[ofa-general] [RFC] [PATCH 0/3] 2.6.23: basic support for IB routers

Hal Rosenstock halr at voltaire.com
Mon May 21 10:26:33 PDT 2007


On Fri, 2007-05-18 at 18:14, Sean Hefty wrote:
> Re-sending - typo in mailing list name...
> 
> I'd like to get feedback about incorporating the following changes to support
> IB routers into 2.6.23.  The goal of the patches is to allow for IB router
> development and prototyping within the current framework of IBA.  The changes
> themselves are fairly minimal, but based on the following concepts:
> 
> * Routing data is maintained by the local SA.  No assumption is made regarding
>   how the SA obtains routing information.  The SA is only expected to respond
>   to cross subnet PR queries by providing a path to the local router.  This
>   matches the behavior in opensm.
> 
> * A ULP connecting to a remote subnet provides path information about both
>   subnets.  For now the implementation simply assumes that the properties of
>   the remote path match that of the local path.  This allows the active side
>   CM to properly format the CM REQ.
> 
> * If the SLID/DLID values in the CM REQ are set to the permissive LID, then
>   the passive side CM uses the SLID/DLID/SL values from the received CM REQ
>   LRH to configure the passive side QP.  This is done to meet C9-54 without
>   requiring communication with the remote SA, but I should note that this
>   behavior is non-compliant.

Should there be some conditionalization of any non IBA compliant code so
it is only turned on if someone really wants this ?

I presume this is to be replaced by the real code some time in the
future once the IBA spec for these router issues is decided.

-- Hal

> These changes were tested by establishing a connection and transferring data
> between two IB subnets connected by an Obsidian router.
> 
> These patches are also available in the ib_router branch of my rdma-dev.git
> tree.  The tree is based on 2.6.21, so include a couple of additional patches
> that were already pushed for 2.6.22.
> 
> Signed-off-by: Sean Hefty <sean.hefty at intel.com>
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
> 
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general




More information about the general mailing list