[linux-2.6-block.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 <rdma/ib_verbs.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 rvt_qp *sqp, struct rvt_swqe *swqe)
{
	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
	struct qib_devdata *dd = ppd->dd;
	struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
	struct rvt_qp *qp;
	struct rdma_ah_attr *ah_attr;
	unsigned long flags;
	struct rvt_sge_state ssge;
	struct rvt_sge *sge;
	struct ib_wc wc;
	u32 length;
	enum ib_qp_type sqptype, dqptype;

	rcu_read_lock();
	qp = rvt_lookup_qpn(rdi, &ibp->rvp, swqe->ud_wr.remote_qpn);
	if (!qp) {
		ibp->rvp.n_pkt_drops++;
		goto drop;
	}

	sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
			IB_QPT_UD : sqp->ibqp.qp_type;
	dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
			IB_QPT_UD : qp->ibqp.qp_type;

	if (dqptype != sqptype ||
	    !(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
		ibp->rvp.n_pkt_drops++;
		goto drop;
	}

	ah_attr = &ibah_to_rvtah(swqe->ud_wr.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 | (rdma_ah_get_path_bits(ah_attr) &
					  ((1 << ppd->lmc) - 1));
			qib_bad_pkey(ibp, pkey1,
				     rdma_ah_get_sl(ah_attr),
				     sqp->ibqp.qp_num, qp->ibqp.qp_num,
				     cpu_to_be16(lid),
				     cpu_to_be16(rdma_ah_get_dlid(ah_attr)));
			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->ud_wr.remote_qkey < 0 ?
			sqp->qkey : swqe->ud_wr.remote_qkey;
		if (unlikely(qkey != qp->qkey))
			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 & RVT_R_REUSE_SGE)
		qp->r_flags &= ~RVT_R_REUSE_SGE;
	else {
		int ret;

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

	if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
		struct ib_grh grh;
		const struct ib_global_route *grd = rdma_ah_read_grh(ah_attr);

		qib_make_grh(ibp, &grh, grd, 0, 0);
		qib_copy_sge(&qp->r_sge, &grh,
			     sizeof(grh), 1);
		wc.wc_flags |= IB_WC_GRH;
	} else
		rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
	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 >= RVT_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;
	}
	rvt_put_ss(&qp->r_sge);
	if (!test_and_clear_bit(RVT_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->ud_wr.pkey_index : 0;
	wc.slid = ppd->lid | (rdma_ah_get_path_bits(ah_attr) &
				((1 << ppd->lmc) - 1));
	wc.sl = rdma_ah_get_sl(ah_attr);
	wc.dlid_path_bits = rdma_ah_get_dlid(ah_attr) & ((1 << ppd->lmc) - 1);
	wc.port_num = qp->port_num;
	/* Signal completion event if the solicited bit is set. */
	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
		     swqe->wr.send_flags & IB_SEND_SOLICITED);
	ibp->rvp.n_loop_pkts++;
bail_unlock:
	spin_unlock_irqrestore(&qp->r_lock, flags);
drop:
	rcu_read_unlock();
}

/**
 * qib_make_ud_req - construct a UD request packet
 * @qp: the QP
 *
 * Assumes the s_lock is held.
 *
 * Return 1 if constructed; otherwise, return 0.
