This patch supports linking powerpc64 glibc with gold, specifically
the __tls_get_addr call in elf/dl-sym.c. That call lacks marker
relocations tying it to the arg setup instructions, but the arg setup
insns are also contructed lacking the usual relocations on a Global
Dynamic TLS code sequence. So there is no chance that anything in
that sequence might be wrongly edited by the linker.
In fact, the aim of linking glibc could have been supported by simply
omitting the error whenever TLS optimisation is disabled, as it is
when linking a shared library. The patch goes further than that,
disabling TLS GD and LD sequence optimisation on a per-object basis
for object files lacking marker relocs.
PR gold/27625
* powerpc.cc (Powerpc_relobj): Add no_tls_marker_, tls_marker_,
and tls_opt_error_ variables and accessors.
(Target_powerpc::Scan::local, global): Call set_tls_marker and
set_no_tls_marker for GD and LD code sequence relocations.
(Target_powerpc::Relocate::relocate): Downgrade the "lacks marker
reloc" error to a warning when safe to do so, and omit the error
entirely if not optimising TLS sequences. Do not optimise GD and
LD sequences for objects lacking marker relocs.
(Target_powerpc::relocate_relocs): Heed no_tls_marker here too.
gold/
PR gold/pr23870
* testsuite/aarch64_pr23870_bar.c: Return a magic value.
* testsuite/aarch64_pr23870_foo.c: Check the magic value and return
success or failure.
The .debug_line secton in DWARF5 has a byte for address size and a byte
for segment selector after DWARF version. Skip them for DWARF5.
PR gold/27246
* dwarf_reader.cc (Sized_dwarf_line_info::read_header_prolog):
Skip address size and segment selector for DWARF5.
A testcase with only ifuncs can result in no plt section (ifunc plt
entries might instead be in iplt), which means we can get to this code
without a static link.
PR 27203
* powerpc.cc (do_plt_fde_location): Remove doing_static_link
assertion.
On Fedora 33 x86-64 with glibc 2.32-3, ifuncmain6pie failed with:
./ifuncmain6pie: IFUNC symbol 'foo' referenced in './ifuncmod6.so' is defined in the executable and creates an unsatisfiable circular dependency.
FAIL ifuncmain6pie (exit status: 127)
Remove non-JUMP_SLOT relocations against foo in ifuncmod6.so, which
trigger the circular IFUNC dependency.
* testsuite/ifuncmain6pie.c: Remove non-JUMP_SLOT relocations
against foo in ifuncmod6.so.
* testsuite/ifuncmod6.c: Likewise.
This fixes a thinko in commit fa40fbe484. st_other global entry bits
are relevant only for 64-bit ELFv2. PowerPC gold leaves local sym
vector of st_other bits as NULL for 32-bit, hence the segfault.
PR 27140
* powerpc.cc (Target_powerpc::Branch_info::make_stub): Only access
object->st_other() when 64-bit.
(Stub_table::add_long_branch_entry): Ignore "other" when 32-bit.
Move sym declaration just before use to avoid -Wmaybe-uninitialized
warning from GCC 11.
PR gold/27097
* incremental.cc (Sized_relobj_incr::do_add_symbols): Move sym
declaration just before use.
(Sized_incr_dynobj::do_add_symbols): Likewise.
* plugin.cc (Sized_pluginobj::do_add_symbols): Likewise.
Add accessor methods to elfcpp::Ehdr class for EI_OSABI and EI_ABIVERSION;
use those to simplify initialization of Osabi class and eliminate the need
to template the class.
elfcpp/
* elfcpp.h (class Ehdr): Add get_ei_osabi and get_ei_abiversion methods.
gold/
* dwp.cc (class Dwo_file): Use new Ehdr::get_ei_osabi and
get_ei_abiversion methods.
* incremental.cc (make_sized_incremental_binary): Likewise.
* object.cc (Sized_relobj_file::Sized_relobj_file): Likewise.
(make_elf_sized_object): Likewise.
* object.h (class Osabi): Make the class untemplated.
