSE_SGE;
		ibp->n_pkt_drops++;
		goto bail_unlock;
	}

	if (ah_attr->ah_flags & IB_AH_GRH) {
		qib_copy_sge(&qp->r_sge, &ah_attr->grh,
			     sizeof(struct ib_grh), 1);
		wc.wc_flags |= IB_WC_GRH;
	} else
		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	ssge.sg_list = swqe->sg_list + 1;
	ssge.sge = *swqe->sg_list;
	ssge.num_sge = swqe->wr.num_sge;
	sge = &ssge.sge;
	while (length) {
		u32 len = sge->length;

		if (len > length)
			len = length;
		if (len > sge->sge_length)
			len = sge->sge_length;
		BUG_ON(len == 0);
		qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
		sge->vaddr += len;
		sge->length -= len;
		sge->sge_length -= len;
		if (sge->sge_length == 0) {
			if (--ssge.num_sge)
				*sge = *ssge.sg_list++;
		} else if (sge->length == 0 && sge->mr->lkey) {
			if (++sge->n >= QIB_SEGSZ) {
				if (++sge->m >= sge->mr->mapsz)
					break;
				sge->n = 0;
			}
			sge->vaddr =
				sge->mr->map[sge->m]->segs[sge->n].vaddr;
			sge->length =
				sge->mr->map[sge->m]->segs[sge->n].length;
		}
		length -= len;
	}
	while (qp->r_sge.num_sge) {
		atomic_dec(&qp->r_sge.sge.mr->refcount);
		if (--qp->r_sge.num_sge)
			qp->r_sge.sge = *qp->r_sge.sg_list++;
	}
	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
		goto bail_unlock;
	wc.wr_id = qp->r_wr_id;
	wc.status = IB_WC_SUCCESS;
	wc.opcode = IB_WC_RECV;
	wc.qp = &qp->ibqp;
	wc.src_qp = sqp->ibqp.qp_num;
	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
		swqe->wr.wr.ud.pkey_index : 0;
	wc.slid = ppd->lid | (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
	wc.sl = ah_attr->sl;
	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
	wc.port_num = qp->port_num;
	/* Signal completion event if the solicited bit is set. */
	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
		     swqe->wr.send_flags & IB_SEND_SOLICITED);
	ibp->n_loop_pkts++;
bail_unlock:
	spin_unlock_irqrestore(&qp->r_lock, flags);
drop:
	if (atomic_dec_and_test(&qp->refcount))
		wake_up(&qp->wait);
}

/**
 * qib_make_ud_req - construct a UD request packet
 * @qp: the QP
 *
 * Return 1 if constructed; otherwise, return 0.
 */
int qib_make_ud_req(struct qib_qp *qp)
{
	struct qib_other_headers *ohdr;
	struct ib_ah_attr *ah_attr;
	struct qib_pportdata *ppd;
	struct qib_ibport *ibp;
	struct qib_swqe *wqe;
	unsigned long flags;
	u32 nwords;
	u32 extra_bytes;
	u32 bth0;
	u16 lrh0;
	u16 lid;
	int ret = 0;
	int next_cur;

	spin_lock_irqsave(&qp->s_lock, flags);

	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
			goto bail;
		/* We are in the error state, flush the work request. */
		if (qp->s_last == qp->s_head)
			goto bail;
		/* If DMAs are in progress, we can't flush immediately. */
		if (atomic_read(&qp->s_dma_busy)) {
			qp->s_flags |= QIB_S_WAIT_DMA;
			goto bail;
		}
		wqe = get_swqe_ptr(qp, qp->s_last);
		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
		goto done;
	}

	if (qp->s_cur == qp->s_head)
		goto bail;

	wqe = get_swqe_ptr(qp, qp->s_cur);
	next_cur = qp->s_cur + 1;
	if (next_cur >= qp->s_size)
		next_cur = 0;

