This patch arranges to have OSABI set to ELFOSABI_GNU (if not set to
some other non-zero value) when gold outputs an ifunc local or global
symbol, or a unique global symbol to either .dynsym or .symtab.
STT_GNU_IFUNC and STB_GNU_UNIQUE have values in the LOOS to HIOS range
and therefore require interpretation according to OSABI.
I'm not sure why parameters->target() is const Target& while
parameters->sized_target() is Sized_target*, but it's inconvenient to
use the latter in Symbol_table::finalize. So this patch adds another
const_cast complained about in layout.cc and gold.cc.
PR 24853
* symtab.h (set_has_gnu_output, has_gnu_output_): New.
* symtab.cc (Symbol_table::Symbol_table): Init has_gnu_output_.
(Symbol_table::finalize): Set ELFOSABI_GNU when has_gnu_output_.
(Symbol_table::set_dynsym_indexes, Symbol_table::sized_finalize):
Call set_has_gnu_output for STT_GNU_IFUNC and STB_GNU_UNIQUE globals.
* object.cc (Sized_relobj_file::do_finalize_local_symbols): Call
set_has_gnu_output when STT_GNU_IFUNC locals will be output.
When generating _end, _edata, etc. symbols, and a version script provides
a version name, and we are linking against another shared library that
provides those symbols with a different version, gold ends up trying to
resolve the other shared library's symbols to the new definitions, resulting
in two copies of each symbol, one as default, and one as non-default.
This patch tests for that condition, and ignores the symbols provided
by the other shared library.
gold/
PR gold/23409
* symtab.cc (Symbol_table::define_special_symbol): Add check for
version name on existing symbol.
* testsuite/Makefile.am (ver_test_pr23409): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/ver_test_pr23409.sh: New test script.
* testsuite/ver_test_pr23409_1.script: New version script.
* testsuite/ver_test_pr23409_2.script: New version script.
When we have a weak reference to a symbol defined in an
as-needed library, and that library ends up not-needed, gold
simply clears the version information in the symbol table, even
if the symbol could have been resolved by a needed library later
in the link order. This results in a loss of version information,
which can cause the program to bind to the wrong version at run
time.
This patch lets a dynamic definition override an earlier one if
the earlier one is from a not-needed library, so that we can
retain the version information from the binding to the needed
library. In order to do that, the tracking of needed/not-needed
had to be moved up to symbol resolution time, instead of during
Symbol_table::set_dynsym_indexes().
In cases where we still end up discarding version information,
I've added a warning.
For the original problem report and discussion, see:
https://stackoverflow.com/questions/50751421/undefined-behavior-in-shared-lib-using-libpthread-but-not-having-it-in-elf-as-d
gold/
* resolve.cc (Symbol_table::resolve): Rename tobinding to
orig_tobinding. Call set_is_needed() for objects that resolve
non-weak references.
(Symbol_table::should_override): Allow a dynamic definition to
override an earlier one in a not-needed library.
* symtab.cc (Symbol_table::set_dynsym_indexes): Remove separate
processing for as-needed symbols. Add warning when discarding
version informatin.
* testsuite/Makefile.am (weak_as_needed): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/weak_as_needed.sh: New test script.
* testsuite/weak_as_needed_a.c: New source file.
* testsuite/weak_as_needed_b.c: New source file.
* testsuite/weak_as_needed_b.script: New version script.
* testsuite/weak_as_needed_c.c: New source file.
* testsuite/weak_as_needed_c.script: New version script.
gold/
PR gold/16504
* dynobj.cc (Versions::symbol_section_contents): Don't set
VERSYM_HIDDEN flag for undefined symbols.
* symtab.cc (Symbol_table::add_from_object): Don't override default
version definition with a different default version.
* symtab.h (Symbol::from_dyn): New method.
* testsuite/plugin_test.c (struct sym_info): Add ver field.
(claim_file_hook): Pass symbol version to plugin API.
(parse_readelf_line): Parse symbol version.
