Document undocumented linker command line options.

ld  * ld.texinfo (--strip-discarded): Document.
(--embedded-relocs): Document.
(--spare-dynamic-tags): Document.
(--task-link): Document.

thread.c: ptid_equal -> operator==

gdb/ChangeLog:
2017-04-10  Pedro Alves  <palves@redhat.com>

* thread.c (add_thread_silent, delete_thread_1, find_thread_ptid)
(ptid_to_global_thread_id, in_thread_list)
(do_captured_list_thread_ids, set_resumed, set_running)
(set_executing, set_stop_requested, finish_thread_state)
(validate_registers_access, can_access_registers_ptid)
(print_thread_info_1, switch_to_thread)
(do_restore_current_thread_cleanup)
(make_cleanup_restore_current_thread, thread_command)
(thread_name_command): Use operator== instead of ptid_equal.

GC gdb/thread.c:current_thread_cleanup_chain

Commit 803bdfe43083475c7df3db38dc96f4e20d05457d ("Don't delete
thread_info if refcount isn't zero") eliminated
restore_current_thread_ptid_changed, so current_thread_cleanup_chain
is no longer necessary either.

gdb/ChangeLog:
2017-04-10  Pedro Alves  <palves@redhat.com>

* thread.c (struct current_thread_cleanup) <next>: Delete field.
(current_thread_cleanup_chain): Delete.
(restore_current_thread_cleanup_dtor)
(make_cleanup_restore_current_thread): Remove references to
current_thread_cleanup_chain.

msp430: Don't use REG_UNKNOWN

gdb/
* msp430-tdep.c (msp430_pseudo_register_read): Never return
REG_UNKNOWN.

Don't delete thread_info if refcount isn't zero

I build GDB with asan, and run test case hook-stop.exp, and threadapply.exp,
I got the following asan error,

=================================================================^M
^[[1m^[[31m==2291==ERROR: AddressSanitizer: heap-use-after-free on address 0x6160000999c4 at pc 0x000000826022 bp 0x7ffd28a8ff70 sp 0x7ffd28a8ff60^M
^[[1m^[[0m^[[1m^[[34mREAD of size 4 at 0x6160000999c4 thread T0^[[1m^[[0m^M
    #0 0x826021 in release_stop_context_cleanup ../../binutils-gdb/gdb/infrun.c:8203^M
    #1 0x72798a in do_my_cleanups ../../binutils-gdb/gdb/common/cleanups.c:154^M
    #2 0x727a32 in do_cleanups(cleanup*) ../../binutils-gdb/gdb/common/cleanups.c:176^M
    #3 0x826895 in normal_stop() ../../binutils-gdb/gdb/infrun.c:8381^M
    #4 0x815208 in fetch_inferior_event(void*) ../../binutils-gdb/gdb/infrun.c:4011^M
    #5 0x868aca in inferior_event_handler(inferior_event_type, void*) ../../binutils-gdb/gdb/inf-loop.c:44^M
....
^[[1m^[[32m0x6160000999c4 is located 68 bytes inside of 568-byte region [0x616000099980,0x616000099bb8)^M
^[[1m^[[0m^[[1m^[[35mfreed by thread T0 here:^[[1m^[[0m^M
    #0 0x7fb0bc1312ca in __interceptor_free (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x982ca)^M
    #1 0xb8c62f in xfree(void*) ../../binutils-gdb/gdb/common/common-utils.c:100^M
    #2 0x83df67 in free_thread ../../binutils-gdb/gdb/thread.c:207^M
    #3 0x83dfd2 in init_thread_list() ../../binutils-gdb/gdb/thread.c:223^M
    #4 0x805494 in kill_command ../../binutils-gdb/gdb/infcmd.c:2595^M
....

Detaching from program: /home/yao.qi/SourceCode/gnu/build-with-asan/gdb/testsuite/outputs/gdb.threads/threadapply/threadapply, process 2399^M
=================================================================^M
^[[1m^[[31m==2387==ERROR: AddressSanitizer: heap-use-after-free on address 0x6160000a98c0 at pc 0x00000083fd28 bp 0x7ffd401c3110 sp 0x7ffd401c3100^M
^[[1m^[[0m^[[1m^[[34mREAD of size 4 at 0x6160000a98c0 thread T0^[[1m^[[0m^M
    #0 0x83fd27 in thread_alive ../../binutils-gdb/gdb/thread.c:741^M
    #1 0x844277 in thread_apply_all_command ../../binutils-gdb/gdb/thread.c:1804^M
....
^M
^[[1m^[[32m0x6160000a98c0 is located 64 bytes inside of 568-byte region [0x6160000a9880,0x6160000a9ab8)^M
^[[1m^[[0m^[[1m^[[35mfreed by thread T0 here:^[[1m^[[0m^M
    #0 0x7f59a7e322ca in __interceptor_free (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x982ca)^M
    #1 0xb8c62f in xfree(void*) ../../binutils-gdb/gdb/common/common-utils.c:100^M
    #2 0x83df67 in free_thread ../../binutils-gdb/gdb/thread.c:207^M
    #3 0x83dfd2 in init_thread_list() ../../binutils-gdb/gdb/thread.c:223^M