	/* Construct the header. */
	ibp = to_iport(qp->ibqp.device, qp->port_num);
	ppd = ppd_from_ibp(ibp);
	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
		if (ah_attr->dlid != QIB_PERMISSIVE_LID)
			ibp->n_multicast_xmit++;
		else
			ibp->n_unicast_xmit++;
	} else {
		ibp->n_unicast_xmit++;
		lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
		if (unlikely(lid == ppd->lid)) {
			/*
			 * If DMAs are in progress, we can't generate
			 * a completion for the loopback packet since
			 * it would be out of order.
			 * XXX Instead of waiting, we could queue a
			 * zero length descriptor so we get a callback.
			 */
			if (atomic_read(&qp->s_dma_busy)) {
				qp->s_flags |= QIB_S_WAIT_DMA;
				goto bail;
			}
			qp->s_cur = next_cur;
			spin_unlock_irqrestore(&qp->s_lock, flags);
			qib_ud_loopback(qp, wqe);
			spin_lock_irqsave(&qp->s_lock, flags);
			qib_send_complete(qp, wqe, IB_WC_SUCCESS);
			goto done;
		}
	}

	qp->s_cur = next_cur;
	extra_bytes = -wqe->length & 3;
	nwords = (wqe->length + extra_bytes) >> 2;

	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
	qp->s_hdrwords = 7;
	qp->s_cur_size = wqe->length;
	qp->s_cur_sge = &qp->s_sge;
	qp->s_srate = ah_attr->static_rate;
	qp->s_wqe = wqe;
	qp->s_sge.sge = wqe->sg_list[0];
	qp->s_sge.sg_list = wqe->sg_list + 1;
	qp->s_sge.num_sge = wqe->wr.num_sge;
	qp->s_sge.total_len = wqe->length;

	if (ah_attr->ah_flags & IB_AH_GRH) {
		/* Header size in 32-bit words. */
		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
					       &ah_attr->grh,
					       qp->s_hdrwords, nwords);
		lrh0 = QIB_LRH_GRH;
		ohdr = &qp->s_hdr.u.l.oth;
		/*
		 * Don't worry about sending to locally attached multicast
		 * QPs.  It is unspecified by the spec. what happens.
		 */
	} else {
		/* Header size in 32-bit words. */
		lrh0 = QIB_LRH_BTH;
		ohdr = &qp->s_hdr.u.oth;
	}
	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
		qp->s_hdrwords++;
		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
	} else
		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
	lrh0 |= ah_attr->sl << 4;
	if (qp->ibqp.qp_type == IB_QPT_SMI)
		lrh0 |= 0xF000; /* Set VL (see ch. 13.5.3.1) */
	else
		lrh0 |= ibp->sl_to_vl[ah_attr->sl] << 12;
	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid);  /* DEST LID */
	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
	lid = ppd->lid;
	if (lid) {
		lid |= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
		qp->s_hdr.lrh[3] = cpu_to_be16(lid);
	} else
		qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
		bth0 |= IB_BTH_SOLICITED;
	bth0 |= extra_bytes << 20;
	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
		qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
			     wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
	ohdr->bth[0] = cpu_to_be32(bth0);
	/*
	 * Use the multicast QP if the destination LID is a multicast LID.
	 */
	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
		ah_attr->dlid != QIB_PERMISSIVE_LID ?
		cpu_to_be32(QIB_MULTICAST_QPN) :
		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
	/*
	 * Qkeys with the high order bit set mean use the
	 * qkey from the QP context instead of the WR (see 10.2.5).
	 */
	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);

done:
	ret = 1;
	goto unlock;

bail:
	qp->s_flags &= ~QIB_S_BUSY;
unlock:
	spin_unlock_irqrestore(&qp->s_lock, flags);
	return ret;
}

static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
{
	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
	struct qib_devdata *dd = ppd->dd;
	unsigned ctxt = ppd->hw_pidx;
	unsigned i;

	pkey &= 0x7fff;	/* remove limited/full membership bit */

	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
		if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
			return i;

	/*
	 * Should not get here, this means hardware failed to validate pkeys.
	 * Punt and return index 0.
	 */
	return 0;
}

/**
 * qib_ud_rcv - receive an incoming UD packet
 * @ibp: the port the packet came in on
 * @hdr: the packet header
 * @has_grh: true if the packet has a GRH
 * @data: the packet data
 * @tlen: the packet length
 * @qp: the QP the packet came on
 *
 * This is called from qib_qp_rcv() to process an incoming UD packet
 * for the given QP.
 * Called at interrupt level.
 */
void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
{
	struct qib_other_headers *ohdr;
	int opcode;
	u32 hdrsize;
	u32 pad;
	struct ib_wc wc;
	u32 qkey;
	u32 src_qp;
	u16 dlid;

