Fix i686-linux --enable-targets=all

Fix i686-linux --enable-targets=all

[Daniel Jacobowitz]

Right now all tests fail if you build an i686-linux compiler with
--enable-targets=all and test the -m64 multilib, while using bleeding
edge binutils.  This was already fixed for x86_64, but different
arrangement of t-* fragments means that the fix didn't cover
i686-linux.

Is this OK?  I have not been able to get my build server to bootstrap
an --enable-targets=all compiler due to the different way Debian /
Ubuntu handle multiple architecture support, but I did build 64-bit
libgcc and verify that crtend.o no longer has a zero terminator
followed by some unwind information.

-- 
Daniel Jacobowitz
CodeSourcery

2007-08-06  Daniel Jacobowitz  <dan@codesourcery.com>

	* config/i386/t-crtstuff (CRTSTUFF_T_CFLAGS): Add
	-fno-asynchronous-unwind-tables.
	* config/i386/t-linux64 (CRTSTUFF_T_CFLAGS): Remove.
	* gcc/config.gcc (x86_64-*-linux*): Use i386/t-crtstuff.

Index: gcc/config/i386/t-crtstuff
===================================================================
--- gcc/config/i386/t-crtstuff	(revision 127159)
+++ gcc/config/i386/t-crtstuff	(working copy)
@@ -1,2 +1,5 @@
 # The pushl in CTOR initialization interferes with frame pointer elimination.
-CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer
+# crtend*.o cannot be compiled without -fno-asynchronous-unwind-tables,
+# because then __FRAME_END__ might not be the last thing in .eh_frame
+# section.
+CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer -fno-asynchronous-unwind-tables
Index: gcc/config/i386/t-linux64
===================================================================
--- gcc/config/i386/t-linux64	(revision 127159)
+++ gcc/config/i386/t-linux64	(working copy)
@@ -17,9 +17,3 @@ EXTRA_MULTILIB_PARTS=crtbegin.o crtend.o
 
 softfp_wrap_start := '\#ifdef __x86_64__'
 softfp_wrap_end := '\#endif'
-
-# The pushl in CTOR initialization interferes with frame pointer elimination.
-# crtend*.o cannot be compiled without -fno-asynchronous-unwind-tables,
-# because then __FRAME_END__ might not be the last thing in .eh_frame
-# section.
-CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer -fno-asynchronous-unwind-tables
Index: gcc/config.gcc
===================================================================
--- gcc/config.gcc	(revision 127159)
+++ gcc/config.gcc	(working copy)
@@ -1166,7 +1166,7 @@ x86_64-*-linux* | x86_64-*-kfreebsd*-gnu
 	x86_64-*-kfreebsd*-gnu) tm_file="${tm_file} kfreebsd-gnu.h" ;;
 	x86_64-*-knetbsd*-gnu) tm_file="${tm_file} knetbsd-gnu.h" ;;
 	esac
-	tmake_file="${tmake_file} i386/t-linux64 i386/t-crtpc i386/t-crtfm i386/t-fprules-softfp64 soft-fp/t-softfp t-dfprules"
+	tmake_file="${tmake_file} i386/t-linux64 i386/t-crtstuff i386/t-crtpc i386/t-crtfm i386/t-fprules-softfp64 soft-fp/t-softfp t-dfprules"
 	;;
 i[34567]86-*-gnu*)
 	;;

Re: Fix i686-linux --enable-targets=all

[Andrew Pinski]

On 8/6/07, Daniel Jacobowitz <drow@false.org> wrote:
> Right now all tests fail if you build an i686-linux compiler with
> --enable-targets=all and test the -m64 multilib, while using bleeding
> edge binutils.  This was already fixed for x86_64, but different
> arrangement of t-* fragments means that the fix didn't cover
> i686-linux.
>
> Is this OK?  I have not been able to get my build server to bootstrap
> an --enable-targets=all compiler due to the different way Debian /
> Ubuntu handle multiple architecture support, but I did build 64-bit
> libgcc and verify that crtend.o no longer has a zero terminator
> followed by some unwind information.

I think this is PR 32840.

