This patch corrects the set of dynamic relocations recognised by gold
as supported by glibc, and teaches ld.bfd to report an error similar
to the gold error. Note that ld --noinhibit-exec can be used to
produce an output, supporting older ld with newer glibc if the set of
supported glibc dynamic relocations changes.
bfd/
* elf64-ppc.c (ppc64_glibc_dynamic_reloc): New function.
(ppc64_elf_relocate_section): Error if emitting unsupported
dynamic relocations.
gold/
* powerpc.cc (Target_powerpc::Scan::check_non_pic): Move REL24
to 32-bit supported.
I get some spurious changes when running autoconf/automake for various
projects in the tree. This is likely because they were generated using
distro-patched tools last time.
I ran `autoreconf -f` in the various automake projects of the
binutils-gdb tree, and this is the result. The tools I am using have
been compiled from source, from the upstream release.
bfd/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
binutils/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
gas/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
gold/ChangeLog:
* testsuite/Makefile.in: Re-generate.
gprof/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
ld/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
opcodes/ChangeLog:
* Makefile.in: Re-generate.
* configure: Re-generate.
Fixes a few leaks in bfd and ld.
bfd/
* mach-o.c (bfd_mach_o_fat_member_init): Likewise. Replace
xstrdup and xmalloc with bfd_strdup and bfd_malloc. Return an
error status. Adjust calls.
* vms-lib.c (_bfd_vms_lib_get_module): Test mhd->id earlier.
Close bfd on failure. Replace xstrdup/bfd_alloc use with
bfd_malloc. Use bfd_set_filename.
gdb/
* solib-spu.c (spu_bfd_open): Use bfd_set_filename.
* spu-linux-nat.c (spu_bfd_open): Likewise.
ld/
* emultempl/pe.em (after_open): Use bfd_set_filename.
* emultempl/pep.em (after_open): Use bfd_set_filename.
Some versions of clang apparently generate non-PIC on powerpc64le to
access common symbols. Since a common symbol and a strong definition
with the same name should resolve to the strong definition we have the
possibility of non-PIC attempting to access shared library variables.
This is really a clanger since powerpc64le is supposed to be PIC by
default, but let's see if ld can cope by generating .dynbss copies.
* elf64-ppc.c (must_be_dyn_reloc): Return 0 for TOC16 relocs.
(ppc64_elf_check_relocs): Support dynamic/copy relocs for TOC16.
(ppc64_elf_adjust_dynamic_symbol): Don't keep dynamic reloc when
needs_copy even if all relocs are in rw sections.
(dec_dynrel_count): Handle TOC16 relocs.
(ppc64_elf_relocate_section): Support dynamic relocs for TOC16.
(ppc64_elf_finish_dynamic_symbol): Adjust to handle needs_copy
semantic change.
This introduces a new bfd_set_filename function, which is then used in
various spots in gdb. This allows for the removal of some casts.
bfd/ChangeLog
2019-09-11 Tom Tromey <tom@tromey.com>
* opncls.c (bfd_set_filename): New function.
* bfd-in2.h: Regenerate.
gdb/ChangeLog
2019-09-11 Tom Tromey <tom@tromey.com>
* symfile-mem.c (symbol_file_add_from_memory): Use
bfd_set_filename.
* solib-darwin.c (darwin_bfd_open): Use bfd_set_filename.
* solib-aix.c (solib_aix_bfd_open): Use bfd_set_filename.
PC-relative relocs typically use the addend in adjusting what they are
relative to. For example:
bcl 20,31,1f
1: mflr 12
addi 12,12,xxx-1b
generates "R_PPC64_REL16 xxx+0x4" for the addi (when little-endian).
The addend reflects the fact that you want the offset relative to the
previous insn not the current one in this case.
So the question is, will we ever want to do something like that for an
instruction using R_PPC64_GOT_PCREL34? I thought so at the time I
first implemented support in ld but at the time I think the hardware
was possibly going to support pcrel+offset+reg addressing. In which
case you might want something like:
load_big_offset_into_r2
pld 3,sym-big_offset@got@pcrel(2)
which would be a way of supporting more than 8G offsets from code to
the GOT. We could do the same with
load_big_offset_into_r2
pla 9,sym-big_offset@got@pcrel
ldx 3,9,2
However, this is really a poor version of TOC-pointer relative code.
