};
static const __be32 mlx4_ib_opcode[] = {
- [IB_WR_SEND] = __constant_cpu_to_be32(MLX4_OPCODE_SEND),
- [IB_WR_LSO] = __constant_cpu_to_be32(MLX4_OPCODE_LSO),
- [IB_WR_SEND_WITH_IMM] = __constant_cpu_to_be32(MLX4_OPCODE_SEND_IMM),
- [IB_WR_RDMA_WRITE] = __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
- [IB_WR_RDMA_WRITE_WITH_IMM] = __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
- [IB_WR_RDMA_READ] = __constant_cpu_to_be32(MLX4_OPCODE_RDMA_READ),
- [IB_WR_ATOMIC_CMP_AND_SWP] = __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_CS),
- [IB_WR_ATOMIC_FETCH_AND_ADD] = __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_FA),
- [IB_WR_SEND_WITH_INV] = __constant_cpu_to_be32(MLX4_OPCODE_SEND_INVAL),
- [IB_WR_LOCAL_INV] = __constant_cpu_to_be32(MLX4_OPCODE_LOCAL_INVAL),
- [IB_WR_FAST_REG_MR] = __constant_cpu_to_be32(MLX4_OPCODE_FMR),
+ [IB_WR_SEND] = cpu_to_be32(MLX4_OPCODE_SEND),
+ [IB_WR_LSO] = cpu_to_be32(MLX4_OPCODE_LSO),
+ [IB_WR_SEND_WITH_IMM] = cpu_to_be32(MLX4_OPCODE_SEND_IMM),
+ [IB_WR_RDMA_WRITE] = cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
+ [IB_WR_RDMA_WRITE_WITH_IMM] = cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
+ [IB_WR_RDMA_READ] = cpu_to_be32(MLX4_OPCODE_RDMA_READ),
+ [IB_WR_ATOMIC_CMP_AND_SWP] = cpu_to_be32(MLX4_OPCODE_ATOMIC_CS),
+ [IB_WR_ATOMIC_FETCH_AND_ADD] = cpu_to_be32(MLX4_OPCODE_ATOMIC_FA),
+ [IB_WR_SEND_WITH_INV] = cpu_to_be32(MLX4_OPCODE_SEND_INVAL),
+ [IB_WR_LOCAL_INV] = cpu_to_be32(MLX4_OPCODE_LOCAL_INVAL),
+ [IB_WR_FAST_REG_MR] = cpu_to_be32(MLX4_OPCODE_FMR),
};
static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
* anymore, so we do this only if selective signaling is off.
*
* Further, on 32-bit platforms, we can't use vmap() to make
- * the QP buffer virtually contigious. Thus we have to use
+ * the QP buffer virtually contiguous. Thus we have to use
* constant-sized WRs to make sure a WR is always fully within
* a single page-sized chunk.
*
}
static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
+ __acquires(&send_cq->lock) __acquires(&recv_cq->lock)
{
- if (send_cq == recv_cq)
+ if (send_cq == recv_cq) {
spin_lock_irq(&send_cq->lock);
- else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
+ __acquire(&recv_cq->lock);
+ } else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
spin_lock_irq(&send_cq->lock);
spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
} else {
}
static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
+ __releases(&send_cq->lock) __releases(&recv_cq->lock)
{
- if (send_cq == recv_cq)
+ if (send_cq == recv_cq) {
+ __release(&recv_cq->lock);
spin_unlock_irq(&send_cq->lock);
- else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
+ } else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
spin_unlock(&recv_cq->lock);
spin_unlock_irq(&send_cq->lock);
} else {
int i;
for (i = 0; i < wr->wr.fast_reg.page_list_len; ++i)
- wr->wr.fast_reg.page_list->page_list[i] =
+ mfrpl->mapped_page_list[i] =
cpu_to_be64(wr->wr.fast_reg.page_list->page_list[i] |
MLX4_MTT_FLAG_PRESENT);
}
static int build_lso_seg(struct mlx4_wqe_lso_seg *wqe, struct ib_send_wr *wr,
- struct mlx4_ib_qp *qp, unsigned *lso_seg_len)
+ struct mlx4_ib_qp *qp, unsigned *lso_seg_len,
+ __be32 *lso_hdr_sz)
{
unsigned halign = ALIGN(sizeof *wqe + wr->wr.ud.hlen, 16);
memcpy(wqe->header, wr->wr.ud.header, wr->wr.ud.hlen);
- /* make sure LSO header is written before overwriting stamping */
- wmb();
-
- wqe->mss_hdr_size = cpu_to_be32((wr->wr.ud.mss - wr->wr.ud.hlen) << 16 |
- wr->wr.ud.hlen);
-
+ *lso_hdr_sz = cpu_to_be32((wr->wr.ud.mss - wr->wr.ud.hlen) << 16 |
+ wr->wr.ud.hlen);
*lso_seg_len = halign;
return 0;
}
int uninitialized_var(stamp);
int uninitialized_var(size);
unsigned uninitialized_var(seglen);
+ __be32 dummy;
+ __be32 *lso_wqe;
+ __be32 uninitialized_var(lso_hdr_sz);
int i;
spin_lock_irqsave(&qp->sq.lock, flags);
ind = qp->sq_next_wqe;
for (nreq = 0; wr; ++nreq, wr = wr->next) {
+ lso_wqe = &dummy;
+
if (mlx4_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
err = -ENOMEM;
*bad_wr = wr;
break;
case IB_WR_LOCAL_INV:
+ ctrl->srcrb_flags |=
+ cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
set_local_inv_seg(wqe, wr->ex.invalidate_rkey);
wqe += sizeof (struct mlx4_wqe_local_inval_seg);
size += sizeof (struct mlx4_wqe_local_inval_seg) / 16;
break;
case IB_WR_FAST_REG_MR:
+ ctrl->srcrb_flags |=
+ cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
set_fmr_seg(wqe, wr);
wqe += sizeof (struct mlx4_wqe_fmr_seg);
size += sizeof (struct mlx4_wqe_fmr_seg) / 16;
size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
if (wr->opcode == IB_WR_LSO) {
- err = build_lso_seg(wqe, wr, qp, &seglen);
+ err = build_lso_seg(wqe, wr, qp, &seglen, &lso_hdr_sz);
if (unlikely(err)) {
*bad_wr = wr;
goto out;
}
+ lso_wqe = (__be32 *) wqe;
wqe += seglen;
size += seglen / 16;
}
for (i = wr->num_sge - 1; i >= 0; --i, --dseg)
set_data_seg(dseg, wr->sg_list + i);
+ /*
+ * Possibly overwrite stamping in cacheline with LSO
+ * segment only after making sure all data segments
+ * are written.
+ */
+ wmb();
+ *lso_wqe = lso_hdr_sz;
+
ctrl->fence_size = (wr->send_flags & IB_SEND_FENCE ?
MLX4_WQE_CTRL_FENCE : 0) | size;
stamp_send_wqe(qp, stamp, size * 16);
ind = pad_wraparound(qp, ind);
}
-
}
out: