This tightens the condition under which ld optimizes PIC entry code
to non-PIC.
bfd/
* elf64-ppc.c (ppc64_elf_relocate_section): Further restrict
ELFv2 entry optimization.
gold/
* powerpc.cc (relocate): Further restrict ELFv2 entry optimization.
This fixes some cases where the linker would incorrectly error on plt
relocs to local ifunc symbols. I've also tidied plt and ifunc
handling for ppc64, where check_relocs was allowing for the
possibility of plt calls via addr14/addr24 relocs but relocate_section
was not.
* elf32-ppc.c (ppc_elf_check_relocs): Don't error on local ifunc
plt call. Wrap long lines.
(ppc_elf_relocate_section): Wrap long lines.
* elf64-ppc.c (ppc64_elf_check_relocs): Don't error on local ifunc
plt calls. Move __tls_get_addr checks later. Don't create plt
for addr14/addr24 relocs.
(ppc64_elf_gc_sweep_hook): Adjust to suit check_relocs changes.
(ppc64_elf_relocate_section): Correct local ifunc handling for
PLT64, PLT32 and PLT16 relocs.
Add a new relocation that marks large-model entry code, for edit back
to medium-model.
include/elf/
* ppc64.h (R_PPC64_ENTRY): Define.
bfd/
* reloc.c (BFD_RELOC_PPC64_ENTRY): New.
* elf64-ppc.c (reloc_howto_type ppc64_elf_howto_raw): Add
entry for R_PPC64_ENTRY.
(LD_R2_0R12, ADD_R2_R2_R12, LIS_R2, ADDIS_R2_R12): Define.
(ppc64_elf_reloc_type_lookup): Handle R_PPC64_ENTRY.
(ppc64_elf_relocate_section): Edit code at R_PPC64_ENTTY. Use
new insn defines.
* libbfd.h: Regenerate.
* bfd-in2.h: Regenerate.
Fixes a segfault in ppc64_elf_tls_optimize found when testing
R_PPC64_ENTRY, and potential for trouble in other places found by
code inspection.
* elf64-ppc.c (ppc64_elf_tls_optimize): Don't segfault on NULL
symbol section or output section.
(ppc64_elf_edit_toc): Similarly for ld -R objects.
(ppc64_elf_size_stubs): Likewise.
The idea here is that instead of using memmove to shuffle the relocs
array every time one is deleted, to add a "wrel" pointer and copy from
rel[0] to wrel[0] as we go.
* elf64-ppc.c (ppc64_elf_relocate_section): Use read and write
pointers to reloc array, rather than memmove when deleting a
reloc. Don't use RELOC_AGAINST_DISCARDED_SECTION. Adjust
reloc counts at end of loop.
* elf32-ppc.c (ppc_elf_relocate_section): Likewise.
This patch fixes all occurences of left-shifting negative constants in C code
which is undefined by the C standard.
bfd/ChangeLog:
* elf64-ppc.c (ppc64_elf_size_stubs, ppc64_elf_build_stubs): Fix left
shift of negative value.
* libbfd.c (safe_read_leb128): Likewise.
* dwarf2.c (place_sections): Likewise.
* bfd-in.h (align_power): Likewise.
* bfd-in2.h (align_power): Likewise.
Sometimes it may be of benefit to force use of the __tls_get_addr_opt
call stub even when the glibc being used during linking does not
advertise __tls_get_addr_opt.
bfd/
* elf64-ppc.h (struct ppc64_elf_params <tls_get_addr_opt>): Rename
from no_tls_get_addr_opt.
* elf64-ppc.c: Update for rename and inversion of tls_get_addr_opt.
(ppc64_elf_tls_setup): Set tls_get_addr_opt to 0 only when at
default of -1.
ld/
* emultempl/ppc64elf.em (params): Init tls_get_addr_opt field to -1.
(OPTION_TLS_GET_ADDR_OPT): Define.
(PARSE_AND_LIST_LONGOPTS): Handle --tls-get-addr-opt.
(PARSE_AND_LIST_OPTIONS, PARSE_AND_LIST_ARGS_CASES): Likewise.
* ld.texinfo: Document --tls-get-addr-optimize and
--no-tls-get-addr-optimize.
