]> git.wh0rd.org Git - elf2flt.git/log
elf2flt.git
16 years agoThis patch allows elf2flt/flthdr's compression options to work in a wider
David McCullough [Wed, 27 Feb 2008 11:41:32 +0000 (11:41 +0000)]
This patch allows elf2flt/flthdr's compression options to work in a wider
variety of environments (e.g. under MinGW/Win32), by linking with zlib
rather than by using external gzip/gunzip executables. The cp binary
isn't used any more either, and libiberty's make_temp_file() is used
instead of mkstemp() as a more portable way of creating a temporary file.

Also the compression logic is made somewhat clearer, IMO.

Signed-off-by: Julian Brown <julian@codesourcery.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
16 years agoWhen moving the xmalloc changes from the Blackfin elf2flt to the upstream
David McCullough [Sun, 24 Feb 2008 23:17:51 +0000 (23:17 +0000)]
When moving the xmalloc changes from the Blackfin elf2flt to the upstream
elf2flt repo, I accidentally dropped the libiberty.h include.  Not a fatal
error, but having proper prototypes is always a good thing.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
16 years agoI hit a random failure in elf2flt (not elf2flt's fault, I was screwing with
David McCullough [Thu, 21 Feb 2008 01:09:05 +0000 (01:09 +0000)]
I hit a random failure in elf2flt (not elf2flt's fault, I was screwing with
ELFs and LMAs/VMAs), but the resulting error message was useless:
malloc: Cannot allocate memory

Since I was executing `...-gcc` at the time, where exactly this error message
was coming from was hard to say.  So instead of using malloc() and then doing
a simple 'perror("malloc"); exit(1);' in the elf2flt.c code, it'd be better
for everyone if we used the xmalloc() funcs from libiberty.  We're already
linking elf2flt against libiberty, so there's no extra headers/libs to link
against.

Now the crash looks like:
bfin-uclinux-elf2flt: out of memory allocating 4221960244 bytes after a total of
 135168 bytes
So much nicer! :)

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
16 years agoThis patch fixes two issues which are triggered when compiling for nios2:
David McCullough [Mon, 3 Dec 2007 23:21:01 +0000 (23:21 +0000)]
This patch fixes two issues which are triggered when compiling for nios2:
* Typo in a goto label
* Misplaced switch statement which probably is left from an older revision.

 - Atle

Signed-off-by: Atle Nissestad <atle@nissestad.no>
16 years agoThe patch below adds two additional DWARF2 debug sections from the standard
David McCullough [Tue, 27 Nov 2007 22:53:00 +0000 (22:53 +0000)]
The patch below adds two additional DWARF2 debug sections from the standard
binutils linker-script.
This is required to fix gdb DWARF errors on nios2.

Signed-off-by: Atle Nissestad <atle@nissestad.no>
16 years agoThis patch is required to successfully link against libstdc++ on the Nios2
David McCullough [Tue, 27 Nov 2007 22:51:30 +0000 (22:51 +0000)]
This patch is required to successfully link against libstdc++ on the Nios2
platform.

Signed-off-by: Atle Nissestad <atle@nissestad.no>
Also remove the duplicated data1 section and add the eh_frame_hdr section.
Source of these changes lost (sorry :-)

17 years agothe intl subdir in binutils may redirect some funcs with libintl_* prefixes
David McCullough [Thu, 11 Oct 2007 03:50:51 +0000 (03:50 +0000)]
the intl subdir in binutils may redirect some funcs with libintl_* prefixes
(like building for win32), so the attached patch adds a stub like the
existing one for dcgettect() ... this fixes building of elf2flt for me for
mingw targets

Mike Frysinger <vapier@gentoo.org>

17 years agoAllow for hex/octal/decimal stack specifications using the usual 0xNN,0NN,NN
David McCullough [Mon, 24 Sep 2007 05:11:48 +0000 (05:11 +0000)]
Allow for hex/octal/decimal stack specifications using the usual 0xNN,0NN,NN
notations.

17 years agoThe short_format var isnt used in flthdr.c, so delete it
David McCullough [Mon, 24 Sep 2007 00:36:17 +0000 (00:36 +0000)]
The short_format var isnt used in flthdr.c, so delete it
Mike Frysinger <vapier@gentoo.org>

17 years agolf2flt crashes on Linux/amd64:
Greg Ungerer [Wed, 1 Aug 2007 03:40:12 +0000 (03:40 +0000)]
lf2flt crashes on Linux/amd64:

(gdb)  run  -a -o links -p links.gdb links.gdb
Starting program: /home/stsp/dslinux/toolchain/prefix/bin/arm-linux-elf-elf2flt -a -o links -p links.gdb links.gdb

Program received signal SIGSEGV, Segmentation fault.
_bfd_elf_canonicalize_reloc (abfd=<value optimized out>, section=0x5f6900,
    relptr=0xffffffffa6360010, symbols=<value optimized out>)
    at /home/stsp/dslinux/toolchain/src/binutils-2.17/bfd/elf.c:6367
6367        *relptr++ = tblptr++;
(gdb) bt
#0  _bfd_elf_canonicalize_reloc (abfd=<value optimized out>, section=0x5f6900,
    relptr=0xffffffffa6360010, symbols=<value optimized out>)
    at /home/stsp/dslinux/toolchain/src/binutils-2.17/bfd/elf.c:6367
#1  0x00000000004006dd in output_relocs (abs_bfd=0x5f5570,
    symbols=0x2b30a5e99010, number_of_symbols=16585, n_relocs=0x7fff04c0fe58,
    text=0x2b30a6102010 "", text_len=<value optimized out>, text_vma=0,
    data=0x2b30a627b010 "", data_len=934480, data_vma=1541824,
    rel_bfd=0x5f4400)
    at /home/stsp/dslinux/toolchain/src/elf2flt-20051225/elf2flt.c:587
#2  0x0000000000401180 in main (argc=<value optimized out>,
    argv=<value optimized out>)
    at /home/stsp/dslinux/toolchain/src/elf2flt-20051225/elf2flt.c:1949

The problem seems to be that the one and only call to xmalloc()
in elf2flt.c does not return a valid pointer for some reason.

