help about cross-compiler

Thu Jul 12 02:58:00 GMT 2001

On Thu, 12 Jul 2001 13:13:26 +0800
"Liu Yong" <yliu@newavetech.com> wrote:

> I think I haven't made things clear.
> I have succeeded in building core-gcc as you described 
> in your mail.
> And succeeded building newlib.

 I would disagree ;-)

> But failed to build cross-gcc,the last step.
> configure: error: installation or configuration problem: C compiler
> cannot create executables.

 This is somehow misleading, the linking phase fails, not the compiler.
Using '--with-newlib' in configure should have worked around the sanity
check in libiberty-configure (where this error comes). Newlib is a known
thing so scanning the 'libc.a' in it can be left undone.

 Before one can create executables for the real target, one must have a
working toolset with a proper linker script, startup, libs and sane
instructions for the link-phase in the 'specs'-file...

> I have looked at the directory /home/yliu/local/m68k-hp-bsd4.3/bin/  ,
> /home/yliu/local/m68k-hp-bsd4.3/lib/ and
> /home/yliu/local/m68k-hp-bsd4.3/include ,there are files in there:
> bin: ar as c++ g++ gcc ld nm ranlib strip
> lib: bcc.ld crt0.o idp.ld libgcc.a libdbug.a libidp.a libmvme135.a
> libmvme162.a libnosys.a mvme135.ld mvme162.ld sbc5204.ld sbc5206.ld

 The C-library ('libc.a') and the C-math library ('libm.a') are missing.
Building and installing newlib should have produced them and they should
be in the 'lib'-subdir... Also subdirs like 'm68000', 'm5200', 'm68040',
'm68060', 'mcpu32', 'msoft-float' etc. should exist for the CPU/FPU-variations.

 Your current 'just some m68k-target' / 'm68k-hp-bsd4.3' doesn't produce
anything else than the '-m68020 -m68881' variation libraries...

 I would suggest using some more common 'generic' target like 'm68k-aout',
'm68k-coff' or 'm68k-elf' for the toolset. And to think about which ones
in the m68k-family should be supported... The BSVC-simulator supports
m68000 and cpu32 (m68360), so if this sim looks promising, the m68000 and
mcpu32 support libs should be produced.  The 'gcc/config/m68k/t-m68kbare'
controls the produced 'multilibs', changing the MULTILIB* definitions
there sets the compiler to support just the CPUs you will want it to
support...

 Or to use a BSVC-specific target 'm68k-bsvc2.1' as described later...

> include:  there is no include directory ,just a sys-include directory that
> contains a file named COPIED (file length is 3 bytes, just has a 
> word 'no')

 Copying the newlib headers into the 'include' shouldn't be a problem, just
use something like:

 cp -r -p -v newlib-1.9.0/newlib/libc/include /home/yliu/local/m68k-hp-bsd4.3

now and rename the 'm68k-hp-bsd4.3' into something else ('m68k-coff',...) if
you change the $target but keep the $prefix...

 The 'sys-include' can be used as the place for the '3rd-party' headers from
Motorola etc. for the chips and their I/O-addresses and specific bit names...

 BTW, I finally decided to make a specific target toolset, 'm68k-bsvc2.1',
for the BSVC 2.1 simulators. So I added this new target to gcc-2.95.3, it
being an alias to the 'm68k-aout', then did the same for newlib-1.9.0 and
built the toolset. Must make an Insight/GDB for this toolset too...  Of
course I could have decided to use my existing 'm68k-coff' and 'm68k-elf'
toolsets with BSVC, but when missing a 'm68k-aout' from my collection, this
was a good enough reason to add it...  The next phase could be to build
glue libraries, 'libbsvc.a's, for the 'BSVC' m68000 and m68360 'boards'...