So let's go with an addend on R_PPC64_GOT_PCREL34 meaning that
sym+addend should be put in a GOT entry, and the relocation calculate
the pc-relative offset to that GOT entry.
Note that this is an extension to the ABI, which says (by the
expression given for GOT relocs) that non-zero addends on GOT and PLT
relocs are ignored. This is true for all GOT/PLT relocs, not just the
pcrel ones.
* elf64-ppc.c (ppc64_elf_check_relocs): Interpret an addend in
GOT_PCREL and PLT_PCREL relocs as affecting the value stored
in the GOT/PLT entry rather than affecting the offset to that
GOI/PLT entry.
(ppc64_elf_edit_toc, ppc64_elf_relocate_section): Likewise.
The loads and stores handled in the second instruction of a sequence
marked by R_PPC64_PCREL_OPT may be a prefix instruction. For example:
pld ra,symbol@got@pcrel
0:
pld rt,off(ra)
.reloc 0b-8,R_PPC64_PCREL_OPT,(.-8)-(0b-8)
can be optimised to
pld rt,symbol+off@pcrel
pnop
* elf64-ppc.c (xlate_pcrel_opt): Handle prefix loads and stores
in second instruction.
(ppc64_elf_relocate_section): Likewise.
This patch fixes the worst of the cases where libbfd might terminate
a program due to calling xstrdup or xmalloc. I've also fixed some
error paths that didn't clean up properly.
PR 24955
* libbfd-in.h (bfd_strdup): New inline function.
* archive.c (_bfd_get_elt_at_filepos): Use bfd_strdup. Close
bfd on error.
* elfcode.h (_bfd_elf_bfd_from_remote_memory): Use bfd_strdup.
* opncls.c (bfd_fopen): Use bfd_strdup. Close fd and stream
on error.
(bfd_openstreamr): Use bfd_strdup.
(bfd_openr_iovec, bfd_openw, bfd_create): Likewise.
* plugin.c (try_load_plugin): Use bfd_malloc.
* libbfd.h: Regenerate.
This fixes a small leak of debug_filename. bfd_openr copies the file
name since git commit 1be5090bca.
PR 11983
* dwarf2.c (_bfd_dwarf2_slurp_debug_info): Free debug_filename
on success. Tidy.
The linker doesn't allocate memory space for sections that are only SEC_ALLOC
and SEC_THREAD_LOCAL. See the IS_TBSS test in ld/ldlang.c. So we need to
pretend that .tdata.dyn sections have contents to get the right result. It
will be marked this way anyways if there is a .tdata section to merge with.
bfd/
PR 23825
* elfnn-riscv.c (riscv_elf_create_dynamic_sections): Add SEC_LOAD,
SEC_DATA, and SEC_HAS_CONTENTS to .tdata.dyn section.
This was noticed while trying to test the compiler -msave-restore support.
Putting non-pic code in a shared library gives a linker error, but doesn't
stop the build.
rohan:2030$ cat libtmp.c
extern int sub2 (int);
int sub (int i) { return sub2 (i + 10); }
rohan:2031$ cat libtmp2.c
extern int sub (int);
int sub2 (int i) { return sub (i + 10); }
rohan:2032$ riscv64-unknown-linux-gnu-gcc --shared -o libtmp.so libtmp.c
rohan:2033$ riscv64-unknown-linux-gnu-gcc --shared -o libtmp2.so libtmp2.c libtmp.so
/home/jimw/FOSS/install-riscv64/lib/gcc/riscv64-unknown-linux-gnu/8.3.0/../../../../riscv64-unknown-linux-gnu/bin/ld: /tmp/cctrsIBe.o(.text+0x18): unresolvable R_RISCV_CALL relocation against symbol `sub'
rohan:2034$ echo $?