This patch fixes the issue by deleting thread_info object if it is
deletable, otherwise, mark it as exited (by set_thread_exited).
Function set_thread_exited is shared from delete_thread_1.  This patch
also moves field "refcount" to private and methods incref and
decref.  Additionally, we stop using "ptid_t" in
"struct current_thread_cleanup" to reference threads, instead we use
"thread_info" directly.  Due to this change, we don't need
restore_current_thread_ptid_changed anymore.

gdb:

2017-04-10  Yao Qi  <yao.qi@linaro.org>

PR gdb/19942
* gdbthread.h (thread_info::deletable): New method.
(thread_info::incref): New method.
(thread_info::decref): New method.
(thread_info::refcount): Move it to private.
* infrun.c (save_stop_context): Call inc_refcount.
(release_stop_context_cleanup): Likewise.
* thread.c (set_thread_exited): New function.
(init_thread_list): Delete "tp" only it is deletable, otherwise
call set_thread_exited.
(delete_thread_1): Call set_thread_exited.
(current_thread_cleanup) <inferior_pid>: Remove.
<thread>: New field.
(restore_current_thread_ptid_changed): Removed.
(do_restore_current_thread_cleanup): Adjust.
(restore_current_thread_cleanup_dtor): Don't call
find_thread_ptid.
(set_thread_refcount): Use dec_refcount.
(make_cleanup_restore_current_thread): Adjust.
(thread_apply_all_command): Call inc_refcount.
(_initialize_thread): Don't call
observer_attach_thread_ptid_changed.

Hoist code on marking thread as exited

This patch hoists code on marking thread as exited, so more code is shared
for two different paths (thread_info is deleted or is not deleted).

gdb:

2017-04-10  Yao Qi  <yao.qi@linaro.org>

* thread.c (delete_thread_1): Hoist code on marking thread as
exited.

gas: xtensa: fix incorrect code generated with auto litpools

* config/tc-xtensa.c (xtensa_maybe_create_literal_pool_frag):
Initialize lps->frag_count with auto_litpool_limit.
(xg_promote_candidate_litpool): New function.
(xtensa_move_literals): Extract candidate litpool promotion code
into separate function. Call it for all possible found
candidates.
(xtensa_switch_to_literal_fragment): Drop 'recursive' flag and
call to xtensa_mark_literal_pool_location that it guards.
Replace it with call to xtensa_maybe_create_literal_pool_frag.
Initialize pool_location with created literal pool candidate.
* testsuite/gas/xtensa/all.exp: Add new tests.
* testsuite/gas/xtensa/auto-litpools-first1.d: New test results.
* testsuite/gas/xtensa/auto-litpools-first1.s: New test.
* testsuite/gas/xtensa/auto-litpools-first2.d: New test results.
* testsuite/gas/xtensa/auto-litpools-first2.s: New test.
* testsuite/gas/xtensa/auto-litpools.d: Fix offsets changed due
to additional jump instruction.

Port the bug fix for PR 19704 (Missing dynamic relocation against undefined weak symbol) to the SPARC architecture.

       * elf32-sparc.c (elf_backend_fixup_symbol): New.
        * elf64-sparc.c (elf_backend_fixup_symbol): New.
        * elfxx-sparc.c (UNDEFINED_WEAK_RESOLVED_TO_ZERO): New.
        (_bfd_sparc_elf_link_hash_entry): Add has_got_reloc and
        has_non_got_reloc.
        (link_hash_newfunc): Initialize has_got_reloc and
has_non_got_reloc.
        (_bfd_sparc_elf_size_dynamic_sections): Set interp to .interp
        section.
        (_bfd_sparc_elf_copy_indirect_symbol): Copy has_got_reloc and
        has_non_got_reloc.
        (_bfd_sparc_elf_check_relocs): Set has_got_reloc and
        has_non_got_reloc.
        (_bfd_sparc_elf_fixup_symbol): New function.
        (allocate_dynrelocs): Don't allocate space for dynamic
        relocations and discard relocations against resolved undefined
        weak symbols in executable.  Don't make resolved undefined weak
        symbols in executable dynamic.  Keep dynamic non-GOT/non-PLT
        relocation against undefined weak symbols in PIE.
        (_bfd_sparc_elf_relocate_section): Don't generate dynamic
        relocations against resolved undefined weak symbols in PIE
        (_bfd_sparc_elf_finish_dynamic_symbol): Keep PLT/GOT entries
        without ynamic PLT/GOT relocations for resolved undefined weak
        symbols.
        Don't generate dynamic relocation against resolved undefined
        weak symbol in executable.
        (pie_finish_undefweak_symbol): New function.
        (_bfd_sparc_elf_finish_dynamic_sections): Call
        pie_finish_undefweak_symbol on all symbols in PIE.
        * elfxx-sparc.h (_bfd_sparc_elf_link_hash_table): Add interp.
        (_bfd_sparc_elf_fixup_symbol): New function.

