On Tue, 2008-04-01 at 12:59 -0700, Roland Dreier wrote:
> > + halign = ALIGN(wr->wr.ud.hlen, 16);
>
> This doesn't seem connected to the problem I see, but is this correct?
> Suppose hlen is 48... then halign will be 48 but it really should be
> 64 I think. Do we really want
>
> halign = ALIGN(wr->wr.ud.hlen + sizeof *wqe, 16);
>
> instead?
>
I don't think so, at least in the case that hlen equals 48 which is a
valid one since the total length used by the LSO segment would be 48 + 4
which requires 4 * 16 bytes chunks. If we'd use the above statement the
send would fail.
Anyway I think this function should look like this:
static int build_lso_seg(struct mlx4_lso_seg *wqe, struct ib_send_wr *wr,
struct mlx4_ib_qp *qp, unsigned *lso_seg_len)
{
unsigned halign = ALIGN(wr->wr.ud.hlen + 4, 16);
/*
* This is a temporary limitation and will be removed in
* a forthcoming FW release:
*/
if (unlikely(halign > 64))
return -EINVAL;
if (unlikely(!(qp->flags & MLX4_IB_QP_LSO) &&
wr->num_sge > qp->sq.max_gs - (halign >> 4)))
return -EINVAL;
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_seg_len = halign;
return 0;
}
And also I suggest to use these too:
@@ -1539,7 +1539,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct
ib_send_wr *wr,
if (wr->opcode == IB_WR_LSO) {
err = build_lso_seg(wqe, wr, qp, &seglen);
- if (err) {
+ if (unlikely(err)) {
*bad_wr = wr;
goto out;
}
@@ -1551,7 +1551,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct
ib_send_wr *wr,
case IB_QPT_SMI:
case IB_QPT_GSI:
err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen);
- if (err) {
+ if (unlikely(err)) {
*bad_wr = wr;
goto out;
}
@@ -1594,7 +1594,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct
ib_send_wr *wr,
*/
wmb();
- if (wr->opcode < 0 || wr->opcode >= ARRAY_SIZE(mlx4_ib_opcode))
{
+ if (unlikely(wr->opcode < 0 || wr->opcode >=
ARRAY_SIZE(mlx4_ib_opcode))) {
err = -EINVAL;
goto out;
}
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
