/*
- * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
+ * 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
u32 length;
qp = ipath_lookup_qpn(&dev->qp_table, swqe->wr.wr.ud.remote_qpn);
- if (!qp) {
+ if (!qp || !(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
dev->n_pkt_drops++;
- goto send_comp;
+ goto done;
}
- rsge.sg_list = NULL;
-
/*
* Check that the qkey matches (except for QP0, see 9.6.1.4.1).
* Qkeys with the high order bit set mean use the
* 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.imm_data = swqe->wr.imm_data;
- } else {
- wc.wc_flags = 0;
- wc.imm_data = 0;
+ wc.ex.imm_data = swqe->wr.ex.imm_data;
}
/*
rq = &qp->r_rq;
}
- if (rq->max_sge > 1) {
- /*
- * XXX We could use GFP_KERNEL if ipath_do_send()
- * was always called from the tasklet instead of
- * from ipath_post_send().
- */
- rsge.sg_list = kmalloc((rq->max_sge - 1) *
- sizeof(struct ipath_sge),
- GFP_ATOMIC);
- if (!rsge.sg_list) {
- dev->n_pkt_drops++;
- goto drop;
- }
- }
-
/*
* 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
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++;
}
wc.status = IB_WC_SUCCESS;
wc.opcode = IB_WC_RECV;
- wc.vendor_err = 0;
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->mkeyprot_resv_lmc & 7)) - 1));
+ ((1 << dev->dd->ipath_lmc) - 1));
wc.sl = ah_attr->sl;
wc.dlid_path_bits =
- ah_attr->dlid & ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
+ 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:
- kfree(rsge.sg_list);
if (atomic_dec_and_test(&qp->refcount))
wake_up(&qp->wait);
-send_comp:
- ipath_send_complete(sqp, swqe, IB_WC_SUCCESS);
+done:;
}
/**
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;
- if (unlikely(!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK)))
- goto bail;
+ 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;
dev->n_unicast_xmit++;
} else {
dev->n_unicast_xmit++;
- lid = ah_attr->dlid &
- ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
+ 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;
- if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
- qp->s_hdrwords++;
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;
ohdr = &qp->s_hdr.u.oth;
}
if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
- ohdr->u.ud.imm_data = wqe->wr.imm_data;
+ 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;
lid = dev->dd->ipath_lid;
if (lid) {
lid |= ah_attr->src_path_bits &
- ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
+ ((1 << dev->dd->ipath_lmc) - 1);
qp->s_hdr.lrh[3] = cpu_to_be16(lid);
} else
qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
*/
ohdr->bth[1] = ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
ah_attr->dlid != IPATH_PERMISSIVE_LID ?
- __constant_cpu_to_be32(IPATH_MULTICAST_QPN) :
+ 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);
/*
ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
done:
- if (++qp->s_cur >= qp->s_size)
- qp->s_cur = 0;
ret = 1;
+ goto unlock;
bail:
+ qp->s_flags &= ~IPATH_S_BUSY;
+unlock:
+ spin_unlock_irqrestore(&qp->s_lock, flags);
return ret;
}
}
}
+ /*
+ * 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))) {
wc.byte_len = tlen + sizeof(struct ib_grh);
/*
- * 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.imm_data = *(__be32 *) data;
- data += sizeof(__be32);
- } else
- wc.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.imm_data = 0;
- wc.wc_flags = 0;
- } else {
- dev->n_pkt_drops++;
- goto bail;
- }
-
- /*
* Get the next work request entry to find where to put the data.
*/
- if (qp->r_reuse_sge)
- qp->r_reuse_sge = 0;
+ 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.
}
/* Silently drop packets which are too big. */
if (wc.byte_len > qp->r_len) {
- qp->r_reuse_sge = 1;
+ qp->r_flags |= IPATH_R_REUSE_SGE;
dev->n_pkt_drops++;
goto bail;
}
ipath_skip_sge(&qp->r_sge, sizeof(struct ib_grh));
ipath_copy_sge(&qp->r_sge, data,
wc.byte_len - sizeof(struct ib_grh));
- qp->r_wrid_valid = 0;
+ 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;
* 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->mkeyprot_resv_lmc & 7)) - 1);
+ 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] &
- __constant_cpu_to_be32(1 << 23)) != 0);
+ cpu_to_be32(1 << 23)) != 0);
bail:;
}