We've decided to standardize on two flags for RISC-V: "-march" sets the
target architecture (which determines which instructions can be
generated), and "-mabi" sets the target ABI. We needed to rework this
because the old flag set didn't support soft-float or single-float ABIs,
and didn't support an x32-style ABI on RISC-V.
Additionally, we've changed the behavior of the -march flag: it's now a
lot stricter and only parses things we can actually understand.
Additionally, it's now lowercase-only: the rationale is that while the
RISC-V ISA manual specifies that ISA strings are case-insensitive, in
Linux-land things are usually case-sensitive. Since this flag can be
used to determine library paths, we didn't want to bake some
case-insensitivity in there that would case trouble later.
This patch implements these two new flags and removes the old flags that
could conflict with these. There wasn't a RISC-V release before, so we
want to just support a clean flag set.
include/
* elf/riscv.h (EF_RISCV_SOFT_FLOAT): Don't define.
(EF_RISCV_FLOAT_ABI, EF_RISCV_FLOAT_ABI_SOFT): Define.
(EF_RISCV_FLOAT_ABI_SINGLE, EF_RISCV_FLOAT_ABI_DOUBLE): Define.
(EF_RISCV_FLOAT_ABI_QUAD): Define.
bfd/
* elfnn-riscv.c (_bfd_riscv_elf_merge_private_bfd_data): Use
EF_RISCV_FLOAT_ABI_SOFT instead of EF_RISCV_SOFT_FLOAT.
binutils/
* readelf.c (get_machine_flags): Use
EF_RISCV_FLOAT_ABI_{SOFT,SINGLE,DOBULE,QUAD) instead of
EF_RISCV_{SOFT,HARD}_FLOAT.
gas/
* config/tc-riscv.h (xlen): Delete.
* config/tc-riscv.c (xlen): Make static.
(abi_xlen): New variable.
(options): Replace OPTION_{M32,M64,MSOFT_FLOAT,MHARD_FLOAT,MRVC}
with OPTION_MABI.
(md_longopts): Likewise.
(md_parse_option): Likewise.
(riscv_elf_final_processing): Likewise.
* doc/as.texinfo (Target RISC-V options): Likewise.
* doc/c-riscv.texi (OPTIONS): Likewise.
* config/tc-riscv.c (float_mode): Removed.
(float_abi): New type, specifies the floating-point ABI.
(riscv_set_abi): New function.
(riscv_add_subset): Only allow lower-case ISA names and require
them to start with "rv".
(riscv_after_parse_args): Likewise.
opcodes/
* riscv-dis.c (riscv_disassemble_insn): Default to the ELF's
XLEN when none is provided.
Before this commit we didn't cleanly support CFI directives because the
internal offsets used to get relaxed which broke them. This patch
significantly reworks how we handle linker relaxations:
* DWARF is now properly supported
* There is a ".option norelax" to disable relaxations, for when users
write assembly that can't be relaxed (if it's to be later patched up,
for example).
* There is an additional _RELAX relocation that specifies when previous
relocations can be relaxed.
We're in the process of documenting the RISC-V ELF ABI, which will
include documentation of our relocations
https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md
but we expect that this relocation set will remain ABI compatible in the
future (ie, it's safe to release).
Thanks to Kuan-Lin Chen for figuring out how to correctly relax the
debug info!
include/
* elf/riscv.h: Add R_RISCV_TPREL_I through R_RISCV_SET32.
bfd/
* reloc.c (BFD_RELOC_RISCV_TPREL_I): New relocation.
(BFD_RELOC_RISCV_TPREL_S): Likewise.
(BFD_RELOC_RISCV_RELAX): Likewise.
(BFD_RELOC_RISCV_CFA): Likewise.
(BFD_RELOC_RISCV_SUB6): Likewise.
(BFD_RELOC_RISCV_SET8): Likewise.
(BFD_RELOC_RISCV_SET8): Likewise.
(BFD_RELOC_RISCV_SET16): Likewise.
(BFD_RELOC_RISCV_SET32): Likewise.
* elfnn-riscv.c (perform_relocation): Handle the new
relocations.
(_bfd_riscv_relax_tls_le): Likewise.
(_bfd_riscv_relax_align): Likewise.
(_bfd_riscv_relax_section): Likewise.