I'm wondering why xmalloc() is used exactly once in elf2flt.c.
All other heap allocations in elf2flt are done with plain malloc().

The attached patch fixes the segfault by replacing the call to xmalloc()
with a call to malloc(). It also makes elf2flt check for return
values of malloc() calls, providing the equivalent behaviour of
using xmalloc().

Patch submitted by Stefan Sperling <stsp@stsp.name>

17 years agoRe-autconf configure script after Mike Frysinger's recent AC_ERROR
Greg Ungerer [Wed, 3 Jan 2007 03:53:51 +0000 (03:53 +0000)]
Re-autconf configure script after Mike Frysinger's recent AC_ERROR
fix to configure.in.

17 years agoThe proper way to do error message output is with AC_MSG_ERROR() ... patch
Greg Ungerer [Wed, 3 Jan 2007 03:49:47 +0000 (03:49 +0000)]
The proper way to do error message output is with AC_MSG_ERROR() ... patch
attached.

Patch submitted by Mike Frysinger <vapier@gentoo.org>

18 years agothe install program is run with the '-s' option which makes `install` run the
David McCullough [Thu, 23 Nov 2006 22:42:07 +0000 (22:42 +0000)]
the install program is run with the '-s' option which makes `install` run the
host `strip` on the installed binary ... this certainly wont work when
cross-compiling and really, the standard is to leave stripping up to the
people doing the actual install

so attached patch simply drops the -s argument to install

Mike Frysinger <vapier@gentoo.org>

18 years agothe bzero() function has been marked as depreciated for quite sometime in
David McCullough [Tue, 14 Nov 2006 22:20:08 +0000 (22:20 +0000)]
the bzero() function has been marked as depreciated for quite sometime in
favor of using memset() with a value of 0 ... patch attached

Mike Frysinger <vapier@gentoo.org>

18 years agoBefore I delve in the long (sorry) RFC below, I've been taking:
David McCullough [Wed, 13 Sep 2006 00:10:30 +0000 (00:10 +0000)]
Before I delve in the long (sorry) RFC below, I've been taking:

    http://www.ucdot.org/article.pl?sid=03/11/25/1126257&mode=thread

and:

    http://docs.blackfin.uclinux.org/doku.php?id=adding_libraries

as more-or-less definitive descriptions of how shared flat libraries
should be built and used.  Please let me know if they aren't, and if
there's another recipe somewhere.

Anyway, the RFC is about two issues:

  (1) Shared libraries usually have some symbols that are internally
      STB_GLOBAL (because they are shared between more than one input
      object) but which should nevertheless be treated as local to
      the library as a whole; they shouldn't be exported to users
      of the library.  According to the article above, the usual
      way to arrange this is to run:

        objcopy -L <sym1> -L <sym2> ... foo.gdb

      on the linked foo.gdb library.  That certainly works, but
      maintaining the list of symbols can be a bit cumbersome.

      I think it would help a lot if ld-elf2flt automatically localised
      symbols with hidden and internal visibility.  This would allow
      users to localise symbols in the library source code -- via gcc
      attributes or assembly directives -- and would eliminate one
      difference between shared flat libraries and shared ELF libraries.
      If the library's source code hides all the appropriate symbols --
      which the source code for many ELF-leaning libraries do, unless their
      build system relies on something like "local:" directives in version
      scripts -- then no separate -L list is needed.

  (2) If a shared library and a user of a shared library both define
      some symbol foo, you don't get a redefinition warning or error.
      However, if the symbols are a different size, you _do_ get a warning
      about that.

      To me, this seems like the worst of both worlds.  I can see cases
      where the user would want to be told about any redefinition, and
      would want a command-line option to enable that behaviour.  That's
      a separate issue that I'm not tackling here.

      But when redefinitions are silently allowed, warning about changes
      in size seems a little pointless.  Shared flat libraries can't
      really refer back to symbols defined by the user of the library
      (such as the executable); all symbol references are resolved
      when the library is statically linked.  Shared flat libraries
      act a lot like a restricted form of ELF -Bsymbolic libraries.

      Thus having a different size of symbol S in shared library X and
      user Y is very unlikely to be an issue unless the redefinition of
      S is itself an issue.  X is not going to refer directly to Y's
      definition of S.  And if you create, say, a shared C library that
      includes all the C library functions needed by at least one
      program on the system, it's likely that some programs will have
      harmless symbol name clashes.  For example, some programs might be
      written with just standard C in mind, and use symbols that clash
      with a POSIX, GNU or uClibc extension.

      I think it would help if we weakened the exported symbols.  This
      would tell the linker that the symbols are only there if needed,
      and that the user of the library is free to use the symbol for
      something else.  This wouldn't affect the redefinition diagnostic;
      if that diagnostic is ever implemented, it should apply to both
      global and weak symbols, because in both cases we'll have two
      definitions of the same symbol.

Comments?  Does this sound reasonable sane?

