[deliverable/linux.git] / drivers / infiniband / hw / ipath / ipath_ud.c

/*
 * Copyright (c) 2006, 2007, 2008 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 <linux/sched.h>
#include <rdma/ib_smi.h>

#include "ipath_verbs.h"
#include "ipath_kernel.h"

/**
 * ipath_ud_loopback - handle send on loopback QPs
 * @sqp: the sending QP
 * @swqe: the send work request
 *
 * This is called from ipath_make_ud_req() to forward a WQE addressed
 * to the same HCA.
 * Note that the receive interrupt handler may be calling ipath_ud_rcv()
 * while this is being called.
 */
static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
{
	struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
	struct ipath_qp *qp;
	struct ib_ah_attr *ah_attr;
	unsigned long flags;
	struct ipath_rq *rq;
	struct ipath_srq *srq;
	struct ipath_sge_state rsge;
	struct ipath_sge *sge;
	struct ipath_rwq *wq;
	struct ipath_rwqe *wqe;
	void (*handler)(struct ib_event *, void *);
	struct ib_wc wc;
	u32 tail;
	u32 rlen;
	u32 length;

	qp = ipath_lookup_qpn(&dev->qp_table, swqe->wr.wr.ud.remote_qpn);
	if (!qp || !(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
		dev->n_pkt_drops++;
		goto done;
	}

	/*
	 * 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 (unlikely(qp->ibqp.qp_num &&
		     ((int) swqe->wr.wr.ud.remote_qkey < 0 ?
		      sqp->qkey : swqe->wr.wr.ud.remote_qkey) != qp->qkey)) {
		/* XXX OK to lose a count once in a while. */
		dev->qkey_violations++;
		dev->n_pkt_drops++;
		goto drop;
	}

	/*
	 * A GRH is expected to preceed 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;
	}

	/*
	 * This would be a lot simpler if we could call ipath_get_rwqe()
	 * but that uses state that the receive interrupt handler uses
	 * so we would need to lock out receive interrupts while doing
	 * local loopback.
	 */
	if (qp->ibqp.srq) {
		srq = to_isrq(qp->ibqp.srq);
		handler = srq->ibsrq.event_handler;
		rq = &srq->rq;
	} else {
		srq = NULL;
		handler = NULL;
		rq = &qp->r_rq;
	}

	/*
	 * Get the next work request entry to find where to put the data.
	 * Note that it is safe to drop the lock after changing rq->tail
	 * since ipath_post_receive() won't fill the empty slot.
	 */
	spin_lock_irqsave(&rq->lock, flags);
	wq = rq->wq;
	tail = wq->tail;
	/* Validate tail before using it since it is user writable. */
	if (tail >= rq->size)
		tail = 0;
	if (unlikely(tail == wq->head)) {
		spin_unlock_irqrestore(&rq->lock, flags);
		dev->n_pkt_drops++;
		goto drop;
	}
	wqe = get_rwqe_ptr(rq, tail);
	rsge.sg_list = qp->r_ud_sg_list;
	if (!ipath_init_sge(qp, wqe, &rlen, &rsge)) {
		spin_unlock_irqrestore(&rq->lock, flags);
		dev->n_pkt_drops++;
		goto drop;
	}
	/* Silently drop packets which are too big. */
	if (wc.byte_len > rlen) {
		spin_unlock_irqrestore(&rq->lock, flags);
		dev->n_pkt_drops++;
		goto drop;
	}
	if (++tail >= rq->size)
		tail = 0;
	wq->tail = tail;
	wc.wr_id = wqe->wr_id;
	if (handler) {
		u32 n;

		/*
		 * validate head pointer value and compute
		 * the number of remaining WQEs.
		 */
		n = wq->head;
		if (n >= rq->size)
			n = 0;
		if (n < tail)
			n += rq->size - tail;
		else
			n -= tail;
		if (n < srq->limit) {
			struct ib_event ev;

			srq->limit = 0;
			spin_unlock_irqrestore(&rq->lock, flags);
			ev.device = qp->ibqp.device;
			ev.element.srq = qp->ibqp.srq;
			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
			handler(&ev, srq->ibsrq.srq_context);
		} else
			spin_unlock_irqrestore(&rq->lock, flags);
	} else
		spin_unlock_irqrestore(&rq->lock, flags);