(howto_table): Likewise.
(riscv_reloc_map): Likewise.
(relax_func_t): New type.
(_bfd_riscv_relax_call): Add reserve_size argument, which
controls the maximal offset pessimism. Correct type of max_alignment.
(_bfd_riscv_relax_lui): Likewise.
(_bfd_riscv_relax_tls_le): Likewise.
(_bfd_riscv_relax_align): Likewise.
(_bfd_riscv_relax_section): Compute the required reserve size
when relocating and use it to when calling relax_func.
* bfd-in2.h: Regenerate.
* libbfd.h: Likewise.
gas/
* config/tc-riscv.c (riscv_set_options): Add relax.
(riscv_opts): Likewise.
(s_riscv_option): Add relax and norelax.
(riscv_apply_const_reloc): New function.
(append_insn): Move constant relocation handling to
riscv_apply_const_reloc.
(md_pcrel_from): Likewise.
(parse_relocation): Skip BFD_RELOC_UNUSED.
(md_pcrel_from): Handle BFD_RELOC_RISCV_SUB6,
BFD_RELOC_RISCV_RELAX, BFD_RELOC_RISCV_CFA.
(md_apply_fix): Likewise.
(riscv_pre_output_hook): New function.
* config/tc-riscv.h (md_pre_output_hook): Define.
(riscv_pre_output_hook): Declare.
(DWARF_CIE_DATA_ALIGNMENT): Always -4.
This is a mixed bag of format changes:
* Replacing constants with macros (0xffffffff with MINUS_ONE, for
example). There's one technically functional change in here (some
MINUS_ONEs are changed to 0), but it only changes the behavior of an
otherwise-unused field.
* Using 0 instead of 0x0 in the relocation table.
* There were some missing spaces before parens, the spaces have been
added.
* A handful of comments are now more descriptive.
* A bunch of whitespace-only changes, mostly alignment and brace
newlines.
bfd/
* elfnn-riscv.c: Formatting and comment fixes throughout.
* elfxx-riscv.c: Likewise.
(howto_table): Change the src_mask field from MINUS_ONE to 0 for
R_RISCV_TLS_DTPMOD32, R_RISCV_TLS_DTPMOD64, R_RISCV_TLS_DTPREL32,
R_RISCV_TLS_DTPREL64, R_RISCV_TLS_TPREL32, R_RISCV_TLS_TPREL64.
opcodes/
* riscv-opc.c: Formatting fixes.
gas/
* config/tc-riscv.c: Formatting and comment fixes throughout.
I recently ran into this error message and found it's not helpful: it
just tells me some temporary file can't be linked. This slightly
improved one at least tells me it's because of an elf32/elf64 conflict.
* elfnn-riscv.c (_bfd_riscv_elf_merge_private_bfd_data): Improve
error message when linking elf32 and elf64.
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.
I recently see the test fails like this,
(gdb) PASS: gdb.gdb/selftest.exp: step over argv initialization
list^M
487 std::vector<struct cmdarg> cmdarg_vec;^M
(gdb) FAIL: gdb.gdb/selftest.exp: unknown source line (after step over argv initialization)
step^M
std::vector<cmdarg, std::allocator<cmdarg> >::vector (this=0x7fffffffdc10) at ../../binutils-gdb/gdb/main.c:487^M
487 std::vector<struct cmdarg> cmdarg_vec;^M
(gdb) FAIL: gdb.gdb/selftest.exp: step into xmalloc call
These fails are caused by using std::vector in commit
f60ee22ea1. selttest.exp should match
the source code of GDB. It is a maintenance pain, so this patch
removes do_steps_and_nexts.
gdb/testsuite:
2016-12-19 Yao Qi <yao.qi@linaro.org>
* gdb.gdb/selftest.exp (do_steps_and_nexts): Remove.
(test_with_self): Don't call do_steps_and_nexts, and remove
code about stepping into xmalloc.
bfd/elf32_arm.c contains a function 'popcount' which conflicts
with a function of the same name in NetBSD's libc.
This change also changes popcount's 'sum' variable to signed
since the function returns a signed integer.
bfd/
* elf32-arm.c (elf32_arm_popcount): Rename from 'popcount'. Make
'sum' local variable signed.