During an incremental link, we should simply ignore the Gnu properties
sections. We were not handling them properly -- failing to process the
properties from the base file. While that could be fixed, the property
sections are meant for deployed binaries, and incremental linking is
for development, so keeping the properties sections just adds to the
likelihood of forcing a full link sooner.
gold/
PR gold/23539
* object.cc (Sized_relobj_file::layout_gnu_property_section): Ignore
Gnu properties during incremental links.
When processing the incremental update, incoming .rodata merge sections
do not match the corresponding section in the base file, because the
SHF_MERGE flag had not been masked out of the latter.
gold/
PR gold/24123
* layout.cc (Layout::init_fixed_output_section): Mask out flags that
should be ignored when matching sections.
I'm unsure why this is deserving of a warning. Not writing the most
efficient code surely can't be a real problem, but that is what
https://gcc.gnu.org/bugzilla//show_bug.cgi?id=88059#c1 seems to say.
plugin.cc:528:10: error: 'char* strncpy(char*, const char*, size_t)' specified bound depends on the length of the source argument [-Werror=stringop-overflow=]
528 | strncpy(tempdir, dir_template, len);
| ~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~
plugin.cc:526:22: note: length computed here
526 | size_t len = strlen(dir_template) + 1;
| ~~~~~~^~~~~~~~~~~~~~
* plugin.cc (Plugin_recorder::init): Replace strncpy with memcpy.
When relocating debug sections, get the section index for the linkonce
section. Since symbols referenced in debugging sections can be defined
a single comdat section with a different section name, also check the
single comdat section.
PR gold/26937
* object.cc (Sized_relobj_file::map_to_kept_section): Get the
section index for linkonce section. Also check the single
comdat section.
* testsuite/Makefile.am (check_SCRIPTS): Add pr26936.sh.
(check_DATA): Add pr26936a.stdout and pr26936b.stdout.
(MOSTLYCLEANFILES): Add pr26936a and pr26936b.
(pr26936a.stdout): New target.
(pr26936a): Likewise.
(pr26936b.stdout): Likewise.
(pr26936b): Likewise.
(pr26936a.o): Likewise.
(pr26936b.o): Likewise.
(pr26936c.o): Likewise.
(pr26936d.o): Likewise.
* testsuite/Makefile.in: Regenerated.
* testsuite/pr26936.sh: New file.
* testsuite/pr26936a.s: Likewise.
* testsuite/pr26936b.s: Likewise.
* testsuite/pr26936c.s: Likewise.
* testsuite/pr26936d.s: Likewise.
incremental-dump wants to instantiate Target_powerpc without options
being set up. This patch fixes
internal error in options, at gold/parameters.h:92
* powerpc.cc (Target_powerpc::no_tprel_opt_): Rename from tprel_opt_.
Init to false.
(Target_powerpc::tprel_opt): Test parameters->options().tls_optimize().
(Target_powerpc::set_tprel_opt): Delete.
(Target_powerpc::set_no_tprel_opt): New function. Update all uses
of set_tprel_opt.
class Plugin_manager has
// A pointer to the current plugin. Used while loading plugins.
Plugin_list::iterator current_;
The same iterator is shared by all threads. It is OK to use it to load
plugins since only one thread loads plugins. Avoid sharing Plugin_list
iterator in all other cases.
PR gold/26200
* plugin.cc (Plugin_manager::claim_file): Don't share Plugin_list
iterator.
(Plugin_manager::all_symbols_read): Likewise.
(Plugin_manager::cleanup): Likewise.
Since upgrading to binutils 2.35 I've been experiencing random memory
corruption related crashes with ld.gold --threads. It's caused by
multiple threads concurrently pushing elements onto the shared
std::vector in File_read::record_file_read(). This vector is supposed to
be protected by file_counts_lock, but that is initialized lazily and
might be NULL when File_read::open() is called, in which case
Hold_optional_lock silently skips locking it.
Fix by calling the initialize() method before attempting to acquire the
lock, the same as other places that use file_counts_lock.
PR 26827
* fileread.cc (File_read::open): Ensure file_counts_lock is
initialized.
