Hi all: I was asked to use floating point in kernel module, and it results in many undefined reference to functions begin with __aeabi. When I supply libgcc.a, there remains only one undefined reference to 'raise'. After consulting Google I found many people confront this problem when building u-boot with gcc, it seems libgcc.a need user space C runtim environment So I tried to fix it by supply this symbol,but insmod failed due to that kernel cannot handle its relocation type. After change some code in kernel it finally worked. I thought it all caused by soft-float library in libgcc.a, but then I noticed functions referencing to 'raise' is actually emitted for divide operation. I wrote a test module involved only complex integer divide(which cannot be optimized by gcc) and the 'raise' problem raised again. So it seems we do not use libgcc.a when building kernel (module), does that mean we can only use those simpy divide operations (which can be traslated to other instructions by gcc without referencing to library functions) in kernel(module)? By the way, if I have to use soft-float library in kernel space anyway, though sounds bad, it should work, right? Does anyone do that before?