Remove the ns32k target from the obsolete list.

* config.bfd: Remove ns32k from obsolete list.

Tidy ppc476 opcodes

PPC_OPCODE_440 being set for ppc476 meant that many opcodes needed to
be deprecated for ppc476.  There are far fewer to add specially for
ppc476 if PPC_OPCODE_440 is not set for ppc476.

* ppc-dis.c (ppc_opts <476>): Remove PPC_OPCODE_440.
* ppc-opc.c (MULHW): Add PPC_OPCODE_476.
(powerpc_opcodes): Adjust PPC440, PPC464 and PPC476 insns to suit
removal of PPC_OPCODE_440 from ppc476 cpu selection bits.

PR21287, Inconsistent section type for .init_array and .init_array.42

PR21287 notes that .init_array is correctly given a type of
SHT_INIT_ARRAY while .init_array.nnn gets SHT_PROGBITS.  This patch
fixes that problem, and properly drops warnings from the compiler that
would cause the testsuite to fail.  My a44d0bd78 change to check
ld_compile status, necessary to pick up compile errors, also meant
warnings were not ignored.

bfd/
PR 21287
* elf.c (special_sections_f): Match .fini_array and .fini_array.*.
(special_sections_i): Likewise for .init_array.
(special_sections_p): Likewise for .preinit_array.
ld/
PR 21287
* testsuite/ld-elf/init-fini-arrays.d: Match INIT_ARRAY and FINI_ARRAY.
* testsuite/ld-elf/init-fini-arrays.s: Use %init_array and %fini_array
section types.
* testsuite/lib/ld-lib.exp (default_ld_compile): Trim assembler
warnings about "ignoring incorrect section type".
(run_ld_link_exec_tests, run_cc_link_tests): Delete old comment.

Clean elfvsb files left over from previous runs

My mips-linux and mips64-linux testsuite runs have been failing a
bunch of visibility tests, seemingly randomly.  It turns out the
problem occurs when object files are left over in ld/tmpdir from a
previous run.

* testsuite/ld-elfvsb/elfvsb.exp (visibility_run): Delete
sh1p.o, sh2p.o, sh1np.o and sh2np.o before compiling.  Use
remote_file host exists rather than file exists.

WebAssembly disassembler support

* wasm32-dis.c (print_insn_wasm32): Avoid DECIMAL_DIG, specify
appropriate floating-point precision directly.

windows-nat.c: Fix bad initialization of ptid

When trying to build for x86_64-w64-mingw32:

/home/simark/src/binutils-gdb/gdb/windows-nat.c: In function ‘void windows_detach(target_ops*, const char*, int)’:
/home/simark/src/binutils-gdb/gdb/windows-nat.c:1915:20: error: converting to ‘ptid_t’ from initializer list would use explicit constructor ‘constexpr ptid_t::ptid_t(int, long int, long int)’
   ptid_t ptid = {-1};
                    ^
Fixed by initializing ptid with the minus_one_ptid variable.

gdb/ChangeLog:

* windows-nat.c (windows_detach): Initialize ptid with
minus_one_ptid.

Automatic date update in version.in

Automatic date update in version.in

Add support for fcvtl and fcvtl2.

sim/aarch64/
* simulator.c (do_vec_FCVTL): New.
(do_vec_op1): Call do_vec_FCVTL.

sim/testsuite/sim/aarch64/
* fcvtl.s: New.

Support the fcmXX zero instructions.

sim/aarch64/
* simulator.c (do_scalar_FCMGE_zero): New.
(do_scalar_FCMLE_zero, do_scalar_FCMGT_zero, do_scalar_FCMEQ_zero)
(do_scalar_FCMLT_zero): Likewise.
(do_scalar_vec): Add calls to new functions.

sim/testsuite/sim/aarch64/
* fcmXX.s: New.