 */
int qib_make_ud_req(struct rvt_qp *qp, unsigned long *flags)
{
	struct qib_qp_priv *priv = qp->priv;
	struct ib_other_headers *ohdr;
	struct rdma_ah_attr *ah_attr;
	struct qib_pportdata *ppd;
	struct qib_ibport *ibp;
	struct rvt_swqe *wqe;
	u32 nwords;
	u32 extra_bytes;
	u32 bth0;
	u16 lrh0;
	u16 lid;
	int ret = 0;
	int next_cur;

	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK)) {
		if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
			goto bail;
		/* We are in the error state, flush the work request. */
		smp_read_barrier_depends(); /* see post_one_send */
		if (qp->s_last == READ_ONCE(qp->s_head))
			goto bail;
		/* If DMAs are in progress, we can't flush immediately. */
		if (atomic_read(&priv->s_dma_busy)) {
			qp->s_flags |= RVT_S_WAIT_DMA;
			goto bail;
		}
		wqe = rvt_get_swqe_ptr(qp, qp->s_last);
		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
		goto done;
	}

	/* see post_one_send() */
	smp_read_barrier_depends();
	if (qp->s_cur == READ_ONCE(qp->s_head))
		goto bail;

	wqe = rvt_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 = &ibah_to_rvtah(wqe->ud_wr.ah)->attr;
	if (rdma_ah_get_dlid(ah_attr) >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
		if (rdma_ah_get_dlid(ah_attr) !=
				be16_to_cpu(IB_LID_PERMISSIVE))
			this_cpu_inc(ibp->pmastats->n_multicast_xmit);
		else
			this_cpu_inc(ibp->pmastats->n_unicast_xmit);
	} else {
		this_cpu_inc(ibp->pmastats->n_unicast_xmit);
		lid = rdma_ah_get_dlid(ah_attr) & ~((1 << ppd->lmc) - 1);
		if (unlikely(lid == ppd->lid)) {
			unsigned long tflags = *flags;
			/*
			 * 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(&priv->s_dma_busy)) {
				qp->s_flags |= RVT_S_WAIT_DMA;
				goto bail;
			}
			qp->s_cur = next_cur;
			spin_unlock_irqrestore(&qp->s_lock, tflags);
			qib_ud_loopback(qp, wqe);
			spin_lock_irqsave(&qp->s_lock, tflags);
			*flags = tflags;
			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 = rdma_ah_get_static_rate(ah_attr);
	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 (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
		/* Header size in 32-bit words. */
		qp->s_hdrwords += qib_make_grh(ibp, &priv->s_hdr->u.l.grh,
					       rdma_ah_read_grh(ah_attr),
					       qp->s_hdrwords, nwords);
		lrh0 = QIB_LRH_GRH;
		ohdr = &priv->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 = &priv->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 |= rdma_ah_get_sl(ah_attr) << 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[rdma_ah_get_sl(ah_attr)] << 12;
	priv->s_hdr->lrh[0] = cpu_to_be16(lrh0);
	priv->s_hdr->lrh[1] =
			cpu_to_be16(rdma_ah_get_dlid(ah_attr));  /* DEST LID */
	priv->s_hdr->lrh[2] =
			cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
	lid = ppd->lid;
	if (lid) {
		lid |= rdma_ah_get_path_bits(ah_attr) &
			((1 << ppd->lmc) - 1);
		priv->s_hdr->lrh[3] = cpu_to_be16(lid);
	} else
		priv->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->ud_wr.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] = rdma_ah_get_dlid(ah_attr) >=
			be16_to_cpu(IB_MULTICAST_LID_BASE) &&
		rdma_ah_get_dlid(ah_attr) != be16_to_cpu(IB_LID_PERMISSIVE) ?
		cpu_to_be32(QIB_MULTICAST_QPN) :
		cpu_to_be32(wqe->ud_wr.remote_qpn);
	ohdr->bth[2] = cpu_to_be32(wqe->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->ud_wr.remote_qkey < 0 ?
					 qp->qkey : wqe->ud_wr.remote_qkey);
	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);

done:
	return 1;
bail:
	qp->s_flags &= ~RVT_S_BUSY;
	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 ib_header *hdr,
		int has_grh, void *data, u32 tlen, struct rvt_qp *qp)
{
	struct ib_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]) & RVT_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_pkey(ibp,
					     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))
			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 & RVT_R_REUSE_SGE)
		qp->r_flags &= ~RVT_R_REUSE_SGE;
	else {
		int ret;

		ret = qib_get_rwqe(qp, 0);
		if (ret < 0) {
			rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
			return;
		}
		if (!ret) {
			if (qp->ibqp.qp_num == 0)
				ibp->rvp.n_vl15_dropped++;
			return;
		}
	}
	/* Silently drop packets which are too big. */
	if (unlikely(wc.byte_len > qp->r_len)) {
		qp->r_flags |= RVT_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
		rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
	rvt_put_ss(&qp->r_sge);
	if (!test_and_clear_bit(RVT_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 >= be16_to_cpu(IB_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. */
	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
		     (ohdr->bth[0] &
			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
	return;

drop:
	ibp->rvp.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>
9ff198f5	35	#include <rdma/ib_verbs.h>
f931551b RC	36
	37	#include "qib.h"
	38	#include "qib_mad.h"
	39
	40	/**
	41	* qib_ud_loopback - handle send on loopback QPs
	42	* @sqp: the sending QP
	43	* @swqe: the send work request
	44	*
	45	* This is called from qib_make_ud_req() to forward a WQE addressed
	46	* to the same HCA.