In case it does, the patch below implements both (1) and (2).  (1) is
simple with CVS objcopy, which now has a --localize-hidden option:

    http://sources.redhat.com/ml/binutils/2006-06/msg00204.html

Sadly, that option won't appear in an FSF release before 2.18, so I've
provided a fall-back that uses objdump and sed instead.  (2) is easy
with the long-standing --weaken option.

FWIW, the patch has been tested fairly extensively on Coldfire with
shared C and C++ libraries, as well as with ad-hoc libraries.  No
post-processing of these libraries was needed: you could just link with
ld-elf2flt and use the output directly.

I've tested both the --localize-hidden and fallback versions.

Richard

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years agoThis patch adds elf2flt support for prioritised constructors and destructors,
David McCullough [Fri, 21 Jul 2006 13:11:04 +0000 (13:11 +0000)]
This patch adds elf2flt support for prioritised constructors and destructors,
but only when --enable-emit-relocs is in effect.

GCC puts prioritised constructors and destructors in special .ctor.N
and .dtor.N sections (where N encodes the priority).  The constructors
and destructors in these sections have a lower priority than those
in the default .ctor and .dtor sections.  However, we need to treat
the crtbegin.o's and crtend.o's sections specially; crtbegin.o's .[cd]tor
sections must come before all the other .[cd]tor* contents and crtend.o's
.[cd]tor sections must come after all other contents.

In --disable-emit-relocs links, it wasn't possible to treat crtbegin.o
and crtend.o specially.  ld-elf2flt would do:

                ld -r -o foo.elf2flt ...input objects...
                    |                               |
            ld -r -o foo.elf foo.elf2flt -T script  |
                                                    |
                                   ld -o foo.gdb foo.elf2flt -T script

so all the input .[cd]tor sections would already be lumped together by the
time the script was used.  This isn't a problem for --enable-emit-relocs as
it only links once.

The patch copies the [cd]tor support from the standard binutils linker
script.  The question is: should the new lines be used unconditionally,
or only when --enable-emit-relocs is in effect?

We could use the new lines for --disable-emit-relocs without breaking
programs that don't use prioritised [cd]tors.  However, it would be a user
interface regression for programs that _do_ use them.  At the moment,
if you try to use prioritised [cd]tors with --disable-emit-relocs,
you get a linker error like this:

   error: no memory region specified for loadable section `.ctors.60535'

With the new lines, the link would silently succeed, and you'd get a
runtime failure instead.  I've therefore guarded the lines with
"SINGLE_LINK:", a bit like the "W_RODAT" stuff that's already there.

Tested on Coldfire, with both --enable-emit-relocs and
--disable-emit-relocs.  It fixes the GCC ecos.exp tests.
Please install if OK.

Richard

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years agoUse the emit-relocs options by default unless the user has configured
David McCullough [Mon, 10 Jul 2006 22:23:46 +0000 (22:23 +0000)]
Use the emit-relocs options by default unless the user has configured
it to do otherwise.

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years agoelf2flt.c (output_relocs) <use_resolved> [TARGET_arm]:
David McCullough [Mon, 10 Jul 2006 22:19:10 +0000 (22:19 +0000)]
elf2flt.c (output_relocs) <use_resolved> [TARGET_arm]:
Add R_ARM_THM_PC11, which does not need a relocation.

Shaun Jackman  <sjackman@gmail.com>

18 years agoThis patch adds m68k support for elf2flt -a. I've tested it with
Greg Ungerer [Mon, 10 Jul 2006 06:00:41 +0000 (06:00 +0000)]
This patch adds m68k support for elf2flt -a.  I've tested it with
load-to-RAM, -msep-data PIC-with-GOT and -mshared-library PIC-with-GOT.

Like ARM, but unlike V850, the new code errors out on relocs it doesn't
understand, rather than assuming that they're 32-bit fields.

It seems likely that every port that supports -a will want to print the
"reloc type not supported in this context" message; we've got two copies
already, and m68k would add a third.  I've therefore moved the v850
"bad_v850_reloc_err:" code outside the #ifdef and renamed the label to
the more neutral "bad_resolved_reloc:".  I've also changed the ARM code
to use this now-common code instead of its own copy.

In a similar vein, I've changed the default 32-bit field handling
so that it is the target of a new label, "good_32bit_resolved_reloc:".
The generic and v850 code can then jump to this code in their default
cases, while m68k can jump to it for R_68K_32 relocs.

The patch was written before Shaun Jackman made -p and -a work together:

    http://mailman.uclinux.org/pipermail/uclinux-dev/2006-June/039029.html

I'd been using this hunk too, but as well as guarding:

    q->address += got_size;

with "!use_resolved", I'd guarded the calculation of got_size itself,
as the value isn't otherwise used.  I've kept this additional check
in the attached patch.  Please apply if OK.

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years ago * elf2flt.c (output_relocs) <use_resolved> [TARGET_arm]: Add
Greg Ungerer [Mon, 3 Jul 2006 05:50:10 +0000 (05:50 +0000)]
* elf2flt.c (output_relocs) <use_resolved> [TARGET_arm]: Add
    R_ARM_REL32, which does not need a relocation.

Patch submitted by Shaun Jackman <sjackman@gmail.com>

18 years agoSupport the -a (use_resolved) option for TARGET_arm.
David McCullough [Mon, 26 Jun 2006 22:18:15 +0000 (22:18 +0000)]
Support the -a (use_resolved) option for TARGET_arm.
Shaun Jackman <sjackman@gmail.com>

18 years agoWe've eliminated the use of the relocation stack from the Blackfin
David McCullough [Mon, 19 Jun 2006 22:00:41 +0000 (22:00 +0000)]
We've eliminated the use of the relocation stack from the Blackfin
toolchain.  That means that the corresponding code in elf2flt can go
away, which is done by the patch below.