	/* Check for GRH */
	if (!has_grh) {
		ohdr = &hdr->u.oth;
		hdrsize = 8 + 12 + 8;   /* LRH + BTH + DETH */
	} else {
		ohdr = &hdr->u.l.oth;
		hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
	}
	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;

	/*
	 * Get the number of bytes the message was padded by
	 * and drop incomplete packets.
	 */
	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
	if (unlikely(tlen < (hdrsize + pad + 4)))
		goto drop;

	tlen -= hdrsize + pad + 4;

	/*
	 * Check that the permissive LID is only used on QP0
	 * and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
	 */
	if (qp->ibqp.qp_num) {
		if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
			     hdr->lrh[3] == IB_LID_PERMISSIVE))
			goto drop;
		if (qp->ibqp.qp_num > 1) {
			u16 pkey1, pkey2;

			pkey1 = be32_to_cpu(ohdr->bth[0]);
			pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
			if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
				qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
					      pkey1,
					      (be16_to_cpu(hdr->lrh[0]) >> 4) &
						0xF,
					      src_qp, qp->ibqp.qp_num,
					      hdr->lrh[3], hdr->lrh[1]);
				return;
			}
		}
		if (unlikely(qkey != qp->qkey)) {
			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
				      src_qp, qp->ibqp.qp_num,
				      hdr->lrh[3], hdr->lrh[1]);
			return;
		}
		/* Drop invalid MAD packets (see 13.5.3.1). */
		if (unlikely(qp->ibqp.qp_num == 1 &&
			     (tlen != 256 ||
			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
			goto drop;
	} else {
		struct ib_smp *smp;

		/* Drop invalid MAD packets (see 13.5.3.1). */
		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
			goto drop;
		smp = (struct ib_smp *) data;
		if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
		     hdr->lrh[3] == IB_LID_PERMISSIVE) &&
		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
			goto drop;
	}

	/*
	 * The opcode is in the low byte when its in network order
	 * (top byte when in host order).
	 */
	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	if (qp->ibqp.qp_num > 1 &&
	    opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
		wc.ex.imm_data = ohdr->u.ud.imm_data;
		wc.wc_flags = IB_WC_WITH_IMM;
		tlen -= sizeof(u32);
	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
		wc.ex.imm_data = 0;
		wc.wc_flags = 0;
	} else
		goto drop;

	/*
	 * A GRH is expected to precede the data even if not
	 * present on the wire.
	 */
	wc.byte_len = tlen + sizeof(struct ib_grh);

	/*
	 * Get the next work request entry to find where to put the data.
	 */
	if (qp->r_flags & QIB_R_REUSE_SGE)
		qp->r_flags &= ~QIB_R_REUSE_SGE;
	else {
		int ret;

		ret = qib_get_rwqe(qp, 0);
		if (ret < 0) {
			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
			return;
		}
		if (!ret) {
			if (qp->ibqp.qp_num == 0)
				ibp->n_vl15_dropped++;
			return;
		}
	}
	/* Silently drop packets which are too big. */
	if (unlikely(wc.byte_len > qp->r_len)) {
		qp->r_flags |= QIB_R_REUSE_SGE;
		goto drop;
	}
	if (has_grh) {
		qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
			     sizeof(struct ib_grh), 1);
		wc.wc_flags |= IB_WC_GRH;
	} else
		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
	while (qp->r_sge.num_sge) {
		atomic_dec(&qp->r_sge.sge.mr->refcount);
		if (--qp->r_sge.num_sge)
			qp->r_sge.sge = *qp->r_sge.sg_list++;
	}
	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
		return;
	wc.wr_id = qp->r_wr_id;
	wc.status = IB_WC_SUCCESS;
	wc.opcode = IB_WC_RECV;
	wc.vendor_err = 0;
	wc.qp = &qp->ibqp;
	wc.src_qp = src_qp;
	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
		qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
	wc.slid = be16_to_cpu(hdr->lrh[3]);
	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
	dlid = be16_to_cpu(hdr->lrh[1]);
	/*
	 * Save the LMC lower bits if the destination LID is a unicast LID.
	 */
	wc.dlid_path_bits = dlid >= QIB_MULTICAST_LID_BASE ? 0 :
		dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
	wc.port_num = qp->port_num;
	/* Signal completion event if the solicited bit is set. */
	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
		     (ohdr->bth[0] &
			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
	return;