0
rohan:2035$ ls -lt libtmp2.so
-rwxr-xr-x 1 jimw jimw 6912 Aug 30 14:32 libtmp2.so
rohan:2036$
The patch fixes this by forcing a linker error. I now get this.
ohan:2059$ sh tmp.script
/home/jimw/FOSS/BINUTILS/X-riscv64-linux/ld/ld-new: libtmp2.o(.text+0x18): unresolvable R_RISCV_CALL relocation against symbol `sub'
/home/jimw/FOSS/BINUTILS/X-riscv64-linux/ld/ld-new: final link failed: bad value
rohan:2060$ echo $?
1
rohan:2061$ ls -lt libtmp2.so
ls: cannot access 'libtmp2.so': No such file or directory
bfd/
* elfnn-riscv.c (riscv_elf_relocate_section): For unresolvable reloc
error, call bfd_set_error, set ret to FALSE, and goto out label.
We need to copy BFD_COMPRESS, BFD_DECOMPRESS and BFD_COMPRESS_GABI flags
for thin archive.
PR ld/24951
* archive.c (_bfd_get_elt_at_filepos): Copy BFD_COMPRESS,
BFD_DECOMPRESS and BFD_COMPRESS_GABI flags for thin archive.
This removes a restriction on various R_PPC_EMB relocations that has
been present for ppc32 since 1996-04-26 git commit e25a798839. As far
as I know, only those relocs that would require addressing via r2 for
.sdata2/.sbss2 access are disallowed in shared libraries.
PR 24697
* elf32-ppc.c (ppc_elf_check_relocs): Call bad_shared_reloc
when !bfd_link_executable for R_PPC_EMB_SDA2I16 and
R_PPC_EMB_SDA2REL. Don't call bad_shared_reloc for any other
reloc.
We can easily support an offset on the second instruction of a
sequence marked with R_PPC64_PCREL_OPT. For example,
pla ra,symbol@pcrel
ld rt,off(ra)
can be optimised to
pld rt,symbol+off@pcrel
nop
* elf64-ppc.c (xlate_pcrel_opt): Add poff parameter. Allow offset
on second insn, return it in poff.
(ppc64_elf_relocate_section): Add offset to paddi addend for
PCREL_OPT.
This was broken when I changed how we compute the value for the gp register.
It used to be computed inside the sdata section. Now it is computed at the
end which makes it an abs section symbol. There is code that tries to use
the alignment of the section that the gp value is in, but this does not work
if it is in the abs section, as the abs section has alignment of 1 byte.
There are people using alternative linker scripts that still define it in the
sdata section, so the code is still useful. Thus adding a check to disable
this when gp is in the abs section.
bfd/
* elfnn-riscv.c (_bfd_riscv_relax_lui): Add check to exclude abs
section when setting max_alignment. Update comment.
(_bfd_riscv_relax_pc): Likewise.
BFD was leaking memory in bfd_check_format_matches. As part of
deciding the proper format of an archive, BFD looks at the format of
the first file stored. That file's bfd was left open for reasons
given in a comment removed in git commit 0e71e4955c that said:
/* We ought to close `first' here, but we can't, because
we have no way to remove it from the archive cache.
It's close to impossible to figure out when we can
release bfd_ardata. FIXME. */
Well, things have changed since that comment was true and we now can
remove files from the archive cache. Closing the first file is good
and cures some of the leaks. Other leaks are caused by
bfd_check_format_matches throwing away bfd tdata before trying a new
match. That lost the element cache set up when format checking the
first element in the archive. The easiest and cleanest fix is to
simply disable the caching when checking the first element.
PR 24891
* bfd.c (struct bfd): Add no_element_cache.
* archive.c (_bfd_get_elt_at_filepos): Don't add element to
archive cache when no_element_cache.
(bfd_generic_archive_p): Set no_element_cache when opening first
element to check format. Close first element too.
(do_slurp_bsd_armap): Don't zero ardata->cache here.
* bfd-in2.h: Regenerate.
Alan Modra
GDB Administrator
Simon Marchi
Phil Blundell
Tom Tromey
Nick Clifton
Jim Wilson
H.J. Lu