[ofa-general] Re: [PATCH 2/2] IB/iSER: Use offset from r2t header for rdma

[Eli Dorfman]

On Sun, Apr 27, 2008 at 3:55 PM, Eli Dorfman <dorfman.eli at gmail.com> wrote:
> Use offset from r2t header for rdma instead of using
>  internal offset counter.
>
>  Signed-off-by: Eli Dorfman <elid at voltaire.com>
>  ---
>   usr/iscsi/iscsi_rdma.c |   16 +++++-----------
>   1 files changed, 5 insertions(+), 11 deletions(-)
>
>  diff --git a/usr/iscsi/iscsi_rdma.c b/usr/iscsi/iscsi_rdma.c
>  index d46ddff..84f5949 100644
>  --- a/usr/iscsi/iscsi_rdma.c
>  +++ b/usr/iscsi/iscsi_rdma.c
>  @@ -1447,28 +1447,22 @@ static int iscsi_rdma_rdma_read(struct
>  iscsi_connection *conn)
>         struct iscsi_r2t_rsp *r2t = (struct iscsi_r2t_rsp *) &conn->rsp.bhs;
>         uint8_t *buf;
>         uint32_t len;
>  +       uint32_t offset;
>         int ret;
>
>         buf = (uint8_t *) task->data + task->offset;
>         len = be32_to_cpu(r2t->data_length);
>  +       offset = be32_to_cpu(r2t->data_offset);
>
>  -       dprintf("len %u stag %x va %llx\n",
>  +       dprintf("len %u stag %x va %llx offset %x\n",
>                 len, itask->rem_write_stag,
>  -               (unsigned long long) itask->rem_write_va);
>  +               (unsigned long long) itask->rem_write_va, offset);
>
>         ret = iser_post_rdma_wr(ci, task, buf, len, IBV_WR_RDMA_READ,
>  -                               itask->rem_write_va, itask->rem_write_stag);
>  +                               itask->rem_write_va + offset, itask->rem_write_stag);
>         if (ret < 0)
>                 return ret;
>
>  -       /*
>  -        * Initiator registers the entire buffer, but gives us a VA that
>  -        * is advanced by immediate + unsolicited data amounts.  Advance
>  -        * rem_va as we read, knowing that the target always grabs segments
>  -        * in order.
>  -        */
>  -       itask->rem_write_va += len;
>  -
>         return 0;
>   }
>
>  --
>  1.5.5
>
Please do not apply this patch until we decide how to sync this with
the initiator side.
See the following discussion for details:
http://www.ietf.org/mail-archive/web/ips/current/msg02506.html

I tend to agree with Pete's option (3) implementing iSER HELLO message
in the initiator and target.
Then adding this patch and the corresponding initiator patch so that we have:
Old initiator working with old target, AND
New initiator working with new target.

Eli