Correct commit 640c0ccdc9 ("some objdump -M options, better reg
dumps"), <https://sourceware.org/ml/binutils/2002-12/msg00706.html>, and
only execute code setting up disassembler options based on ELF file
structures if SYMTAB_AVAILABLE is set.
opcodes/
* mips-dis.c (set_default_mips_dis_options) [SYMTAB_AVAILABLE]:
Only examine ELF file structures here.
Complement commit 5e7fc731f8 ("MIPS/opcodes: Also set disassembler's
ASE flags from ELF structures") and fix an `--enable-targets=all' GDB
build regression on 32-bit hosts where the MIPS target is a secondary:
../opcodes/libopcodes.a(mips-dis.o): In function `set_default_mips_dis_options':
mips-dis.c:(.text+0x906): undefined reference to `bfd_mips_elf_get_abiflags'
collect2: error: ld returned 1 exit status
make[2]: *** [gdb] Error 1
by avoiding making a call to the `bfd_mips_elf_get_abiflags' function,
which is not available, because there is no MIPS/ELF BFD included in
32-bit BFD builds. This call is only made from a conditional code block
guarded by a check against `bfd_target_elf_flavour', which is dead in
such a configuration, however cannot be optimized away by the compiler.
Also some other MIPS BFDs may be available, such as a.out, ECOFF or PE,
so the disassembler has to remain functional.
opcodes/
* mips-dis.c (set_default_mips_dis_options) [BFD64]: Only call
`bfd_mips_elf_get_abiflags' here.
Due to changes introduced by
commit 4d01a485d2
('struct expression *' -> gdb::unique_xmalloc_ptr<expression>)
compilation is broken on Darwin.
../gdb/darwin-nat-info.c:733:8: error: assigning to 'struct expression *'
from incompatible type
'expression_up' (aka 'std::__1::unique_ptr<expression, gdb::xfree_deleter<expression> >')
expr = parse_expression (exp);
Beside compilation, memory leak was solved as 'make_clean_up' was not called in previous
version.
2016-12-16 Bernhard Heckel <bernhard.heckel@intel.com>
gdb/Changelog:
* darwin-nat-info.c (info_mach_region_command): Use expression_up.
Respect any ASE flags recorded in ELF file structures for the purpose of
selecting instructions to be disassembled, preventing code from being
hex-dumped even though having been clearly indicated as valid at the
assembly time. Use date from the MIPS ABI flags structure if present,
and otherwise there may be an MDMX ASE flag set in the ELF file header.
For backwards compatibility only set extra flags and do not clear any,
preserving all previously set by the architecture selected to be
disassembled for.
include/
* elf/mips.h (Elf_Internal_ABIFlags_v0): Also declare struct
typedef as `elf_internal_abiflags_v0'.
bfd/
* bfd-in.h (elf_internal_abiflags_v0): New struct declaration.
(bfd_mips_elf_get_abiflags): New prototype.
* elfxx-mips.c (bfd_mips_elf_get_abiflags): New function.
* bfd-in2.h: Regenerate.
opcodes/
* mips-dis.c (mips_convert_abiflags_ases): New function.
(set_default_mips_dis_options): Also infer ASE flags from ELF
file structures.
binutils/
* testsuite/binutils-all/mips/mips-ase-1.d: New test.
* testsuite/binutils-all/mips/mips-ase-2.d: New test.
* testsuite/binutils-all/mips/mips-ase-3.d: New test.
* testsuite/binutils-all/mips/mips-ase-1.s: New test source.
* testsuite/binutils-all/mips/mips-ase-2.s: New test source.
* testsuite/binutils-all/mips/mips.exp: Run the new tests.
Move ELF file header flag interpretation code, used to set disassembler
options, beyond architecture setup. No functional change as the effects
of both code sections are disjoint from each other, but this provides
for a further expansion of ELF file header flag interpretation.
opcodes/
* mips-dis.c (set_default_mips_dis_options): Reorder ELF file
header flag interpretation code.
Complement commit c9775dde32 ("MIPS16: Add R_MIPS16_PC16_S1 branch
relocation support)" and report an assembly error when a relocation is
required for an instruction, currently a branch only, that has been
forced to use its unextended encoding, either with the use of an
explicit `.t' mnemonic suffix, or by means of `.set noautoextend' being
active, fixing an assertion failure currently caused instead.
gas/
* config/tc-mips.c (md_convert_frag): Report an error instead of
asserting on `ext'.