	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
	if (ah_attr->ah_flags & IB_AH_GRH) {
		ipath_copy_sge(&rsge, &ah_attr->grh, sizeof(struct ib_grh));
		wc.wc_flags |= IB_WC_GRH;
	} else
		ipath_skip_sge(&rsge, sizeof(struct ib_grh));
	sge = swqe->sg_list;
	while (length) {
		u32 len = sge->length;

		if (len > length)
			len = length;
		if (len > sge->sge_length)
			len = sge->sge_length;
		BUG_ON(len == 0);
		ipath_copy_sge(&rsge, sge->vaddr, len);
		sge->vaddr += len;
		sge->length -= len;
		sge->sge_length -= len;
		if (sge->sge_length == 0) {
			if (--swqe->wr.num_sge)
				sge++;
		} else if (sge->length == 0 && sge->mr != NULL) {
			if (++sge->n >= IPATH_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;
	}
	wc.status = IB_WC_SUCCESS;
	wc.opcode = IB_WC_RECV;
	wc.qp = &qp->ibqp;
	wc.src_qp = sqp->ibqp.qp_num;
	/* XXX do we know which pkey matched? Only needed for GSI. */
	wc.pkey_index = 0;
	wc.slid = dev->dd->ipath_lid |
		(ah_attr->src_path_bits &
		 ((1 << dev->dd->ipath_lmc) - 1));
	wc.sl = ah_attr->sl;
	wc.dlid_path_bits =
		ah_attr->dlid & ((1 << dev->dd->ipath_lmc) - 1);
	wc.port_num = 1;
	/* Signal completion event if the solicited bit is set. */
	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
		       swqe->wr.send_flags & IB_SEND_SOLICITED);
drop:
	if (atomic_dec_and_test(&qp->refcount))
		wake_up(&qp->wait);
done:;
}

/**
 * ipath_make_ud_req - construct a UD request packet
 * @qp: the QP
 *
 * Return 1 if constructed; otherwise, return 0.
 */
int ipath_make_ud_req(struct ipath_qp *qp)
{
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	struct ipath_other_headers *ohdr;
	struct ib_ah_attr *ah_attr;
	struct ipath_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_ipath_state_ops[qp->state] & IPATH_PROCESS_NEXT_SEND_OK)) {
		if (!(ib_ipath_state_ops[qp->state] & IPATH_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 |= IPATH_S_WAIT_DMA;
			goto bail;
		}
		wqe = get_swqe_ptr(qp, qp->s_last);
		ipath_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. */
	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
	if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE) {
		if (ah_attr->dlid != IPATH_PERMISSIVE_LID)
			dev->n_multicast_xmit++;
		else
			dev->n_unicast_xmit++;
	} else {
		dev->n_unicast_xmit++;
		lid = ah_attr->dlid & ~((1 << dev->dd->ipath_lmc) - 1);
		if (unlikely(lid == dev->dd->ipath_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 |= IPATH_S_WAIT_DMA;
				goto bail;
			}
			qp->s_cur = next_cur;
			spin_unlock_irqrestore(&qp->s_lock, flags);
			ipath_ud_loopback(qp, wqe);
			spin_lock_irqsave(&qp->s_lock, flags);
			ipath_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_dmult = 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;

	if (ah_attr->ah_flags & IB_AH_GRH) {
		/* Header size in 32-bit words. */
		qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
						 &ah_attr->grh,
						 qp->s_hdrwords, nwords);
		lrh0 = IPATH_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 = IPATH_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) */
	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 = dev->dd->ipath_lid;
	if (lid) {
		lid |= ah_attr->src_path_bits &
			((1 << dev->dd->ipath_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 |= 1 << 23;
	bth0 |= extra_bytes << 20;
	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY :
		ipath_get_pkey(dev->dd, 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 >= IPATH_MULTICAST_LID_BASE &&
		ah_attr->dlid != IPATH_PERMISSIVE_LID ?
		cpu_to_be32(IPATH_MULTICAST_QPN) :
		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPATH_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 &= ~IPATH_S_BUSY;
unlock:
	spin_unlock_irqrestore(&qp->s_lock, flags);
	return ret;
}

/**
 * ipath_ud_rcv - receive an incoming UD packet
 * @dev: the device 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 ipath_qp_rcv() to process an incoming UD packet
 * for the given QP.
 * Called at interrupt level.
 */
void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
		  int has_grh, void *data, u32 tlen, struct ipath_qp *qp)
{
	struct ipath_other_headers *ohdr;
	int opcode;
	u32 hdrsize;
	u32 pad;
	struct ib_wc wc;
	u32 qkey;
	u32 src_qp;
	u16 dlid;
	int header_in_data;

	/* Check for GRH */
	if (!has_grh) {
		ohdr = &hdr->u.oth;
		hdrsize = 8 + 12 + 8;	/* LRH + BTH + DETH */
		qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
		src_qp = be32_to_cpu(ohdr->u.ud.deth[1]);
		header_in_data = 0;
	} else {
		ohdr = &hdr->u.l.oth;
		hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
		/*
		 * The header with GRH is 68 bytes and the core driver sets
		 * the eager header buffer size to 56 bytes so the last 12
		 * bytes of the IB header is in the data buffer.
		 */
		header_in_data = dev->dd->ipath_rcvhdrentsize == 16;
		if (header_in_data) {
			qkey = be32_to_cpu(((__be32 *) data)[1]);
			src_qp = be32_to_cpu(((__be32 *) data)[2]);
			data += 12;
		} else {
			qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
			src_qp = be32_to_cpu(ohdr->u.ud.deth[1]);
		}
	}
	src_qp &= IPATH_QPN_MASK;

