Build for ARM 926T fails if option software floating point is enabled

[Rob Landley]

On Friday 07 September 2007 3:46:41 pm Pranav Sanghadia wrote:
> Hello Guys,
>
> I am new to embedded Linux, trying to build cross platform tools for ARM
> 920t. I got following error message when option "software floating point"
> is enabled. I found this issue reported as bug but do not see clear
> solution. Can some one please help me?

It works fine for me in http://landley.net/code/firmware

Currently the hg repository for that is screwed up by 2.6.23-rc5, I've only 
gotten four of the targets working with the new kernel yet.  I should have 
the rest working again by the time 2.6.23 final is released.

Try the "armv4l" target in the release version.  That uses soft float.


P.S.  You might have to add this patch to sources/patches (in current -hg it's 
sources/patches/uClibc-softarm.patch):

--- uClibc-0.9.29/Rules.mak
+++ uClibc-0.9.29-new/Rules.mak
@@ -378,9 +378,11 @@
 # soft float encodings.
 ifneq ($(TARGET_ARCH),nios)
 ifneq ($(TARGET_ARCH),nios2)
+ifneq ($(TARGET_ARCH),arm)
 CFLAGS += -msoft-float
 endif
 endif
+endif
 ifeq ($(TARGET_ARCH),arm)
 # No longer needed with current toolchains, but leave it here for now.
 # If anyone is actually still using gcc 2.95 (say), they can uncomment it.

(Apparently gcc 4.x now has two soft-float APIs, and -msoft-float tells gcc 
4.1 to use the soft-fpa API when libgcc was built with the default soft-vfp.  
It seems the toolchain uses soft float by default when it's built for it 
anyway (at least on gcc 4.x: the fix Peter has in his tree should still work 
with older versions of gcc...))

Rob
-- 
"One of my most productive days was throwing away 1000 lines of code."
  - Ken Thompson.