* testsuite/gas/mips/mips16-branch-unextended-1.d: New test.
* testsuite/gas/mips/mips16-branch-unextended-2.d: New test.
* testsuite/gas/mips/mips16-branch-unextended-1.s: New test
source.
* testsuite/gas/mips/mips16-branch-unextended-2.s: New test.
* testsuite/gas/mips/mips16-branch-unextended.l: New stderr
output.
* testsuite/gas/mips/mips.exp: Run the new tests.
Fix the annotation of SP-relative SD instructions incorrectly marked as
reading from the PC rather than SP, which in turn prevented their 16-bit
forms from being scheduled into jump delay slots. This bug has been
there since forever.
opcodes/
* mips16-opc.c (mips16_opcodes): Set RD_SP rather than RD_PC in
`pinfo2' with SP-relative "sd" entries.
gas/
* testsuite/gas/mips/mips16-sprel-swap.d: New test.
* testsuite/gas/mips/mips16-sprel-swap.s: New test source.
* testsuite/gas/mips/mips.exp: Run the new test.
Correct the note about JALRC/JRC being compact jumps rather than
branches, and add a reference from where the remaining MIPS16e additions
live and the jumps used to be too, complementing commit ceb94aa50d
("Update insn_mo when converting to a MIPS16e compact jump"),
<https://sourceware.org/ml/binutils/2011-06/msg00369.html>.
opcodes/
* mips16-opc.c (mips16_opcodes): Update comments on MIPS16e
compact jumps.
TCB_SIZE is 2*sizeof(void *), which is 0x10 for lp64, and 0x8 for
ilp32. During relaxation, ld goes to do a replace:
bl __tls_get_addr => add R0, R0, TCB_SIZE
But actual implementation is:
bfd_putl32 (0x91004000, contents + rel->r_offset + 4);
Which is equivalent of add x0, x0, 0x10. This is wrong for ilp32.
The possible fix for it is:
bfd_putl32 (0x91000000 | (TCB_SIZE<<10), contents + rel->r_offset + 4);
But ilp32 also needs w-registers, so it's simpler to put proper
instruction in #if/#else condition.
There are 2 such relaxations in elfNN_aarch64_tls_relax(), and so 2 new
tests added for ilp32 mode to test it.
Yury
* bfd/elfnn-aarch64.c: fix TLS relaxations for ilp32 where
TCB_SIZE is used.
* ld/testsuite/ld-aarch64/aarch64-elf.exp: Add tests for the case.
* ld/testsuite/ld-aarch64/tls-relax-ld-le-small-ilp32.d: New file.
* ld/testsuite/ld-aarch64/tls-relax-ld-le-tiny-ilp32.d: New file.
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
TCB_SIZE is 2*sizeof(void *), which is 0x10 for lp64, and 0x8 for
ilp32. During relaxation, ld goes to do a replace:
bl __tls_get_addr => add R0, R0, TCB_SIZE
But actual implementation is:
bfd_putl32 (0x91004000, contents + rel->r_offset + 4);
Which is equivalent of add x0, x0, 0x10. This is wrong for ilp32.
The possible fix for it is:
bfd_putl32 (0x91000000 | (TCB_SIZE<<10), contents + rel->r_offset + 4);
But ilp32 also needs w-registers, so it's simpler to put proper
instruction in #if/#else condition.
THere are 2 such relaxations in elfNN_aarch64_tls_relax(), and so 2 new
tests added for ilp32 mode to test it.
Yury
gold/
PR gold/20749
* options.h (--orphan-handling): New option.
(General_options::Orphan_handling): New enum.
(General_options::orphan_handling_enum): New method.
(General_options::set_orphan_handling_enum): New method.
(General_options::orphan_handling_enum_): New data member.
* options.cc (General_options::General_options): Initialize new member.
(General_options::finalize): Convert --orphan-handling argument to enum.
* script-sections.cc (Script_sections::output_section_name): Check it.
The internal CN register representation for coprocessor fields used in aarch64
sys, sysl instructions are removed in this patch.
After the change, those fields are represented as immediate. Related checks are
added as well.
opcodes/
* aarch64-opc.c (aarch64_opnd_qualifiers): New CR value range
qualifier.