	/*
	 * 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)) {
			dev->n_pkt_drops++;
			goto bail;
		}
		if (unlikely(qkey != qp->qkey)) {
			/* XXX OK to lose a count once in a while. */
			dev->qkey_violations++;
			dev->n_pkt_drops++;
			goto bail;
		}
	} else if (hdr->lrh[1] == IB_LID_PERMISSIVE ||
		   hdr->lrh[3] == IB_LID_PERMISSIVE) {
		struct ib_smp *smp = (struct ib_smp *) data;

		if (smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
			dev->n_pkt_drops++;
			goto bail;
		}
	}

	/*
	 * 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) {
		if (header_in_data) {
			wc.ex.imm_data = *(__be32 *) data;
			data += sizeof(__be32);
		} else
			wc.ex.imm_data = ohdr->u.ud.imm_data;
		wc.wc_flags = IB_WC_WITH_IMM;
		hdrsize += sizeof(u32);
	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
		wc.ex.imm_data = 0;
		wc.wc_flags = 0;
	} else {
		dev->n_pkt_drops++;
		goto bail;
	}

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

	/* Drop invalid MAD packets (see 13.5.3.1). */
	if (unlikely((qp->ibqp.qp_num == 0 &&
		      (tlen != 256 ||
		       (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)) ||
		     (qp->ibqp.qp_num == 1 &&
		      (tlen != 256 ||
		       (be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))) {
		dev->n_pkt_drops++;
		goto bail;
	}

	/*
	 * A GRH is expected to preceed 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 & IPATH_R_REUSE_SGE)
		qp->r_flags &= ~IPATH_R_REUSE_SGE;
	else if (!ipath_get_rwqe(qp, 0)) {
		/*
		 * Count VL15 packets dropped due to no receive buffer.
		 * Otherwise, count them as buffer overruns since usually,
		 * the HW will be able to receive packets even if there are
		 * no QPs with posted receive buffers.
		 */
		if (qp->ibqp.qp_num == 0)
			dev->n_vl15_dropped++;
		else
			dev->rcv_errors++;
		goto bail;
	}
	/* Silently drop packets which are too big. */
	if (wc.byte_len > qp->r_len) {
		qp->r_flags |= IPATH_R_REUSE_SGE;
		dev->n_pkt_drops++;
		goto bail;
	}
	if (has_grh) {
		ipath_copy_sge(&qp->r_sge, &hdr->u.l.grh,
			       sizeof(struct ib_grh));
		wc.wc_flags |= IB_WC_GRH;
	} else
		ipath_skip_sge(&qp->r_sge, sizeof(struct ib_grh));
	ipath_copy_sge(&qp->r_sge, data,
		       wc.byte_len - sizeof(struct ib_grh));
	if (!test_and_clear_bit(IPATH_R_WRID_VALID, &qp->r_aflags))
		goto bail;
	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;
	/* XXX do we know which pkey matched? Only needed for GSI. */
	wc.pkey_index = 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 >= IPATH_MULTICAST_LID_BASE ? 0 :
		dlid & ((1 << dev->dd->ipath_lmc) - 1);
	wc.port_num = 1;
	/* Signal completion event if the solicited bit is set. */
	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
		       (ohdr->bth[0] &
			cpu_to_be32(1 << 23)) != 0);

