jcm asked for details of the omap issue that we've fixed. The find
what by shear luck.
Basically I was working on getting a 3.5 kernel building for all
arches on rawhide. I ran into an issue with "arm_memblock_steal" so
was googling to try and work out a fix. I came across the following
thread from Russell about the introduction of arm_memblock_steal and
he mentioned in this post  where "errata i688" was being marked
BROKEN due to it's use of memory (see thread) and with a little
reading it looks to like our issue to at least not try. So pinged it
to dgilmore and we both tested in parallel :-) With other searching
today this  thread also covers it pretty well, and the commit
message is pretty damning :-)
A side note also that in the short term IMX 3.5 kernels are broken
until this patch  land likely very soon.