Thanks,
Andrew Pinski

Re: Fix i686-linux --enable-targets=all

[H.J. Lu]

On Mon, Aug 06, 2007 at 09:36:55AM -0400, Daniel Jacobowitz wrote:
> Right now all tests fail if you build an i686-linux compiler with
> --enable-targets=all and test the -m64 multilib, while using bleeding
> edge binutils.  This was already fixed for x86_64, but different
> arrangement of t-* fragments means that the fix didn't cover
> i686-linux.
> 
> Is this OK?  I have not been able to get my build server to bootstrap
> an --enable-targets=all compiler due to the different way Debian /
> Ubuntu handle multiple architecture support, but I did build 64-bit
> libgcc and verify that crtend.o no longer has a zero terminator
> followed by some unwind information.
> 

It is the part of

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31868

and a complete patch is posted at

http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00563.html

Could someone please review it?


H.J.

Re: Fix i686-linux --enable-targets=all

[Daniel Jacobowitz]

On Mon, Aug 06, 2007 at 06:41:28AM -0700, H.J. Lu wrote:
> It is the part of
> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31868
> 
> and a complete patch is posted at
> 
> http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00563.html
> 
> Could someone please review it?

I see.  I don't think your patch actually handles this case; you set
CRTSTUFF_T_FLAGS in t-crtstuff64, but i386/t-crtstuff will be included
afterwards, and that will set it again.  That's why I did it more
centrally.  The problem is the same for any target that forces
creation of .eh_frame; my patch won't affect i386-linux crtend.o even
though it changes the compile flags there.

Also, I think += is preferred.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Fix i686-linux --enable-targets=all

[H.J. Lu]

On Mon, Aug 06, 2007 at 09:51:41AM -0400, Daniel Jacobowitz wrote:
> On Mon, Aug 06, 2007 at 06:41:28AM -0700, H.J. Lu wrote:
> > It is the part of
> > 
> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31868
> > 
> > and a complete patch is posted at
> > 
> > http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00563.html
> > 
> > Could someone please review it?
> 
> I see.  I don't think your patch actually handles this case; you set
> CRTSTUFF_T_FLAGS in t-crtstuff64, but i386/t-crtstuff will be included
> afterwards, and that will set it again.  That's why I did it more

Could you tell me where i386/t-crtstuff is included after
i386/t-crtstuff64? I must have missed it.

> centrally.  The problem is the same for any target that forces
> creation of .eh_frame; my patch won't affect i386-linux crtend.o even
> though it changes the compile flags there.

-fno-asynchronous-unwind-tables is turned off for i386 by default.
That is why i386's crtend.o is OK. Turning it off explicitly should
be OK. I have been using it for 3 months.

> 
> Also, I think += is preferred.

My patch is supposed to set CRTSTUFF_T_CFLAGS only once. += isn't
very useful.


H.J.

Re: Fix i686-linux --enable-targets=all

[Daniel Jacobowitz]

On Mon, Aug 06, 2007 at 07:12:55AM -0700, H.J. Lu wrote:
> Could you tell me where i386/t-crtstuff is included after
> i386/t-crtstuff64? I must have missed it.

i[34567]86-*-linux* | i[34567]86-*-kfreebsd*-gnu | i[34567]86-*-knetbsd*-gnu)
...
                if test x$enable_targets = xall; then
...
                        tmake_file="${tmake_file} i386/t-linux64 i386/t-fprules-softfp64 soft-fp/t-softfp"
...
        tmake_file="${tmake_file} i386/t-crtstuff i386/t-crtpc i386/t-crtfm t-dfprules"
	;;

I think you should be able to add it to the existing t-crtstuff, as in
my patch.  I just didn't fix the other non-Linux targets.

> > Also, I think += is preferred.
> 
> My patch is supposed to set CRTSTUFF_T_CFLAGS only once. += isn't
> very useful.