Bernd Schmidt <bernds_cb1@t-online.de>

18 years agoNewly autoconf generated configure script. It was out-of-date relative
Greg Ungerer [Thu, 15 Jun 2006 01:48:46 +0000 (01:48 +0000)]
Newly autoconf generated configure script. It was out-of-date relative
to the current configure.in. It didn't properly process
@binutils_ldscript_dir@, and that meant it was not properly set
in the ld-elf2flt.in/elf2flt.ld processing.

18 years agoThe patch below teaches elf2flt to handle R_68K_PC16 relocs.
David McCullough [Tue, 23 May 2006 22:28:31 +0000 (22:28 +0000)]
The patch below teaches elf2flt to handle R_68K_PC16 relocs.
Our immediate need for this was to deal with relocs for uClibc's
libc/sysdeps/linux/m68k/clone.S, which has "bcc.w __syscall_error"
instructions.  It might be argued that clone.S should be using jbcc
instead, in case __syscall_error ends up too far away.  But even if
that's true, elf2flt should still support such branches in cases
whether the user _knows_ that the target is within range.
So I think elf2flt should be patched either way.

We've been using this patch for a while now without problems.
Please install if OK.

Richard

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years agoThe current elf2flt.ld forgets to allow to put `gnu.linkonce.r*' sections
David McCullough [Thu, 6 Apr 2006 00:37:45 +0000 (00:37 +0000)]
The current elf2flt.ld forgets to allow to put `gnu.linkonce.r*' sections
in the read-only data segment.

Philippe De Muyter <phdm@macqel.be>

18 years agoAdd the "b" back to files opens for WIN32 builds as I am fairly sure it was
David McCullough [Mon, 3 Apr 2006 02:58:44 +0000 (02:58 +0000)]
Add the "b" back to files opens for WIN32 builds as I am fairly sure it was
required.

Fix the remain rb and wb cases that were missed.

18 years agoChanged the mode argument for all the popen() calls to be just "w",
Greg Ungerer [Mon, 3 Apr 2006 00:24:56 +0000 (00:24 +0000)]
Changed the mode argument for all the popen() calls to be just "w",
instead of "wb". According to the Linux manual page posix implementations
must accept only "r" or "w", no other characters. And you will get a
return error with EINVAL with modern version og glibc if this is not
the case.

It may well be that this is fine for CYGWIN too, someone needs to
test and report.

Calls to fopen and the like are unchanged, and they certainly do
support the "b" mode specifier.

18 years agoWhee. What a most entertaining bug.
David McCullough [Mon, 3 Apr 2006 00:06:38 +0000 (00:06 +0000)]
Whee.  What a most entertaining bug.

For some (unfathomable) reason, the flat loader in the kernel likes to
use network (big endian) byte order, but only sometimes.  In particular,
its assumptions about relocation byte order depend on whether the
executable uses ID shared libraries or not.

It took me most of today to figure out why the kernel was trying to
relocate something at address 0x94160000, until I noticed that that's
the byteswapped version of 0x1694, and further noticed that it was
omitting a byteswap due to the flag Has-PIC-GOT in the flat header.

So, why is that flag set?  After all, opreport wasn't compiled with
-mid-shared-library or anything.

It turns out that elf2flt sets that flag if grepping the executable for
GLOBAL_OFFSET_TABLE succeeds.  That's a reasonable thing to do, except
when we link an executable against libbfd.a - which contains
GLOBAL_OFFSET_TABLE not as a symbol, but as a string, as it needs to
produce the symbol in linker output files.

Changing elf2flt to grep nm output, rather than the executable directly,
fixes the problem.  So, in the end it turns out this wasn't related to
the 20030928 failure at all...

Bernd Schmidt <bernds_cb1@t-online.de>

18 years agoThis patch adds support for .preinit_array, .init_array and .fini_array
David McCullough [Thu, 23 Mar 2006 23:06:49 +0000 (23:06 +0000)]
This patch adds support for .preinit_array, .init_array and .fini_array
to the elf2flt linker script.  This seems useful in itself, and is
actually required for recent versions of uclibc, where UCLIBC_CTOR_DTOR=y
depends on the associated array symbols being defined.

Tested on m68k-uclinux using the gcc and g++ DejaGNU testsuites.  I also
ran some tests by hand to make sure that callbacks in the array sections
were being called at the right time.  Please install if OK.

Richard

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
18 years agoHello,
David McCullough [Wed, 22 Feb 2006 01:29:27 +0000 (01:29 +0000)]
Hello,

the following trivial patch is required to build elf2flt
on Darwin.  Weirdly, static compilation is intentionally
unsupported by Apple for better compatibility:

http://developer.apple.com/qa/qa2001/qa1118.html

As elf2flt binaries are usually shipped along with other
dynamic binaries such as gcc and gdb, we don't loose
anything on Linux.

I'd commit it myself, but I can't access the CVS server
right now.

Bernardo Innocenti <bernie@develer.com>

18 years agoThis patch allows cross-compilation of elf2flt for MinGW hosts when a
David McCullough [Wed, 1 Feb 2006 22:34:46 +0000 (22:34 +0000)]
This patch allows cross-compilation of elf2flt for MinGW hosts when a
non-MinGW build system is in use. By using uname, there is currently an
assumption in 'Makefile.in' that (build system) == (host system), which
isn't necessarily the case in a Canadian cross environment. (At
CodeSourcery we use Linux for building our MinGW-hosted toolchains, some
of which will soon include elf2flt).