drop:
	ibp->n_pkt_drops++;
}
Commit	Line	Data
f931551b RC	1	/*
	2	* Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
	3	* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
	4	*
	5	* This software is available to you under a choice of one of two
	6	* licenses. You may choose to be licensed under the terms of the GNU
	7	* General Public License (GPL) Version 2, available from the file
	8	* COPYING in the main directory of this source tree, or the
	9	* OpenIB.org BSD license below:
	10	*
	11	* Redistribution and use in source and binary forms, with or
	12	* without modification, are permitted provided that the following
	13	* conditions are met:
	14	*
	15	* - Redistributions of source code must retain the above
	16	* copyright notice, this list of conditions and the following
	17	* disclaimer.
	18	*
	19	* - Redistributions in binary form must reproduce the above
	20	* copyright notice, this list of conditions and the following
	21	* disclaimer in the documentation and/or other materials
	22	* provided with the distribution.
	23	*
	24	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	25	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	26	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	27	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	28	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	29	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	30	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	31	* SOFTWARE.
	32	*/
	33
	34	#include <rdma/ib_smi.h>
	35
	36	#include "qib.h"
	37	#include "qib_mad.h"
	38
	39	/**
	40	* qib_ud_loopback - handle send on loopback QPs
	41	* @sqp: the sending QP
	42	* @swqe: the send work request
	43	*
	44	* This is called from qib_make_ud_req() to forward a WQE addressed
	45	* to the same HCA.
	46	* Note that the receive interrupt handler may be calling qib_ud_rcv()
	47	* while this is being called.
	48	*/
	49	static void qib_ud_loopback(struct qib_qp sqp, struct qib_swqe swqe)
	50	{
	51	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
	52	struct qib_pportdata *ppd;
	53	struct qib_qp *qp;
	54	struct ib_ah_attr *ah_attr;
	55	unsigned long flags;
	56	struct qib_sge_state ssge;
	57	struct qib_sge *sge;
	58	struct ib_wc wc;
	59	u32 length;
	60
	61	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
	62	if (!qp) {
	63	ibp->n_pkt_drops++;
	64	return;
65	}
66	if (qp->ibqp.qp_type != sqp->ibqp.qp_type \|\|
67	!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
68	ibp->n_pkt_drops++;
69	goto drop;
70	}
71
72	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
73	ppd = ppd_from_ibp(ibp);
74
75	if (qp->ibqp.qp_num > 1) {
76	u16 pkey1;
77	u16 pkey2;
78	u16 lid;
79
80	pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
81	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
82	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
83	lid = ppd->lid \| (ah_attr->src_path_bits &
84	((1 << ppd->lmc) - 1));
85	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
86	ah_attr->sl,
87	sqp->ibqp.qp_num, qp->ibqp.qp_num,
88	cpu_to_be16(lid),
89	cpu_to_be16(ah_attr->dlid));
90	goto drop;
91	}
92	}
93
94	/*
95	* Check that the qkey matches (except for QP0, see 9.6.1.4.1).
96	* Qkeys with the high order bit set mean use the
97	* qkey from the QP context instead of the WR (see 10.2.5).
98	*/
99	if (qp->ibqp.qp_num) {
100	u32 qkey;
101
102	qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
103	sqp->qkey : swqe->wr.wr.ud.remote_qkey;
104	if (unlikely(qkey != qp->qkey)) {
105	u16 lid;
106
107	lid = ppd->lid \| (ah_attr->src_path_bits &
108	((1 << ppd->lmc) - 1));
109	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
110	ah_attr->sl,
111	sqp->ibqp.qp_num, qp->ibqp.qp_num,
112	cpu_to_be16(lid),
