Note... modified my mjn3 to not conflict with the big endian arm patch.
Warning!!!  Only the linux target is aware of TARGET_ENDIAN_DEFAULT.
Also changed
  #define SUBTARGET_EXTRA_ASM_SPEC "\
  %{!mcpu=*:-mcpu=xscale} \
  %{mhard-float:-mfpu=fpa} \
  %{!mhard-float: %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}}"
to
  #define SUBTARGET_EXTRA_ASM_SPEC "\
  %{mhard-float:-mfpu=fpa} \
  %{!mhard-float: %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}}"
in gcc/config/arm/linux-elf.h.
#
# Submitted:
#
# Dimitry Andric <dimitry@andric.com>, 2004-05-01
#
# Description:
#
# Nicholas Pitre released this patch for gcc soft-float support here: 
# http://lists.arm.linux.org.uk/pipermail/linux-arm/2003-October/006436.html
#
# This version has been adapted to work with gcc 3.4.0.
#
# The original patch doesn't distinguish between softfpa and softvfp modes
# in the way Nicholas Pitre probably meant.  His description is:
#
# "Default is to use APCS-32 mode with soft-vfp.  The old Linux default for
# floats can be achieved with -mhard-float or with the configure
# --with-float=hard option.  If -msoft-float or --with-float=soft is used then
# software float support will be used just like the default but with the legacy
# big endian word ordering for double float representation instead."
#
# Which means the following:
#
# * If you compile without -mhard-float or -msoft-float, you should get
#   software floating point, using the VFP format.  The produced object file
#   should have these flags in its header:
#
#     private flags = 600: [APCS-32] [VFP float format] [software FP]
#
# * If you compile with -mhard-float, you should get hardware floating point,
#   which always uses the FPA format.  Object file header flags should be:
#
#     private flags = 0: [APCS-32] [FPA float format]
#
# * If you compile with -msoft-float, you should get software floating point,
#   using the FPA format.  This is done for compatibility reasons with many
#   existing distributions.  Object file header flags should be:
#
#     private flags = 200: [APCS-32] [FPA float format] [software FP]
#
# The original patch from Nicholas Pitre contained the following constructs:
#
#   #define SUBTARGET_EXTRA_ASM_SPEC "%{!mcpu=*:-mcpu=xscale} \
#     %{mhard-float:-mfpu=fpa} \
#     %{!mhard-float: %{msoft-float:-mfpu=softfpa;:-mfpu=softvfp}}"
#
# However, gcc doesn't accept this ";:" notation, used in the 3rd line.  This
# is probably the reason Robert Schwebel modified it to:
#
#   #define SUBTARGET_EXTRA_ASM_SPEC "%{!mcpu=*:-mcpu=xscale} \
#     %{mhard-float:-mfpu=fpa} \
#     %{!mhard-float: %{msoft-float:-mfpu=softfpa -mfpu=softvfp}}"
#
# But this causes the following behaviour:
#
# * If you compile without -mhard-float or -msoft-float, the compiler generates
#   software floating point instructions, but *nothing* is passed to the
#   assembler, which results in an object file which has flags:
#
#     private flags = 0: [APCS-32] [FPA float format]
#
#   This is not correct!
#
# * If you compile with -mhard-float, the compiler generates hardware floating
#   point instructions, and passes "-mfpu=fpa" to the assembler, which results
#   in an object file which has the same flags as in the previous item, but now
#   those *are* correct.
#    
# * If you compile with -msoft-float, the compiler generates software floating
#   point instructions, and passes "-mfpu=softfpa -mfpu=softvfp" (in that
#   order) to the assembler, which results in an object file with flags:
#
#   private flags = 600: [APCS-32] [VFP float format] [software FP]
#
#   This is not correct, because the last "-mfpu=" option on the assembler
#   command line determines the actual FPU convention used (which should be FPA
#   in this case).
#
# Therefore, I modified this patch to get the desired behaviour.  Every
# instance of the notation:
#
#   %{msoft-float:-mfpu=softfpa -mfpu=softvfp}
#
# was changed to:
#
#   %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}
#
# I also did the following:
# 
# * Modified all TARGET_DEFAULT macros I could find to include ARM_FLAG_VFP, to
#   be consistent with Nicholas' original patch.
# * Removed any "msoft-float" or "mhard-float" from all MULTILIB_DEFAULTS
#   macros I could find.  I think that if you compile without any options, you
#   would like to get the defaults. :)
# * Removed the extra -lfloat option from LIBGCC_SPEC, since it isn't needed
#   anymore.  (The required functions are now in libgcc.)

diff -urN gcc-3.4.1-old/gcc/config/arm/coff.h gcc-3.4.1/gcc/config/arm/coff.h
--- gcc-3.4.1-old/gcc/config/arm/coff.h	2004-02-24 08:25:22.000000000 -0600
+++ gcc-3.4.1/gcc/config/arm/coff.h	2004-09-02 21:51:15.000000000 -0500
@@ -31,11 +31,16 @@
 #define TARGET_VERSION fputs (" (ARM/coff)", stderr)
 