Automatic date update in version.in

Use NOPIE_CFLAGS and NOPIE_LDFLAGS to disable PIE

Since not all compilers support -no-pie, NOPIE_CFLAGS and NOPIE_LDFLAGS
should be used to disable PIE.

PR ld/21090
* testsuite/ld-x86-64/x86-64.exp (undefined_weak): Use
NOPIE_CFLAGS and NOPIE_LDFLAGS to disable PIE for the non-pie
version of the test.

ELF: Check ELF_COMMON_DEF_P for common symbols

Since common symbols that are turned into definitions don't have the
DEF_REGULAR flag set, we need to check ELF_COMMON_DEF_P for common
symbols.

bfd/

PR ld/19579
PR ld/21306
* elf32-s390.c (elf_s390_finish_dynamic_symbol): Check
ELF_COMMON_DEF_P for common symbols.
* elf64-s390.c (elf_s390_finish_dynamic_symbol): Likewise.
* elf64-x86-64.c (elf_x86_64_relocate_section): Likewise.
* elflink.c (_bfd_elf_merge_symbol): Revert commits
202ac193bbbecc96a4978d1ac3d17148253f9b01 and
07492f668d2173da7a2bda3707ff0985e0f460b6.

ld/

PR ld/19579
PR ld/21306
* testsuite/ld-elf/pr19579a.c (main): Updated.

ptid-selftests: Fix erroneous assert messages

gdb/ChangeLog:

* unittests/ptid-selftests.c: Fix erroneous assert messages.

Fix building the BFD library for Win64 by reqorking the find_separate_debug_file interface.

* opncls.c (bfd_get_debug_link_info): Rename to...
(bfd_get_debug_link_info_1): ... this.  Change type of second
parameter to void pointer.  Adjust.
(bfd_get_debug_link_info): Reimplement on top of
bfd_get_debug_link_info_1.
(separate_debug_file_exists, separate_alt_debug_file_exists):
Change type of second parameter to void pointer.  Adjust.
(get_func_type, check_func_type): Change type of second parameter
to void pointer.
(find_separate_debug_file): Add 'func_data' parameter.  Pass it to
the callback functions instead of passing the address of a local.
(bfd_follow_gnu_debuglink): Pass address of unsigned long local to
find_separate_debug_file.
(get_alt_debug_link_info_shim): Change type of second parameter to
void pointer.  Adjust.
(bfd_follow_gnu_debugaltlink): Adjust to pass NULL to
find_separate_debug_file.
(get_build_id_name, bfd_boolean check_build_id_file): Change type
of second parameter to void pointer.  Adjust.
(bfd_follow_build_id_debuglink): Pass address of bfd_build_id
pointer local to find_separate_debug_file.

Fix failure in x86_64 linker tests when compiling with a PIE enabled compiler.

PR 21090
* testsuite/ld-x86-64/x86-64.exp (undefined_weak): Explicitly
disable PIE for the non-pie version of the test.

Remove E6500 insns from PPC_OPCODE_ALTIVEC2

This isn't losing anything from the testsuite.  All of these insns
appear in testsuite/gas/ppc/e6500.s

opcodes/
* ppc-opc.c (powerpc_opcodes <mviwsplt, mvidsplt, lvexbx, lvepxl,
lvexhx, lvepx, lvexwx, stvexbx, stvexhx, stvexwx, lvtrx, lvtlx,
lvswx, stvfrx, stvflx, stvswx, lvsm, stvepxl, lvtrxl, stvepx,
lvtlxl, lvswxl, stvfrxl, stvflxl, stvswxl>): Enable E6500 only
vector instructions with E6500 not PPCVEC2.
gas/
* testsuite/gas/ppc/altivec2.s: Delete E6500 vector insns.
* testsuite/gas/ppc/altivec2.d: Adjust to suit.

Tweak MBIND ld test for ARM.

* testsuite/ld-elf/mbind2a.s: Don't use @, the ARM comment char.

MBIND gas test tweak

score-elf aligns text sections.

* testsuite/gas/elf/section12a.d: Don't expect alignment of 1
for .mbind.text.

Add BFIN_MAX_REGISTER_SIZE

gdb/
* bfin-tdep.c (BFIN_MAX_REGISTER_SIZE): Add.
(bfin_pseudo_register_read): Use BFIN_MAX_REGISTER_SIZE.
(bfin_pseudo_register_write): Likewise

pe/coff: handle weak defined symbol for gc-sections.

bfd/
* coffgen.c (_bfd_coff_gc_mark_hook): Handle PE weak
external symbols with a definition.
(_bfd_coff_gc_mark_extra_sections): Fix typo.

ld/
* testsuite/ld-pe/pe.exp: New test.
* testsuite/ld-pe/weakdef-1.s: New test source.
* testsuite/ld-pe/weakdef-1.d: New test.