Oh, I misread this, sorry:

-CRTSTUFF_T_CFLAGS_S = -fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC

-- 
Daniel Jacobowitz
CodeSourcery

Re: Fix i686-linux --enable-targets=all

[H.J. Lu]

On Mon, Aug 06, 2007 at 10:23:59AM -0400, Daniel Jacobowitz wrote:
> On Mon, Aug 06, 2007 at 07:12:55AM -0700, H.J. Lu wrote:
> > Could you tell me where i386/t-crtstuff is included after
> > i386/t-crtstuff64? I must have missed it.
> 
> i[34567]86-*-linux* | i[34567]86-*-kfreebsd*-gnu | i[34567]86-*-knetbsd*-gnu)
> ...
>                 if test x$enable_targets = xall; then
> ...
>                         tmake_file="${tmake_file} i386/t-linux64 i386/t-fprules-softfp64 soft-fp/t-softfp"
> ...
>         tmake_file="${tmake_file} i386/t-crtstuff i386/t-crtpc i386/t-crtfm t-dfprules"
> 	;;
> 
> I think you should be able to add it to the existing t-crtstuff, as in
> my patch.  I just didn't fix the other non-Linux targets.
> 

I am testing this patch now.


H.J.
-----
2007-08-06  H.J. Lu  <hongjiu.lu@intel.com>
	    Daniel Jacobowitz  <dan@codesourcery.com>

	PR target/31868
	* config.gcc (x86_64-*-freebsd*): Add i386/t-crtstuff to
	tmake_file.
	(x86_64-*-netbsd*): Likewise.
	(x86_64-*-linux*): Likewise.
	(x86_64-*-kfreebsd*-gnu): Likewise.
	(x86_64-*-knetbsd*-gnu): Likewise.
	(i[34567]86-*-solaris2.1[0-9]*): Likewise.

	* config/i386/t-linux64 (CRTSTUFF_T_CFLAGS): Removed.

	* config/i386/t-crtstuff (CRTSTUFF_T_CFLAGS): Update comments.
	Add -fno-asynchronous-unwind-tables.

	* config/t-freebsd (CRTSTUFF_T_CFLAGS_S): Add $(CRTSTUFF_T_CFLAGS).
	* config/t-libc-ok (CRTSTUFF_T_CFLAGS_S): Likewise.
	* config/t-lynx (CRTSTUFF_T_CFLAGS_S): Likewise.
	* config/t-netbsd (CRTSTUFF_T_CFLAGS_S): Likewise.
	* config/t-svr4 (CRTSTUFF_T_CFLAGS_S): Likewise.

--- gcc/config.gcc.crtend	2007-08-04 07:57:43.000000000 -0700
+++ gcc/config.gcc	2007-08-06 07:49:51.000000000 -0700
@@ -1099,6 +1099,7 @@ i[34567]86-*-freebsd*)
 	;;
 x86_64-*-freebsd*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h ${fbsd_tm_file} i386/x86-64.h i386/freebsd.h i386/freebsd64.h"
+	tmake_file="${tmake_file} i386/t-crtstuff"
 	;;
 i[34567]86-*-netbsdelf*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h netbsd.h netbsd-elf.h i386/netbsd-elf.h"
@@ -1111,6 +1112,7 @@ i[34567]86-*-netbsd*)
 	;;
 x86_64-*-netbsd*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h netbsd.h netbsd-elf.h i386/x86-64.h i386/netbsd64.h"
+	tmake_file="${tmake_file} i386/t-crtstuff"
 	;;
 i[34567]86-*-openbsd2.*|i[34567]86-*openbsd3.[0123])
 	tm_file="i386/i386.h i386/unix.h i386/bsd.h i386/gas.h i386/gstabs.h openbsd-oldgas.h openbsd.h i386/openbsd.h"
@@ -1170,7 +1172,7 @@ x86_64-*-linux* | x86_64-*-kfreebsd*-gnu
 	x86_64-*-kfreebsd*-gnu) tm_file="${tm_file} kfreebsd-gnu.h" ;;
 	x86_64-*-knetbsd*-gnu) tm_file="${tm_file} knetbsd-gnu.h" ;;
 	esac
-	tmake_file="${tmake_file} i386/t-linux64 i386/t-crtpc i386/t-crtfm i386/t-fprules-softfp64 soft-fp/t-softfp t-dfprules"
+	tmake_file="${tmake_file} i386/t-linux64 i386/t-crtstuff i386/t-crtpc i386/t-crtfm i386/t-fprules-softfp64 soft-fp/t-softfp t-dfprules"
 	;;
 i[34567]86-*-gnu*)
 	;;
@@ -1267,7 +1269,7 @@ i[34567]86-*-solaris2*)
 	*-*-solaris2.1[0-9]*)
 		tm_file="${tm_file} i386/x86-64.h i386/sol2-10.h"
 		tm_defines="${tm_defines} TARGET_BI_ARCH=1"
-		tmake_file="$tmake_file i386/t-sol2-10"
+		tmake_file="$tmake_file i386/t-crtstuff i386/t-sol2-10"
 		need_64bit_hwint=yes
 		# FIXME: -m64 for i[34567]86-*-* should be allowed just
 		# like -m32 for x86_64-*-*.
--- gcc/config/i386/t-crtstuff.crtend	2006-09-27 22:37:59.000000000 -0700
+++ gcc/config/i386/t-crtstuff	2007-08-06 07:47:49.000000000 -0700
@@ -1,2 +1,7 @@
 # The pushl in CTOR initialization interferes with frame pointer elimination.
-CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer
+# crtend*.o cannot be compiled without -fno-asynchronous-unwind-tables,
+# because then __FRAME_END__ might not be the last thing in .eh_frame
+# section.  -fno-asynchronous-unwind-tables is off by default for i386
+# and is on by default for x86-64.  We turn it off for both i386 and
+# x86-64.
+CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer -fno-asynchronous-unwind-tables
--- gcc/config/i386/t-linux64.crtend	2007-05-18 07:10:43.000000000 -0700
+++ gcc/config/i386/t-linux64	2007-08-04 07:57:44.000000000 -0700
@@ -17,9 +17,3 @@ EXTRA_MULTILIB_PARTS=crtbegin.o crtend.o
 
