[ofw] SRP Initiator Issue

Josh Morris josh.morris at texmemsys.com
Wed May 9 09:24:36 PDT 2007


In addition to the Send Message Depth, Jim Mott pointed out that the 
RDMA Transfer Size should also be greater than 0. I have changed these 
values and can now connect and run data to and from the target. Thanks 
to everyone for the help.

I do have one more quick question before I commit these changes to the 
SRP target. The SRP spec states that the Send Message Depth and RDMA 
Transfer Size fields are reserved. I'm guessing that in this case 
"reserved" does not mean the field should be set to 0, but the field 
should instead follow the IBTA spec requirements. Is this correct?

Thanks,
Josh


Yossi Leybovich wrote:
>
> Hi
>
> Look at his code from srp_hba.c
>
> __srp_validate_ioc()
>
> *…...*
>
> * // Can it handle our IO requirements?*
>
> * if ( cl_ntoh32( p_ioc_rec->info.profile.send_msg_size ) < 
> SRP_MIN_TGT_TO_INI_IU ||*
>
> * cl_ntoh16( p_ioc_rec->info.profile.send_msg_depth ) == 0 ||*
>
> * cl_ntoh32( p_ioc_rec->info.profile.rdma_size ) < 
> SRP_MIN_TGT_TO_INI_DMA )*
>
> * {*
>
> * SRP_PRINT_EXIT( TRACE_LEVEL_ERROR, SRP_DBG_ERROR,*
>
> * ("Device Not Capable.\n") );*
>
> * return FALSE;*
>
> * }*
>
> From what I scan see the send_msg_depth is 0* *
>
> Thanks
>
> Yossi
>
> * *
>
> > -----Original Message-----
>
> > From: Josh Morris [_mailto:josh.morris at texmemsys.com_]
>
> > Sent: Wednesday, May 09, 2007 2:10 AM
>
> > To: Fab Tillier; Yossi Leybovich
>
> > Cc: ofw at lists.openfabrics.org
>
> > Subject: Re: [SPAM] - RE: [ofw] SRP Initiator Issue - Email
>
> > found in subject
>
> >
>
> > Thank you for the responses.
>
> >
>
> > I double checked the IOControllerProfile returned by the
>
> > target and did notice that the "Service Connections" and the
>
> > "Initiators Supported"
>
> > fields were 0 when the specification states they should be at
>
> > least one.
>
> > A little more investigation revealed the target was using the
>
> > IOControllerProfile defined in IBTA 1.2, but the SRP spec
>
> > states that the IOControllerProfile in IBTA 1.0.a should be
>
> > used. Unfortunately correcting this did not fix the problem.
>
> > Also the ServiceEntries packet seems to be formated correctly.
>
> >
>
> > I've attached an IB trace taken with a CATC IB analyzer.
>
> > Please let me know if a different format is needed. The
>
> > trace runs from when the SM was started to after the SRP
>
> > Miniport driver installation fails. I'm not familiar with the
>
> > traceview utility so it may take me a couple of days to get
>
> > WPP traces.
>
> >
>
> > Josh
>
> >
>
> > Fab Tillier wrote:
>
> > > When you look at the DM queries, do you see any service entries
>
> > > reported in the IOC profile of the SRP target? Are these service
>
> > > entries properly formatted (i.e. follow the SRP spec)?
>
> > >
>
> > > Without service entries you won't get any connections.
>
> > >
>
> > > -Fab
>
> > >
>
> > > -----Original Message-----
>
> > > From: ofw-bounces at lists.openfabrics.org
>
> > > [_mailto:ofw-bounces at lists.openfabrics.org_] On Behalf Of Josh Morris
>
> > > Sent: Tuesday, May 08, 2007 12:19 PM
>
> > > To: ofw at lists.openfabrics.org
>
> > > Subject: [ofw] SRP Initiator Issue
>
> > >
>
> > > I am having some issues connecting to an SRP target using
>
> > the OpenIB
>
> > > Windows drivers (version 1.0.0.614). The setup consists of
>
> > an server
>
> > > running Windows Server 2003 directly connected to the SRP
>
> > target. The
>
> > > IB
>
> > >
>
> > > drivers install, however when installing the InfiniBand SRP
>
> > Miniport a
>
> > > failure occurs with the error "Device cannot start (Code 10)." The
>
> > > subnet manager is openSM and is running on the Windows server.
>
> > >
>
> > > Some digging in the list archives revealed that this error is
>
> > > typically
>
> > >
>
> > > caused when the target does not accept the initiator's
>
> > login request.
>
> > > I captured a packet trace and do not see the initiator transmit the
>
> > > login request. I do see subnet management and device
>
> > management MADs,
>
> > > but there are no communication management MADs. I also see some
>
> > > suspicious packets containing a GRH scattered throughout the trace.
>
> > > The destination
>
> > >
>
> > > GID of the GRH does not match the GID of the target.
>
> > >
>
> > > The HCA is a Mellanox PCIe card(MHEL-CF128-T) with firmware version
>
> > > 4.8.200. If I can provide any further information please
>
> > let me know.
>
> > >
>
> > > Thank you,
>
> > >
>
> > > Josh Morris
>
> > > josh.morris at texmemsys.com
>
> > > _______________________________________________
>
> > > ofw mailing list
>
> > > ofw at lists.openfabrics.org
>
> > > _http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw_
>
> > >
>
> >
>
> >
>



More information about the ofw mailing list