Re: [RFA for trunk, 3.4 branch] enable mips-vxworks config.

[Eric Christopher <echristo at redhat dot com>]

The mips parts are OK. Need someone (I think Zack) to approve the
vxworks bits.

-eric

On Fri, 2004-06-04 at 23:43, cgd@broadcom.com wrote:
> The following patch enables mips-vxworks support.  Ideally, I'd like
> to see it approved for trunk and the 3.4 branch (it applies
> almost-cleanly to the branch -- just a few offset changes, IIRC).
> 
> With this patch, mips-vxworks-gcc works well enough to compile a
> couple of my favorite vxworks BSPs into working vxWorks kernels.  The
> mips-vxworks compiler still has to use the vxworks assembler and
> linker, since I didn't want to work through incompatibilities with the
> current binutils.
> 
> I've also verified that an i686-linux native compiler still builds w/
> the change.
> 
> Because of the config/vxworks.h problem, i don't think that ppc
> vxworks support works anyplace right now (can't test it myself), so
> i'm not worried about breaking it w/ the longlong.h changes.  But
> based on the #defines provided by the vxworks hdrs and by the rs6000
> backend, i think they're correct.
> 
> 
> 
> chris
> --
> 2004-06-04  Chris Demetriou  <cgd@broadcom.com>
> 
> 	* gcov-io.c (gcov_open) [GCOV_LOCKED]: Always supply three
> 	arguments to open().
> 
> 	* longlong.h: Don't assume that "PPC" being defined means PowerPC;
> 	on VxWorks it doesn't.  Check for '__ppc' for VxWorks.
> 
> 	* config/vxworks.h (HANDLE_SYSV_PRAGMA): Define with value "1"
> 	for consistency with other definitions.
> 
> 	* config/mips/vxworks.h (EXTRA_SECTIONS, EXTRA_SECTION_FUNCTIONS):
> 	Do not redefine.
> 	(SUBTARGET_CC1_SPEC): Use -mabi=o64 for 64-bit ISAs, if no other
> 	ABI is specified.
> 	(MDEBUG_ASM_SPEC): Redefine (now empty).
> 	
> Index: gcov-io.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/gcov-io.c,v
> retrieving revision 1.15
> diff -u -p -r1.15 gcov-io.c
> --- gcov-io.c	23 Feb 2004 17:02:50 -0000	1.15
> +++ gcov-io.c	5 Jun 2004 06:18:17 -0000
> @@ -84,8 +84,11 @@ gcov_open (const char *name, int mode)
>    gcov_var.endian = 0;
>  #endif
>  #if GCOV_LOCKED
> +  /* Some systems, e.g. VxWorks, prototype open() as having 3 arguments.
> +     To accomodate these, we supply the third argument here even when it's
> +     unnecessary.  This should cause no harm.  */
>    if (mode > 0)
> -    fd = open (name, O_RDWR);
> +    fd = open (name, O_RDWR, 0);
>    else
>      fd = open (name, O_RDWR | O_CREAT, 0666);
>    if (fd < 0)
> Index: longlong.h
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/longlong.h,v
> retrieving revision 1.43
> diff -u -p -r1.43 longlong.h
> --- longlong.h	16 Mar 2004 04:12:25 -0000	1.43
> +++ longlong.h	5 Jun 2004 06:18:18 -0000
> @@ -631,15 +631,17 @@ UDItype __umulsidi3 (USItype, USItype);
>  
>  /* FIXME: We should test _IBMR2 here when we add assembly support for the
>     system vendor compilers.
> -   FIXME: What's needed for gcc PowerPC VxWorks?  __vxworks__ is not good
> -   enough, since that hits ARM and m68k too.  */
> +   Note: on vxWorks, because of headers included before longlong.h
> +   (vxCpu.h), "PPC" will be defined for all architectures!  So, we exclude
> +   it here.  (Current) GCC defines __ppc for vxWorks.  */
>  #if (defined (_ARCH_PPC)	/* AIX */				\
>       || defined (_ARCH_PWR)	/* AIX */				\
>       || defined (_ARCH_COM)	/* AIX */				\
>       || defined (__powerpc__)	/* gcc */				\
>       || defined (__POWERPC__)	/* BEOS */				\
>       || defined (__ppc__)	/* Darwin */				\
> -     || defined (PPC)		/* GNU/Linux, SysV */			\
> +     || (defined (PPC) && !defined (__vxworks)) /* GNU/Linux, SysV */	\
> +     || defined (__ppc)		/* vxWorks */				\
>       ) && W_TYPE_SIZE == 32
>  #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
>    do {									\
> Index: config/vxworks.h
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/config/vxworks.h,v
> retrieving revision 1.5
> diff -u -p -r1.5 vxworks.h
> --- config/vxworks.h	24 Dec 2003 00:14:19 -0000	1.5
> +++ config/vxworks.h	5 Jun 2004 06:18:18 -0000
> @@ -43,7 +43,7 @@ Boston, MA 02111-1307, USA.  */
>  #define NO_DOT_IN_LABEL
>  
>  /* enable #pragma pack(n) */
> -#define HANDLE_SYSV_PRAGMA
> +#define HANDLE_SYSV_PRAGMA 1
>  
>  /* No underscore is prepended to any C symbol name.  */
>  #undef USER_LABEL_PREFIX
> Index: config/mips/vxworks.h
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/config/mips/vxworks.h,v
> retrieving revision 1.9
> diff -u -p -r1.9 vxworks.h
> --- config/mips/vxworks.h	22 Apr 2004 07:03:00 -0000	1.9
> +++ config/mips/vxworks.h	5 Jun 2004 06:18:18 -0000
> @@ -17,14 +17,6 @@ along with GCC; see the file COPYING.  I
>  the Free Software Foundation, 59 Temple Place - Suite 330,
>  Boston, MA 02111-1307, USA.  */
>  
> -#undef  EXTRA_SECTIONS
> -#define EXTRA_SECTIONS in_sdata, in_sbss
> -
> -#undef  EXTRA_SECTION_FUNCTIONS
> -#define EXTRA_SECTION_FUNCTIONS                                            \
> -  SECTION_FUNCTION_TEMPLATE(sdata_section, in_sdata, SDATA_SECTION_ASM_OP) \
> -  SECTION_FUNCTION_TEMPLATE(sbss_section, in_sbss, SBSS_SECTION_ASM_OP)
> -
>  
>  #undef  TARGET_VERSION
>  #define TARGET_VERSION fprintf (stderr, " (MIPS, VxWorks syntax)");
> @@ -73,6 +65,17 @@ Boston, MA 02111-1307, USA.  */
>    %{msoft-float:-DSOFT_FLOAT} \
>    %{mips1:-D_WRS_R3K_EXC_SUPPORT}"
>  
> +/* Fake out the ABI.  64-bit ISAs use O64, but it seems to be marked
> +   like o32 in the vxworks kernel libraries..  Therefore, when building
> +   64-bit code we tell cc1 to use the o64 ABI but don't tell
> +   the assembler anything.  */
> +#undef SUBTARGET_CC1_SPEC
> +#define SUBTARGET_CC1_SPEC "%{!mabi=*:%{mips3|mips4|mips64:-mabi=o64}}"
> +
>  /* No sdata.  */
>  #undef MIPS_DEFAULT_GVALUE
>  #define MIPS_DEFAULT_GVALUE 0
> +
> +/* -mdebug / -no-mdebug aren't supported by the assembler.  */
> +#undef MDEBUG_ASM_SPEC
> +#define MDEBUG_ASM_SPEC ""
-- 
Eric Christopher <echristo@redhat.com>