[ofa-general] Re: [ewg] rhel 5.2 iSER support?

[Or Gerlitz]

Sameer Mehta wrote:
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: iser_connect:connecting 
> to: 192.168.0.5, port 0xbc0c
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: iser_cma_handler:event 0 
> conn ffff81015de00bc0 id ffff81017fc8e200
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: iser_cma_handler:event 2 
> conn ffff81015de00bc0 id ffff81017fc8e200
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: 
> iser_create_ib_conn_res:setting conn ffff81015de00bc0 cma_id 
> ffff81017fc8e200: fmr_pool ffff810140c9aec0 qp ffff810168974e00
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: iser_cma_handler:event 8 
> conn ffff81015de00bc0 id ffff81017fc8e200
> Dec  2 16:44:52 nws-bur-25-46 kernel: iser: iser_cma_handler:event: 8, 
> error: 8
>
> Am I missing something here? is iSER transport available in v1.4?
You are getting REJECTED (8) event with the reject reason being 
INVALID_SERVICE_ID (8), see include/rdma/ib_cm.h. This means there's no 
one listening on the Service-ID you are attempting to connect to, eg 
your target didn't issue a  listen call on the SID (service id) you are 
trying to connect to or there's some mismatch is the SID as constructed 
by the initiator, etc.

Related inter-op issue has been brought by Jesse Butler from Sun couple 
of months ago,  
http://lists.openfabrics.org/pipermail/general/2008-October/054487.html 
but I am not sure where it stands.

The code that builds the SID from the tcp port is cma_get_service_id 
(drivers/infiniband/core/cma.c, below) where in this case the resulted 
SID is 0x0000000001060cbc

Or.
> static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
> {
> 	return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
> }