Class-ify ptid_t

I grew a bit tired of using ptid_get_{lwp,pid,tid} and friends, so I decided to
make it a bit easier to use by making it a proper class.  The fields are now
private, so it's not possible to change a ptid_t field by mistake.

The new methods of ptid_t map to existing functions/practice like this:

  ptid_t (pid, lwp, tid) -> ptid_build (pid, lwp, tid)
  ptid_t (pid) -> pid_to_ptid (pid)
  ptid.is_pid () -> ptid_is_pid (ptid)
  ptid == other -> ptid_equal (ptid, other)
  ptid != other -> !ptid_equal (ptid, other)
  ptid.pid () -> ptid_get_pid (ptid)
  ptid.lwp_p () -> ptid_lwp_p (ptid)
  ptid.lwp () -> ptid_get_lwp (ptid)
  ptid.tid_p () -> ptid_tid_p (ptid)
  ptid.tid () -> ptid_get_tid (ptid)
  ptid.matches (filter) -> ptid_match (ptid, filter)

I've replaced the implementation of the existing functions with calls to
the new methods.  People are encouraged to gradually switch to using the
ptid_t methods instead of the functions (or we can change them all in
one pass eventually).

Also, I'm not sure if it's worth it (because of ptid_t's relatively
small size), but I have made the functions and methods take ptid_t
arguments by const reference instead of by value.

gdb/ChangeLog:

* common/ptid.h (struct ptid): Change to...
(class ptid_t): ... this.
<ptid_t>: New constructors.
<pid, lwp_p, lwp, tid_p, tid, is_pid, operator==, operator!=,
matches>: New methods.
<make_null, make_minus_one>: New static methods.
<pid>: Rename to...
<m_pid>: ...this.
<lwp>: Rename to...
<m_lwp>: ...this.
<tid>: Rename to...
<m_tid>: ...this.
(ptid_build, ptid_get_pid, ptid_get_lwp, ptid_get_tid, ptid_equal,
ptid_is_pid, ptid_lwp_p, ptid_tid_p, ptid_match): Take ptid arguments
as references, move comment to class ptid_t.
* common/ptid.c (null_ptid, minus_one_ptid): Initialize with
ptid_t static methods.
(ptid_build, pid_to_ptid, ptid_get_pid, ptid_get_tid,
ptid_equal, ptid_is_pid, ptid_lwp_p, ptid_tid_p, ptid_match):
Take ptid arguments as references, implement using ptid_t methods.
* unittests/ptid-selftests.c: New file.
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
unittests/ptid-selftests.c.
(SUBDIR_UNITTESTS_OBS): Add unittests/ptid-selftests.o.

gdb/gdbserver/ChangeLog:

* server.c (handle_v_cont): Initialize thread_resume::thread
with null_ptid.

Automatic date update in version.in

Add support for disassembling WebAssembly opcodes.

include * dis-asm.h: Add prototypes for wasm32 disassembler.

opcodes * Makefile.am: Add wasm32-dis.c.
* configure.ac: Add wasm32-dis.c to wasm32 target.
* disassemble.c: Add wasm32 disassembler code.
* wasm32-dis.c: New file.
* Makefile.in: Regenerate.
* configure: Regenerate.
* po/POTFILES.in: Regenerate.
* po/opcodes.pot: Regenerate.

gas * testsuite/gas/wasm32/allinsn.d: Adjust test for disassembler
changes.
* testsuite/gas/wasm32/disass.d: New test.
* testsuite/gas/wasm32/disass.s: New test.
* testsuite/gas/wasm32/disass-2.d: New test.
* testsuite/gas/wasm32/disass-2.s: New test.
* testsuite/gas/wasm32/reloc.d: Adjust test for changed reloc
names.
* testsuite/gas/wasm32/reloc.s: Update test for changed assembler
syntax.
* testsuite/gas/wasm32/wasm32.exp: Run new tests.  Expect allinsn
test to succeed.

Fix Windows gdb build failure with Python 2 support

GDB fails to build for Windows host with Python 2 support enabled due
to PyFile_FromString's second argument being of type char * and being
passed a (const) string literal. This parameter is input only so this
commit fixes the issue by casting to char *.

2017-04-06  Thomas Preud'homme  <thomas.preudhomme@arm.com>

gdb/
* python/python.c (python_run_simple_file): Cast mode literal to
non-const char pointer as expected by PyFile_FromString.

mbind tests: don't check PT_LOAD headers

On powerpc-linux, the second PT_LOAD header is RWE.  On hppa64-linux,
only one PT_LOAD header is present.  Since the tests are really about
the new GNU_MBIND headers, ignore PT_LOAD and its section mapping.