	47	* Note that the receive interrupt handler may be calling qib_ud_rcv()
	48	* while this is being called.
	49	*/
7c2e11fe	50	static void qib_ud_loopback(struct rvt_qp sqp, struct rvt_swqe swqe)
f931551b RC	51	{
f931551b RC	52	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
1cefc2cd HC	53	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
	54	struct qib_devdata *dd = ppd->dd;
	55	struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
7c2e11fe	56	struct rvt_qp *qp;
90898850	57	struct rdma_ah_attr *ah_attr;
f931551b	58	unsigned long flags;
7c2e11fe DD	59	struct rvt_sge_state ssge;
7c2e11fe DD	60	struct rvt_sge *sge;
f931551b RC	61	struct ib_wc wc;
f931551b RC	62	u32 length;
6e0ea9e6	63	enum ib_qp_type sqptype, dqptype;
f931551b	64
1cefc2cd HC	65	rcu_read_lock();
1cefc2cd HC	66	qp = rvt_lookup_qpn(rdi, &ibp->rvp, swqe->ud_wr.remote_qpn);
f931551b	67	if (!qp) {
f24a6d48	68	ibp->rvp.n_pkt_drops++;
13d84914	69	goto drop;
f931551b	70	}
6e0ea9e6 IW	71
	72	sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
	73	IB_QPT_UD : sqp->ibqp.qp_type;
	74	dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
	75	IB_QPT_UD : qp->ibqp.qp_type;
	76
	77	if (dqptype != sqptype \|\|
db3ef0eb	78	!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
f24a6d48	79	ibp->rvp.n_pkt_drops++;
f931551b RC	80	goto drop;
	81	}
	82
96ab1ac1	83	ah_attr = &ibah_to_rvtah(swqe->ud_wr.ah)->attr;
f931551b RC	84	ppd = ppd_from_ibp(ibp);
	85
	86	if (qp->ibqp.qp_num > 1) {
	87	u16 pkey1;
	88	u16 pkey2;
	89	u16 lid;
	90
	91	pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
	92	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
	93	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
d8966fcd	94	lid = ppd->lid \| (rdma_ah_get_path_bits(ah_attr) &
f931551b	95	((1 << ppd->lmc) - 1));
13d84914 DD	96	qib_bad_pkey(ibp, pkey1,
	97	rdma_ah_get_sl(ah_attr),
	98	sqp->ibqp.qp_num, qp->ibqp.qp_num,
	99	cpu_to_be16(lid),
	100	cpu_to_be16(rdma_ah_get_dlid(ah_attr)));
f931551b RC	101	goto drop;
	102	}
	103	}
	104
	105	/*
	106	* Check that the qkey matches (except for QP0, see 9.6.1.4.1).
	107	* Qkeys with the high order bit set mean use the
	108	* qkey from the QP context instead of the WR (see 10.2.5).
	109	*/
	110	if (qp->ibqp.qp_num) {
	111	u32 qkey;
	112
e622f2f4 CH	113	qkey = (int)swqe->ud_wr.remote_qkey < 0 ?
e622f2f4 CH	114	sqp->qkey : swqe->ud_wr.remote_qkey;
13d84914	115	if (unlikely(qkey != qp->qkey))
f931551b	116	goto drop;
f931551b RC	117	}
	118
	119	/*
25985edc	120	* A GRH is expected to precede the data even if not
f931551b RC	121	* present on the wire.
	122	*/
	123	length = swqe->length;
041af0bb	124	memset(&wc, 0, sizeof(wc));
f931551b RC	125	wc.byte_len = length + sizeof(struct ib_grh);
	126
	127	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
	128	wc.wc_flags = IB_WC_WITH_IMM;
	129	wc.ex.imm_data = swqe->wr.ex.imm_data;
	130	}
	131
	132	spin_lock_irqsave(&qp->r_lock, flags);
	133
	134	/*
	135	* Get the next work request entry to find where to put the data.
