[openib-general] Re: 2/2 libibverbs + libmthca changes for query QP

[Caitlin Bestler]

openib-general-bounces at openib.org wrote:
> Roland Dreier wrote:
>> This reminds me of something I've been meaning to ask: what is the
>> attr_mask of the query QP function used for?  The verbs definition in
>> the IB spec does not include an attribute mask.  Obviously we don't
>> have to follow the spec precisely but I'm wondering what the
>> advantage is in having an attribute mask for query QP.  None of the
>> other query methods (including query SRQ) have an attribute mask.
> 
> My understanding of the original code that this came from is
> that it permitted querying attributes that may be stored in
> memory, rather than requiring reading the card.  E.g. if all
> you needed was the QPN, then you could obtain only that
> information.  Given that things like the QPN, QP type, etc.
> are already available through the API, I'm not sure that the mask is
> needed. 
> 

Anticipating what queries are expensive (require a round-trip
to the card) is inherently model dependent. Having an attribute
mask allows the consumer to state *exactly* what informatino is
required, and allowing each driver to decide whether a round-trip
to the card is needed for *this* set of information.

It should be clear that the provider MAY supply answers that
were not requested but SHOULD NOT take extra steps to obtain
that information. That is, a provider could first copy all
the data in had on-host, and then check a flag to see if
a round-trip to the device was required. If so it might
just copy everything from the device.

This style was selected in both DAT and IT-API because it 
simpler to use this type of attribute uniformly rather than
expecting the consumer to remember which queries might be
issues on *some* devices.