Macros like READ_LEB128 in wasm-module.c that alter control flow are
evil. Maintainers will break your code if you have hidden ways to
reach labels.
* wasm-module.c (wasm_scan_name_function_section): Don't
attempt to bfd_release NULL.
The bpf reloc howtos are a bit weird, using bitpos to specify an
offset from r_offset that is outside the size of the reloc as given by
howto.size. That means bfd_get_reloc_size gives the wrong answer for
range checking, and thus bfd_reloc_offset_in_range can't be used.
* elf64-bpf.c (bpf_elf_generic_reloc): Handle bitpos offset reloc
range checking.
We can't assume .dynamic is a multiple of ElfNN_External_Dyn, at least
not when presented with fuzzed object files.
* elfnn-aarch64.c (get_plt_type): Don't access past end of
improperly sized .dynamic.
According to the privileged spec, there are five new instructions for
svinval extension. Two of them (HINVAL.VVMA and HINVAL.GVMA) need to
enable the hypervisor extension. But there is no implementation of
hypervisor extension in mainline for now, so let's consider the related
issues later.
31..25 24..20 19..15 14..12 11...7 6..2 1..0
sinval.vma 0001011 rs2 rs1 000 00000 11100 11
sfence.w.inval 0001100 00000 00000 000 00000 11100 11
sfence.inval.ir 0001100 00001 00000 000 00000 11100 11
hinval.vvma 0010011 rs2 rs1 000 00000 11100 11
hinval.gvma 0110011 rs2 rs1 000 00000 11100 11
This patch is cherry-picked from the riscv integration branch since the
svinval extension is frozen for now. Besides, we fix the funct7 encodings
of hinval.vvma and hinval.gvma, from 0x0011011 and 0x0111011 to 0x0010011
and 0x0110011.
bfd/
* elfxx-riscv.c (riscv_supported_std_s_ext): Added svinval.
(riscv_multi_subset_supports): Handle INSN_CLASS_SVINVAL.
gas/
* testsuite/gas/riscv/svinval.d: New testcase.
* testsuite/gas/riscv/svinval.s: Likewise.
include/
* opcode/riscv-opc.h: Added encodings for svinval.
* opcode/riscv.h (enum riscv_insn_class): Added INSN_CLASS_SVINVAL.
opcodes/
* riscv-opc.c (riscv_opcodes): Added svinval instructions.
The logic between these rules are extremely similar, so unify them
into a single variable by leveraging make $@ and $< variables.
Also add automake silent rule support while we're here.
Using an unsigned int cast (to 32 bits) on a pointer difference (of
possibly 64 bits) is wrong. Even though it will work on all real
object files, the fuzzers will eventually find this hole.
PR 28687
* dwarf1.c (parse_die): Cast pointer difference to size_t.
Catch another possible pointer overflow.
PR28691 is a fuzzing PR that triggers a non-problem of "output changes
per run" with PIEs and/or different compilers. I've closed similar
PRs before as wontfix, but I guess there will be no end of this type
of PR. The trigger is an attribute that usually takes one of the
offset/constant reference DW_FORMs being given an indexed string
DW_FORM. The bfd reader doesn't support indexed strings and returns
an error string instead. The address of the string varies with PIE
runs and/or compiler, and we allow that address to appear in output.
Fix this by validating integer attribute forms, as we do for string
form attributes.
PR 28691
* dwarf2.c (is_str_attr): Rename to..
(is_str_form): ..this. Change param type. Update calls.
(is_int_form): New function.
(read_attribute_value): Handle DW_FORM_addrx2.
(find_abstract_instance): Validate form when using attr.u.val.
(scan_unit_for_symbols, parse_comp_unit): Likewise.
gas/config/tc-loongarch.c: In function ‘assember_macro_helper’:
gas/config/tc-loongarch.c:915:28: error: right shift count >= width of type [-Werror=shift-count-overflow]
915 | hi32 = insn->args[1] >> 32;
| ^~
One possible fix is to make offsetT a 64-bit type for loongarch32.
This also makes bfd/targmatch.h (generated from bfd/config.bfd)
consistent since the loongarch32 match is inside #ifdef BFD64.
* config.bfd (loongarch32-*): Set want64.
This test was failing here and on another similar symbol:
[ 4](sec 1)(fl 0x00)(ty 0)(scl 143) (nx 0) 0x05d1745d11745d21 .bs
where correct output is
[ 4](sec 1)(fl 0x00)(ty 0)(scl 143) (nx 0) 0x000000000000000a .bs
The problem is caused by a 32-bit host pointer being sign-extended
when stored into a 64-bit bfd_vma, and then that value not being
trimmed back to 32 bits when used. The following belt-and-braces
patch fixes both the store and subsequent reads.
* coffcode.h (coff_slurp_symbol_table): Do not sign extend
when storing a host pointer to syment.n_value.
* coffgen.c (coff_get_symbol_info): Cast syment.n_value to a
bfd_hostptr_t before using in arithmetic.
(coff_print_symbol): Likewise.
Clang 13 and 14 produce some -Wunused-but-set-{variable,parameter} for
situations where gcc doesn't. In particular, when a variable is set and
then used in a way to update its own value. For example, if `i` is only
used in this way:
int i = 2;
i++;
i = i + 1;
gcc won't warn, but clang will.