113	cpu_to_be16(ah_attr->dlid));
114	goto drop;
115	}
116	}
117
118	/*
25985edc	119	* A GRH is expected to precede the data even if not
f931551b RC	120	* present on the wire.
	121	*/
	122	length = swqe->length;
	123	memset(&wc, 0, sizeof wc);
	124	wc.byte_len = length + sizeof(struct ib_grh);
	125
	126	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
	127	wc.wc_flags = IB_WC_WITH_IMM;
	128	wc.ex.imm_data = swqe->wr.ex.imm_data;
	129	}
	130
	131	spin_lock_irqsave(&qp->r_lock, flags);
	132
	133	/*
	134	* Get the next work request entry to find where to put the data.
	135	*/
	136	if (qp->r_flags & QIB_R_REUSE_SGE)
	137	qp->r_flags &= ~QIB_R_REUSE_SGE;
	138	else {
	139	int ret;
	140
	141	ret = qib_get_rwqe(qp, 0);
	142	if (ret < 0) {
	143	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
	144	goto bail_unlock;
	145	}
	146	if (!ret) {
	147	if (qp->ibqp.qp_num == 0)
	148	ibp->n_vl15_dropped++;
	149	goto bail_unlock;
	150	}
	151	}
	152	/* Silently drop packets which are too big. */
	153	if (unlikely(wc.byte_len > qp->r_len)) {
	154	qp->r_flags \|= QIB_R_REUSE_SGE;
	155	ibp->n_pkt_drops++;
	156	goto bail_unlock;
	157	}
	158
	159	if (ah_attr->ah_flags & IB_AH_GRH) {
	160	qib_copy_sge(&qp->r_sge, &ah_attr->grh,
	161	sizeof(struct ib_grh), 1);
	162	wc.wc_flags \|= IB_WC_GRH;
	163	} else
	164	qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	165	ssge.sg_list = swqe->sg_list + 1;
	166	ssge.sge = *swqe->sg_list;
	167	ssge.num_sge = swqe->wr.num_sge;
	168	sge = &ssge.sge;
	169	while (length) {
	170	u32 len = sge->length;
	171
	172	if (len > length)
	173	len = length;
	174	if (len > sge->sge_length)
	175	len = sge->sge_length;
	176	BUG_ON(len == 0);
	177	qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
	178	sge->vaddr += len;
	179	sge->length -= len;
	180	sge->sge_length -= len;
	181	if (sge->sge_length == 0) {
	182	if (--ssge.num_sge)
	183	sge = ssge.sg_list++;
184	} else if (sge->length == 0 && sge->mr->lkey) {
185	if (++sge->n >= QIB_SEGSZ) {
186	if (++sge->m >= sge->mr->mapsz)
187	break;
188	sge->n = 0;
189	}
190	sge->vaddr =
191	sge->mr->map[sge->m]->segs[sge->n].vaddr;
192	sge->length =
193	sge->mr->map[sge->m]->segs[sge->n].length;
194	}
195	length -= len;
196	}
197	while (qp->r_sge.num_sge) {
198	atomic_dec(&qp->r_sge.sge.mr->refcount);
199	if (--qp->r_sge.num_sge)
200	qp->r_sge.sge = *qp->r_sge.sg_list++;
201	}
202	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
203	goto bail_unlock;
204	wc.wr_id = qp->r_wr_id;
205	wc.status = IB_WC_SUCCESS;
206	wc.opcode = IB_WC_RECV;
207	wc.qp = &qp->ibqp;
208	wc.src_qp = sqp->ibqp.qp_num;
209	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
210	swqe->wr.wr.ud.pkey_index : 0;
211	wc.slid = ppd->lid \| (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
212	wc.sl = ah_attr->sl;
213	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
214	wc.port_num = qp->port_num;
215	/* Signal completion event if the solicited bit is set. */
216	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
217	swqe->wr.send_flags & IB_SEND_SOLICITED);
218	ibp->n_loop_pkts++;
219	bail_unlock:
220	spin_unlock_irqrestore(&qp->r_lock, flags);
221	drop:
222	if (atomic_dec_and_test(&qp->refcount))
223	wake_up(&qp->wait);
224	}
225
226	/**
227	* qib_make_ud_req - construct a UD request packet
228	* @qp: the QP
229	*
230	* Return 1 if constructed; otherwise, return 0.
231	*/
232	int qib_make_ud_req(struct qib_qp *qp)
233	{
234	struct qib_other_headers *ohdr;
235	struct ib_ah_attr *ah_attr;
236	struct qib_pportdata *ppd;
237	struct qib_ibport *ibp;
238	struct qib_swqe *wqe;
239	unsigned long flags;
240	u32 nwords;
241	u32 extra_bytes;
242	u32 bth0;
243	u16 lrh0;
244	u16 lid;
245	int ret = 0;