 softfp_wrap_start := '\#ifdef __x86_64__'
 softfp_wrap_end := '\#endif'
-
-# The pushl in CTOR initialization interferes with frame pointer elimination.
-# crtend*.o cannot be compiled without -fno-asynchronous-unwind-tables,
-# because then __FRAME_END__ might not be the last thing in .eh_frame
-# section.
-CRTSTUFF_T_CFLAGS = -fno-omit-frame-pointer -fno-asynchronous-unwind-tables
--- gcc/config/t-freebsd.crtend	2006-09-27 22:38:23.000000000 -0700
+++ gcc/config/t-freebsd	2007-08-04 07:57:44.000000000 -0700
@@ -1,5 +1,5 @@
 # Compile crtbeginS.o and crtendS.o with pic.
-CRTSTUFF_T_CFLAGS_S = -fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
 
 # Compile libgcc.a with pic.
 TARGET_LIBGCC2_CFLAGS += -fPIC
--- gcc/config/t-libc-ok.crtend	2006-09-27 22:38:23.000000000 -0700
+++ gcc/config/t-libc-ok	2007-08-04 07:57:44.000000000 -0700
@@ -1 +1 @@
-CRTSTUFF_T_CFLAGS_S=-fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
--- gcc/config/t-lynx.crtend	2006-09-27 22:38:23.000000000 -0700
+++ gcc/config/t-lynx	2007-08-04 07:57:44.000000000 -0700
@@ -1,5 +1,5 @@
 # Compile crtbeginS.o and crtendS.o with pic.
-CRTSTUFF_T_CFLAGS_S = -fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
 
 # Compile libgcc2.a with pic.
 TARGET_LIBGCC2_CFLAGS = -fPIC
--- gcc/config/t-netbsd.crtend	2006-09-27 22:38:23.000000000 -0700
+++ gcc/config/t-netbsd	2007-08-04 07:57:44.000000000 -0700
@@ -1,2 +1,2 @@
 # Always build crtstuff with PIC.
-CRTSTUFF_T_CFLAGS = -fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
--- gcc/config/t-svr4.crtend	2006-09-27 22:38:23.000000000 -0700
+++ gcc/config/t-svr4	2007-08-04 07:57:44.000000000 -0700
@@ -4,7 +4,7 @@
 # we will be doing that, we just always use -fPIC when compiling the
 # routines in crtstuff.c.  Likewise for libgcc2.c.
 
