[openib-general] Re: Some CM Comments

[Sean Hefty]

Hal Rosenstock wrote:
> 2. The following CM states appear to be missing:
> peer compare
> reject retry
> timeout
> DREQ timeout (is an event) It is also a state in the spec
> RTU timeout (nor is it an event)

> 3. There is no REJ received CM event type. Shouldn't there be ?

I'll add in needed CM states when I get to the relevant portions of the 
code.  I may have left out some of the states, since they are temporary 
and would transition the connection object immediately to a new state. 
  E.g. I don't think that we really need a peer compare state.

> 5. REQ message is missing transport service type (connection type).
> Also, what about ack timeout ? Also, SRQ. Is starting PSN chosen
> internally by the CM ?

The CM can get the transport service type from the QP type.  I have a 
note to add in SRQ based on our previous discussions, but we need to 
add in verb support as well.

I need to think about the PSN.  The Topspin CM generated the PSN 
internal to the CM, but I'm not certain that it makes sense to still do 
this.

> Is there a danger to have the client set the CM response timeout ? Also,
> is the lack of remote response timeout for LAP in inconsistent with this
> ?

Since the client is responsible for generating the response, I thought 
that it made more sense for it to set that value.  We could have the CM 
set a reasonable upper bound to what a client can provide.

The lack of a remote response timeout for LAP does seem inconsistent. 
It looks like the code could use the value from the REQ, but it may be 
better to expose this field.

Thanks for the feedback.  Once there's more code, most of these issues 
should get flushed out.

- Sean