246	int next_cur;
247
248	spin_lock_irqsave(&qp->s_lock, flags);
249
250	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
251	if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
252	goto bail;
253	/* We are in the error state, flush the work request. */
254	if (qp->s_last == qp->s_head)
255	goto bail;
256	/* If DMAs are in progress, we can't flush immediately. */
257	if (atomic_read(&qp->s_dma_busy)) {
258	qp->s_flags \|= QIB_S_WAIT_DMA;
259	goto bail;
260	}
261	wqe = get_swqe_ptr(qp, qp->s_last);
262	qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
263	goto done;
264	}
265
266	if (qp->s_cur == qp->s_head)
267	goto bail;
268
269	wqe = get_swqe_ptr(qp, qp->s_cur);
270	next_cur = qp->s_cur + 1;
271	if (next_cur >= qp->s_size)
272	next_cur = 0;
273
274	/* Construct the header. */
275	ibp = to_iport(qp->ibqp.device, qp->port_num);
276	ppd = ppd_from_ibp(ibp);
277	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
278	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
279	if (ah_attr->dlid != QIB_PERMISSIVE_LID)
280	ibp->n_multicast_xmit++;
281	else
282	ibp->n_unicast_xmit++;
283	} else {
284	ibp->n_unicast_xmit++;
285	lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
286	if (unlikely(lid == ppd->lid)) {
287	/*
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.
293	*/
294	if (atomic_read(&qp->s_dma_busy)) {
295	qp->s_flags \|= QIB_S_WAIT_DMA;
296	goto bail;
297	}
298	qp->s_cur = next_cur;
299	spin_unlock_irqrestore(&qp->s_lock, flags);
300	qib_ud_loopback(qp, wqe);
301	spin_lock_irqsave(&qp->s_lock, flags);
302	qib_send_complete(qp, wqe, IB_WC_SUCCESS);
303	goto done;
304	}
305	}
306
307	qp->s_cur = next_cur;
308	extra_bytes = -wqe->length & 3;
309	nwords = (wqe->length + extra_bytes) >> 2;
310
311	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
312	qp->s_hdrwords = 7;
313	qp->s_cur_size = wqe->length;
314	qp->s_cur_sge = &qp->s_sge;
315	qp->s_srate = ah_attr->static_rate;
316	qp->s_wqe = wqe;
317	qp->s_sge.sge = wqe->sg_list[0];
318	qp->s_sge.sg_list = wqe->sg_list + 1;
319	qp->s_sge.num_sge = wqe->wr.num_sge;
320	qp->s_sge.total_len = wqe->length;
321
322	if (ah_attr->ah_flags & IB_AH_GRH) {
323	/* Header size in 32-bit words. */
324	qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
325	&ah_attr->grh,
326	qp->s_hdrwords, nwords);
327	lrh0 = QIB_LRH_GRH;
328	ohdr = &qp->s_hdr.u.l.oth;
329	/*
330	* Don't worry about sending to locally attached multicast
331	* QPs. It is unspecified by the spec. what happens.
332	*/
333	} else {
334	/* Header size in 32-bit words. */
335	lrh0 = QIB_LRH_BTH;
336	ohdr = &qp->s_hdr.u.oth;
337	}
338	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
339	qp->s_hdrwords++;
340	ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
341	bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
342	} else
343	bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
344	lrh0 \|= ah_attr->sl << 4;
345	if (qp->ibqp.qp_type == IB_QPT_SMI)
346	lrh0 \|= 0xF000; /* Set VL (see ch. 13.5.3.1) */
347	else
348	lrh0 \|= ibp->sl_to_vl[ah_attr->sl] << 12;
349	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
350	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid); /* DEST LID */
351	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
352	lid = ppd->lid;
353	if (lid) {
354	lid \|= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
355	qp->s_hdr.lrh[3] = cpu_to_be16(lid);
356	} else
357	qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
358	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
359	bth0 \|= IB_BTH_SOLICITED;
360	bth0 \|= extra_bytes << 20;
361	bth0 \|= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
362	qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