	136	*/
01ba79d4 HC	137	if (qp->r_flags & RVT_R_REUSE_SGE)
01ba79d4 HC	138	qp->r_flags &= ~RVT_R_REUSE_SGE;
f931551b RC	139	else {
	140	int ret;
	141
	142	ret = qib_get_rwqe(qp, 0);
	143	if (ret < 0) {
beb5a042	144	rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
f931551b RC	145	goto bail_unlock;
	146	}
	147	if (!ret) {
	148	if (qp->ibqp.qp_num == 0)
f24a6d48	149	ibp->rvp.n_vl15_dropped++;
f931551b RC	150	goto bail_unlock;
	151	}
	152	}
	153	/* Silently drop packets which are too big. */
	154	if (unlikely(wc.byte_len > qp->r_len)) {
01ba79d4	155	qp->r_flags \|= RVT_R_REUSE_SGE;
f24a6d48	156	ibp->rvp.n_pkt_drops++;
f931551b RC	157	goto bail_unlock;
	158	}
	159
d8966fcd	160	if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
527dbf12	161	struct ib_grh grh;
d8966fcd	162	const struct ib_global_route *grd = rdma_ah_read_grh(ah_attr);
527dbf12	163
d8966fcd	164	qib_make_grh(ibp, &grh, grd, 0, 0);
527dbf12 DC	165	qib_copy_sge(&qp->r_sge, &grh,
527dbf12 DC	166	sizeof(grh), 1);
f931551b RC	167	wc.wc_flags \|= IB_WC_GRH;
f931551b RC	168	} else
3fc4a090	169	rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
f931551b RC	170	ssge.sg_list = swqe->sg_list + 1;
	171	ssge.sge = *swqe->sg_list;
	172	ssge.num_sge = swqe->wr.num_sge;
	173	sge = &ssge.sge;
	174	while (length) {
	175	u32 len = sge->length;
	176
	177	if (len > length)
	178	len = length;
	179	if (len > sge->sge_length)
	180	len = sge->sge_length;
	181	BUG_ON(len == 0);
	182	qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
	183	sge->vaddr += len;
	184	sge->length -= len;
	185	sge->sge_length -= len;
	186	if (sge->sge_length == 0) {
	187	if (--ssge.num_sge)
	188	sge = ssge.sg_list++;
	189	} else if (sge->length == 0 && sge->mr->lkey) {
7c2e11fe	190	if (++sge->n >= RVT_SEGSZ) {
f931551b RC	191	if (++sge->m >= sge->mr->mapsz)
	192	break;
	193	sge->n = 0;
	194	}
	195	sge->vaddr =
	196	sge->mr->map[sge->m]->segs[sge->n].vaddr;
	197	sge->length =
	198	sge->mr->map[sge->m]->segs[sge->n].length;
	199	}
	200	length -= len;
	201	}
70696ea7	202	rvt_put_ss(&qp->r_sge);
01ba79d4	203	if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
f931551b RC	204	goto bail_unlock;
	205	wc.wr_id = qp->r_wr_id;
	206	wc.status = IB_WC_SUCCESS;
	207	wc.opcode = IB_WC_RECV;
	208	wc.qp = &qp->ibqp;
	209	wc.src_qp = sqp->ibqp.qp_num;
	210	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
e622f2f4	211	swqe->ud_wr.pkey_index : 0;
d8966fcd DC	212	wc.slid = ppd->lid \| (rdma_ah_get_path_bits(ah_attr) &
	213	((1 << ppd->lmc) - 1));
	214	wc.sl = rdma_ah_get_sl(ah_attr);
	215	wc.dlid_path_bits = rdma_ah_get_dlid(ah_attr) & ((1 << ppd->lmc) - 1);
f931551b RC	216	wc.port_num = qp->port_num;
f931551b RC	217	/* Signal completion event if the solicited bit is set. */
4bb88e5f	218	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
f931551b	219	swqe->wr.send_flags & IB_SEND_SOLICITED);
f24a6d48	220	ibp->rvp.n_loop_pkts++;
f931551b RC	221	bail_unlock:
	222	spin_unlock_irqrestore(&qp->r_lock, flags);
	223	drop:
1cefc2cd	224	rcu_read_unlock();
f931551b RC	225	}
	226
	227	/**
	228	* qib_make_ud_req - construct a UD request packet
	229	* @qp: the QP
	230	*
46a80d62 MM	231	* Assumes the s_lock is held.
