RE: verb spec clarification

Caitlin,

Yes, the RDMA Read Response for a zero length read will obey the ordering rules. That is what allows it to be used as a fence which is its primary purpose. The only checks turned off (in the sections Paul quoted) are those that check the appropriateness of the memory region/window targeted since it doesn't really access the region/window.

A zero-length RDMA Read needs to respect the negotiated RDMA Read queue depth. If one arrives and there is no buffer it will cause an error. The resource checks related to that behavior apply.

Regards,
Pat

-----Original Message-----
From: [email protected] [mailto:[email protected]]On Behalf Of
Caitlin Bestler
Sent: Tuesday, 29 March, 2005 6:39 AM
To: Culley, Paul
Cc: Patricio Kaplan; [email protected]
Subject: Re: [rddp] verb spec clarification


Culley, Paul wrote:
> For Verbs, The answer is no.
> 
> Section 7.6.3 pg 112 near top
> "If the length of the access is zero, the RI MUST NOT perform any of the
> above checks on the Memory Region."
> 
> Section 7.10.6.2 pg 124, 2/3 down
> "If the length of the access is zero, the RI MUST NOT perform any of the
> above checks on the Memory Window."
> 
> Paul R. Culley
> HP Fellow
> 281-514-5543
>  

Putting aside any overly broad use of the phrase "above checks",
would you agree that the *only* thing special about zero length
reads is that the STag is not validated?

Specifically, the RDMA Read Response will still be ordered just
as the response to any other RDMA Read Request, correct?

Therefore, it will consume a resource (on either a connection
or wider scale). That resource SHOULD/MUST be properly accounted
for by *some* mechanism.

I can think of nothing in the wire specifications that would
justify suppressing of normal resource checking for zero length
reads. The wire protocol does not specify *how* the local
interface tracks resources, but the security draft requires
that they be managed somehow whenever they are used across
multiple connections or are in any way a resource more
limited than just host memory.

_______________________________________________
rddp mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/rddp