* testsuite/ld-elf/mbind1a.d: Remove matches for PT_LOAD segments.
* testsuite/ld-elf/mbind1b.d: Likewise.
* testsuite/ld-elf/mbind1c.d: Likewise.

Regen SRC-POTFILES.in

Missed when elf-properties.c was added.

* po/SRC-POTFILES.in: Regenerate.

[objcopy] Fix quadratic-time when handling --redefine-syms

objcopy/
* objcopy.c (struct redefine_node): Delete the field "next".
(redefine_sym_list): Deleted.
(redefine_specific_htab): New hash table.
(redefine_specific_reverse_htab): Likewise.
(eq_string_redefnode): New function.
(htab_hash_redefnode): Likewise.
(create_symbol2redef_htab): Likewise.
(add_specific_symbol_node): Likewise.
(create_symbol_htabs): Create redefine_specific_htab and
redefine_specific_reverse_htab.
(lookup_sym_redefinition): Use hash table instead of list.
(redefine_list_append): Likewise, and rename to add_redefine_and_check.
(copy_main): Use redefine_specific_htab instead of redefine_sym_list.
Update comments.

Automatic date update in version.in

PR ld/21233: change xfails to kfails, fix the indentation issue

ptid_{lwp,tid}_p: Remove unnecessary checks

The calls to ptid_equal in ptid_lwp_p and ptid_tid_p that compare the
argument to minus_one_ptid and null_ptid are not necessary.  The calls
in question are:

   if (ptid_equal (minus_one_ptid, ptid)
       || ptid_equal (null_ptid, ptid))
     return 0;

minus_one_ptid is { .pid = -1, .lwp = 0, .tid = 0 }
null_ptid is { .pid = 0, .lwp = 0, .tid = 0 }

If the ptid argument is either of them, the statements

  return (ptid_get_lwp (ptid) != 0);

and

  return (ptid_get_tid (ptid) != 0);

will yield the same result (0/false).

gdb/ChangeLog:

* common/ptid.c (ptid_lwp_p, ptid_tid_p): Remove comparison with
minus_one_ptid and null_ptid.

-Wwrite-strings: Remove -Wno-write-strings

AFAIK GDB is now free from -Wwrite-strings warnings.  A few warnings may
be left behind in some host-specific code, but those should be few and
easy to fix.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* warning.m4 (build_warnings): Remove -Wno-write-strings.
* configure: Regenerate.

gdb/gdbserver/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* configure: Regenerate.