* testsuite/Makefile.am (ver_test_pr16504): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/ver_test_pr16504.sh: New test script.
* testsuite/ver_test_pr16504_a.c: New source file.
* testsuite/ver_test_pr16504_a.script: New version script.
* testsuite/ver_test_pr16504_b.c: New source file.
* testsuite/ver_test_pr16504_b.script: New version script.
If a shared library contains an undefined symbol and LTO adds
a new reference to that same undefined symbol, the reference in the new
object added by the plugin would not trigger a rescan of the archive
containing the symbol.
2017-11-17 Stephen Crane <sjc@immunant.com>
gold/
PR gold/22448
* symtab.cc (Symbol_table::add_from_object): Only rescan for
undefined symbols in regular, not dynamic, objects.
Since special symbol must be defined in a regular object, definition
from a dynamic object should be ignored. If special symbol has the
hidden or internal visibility, reference from a dynamic object should
also be ignored. Also __start and __stop symbols in a dynamic object
shouldn't be preempted.
PR gold/22291
* layout.cc (Layout::define_section_symbols): Use STV_PROTECTED
for __start and __stop symbols.
* symtab.cc (Symbol_table::define_special_symbol): Add an
argument, visibility. Ignore definition and reference from
a dynamic object, depending on visibility.
(Symbol_table::do_define_in_output_data): Pass visibility to
define_special_symbol.
(Symbol_table::do_define_in_output_segment): Likewise.
(Symbol_table::do_define_as_constant): Likewise.
(Symbol_table::add_undefined_symbol_from_command_line): Pass
STV_DEFAULT to define_special_symbol.
* symtab.h (Symbol_table::define_special_symbol): Add an
argument, visibility.
Some targets prefix global symbols with "_".
bfd/
* archive.c (_bfd_compute_and_write_armap): Match "__gnu_lto_slim"
optionally prefixed with "_".
* linker.c (_bfd_generic_link_add_one_symbol): Likewise.
binutils/
* nm.c (filter_symbols): Match "__gnu_lto_slim" optionally prefixed
with "_".
gold/
* symtab.cc (Symbol_table::add_from_relobj): Match "__gnu_lto_slim"
optionally prefixed with "_".
ld/
* testsuite/ld-plugin/lto-3r.d: Match "__gnu_lto_v" optionally
prefixed with "_".
* testsuite/ld-plugin/lto-5r.d: Likewise.
On 64-bit targets there is a 32-bit hole in symbol->u_, and another
due to symbol flags exceeding 32 bits. By splitting the union,
the total size of the class reduces by one 64-bit word.
* symtab.h (Symbol): Split u_ into u1_ and u2_. Adjust accessors
to suit. Move plt_offset_ before got_offsets_.
* symtab.cc (Symbol::init_fields): Adjust for union change.
(Symbol::init_base_output_data): Likewise.
(Symbol::init_base_output_segment): Likewise.
(Symbol::allocate_base_common): Likewise.
(Symbol::output_section): Likewise.
(Symbol::set_output_section): Likewise.
(Symbol::set_output_segment): Likewise.
* resolve.cc (Symbol::override_base): Likewise.
(Symbol::override_base_with_special): Likewise.
This patch provides a flag for PowerPC64 ELFv2 use in class Symbol,
and modifies Sized_target::resolve to return whether the symbol has
been resolved. If not, normal processing continues. I use this for
PowerPC64 ELFv2 to keep track of whether a symbol has any definition
with non-zero localentry, in order to disable --plt-localentry for
that symbol.
PR 21847
* powerpc.cc (Target_powerpc::is_elfv2_localentry0): Test
non_zero_localentry.
(Target_powerpc::resolve): New function.
(powerpc_info): Set has_resolve for 64-bit.
* target.h (Sized_target::resolve): Return bool.
* resolve.cc (Symbol_table::resolve): Continue with normal
processing when target resolve returns false.
* symtab.h (Symbol::non_zero_localentry, set_non_zero_localentry):
New accessors.
(Symbol::non_zero_localentry_): New flag bit.
