Some changes to recent 2.3.4-pre kernels pointed out this bug. The mkswap tool on sparc64 would set the wrong bits in the swappage bitmap in the signature for version 0 swap partitions (the issue is that on big endian it matters what size the unit of words is for bitmaps). This is deadly, and if users report bugs where their UltraSparc's spit out messages such as "attempt to read past end of device" or similar, and the device number printed out turns out to be one of their swap devices, this is indeed the bug they are seeing. This is against the current 6.0 util-linux SRPM and builds just fine. Please put out an errata for this at next convenience, thanks a lot.
Fixed (by applying patch) in util-linux-2.9o-14.