46a80d62 MM	232	*
f931551b RC	233	* Return 1 if constructed; otherwise, return 0.
f931551b RC	234	*/
747f4d7a	235	int qib_make_ud_req(struct rvt_qp qp, unsigned long flags)
f931551b	236	{
ffc26907	237	struct qib_qp_priv *priv = qp->priv;
261a4351	238	struct ib_other_headers *ohdr;
90898850	239	struct rdma_ah_attr *ah_attr;
f931551b RC	240	struct qib_pportdata *ppd;
f931551b RC	241	struct qib_ibport *ibp;
7c2e11fe	242	struct rvt_swqe *wqe;
f931551b RC	243	u32 nwords;
	244	u32 extra_bytes;
	245	u32 bth0;
	246	u16 lrh0;
	247	u16 lid;
	248	int ret = 0;
	249	int next_cur;
	250
db3ef0eb HC	251	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK)) {
db3ef0eb HC	252	if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
f931551b RC	253	goto bail;
f931551b RC	254	/* We are in the error state, flush the work request. */
46a80d62	255	smp_read_barrier_depends(); /* see post_one_send */
6aa7de05	256	if (qp->s_last == READ_ONCE(qp->s_head))
f931551b RC	257	goto bail;
f931551b RC	258	/* If DMAs are in progress, we can't flush immediately. */
ffc26907	259	if (atomic_read(&priv->s_dma_busy)) {
01ba79d4	260	qp->s_flags \|= RVT_S_WAIT_DMA;
f931551b RC	261	goto bail;
f931551b RC	262	}
db3ef0eb	263	wqe = rvt_get_swqe_ptr(qp, qp->s_last);
f931551b RC	264	qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
	265	goto done;
	266	}
	267
46a80d62 MM	268	/* see post_one_send() */
46a80d62 MM	269	smp_read_barrier_depends();
6aa7de05	270	if (qp->s_cur == READ_ONCE(qp->s_head))
f931551b RC	271	goto bail;
f931551b RC	272
db3ef0eb	273	wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
f931551b RC	274	next_cur = qp->s_cur + 1;
	275	if (next_cur >= qp->s_size)
	276	next_cur = 0;
	277
	278	/* Construct the header. */
	279	ibp = to_iport(qp->ibqp.device, qp->port_num);
	280	ppd = ppd_from_ibp(ibp);
96ab1ac1	281	ah_attr = &ibah_to_rvtah(wqe->ud_wr.ah)->attr;
d8966fcd DC	282	if (rdma_ah_get_dlid(ah_attr) >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
	283	if (rdma_ah_get_dlid(ah_attr) !=
	284	be16_to_cpu(IB_LID_PERMISSIVE))
7d7632ad	285	this_cpu_inc(ibp->pmastats->n_multicast_xmit);
f931551b	286	else
7d7632ad	287	this_cpu_inc(ibp->pmastats->n_unicast_xmit);
f931551b	288	} else {
7d7632ad	289	this_cpu_inc(ibp->pmastats->n_unicast_xmit);
d8966fcd	290	lid = rdma_ah_get_dlid(ah_attr) & ~((1 << ppd->lmc) - 1);
f931551b	291	if (unlikely(lid == ppd->lid)) {
747f4d7a	292	unsigned long tflags = *flags;
f931551b RC	293	/*
	294	* If DMAs are in progress, we can't generate
	295	* a completion for the loopback packet since
	296	* it would be out of order.
	297	* XXX Instead of waiting, we could queue a
	298	* zero length descriptor so we get a callback.