OPD lookup goes awry when .opd has been shuffled to remove unused
functions.
* elf64-ppc.c (ppc64_elf_maybe_function_sym): Adjust symbol value
if .opd section has been edited.
Don't emit long branch or plt branch stubs to save/restore functions.
Copy them instead. The problem is that plt branch stubs currently
trash r12, one of the parameters to some of the save/restore
functions, and there is no free register available to use instead of
r12.
6f20ed8a is prerequisite for this patch.
PR 18878
* elf64-ppc.c (ARRAY_SIZE): Define. Use throughout.
(enum ppc_stub_type): Add ppc_stub_save_res.
(struct map_stub): Add "next" and "needs_save_res".
(struct ppc_link_hash_entry): Add "save_res" flag.
(struct ppc_link_hash_table): Add "group".
(sfpr_define): Add stub_sec param. Define symbol in stub_sec if
stub_sec is non-null. Set "save_res".
(save_res_funcs): Make file scope, rename from funcs. Adjust uses.
(ppc64_elf_adjust_dynamic_symbol): Prohibit plt call to save_res syms.
(ppc_build_one_stub): Handle ppc_stub_save_res.
(ppc_size_one_stub): Set stub type to ppc_size_one_stub on finding
stub for linker defined save_res sym.
(group_sections): Init new fields of struct map_stub.
(ppc64_elf_size_stubs): Reserve space for save/restore func copy.
(ppc64_elf_build_stubs): Copy save/restore funcs to groups. Emit
alias syms too.
(ppc64_elf_relocate_section): Set destination for ppc_stub_save_res.
Sometimes these stubs don't need to change the low 16-bits of r2, so in
that case omit a useless addi r2,r2,0 insn. Also, change the get_r2off
error return from 0 to -1 since 0 is a valid return for ELFv2 -R objects.
* elf64-ppc.c (get_r2off): Return -1 on error.
(ppc_build_one_stub): Adjust for get_r2off change. Don't emit
addi r2,r2,0 on r2off stubs when the low 16-bit delta is zero.
(ppc_size_one_stub): Corresponding size changes for r2off stubs.
Add condition in test for -R objects.
Two organizational changes to the array of additional info kept for
sections.
1) Move group info into a per-group allocated struct, in preparation
for future changes that need per-group accounting.
2) Expand the array to include output sections, which simplifies
sizing and removes the need for a separate output section array.
* section.c (section_id): Make file scope.
(bfd_get_next_section_id): New function.
* elf64-ppc.c (struct map_stub): Remove toc_off field. Move decl.
(struct ppc_stub_hash_entry): Delete stub_sec and id_sec. Add
group. Update all uses.
(struct ppc_link_hash_table): Delete top_id, top_index, and
input_list. Add sec_info_arr_size. Rename stub_group to
sec_info, and make group info indirect. Update stub_group refs
throughout file.
(ppc_add_stub): Don't look for stub_sec on link_sec stub_group
entry.
(ppc_build_one_stub): Delete FIXME.
(ppc64_elf_setup_section_lists): Size htab->sec_info for all
sections, not just input sections. Don't create htab->input_list.
(ppc64_elf_next_input_section): Update to use sec_info union as
list pointer.
(PREV_SEC): Delete.
(group_sections): Pass "info" param rather than "htab". Iterate
over output sections rather than input_list. Use sec_info union
as list pointers. Alloc atruct map_stub, and return fail status.
* bfd-in2.h: Regenerate.
These int vars are really unsigned, so make them so.
* section.c (struct bfd_section): Make "id" and "index" unsigned.
* coff-rs6000.c (_bfd_xcoff_sizeof_headers): Adjust local var to suit.
* elf.c (elf_map_symbols): Likewise.
* elf64-ppc.c (sym_exists_at): Make "id" param unsigned.
(struct ppc_link_hash_table): Make "top_id" and "top_index" unsigned.
(ppc64_elf_setup_section_lists): Ditto for local vars.
* elf32-arm.c: Similarly to elf64-ppc.c.
* elf32-avr.c: Likewise.
* elf32-hppa.c: Likewise.
* elf32-m68hc1x.c: Likewise.
* elf32-metag.c: Likewise.
