> > if (attr_mask & IB_QP_ACCESS_FLAGS) {
> > if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE) {
> > nesqp->nesqp_context->misc |=
> > cpu_to_le32(NES_QPCONTEXT_MISC_RDMA_WRITE_EN |
> >
> > NES_QPCONTEXT_MISC_RDMA_READ_EN);
> > issue_modify_qp = 1;
> > }
>
> This one maybe a workaround for a silicon bug. I have to look
> into it to be sure.
not sure I see how this could be a HW bug workaround... this is setting
some QP flags based on some random flag that a user shouldn't be
setting.
> > if (nesqp->user_mode) {
> > nesqp->nesqp_context->misc |=
> > cpu_to_le32(NES_QPCONTEXT_MISC_RDMA_WRITE_EN |
> >
> > NES_QPCONTEXT_MISC_RDMA_READ_EN);
> > issue_modify_qp = 1;
> > }
> > }
> There is a comment in our CVS source indicating that was a
> workaround for a CM bug in OFED 1.2. We may be able to remove this one.
would be nice to fix the underlying bug if it's still present. I would
note that cxgb3 has no corresponding code for usermode QPs and appears
to work fine.
- R.
_______________________________________________
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