(operand_general_constraint_met_p): Remove case for CP_REG.
(aarch64_print_operand): Print CRn, CRm operand using imm field.
* aarch64-tbl.h (QL_SYS): Use CR qualifier.
(QL_SYSL): Likewise.
(aarch64_opcode_table): Change CRn, CRm operand class and type.
* aarch64-opc-2.c : Regenerate.
* aarch64-asm-2.c : Likewise.
* aarch64-dis-2.c : Likewise.
include/
* opcode/aarch64.h (aarch64_operand_class): Remove
AARCH64_OPND_CLASS_CP_REG.
(enum aarch64_opnd): Change AARCH64_OPND_Cn to AARCH64_OPND_CRn,
AARCH64_OPND_Cm to AARCH64_OPND_CRm.
(aarch64_opnd_qualifier): Define AARCH64_OPND_QLF_CR qualifier.
gas/
* config/tc-aarch64.c (AARCH64_REG_TYPES): Remove CN register.
(get_reg_expected_msg): Remove CN register case.
(parse_operands): rewrite parser for CRn, CRm operand.
(reg_names): Remove CN register.
* testsuite/gas/aarch64/diagnostic.s: Add a new test case.
* testsuite/gas/aarch64/diagnostic.l: Adjust error message.
sim/aarch64
* simulator.c (NEG, POS): Move before set_flags_for_add64.
(set_flags_for_add64): Replace with a modified copy of
set_flags_for_sub64.
sim/testsuite/sim/aarch64
* testutils.inc (pass): Move .Lpass to start.
(fail): Move .Lfail to start. Return 1 instead of 0.
(start): Moved .Lpass and .Lfail to here.
* adds.s: New.
* fstur.s: New.
* tbnz.s: New.
There were still some cases I found where orphan section placement
was screwy -- where the script has no output section description for
either .data or .bss, a .bss orphan section ends up getting placed
before the .data section. In addition, if there is an output section
description for a data section not named .data (e.g., .rela.dyn),
the orphan .bss gets placed before it. This patch cleans that up,
by tracking the last allocated section even as we're adding orphans.
I've also improved segment layout in the absence of a PHDRS clause.
A zero-length NOBITS section will no longer force a new segment
when followed by a PROGBITS section.
2016-12-12 Cary Coutant <ccoutant@gmail.com>
gold/
* script-sections.cc (Orphan_section_placement::update_last_alloc):
New method.
(Orphan_section_placement::find_place): Place orphan .data section
after either RODATA or TEXT.
(Script_sections::place_orphan): Call update_last_alloc for allocated
sections.
(Script_sections::create_segments): Improve handling of BSS.
HPPA64 needs to add a DT_PHDR header for shared libs. That's fine
when linking but shouldn't happen for strip/objcopy. Also PHDR must
come first so there's no need to look at all program headers.
bfd/
* elf64-hppa.c (elf64_hppa_modify_segment_map): Don't add PHDR
for objcopy/strip or when a ld script specifies PHDRS.
ld/
* testsuite/ld-elf/nobits-1.d: Remove xfail for hppa64.
* testsuite/ld-elf/note-1.d: Likewise.
* testsuite/ld-elf/note-2.d: Likewise.
RX does horrible fudges to PT_LOAD p_vaddr, that affect the testsuite
and mean the target won't support dynamic objects. The latter
probably doesn't matter too much since RX is an embedded target, but
it's easy to stop some of the fudges in order to reduce special cases
for RX in the testsuite. The changes make sense in isolation too.
bfd/
* elf32-rx.c (elf32_rx_modify_program_headers): Don't adjust
segments that include the ELF file header or program headers.
ld/
* testsuite/ld-elf/flags1.d: Run for RX.
* testsuite/ld-scripts/phdrs.exp: Likewise.
* testsuite/ld-scripts/pr14962.d: Likewise.
* testsuite/ld-scripts/pr14962-2.d: Likewise.
PowerPC64 aligns .got to a 256 byte boundary. This tends to bump the
data segment file size.
PR gold/16711
* testsuite/script_test_15a.sh: Allows larger p_filesz.
* testsuite/script_test_15b.sh: Likewise.
* testsuite/script_test_15c.sh: Likewise.