* elf32-nios2.c: Likewise.
* elfnn-aarch64.c: Likewise.
* simple.c (struct saved_offsets): Make "section_count" unsigned.
* bfd-in2.h: Regenerate.
bfd/
* elf64-ppc.c (ppc64_elf_func_desc_adjust): Don't redefine .TOC.
if already defined, and set linker_def.
(ppc64_elf_set_toc): Use .TOC. value if defined other than by
the backend.
ld/
* ldexp.c (exp_fold_tree_1): Clear linker_def on symbol assignment.
Since the backend elf_add_symbol_hook isn't called on local symbols,
the EI_OSABI field isn't to ELFOSABI_GNU where are local IFUNC symbols.
This patch changes the x86 backends to set has_gnu_symbols if there are
relocations against IFUNC symbols. Other backends with IFUNC support
may need a similar change.
This patch also changes the type of has_gnu_symbols from bfd_boolean to
enum elf_gnu_symbols.
bfd/
PR ld/18815
* elf-bfd.h (elf_gnu_symbols): New enum.
(elf_obj_tdata): Use elf_gnu_symbols on has_gnu_symbols.
* elf-s390-common.c (elf_s390_add_symbol_hook): Set
has_gnu_symbols to elf_gnu_symbol_any.
* elf32-arm.c (elf32_arm_add_symbol_hook): Likewise.
* elf32-m68k.c (elf_m68k_add_symbol_hook): Likewise.
* elf32-ppc.c (ppc_elf_add_symbol_hook): Likewise.
* elf32-sparc.c (elf32_sparc_add_symbol_hook): Likewise.
* elf64-ppc.c (ppc64_elf_add_symbol_hook): Likewise.
* elf64-sparc.c (elf64_sparc_add_symbol_hook): Likewise.
* lfxx-aarch64.c (_bfd_aarch64_elf_add_symbol_hook): Likewise.
* elf32-i386.c (elf_i386_check_relocs): Update has_gnu_symbols
if there are relocations against IFUNC symbols.
(elf_i386_add_symbol_hook): Don't check STT_GNU_IFUNC here.
* elf64-x86-64. (elf_x86_64_check_relocs): Update has_gnu_symbols
if there are relocations against IFUNC symbols.
(elf_x86_64_add_symbol_hook): Don't check STT_GNU_IFUNC here.
ld/testsuite/
PR ld/18815
* ld-i386/i386.exp: Run pr18815.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr18815.d: New file.
* ld-i386/pr18815.s: Likewise.
* ld-x86-64/pr18815.d: Likewise.
* ld-x86-64/pr18815.s: Likewise.
Bogus assembly can hit an assertion in opd_entry_value when the symbol
referenced by a function descriptor is undefined. Worse, the code
after the assert copies unitialised memory to return the code section.
This uninitialised pointer can later be dereferencd, possibly causing
a linker segmentation fault.
* elf64-ppc.c (opd_entry_value): Remove assertion. Instead,
return -1 if symbol referenced is not defined. Tidy.
bfd/
* elf64-ppc.c (toc_adjusting_stub_needed): Use the symbol value
plus addend rather than the original st_value when looking up
entries in opd->adjust.
ld/testsuite/
* ld-powerpc/tocopt6-inc.s, ld-powerpc/tocopt6a.s,
ld-powerpc/tocopt6b.s, ld-powerpc/tocopt6c.s,
ld-powerpc/tocopt6.d: New test.
* ld-powerpc/powerpc.exp (ppc64elftests): Add it.
* elf64-ppc.c (ppc64_elf_relocate_section): Report overflow to
stubs, even those for undefined weak symbols. Otherwise, don't
report relocation overflow on branches to undefined strong
symbols. Fix memory leak.
* elf32-ppc.c (ppc_elf_relocate_section): Don't report relocation
overflow on branches to undefined strong symbols.
I was looking at a current glibc using objdump today and saw an odd
plt call stub.
0000000000044d80 <00000033.plt_call.__strchrnul>:
44d80: f8 41 00 28 std r2,40(r1)
44d84: e9 82 8c f8 ld r12,-29448(r2)
44d88: 7d 89 03 a6 mtctr r12
44d8c: e8 42 8d 00 ld r2,-29440(r2)
44d90: 28 22 00 00 cmpldi r2,0
44d94: 4c e2 04 20 bnectr+
44d98: 48 13 84 f0 b 17d288 <realloc@plt>
What? It doesn't branch to __strchrnul@plt on finding a zero r2?