Julian Brown <julian@codesourcery.com>

18 years agoPatch to add configure option:
David McCullough [Tue, 17 Jan 2006 23:12:08 +0000 (23:12 +0000)]
Patch to add configure option:

--with-binutils-ldscript-dir

to allow the user to change the location of the elf2flt.ld script.

Mike Frysinger <vapier@gentoo.org>

18 years agoseems my previous attempt at updating the current elf2flt code with nios
Greg Ungerer [Fri, 13 Jan 2006 06:08:01 +0000 (06:08 +0000)]
seems my previous attempt at updating the current elf2flt code with nios
relocations was slightly wrong ... the code needs to be moved down a few
hundred lines  :)

Patch submitted by Mike Frysinger <vapier@gentoo.org>

18 years agonios/nios2 support by Wentao Xu <wentao@microtronix.com>
David McCullough [Wed, 11 Jan 2006 03:39:16 +0000 (03:39 +0000)]
nios/nios2 support by Wentao Xu <wentao@microtronix.com>
cleaned up for inclusion by Mike Frysinger <vapier@gentoo.org>

19 years agoHi,
David McCullough [Wed, 28 Sep 2005 05:26:02 +0000 (05:26 +0000)]
Hi,

I'm the maintainer of a windows hosted arm-elf gcc toolchain used mostly
for homebrew console development. One of my users recently asked if I
could have a look a getting elf2flt to build under mingw/msys to help
with his uclinux Nintendo DS port. I've attached a patch containing the
changes I made to the source in CVS, thought it might be useful.

2005-09-25  Dave Murphy  <davem@devkitpro.org>
   * Makefile.in: add ws2_32 library when building from msys shell.
   * cygwin-elf.h: On mingw.include stdint.h and add extra typedefs
   * elf2flt.c: Include cygwin-elf.h on mingw too.
   * flthdr.c: For mingw use winsock2.h for internet consts & structs ,
     use mktemp instead of mkstemp

Dave

19 years agoBlackfin support merged in by Bernd Schmidt <bernds_cb1@t-online.de>
David McCullough [Thu, 25 Aug 2005 02:19:42 +0000 (02:19 +0000)]
Blackfin support merged in by Bernd Schmidt <bernds_cb1@t-online.de>

19 years agoChange all file format related unsigned longs to uint32_t, so that 64 bits
David McCullough [Mon, 1 Aug 2005 00:02:33 +0000 (00:02 +0000)]
Change all file format related unsigned longs to uint32_t,  so that 64 bits
hosts can build this code correctly,  plus a little bit of Blackfin code.

Most of the changes provided Bernd Schmidt <bernds_cb1 at t-online.de>

19 years agoA new config.sub with blackfin support
David McCullough [Fri, 15 Jul 2005 10:33:32 +0000 (10:33 +0000)]
A new config.sub with blackfin support
Bernd Schmidt <bernds_cb1@t-online.de>

19 years agoSmall spelling fixup from Dylan Griffiths.
David McCullough [Wed, 6 Jul 2005 23:00:20 +0000 (23:00 +0000)]
Small spelling fixup from Dylan Griffiths.

19 years agoTweaked Makefile.in so that post-configured Makefile tests for a Cygwin build
John Williams [Tue, 3 May 2005 23:26:11 +0000 (23:26 +0000)]
Tweaked Makefile.in so that post-configured Makefile tests for a Cygwin build
environment, and if detected, explicitly adjusts the link order so that
libcygwin links *before* libiberty.a.  This gets around a Cygwin problem with
getopt()'s optind/optarg symbols.  See, for example:

http://sources.redhat.com/ml/cygwin/2004-04/msg01179.html

Also tweaked cygwin-elf.h and elf2flt.c so that Cygwin builds for MicroBlaze
target work as expected.

Patch sumbitted by John Williams <jwilliams@itee.uq.edu.au>

19 years agoExplicit assignment of elf2flt package under GPL v2.0.
John Williams [Mon, 18 Apr 2005 22:00:48 +0000 (22:00 +0000)]
Explicit assignment of elf2flt package under GPL v2.0.

19 years agoAdd missing headers.
Bernardo Innocenti [Wed, 16 Mar 2005 03:26:26 +0000 (03:26 +0000)]
Add missing headers.

Patch submitted by Bernardo Innocenti <bernie@develer.com>.

19 years agoBuild elf2flt with headers from binutils 2.15.94+.
Bernardo Innocenti [Wed, 23 Feb 2005 01:33:26 +0000 (01:33 +0000)]
Build elf2flt with headers from binutils 2.15.94+.
Should still work with older versions of binutils.

Patch submitted by Bernardo Innocenti <bernie@develer.com>.

19 years agoCygwin fixes.
Bernardo Innocenti [Wed, 23 Feb 2005 01:31:04 +0000 (01:31 +0000)]
Cygwin fixes.

Patch submitted by Stig Petter Olsrød <stigpo@users.sourceforge.net>.

20 years agoAdd support for win32 (non-cygwin) builds of elf2flt
John Williams [Mon, 8 Nov 2004 00:25:31 +0000 (00:25 +0000)]
Add support for win32 (non-cygwin) builds of elf2flt

Fix minor typo in error message

Patch submitted by Scott Thibault <thibault@gmvhdl.com>

20 years agoh8/300 forbid relocation of absolute symbol
Yoshinori Sato [Thu, 28 Oct 2004 16:54:25 +0000 (16:54 +0000)]
h8/300 forbid relocation of absolute symbol

20 years ago- add target h8300h
David McCullough [Thu, 13 May 2004 00:45:07 +0000 (00:45 +0000)]
- add target h8300h
- shared flat binary support for h8

Yoshinori Sato <ysato@users.sourceforge.jp>

20 years agoUse headers direct from binutils
David McCullough [Tue, 23 Mar 2004 04:00:15 +0000 (04:00 +0000)]
Use headers direct from binutils

Patch also includes a horrible kludge to get around some wierdness in the
elf files produced by mb-ld.

