I debated whether I should mention CVM_FORCE_HARD_FLOAT = true. For one, it is not officially supported. We've never tested or released a product with it. We did some experimenting with it on the netwinders a few years back, and after some effort still had 20 or so TCK failures. Perhaps these are resolved by newer versions of the kernel softfloat library and gcc, but I'm not sure. The other reason I didn't mention it is because of the performance issues I brought up.
You might want to still consider the -msoft-float option. I've been told newer toolchains do not require -lfloat, so removing this should resolve the problem you ran into.
[Message sent by forum member 'cjplummer' (cjplummer)]