* testsuite/Makefile.am (check_PROGRAMS): Add a test that passes
-Wl,--threads.
* testsuite/Makefile.in: Regenerate.
Skip bootstrap-test and bootstrap-test-r for LTO build due to different
build IDs.
* Makefile.am (bootstrap-test): Skip for LTO build.
(bootstrap-test-r): Likewise.
Skip incremental_test_2, incremental_test_3, incremental_test_4,
incremental_test_5, incremental_copy_test, incremental_common_test_1
and incremental_comdat_test_1 when -fcf-protection is used to compile
gold since gold doesn't properly support -fcf-protection on Intel CET
enabled OS.
Also skip incremental_copy_test and incremental_comdat_test_1 for GCC 9
or later since they failed with GCC 9 or later.
PR gold/23539
* configure.ac: Check for GCC 9 or later and for -fcf-protection.
* configure: Regenerated.
* testsuite/Makefile.am (check_PROGRAMS): Skip incremental_test_2,
incremental_test_3, incremental_test_4, incremental_test_5,
incremental_copy_test, incremental_common_test_1 and
incremental_comdat_test_1 for -fcf-protection. Also Skip
incremental_copy_test and incremental_comdat_test_1 for GCC 9 or
later.
* testsuite/Makefile.in: Regenerated.
This patch updates GNU_PROPERTY_X86_XXX macros for gold:
1. GNU_PROPERTY_X86_UINT32_AND_XXX: A 4-byte unsigned integer property.
A bit is set if it is set in all relocatable inputs:
#define GNU_PROPERTY_X86_UINT32_AND_LO 0xc0000002
#define GNU_PROPERTY_X86_UINT32_AND_HI 0xc0007fff
2. GNU_PROPERTY_X86_UINT32_OR_XXX: A 4-byte unsigned integer property.
A bit is set if it is set in any relocatable inputs:
#define GNU_PROPERTY_X86_UINT32_OR_LO 0xc0008000
#define GNU_PROPERTY_X86_UINT32_OR_HI 0xc000ffff
3. GNU_PROPERTY_X86_UINT32_OR_AND_XXX: A 4-byte unsigned integer property.
A bit is set if it is set in any relocatable inputs and the property is
present in all relocatable inputs:
#define GNU_PROPERTY_X86_UINT32_OR_AND_LO 0xc0010000
#define GNU_PROPERTY_X86_UINT32_OR_AND_HI 0xc0017fff
4. GNU_PROPERTY_X86_FEATURE_2_NEEDED, GNU_PROPERTY_X86_FEATURE_2_USED
and GNU_PROPERTY_X86_FEATURE_2_XXX bits.
GNU_PROPERTY_X86_FEATURE_1_AND is unchanged. GNU_PROPERTY_X86_ISA_1_USED
and GNU_PROPERTY_X86_ISA_1_NEEDED are updated to better support targeted
processors since GNU_PROPERTY_X86_ISA_1_?86 aren't isn't very useful.
A new set of GNU_PROPERTY_X86_ISA_1_XXX bits are defined. The previous
GNU_PROPERTY_X86_ISA_1_XXX macros are deprecated and renamed to
GNU_PROPERTY_X86_COMPAT_ISA_1_XXX and GNU_PROPERTY_X86_COMPAT_2_ISA_1_XXX.
elfcpp/
* elfcpp.h (GNU_PROPERTY_X86_ISA_1_USED): Renamed to ...
(GNU_PROPERTY_X86_COMPAT_ISA_1_USED): This.
(GNU_PROPERTY_X86_ISA_1_NEEDED): Renamed to ...
(GNU_PROPERTY_X86_COMPAT_ISA_1_NEEDED): This.
(GNU_PROPERTY_X86_UINT32_AND_LO): New.
(GNU_PROPERTY_X86_UINT32_AND_HI): Likewise.
(GNU_PROPERTY_X86_UINT32_OR_LO): Likewise.
(GNU_PROPERTY_X86_UINT32_OR_HI): Likewise.
