[openib-general] smpdump and current MAD layer

[Sean Hefty]

Hal Rosenstock wrote:

>>This is something that was briefly discussed before.  I think that I 
>>would support snooping by extending the ib_mad_reg_reg structure to 
>>indicate a registration type, possibly along with some additional 
>>filtering parameters.  (We could also create a new snoop routine.)
> 
> Maybe a single bit field in the registration request to indicate snoop.
> 
> Another question is what is the granularity of snoop registrations
> needed to be supported ? Is one SMP snooper and one GMP snooper
> sufficient ? Should the snoopers be per class ? It seems to me that
> going down to the method level is too much for snoopers. This is just
> another way of expressing the filtering parameters you mention.

I guess filtering can be done above the MAD layer, so just letting the 
user specify the qp_type may be all that's needed, beyond indicating 
that snooping is desired.  If we go this route, we can probably support 
any number of snoopers.

>>One issue with snooping MAD is whether the snooping occurs above or 
>>below RMPP, or possibly in both places.
> 
> In general, I would think the GMP snooping would specify whether it is
> to be done above or below RMPP and perhaps the class or all GS classes
> (some combinations wouldn't make sense). I could see if one thought one
> was having problems with RMPP handling doing it below RMPP and otherwise
> above (the normal case).

Hmm... we could let the client decide through the rmpp_version 
parameter.  Also, would snooping include redirected QPs?  I think that 
we can support this.

- Sean