 #undef  TARGET_DEFAULT
-#define TARGET_DEFAULT (ARM_FLAG_SOFT_FLOAT | ARM_FLAG_APCS_32 | ARM_FLAG_APCS_FRAME | ARM_FLAG_MMU_TRAPS)
+#define TARGET_DEFAULT		\
+	( ARM_FLAG_SOFT_FLOAT	\
+	| ARM_FLAG_VFP		\
+	| ARM_FLAG_APCS_32	\
+	| ARM_FLAG_APCS_FRAME	\
+	| ARM_FLAG_MMU_TRAPS )
 
 #ifndef MULTILIB_DEFAULTS
 #define MULTILIB_DEFAULTS \
-  { "marm", "mlittle-endian", "msoft-float", "mapcs-32", "mno-thumb-interwork" }
+  { "marm", "mlittle-endian", "mapcs-32", "mno-thumb-interwork" }
 #endif
 
 /* This is COFF, but prefer stabs.  */
diff -urN gcc-3.4.1-old/gcc/config/arm/elf.h gcc-3.4.1/gcc/config/arm/elf.h
--- gcc-3.4.1-old/gcc/config/arm/elf.h	2004-02-24 08:25:22.000000000 -0600
+++ gcc-3.4.1/gcc/config/arm/elf.h	2004-09-02 21:51:15.000000000 -0500
@@ -46,7 +46,9 @@
 
 #ifndef SUBTARGET_ASM_FLOAT_SPEC
 #define SUBTARGET_ASM_FLOAT_SPEC "\
-%{mapcs-float:-mfloat} %{msoft-float:-mfpu=softfpa}"
+%{mapcs-float:-mfloat} \
+%{mhard-float:-mfpu=fpa} \
+%{!mhard-float: %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}}"
 #endif
 
 #ifndef ASM_SPEC
@@ -106,12 +108,17 @@
 #endif
 
 #ifndef TARGET_DEFAULT
-#define TARGET_DEFAULT (ARM_FLAG_SOFT_FLOAT | ARM_FLAG_APCS_32 | ARM_FLAG_APCS_FRAME | ARM_FLAG_MMU_TRAPS)
+#define TARGET_DEFAULT		\
+	( ARM_FLAG_SOFT_FLOAT	\
+	| ARM_FLAG_VFP		\
+	| ARM_FLAG_APCS_32	\
+	| ARM_FLAG_APCS_FRAME	\
+	| ARM_FLAG_MMU_TRAPS )
 #endif
 
 #ifndef MULTILIB_DEFAULTS
 #define MULTILIB_DEFAULTS \
-  { "marm", "mlittle-endian", "msoft-float", "mapcs-32", "mno-thumb-interwork", "fno-leading-underscore" }
+  { "marm", "mlittle-endian", "mapcs-32", "mno-thumb-interwork", "fno-leading-underscore" }
 #endif
 
 #define TARGET_ASM_FILE_START_APP_OFF true
diff -urN gcc-3.4.1-old/gcc/config/arm/linux-elf.h gcc-3.4.1/gcc/config/arm/linux-elf.h
--- gcc-3.4.1-old/gcc/config/arm/linux-elf.h	2004-09-02 21:50:52.000000000 -0500
+++ gcc-3.4.1/gcc/config/arm/linux-elf.h	2004-09-02 22:00:49.000000000 -0500
@@ -44,12 +44,26 @@
 #define TARGET_LINKER_EMULATION "armelf_linux"
 #endif
 
-/* Default is to use APCS-32 mode.  */
+/*
+ * Default is to use APCS-32 mode with soft-vfp.
+ * The old Linux default for floats can be achieved with -mhard-float
+ * or with the configure --with-float=hard option.
+ * If -msoft-float or --with-float=soft is used then software float 
+ * support will be used just like the default but with the legacy
+ * big endian word ordering for double float representation instead.
+ */
 #undef  TARGET_DEFAULT
-#define TARGET_DEFAULT \
-		( ARM_FLAG_APCS_32 | \
-		  ARM_FLAG_MMU_TRAPS | \
-		  TARGET_ENDIAN_DEFAULT )
+#define TARGET_DEFAULT		\
+	( ARM_FLAG_APCS_32	\
+	| ARM_FLAG_SOFT_FLOAT	\
+	| TARGET_ENDIAN_DEFAULT	\
+	| ARM_FLAG_VFP		\
+	| ARM_FLAG_MMU_TRAPS )
+
+#undef  SUBTARGET_EXTRA_ASM_SPEC
+#define SUBTARGET_EXTRA_ASM_SPEC "\
+%{mhard-float:-mfpu=fpa} \
+%{!mhard-float: %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}}"
 