	299	*/
ffc26907	300	if (atomic_read(&priv->s_dma_busy)) {
01ba79d4	301	qp->s_flags \|= RVT_S_WAIT_DMA;
f931551b RC	302	goto bail;
	303	}
	304	qp->s_cur = next_cur;
747f4d7a	305	spin_unlock_irqrestore(&qp->s_lock, tflags);
f931551b	306	qib_ud_loopback(qp, wqe);
747f4d7a MM	307	spin_lock_irqsave(&qp->s_lock, tflags);
747f4d7a MM	308	*flags = tflags;
f931551b RC	309	qib_send_complete(qp, wqe, IB_WC_SUCCESS);
	310	goto done;
	311	}
	312	}
	313
	314	qp->s_cur = next_cur;
	315	extra_bytes = -wqe->length & 3;
	316	nwords = (wqe->length + extra_bytes) >> 2;
	317
	318	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
	319	qp->s_hdrwords = 7;
	320	qp->s_cur_size = wqe->length;
	321	qp->s_cur_sge = &qp->s_sge;
d8966fcd	322	qp->s_srate = rdma_ah_get_static_rate(ah_attr);
f931551b RC	323	qp->s_wqe = wqe;
	324	qp->s_sge.sge = wqe->sg_list[0];
	325	qp->s_sge.sg_list = wqe->sg_list + 1;
	326	qp->s_sge.num_sge = wqe->wr.num_sge;
	327	qp->s_sge.total_len = wqe->length;
	328
d8966fcd	329	if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
f931551b	330	/* Header size in 32-bit words. */
ffc26907	331	qp->s_hdrwords += qib_make_grh(ibp, &priv->s_hdr->u.l.grh,
d8966fcd	332	rdma_ah_read_grh(ah_attr),
f931551b RC	333	qp->s_hdrwords, nwords);
f931551b RC	334	lrh0 = QIB_LRH_GRH;
ffc26907	335	ohdr = &priv->s_hdr->u.l.oth;
f931551b RC	336	/*
	337	* Don't worry about sending to locally attached multicast
	338	* QPs. It is unspecified by the spec. what happens.
	339	*/
	340	} else {
	341	/* Header size in 32-bit words. */
	342	lrh0 = QIB_LRH_BTH;
ffc26907	343	ohdr = &priv->s_hdr->u.oth;
f931551b RC	344	}
	345	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
	346	qp->s_hdrwords++;
	347	ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
	348	bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
	349	} else
	350	bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
d8966fcd	351	lrh0 \|= rdma_ah_get_sl(ah_attr) << 4;
f931551b RC	352	if (qp->ibqp.qp_type == IB_QPT_SMI)
	353	lrh0 \|= 0xF000; /* Set VL (see ch. 13.5.3.1) */
	354	else
d8966fcd	355	lrh0 \|= ibp->sl_to_vl[rdma_ah_get_sl(ah_attr)] << 12;
ffc26907	356	priv->s_hdr->lrh[0] = cpu_to_be16(lrh0);
d8966fcd DC	357	priv->s_hdr->lrh[1] =
d8966fcd DC	358	cpu_to_be16(rdma_ah_get_dlid(ah_attr)); /* DEST LID */
ffc26907 DD	359	priv->s_hdr->lrh[2] =
ffc26907 DD	360	cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
f931551b RC	361	lid = ppd->lid;
f931551b RC	362	if (lid) {
d8966fcd DC	363	lid \|= rdma_ah_get_path_bits(ah_attr) &
d8966fcd DC	364	((1 << ppd->lmc) - 1);
ffc26907	365	priv->s_hdr->lrh[3] = cpu_to_be16(lid);
f931551b	366	} else
ffc26907	367	priv->s_hdr->lrh[3] = IB_LID_PERMISSIVE;
f931551b RC	368	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
	369	bth0 \|= IB_BTH_SOLICITED;
	370	bth0 \|= extra_bytes << 20;
	371	bth0 \|= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
	372	qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
e622f2f4	373	wqe->ud_wr.pkey_index : qp->s_pkey_index);
f931551b RC	374	ohdr->bth[0] = cpu_to_be32(bth0);
	375	/*
	376	* Use the multicast QP if the destination LID is a multicast LID.