Turns out this isn't a real problem since the stub is for loading an
ifunc, so will not be lazily resolved and thus r2 will never be zero.
Of course, that means the thread-safety check is unnecessary.
I also tweak the special __tls_get_addr_opt call stub here, to
restore r2 immediately after the call. Not doing that might affect
eh_frame unwinding.
* elf64-ppc.c (plt_stub_size, build_plt_stub): Don't build
thread-safe stubs for iplt.
(build_tls_get_addr_stub): Restore r2 immediately after call.
Rather than reporting a link error on attempting to use dynbss for
protected vars, use dynamic text relocs.
* elf32-ppc.c (ppc_elf_adjust_dynamic_symbol): Heed -z nocopyreloc.
Use text relocs rather than giving an error on trying to use
.dynbss for protected shared lib vars.
* elf64-ppc.c (ppc64_elf_adjust_dynamic_symbol): Likewise.
I was having a little closer look at what is going on here and noticed
that HJ unconditionally emits a NULL STT_FILE symbol before emitting
forced local symbols. That means we really don't need a second pass
over forced local symbols. The only reason for two passes is when
some forced local symbol can be emitted before the NULL STT_FILE. So
I set about removing the second pass, updating the testsuite all over
again. It's also unnecessary to emit the NULL STT_FILE when no
previous file symbol has been emitted.
bfd/
PR ld/17975
* elflink.c (struct elf_outext_info): Remove need_second_pass
and second_pass.
(elf_link_output_extsym): Delete code handling second forced
local pass. Move code emitting NULL STT_FILE symbol later, so
that it can be omitted if forced local is stripped. Don't
emit the NULL STT_FILE if no file symbols have been output.
(bfd_elf_final_link): Remove second forced local pass.
* elf32-ppc.c (add_stub_sym): Set linker_def on linker syms.
(ppc_elf_size_dynamic_sections): Likewise.
* elf64-ppc.c (ppc_build_one_stub): Likewise.
(build_global_entry_stubs): Likewise.
(ppc64_elf_build_stubs): Likewise.
ld/testsuite/
PR ld/17975
* ld-aarch64/gc-tls-relocs.d, * ld-alpha/tlspic.rd,
* ld-cris/libdso-2.d, * ld-i386/tlsdesc-nacl.rd, * ld-i386/tlsdesc.rd,
* ld-i386/tlsnopic-nacl.rd, * ld-i386/tlsnopic.rd,
* ld-i386/tlspic-nacl.rd, * ld-i386/tlspic.rd, * ld-ia64/tlspic.rd,
* ld-powerpc/tlsexe.r, * ld-powerpc/tlsexetoc.r,
* ld-powerpc/tlsso.r, * ld-powerpc/tlstocso.r,
* ld-s390/tlspic.rd, * ld-s390/tlspic_64.rd,
* ld-sparc/tlssunnopic32.rd, * ld-sparc/tlssunnopic64.rd,
* ld-sparc/tlssunpic32.rd, * ld-sparc/tlssunpic64.rd,
* ld-tic6x/shlib-1.rd, * ld-tic6x/shlib-1b.rd, * ld-tic6x/shlib-1r.rd,
* ld-tic6x/shlib-1rb.rd, * ld-tic6x/shlib-noindex.rd,
* ld-x86-64/tlsdesc-nacl.rd, * ld-x86-64/tlsdesc.rd,
* ld-x86-64/tlspic-nacl.rd, * ld-x86-64/tlspic.rd: Update.
PR binutils/17512
* dwarf.c (read_1_byte, read_1_signed_byte, read_2_bytes)
(read_4_bytes, read_8_bytes, read_n_bytes, read_string)
(read_indirect_string, read_alt_indirect_string)
(read_alt_indirect_ref, read_address, read_abbrevs)
(read_attribute_value, read_attribute, decode_line_info)
(find_abstract_instance_name, read_rangelist)
(scan_unit_for_symbols, parse_comp_unit)
(_bfd_dwarf2_find_nearest_line): Harden DWARF reading code. Pass
end pointers to reading functions and check for offsets taking
pointers out of range. Replace calls to read_*_leb128 with calls
to safe_read_leb128.
