[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;
	struct rvt_qp *qp;
	struct ib_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;

	qp = qib_lookup_qpn(ibp, swqe->ud_wr.remote_qpn);
	if (!qp) {
		ibp->rvp.n_pkt_drops++;
		return;
	}

	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 | (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->ud_wr.remote_qkey < 0 ?
			sqp->qkey : swqe->ud_wr.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 & RVT_R_REUSE_SGE)
		qp->r_flags &= ~RVT_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->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 (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 >= 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;
	}
	qib_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 | (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. */
	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:
	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 rvt_qp *qp)
{
	struct qib_qp_priv *priv = qp->priv;
	struct qib_other_headers *ohdr;
	struct ib_ah_attr *ah_attr;
	struct qib_pportdata *ppd;
	struct qib_ibport *ibp;
	struct rvt_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_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. */
		if (qp->s_last == 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;
	}

	if (qp->s_cur == 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 (ah_attr->dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
		if (ah_attr->dlid != 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 = 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(&priv->s_dma_busy)) {
				qp->s_flags |= RVT_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, &priv->s_hdr->u.l.grh,
					       &ah_attr->grh,
					       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 |= 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;
	priv->s_hdr->lrh[0] = cpu_to_be16(lrh0);
	priv->s_hdr->lrh[1] = cpu_to_be16(ah_attr->dlid);  /* DEST LID */
	priv->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);
		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] = ah_attr->dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE) &&
		ah_attr->dlid != 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(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->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:
	ret = 1;
	goto unlock;

bail:
	qp->s_flags &= ~RVT_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 rvt_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 & RVT_R_REUSE_SGE)
		qp->r_flags &= ~RVT_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->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
		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);
	qib_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	{
	52	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
	53	struct qib_pportdata *ppd;
7c2e11fe	54	struct rvt_qp *qp;
f931551b RC	55	struct ib_ah_attr *ah_attr;
f931551b RC	56	unsigned long flags;
7c2e11fe DD	57	struct rvt_sge_state ssge;
7c2e11fe DD	58	struct rvt_sge *sge;
f931551b RC	59	struct ib_wc wc;
f931551b RC	60	u32 length;
6e0ea9e6	61	enum ib_qp_type sqptype, dqptype;
f931551b	62
e622f2f4	63	qp = qib_lookup_qpn(ibp, swqe->ud_wr.remote_qpn);
f931551b	64	if (!qp) {
f24a6d48	65	ibp->rvp.n_pkt_drops++;
f931551b RC	66	return;
f931551b RC	67	}
6e0ea9e6 IW	68
	69	sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
	70	IB_QPT_UD : sqp->ibqp.qp_type;
	71	dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
	72	IB_QPT_UD : qp->ibqp.qp_type;
	73
	74	if (dqptype != sqptype \|\|
db3ef0eb	75	!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
f24a6d48	76	ibp->rvp.n_pkt_drops++;
f931551b RC	77	goto drop;
	78	}
	79
96ab1ac1	80	ah_attr = &ibah_to_rvtah(swqe->ud_wr.ah)->attr;
f931551b RC	81	ppd = ppd_from_ibp(ibp);
	82
	83	if (qp->ibqp.qp_num > 1) {
	84	u16 pkey1;
	85	u16 pkey2;
	86	u16 lid;
	87
	88	pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
	89	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
	90	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
	91	lid = ppd->lid \| (ah_attr->src_path_bits &
	92	((1 << ppd->lmc) - 1));
	93	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
	94	ah_attr->sl,
	95	sqp->ibqp.qp_num, qp->ibqp.qp_num,
	96	cpu_to_be16(lid),
	97	cpu_to_be16(ah_attr->dlid));
	98	goto drop;
	99	}
	100	}
	101
	102	/*
	103	* Check that the qkey matches (except for QP0, see 9.6.1.4.1).
	104	* Qkeys with the high order bit set mean use the
	105	* qkey from the QP context instead of the WR (see 10.2.5).