	377	*/
d8966fcd DC	378	ohdr->bth[1] = rdma_ah_get_dlid(ah_attr) >=
	379	be16_to_cpu(IB_MULTICAST_LID_BASE) &&
	380	rdma_ah_get_dlid(ah_attr) != be16_to_cpu(IB_LID_PERMISSIVE) ?
f931551b	381	cpu_to_be32(QIB_MULTICAST_QPN) :
e622f2f4	382	cpu_to_be32(wqe->ud_wr.remote_qpn);
46a80d62	383	ohdr->bth[2] = cpu_to_be32(wqe->psn & QIB_PSN_MASK);
f931551b RC	384	/*
	385	* Qkeys with the high order bit set mean use the
	386	* qkey from the QP context instead of the WR (see 10.2.5).
	387	*/
e622f2f4 CH	388	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->ud_wr.remote_qkey < 0 ?
e622f2f4 CH	389	qp->qkey : wqe->ud_wr.remote_qkey);
f931551b RC	390	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
	391
	392	done:
46a80d62	393	return 1;
f931551b	394	bail:
01ba79d4	395	qp->s_flags &= ~RVT_S_BUSY;
f931551b RC	396	return ret;
	397	}
	398
	399	static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
	400	{
	401	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
	402	struct qib_devdata *dd = ppd->dd;
	403	unsigned ctxt = ppd->hw_pidx;
	404	unsigned i;
	405
	406	pkey &= 0x7fff; /* remove limited/full membership bit */
	407
	408	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
	409	if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
	410	return i;
	411
	412	/*
	413	* Should not get here, this means hardware failed to validate pkeys.
	414	* Punt and return index 0.
	415	*/
	416	return 0;
	417	}
	418
	419	/**
	420	* qib_ud_rcv - receive an incoming UD packet
	421	* @ibp: the port the packet came in on
	422	* @hdr: the packet header
	423	* @has_grh: true if the packet has a GRH
	424	* @data: the packet data
	425	* @tlen: the packet length
	426	* @qp: the QP the packet came on
	427	*
	428	* This is called from qib_qp_rcv() to process an incoming UD packet
	429	* for the given QP.
	430	* Called at interrupt level.
	431	*/
261a4351	432	void qib_ud_rcv(struct qib_ibport ibp, struct ib_header hdr,
7c2e11fe	433	int has_grh, void data, u32 tlen, struct rvt_qp qp)
f931551b	434	{
261a4351	435	struct ib_other_headers *ohdr;
f931551b RC	436	int opcode;
	437	u32 hdrsize;
	438	u32 pad;
	439	struct ib_wc wc;
	440	u32 qkey;
	441	u32 src_qp;
	442	u16 dlid;
	443
	444	/* Check for GRH */
	445	if (!has_grh) {
	446	ohdr = &hdr->u.oth;
	447	hdrsize = 8 + 12 + 8; /* LRH + BTH + DETH */
	448	} else {
	449	ohdr = &hdr->u.l.oth;
	450	hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
	451	}
	452	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
70696ea7	453	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK;
f931551b	454
057ae62f MM	455	/*
	456	* Get the number of bytes the message was padded by
	457	* and drop incomplete packets.
	458	*/
f931551b	459	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
057ae62f MM	460	if (unlikely(tlen < (hdrsize + pad + 4)))
	461	goto drop;
	462
f931551b RC	463	tlen -= hdrsize + pad + 4;
	464
	465	/*
	466	* Check that the permissive LID is only used on QP0
	467	* and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
	468	*/
	469	if (qp->ibqp.qp_num) {
	470	if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
057ae62f MM	471	hdr->lrh[3] == IB_LID_PERMISSIVE))
057ae62f MM	472	goto drop;
f931551b RC	473	if (qp->ibqp.qp_num > 1) {
	474	u16 pkey1, pkey2;
	475
	476	pkey1 = be32_to_cpu(ohdr->bth[0]);
	477	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
	478	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
13d84914 DD	479	qib_bad_pkey(ibp,
	480	pkey1,
	481	(be16_to_cpu(hdr->lrh[0]) >> 4) &
f931551b	482	0xF,
13d84914 DD	483	src_qp, qp->ibqp.qp_num,
13d84914 DD	484	hdr->lrh[3], hdr->lrh[1]);
057ae62f	485	return;
f931551b RC	486	}
f931551b RC	487	}
13d84914	488	if (unlikely(qkey != qp->qkey))
057ae62f	489	return;
13d84914	490
f931551b RC	491	/* Drop invalid MAD packets (see 13.5.3.1). */
	492	if (unlikely(qp->ibqp.qp_num == 1 &&
	493	(tlen != 256 \|\|
057ae62f MM	494	(be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
057ae62f MM	495	goto drop;
f931551b RC	496	} else {
	497	struct ib_smp *smp;
	498
	499	/* Drop invalid MAD packets (see 13.5.3.1). */
057ae62f MM	500	if (tlen != 256 \|\| (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
057ae62f MM	501	goto drop;
f931551b RC	502	smp = (struct ib_smp *) data;
	503	if ((hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
	504	hdr->lrh[3] == IB_LID_PERMISSIVE) &&
057ae62f MM	505	smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
057ae62f MM	506	goto drop;
f931551b RC	507	}
	508
	509	/*
	510	* The opcode is in the low byte when its in network order
	511	* (top byte when in host order).