bail:;
}
Commit	Line	Data
74ed6b5e	1	/*
e7eacd36	2	* Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
74ed6b5e BS	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
d43c36dc	34	#include <linux/sched.h>
74ed6b5e BS	35	#include <rdma/ib_smi.h>
	36
	37	#include "ipath_verbs.h"
373d9915 RC	38	#include "ipath_kernel.h"
373d9915 RC	39
74ed6b5e BS	40	/**
74ed6b5e BS	41	* ipath_ud_loopback - handle send on loopback QPs
4ee97180 RC	42	* @sqp: the sending QP
4ee97180 RC	43	* @swqe: the send work request
74ed6b5e	44	*
4ee97180	45	* This is called from ipath_make_ud_req() to forward a WQE addressed
74ed6b5e	46	* to the same HCA.
373d9915 RC	47	* Note that the receive interrupt handler may be calling ipath_ud_rcv()
373d9915 RC	48	* while this is being called.
74ed6b5e	49	*/
4ee97180	50	static void ipath_ud_loopback(struct ipath_qp sqp, struct ipath_swqe swqe)
74ed6b5e BS	51	{
	52	struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
	53	struct ipath_qp *qp;
	54	struct ib_ah_attr *ah_attr;
	55	unsigned long flags;
	56	struct ipath_rq *rq;
	57	struct ipath_srq *srq;
	58	struct ipath_sge_state rsge;
	59	struct ipath_sge *sge;
373d9915	60	struct ipath_rwq *wq;
74ed6b5e	61	struct ipath_rwqe *wqe;
373d9915	62	void (handler)(struct ib_event , void *);
4ee97180	63	struct ib_wc wc;
373d9915 RC	64	u32 tail;
373d9915 RC	65	u32 rlen;
4ee97180	66	u32 length;
74ed6b5e	67
4ee97180	68	qp = ipath_lookup_qpn(&dev->qp_table, swqe->wr.wr.ud.remote_qpn);
e509be89	69	if (!qp \|\| !(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
4ee97180	70	dev->n_pkt_drops++;
e509be89	71	goto done;
4ee97180 RC	72	}
4ee97180 RC	73
74ed6b5e BS	74	/*
	75	* Check that the qkey matches (except for QP0, see 9.6.1.4.1).
	76	* Qkeys with the high order bit set mean use the
	77	* qkey from the QP context instead of the WR (see 10.2.5).
	78	*/
	79	if (unlikely(qp->ibqp.qp_num &&
4ee97180 RC	80	((int) swqe->wr.wr.ud.remote_qkey < 0 ?
4ee97180 RC	81	sqp->qkey : swqe->wr.wr.ud.remote_qkey) != qp->qkey)) {
74ed6b5e BS	82	/* XXX OK to lose a count once in a while. */
	83	dev->qkey_violations++;
	84	dev->n_pkt_drops++;
4ee97180	85	goto drop;
74ed6b5e BS	86	}
	87
	88	/*
	89	* A GRH is expected to preceed the data even if not
	90	* present on the wire.
	91	*/
4ee97180	92	length = swqe->length;
e509be89	93	memset(&wc, 0, sizeof wc);
4ee97180	94	wc.byte_len = length + sizeof(struct ib_grh);
74ed6b5e	95
4ee97180 RC	96	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
4ee97180 RC	97	wc.wc_flags = IB_WC_WITH_IMM;
00f7ec36	98	wc.ex.imm_data = swqe->wr.ex.imm_data;
74ed6b5e BS	99	}
	100
	101	/*
4ee97180 RC	102	* This would be a lot simpler if we could call ipath_get_rwqe()
	103	* but that uses state that the receive interrupt handler uses
	104	* so we would need to lock out receive interrupts while doing
	105	* local loopback.
74ed6b5e BS	106	*/
	107	if (qp->ibqp.srq) {
	108	srq = to_isrq(qp->ibqp.srq);
373d9915	109	handler = srq->ibsrq.event_handler;
74ed6b5e BS	110	rq = &srq->rq;
	111	} else {
	112	srq = NULL;
373d9915	113	handler = NULL;
74ed6b5e BS	114	rq = &qp->r_rq;
74ed6b5e BS	115	}
373d9915	116
4ee97180 RC	117	/*
	118	* Get the next work request entry to find where to put the data.
	119	* Note that it is safe to drop the lock after changing rq->tail
	120	* since ipath_post_receive() won't fill the empty slot.
	121	*/
74ed6b5e	122	spin_lock_irqsave(&rq->lock, flags);
373d9915 RC	123	wq = rq->wq;
373d9915 RC	124	tail = wq->tail;
4ee97180 RC	125	/* Validate tail before using it since it is user writable. */
	126	if (tail >= rq->size)
	127	tail = 0;
	128	if (unlikely(tail == wq->head)) {
	129	spin_unlock_irqrestore(&rq->lock, flags);
	130	dev->n_pkt_drops++;
	131	goto drop;
	132	}
	133	wqe = get_rwqe_ptr(rq, tail);
7c37d744	134	rsge.sg_list = qp->r_ud_sg_list;
4ee97180 RC	135	if (!ipath_init_sge(qp, wqe, &rlen, &rsge)) {