	106	*/
	107	if (qp->ibqp.qp_num) {
	108	u32 qkey;
	109
e622f2f4 CH	110	qkey = (int)swqe->ud_wr.remote_qkey < 0 ?
e622f2f4 CH	111	sqp->qkey : swqe->ud_wr.remote_qkey;
f931551b RC	112	if (unlikely(qkey != qp->qkey)) {
	113	u16 lid;
	114
	115	lid = ppd->lid \| (ah_attr->src_path_bits &
	116	((1 << ppd->lmc) - 1));
	117	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
	118	ah_attr->sl,
	119	sqp->ibqp.qp_num, qp->ibqp.qp_num,
	120	cpu_to_be16(lid),
	121	cpu_to_be16(ah_attr->dlid));
	122	goto drop;
	123	}
	124	}
	125
	126	/*
25985edc	127	* A GRH is expected to precede the data even if not
f931551b RC	128	* present on the wire.
	129	*/
	130	length = swqe->length;
041af0bb	131	memset(&wc, 0, sizeof(wc));
f931551b RC	132	wc.byte_len = length + sizeof(struct ib_grh);
	133
	134	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
	135	wc.wc_flags = IB_WC_WITH_IMM;
	136	wc.ex.imm_data = swqe->wr.ex.imm_data;
	137	}
	138
	139	spin_lock_irqsave(&qp->r_lock, flags);
	140
	141	/*
	142	* Get the next work request entry to find where to put the data.
	143	*/
01ba79d4 HC	144	if (qp->r_flags & RVT_R_REUSE_SGE)
01ba79d4 HC	145	qp->r_flags &= ~RVT_R_REUSE_SGE;
f931551b RC	146	else {
	147	int ret;
	148
	149	ret = qib_get_rwqe(qp, 0);
	150	if (ret < 0) {
	151	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
	152	goto bail_unlock;
	153	}
	154	if (!ret) {
	155	if (qp->ibqp.qp_num == 0)
f24a6d48	156	ibp->rvp.n_vl15_dropped++;
f931551b RC	157	goto bail_unlock;
	158	}
	159	}
	160	/* Silently drop packets which are too big. */
	161	if (unlikely(wc.byte_len > qp->r_len)) {
01ba79d4	162	qp->r_flags \|= RVT_R_REUSE_SGE;
f24a6d48	163	ibp->rvp.n_pkt_drops++;
f931551b RC	164	goto bail_unlock;
	165	}
	166
	167	if (ah_attr->ah_flags & IB_AH_GRH) {
	168	qib_copy_sge(&qp->r_sge, &ah_attr->grh,
	169	sizeof(struct ib_grh), 1);
	170	wc.wc_flags \|= IB_WC_GRH;
	171	} else
	172	qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	173	ssge.sg_list = swqe->sg_list + 1;
	174	ssge.sge = *swqe->sg_list;
	175	ssge.num_sge = swqe->wr.num_sge;
	176	sge = &ssge.sge;
	177	while (length) {
	178	u32 len = sge->length;
	179
	180	if (len > length)
	181	len = length;
	182	if (len > sge->sge_length)
	183	len = sge->sge_length;
	184	BUG_ON(len == 0);
	185	qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
	186	sge->vaddr += len;
	187	sge->length -= len;
	188	sge->sge_length -= len;
	189	if (sge->sge_length == 0) {
	190	if (--ssge.num_sge)
	191	sge = ssge.sg_list++;
	192	} else if (sge->length == 0 && sge->mr->lkey) {
7c2e11fe	193	if (++sge->n >= RVT_SEGSZ) {
f931551b RC	194	if (++sge->m >= sge->mr->mapsz)
	195	break;
	196	sge->n = 0;
	197	}
	198	sge->vaddr =
	199	sge->mr->map[sge->m]->segs[sge->n].vaddr;
	200	sge->length =
	201	sge->mr->map[sge->m]->segs[sge->n].length;
	202	}
	203	length -= len;
	204	}
6a82649f	205	qib_put_ss(&qp->r_sge);
01ba79d4	206	if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
f931551b RC	207	goto bail_unlock;
	208	wc.wr_id = qp->r_wr_id;
	209	wc.status = IB_WC_SUCCESS;