 #define SUBTARGET_CPU_DEFAULT TARGET_CPU_arm6
 
@@ -57,7 +71,7 @@
 
 #undef  MULTILIB_DEFAULTS
 #define MULTILIB_DEFAULTS \
-	{ "marm", TARGET_ENDIAN_OPTION, "mhard-float", "mapcs-32", "mno-thumb-interwork" }
+	{ "marm", TARGET_ENDIAN_OPTION, "mapcs-32", "mno-thumb-interwork" }
 
 #define CPP_APCS_PC_DEFAULT_SPEC "-D__APCS_32__"
 
@@ -72,7 +86,7 @@
    %{shared:-lc} \
    %{!shared:%{profile:-lc_p}%{!profile:-lc}}"
 
-#define LIBGCC_SPEC "%{msoft-float:-lfloat} -lgcc"
+#define LIBGCC_SPEC "-lgcc"
 
 /* Provide a STARTFILE_SPEC appropriate for GNU/Linux.  Here we add
    the GNU/Linux magical crtbegin.o file (see crtstuff.c) which
diff -urN gcc-3.4.1-old/gcc/config/arm/t-linux gcc-3.4.1/gcc/config/arm/t-linux
--- gcc-3.4.1-old/gcc/config/arm/t-linux	2003-09-20 16:09:07.000000000 -0500
+++ gcc-3.4.1/gcc/config/arm/t-linux	2004-09-02 21:51:15.000000000 -0500
@@ -4,7 +4,10 @@
 LIBGCC2_DEBUG_CFLAGS = -g0
 
 LIB1ASMSRC = arm/lib1funcs.asm
-LIB1ASMFUNCS = _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_lnx
+LIB1ASMFUNCS = _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_lnx \
+	_negdf2 _addsubdf3 _muldivdf3 _cmpdf2 _unorddf2 _fixdfsi _fixunsdfsi \
+	_truncdfsf2 _negsf2 _addsubsf3 _muldivsf3 _cmpsf2 _unordsf2 \
+	_fixsfsi _fixunssfsi
 
 # MULTILIB_OPTIONS = mhard-float/msoft-float
 # MULTILIB_DIRNAMES = hard-float soft-float
diff -urN gcc-3.4.1-old/gcc/config/arm/unknown-elf.h gcc-3.4.1/gcc/config/arm/unknown-elf.h
--- gcc-3.4.1-old/gcc/config/arm/unknown-elf.h	2004-02-24 08:25:22.000000000 -0600
+++ gcc-3.4.1/gcc/config/arm/unknown-elf.h	2004-09-02 21:51:15.000000000 -0500
@@ -30,7 +30,12 @@
 
 /* Default to using APCS-32 and software floating point.  */
 #ifndef TARGET_DEFAULT
-#define TARGET_DEFAULT	(ARM_FLAG_SOFT_FLOAT | ARM_FLAG_APCS_32 | ARM_FLAG_APCS_FRAME | ARM_FLAG_MMU_TRAPS)
+#define TARGET_DEFAULT		\
+	( ARM_FLAG_SOFT_FLOAT	\
+	| ARM_FLAG_VFP		\
+	| ARM_FLAG_APCS_32	\
+	| ARM_FLAG_APCS_FRAME	\
+	| ARM_FLAG_MMU_TRAPS )
 #endif
 
 /* Now we define the strings used to build the spec file.  */
diff -urN gcc-3.4.1-old/gcc/config/arm/xscale-elf.h gcc-3.4.1/gcc/config/arm/xscale-elf.h
--- gcc-3.4.1-old/gcc/config/arm/xscale-elf.h	2003-07-01 18:26:43.000000000 -0500
+++ gcc-3.4.1/gcc/config/arm/xscale-elf.h	2004-09-02 21:51:15.000000000 -0500
@@ -49,11 +49,12 @@
 		     endian, regardless of the endian-ness of the memory
 		     system.  */
 		     
-#define SUBTARGET_EXTRA_ASM_SPEC "%{!mcpu=*:-mcpu=xscale} \
-  %{mhard-float:-mfpu=fpa} \
-  %{!mhard-float: %{msoft-float:-mfpu=softfpa;:-mfpu=softvfp}}"
+#define SUBTARGET_EXTRA_ASM_SPEC "\
+%{!mcpu=*:-mcpu=xscale} \
+%{mhard-float:-mfpu=fpa} \
+%{!mhard-float: %{msoft-float:-mfpu=softfpa} %{!msoft-float:-mfpu=softvfp}}"
 
 #ifndef MULTILIB_DEFAULTS
 #define MULTILIB_DEFAULTS \
-  { "mlittle-endian", "mno-thumb-interwork", "marm", "msoft-float" }
+  { "mlittle-endian", "mno-thumb-interwork", "marm" }
 #endif