	136	spin_unlock_irqrestore(&rq->lock, flags);
	137	dev->n_pkt_drops++;
	138	goto drop;
74ed6b5e BS	139	}
74ed6b5e BS	140	/* Silently drop packets which are too big. */
4ee97180	141	if (wc.byte_len > rlen) {
74ed6b5e BS	142	spin_unlock_irqrestore(&rq->lock, flags);
74ed6b5e BS	143	dev->n_pkt_drops++;
4ee97180	144	goto drop;
74ed6b5e	145	}
4ee97180 RC	146	if (++tail >= rq->size)
4ee97180 RC	147	tail = 0;
373d9915	148	wq->tail = tail;
4ee97180	149	wc.wr_id = wqe->wr_id;
373d9915	150	if (handler) {
74ed6b5e BS	151	u32 n;
74ed6b5e BS	152
373d9915 RC	153	/*
	154	* validate head pointer value and compute
	155	* the number of remaining WQEs.
	156	*/
	157	n = wq->head;
	158	if (n >= rq->size)
	159	n = 0;
	160	if (n < tail)
	161	n += rq->size - tail;
74ed6b5e	162	else
373d9915	163	n -= tail;
74ed6b5e BS	164	if (n < srq->limit) {
	165	struct ib_event ev;
	166
	167	srq->limit = 0;
	168	spin_unlock_irqrestore(&rq->lock, flags);
	169	ev.device = qp->ibqp.device;
	170	ev.element.srq = qp->ibqp.srq;
	171	ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
373d9915	172	handler(&ev, srq->ibsrq.srq_context);
74ed6b5e BS	173	} else
	174	spin_unlock_irqrestore(&rq->lock, flags);
	175	} else
	176	spin_unlock_irqrestore(&rq->lock, flags);
373d9915	177
4ee97180	178	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
74ed6b5e BS	179	if (ah_attr->ah_flags & IB_AH_GRH) {
74ed6b5e BS	180	ipath_copy_sge(&rsge, &ah_attr->grh, sizeof(struct ib_grh));
4ee97180	181	wc.wc_flags \|= IB_WC_GRH;
74ed6b5e BS	182	} else
74ed6b5e BS	183	ipath_skip_sge(&rsge, sizeof(struct ib_grh));
4ee97180	184	sge = swqe->sg_list;
74ed6b5e BS	185	while (length) {
	186	u32 len = sge->length;
	187
	188	if (len > length)
	189	len = length;
30d149ab RC	190	if (len > sge->sge_length)
30d149ab RC	191	len = sge->sge_length;
74ed6b5e BS	192	BUG_ON(len == 0);
	193	ipath_copy_sge(&rsge, sge->vaddr, len);
	194	sge->vaddr += len;
	195	sge->length -= len;
	196	sge->sge_length -= len;
	197	if (sge->sge_length == 0) {
4ee97180 RC	198	if (--swqe->wr.num_sge)
4ee97180 RC	199	sge++;
74ed6b5e BS	200	} else if (sge->length == 0 && sge->mr != NULL) {
	201	if (++sge->n >= IPATH_SEGSZ) {
	202	if (++sge->m >= sge->mr->mapsz)
	203	break;
	204	sge->n = 0;
	205	}
	206	sge->vaddr =
	207	sge->mr->map[sge->m]->segs[sge->n].vaddr;
	208	sge->length =
	209	sge->mr->map[sge->m]->segs[sge->n].length;
	210	}
	211	length -= len;
	212	}
4ee97180 RC	213	wc.status = IB_WC_SUCCESS;
4ee97180 RC	214	wc.opcode = IB_WC_RECV;
4ee97180 RC	215	wc.qp = &qp->ibqp;
4ee97180 RC	216	wc.src_qp = sqp->ibqp.qp_num;
74ed6b5e	217	/* XXX do we know which pkey matched? Only needed for GSI. */
4ee97180 RC	218	wc.pkey_index = 0;
4ee97180 RC	219	wc.slid = dev->dd->ipath_lid \|
74ed6b5e	220	(ah_attr->src_path_bits &
542869a1	221	((1 << dev->dd->ipath_lmc) - 1));
4ee97180 RC	222	wc.sl = ah_attr->sl;
4ee97180 RC	223	wc.dlid_path_bits =
542869a1	224	ah_attr->dlid & ((1 << dev->dd->ipath_lmc) - 1);
4ee97180	225	wc.port_num = 1;
74ed6b5e	226	/* Signal completion event if the solicited bit is set. */
4ee97180 RC	227	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
	228	swqe->wr.send_flags & IB_SEND_SOLICITED);
	229	drop:
74ed6b5e BS	230	if (atomic_dec_and_test(&qp->refcount))
74ed6b5e BS	231	wake_up(&qp->wait);
e509be89	232	done:;
74ed6b5e BS	233	}
	234
	235	/**
4ee97180	236	* ipath_make_ud_req - construct a UD request packet
74ed6b5e	237	* @qp: the QP
74ed6b5e	238	*
4ee97180	239	* Return 1 if constructed; otherwise, return 0.
74ed6b5e	240	*/
4ee97180	241	int ipath_make_ud_req(struct ipath_qp *qp)
74ed6b5e BS	242	{
	243	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	244	struct ipath_other_headers *ohdr;
	245	struct ib_ah_attr *ah_attr;
4ee97180	246	struct ipath_swqe *wqe;
e509be89	247	unsigned long flags;
74ed6b5e	248	u32 nwords;