(* elf64-ppc.c (opd_entry_value): Add a check for an overlarge
offset.
* syms.c (_bfd_stab_section_find_nearest_line): Add checks for
computed file_name address being before the start of the string
table.
Real code won't hit these, but it's possible to contrive a testcase..
* elf32-ppc.c (ppc_elf_relocate_section): Don't segfault on NULL
tls_sec.
* elf64-ppc.c (ppc64_elf_relocate_section): Likewise.
* elflink.c (elf_link_output_extsym): Don't assert on NULL tls_sec.
The linker hardcoded r3 into a local-dynamic to local-exec TLS
optimization sequence. This is normally the case since r3 is required
as a parameter to (the optimized out) __tls_get_addr call. However,
it is possible for a compiler, LLVM in this case, to set up the
parameter value in another register then copy it to r3 before the
call.
When fixing this problem, I noticed that ppc32 had another bug when
optimizing away one of the TLS insns to a nop.
The patch also tidies a mask used by global-dynamic to initial-exec
TLS optimization, to just select the fields needed. Leaving the
offset in the instruction wasn't a bug since it will be overwritten
anyway.
bfd/
* elf64-ppc.c (ppc64_elf_relocate_section): Correct GOT_TLSLD
optimization. Tidy mask for GOT_TLSGD optimization.
* elf32-ppc.c (ppc_elf_relocate_section): Likewise. Correct
location of nop zapping high insn too.
ld/testsuite/
* ld-powerpc/tlsld.d, * ld-powerpc/tlsld.s: New test.
* ld-powerpc/tlsld32.d, * ld-powerpc/tlsld32.s: New test.
* ld-powerpc/powerpc.exp: Run them. Move tocvar and tocnovar.
The changes to reorder sections for better relro protection on powerpc64,
3e2b0f31, 23283c1b, and 5ad18f16, run into a problem with xlc.
xlc -qdatalocal puts global variables into .toc, which means that .toc
must be writable. The simplest way to accomplish this is to edit the
linker script to remove .toc sections from .got on detecting xlc object
files.
bfd/
* elf64-ppc.h (struct ppc64_elf_params): Add "object_in_toc".
* elf64-ppc.c (ppc64_elf_add_symbol_hook): Assume that global symbols
in .toc indicate xlc compiled code that might require a rw .toc.
ld/
* emulparams/elf64ppc.sh (INITIAL_READWRITE_SECTIONS): Define.
* emultempl/ppc64elf.em (params): Init new field.
(ppc_after_open): New function.
(LDEMUL_AFTER_OPEN): Define.
* ldlang.c (lang_final): Whitespace fix.
ld/testsuite/
* ld-powerpc/tocvar.d, * ld-powerpc/tocvar.s: New test.
* ld-powerpc/tocnovar.d, * ld-powerpc/tocnovar.s: New test.
* ld-powerpc/powerpc.exp: Run tocvar and tocnovar.
PR binutils/17512
* addr2line.c (slurp_symtab): If the symcount is zero, free the
symbol table pointer.
* dwarf2.c (concat_filename): Check for an empty directory table.
(scan_unit_for_symbols): Check for reading off the end of the
unit.
(parse_comp_unit): Check for a DW_AT_comp_dir attribute with a
non-string form.
* elf64-ppc.c (opd_entry_value): Fail if there are no relocs
available.
Copy relocs are used in a scheme to avoid dynamic text relocations in
non-PIC executables that refer to variables defined in shared
libraries. The idea is to have the linker define any such variable in
the executable, with a copy reloc copying the initial value, then have
both the executable and shared library refer to the executable copy.
If the shared library defines the variable as protected then we have
two copies of the variable being used.
PR 15228
* elflink.c (_bfd_elf_adjust_dynamic_copy): Add "info" param.
Error on copy relocs against protected symbols.
(elf_merge_st_other): Set h->protected_def.
* elf-bfd.h (struct elf_link_hash_entry): Add "protected_def".
