Hi Hal,

On 06:18 Mon 15 Oct     , Hal Rosenstock wrote:
> 
> I don't recall as this is from a very very long time ago but in looking
> at this, I agree with your assessment that it can be simplified (and
> there appears to be no real need for what is contained in struct Port
> other than the fd). The only downside I see is the subtle change in the
> public umad_ APIs changing int portid -> int fd.

There is no API change at all - umad_open_port() still return unique
integer descriptor as it was before. Here we are only changing
undocumented (at least I'm not able to find any public description about
what umad_open_port() should return) behavior of this API (by replacing
mad device number as umad_open_port() return value, but if we want to
support multiple open()s there is no choice - device number is not
suitable for this).

> I suppose all the tools
> would continue to work without change here even if libibumad were
> changed underneath it, right ?

Right.

> BTW, when you do this, the umad man pages
> should all be updated for this change.

I see only that umad_open_port.3 should be fixed - it says that return
value is "0" on success, which is not correct anyway. Not really related
to the patch. Do you see another places to fix in man?

Sasha
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to