	210	wc.opcode = IB_WC_RECV;
	211	wc.qp = &qp->ibqp;
	212	wc.src_qp = sqp->ibqp.qp_num;
	213	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
e622f2f4	214	swqe->ud_wr.pkey_index : 0;
f931551b RC	215	wc.slid = ppd->lid \| (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
	216	wc.sl = ah_attr->sl;
	217	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
	218	wc.port_num = qp->port_num;
	219	/* Signal completion event if the solicited bit is set. */
4bb88e5f	220	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
f931551b	221	swqe->wr.send_flags & IB_SEND_SOLICITED);
f24a6d48	222	ibp->rvp.n_loop_pkts++;
f931551b RC	223	bail_unlock:
	224	spin_unlock_irqrestore(&qp->r_lock, flags);
	225	drop:
	226	if (atomic_dec_and_test(&qp->refcount))
	227	wake_up(&qp->wait);
	228	}
	229
	230	/**
	231	* qib_make_ud_req - construct a UD request packet
	232	* @qp: the QP
	233	*
	234	* Return 1 if constructed; otherwise, return 0.
	235	*/
7c2e11fe	236	int qib_make_ud_req(struct rvt_qp *qp)
f931551b	237	{
ffc26907	238	struct qib_qp_priv *priv = qp->priv;
f931551b RC	239	struct qib_other_headers *ohdr;
	240	struct ib_ah_attr *ah_attr;
	241	struct qib_pportdata *ppd;
	242	struct qib_ibport *ibp;
7c2e11fe	243	struct rvt_swqe *wqe;
f931551b RC	244	unsigned long flags;
	245	u32 nwords;
	246	u32 extra_bytes;
	247	u32 bth0;
	248	u16 lrh0;
	249	u16 lid;
	250	int ret = 0;
	251	int next_cur;
	252
	253	spin_lock_irqsave(&qp->s_lock, flags);
	254
db3ef0eb HC	255	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK)) {
db3ef0eb HC	256	if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
f931551b RC	257	goto bail;
	258	/* We are in the error state, flush the work request. */
	259	if (qp->s_last == qp->s_head)
	260	goto bail;
	261	/* If DMAs are in progress, we can't flush immediately. */
ffc26907	262	if (atomic_read(&priv->s_dma_busy)) {
01ba79d4	263	qp->s_flags \|= RVT_S_WAIT_DMA;
f931551b RC	264	goto bail;
f931551b RC	265	}
db3ef0eb	266	wqe = rvt_get_swqe_ptr(qp, qp->s_last);
f931551b RC	267	qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
	268	goto done;
	269	}
	270
	271	if (qp->s_cur == qp->s_head)
	272	goto bail;
	273
db3ef0eb	274	wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
f931551b RC	275	next_cur = qp->s_cur + 1;
	276	if (next_cur >= qp->s_size)
	277	next_cur = 0;
	278
	279	/* Construct the header. */
	280	ibp = to_iport(qp->ibqp.device, qp->port_num);
	281	ppd = ppd_from_ibp(ibp);
96ab1ac1	282	ah_attr = &ibah_to_rvtah(wqe->ud_wr.ah)->attr;
9ff198f5 DD	283	if (ah_attr->dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
9ff198f5 DD	284	if (ah_attr->dlid != 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);
f931551b RC	290	lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
	291	if (unlikely(lid == ppd->lid)) {
	292	/*
	293	* If DMAs are in progress, we can't generate
	294	* a completion for the loopback packet since
	295	* it would be out of order.
	296	* XXX Instead of waiting, we could queue a
	297	* zero length descriptor so we get a callback.