(_bfd_elf_adjust_dynamic_copy): Update prototype.
* elf-m10300.c (_bfd_mn10300_elf_adjust_dynamic_symbol): Update
_bfd_elf_adjust_dynamic_copy call.
* elf32-arm.c (elf32_arm_adjust_dynamic_symbol): Likewise.
* elf32-cr16.c (_bfd_cr16_elf_adjust_dynamic_symbol): Likewise.
* elf32-cris.c (elf_cris_adjust_dynamic_symbol): Likewise.
* elf32-hppa.c (elf32_hppa_adjust_dynamic_symbol): Likewise.
* elf32-i370.c (i370_elf_adjust_dynamic_symbol): Likewise.
* elf32-i386.c (elf_i386_adjust_dynamic_symbol): Likewise.
* elf32-lm32.c (lm32_elf_adjust_dynamic_symbol): Likewise.
* elf32-m32r.c (m32r_elf_adjust_dynamic_symbol): Likewise.
* elf32-m68k.c (elf_m68k_adjust_dynamic_symbol): Likewise.
* elf32-metag.c (elf_metag_adjust_dynamic_symbol): Likewise.
* elf32-or1k.c (or1k_elf_adjust_dynamic_symbol): Likewise.
* elf32-ppc.c (ppc_elf_adjust_dynamic_symbol): Likewise.
* elf32-s390.c (elf_s390_adjust_dynamic_symbol): Likewise.
* elf32-sh.c (sh_elf_adjust_dynamic_symbol): Likewise.
* elf32-tic6x.c (elf32_tic6x_adjust_dynamic_symbol): Likewise.
* elf32-tilepro.c (tilepro_elf_adjust_dynamic_symbol): Likewise.
* elf32-vax.c (elf_vax_adjust_dynamic_symbol): Likewise.
* elf64-ppc.c (ppc64_elf_adjust_dynamic_symbol): Likewise.
* elf64-s390.c (elf_s390_adjust_dynamic_symbol): Likewise.
* elf64-sh64.c (sh64_elf64_adjust_dynamic_symbol): Likewise.
* elf64-x86-64.c (elf_x86_64_adjust_dynamic_symbol): Likewise.
* elfnn-aarch64.c (elfNN_aarch64_adjust_dynamic_symbol): Likewise.
* elfxx-mips.c (_bfd_mips_elf_adjust_dynamic_symbol): Likewise.
* elfxx-sparc.c (_bfd_sparc_elf_adjust_dynamic_symbol): Likewise.
* elfxx-tilegx.c (tilegx_elf_adjust_dynamic_symbol): Likewise.
Now that ld -r relocs are sorted by the pr17666 fix, there isn't so
much need to sort incoming relocs.
* elf64-ppc.c (sort_r_offset): Delete.
(ppc64_elf_edit_opd): Don't sort input relocs.
This patch sorts .opd relocs (see pr17666) and allows .opd sections
with a mix of 16 and 24 byte entries to be edited.
* elf64-ppc.c (OPD_NDX): Define. Use throughout for sizing/indexing
_opd_sec_data array, halving required memory.
(sort_r_offset): New function.
(ppc64_elf_edit_opd): Sort incoming relocs. Accept .opd
sections with a mix of 16 and 24 byte OPD entries. Don't
attempt to honour --non-overlapping-opd for .opd sections with
unexpected relocs. Simplify opd entry size calculations by
first finding the reloc for the next entry. Make edit loop
handle one opd entry per iteration, with an inner loop
handling relocs per entry.
Alignment of plt stubs was broken, firstly because the option was
being dropped due to the alignment value not being set in the "params"
struct used in elf64-ppc.c, and secondly due to not calculating the
number of alignment boundary crossings correctly.
bfd/
* elf64-ppc.c (plt_stub_pad): Correct.
ld/
* ld.texinfo: Correct --plt-align documentation.
* emultempl/ppc64elf.em (plt_stub_align): Delete. Use and set
params.plt_stub_align instead.
Corrects overflow test for rel14, addr14, rel24, addr24 branch relocs,
and prints an information message to give a hint as to how a branch
that can't reach a stub might be cured.
bfd/
* elf64-ppc.c (group_sections): Init stub14_group_size from
--stub-group-size parameter divided by 1024.
gold/
* powerpc.cc (Stub_control::Stub_control): Init stub14_group_size_
from --stub-group-size parameter divided by 1024.