	512	*/
	513	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	514	if (qp->ibqp.qp_num > 1 &&
	515	opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
	516	wc.ex.imm_data = ohdr->u.ud.imm_data;
	517	wc.wc_flags = IB_WC_WITH_IMM;
c7665e5a	518	tlen -= sizeof(u32);
f931551b RC	519	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
	520	wc.ex.imm_data = 0;
	521	wc.wc_flags = 0;
057ae62f MM	522	} else
057ae62f MM	523	goto drop;
f931551b RC	524
f931551b RC	525	/*
25985edc	526	* A GRH is expected to precede the data even if not
f931551b RC	527	* present on the wire.
	528	*/
	529	wc.byte_len = tlen + sizeof(struct ib_grh);
	530
f931551b RC	531	/*
	532	* Get the next work request entry to find where to put the data.
	533	*/
01ba79d4 HC	534	if (qp->r_flags & RVT_R_REUSE_SGE)
01ba79d4 HC	535	qp->r_flags &= ~RVT_R_REUSE_SGE;
f931551b RC	536	else {
	537	int ret;
	538
	539	ret = qib_get_rwqe(qp, 0);
	540	if (ret < 0) {
beb5a042	541	rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
a5210c12	542	return;
f931551b RC	543	}
	544	if (!ret) {
	545	if (qp->ibqp.qp_num == 0)
f24a6d48	546	ibp->rvp.n_vl15_dropped++;
a5210c12	547	return;
f931551b RC	548	}
	549	}
	550	/* Silently drop packets which are too big. */
	551	if (unlikely(wc.byte_len > qp->r_len)) {
01ba79d4	552	qp->r_flags \|= RVT_R_REUSE_SGE;
057ae62f	553	goto drop;
f931551b RC	554	}
	555	if (has_grh) {
	556	qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
	557	sizeof(struct ib_grh), 1);
	558	wc.wc_flags \|= IB_WC_GRH;
	559	} else
3fc4a090	560	rvt_skip_sge(&qp->r_sge, sizeof(struct ib_grh), true);
f931551b	561	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
70696ea7	562	rvt_put_ss(&qp->r_sge);
01ba79d4	563	if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
a5210c12	564	return;
f931551b RC	565	wc.wr_id = qp->r_wr_id;
	566	wc.status = IB_WC_SUCCESS;
	567	wc.opcode = IB_WC_RECV;
	568	wc.vendor_err = 0;
	569	wc.qp = &qp->ibqp;
	570	wc.src_qp = src_qp;
	571	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
	572	qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
	573	wc.slid = be16_to_cpu(hdr->lrh[3]);
	574	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
	575	dlid = be16_to_cpu(hdr->lrh[1]);
	576	/*
	577	* Save the LMC lower bits if the destination LID is a unicast LID.
	578	*/
9ff198f5	579	wc.dlid_path_bits = dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE) ? 0 :
f931551b RC	580	dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
	581	wc.port_num = qp->port_num;
	582	/* Signal completion event if the solicited bit is set. */
4bb88e5f	583	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
f931551b RC	584	(ohdr->bth[0] &
f931551b RC	585	cpu_to_be32(IB_BTH_SOLICITED)) != 0);
057ae62f MM	586	return;
	587
	588	drop:
f24a6d48	589	ibp->rvp.n_pkt_drops++;
f931551b	590	}