John Williams <jwilliams@itee.uq.edu.au>

20 years agoA small patch to ld-elf2flt.in, that lets an elf2flt-ified toolchain run
David McCullough [Fri, 19 Mar 2004 00:17:18 +0000 (00:17 +0000)]
A small patch to ld-elf2flt.in, that lets an elf2flt-ified toolchain run
from a directory other than that specifed in the original ./configure.
Helpful if you distribute toolchains and can't control where people
install them! :)  Also helpful if you are testing toolchains and need to
move them around after the elf2flt "make install".

John Williams <jwilliams@itee.uq.edu.au>

20 years agoSomehow the microblaze support in elf2flt/config.sub was removed in the last revision...
Greg Ungerer [Tue, 9 Mar 2004 04:49:56 +0000 (04:49 +0000)]
Somehow the microblaze support in elf2flt/config.sub was removed in the last revision (1.4 -> 1.5).

Attached patch restores it.

Submitted by John Williams <jwilliams@itee.uq.edu.au>.

20 years agoWhen the linker "-v" flag is passed, turn on shell debugging to show what
David McCullough [Tue, 10 Feb 2004 00:22:19 +0000 (00:22 +0000)]
When the linker "-v" flag is passed,  turn on shell debugging to show what
we are doing.

20 years agothis patch for elf2flt makes sure that all install dirs exist before
David McCullough [Mon, 9 Feb 2004 23:59:40 +0000 (23:59 +0000)]
this patch for elf2flt makes sure that all install dirs exist before
copying files to them.

Bernardo Innocenti <bernie@develer.com>

20 years agoIn order to use m68k-uclinux globally as the new toolchain target,
David McCullough [Mon, 5 Jan 2004 23:41:50 +0000 (23:41 +0000)]
In order to use m68k-uclinux globally as the new toolchain target,
I had to replace config.sub and config.guess in elf2flt with updated
versions already containing the required bits.

Bernardo Innocenti <bernie@develer.com>

20 years agoA slightly updated configure script
David McCullough [Mon, 5 Jan 2004 23:38:14 +0000 (23:38 +0000)]
A slightly updated configure script

21 years agoFind attached the patch for the elf2flt utility. Except the elf2flt.c file, I have...
Greg Ungerer [Tue, 11 Nov 2003 07:23:04 +0000 (07:23 +0000)]
Find attached the patch for the elf2flt utility. Except the elf2flt.c file, I have slightly modified the Makefile.in. The reason for this is to check the target architecture and if it is set to "e1", another linker script is installed. The reason for this alchemy is that our linker needs some sections that are included in the e1-elf2flt.ld script.

Patch submitted by Yannis Mitsos <gmitsos@telecom.ntua.gr>.

21 years agoAllow installing elf2flt with build directory out of the source directory
David McCullough [Tue, 23 Sep 2003 00:00:26 +0000 (00:00 +0000)]
Allow installing elf2flt with build directory out of the source directory
Bernardo Innocenti <bernie@develer.com>

21 years agoCheck in stubs.c from Miles Bader's last patch
David McCullough [Sun, 21 Sep 2003 23:19:08 +0000 (23:19 +0000)]
Check in stubs.c from Miles Bader's last patch

21 years agoAdd support for possibly missing dcgettext function
David McCullough [Tue, 9 Sep 2003 11:21:48 +0000 (11:21 +0000)]
Add support for possibly missing dcgettext function
Miles Bader <miles@gnu.org>

21 years agoAllow for really big apps, Miles Bader <miles@gnu.org>
David McCullough [Tue, 9 Sep 2003 11:20:14 +0000 (11:20 +0000)]
Allow for really big apps, Miles Bader <miles@gnu.org>

21 years agoI've now tested building the uClinux tools on a Windows box. These are
David McCullough [Wed, 3 Sep 2003 12:29:29 +0000 (12:29 +0000)]
I've now tested building the uClinux tools on a Windows box. These are
some more elf2flt fixes required for Cygwin:

Bernardo Innocenti <bernie@develer.com>

21 years agoI've found the bug! A BSS link-once section was missing in elf2flt.ld.
David McCullough [Sun, 17 Aug 2003 23:41:40 +0000 (23:41 +0000)]
I've found the bug! A BSS link-once section was missing in elf2flt.ld.
The old version of GCC probably didn't ever generate this kind of
section because it was placing everything in the data link-once section.

Patch from Bernardo Innocenti <bernie@develer.com>

21 years agoFixup after last update
David McCullough [Wed, 6 Aug 2003 16:50:41 +0000 (16:50 +0000)]
Fixup after last update

21 years agothis patch adds Cygwin support to elf2flt. Also include a copy
David McCullough [Mon, 4 Aug 2003 18:06:49 +0000 (18:06 +0000)]
this patch adds Cygwin support to elf2flt. Also include a copy
of the elf.h header which is missing on Win32.