	298	*/
ffc26907	299	if (atomic_read(&priv->s_dma_busy)) {
01ba79d4	300	qp->s_flags \|= RVT_S_WAIT_DMA;
f931551b RC	301	goto bail;
	302	}
	303	qp->s_cur = next_cur;
	304	spin_unlock_irqrestore(&qp->s_lock, flags);
	305	qib_ud_loopback(qp, wqe);
	306	spin_lock_irqsave(&qp->s_lock, flags);
	307	qib_send_complete(qp, wqe, IB_WC_SUCCESS);
	308	goto done;
	309	}
	310	}
	311
	312	qp->s_cur = next_cur;
	313	extra_bytes = -wqe->length & 3;
	314	nwords = (wqe->length + extra_bytes) >> 2;
	315
	316	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
	317	qp->s_hdrwords = 7;
	318	qp->s_cur_size = wqe->length;
	319	qp->s_cur_sge = &qp->s_sge;
	320	qp->s_srate = ah_attr->static_rate;
	321	qp->s_wqe = wqe;
	322	qp->s_sge.sge = wqe->sg_list[0];
	323	qp->s_sge.sg_list = wqe->sg_list + 1;
	324	qp->s_sge.num_sge = wqe->wr.num_sge;
	325	qp->s_sge.total_len = wqe->length;
	326
	327	if (ah_attr->ah_flags & IB_AH_GRH) {
	328	/* Header size in 32-bit words. */
ffc26907	329	qp->s_hdrwords += qib_make_grh(ibp, &priv->s_hdr->u.l.grh,
f931551b RC	330	&ah_attr->grh,
	331	qp->s_hdrwords, nwords);
	332	lrh0 = QIB_LRH_GRH;
ffc26907	333	ohdr = &priv->s_hdr->u.l.oth;
f931551b RC	334	/*
	335	* Don't worry about sending to locally attached multicast
	336	* QPs. It is unspecified by the spec. what happens.
	337	*/
	338	} else {
	339	/* Header size in 32-bit words. */
	340	lrh0 = QIB_LRH_BTH;
ffc26907	341	ohdr = &priv->s_hdr->u.oth;
f931551b RC	342	}
	343	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
	344	qp->s_hdrwords++;
	345	ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
	346	bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
	347	} else
	348	bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
	349	lrh0 \|= ah_attr->sl << 4;
	350	if (qp->ibqp.qp_type == IB_QPT_SMI)
	351	lrh0 \|= 0xF000; /* Set VL (see ch. 13.5.3.1) */
	352	else
	353	lrh0 \|= ibp->sl_to_vl[ah_attr->sl] << 12;
ffc26907 DD	354	priv->s_hdr->lrh[0] = cpu_to_be16(lrh0);
	355	priv->s_hdr->lrh[1] = cpu_to_be16(ah_attr->dlid); /* DEST LID */
	356	priv->s_hdr->lrh[2] =
	357	cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
f931551b RC	358	lid = ppd->lid;
	359	if (lid) {
	360	lid \|= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
ffc26907	361	priv->s_hdr->lrh[3] = cpu_to_be16(lid);
f931551b	362	} else
ffc26907	363	priv->s_hdr->lrh[3] = IB_LID_PERMISSIVE;
f931551b RC	364	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
	365	bth0 \|= IB_BTH_SOLICITED;
	366	bth0 \|= extra_bytes << 20;
	367	bth0 \|= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
	368	qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
e622f2f4	369	wqe->ud_wr.pkey_index : qp->s_pkey_index);
f931551b RC	370	ohdr->bth[0] = cpu_to_be32(bth0);
	371	/*
	372	* Use the multicast QP if the destination LID is a multicast LID.
	373	*/
9ff198f5 DD	374	ohdr->bth[1] = ah_attr->dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE) &&
9ff198f5 DD	375	ah_attr->dlid != be16_to_cpu(IB_LID_PERMISSIVE) ?
f931551b	376	cpu_to_be32(QIB_MULTICAST_QPN) :
e622f2f4	377	cpu_to_be32(wqe->ud_wr.remote_qpn);
f931551b RC	378	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
	379	/*
	380	* Qkeys with the high order bit set mean use the
	381	* qkey from the QP context instead of the WR (see 10.2.5).
