[ofa-general] ib_umad method mask problems on big-endian 64-bit archs

[Roland Dreier]

 > Do we have another another user_umad users (OFA or another known, where
 > switch could be painful)? If not, I will prefer this way instead of
 > keeping two OpenSMs for ppc64.

I agree.  After thinking it over, the least painful thing to do to
update "documentation" (ib_user_mad.h) to match what the kernel does
now for 64-bit big endian systems, and fix the case of 32-bit big
endian userspace on a 64-bit kernel.

I'll post a patch for 2.6.24.