74ed6b5e BS	249	u32 extra_bytes;
	250	u32 bth0;
	251	u16 lrh0;
	252	u16 lid;
4ee97180	253	int ret = 0;
7ec01ff9	254	int next_cur;
74ed6b5e	255
e509be89 RC	256	spin_lock_irqsave(&qp->s_lock, flags);
	257
	258	if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_NEXT_SEND_OK)) {
	259	if (!(ib_ipath_state_ops[qp->state] & IPATH_FLUSH_SEND))
	260	goto bail;
	261	/* We are in the error state, flush the work request. */
	262	if (qp->s_last == qp->s_head)
	263	goto bail;
	264	/* If DMAs are in progress, we can't flush immediately. */
	265	if (atomic_read(&qp->s_dma_busy)) {
	266	qp->s_flags \|= IPATH_S_WAIT_DMA;
	267	goto bail;
	268	}
	269	wqe = get_swqe_ptr(qp, qp->s_last);
	270	ipath_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
	271	goto done;
	272	}
74ed6b5e	273
4ee97180	274	if (qp->s_cur == qp->s_head)
6ce73b07	275	goto bail;
6ce73b07	276
4ee97180	277	wqe = get_swqe_ptr(qp, qp->s_cur);
7ec01ff9 RC	278	next_cur = qp->s_cur + 1;
	279	if (next_cur >= qp->s_size)
	280	next_cur = 0;
74ed6b5e BS	281
74ed6b5e BS	282	/* Construct the header. */
4ee97180	283	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
27b678dd BS	284	if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE) {
27b678dd BS	285	if (ah_attr->dlid != IPATH_PERMISSIVE_LID)
74ed6b5e BS	286	dev->n_multicast_xmit++;
	287	else
	288	dev->n_unicast_xmit++;
	289	} else {
	290	dev->n_unicast_xmit++;
e509be89	291	lid = ah_attr->dlid & ~((1 << dev->dd->ipath_lmc) - 1);
34b2aafe	292	if (unlikely(lid == dev->dd->ipath_lid)) {
e509be89 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	*/
	300	if (atomic_read(&qp->s_dma_busy)) {
	301	qp->s_flags \|= IPATH_S_WAIT_DMA;
	302	goto bail;
	303	}
7ec01ff9	304	qp->s_cur = next_cur;
e509be89	305	spin_unlock_irqrestore(&qp->s_lock, flags);
4ee97180	306	ipath_ud_loopback(qp, wqe);
e509be89 RC	307	spin_lock_irqsave(&qp->s_lock, flags);
e509be89 RC	308	ipath_send_complete(qp, wqe, IB_WC_SUCCESS);
74ed6b5e BS	309	goto done;
	310	}
	311	}
4ee97180	312
7ec01ff9	313	qp->s_cur = next_cur;
4ee97180 RC	314	extra_bytes = -wqe->length & 3;
	315	nwords = (wqe->length + extra_bytes) >> 2;
	316
	317	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
	318	qp->s_hdrwords = 7;
4ee97180 RC	319	qp->s_cur_size = wqe->length;
4ee97180 RC	320	qp->s_cur_sge = &qp->s_sge;
124b4dcb	321	qp->s_dmult = ah_attr->static_rate;
4ee97180 RC	322	qp->s_wqe = wqe;
	323	qp->s_sge.sge = wqe->sg_list[0];
	324	qp->s_sge.sg_list = wqe->sg_list + 1;
	325	qp->s_sge.num_sge = wqe->wr.num_sge;
	326
74ed6b5e BS	327	if (ah_attr->ah_flags & IB_AH_GRH) {
74ed6b5e BS	328	/* Header size in 32-bit words. */
4ee97180 RC	329	qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
	330	&ah_attr->grh,
	331	qp->s_hdrwords, nwords);
27b678dd	332	lrh0 = IPATH_LRH_GRH;
74ed6b5e	333	ohdr = &qp->s_hdr.u.l.oth;
74ed6b5e BS	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. */
27b678dd	340	lrh0 = IPATH_LRH_BTH;
74ed6b5e BS	341	ohdr = &qp->s_hdr.u.oth;
74ed6b5e BS	342	}
4ee97180	343	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
4e1e93a4	344	qp->s_hdrwords++;
0f39cf3d	345	ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
74ed6b5e	346	bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
4ee97180	347	} else
74ed6b5e	348	bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
74ed6b5e BS	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	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
	353	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid); /* DEST LID */
4ee97180 RC	354	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords +
4ee97180 RC	355	SIZE_OF_CRC);
34b2aafe	356	lid = dev->dd->ipath_lid;
74ed6b5e BS	357	if (lid) {
74ed6b5e BS	358	lid \|= ah_attr->src_path_bits &
542869a1	359	((1 << dev->dd->ipath_lmc) - 1);
74ed6b5e BS	360	qp->s_hdr.lrh[3] = cpu_to_be16(lid);
	361	} else