	382	*/
e622f2f4 CH	383	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->ud_wr.remote_qkey < 0 ?
e622f2f4 CH	384	qp->qkey : wqe->ud_wr.remote_qkey);
f931551b RC	385	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
	386
	387	done:
	388	ret = 1;
	389	goto unlock;
	390
	391	bail:
01ba79d4	392	qp->s_flags &= ~RVT_S_BUSY;
f931551b RC	393	unlock:
	394	spin_unlock_irqrestore(&qp->s_lock, flags);
	395	return ret;
	396	}
	397
	398	static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
	399	{
	400	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
	401	struct qib_devdata *dd = ppd->dd;
	402	unsigned ctxt = ppd->hw_pidx;
	403	unsigned i;
	404
	405	pkey &= 0x7fff; /* remove limited/full membership bit */
	406
	407	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
	408	if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
	409	return i;
	410
	411	/*
	412	* Should not get here, this means hardware failed to validate pkeys.
	413	* Punt and return index 0.
	414	*/
	415	return 0;
	416	}
	417
	418	/**
	419	* qib_ud_rcv - receive an incoming UD packet
	420	* @ibp: the port the packet came in on
	421	* @hdr: the packet header
	422	* @has_grh: true if the packet has a GRH
	423	* @data: the packet data
	424	* @tlen: the packet length
	425	* @qp: the QP the packet came on
	426	*
	427	* This is called from qib_qp_rcv() to process an incoming UD packet
	428	* for the given QP.
	429	* Called at interrupt level.
	430	*/
	431	void qib_ud_rcv(struct qib_ibport ibp, struct qib_ib_header hdr,
7c2e11fe	432	int has_grh, void data, u32 tlen, struct rvt_qp qp)
f931551b RC	433	{
	434	struct qib_other_headers *ohdr;
	435	int opcode;
	436	u32 hdrsize;
	437	u32 pad;
	438	struct ib_wc wc;
	439	u32 qkey;
	440	u32 src_qp;
	441	u16 dlid;
	442
	443	/* Check for GRH */
	444	if (!has_grh) {
	445	ohdr = &hdr->u.oth;
	446	hdrsize = 8 + 12 + 8; /* LRH + BTH + DETH */
	447	} else {
	448	ohdr = &hdr->u.l.oth;
	449	hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
	450	}
	451	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
	452	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
	453
057ae62f MM	454	/*
	455	* Get the number of bytes the message was padded by
	456	* and drop incomplete packets.
	457	*/
f931551b	458	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
057ae62f MM	459	if (unlikely(tlen < (hdrsize + pad + 4)))
	460	goto drop;
	461
f931551b RC	462	tlen -= hdrsize + pad + 4;
	463
	464	/*
	465	* Check that the permissive LID is only used on QP0
	466	* and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
	467	*/
	468	if (qp->ibqp.qp_num) {
	469	if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
057ae62f MM	470	hdr->lrh[3] == IB_LID_PERMISSIVE))
057ae62f MM	471	goto drop;
f931551b RC	472	if (qp->ibqp.qp_num > 1) {
	473	u16 pkey1, pkey2;
	474
	475	pkey1 = be32_to_cpu(ohdr->bth[0]);
	476	pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
	477	if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
	478	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
	479	pkey1,
	480	(be16_to_cpu(hdr->lrh[0]) >> 4) &
	481	0xF,
	482	src_qp, qp->ibqp.qp_num,
	483	hdr->lrh[3], hdr->lrh[1]);
057ae62f	484	return;
f931551b RC	485	}
	486	}
	487	if (unlikely(qkey != qp->qkey)) {
	488	qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
	489	(be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
	490	src_qp, qp->ibqp.qp_num,
	491	hdr->lrh[3], hdr->lrh[1]);
057ae62f	492	return;
f931551b RC	493	}
	494	/* Drop invalid MAD packets (see 13.5.3.1). */
	495	if (unlikely(qp->ibqp.qp_num == 1 &&
	496	(tlen != 256 \|\|
057ae62f MM	497	(be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
057ae62f MM	498	goto drop;
f931551b RC	499	} else {
	500	struct ib_smp *smp;
	501
	502	/* Drop invalid MAD packets (see 13.5.3.1). */
057ae62f MM	503	if (tlen != 256 \|\| (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
057ae62f MM	504	goto drop;
f931551b RC	505	smp = (struct ib_smp *) data;
	506	if ((hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
	507	hdr->lrh[3] == IB_LID_PERMISSIVE) &&
057ae62f MM	508	smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
057ae62f MM	509	goto drop;
f931551b RC	510	}
	511
	512	/*
	513	* The opcode is in the low byte when its in network order
	514	* (top byte when in host order).