Fix all such errors found in an --enable-targets=all build. It would be
important for somebody who knows what they're doing to just make sure
that these variables can indeed be deleted, and that there a no cases
where it's a bug, and the variable should actually be used.
The first instance of this error fix by this patch is:
CC elf32-score.lo
/home/simark/src/binutils-gdb/bfd/elf32-score.c:450:11: error: variable 'relocation' set but not used [-Werror,-Wunused-but-set-variable]
bfd_vma relocation;
^
Change-Id: I2f233ce20352645cf388aff3dfa08a651d21a6b6
This reverts a 1995 fix to handle bogus object files. Presumably such
object files have long gone.
* elf.c (bfd_section_from_shdr): Remove old hack for Oracle
libraries.
Not returning an error indication here leaves the attribute
uninitialised, which then leads to intemperate behaviour.
PR 28674
* dwarf2.c (read_attribute_value): Return NULL on trying to read
past end of attributes.
binutils-all/strip-13 and binutils-all/strip-14 tests create
SHT_REL/SHT_RELA sections by hand. These don't have sh_link set to
the .symtab section as they should, leading to readelf warnings if you
happen to be looking at the object files.
* elf.c (assign_section_numbers): Formatting. Set sh_link for
reloc sections created as normal sections in relocatable
objects.
* To be consistent with -march option, removed the "=" operator when
user want to reset the whole architecture string. So the formats are,
.option arch, +<extension><version>, ...
.option arch, -<extension>
.option arch, <ISA string>
* Don't allow to add or remove the base extensions in the .option arch
directive. Instead, users should reset the whole architecture string
while they want to change the base extension.
* The operator "+" won't update the version of extension, if the
extension is already in the subset list.
bfd/
* elfxx-riscv.c (riscv_add_subset): Don't update the version
if the extension is already in the subset list.
(riscv_update_subset): To be consistent with -march option,
removed the "=" operator when user want to reset the whole
architecture string. Besides, Don't allow to add or remove
the base extensions in the .option arch directive.
gas/
* testsuite/gas/riscv/option-arch-01.s: Updated since we cannot
add or remove the base extensions in the .option arch directive.
* testsuite/gas/riscv/option-arch-02.s: Likewise.
* testsuite/gas/riscv/option-arch-fail.l: Likewise.
* testsuite/gas/riscv/option-arch-fail.s: Likewise.
* testsuite/gas/riscv/option-arch-01a.d: Set -misa-spec=2.2.
* testsuite/gas/riscv/option-arch-01b.d: Likewise.
* testsuite/gas/riscv/option-arch-02.d: Updated since the .option
arch, + won't change the version of extension, if the extension is
already in the subset list.
* testsuite/gas/riscv/option-arch-03.s: Removed the "=" operator
when resetting the whole architecture string.
If a.so contains an SHT_RELR section, objcopy a.so will fail with:
a.so: unknown type [0x13] section `.relr.dyn'
This change allows objcopy to work.
bfd/
* elf.c (bfd_section_from_shdr): Support SHT_RELR.
The rules for rebuilding the bfd headers live in the doc/ subdir
(most likely) because they rely on the chew & related tools. But
we can collapse them into the main Makefile while keeping the tools
in the doc subdir easily enough. This makes the code simpler and
allows for rebuilding them in parallel.
Also add automake silent rule support while we're here.
These have been around for decades but don't appear to be used, and
trying to build them (e.g. `make archive.p archive.ip`) doesn't work,
so just delete it all.
Section alignment for coff-go32 is inconsistent - The '.text' and
'.data' sections are 16-byte aligned, but named sections '.text.*' and
'.data.*' are only 4-byte aligned. '.gnu.linkonce.r.*' is aligned to
16 bytes, yet '.rodata' and '.rodata.*' are aligned to 4 bytes. For
'.bss' all input sections are only aligned to 4 bytes.
This primarily can cause trouble when using SSE instructions, which
require their memory operands to be aligned to 16-byte boundaries.
This patch solves the issue simply by setting the section alignment
to 16 bytes, for all code and data sections referenced in the default
linker script.
* coff-go32.c (COFF_SECTION_ALIGNMENT_ENTRIES): Use partial
name match for .text, .data. Add entries for .const, .rodata,
.bss, .gnu.linkonce.b.
Commit b69c9d41e8 edited bfd/Makefile.in rather than using automake,
which meant a typo in Makefile.am was not discovered and other
differences in Makefile.in are seen with a proper regeneration. One
difference was lack of an empty line between the pe-aarch64igen.c rule
and the following $(BFD32_LIBS) etc. dependency rule, in the
regenerated file. Not that it matters for proper "make" behaviour,
but it's nicer with a line between those rules. Moving the rule
earlier seems to cure the missing empty line.
* Makefile.am (BFD64_BACKENDS): Correct typo.
(BFD_H_DEPS, LOCAL_H_DEPS): Move earlier. Move rule using these
deps earlier too.
* Makefile.in: Regenerate.
* po/BLD-POTFILES.in: Regenerate.
* po/SRC-POTFILES.in: Regenerate.