-CRTSTUFF_T_CFLAGS = -fPIC
+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
 TARGET_LIBGCC2_CFLAGS = -fPIC
 
 # See all the declarations.

Re: Fix i686-linux --enable-targets=all

[Daniel Jacobowitz]

On Mon, Aug 06, 2007 at 07:56:47AM -0700, H.J. Lu wrote:
> 2007-08-06  H.J. Lu  <hongjiu.lu@intel.com>
> 	    Daniel Jacobowitz  <dan@codesourcery.com>
> 
> 	PR target/31868
> 	* config.gcc (x86_64-*-freebsd*): Add i386/t-crtstuff to
> 	tmake_file.
> 	(x86_64-*-netbsd*): Likewise.
> 	(x86_64-*-linux*): Likewise.
> 	(x86_64-*-kfreebsd*-gnu): Likewise.
> 	(x86_64-*-knetbsd*-gnu): Likewise.
> 	(i[34567]86-*-solaris2.1[0-9]*): Likewise.
> 
> 	* config/i386/t-linux64 (CRTSTUFF_T_CFLAGS): Removed.
> 
> 	* config/i386/t-crtstuff (CRTSTUFF_T_CFLAGS): Update comments.
> 	Add -fno-asynchronous-unwind-tables.
> 
> 	* config/t-freebsd (CRTSTUFF_T_CFLAGS_S): Add $(CRTSTUFF_T_CFLAGS).
> 	* config/t-libc-ok (CRTSTUFF_T_CFLAGS_S): Likewise.
> 	* config/t-lynx (CRTSTUFF_T_CFLAGS_S): Likewise.
> 	* config/t-netbsd (CRTSTUFF_T_CFLAGS_S): Likewise.
> 	* config/t-svr4 (CRTSTUFF_T_CFLAGS_S): Likewise.

Thank you!  This looks good to me.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Fix i686-linux --enable-targets=all

[Paolo Bonzini]

> -CRTSTUFF_T_CFLAGS_S = -fPIC
> +CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC

This

CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS_S) -fPIC

wouldn't actually work at all (infinite recursion). :-)

Paolo

Re: Fix i686-linux --enable-targets=all

[Daniel Jacobowitz]

On Mon, Aug 06, 2007 at 06:05:17PM +0200, Paolo Bonzini wrote:
> 
> > -CRTSTUFF_T_CFLAGS_S = -fPIC
> > +CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
> 
> This
> 
> CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS_S) -fPIC
> 
> wouldn't actually work at all (infinite recursion). :-)

I made the same mistake reading it.  CRTSTUFF_T_CFLAGS on the RHS
should be defined independently of CRTSTUFF_T_CFLAGS_S on the right.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Fix i686-linux --enable-targets=all

[H.J. Lu]

On Mon, Aug 06, 2007 at 06:05:17PM +0200, Paolo Bonzini wrote:
> 
> >-CRTSTUFF_T_CFLAGS_S = -fPIC
> >+CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS) -fPIC
> 
> This
> 
> CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS_S) -fPIC
> 
> wouldn't actually work at all (infinite recursion). :-)

Why do we want to use

CRTSTUFF_T_CFLAGS_S = $(CRTSTUFF_T_CFLAGS_S) -fPIC

CRTSTUFF_T_CFLAGS is for static crt*.o and CRTSTUFF_T_CFLAGS_S is
for shared crt*.o. The difference is CRTSTUFF_T_CFLAGS_S has
-fPIC and CRTSTUFF_T_CFLAGS doesn't. Otherwise, they should be
the same.


H.J.

Re: Fix i686-linux --enable-targets=all

[Paolo Bonzini]

H.J. Lu wrote:

> I am testing this patch now.

Ok if it passes testing.  Best would be building both i686-pc-linux-gnu 
(with and without --enable-targets=all) and x86_64-pc-linux-gnu, but I 
guess testing only i686-pc-linux-gnu --enable-targets=all is also fine.

Paolo

Re: Fix i686-linux --enable-targets=all

[H.J. Lu]

On Mon, Aug 06, 2007 at 06:28:35PM +0200, Paolo Bonzini wrote:
> H.J. Lu wrote:
> 
> >I am testing this patch now.
> 
> Ok if it passes testing.  Best would be building both i686-pc-linux-gnu 
> (with and without --enable-targets=all) and x86_64-pc-linux-gnu, but I 
> guess testing only i686-pc-linux-gnu --enable-targets=all is also fine.
> 

I don't have setup to test i686-pc-linux-gnu --enable-targets=all.
Since Dan has no problem with i686-pc-linux-gnu --enable-targets=all,
I am checking it in.


H.J.