* symtab.cc (Symbol::init_fields): Init non_zero_localentry_.
Gold was not placing forced-local symbols (e.g., hidden visibility)
at the front of the dynamic symbol table, or including them in the
count of local symbols recorded in the .dynsym section's sh_info field.
gold/
* layout.cc (Layout::finalize): Track count of forced-local symbols
in .dynsym.
(Layout::create_symtab_sections): Add local_dynamic_count parameter;
use that instead of sh_info value.
(Layout::create_dynamic_symtab): Add pforced_local_dynamic_count
parameter; pass it to Symtab::set_dynsym_indexes(). Include forced
locals in sh_info value. Pass index of first real global to
Dynobj::create_gnu_hash_table() and Dynobj::create_elf_hash_table().
* layout.h (Layout::create_symtab_sections): Add local_dynamic_count
parameter.
(Layout::create_dynamic_symtab): Add pforced_local_dynamic_count
parameter.
* symtab.cc (Symbol_table::set_dynsym_indexes): Add pforced_local_count
parameter. Process forced-local symbols first and return the count.
(Symbol_table::finalize): Update comments.
* symtab.h (Symbol_table::set_dynsym_indexes): Add pforced_local_count
parameter.
(Symbol_table::first_dynamic_global_index_): Update comment.
(Symbol_table::dynamic_count_): Update comment.
* testsuite/Makefile.am (ifuncmod1.sh): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/ifuncmod1.sh: New shell script.
When creating a predefined hidden symbol like _GLOBAL_OFFSET_TABLE_, gold
was incorrectly letting a version script add a version to the symbol,
resulting in two copies of the symbol, both STB_LOCAL, but one of which
was grouped in the globals part of the symbol table.
gold/
* symtab.cc (Symbol_table::define_special_symbol): Add is_forced_local
parameter; if set, do not check version script.
(Symbol_table::do_define_in_output_data): Pass is_forced_local for
STB_LOCAL predefined symbols.
(Symbol_table::do_define_in_output_segment): Likewise.
(Symbol_table::do_define_in_output_segment): Likewise.
(Symbol_table::do_define_as_constant): Likewise.
* symtab.h (Symbol_table::define_special_symbol): Add is_forced_local
parameter. Adjust all callers.
* testsuite/Makefile.am (ver_test_8.sh): New test case.
* testsuite/Makefile.in: Regenerate.
* ver_test_8.sh: New test script.
Linker-created symbols like __ehdr_start, __etext, __edata, and end
should be relocatable, not absolute.
gold/
* output.cc (Output_segment::first_section): Return NULL if there are
no sections in the segment.
* output.h (Output_segment::first_section_load_address): Assert that
first section is not NULL.
* symtab.cc (Symbol_table::sized_write_globals): Attach linker-created
segment-relative symbols to first section of the segment.
If the .bss section has other data in it besides common allocations,
gold was subtracting the wrong section start address from the symbol
value.
gold/
PR gold/20976
* symtab.cc (Symbol_table::sized_write_globals): Use address of
output section, not input section.
* testsuite/Makefile.am (pr20976): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/pr20976.c: New source file.
When we see an unversioned symbol reference in a shared library, followed
by a default definition of the symbol in another shared library, we were
treating them as separate symbols. That should only happen when both are
definitions.
gold/
PR gold/20238
* symtab.cc (Symbol_table::define_default_version): Check that
unversioned symbol is defined.
2016-08-10 James Clarke <jrtc27@jrtc27.com>
gold/
PR gold/20443
* symtab.cc (Symbol_table::add_from_relobj): Handle NULL symbols,
which will be present for STT_SPARC_REGISTER.
(Symbol_table::add_from_pluginobj): Likewise.
(Symbol_table::add_from_dynobj): Likewise.
(Symbol_table::add_from_incrobj): Removed dead code.
In PR 16979, a reference to malloc is being resolved to an unversioned
reference in libmalloc.so. When linked with --as-needed, however, the
dynamic table does not list libmalloc.so as a DT_NEEDED library.
