[ofa-general] [PATCH RFC] RDMA/cxgb3: Fixes for zero-stag.

[Steve Wise]

Or Gerlitz wrote:
> Steve Wise wrote:
>> NFSRDMA was planning on using the zero stag for recvs on iwarp 
>> transports.  This boils down to only needing the zero stag + fastreg 
>> mrs for NFSRDMA server over T3, avoiding dma_mrs which have limits in 
>> the T3 HW.
>>> the zero STag breaks the semantics slightly in that there is no 
>>> longer any PD attached to the key returned by ib_get_dma_mr 
> Steve,
>
> I am fine with the approach of new verb, however, I still want to make 
> sure that this can be used under a unified IB/iWARP design of a ULP 
> without much pain... the kernel ULPs: rNFS, iSER initiator (Lustre, 
> RDS, etc) use the dma mr for recv nd send work-requests.
>
> So how you envision the ULP code in the presence of the new verb - is 
> it the ULP to check if the device supports the capability and if yes, 
> use dev->local_dma_lkey else call get_dma_mr and use the returned lkey?
>

Yes.

> in you comment above you say "using the zero stag for recvs on iwarp 
> transports" does zero stag applies also to sends? if not, will sends 
> use dma-mr?
>
>
Yes, NFSRDMA server will use this local_dma_lkey and/or a dma_mr for 
local access SGES.  That includes recvs, sends, and the source of writes.

Steve.