On 8/19/19 9:16 AM, Jason Gunthorpe wrote:
On Mon, Aug 19, 2019 at 08:45:58AM -0700, Bart Van Assche wrote:
On 8/19/19 8:17 AM, Jason Gunthorpe wrote:
On Mon, Aug 19, 2019 at 08:11:21AM -0700, Bart Van Assche wrote:
Does uniqueness of the I/O controller GUID only matter in InfiniBand
networks or does it also matter in other RDMA networks?
How about using 0 as default value for the srpt_service_guid in RoCE
networks?
How does SRP connection management even work on RoCE?? The CM MADs
still carry a service_id? How do the sides exchange the service ID to
start the connection? Or is it ultimately overriden in the CM to use
an IP port based service ID?
The ib_srpt kernel driver would have to set id_ext to a unique value if
srpt_service_guid would be zero since the SRP initiator kernel driver uses
the IOC GUID + id_ext + initiator_ext combination in its connection
uniqueness check (srp_conn_unique()).
Sounds like you should just generate something random for RDMA/CM mode ?
Still a bit confused how this is usable though if the initiating side
needs the service ID?
Hi Jason,
When I read Lijun Ou's e-mails for the first time I was assuming that my
patch had been tested on top of a recent kernel. After having reread
these e-mails I think my patch had been tested on top of kernel v4.14
and is not necessary for more recent kernels. So I think we can drop the
patch at the start of this e-mail thread.
Bart.