(GNU_PROPERTY_X86_UINT32_OR_AND_LO): Likewise.
(GNU_PROPERTY_X86_UINT32_OR_AND_HI): Likewise.
(GNU_PROPERTY_X86_COMPAT_2_ISA_1_NEEDED): New.
(GNU_PROPERTY_X86_COMPAT_2_ISA_1_NEEDED): Likewise.
(GNU_PROPERTY_X86_FEATURE_1_AND): Updated to
(GNU_PROPERTY_X86_UINT32_AND_LO + 0).
(GNU_PROPERTY_X86_ISA_1_NEEDED): New. Defined to
GNU_PROPERTY_X86_UINT32_OR_LO + 2.
(GNU_PROPERTY_X86_FEATURE_2_NEEDED): New. Defined to
(GNU_PROPERTY_X86_UINT32_OR_LO + 1).
(GNU_PROPERTY_X86_ISA_1_USED): New. Defined to
GNU_PROPERTY_X86_UINT32_OR_AND_LO + 2.
(GNU_PROPERTY_X86_FEATURE_2_USED): New. Defined to
(GNU_PROPERTY_X86_UINT32_OR_AND_LO + 1).
gold/
* x86_64.cc (Target_x86_64::Target_x86_64): Initialize
feature_2_used_, feature_2_needed_ and object_feature_2_used_.
(Target_x86_64::feature_2_used_): New data member.
(Target_x86_64::feature_2_needed_): Likewise.
(Target_x86_64::object_isa_1_used_): Likewise.
(Target_x86_64::record_gnu_property): Support
GNU_PROPERTY_X86_COMPAT_ISA_1_USED,
GNU_PROPERTY_X86_COMPAT_ISA_1_NEEDED,
GNU_PROPERTY_X86_COMPAT_2_ISA_1_USED,
GNU_PROPERTY_X86_COMPAT_2_ISA_1_NEEDED,
GNU_PROPERTY_X86_FEATURE_2_USED and
GNU_PROPERTY_X86_FEATURE_2_NEEDED.
(Target_x86_64::merge_gnu_properties): Merge FEATURE_2_USED bits.
Initialize object_feature_2_used_.
(Target_x86_64::do_finalize_gnu_properties): Support
GNU_PROPERTY_X86_FEATURE_2_USED and
GNU_PROPERTY_X86_FEATURE_2_NEEDED.
* testsuite/gnu_property_a.S (GNU_PROPERTY_X86_ISA_1_USED): Set
to 0xc0010002.
(GNU_PROPERTY_X86_ISA_1_NEEDED): Set to 0xc0008002.
* testsuite/gnu_property_b.S (GNU_PROPERTY_X86_ISA_1_USED): Set
to 0xc0010002.
(GNU_PROPERTY_X86_ISA_1_NEEDED): Set to 0xc0008002.
* testsuite/gnu_property_c.S (GNU_PROPERTY_X86_ISA_1_USED): Set
to 0xc0010002.
(GNU_PROPERTY_X86_ISA_1_NEEDED): Set to 0xc0008002.
* testsuite/gnu_property_test.sh: Updated.
The NT_GNU_PROPERTY_TYPE_0 note should be aligned to 8 bytes for 64-bit
ELF as specified by gABI. A note section can be only placed in a PT_NOTE
segment with the same alignment.
PR gold/22914
PR gold/23535
* layout.cc (Layout::attach_allocated_section_to_segment): Place
a note section in a PT_NOTE segment with the same alignment. Set
the alignment of the PT_NOTE segment from the alignment of the
note section.
(Layout::create_note): Align the NT_GNU_PROPERTY_TYPE_0 note to 8
bytes for 64-bit ELF.
(Layout::segment_precedes): Place segments with larger alignments
first.
* output.cc (Output_segment::Output_segment): Initialize align_.
* output.h (Output_segment): Add align, set_align and align_.
* testsuite/Makefile.am (gnu_property_test.stdout): Pass -lhSWn
to $(TEST_READELF).
(gnu_property_test): Pass --build-id to ld.
* testsuite/Makefile.in: Regenerated.