	515	*/
	516	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	517	if (qp->ibqp.qp_num > 1 &&
	518	opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
	519	wc.ex.imm_data = ohdr->u.ud.imm_data;
	520	wc.wc_flags = IB_WC_WITH_IMM;
c7665e5a	521	tlen -= sizeof(u32);
f931551b RC	522	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
	523	wc.ex.imm_data = 0;
	524	wc.wc_flags = 0;
057ae62f MM	525	} else
057ae62f MM	526	goto drop;
f931551b RC	527
f931551b RC	528	/*
25985edc	529	* A GRH is expected to precede the data even if not
f931551b RC	530	* present on the wire.
	531	*/
	532	wc.byte_len = tlen + sizeof(struct ib_grh);
	533
f931551b RC	534	/*
	535	* Get the next work request entry to find where to put the data.
	536	*/
01ba79d4 HC	537	if (qp->r_flags & RVT_R_REUSE_SGE)
01ba79d4 HC	538	qp->r_flags &= ~RVT_R_REUSE_SGE;
f931551b RC	539	else {
	540	int ret;
	541
	542	ret = qib_get_rwqe(qp, 0);
	543	if (ret < 0) {
	544	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
a5210c12	545	return;
f931551b RC	546	}
	547	if (!ret) {
	548	if (qp->ibqp.qp_num == 0)
f24a6d48	549	ibp->rvp.n_vl15_dropped++;
a5210c12	550	return;
f931551b RC	551	}
	552	}
	553	/* Silently drop packets which are too big. */
	554	if (unlikely(wc.byte_len > qp->r_len)) {
01ba79d4	555	qp->r_flags \|= RVT_R_REUSE_SGE;
057ae62f	556	goto drop;
f931551b RC	557	}
	558	if (has_grh) {
	559	qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
	560	sizeof(struct ib_grh), 1);
	561	wc.wc_flags \|= IB_WC_GRH;
	562	} else
	563	qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
	564	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
6a82649f	565	qib_put_ss(&qp->r_sge);
01ba79d4	566	if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
a5210c12	567	return;
f931551b RC	568	wc.wr_id = qp->r_wr_id;
	569	wc.status = IB_WC_SUCCESS;
	570	wc.opcode = IB_WC_RECV;
	571	wc.vendor_err = 0;
	572	wc.qp = &qp->ibqp;
	573	wc.src_qp = src_qp;
	574	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
	575	qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
	576	wc.slid = be16_to_cpu(hdr->lrh[3]);
	577	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
	578	dlid = be16_to_cpu(hdr->lrh[1]);
	579	/*
	580	* Save the LMC lower bits if the destination LID is a unicast LID.
	581	*/
9ff198f5	582	wc.dlid_path_bits = dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE) ? 0 :
f931551b RC	583	dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
	584	wc.port_num = qp->port_num;
	585	/* Signal completion event if the solicited bit is set. */
4bb88e5f	586	rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
f931551b RC	587	(ohdr->bth[0] &
f931551b RC	588	cpu_to_be32(IB_BTH_SOLICITED)) != 0);
057ae62f MM	589	return;
	590
	591	drop:
f24a6d48	592	ibp->rvp.n_pkt_drops++;
f931551b	593	}