If we have a reference to an unversioned symbol in a shared object,
and we later see a versioned definition in another shared object, we
were overriding the first definition with the second in the process of
defining the default version. As a result, we no longer think that the
first shared object was actually needed to resolve any symbols, and we
don't list it as a DT_NEEDED library.
This patch fixes the problem by treating the two definitions as separate
symbols, so the second definition does not override the first.
2016-03-30 Cary Coutant <ccoutant@gmail.com>
gold/
PR gold/16979
* symtab.cc (Symbol_table::define_default_version): Check for case
where symbols are both in different shared objects.
gold/
PR gold/19019
* layout.h (Layout::add_target_specific_dynamic_tag): New function.
* layout.cc (Layout::add_target_specific_dynamic_tag): New function.
* mips.cc (Target_mips::make_symbol): Adjust function signature.
* sparc.cc (Target_sparc::Target_sparc): Initialize register_syms_.
(Target_sparc::do_is_defined_by_abi): Remove test for
STT_SPARC_REGISTER.
(Target_sparc::Register_symbol): New struct type.
(Target_sparc::register_syms_): New data member.
(Target_sparc<64, true>::sparc_info): Set has_make_symbol to true.
(Target_sparc::make_symbol): New function.
(Target_sparc::do_finalize_sections): Add register symbols and new
dynamic table entries.
* symtab.h (Sized_symbol::init_undefined): Add value parameter.
(Symbol_table::add_target_global_symbol): New function.
(Symbol_table::target_symbols_): New data member.
* symtab.cc (Sized_symbol::init_undefined): Add value parameter.
(Symbol_table::Symbol_table): Initialize target_symbols_.
(Symbol_table::add_from_object): Pass additional parameters to
Target::make_symbol.
(Symbol_table::define_special_symbol): Likewise.
(Symbol_table::add_undefined_symbol_from_command_line): Pass 0 for
undefined symbol value.
(Symbol_table::set_dynsym_indexes): Process target-specific symbols.
(Symbol_table::sized_finalize): Likewise.
(Symbol_table::sized_write_globals): Likewise.
* target.h (Sized_target::make_symbol): Add name, st_type, object,
st_shndx, and value parameters.
The previous commit to fix PR gold/18886 converted STT_IFUNC
to STT_FUNC when resolving to a symbol defined in a shared library.
This leads to an internal error if the shared library symbol is
seen first, as we do not convert the symbol at all.
We need to override the STT_IFUNC in add_from_dynobj() instead of
in override_base().
gold/
PR gold/18930
PR gold/18886
* resolve.cc (Symbol::override_base): Don't convert IFUNC symbols here.
* symtab.cc (Symbol_table::add_from_dynobj): Convert them here instead.
Turn IFUNC symbols from shared libraries into normal FUNC symbols when
we are resolving symbol references, instead of when we are writing out
the symbol table.
PR gold/18886
* resolve.cc (Symbol::override_base): Turn IFUNC symbols from
shared libraries into normal FUNC symbols.
* symtab.cc (Symbol_table::sized_write_symbol): Assert IFUNC
symbols aren't from shared libraries.
When a user object overrides a versioned definition in a shared object,
it's possible to trigger an internal error because the versioned
definition we saw didn't override the unversioned definition. This
was introduced by the fix for PR 18703. This patch fixes the problem
by reordering the tests so that we do not check ret->is_default()
until we're confident that the symbol is in fact versioned.
gold/
PR gold/18866
PR gold/18703
* symtab.cc (Symbol_table): Reorder conditions to avoid internal error.
If a symbol is defined with ".symver foo,foo@VER", the assembler
creates two symbols in the object: one unversioned, and one with
the (non-default) version "VER". If foo is listed in a version
script, gold would then make the first of those symbols the
default version, and would ignore the second symbol as a
duplicate, without making it a non-default version. While this is
arguably reasonable behavior, it doesn't match Gnu ld behavior,
so this patch fixes that by allowing the second definition to
override the first by resetting the "default version" indication.
