[openib-general] Re: Dual Sided RMPP Support as well as OpenSM Implications

[Hal Rosenstock]

On Mon, 2006-04-10 at 12:34, Roland Dreier wrote:
>     Hal> Right now, it needs to be burnt into the MAD core as there is
>     Hal> no other way to determine this (as it is class/method
>     Hal> dependent). I have made a spec comment on this lack of
>     Hal> flexibility but don't have a good solution for this at least
>     Hal> yet...
> 
> Now you've lost me.  What if one process sets the ioctl and another
> process doesn't?  What happens to dual-sided RMPPs then?

It's a readable thing rather than a settable thing.

> If it's really a system-wide setting

It is.

> then I guess it should be a file under /sys/class/infiniband_mad.

Ah, that's better than an ioctl and no file would certainly be the
backward compatibility mode (no dual sided RMPP support).

-- Hal