363	wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
364	ohdr->bth[0] = cpu_to_be32(bth0);
365	/*
366	* Use the multicast QP if the destination LID is a multicast LID.
367	*/
368	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
369	ah_attr->dlid != QIB_PERMISSIVE_LID ?
370	cpu_to_be32(QIB_MULTICAST_QPN) :
371	cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
372	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
373	/*
374	* Qkeys with the high order bit set mean use the
375	* qkey from the QP context instead of the WR (see 10.2.5).
376	*/
377	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
378	qp->qkey : wqe->wr.wr.ud.remote_qkey);
379	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
380
381	done:
382	ret = 1;
383	goto unlock;
384
385	bail:
386	qp->s_flags &= ~QIB_S_BUSY;
387	unlock:
388	spin_unlock_irqrestore(&qp->s_lock, flags);
389	return ret;
390	}
391
392	static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
393	{
394	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
395	struct qib_devdata *dd = ppd->dd;
396	unsigned ctxt = ppd->hw_pidx;
397	unsigned i;
398
399	pkey &= 0x7fff; /* remove limited/full membership bit */
400
401	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
402	if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
403	return i;
404
405	/*
406	* Should not get here, this means hardware failed to validate pkeys.
407	* Punt and return index 0.
408	*/
409	return 0;
410	}
411
412	/**
413	* qib_ud_rcv - receive an incoming UD packet
414	* @ibp: the port the packet came in on
415	* @hdr: the packet header
416	* @has_grh: true if the packet has a GRH
417	* @data: the packet data
418	* @tlen: the packet length
419	* @qp: the QP the packet came on
420	*
421	* This is called from qib_qp_rcv() to process an incoming UD packet
422	* for the given QP.
423	* Called at interrupt level.
424	*/
425	void qib_ud_rcv(struct qib_ibport ibp, struct qib_ib_header hdr,
426	int has_grh, void data, u32 tlen, struct qib_qp qp)
427	{
428	struct qib_other_headers *ohdr;
429	int opcode;
430	u32 hdrsize;
431	u32 pad;
432	struct ib_wc wc;
433	u32 qkey;
434	u32 src_qp;
435	u16 dlid;
436
437	/* Check for GRH */
438	if (!has_grh) {
439	ohdr = &hdr->u.oth;
440	hdrsize = 8 + 12 + 8; /* LRH + BTH + DETH */
441	} else {
442	ohdr = &hdr->u.l.oth;
443	hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
444	}
445	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
446	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
447
057ae62f MM	448	/*
	449	* Get the number of bytes the message was padded by
	450	* and drop incomplete packets.
	451	*/
f931551b	452	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
057ae62f MM	453	if (unlikely(tlen < (hdrsize + pad + 4)))
	454	goto drop;
	455
f931551b RC	456	tlen -= hdrsize + pad + 4;
	457
	458	/*
	459	* Check that the permissive LID is only used on QP0
	460	* and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
	461	*/
	462	if (qp->ibqp.qp_num) {
	463	if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
057ae62f MM	464	hdr->lrh[3] == IB_LID_PERMISSIVE))
057ae62f MM	465	goto drop;
f931551b RC	466	if (qp->ibqp.qp_num > 1) {
	467	u16 pkey1, pkey2;
	468
	469	pkey1 = be32_to_cpu(ohdr->bth[0]);
	470	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
	471	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
	472	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
	473	pkey1,
	474	(be16_to_cpu(hdr->lrh[0]) >> 4) &
	475	0xF,
	476	src_qp, qp->ibqp.qp_num,
	477	hdr->lrh[3], hdr->lrh[1]);
057ae62f	478	return;
f931551b RC	479	}
	480	}
	481	if (unlikely(qkey != qp->qkey)) {
	482	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
	483	(be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