This is just a cleaned-up version of Leon's former Cygwin patch
and I couldn't really test it on Windows. At least I can ensure
it doesn't break anything on Linux :-)

Bernardo Innocenti <bernie@develer.com>

21 years agoFixed handling of MICROBLAZE_32 reloc type. The mb-gcc distributes
David McCullough [Tue, 29 Jul 2003 10:47:48 +0000 (10:47 +0000)]
Fixed handling of MICROBLAZE_32 reloc type.  The mb-gcc distributes
relocations across several places - the "standard" ones such as
q->addend and bfd_section_vma(), but also embeds small offsets in the
actual text (code) itself.  Thus, it is necessary to examine the text to
get an initial offset, then add that to the others to get the final
fixup location.

This is required for at least MICROBLAZE_64 and MICROBLAZE_32 - others
may also need it.

Patch from John Williams <jwilliams@itee.uq.edu.au>

21 years agoPatch from Bernardo Innocenti <bernie@develer.com>:
David McCullough [Tue, 22 Jul 2003 11:31:20 +0000 (11:31 +0000)]
Patch from Bernardo Innocenti <bernie@develer.com>:

 - Use implicit a.out as output when -o option is not specified, like
   real ld does (this helps with configure tests);

 - Protect OFILE usage with quotes in case the name contains spaces;

21 years agoGet rid of false/FALSE and true/TRUE usage so we work with all versions of
David McCullough [Mon, 14 Jul 2003 00:19:02 +0000 (00:19 +0000)]
Get rid of false/FALSE and true/TRUE usage so we work with all versions of
the binutils.  Patch from Bernardo Innocenti <bernie@develer.com>.

21 years agoIgnore the -relax option. Done for the MicroBlaze, enabled for everyone,
David McCullough [Thu, 10 Jul 2003 23:17:30 +0000 (23:17 +0000)]
Ignore the -relax option.  Done for the MicroBlaze,  enabled for everyone,
hopefully this won't be a problem.

21 years agoMake sure "-relax" is only used on FINAL links, let's see if this is good
David McCullough [Thu, 10 Jul 2003 06:42:49 +0000 (06:42 +0000)]
Make sure "-relax" is only used on FINAL links, let's see if this is good
enough for the Microblaze -r/-relax incompatibility.

21 years agoMicroblaze changes from John Williams.
David McCullough [Wed, 25 Jun 2003 04:49:40 +0000 (04:49 +0000)]
Microblaze changes from John Williams.

21 years agoTwo changes from Miles Bader <miles@gnu.org>
David McCullough [Mon, 2 Jun 2003 11:11:08 +0000 (11:11 +0000)]
Two changes from Miles Bader <miles@gnu.org>

1.) Previously it was writing the relocated pointer at the pointer's section
    offset from the start of the containing _segment_, not the start of the
    containing section (it doesn't affect most people because for normal
    usage, they're the same).

2)  Add v850 HI16S/LO16 relocation type

21 years agoFix the zero padding so that it doesn't crash if tehre is more than 1K of
David McCullough [Fri, 30 May 2003 07:02:26 +0000 (07:02 +0000)]
Fix the zero padding so that it doesn't crash if tehre is more than 1K of
padding required.

Fix the shared library support which,  although the library is at a high
address,  we don't want to pad the file to +16Mb.  We also do not want the
relocations starting +16Mb into the file :-)

21 years agoAdd ktrace option so that individual flat files will produce the useful
David McCullough [Thu, 29 May 2003 03:17:56 +0000 (03:17 +0000)]
Add ktrace option so that individual flat files will produce the useful
kernel trace for debugging.

Update to latest flat.h.

21 years agoPatches from Miles Bader <miles@gnu.org>:
David McCullough [Tue, 27 May 2003 10:42:57 +0000 (10:42 +0000)]
Patches from Miles Bader <miles@gnu.org>:

* replace occurances of sectionp + q->address with r_mem.

* fix all hex debug to have 0x prefix

* Add new v850 split relocation support

21 years agoMultiple sections of a given type are combined, and any holes between
Greg Ungerer [Thu, 22 May 2003 07:10:54 +0000 (07:10 +0000)]
Multiple sections of a given type are combined, and any holes between
sections are filled with zeroes.

This changes the definition of `text', `data', and `bss' to use the
flags set by BFD, rather than requiring hardwired input section names
(for the normal case where one is using GNU ld and the elf2flt.ld linker
script, this should produce identical results, as that only ever
produces the three well-defined sections anyway).

Patch from Miles Bader <miles@gnu.org>.

21 years agoThis capability was already sort of in the code, but intertwined with
Greg Ungerer [Thu, 22 May 2003 07:07:52 +0000 (07:07 +0000)]
This capability was already sort of in the code, but intertwined with
the GOT/PIC processing option.

This patch enables a separate ELF file for relocation info to be
specified independently using a -R option.

Patch from Miles Bader <miles@gnu.org>.

21 years agoAdded support for the Microblaze archiecture.
Greg Ungerer [Thu, 22 May 2003 05:42:19 +0000 (05:42 +0000)]
Added support for the Microblaze archiecture.

The only mod outside the #ifdef TARGET_microblaze stuff in elf2flt.c is a new
entry in config.sub so it knows about the target, and a new entry in the
elf2flt.ld because the microblaze gnu tools create sections called (.bss.*).

Patch from John Williams <jwilliams@itee.uq.edu.au>.

21 years agoAdded SH support from Paul Mundt <lethal@linux-sh.org>
David McCullough [Thu, 8 May 2003 03:36:16 +0000 (03:36 +0000)]
Added SH support from Paul Mundt <lethal@linux-sh.org>

21 years agoFixup some host alignment errors from Miles Bader <miles@gnu.org>
David McCullough [Tue, 6 May 2003 07:57:20 +0000 (07:57 +0000)]
Fixup some host alignment errors from Miles Bader <miles@gnu.org>

21 years agoAdded a --disable-got-check for platforms that can't do PIC/GOT yet, and
David McCullough [Wed, 16 Apr 2003 06:16:00 +0000 (06:16 +0000)]
Added a --disable-got-check for platforms that can't do PIC/GOT yet, and
more so for h8300 binutils that cannot do the double link stage normally
used by ld-elf2flt.