	362	qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
4ee97180	363	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
74ed6b5e BS	364	bth0 \|= 1 << 23;
74ed6b5e BS	365	bth0 \|= extra_bytes << 20;
27b678dd	366	bth0 \|= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY :
34b2aafe	367	ipath_get_pkey(dev->dd, qp->s_pkey_index);
74ed6b5e BS	368	ohdr->bth[0] = cpu_to_be32(bth0);
	369	/*
	370	* Use the multicast QP if the destination LID is a multicast LID.
	371	*/
27b678dd BS	372	ohdr->bth[1] = ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
27b678dd BS	373	ah_attr->dlid != IPATH_PERMISSIVE_LID ?
9c3da099	374	cpu_to_be32(IPATH_MULTICAST_QPN) :
4ee97180	375	cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
27b678dd	376	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPATH_PSN_MASK);
74ed6b5e BS	377	/*
	378	* Qkeys with the high order bit set mean use the
	379	* qkey from the QP context instead of the WR (see 10.2.5).
	380	*/
4ee97180 RC	381	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
4ee97180 RC	382	qp->qkey : wqe->wr.wr.ud.remote_qkey);
74ed6b5e	383	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
74ed6b5e BS	384
74ed6b5e BS	385	done:
4ee97180	386	ret = 1;
e509be89	387	goto unlock;
74ed6b5e BS	388
74ed6b5e BS	389	bail:
e509be89 RC	390	qp->s_flags &= ~IPATH_S_BUSY;
	391	unlock:
	392	spin_unlock_irqrestore(&qp->s_lock, flags);
74ed6b5e BS	393	return ret;
	394	}
	395
	396	/**
	397	* ipath_ud_rcv - receive an incoming UD packet
	398	* @dev: the device the packet came in on
	399	* @hdr: the packet header
	400	* @has_grh: true if the packet has a GRH
	401	* @data: the packet data
	402	* @tlen: the packet length
	403	* @qp: the QP the packet came on
	404	*
	405	* This is called from ipath_qp_rcv() to process an incoming UD packet
	406	* for the given QP.
	407	* Called at interrupt level.
	408	*/
	409	void ipath_ud_rcv(struct ipath_ibdev dev, struct ipath_ib_header hdr,
	410	int has_grh, void data, u32 tlen, struct ipath_qp qp)
	411	{
	412	struct ipath_other_headers *ohdr;
	413	int opcode;
	414	u32 hdrsize;
	415	u32 pad;
74ed6b5e BS	416	struct ib_wc wc;
	417	u32 qkey;
	418	u32 src_qp;
74ed6b5e BS	419	u16 dlid;
	420	int header_in_data;
	421
	422	/* Check for GRH */
	423	if (!has_grh) {
	424	ohdr = &hdr->u.oth;
	425	hdrsize = 8 + 12 + 8; /* LRH + BTH + DETH */
	426	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
	427	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]);
	428	header_in_data = 0;
	429	} else {
	430	ohdr = &hdr->u.l.oth;
	431	hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
	432	/*
	433	* The header with GRH is 68 bytes and the core driver sets
	434	* the eager header buffer size to 56 bytes so the last 12
	435	* bytes of the IB header is in the data buffer.
	436	*/
34b2aafe	437	header_in_data = dev->dd->ipath_rcvhdrentsize == 16;
74ed6b5e BS	438	if (header_in_data) {
	439	qkey = be32_to_cpu(((__be32 *) data)[1]);
	440	src_qp = be32_to_cpu(((__be32 *) data)[2]);
	441	data += 12;
	442	} else {
	443	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
	444	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]);
	445	}
	446	}
27b678dd	447	src_qp &= IPATH_QPN_MASK;
74ed6b5e BS	448
	449	/*
	450	* Check that the permissive LID is only used on QP0
	451	* and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
	452	*/
	453	if (qp->ibqp.qp_num) {
	454	if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
	455	hdr->lrh[3] == IB_LID_PERMISSIVE)) {
	456	dev->n_pkt_drops++;
	457	goto bail;
	458	}
	459	if (unlikely(qkey != qp->qkey)) {
	460	/* XXX OK to lose a count once in a while. */
	461	dev->qkey_violations++;
	462	dev->n_pkt_drops++;
	463	goto bail;
	464	}
	465	} else if (hdr->lrh[1] == IB_LID_PERMISSIVE \|\|
	466	hdr->lrh[3] == IB_LID_PERMISSIVE) {
	467	struct ib_smp smp = (struct ib_smp ) data;
