[ofw] Re: saquery & osm vendor AL - ca_names missing from osm_vendor_t ?

[Yevgeny Kliteynik]

Hi Stan,

Adding Sasha (OFED management maintainer)
and the openib mailing list.

Stan C. Smith wrote:
> Hello,
>   The Windows OpenSM vendor AL struct 'osm_vendor_t' (defined in opensm\user\include\vendor\osm_vendor_al.h) is missing
> the entry 'ca_names[UMAD_MAX_DEVICES][UMAD_CA_NAME_LEN]'.
> saquery.c expects to find ca_names in osm_vendor_t.
> 
> A couple of observations:
> 1) Windows currently supports a much older version of opensm than what OFED 1.4 tools expect.

Correct. Windows OpenSM is a ported pre-OFED 1.2 OpenSM with couple of minor fixes.

> 2) saquery uses OpenSM mad interfaces while it 'could' be using libibmad interfaces.

By "OpenSM mad interfaces" you mean libosmvendor?

>    If libibmad from saquery, then OpenSM would not need libibmad references for Windows.

Not sure what you mean here. You mean removing libibmad dependency from saquery?

> 3) How bad is it to create libibmad dependencies for OpenSM?

Pretty bad. I don't think we should add a new dependency unless there's a
really good reason for it.

> 4) saquery.c is the only diags pgms (so far) which uses OpenSM MAD interfaces; the rest use
>    libibmad.
>
> Most of the OFED diagnostic tools support the cmd-line option '-C ca_name'. This cmd-line input is resolved thru
> libibmad interfaces.
> Saquery is no exception in that it expects to match the '-C ca_name' against osm_vendor_t.ca_names[]. 'ibstat -l' lists
> CA names.
> 
> The question becomes how best to resolve the missing ca_names?
> 
> 1) modify saquery to call libibmad interface to get CA names; leaves osm_vendor_t unmodified.
>    So far, saquery is the only diag pgm which uses OSM mad interfaces; expecting ca_names
>    in osm_vendor_t.
> 
> 2) Modify OpenSM vendor AL osm_vendor_t struct to include CA names and populate ca_names
>    from OpenSM code?

I'd say that this option is much better.

>    Creates libibmad dependencies for opensm.

But it doesn't have to. Can IBAL expose some function to get these names,
so that Win osmvendor will use this function instead of libibmad?

Also, Linux osmvendor doesn't have libibmad dependency.
It uses umad function umad_get_cas_names() to obtain the CA names.
I know that there is a Windows version of umad, but I don't know what is
its status. If we *have* to add an additional dependency, then it should
be libibumad and not libibmad.

At some point in the future we would really want to have the new version
of OFED OpenSM ported to WinOF. If there will be a match between Linux and
Windows libraries, then the whole vendor concept can be simplified and
there won't be a need to have a separate vendor for IBAL. The things
that would be different are platform-dependent issues like threads, locks,
syslog, but not IB-related code.

-- Yevgeny


> Comments?
> 
> Thanks,
> 
> Stan.
> 
> 
> 
>