(Powerpc_relocate_functions::rela, rela_ua): Add fieldsize
template parameter. Update all uses.
(Target_powerpc::Relocate::relocate): Rename has_plt_value to
has_stub_value. Set for long branches. Don't report overflow for
branch to undefined weak symbols. Print info message on
overflowing branch to stub.
Go is multi-threaded, so use thread-safe plt stubs. __go_go doesn't
actually start threads, but is a convenient libgo function with a
reasonably unique name that is called by all Go executables. (Threads
are started by a static function of libgo, runtime_newosproc, called
by a whole lot of global functions, including runtime_starttheworld
and runtime_main.)
bfd/
* elf64-ppc.c (ppc64_elf_size_stubs): Add __go_go to thread_starters.
gold/
* powerpc.cc (Target_powerpc::do_relax): Add __go_go to thread_starters.
The code in ppc64_elf_tls_optimize looking at the .toc is only
interested in .toc entries that are addresses. .toc can contain more
than just an array of addresses, so if we have items that aren't
8-byte aligned, ignore them.
* elf64-ppc.c (ppc64_elf_tls_optimize): Ignore relocs against toc
entries that aren't a multiple of 8 rather than failing assertion.
powerpc64 ld builds plt call stubs with a read barrier to provide
thread safety on lazy plt updates, necessary on multi-threaded apps
with power7 or later weakly ordered memory. gcc-4.9 libgomp
introduced more functions that could call pthread_create, which means
we have more functions that if referenced in an executable should
cause a default of --plt-thread-safe.
* elf64-ppc.c (ppc64_elf_size_stubs): Add gcc-4.9 libgomp functions
to thread_starter.
When functions are emitted in comdat groups, global symbols defined in
duplicates of the group are treated as if they were undefined. That
prevents the symbols in the discarded sections from affecting the
linker's global symbol hash table or causing duplicate symbol errors.
Annoyingly, when gcc emits a function to a comdat group, it does not
put *all* of a function's code and data in the comdat group.
Typically, constant tables, exception handling info, and debug info
are emitted to normal sections outside of the group, which is a
perennial source of linker problems due to the special handling needed
to deal with the extra-group pieces that ought to be discarded. In
the case of powerpc64-gcc, the OPD entry for a function is not put in
the group. Since the function symbol is defined on the OPD entry this
means we need to handle symbols in .opd specially.
To see how this affects LTO in particular, consider the linker
testcase PR ld/12942 (1). This testcase links an LTO object file
pr12942a.o with a normal (non-LTO) object pr12942b.o. Both objects
contain a definition for _Z4testv in a comdat group. On loading
pr12942a.o, the linker sees a comdat group (actually linkonce section)
for _Z4testv and a weak _Z4testv defined in the IR. On loading
pr12942b.o, the linker sees the same comdat group, and thus discards
it. However, _Z4testv is a weak symbol defined in .opd, not part of
the group, so this weak symbol overrides the weak IR symbol. On
(re)loading the LTO version of pr12942a.o, the linker sees another
weak _Z4testv, but this one does not override the value we have from
pr12942b.o. The result is a linker complaint about "`_Z4testv'
... defined in discarded section `.group' of tmpdir/pr12942b.o".
* elf64-ppc.c (ppc64_elf_add_symbol_hook): If function code
section for function symbols defined in .opd is discarded, let
the symbol appear to be undefined.
(opd_entry_value): Ensure the result section is that for the
function code section in the same object as the OPD entry.
For binary ouput, we don't have an ELF bfd output so can't access
elf_elfheader. The elf64-ppc.c changes are really just a tidy,
triggered by looking at all places where the abiversion bits are
accessed.
bfd/
* elf64-ppc.c (ppc64_elf_before_check_relocs): Do .opd processing
even when output is not ppc64 ELF. Remove redundant tests on
type of input bfd.
ld/
PR 17488
* emultempl/ppc64elf.em (gld${EMULATION_NAME}_finish): Don't attempt
to access ELF header e_flags when not ppc64 ELF output.