-Wwrite-strings: The Rest

This is the remainder boring constification that all looks more of less
borderline obvious IMO.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* ada-exp.y (yyerror): Constify.
* ada-lang.c (bound_name, get_selections)
(ada_variant_discrim_type)
(ada_variant_discrim_name, ada_value_struct_elt)
(ada_lookup_struct_elt_type, is_unchecked_variant)
(ada_which_variant_applies, standard_exc, ada_get_next_arg)
(catch_ada_exception_command_split)
(catch_ada_assert_command_split, catch_assert_command)
(ada_op_name): Constify.
* ada-lang.h (ada_yyerror, get_selections)
(ada_variant_discrim_name, ada_value_struct_elt): Constify.
* arc-tdep.c (arc_print_frame_cache): Constify.
* arm-tdep.c (arm_skip_stub): Constify.
* ax-gdb.c (gen_binop, gen_struct_ref_recursive, gen_struct_ref)
(gen_aggregate_elt_ref): Constify.
* bcache.c (print_bcache_statistics): Constify.
* bcache.h (print_bcache_statistics): Constify.
* break-catch-throw.c (catch_exception_command_1):
* breakpoint.c (struct ep_type_description::description):
Constify.
(add_solib_catchpoint): Constify.
(catch_fork_command_1): Add cast.
(add_catch_command): Constify.
* breakpoint.h (add_catch_command, add_solib_catchpoint):
Constify.
* bsd-uthread.c (bsd_uthread_state): Constify.
* buildsym.c (patch_subfile_names): Constify.
* buildsym.h (next_symbol_text_func, patch_subfile_names):
Constify.
* c-exp.y (yyerror): Constify.
(token::oper): Constify.
* c-lang.h (c_yyerror, cp_print_class_member): Constify.
* c-varobj.c (cplus_describe_child): Constify.
* charset.c (find_charset_names): Add cast.
(find_charset_names): Constify array and add const_cast.
* cli/cli-cmds.c (complete_command, cd_command): Constify.
(edit_command): Constify.
* cli/cli-decode.c (lookup_cmd): Constify.
* cli/cli-dump.c (dump_memory_command, dump_value_command):
Constify.
(struct dump_context): Constify.
(add_dump_command, restore_command): Constify.
* cli/cli-script.c (get_command_line): Constify.
* cli/cli-script.h (get_command_line): Constify.
* cli/cli-utils.c (check_for_argument): Constify.
* cli/cli-utils.h (check_for_argument): Constify.
* coff-pe-read.c (struct read_pe_section_data): Constify.
* command.h (lookup_cmd): Constify.
* common/print-utils.c (decimal2str): Constify.
* completer.c (gdb_print_filename): Constify.
* corefile.c (set_gnutarget): Constify.
* cp-name-parser.y (yyerror): Constify.
* cp-valprint.c (cp_print_class_member): Constify.
* cris-tdep.c (cris_register_name, crisv32_register_name):
Constify.
* d-exp.y (yyerror): Constify.
(struct token::oper): Constify.
* d-lang.h (d_yyerror): Constify.
* dbxread.c (struct header_file_location::name): Constify.
(add_old_header_file, add_new_header_file, last_function_name)
(dbx_next_symbol_text, add_bincl_to_list)
(find_corresponding_bincl_psymtab, set_namestring)
(find_stab_function_addr, read_dbx_symtab, start_psymtab)
(dbx_end_psymtab, read_ofile_symtab, process_one_symbol):
* defs.h (command_line_input, print_address_symbolic)
(deprecated_readline_begin_hook): Constify.
* dwarf2read.c (anonymous_struct_prefix, dwarf_bool_name):
Constify.
* event-top.c (handle_line_of_input): Constify and add cast.
* exceptions.c (catch_errors): Constify.
* exceptions.h (catch_errors): Constify.
* expprint.c (print_subexp_standard, op_string, op_name)
(op_name_standard, dump_raw_expression, dump_raw_expression):
* expression.h (op_name, op_string, dump_raw_expression):
Constify.
* f-exp.y (yyerror): Constify.
(struct token::oper): Constify.
(struct f77_boolean_val::name): Constify.
* f-lang.c (f_word_break_characters): Constify.
* f-lang.h (f_yyerror): Constify.
* fork-child.c (fork_inferior): Add cast.
* frv-tdep.c (struct gdbarch_tdep::register_names): Constify.
(new_variant): Constify.
* gdbarch.sh (pstring_ptr, pstring_list): Constify.
* gdbarch.c: Regenerate.
* gdbcore.h (set_gnutarget): Constify.
* go-exp.y (yyerror): Constify.
(token::oper): Constify.
* go-lang.h (go_yyerror): Constify.
* go32-nat.c (go32_sysinfo): Constify.
* guile/scm-breakpoint.c (gdbscm_breakpoint_expression): Constify.
* guile/scm-cmd.c (cmdscm_function): Constify.
* guile/scm-param.c (pascm_param_value): Constify.
* h8300-tdep.c (h8300_register_name, h8300s_register_name)
(h8300sx_register_name): Constify.
* hppa-tdep.c (hppa32_register_name, hppa64_register_name):
Constify.
* ia64-tdep.c (ia64_register_names): Constify.
* infcmd.c (construct_inferior_arguments): Constify.
(path_command, attach_post_wait): Constify.
* language.c (show_range_command, show_case_command)
(unk_lang_error): Constify.
* language.h (language_defn::la_error)
(language_defn::la_name_of_this): Constify.
* linespec.c (decode_line_2): Constify.
* linux-thread-db.c (thread_db_err_str): Constify.
* lm32-tdep.c (lm32_register_name): Constify.
* m2-exp.y (yyerror): Constify.