 Adding the 'm68k-bsvc*' target into gcc-2.95.3 happens as follows:

 1. Edit the main 'config.sub' by adding a ' | -bsvc*' somewhere where the
    script checks for the known 'systems', like :

--------------------- clip -----------------------------------------------
*** config.bak	Mon Jun 18 18:44:14 2001
--- config.sub	Tue Jul 10 19:59:56 2001
***************
*** 940,946 ****
  	      | -*vms* | -sco* | -esix* | -isc* | -aix* | -sunos | -sunos[34]*\
  	      | -hpux* | -unos* | -osf* | -luna* | -dgux* | -solaris* | -sym* \
  	      | -amigaos* | -amigados* | -msdos* | -newsos* | -unicos* | -aof* \
! 	      | -aos* | -djgpp2* \
  	      | -nindy* | -vxsim* | -vxworks* | -ebmon* | -hms* | -mvs* \
  	      | -clix* | -riscos* | -uniplus* | -iris* | -rtu* | -xenix* \
  	      | -hiux* | -386bsd* | -netbsd* | -openbsd* | -freebsd* | -riscix* \
--- 940,946 ----
  	      | -*vms* | -sco* | -esix* | -isc* | -aix* | -sunos | -sunos[34]*\
  	      | -hpux* | -unos* | -osf* | -luna* | -dgux* | -solaris* | -sym* \
  	      | -amigaos* | -amigados* | -msdos* | -newsos* | -unicos* | -aof* \
! 	      | -aos* | -djgpp2* | -bsvc* \
  	      | -nindy* | -vxsim* | -vxworks* | -ebmon* | -hms* | -mvs* \
  	      | -clix* | -riscos* | -uniplus* | -iris* | -rtu* | -xenix* \
  	      | -hiux* | -386bsd* | -netbsd* | -openbsd* | -freebsd* | -riscix* \
--------------------- clip -----------------------------------------------

 2. Add the '| m68k-*-bsvc*' into the 'm68k-*-aout*' config-entry in 'gcc/configure' :

--------------------- clip -----------------------------------------------
*** configure.bak	Tue Jul  3 15:31:22 2001
--- configure	Tue Jul 10 19:47:51 2001
***************
*** 4413,4427 ****
  		extra_headers=math-68881.h
  		;;
  	m68k-wrs-vxworks*)
- 		tm_file=m68k/vxm68k.h
  		tmake_file=m68k/t-vxworks68
  		extra_headers=math-68881.h
- 		thread_file='vxworks'
  		float_format=m68k
  		;;
! 	m68k-*-aout*)
  		tmake_file=m68k/t-m68kbare
! 		tm_file="m68k/m68k-aout.h libgloss.h"
  		extra_headers=math-68881.h
  		float_format=m68k
  		;;
--- 4413,4427 ----
  		extra_headers=math-68881.h
  		;;
  	m68k-wrs-vxworks*)
  		tmake_file=m68k/t-vxworks68
+ 		tm_file=m68k/vxm68k.h
  		extra_headers=math-68881.h
  		float_format=m68k
+ 		thread_file='vxworks'
  		;;
! 	m68k-*-aout* | m68k-*-bsvc*)
  		tmake_file=m68k/t-m68kbare
! 		tm_file="m68k/m68k-aout.h"
  		extra_headers=math-68881.h
  		float_format=m68k
  		;;
--------------------- clip -----------------------------------------------

 The stuff order in the 'm68k-wrs-vxworks*' was also rearranged for my
artistic eyes...

 Editing the 'config.sub' in 'newlib-1.9.0' sources quite in the same
way can be left as an homework... No other changes to newlib needed.

 Perhaps a totally 'new' target for GCC could have been added, like:

 	m68k-*-bsvc*)
  		tmake_file=m68k/t-m68kbsvc
 		tm_file="m68k/m68k-aout.h"
  		extra_headers=math-68881.h
  		float_format=m68k
  		;;

and the 't-m68kbsvc' (using 't-m68kbare' as the template) edited to
produce only the extra 'm68000' support with this target, the 'mcpu32'
being the default option (not the 'm68020' and 'm68881' as in the plain
vanilla GCC-sources)...

 The 'm68332' or 'mcpu32' can be set as the default CPU-type in
'gcc/config/m68k/m68k-none.h' and I think the things to do this were
handled on this list earlier, but a quick reminder here would be to edit
the :

  #ifndef TARGET_CPU_DEFAULT
  #define TARGET_CPU_DEFAULT  M68K_CPU_m68332
  #endif

in the beginning of 'm68k-none.h' to be as shown...

Cheers, Kai

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com