	484	src_qp, qp->ibqp.qp_num,
	485	hdr->lrh[3], hdr->lrh[1]);
057ae62f	486	return;
f931551b RC	487	}
	488	/* Drop invalid MAD packets (see 13.5.3.1). */
	489	if (unlikely(qp->ibqp.qp_num == 1 &&
	490	(tlen != 256 \|\|
057ae62f MM	491	(be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
057ae62f MM	492	goto drop;
f931551b RC	493	} else {
	494	struct ib_smp *smp;
	495
	496	/* Drop invalid MAD packets (see 13.5.3.1). */
057ae62f MM	497	if (tlen != 256 \|\| (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
057ae62f MM	498	goto drop;
f931551b RC	499	smp = (struct ib_smp *) data;
	500	if ((hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
	501	hdr->lrh[3] == IB_LID_PERMISSIVE) &&
057ae62f MM	502	smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
057ae62f MM	503	goto drop;
f931551b RC	504	}
	505
	506	/*
	507	* The opcode is in the low byte when its in network order
	508	* (top byte when in host order).
	509	*/
	510	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	511	if (qp->ibqp.qp_num > 1 &&
	512	opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
	513	wc.ex.imm_data = ohdr->u.ud.imm_data;
	514	wc.wc_flags = IB_WC_WITH_IMM;
c7665e5a	515	tlen -= sizeof(u32);
f931551b RC	516	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
	517	wc.ex.imm_data = 0;
	518	wc.wc_flags = 0;
057ae62f MM	519	} else
057ae62f MM	520	goto drop;
f931551b RC	521
f931551b RC	522	/*
25985edc	523	* A GRH is expected to precede the data even if not
f931551b RC	524	* present on the wire.
	525	*/
	526	wc.byte_len = tlen + sizeof(struct ib_grh);
	527
f931551b RC	528	/*
	529	* Get the next work request entry to find where to put the data.
	530	*/
	531	if (qp->r_flags & QIB_R_REUSE_SGE)
	532	qp->r_flags &= ~QIB_R_REUSE_SGE;
	533	else {
	534	int ret;
	535
	536	ret = qib_get_rwqe(qp, 0);
	537	if (ret < 0) {
	538	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
a5210c12	539	return;
f931551b RC	540	}
	541	if (!ret) {
	542	if (qp->ibqp.qp_num == 0)
	543	ibp->n_vl15_dropped++;
a5210c12	544	return;
f931551b RC	545	}
	546	}
	547	/* Silently drop packets which are too big. */
	548	if (unlikely(wc.byte_len > qp->r_len)) {
	549	qp->r_flags \|= QIB_R_REUSE_SGE;
057ae62f	550	goto drop;
f931551b RC	551	}
	552	if (has_grh) {
	553	qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
	554	sizeof(struct ib_grh), 1);
	555	wc.wc_flags \|= IB_WC_GRH;
	556	} else
	557	qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	558	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
	559	while (qp->r_sge.num_sge) {
	560	atomic_dec(&qp->r_sge.sge.mr->refcount);
	561	if (--qp->r_sge.num_sge)
	562	qp->r_sge.sge = *qp->r_sge.sg_list++;
	563	}
	564	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
a5210c12	565	return;
f931551b RC	566	wc.wr_id = qp->r_wr_id;
	567	wc.status = IB_WC_SUCCESS;
	568	wc.opcode = IB_WC_RECV;
	569	wc.vendor_err = 0;
	570	wc.qp = &qp->ibqp;
	571	wc.src_qp = src_qp;
	572	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
	573	qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
	574	wc.slid = be16_to_cpu(hdr->lrh[3]);
	575	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
	576	dlid = be16_to_cpu(hdr->lrh[1]);
	577	/*
	578	* Save the LMC lower bits if the destination LID is a unicast LID.
	579	*/
	580	wc.dlid_path_bits = dlid >= QIB_MULTICAST_LID_BASE ? 0 :
	581	dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
	582	wc.port_num = qp->port_num;
	583	/* Signal completion event if the solicited bit is set. */
	584	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
	585	(ohdr->bth[0] &
	586	cpu_to_be32(IB_BTH_SOLICITED)) != 0);
057ae62f MM	587	return;
	588
	589	drop:
	590	ibp->n_pkt_drops++;
f931551b	591	}