[ofa-general] Re: setting iWarp IRD and ORD
Kanevsky, Arkady
Arkady.Kanevsky at netapp.com
Fri Jan 4 06:09:17 PST 2008
And what happens when RNIC on two sides have a different upper limits?
Specifically, if requestor asks for ORD which is bigger than responder
can handle?
Is it user responsibility to pass ORD request out of bound to responder
and
if responder can not satisfy it then reject the request?
Thanks,
Arkady Kanevsky email: arkady at netapp.com
Network Appliance Inc. phone: 781-768-5395
1601 Trapelo Rd. - Suite 16. Fax: 781-895-1195
Waltham, MA 02451 central phone: 781-768-5300
________________________________
From: Steve Wise [mailto:swise at opengridcomputing.com]
Sent: Thursday, January 03, 2008 9:42 PM
To: Sean Hefty
Cc: OpenFabrics General
Subject: [ofa-general] Re: setting iWarp IRD and ORD
Sean Hefty wrote:
What exactly are you fixing?
It may help to look at the patch for the librdmacm that
I just posted. The
problem that was seen was that uDAPL set IRD/ORD to 16
in the connection
request, but the passive side could only support 4. The
rdma_cm was supposed to
use the lower value (provided by the user when calling
rdma_accept) when
transitioning the QP, but instead used the value from
the request.
Since iWarp connect request doesn't carry the IRD/ORD
and always uses the value
provided by the user through rdma_connect or
rdma_accept, it doesn't sound like
it would hit this problem. I just don't quite follow
why
iwcm_init_qp_rts_attr() doesn't set the QP attribute
mask for IRD/ORD, or how
the QP is programmed with these values.
Perhaps this is a bug. But the chelsio driver just saves off
the ord/ird from the connection parameters and then programs the qp with
these values when the qp is associated with the connection (just after
the connection goes into rdma mode)...
- Sean
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20080104/9ff531df/attachment.html>
More information about the general
mailing list