	468
	469	if (smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
	470	dev->n_pkt_drops++;
	471	goto bail;
	472	}
	473	}
	474
0a69631b RC	475	/*
	476	* The opcode is in the low byte when its in network order
	477	* (top byte when in host order).
	478	*/
	479	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	480	if (qp->ibqp.qp_num > 1 &&
	481	opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
	482	if (header_in_data) {
00f7ec36	483	wc.ex.imm_data = (__be32 ) data;
0a69631b RC	484	data += sizeof(__be32);
0a69631b RC	485	} else
00f7ec36	486	wc.ex.imm_data = ohdr->u.ud.imm_data;
0a69631b RC	487	wc.wc_flags = IB_WC_WITH_IMM;
	488	hdrsize += sizeof(u32);
	489	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
00f7ec36	490	wc.ex.imm_data = 0;
0a69631b RC	491	wc.wc_flags = 0;
	492	} else {
	493	dev->n_pkt_drops++;
	494	goto bail;
	495	}
	496
74ed6b5e BS	497	/* Get the number of bytes the message was padded by. */
	498	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
	499	if (unlikely(tlen < (hdrsize + pad + 4))) {
	500	/* Drop incomplete packets. */
	501	dev->n_pkt_drops++;
	502	goto bail;
	503	}
	504	tlen -= hdrsize + pad + 4;
	505
	506	/* Drop invalid MAD packets (see 13.5.3.1). */
	507	if (unlikely((qp->ibqp.qp_num == 0 &&
	508	(tlen != 256 \|\|
	509	(be16_to_cpu(hdr->lrh[0]) >> 12) != 15)) \|\|
	510	(qp->ibqp.qp_num == 1 &&
	511	(tlen != 256 \|\|
	512	(be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))) {
	513	dev->n_pkt_drops++;
	514	goto bail;
	515	}
	516
	517	/*
	518	* A GRH is expected to preceed the data even if not
	519	* present on the wire.
	520	*/
	521	wc.byte_len = tlen + sizeof(struct ib_grh);
	522
74ed6b5e BS	523	/*
74ed6b5e BS	524	* Get the next work request entry to find where to put the data.
74ed6b5e	525	*/
e509be89 RC	526	if (qp->r_flags & IPATH_R_REUSE_SGE)
e509be89 RC	527	qp->r_flags &= ~IPATH_R_REUSE_SGE;
373d9915	528	else if (!ipath_get_rwqe(qp, 0)) {
fba75200 BS	529	/*
	530	* Count VL15 packets dropped due to no receive buffer.
	531	* Otherwise, count them as buffer overruns since usually,
	532	* the HW will be able to receive packets even if there are
	533	* no QPs with posted receive buffers.
	534	*/
	535	if (qp->ibqp.qp_num == 0)
	536	dev->n_vl15_dropped++;
	537	else
	538	dev->rcv_errors++;
74ed6b5e BS	539	goto bail;
	540	}
	541	/* Silently drop packets which are too big. */
373d9915	542	if (wc.byte_len > qp->r_len) {
e509be89	543	qp->r_flags \|= IPATH_R_REUSE_SGE;
74ed6b5e BS	544	dev->n_pkt_drops++;
	545	goto bail;
	546	}
74ed6b5e BS	547	if (has_grh) {
	548	ipath_copy_sge(&qp->r_sge, &hdr->u.l.grh,
	549	sizeof(struct ib_grh));
	550	wc.wc_flags \|= IB_WC_GRH;
	551	} else
	552	ipath_skip_sge(&qp->r_sge, sizeof(struct ib_grh));
	553	ipath_copy_sge(&qp->r_sge, data,
	554	wc.byte_len - sizeof(struct ib_grh));
e509be89 RC	555	if (!test_and_clear_bit(IPATH_R_WRID_VALID, &qp->r_aflags))
e509be89 RC	556	goto bail;
373d9915	557	wc.wr_id = qp->r_wr_id;
74ed6b5e BS	558	wc.status = IB_WC_SUCCESS;
	559	wc.opcode = IB_WC_RECV;
	560	wc.vendor_err = 0;
062dbb69	561	wc.qp = &qp->ibqp;
74ed6b5e BS	562	wc.src_qp = src_qp;
	563	/* XXX do we know which pkey matched? Only needed for GSI. */
	564	wc.pkey_index = 0;
	565	wc.slid = be16_to_cpu(hdr->lrh[3]);
	566	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
	567	dlid = be16_to_cpu(hdr->lrh[1]);
	568	/*
	569	* Save the LMC lower bits if the destination LID is a unicast LID.
	570	*/
27b678dd	571	wc.dlid_path_bits = dlid >= IPATH_MULTICAST_LID_BASE ? 0 :
542869a1	572	dlid & ((1 << dev->dd->ipath_lmc) - 1);
4ee97180	573	wc.port_num = 1;
74ed6b5e BS	574	/* Signal completion event if the solicited bit is set. */
	575	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
	576	(ohdr->bth[0] &
9c3da099	577	cpu_to_be32(1 << 23)) != 0);
74ed6b5e BS	578
	579	bail:;
	580	}