* testsuite/gnu_property_test.sh (check_alignment): New.
Use check_alignment to check the NT_GNU_PROPERTY_TYPE_0 note
alignment. Verify that there are 2 PT_NOTE segments.
With
commit 4aebb6312e
Author: Rahul Chaudhry <rahulchaudhry@google.com>
Date: Wed Feb 15 00:37:10 2017 -0800
Improved support for --icf=safe when used with -pie.
we now check opcode with R_X86_64_PC32 relocation, which tell branches
from other instructions. We can enable safe ICF for shared object on
x86-64. Also, global symbols with non-default visibility should be
folded like local symbols.
PR gold/21452
* x86_64.cc (Scan::local_reloc_may_be_function_pointer): Remove
check for shared library.
(Scan::global_reloc_may_be_function_pointer): Remove check for
shared library and symbol visibility.
* testsuite/icf_safe_so_test.cc (bar_static): New function.
(main): Take function address of bar_static and use it.
* testsuite/icf_safe_so_test.sh (arch_specific_safe_fold): Also
check fold on x86-64. Check bar_static isn't folded.
Calling non-pcrel functions from pcrel code requires a stub to set up
r2. Gold created the stub, but an "optimisation" made the stub jump
to the function local entry, ie. r2 was not initialised.
This patch fixes that long branch stub problem, and another that might
occur for plt call stubs to local functions.
bfd/
* elf64-ppc.c (write_plt_relocs_for_local_syms): Don't do local
entry offset optimisation.
gold/
* powerpc.cc (Powerpc_relobj::do_relocate_sections): Don't do
local entry offset optimisation for lplt_section.
(Target_powerpc::Branch_info::make_stub): Don't add local
entry offset to long branch dest passed to
add_long_branch_entry. Do pass st_other bits.
(Stub_table::Branch_stub_ent): Add "other_" field.
(Stub_table::add_long_branch_entry): Add "other" param, and
save.
(Stub_table::branch_stub_size): Adjust long branch offset.
(Stub_table::do_write): Likewise.
(Target_powerpc::Relocate::relocate): Likewise.
GOT relocations can refer directly to a function in a fixed position
executable, unlike ADDR64 which needs a global entry stub, or branch
relocs, which need PLT stubs.
* powerpc.cc (is_got_reloc): New function.
(Target_powerpc::Relocate::relocate): Use it here, exclude GOT
relocs when looking for stubs.
For functions with small (< 256 bytes) stack frames, the current x86
do_calls_non_split ignores --split-stack-adjust-size and, in
combination with __morestack_non_split, supplies a non-split-stack
function with at least 0x100000 (1M) available stack. On powerpc64, a
default of 0x4000 is not large enough to reliably work with the golang
testsuite. This increase the default size to the defacto x86 value.
* options.h (split_stack_adjust_size): Default to 0x100000.
Gold version of commit 3cd7c7d7ef.
* powerpc.cc (Target_powerpc): Rename power10_stubs_ to
power10_relocs_.
(Target_powerpc::set_power10_relocs): New accessor.
(Target_powerpc::set_power10_stubs): Delete.
(Target_powerpc::power10_stubs): Adjust.
(Target_powerpc::has_localentry0): New accessor.
(ld_0_11): New constant.
(glink_eh_frame_fde_64v1, glink_eh_frame_fde_64v2): Adjust.
(glink_eh_frame_fde_64v2_localentry0): New.
(Output_data_glink::pltresolve_size): Update.
(Output_data_glink::add_eh_frame): Use localentry0 version eh_frame.
(Output_data_glink::do_write): Move r2 save to start of ELFv2 stub
and only emit for has_localentry0. Don't use r2 in the stub.
(Target_powerpc::Scan::local, global): Adjust for
set_power10_relocs renaming.
(Target_powerpc::scan_relocs): Warn and reset plt_localentry0_.
This stub doesn't have the r2 store at the beginning.
* powerpc.cc (Target_powerpc::Relocate::relocate): Don't skip
first insn of __tls_get_addr_opt stub.
