On 20081011, Khem RAJ writes:
You have to bar gcc trying to build unwinding at stage 1.

/trunk/patches/gcc/4.2.0/290-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.2.1/290-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.2.2/290-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.3.1/350-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.2.3/310-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.3.2/350-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
/trunk/patches/gcc/4.2.4/310-sh-without-headers.patch | 26 26 0 0 ++++++++++++++++++++++++++
7 files changed, 182 insertions(+)

 Original patch from H.J. Lu, as reported there:

 http://www.nabble.com/PATCH:-Use-short-for-fnstsw-td14775171.html

 fnstsw stores 16bit into %ax. The upper 16bit of %eax is unchanged.

 The new assembler (binutils-2.18.50.0.4 and up) will disallow "fnstsw %eax".

 diff -dur glibc-2.6.1.orig/sysdeps/i386/fpu/ftestexcept.c glibc-2.6.1/sysdeps/i386/fpu/ftestexcept.c

 --- glibc-2.6.1.orig/sysdeps/i386/fpu/ftestexcept.c	2004-03-05 11:14:48.000000000 +0100

 +++ glibc-2.6.1/sysdeps/i386/fpu/ftestexcept.c	2008-06-24 10:56:14.000000000 +0200

 @@ -26,7 +26,7 @@

  int

  fetestexcept (int excepts)

  {

 -  int temp;

 +  short temp;

    int xtemp = 0;

    /* Get current exceptions.  */