Several test cases from the Gnu ld testsuite also exposed another
related problem, where a symbol defined with ".symver foo,foo@",
placed into a shared library, is not handled properly by gold.
This patch also fixes that case, binding the symbol to the base
version.
gold/
PR gold/18703
* dynobj.cc (Versions::record_version): Handle symbol defined with
base version.
(Versions::symbol_section_contents): Likewise.
* symtab.h (Symbol::set_is_not_default): New class method.
(Symbol_table::resolve): Add is_default_version parameter.
(Symbol_table::should_override): Likewise.
* resolve.cc (Symbol_table::resolve): Add is_default_version parameter,
and pass to should_override. Adjust all callers and explicit
instantiations.
(Symbol_table::should_override): Add is_default_value parameter;
allow default version in a dynamic object to override existing
definition from same object.
* symtab.cc (Symbol_table::add_from_object): Handle case where same
symbol is defined as unversioned and non-default version in the same
object.
* testsuite/Makefile.am (ver_test_13): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/ver_test_4.cc: Add test for symbol with base version.
* testsuite/ver_test_4.sh: Likewise.
* testsuite/ver_test_13.c: New source file.
* testsuite/ver_test_13.script: New version script.
* testsuite/ver_test_13.sh: New test case.
When creating the special __start_ and __stop_ section symbols, gold
does not check the version script to see if they should be local
instead of global.
2015-07-21 Cary Coutant <ccoutant@gmail.com>
gold/
PR gold/18548
* symtab.cc (Symbol_table::do_define_in_output_data): Check for
forced local symbol even when oldsym != NULL.
(Symbol_table::do_define_in_output_segment): Likewise.
(Symbol_table::do_define_as_constant): Likewise.
This patch adds option --weak-unresolved-symbols to treat unresolved symbols as
weak references. This is helpful when we want the link to succeed with unresolved
symbols and the dynamic loader to not complain at run-time. Option
--warn-unresolved-symbols lets the link succeed but could fail at run-time with
unresolved symbol warnings especially when the unresolved symbols have GOT entries
and dynamic relocations against them, like when -fPIE is used.
This patch fixes Symbol::final_value_is_known so that a defined TLS symbol
in a PIE link is treated as having a known final value, thus allowing
GD->LE TLS optimization, eliminating an unnecessary GOT entry and dynamic
relocation.
gold/
* symtab.cc (Symbol::final_value_is_known): Check for TLS symbol
in a PIE link.
* testsuite/Makefile.am (tls_pie_test.sh): New test.
* testsuite/Makefile.in: Regenerate.
* testsuite/tls_pie_test.sh: New.
gold/
* debug.h (DEBUG_LOCATION): New.
(DEBUG_ALL): Include DEBUG_LOCATION.
(debug_string_to_enum): Add DEBUG_LOCATION.
* dwarf_reader.cc (Sized_dwarf_line_info::read_lines): Fix debug
output to print correct context.
(Sized_dwarf_line_info::do_addr2line): Add debug output. Return
up to 4 more locations at the beginning of the function.
* symtab.cc (Symbol_table::detect_odr_violations): Get canonical
result before sorting list of line numbers.
* testsuite/debug_msg.sh: Allow range of line numbers for
canonical results on optimized code.
pr17670 is about an assert triggering on a branch to an undefined
weak symbol, the symbol being undefined due to dropping its comdat
group section. (Well sort of. The symbol is actually defined in
an .opd section which isn't part of the group, but the code section
the opd entry points at is dropped.) So don't assert.
Also, don't make long branch stubs to such symbols, and arrange to
have target-reloc.h code warn when applying relocs that use the sym.
PR 17670
* symtab.cc (Symbol::set_undefined): Remove assertion.
* powerpc.cc (Target_powerpc::symval_for_branch): Don't assert
on symbols defined in discarded sections, instead return false.
Rearrange params, update all callers.
(Target_powerpc::Branch_info::make_stub): Don't make stubs for
branches to syms in discarded sections.