This should be it for h8300 support in elf2flt.

21 years agoAdd H8 support (tested on H8300S).
David McCullough [Wed, 9 Apr 2003 05:18:13 +0000 (05:18 +0000)]
Add H8 support (tested on H8300S).

Remove GOT limit from all but the m68k targets.

Added bad reloc error count and fail

Added --with-binutils-include-dir so we can get the elf/XX headers
from binutils directly and eventually dispense with elf.h.

21 years agoarm expects the data segment on a 32 byte boundary, otherwise the GOT
David McCullough [Fri, 14 Mar 2003 04:26:42 +0000 (04:26 +0000)]
arm expects the data segment on a 32 byte boundary,  otherwise the GOT
entries can throw the alignment of the relocations out and things get
pretty ugly from there.

21 years agoput in the full contructor support by default.
David McCullough [Thu, 13 Mar 2003 06:14:40 +0000 (06:14 +0000)]
put in the full contructor support by default.

For arm-elf remove all lines containing __[CD]TOR_END__, I'm not sure why
it is different yet.

21 years agoChanges needed to fix arm-elf debugging and pthreads on ARM and m68k
David McCullough [Tue, 4 Mar 2003 10:48:37 +0000 (10:48 +0000)]
Changes needed to fix arm-elf debugging and pthreads on ARM and m68k
It also fixes random crashes on ARM where the end of data was not
aligned on a 16byte boundary.

22 years agogcc 3.1 and 3.1.1 create a .jcr ELF section, which apparently is intended to
Erik Andersen [Wed, 31 Jul 2002 09:07:41 +0000 (09:07 +0000)]
gcc 3.1 and 3.1.1 create a .jcr ELF section, which apparently is intended to
contain a list of pointers to classes to be registered during constructor
invoction time.  Sigh.  So make elf2flt work with newer gcc versions.
 -Erik

22 years agoPass through R_ARM_GOT32 and R_ARM_GOTPC relocations
Erik Andersen [Wed, 31 Jul 2002 05:13:08 +0000 (05:13 +0000)]
Pass through R_ARM_GOT32 and R_ARM_GOTPC relocations

22 years agoClean up the absolute path problems in ld-elf2flt
David McCullough [Thu, 11 Jul 2002 14:09:32 +0000 (14:09 +0000)]
Clean up the absolute path problems in ld-elf2flt

Tested with arm-elf toolchain build,  apps build
and run ok for me :-)

22 years agoFixed some spelling mistakes.
Greg Ungerer [Fri, 12 Apr 2002 13:00:44 +0000 (13:00 +0000)]
Fixed some spelling mistakes.

22 years agocleanup a few things while adding support for the shared library flat file
David McCullough [Thu, 11 Apr 2002 00:05:51 +0000 (00:05 +0000)]
cleanup a few things while adding support for the shared library flat file
stuff.

22 years agoFixup for ARM PLT32 relocation type, patch from David Byron
Greg Ungerer [Tue, 19 Mar 2002 02:21:30 +0000 (02:21 +0000)]
Fixup for ARM PLT32 relocation type, patch from David Byron
<dbyron@dbyron.com> (was <dbyron@coactive.com>).

22 years agoAdded -with-bfd-include-dir=<dir> option as we have to use the bfd.h
David McCullough [Mon, 18 Feb 2002 04:58:36 +0000 (04:58 +0000)]
Added -with-bfd-include-dir=<dir> option as we have to use the bfd.h
that matches the libbfd.a we are using,  otherwise, elf2flt may coredump if
the system header is incompatible with the libbfd.a used.

22 years agoChanges from Pauli to enable moving strings into the text area (if possible).
David McCullough [Sun, 17 Feb 2002 23:26:11 +0000 (23:26 +0000)]
Changes from Pauli to enable moving strings into the text area (if possible).

22 years agoChanges to compile under Solaris from The LEOX team <lpm@leox.org>.
David McCullough [Sun, 17 Feb 2002 23:20:14 +0000 (23:20 +0000)]
Changes to compile under Solaris from The LEOX team <lpm@leox.org>.

22 years agoImprovements to the configure system from Miles Bader (miles@lsi.nec.co.jp).
Greg Ungerer [Thu, 14 Feb 2002 13:07:29 +0000 (13:07 +0000)]
Improvements to the configure system from Miles Bader (miles@lsi.nec.co.jp).

22 years agoSome more v850 tweaks from Miles Bader <miles@gnu.org>
David McCullough [Thu, 14 Feb 2002 09:56:05 +0000 (09:56 +0000)]
Some more v850 tweaks from Miles Bader <miles@gnu.org>

22 years agoFix an error in the last update.
David McCullough [Thu, 14 Feb 2002 09:44:56 +0000 (09:44 +0000)]
Fix an error in the last update.

22 years agoFixups for the v850 from Miles Bader <miles@gnu.org>.
David McCullough [Thu, 14 Feb 2002 01:59:17 +0000 (01:59 +0000)]
Fixups for the v850 from Miles Bader <miles@gnu.org>.

22 years agoInitial revision
David McCullough [Thu, 14 Feb 2002 01:40:50 +0000 (01:40 +0000)]
Initial revision