(Global_symbol_visitor_opd::operator()): Set discarded opd syms
undefined and flag as discarded.
(Target_powerpc::Relocate::relocate): Localize variable.
The plugin API doesn't provide a way for the claimed file handler to
identify a TLS symbol, so when adding a common TLS symbol, gold
mistakenly places the symbol in the non-TLS commons list, and does
not override it when we see the replacement symbol that is marked
as TLS. Consequently, we allocate the TLS common symbol as a regular
common, and, if it's the only TLS in the program, we'll give an
internal error because we haven't allocated a TLS segment.
This patch fixes the problem by removing an exclusion where common
symbols would not override the placeholder symbols, but checking to
see if the size needs adjusting (the original reason for the exclusion).
Furthermore, we need to avoid putting placeholder symbols in the common
list, and wait until we see a real common symbol with a type we can
trust.
gold/
PR gold/17432
* resolve.cc (Symbol_table::resolve): Override common placeholder
symbols, but adjust sizes.
* symtab.cc (Symbol_table::add_from_object): Don't add placeholder
symbols to common lists.
2014-03-10 Sasa Stankovic <Sasa.Stankovic@imgtec.com>
gold/
* symtab.cc (Symbol_table::sized_write_globals): Allow a target to
adjust dynamic symbol value.
* target.h (Target::adjust_dyn_symbol): New function.
(Target::do_adjust_dyn_symbol): New function.
2014-03-10 Sasa Stankovic <Sasa.Stankovic@imgtec.com>
gold/
* symtab.cc (Symbol_table::set_dynsym_indexes): Allow a target to set
dynsym indexes.
* target.h (Target::has_custom_set_dynsym_indexes): New function.
(Target::do_has_custom_set_dynsym_indexes): New function.
(Target::set_dynsym_indexes): New function.
(Target::do_set_dynsym_indexes): New function.
PR gold/13577 complains that even though symbols listed in
the --dynamic-list script are exported, they are still bound symbolically
if -Bsymbolic is also used. There are two underlying problems here.
First, -Bsymbolic should be overridden by --dynamic-list, since the
dynamic list provides an explicit list of symbols that are not bound
within the library, and if we go ahead and set DT_SYMBOLIC, then the
dynamic loader will bind it within the library anyway. Second, gold
did not properly identify the symbols listed in the file as preemptible.
PR gold/16530 complains that symbols listed in the --dynamic-list script
can still be garbage collected. I've fixed this by checking the symbols
as they're added to the symbol table. (Unlike the --export-dynamic-symbol
option, we can't iterate over the list, because the --dynamic-list script
can have wildcards in it.)
gold/
2014-02-05 Cary Coutant <ccoutant@google.com>
PR gold/13577
* options.cc (General_options::parse_dynamic_list):
Set have_dynamic_list_.
(General_options::General_options): Initialize have_dynamic_list_.
(General_options::finalize): Turn off -Bsymbolic and
-Bsymbolic-functions if --dynamic-list provided.
* options.h (General_options::have_dynamic_list): New function.
(General_options::have_dynamic_list_): New data member.
* symtab.h (Symbol::is_preemptible): Handle --dynamic-list
correctly.
PR gold/16530
* symtab.cc (Symbol_table::add_from_relobj): If symbol is named
in --dynamic-list, mark it.
* testsuite/Makefile.am (gc_dynamic_list_test.sh): New test case.
(dynamic_list_2): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/dynamic_list_2.cc: New file.
* testsuite/dynamic_list_2.t: New file.
* testsuite/dynamic_list_lib1.cc: New file.
* testsuite/dynamic_list_lib2.cc: New file.
* testsuite/gc_dynamic_list_test.c: New file.
* testsuite/gc_dynamic_list_test.sh: New file.
* testsuite/gc_dynamic_list_test.t: New file.
Alan Modra
Cary Coutant
Stephen Crane
H.J. Lu
James Clarke
Sriraman Tallam
Ian Coolidge
Rafael Ávila de Espíndola
Cary Coutant