* m2-lang.h (m2_yyerror): Constify.
* m32r-tdep.c (m32r_register_names): Constify and make static.
* m68hc11-tdep.c (m68hc11_register_names): Constify.
* m88k-tdep.c (m88k_register_name): Constify.
* macroexp.c (appendmem): Constify.
* mdebugread.c (fdr_name, add_data_symbol, parse_type)
(upgrade_type, parse_external, parse_partial_symbols)
(mdebug_next_symbol_text, cross_ref, mylookup_symbol, new_psymtab)
(new_symbol): Constify.
* memattr.c (mem_info_command): Constify.
* mep-tdep.c (register_name_from_keyword): Constify.
* mi/mi-cmd-env.c (mi_cmd_env_path, _initialize_mi_cmd_env):
Constify.
* mi/mi-cmd-stack.c (list_args_or_locals): Constify.
* mi/mi-cmd-var.c (mi_cmd_var_show_attributes): Constify.
* mi/mi-main.c (captured_mi_execute_command): Constify and add
cast.
(mi_execute_async_cli_command): Constify.
* mips-tdep.c (mips_register_name): Constify.
* mn10300-tdep.c (register_name, mn10300_generic_register_name)
(am33_register_name, am33_2_register_name)
* moxie-tdep.c (moxie_register_names): Constify.
* nat/linux-osdata.c (osdata_type): Constify fields.
* nto-tdep.c (nto_parse_redirection): Constify.
* objc-lang.c (lookup_struct_typedef, lookup_objc_class)
(lookup_child_selector): Constify.
(objc_methcall::name): Constify.
* objc-lang.h (lookup_objc_class, lookup_child_selector)
(lookup_struct_typedef): Constify.
* objfiles.c (pc_in_section): Constify.
* objfiles.h (pc_in_section): Constify.
* p-exp.y (struct token::oper): Constify.
(yyerror): Constify.
* p-lang.h (pascal_yyerror): Constify.
* parser-defs.h (op_name_standard): Constify.
(op_print::string): Constify.
(exp_descriptor::op_name): Constify.
* printcmd.c (print_address_symbolic): Constify.
* psymtab.c (print_partial_symbols): Constify.
* python/py-breakpoint.c (stop_func): Constify.
(bppy_get_expression): Constify.
* python/py-cmd.c (cmdpy_completer::name): Constify.
(cmdpy_function): Constify.
* python/py-event.c (evpy_add_attribute)
(gdbpy_initialize_event_generic): Constify.
* python/py-event.h (evpy_add_attribute)
(gdbpy_initialize_event_generic): Constify.
* python/py-evts.c (add_new_registry): Constify.
* python/py-finishbreakpoint.c (outofscope_func): Constify.
* python/py-framefilter.c (get_py_iter_from_func): Constify.
* python/py-inferior.c (get_buffer): Add cast.
* python/py-param.c (parm_constant::name): Constify.
* python/py-unwind.c (fprint_frame_id): Constify.
* python/python.c (gdbpy_parameter_value): Constify.
* remote-fileio.c (remote_fio_func_map): Make 'name' const.
* remote.c (memory_packet_config::name): Constify.
(show_packet_config_cmd, remote_write_bytes)
(remote_buffer_add_string):
* reverse.c (exec_reverse_once): Constify.
* rs6000-tdep.c (variant::name, variant::description): Constify.
* rust-exp.y (rustyyerror): Constify.
* rust-lang.c (rust_op_name): Constify.
* rust-lang.h (rustyyerror): Constify.
* serial.h (serial_ops::name): Constify.
* sh-tdep.c (sh_sh_register_name, sh_sh3_register_name)
(sh_sh3e_register_name, sh_sh2e_register_name)
(sh_sh2a_register_name, sh_sh2a_nofpu_register_name)
(sh_sh_dsp_register_name, sh_sh3_dsp_register_name)
(sh_sh4_register_name, sh_sh4_nofpu_register_name)
(sh_sh4al_dsp_register_name): Constify.
* sh64-tdep.c (sh64_register_name): Constify.
* solib-darwin.c (lookup_symbol_from_bfd): Constify.
* spu-tdep.c (spu_register_name, info_spu_dma_cmdlist): Constify.
* stabsread.c (patch_block_stabs, read_type_number)
(ref_map::stabs, ref_add, process_reference)
(symbol_reference_defined, define_symbol, define_symbol)
(error_type, read_type, read_member_functions, read_cpp_abbrev)
(read_one_struct_field, read_struct_fields, read_baseclasses)
(read_tilde_fields, read_struct_type, read_array_type)
(read_enum_type, read_sun_builtin_type, read_sun_floating_type)
(read_huge_number, read_range_type, read_args, common_block_start)
(find_name_end): Constify.
* stabsread.h (common_block_start, define_symbol)
(process_one_symbol, symbol_reference_defined, ref_add):
* symfile.c (get_section_index, add_symbol_file_command):
* symfile.h (get_section_index): Constify.
* target-descriptions.c (tdesc_type::name): Constify.
(tdesc_free_type): Add cast.
* target.c (find_default_run_target):
(add_deprecated_target_alias, find_default_run_target)
(target_announce_detach): Constify.
(do_option): Constify.
* target.h (add_deprecated_target_alias): Constify.
* thread.c (print_thread_info_1): Constify.
* top.c (deprecated_readline_begin_hook, command_line_input):
Constify.
(init_main): Add casts.
* top.h (handle_line_of_input): Constify.
* tracefile-tfile.c (tfile_write_uploaded_tsv): Constify.
* tracepoint.c (tvariables_info_1, trace_status_mi): Constify.
(tfind_command): Rename to ...
(tfind_command_1): ... this and constify.
(tfind_command): New function.
(tfind_end_command, tfind_start_command): Adjust.
(encode_source_string): Constify.
* tracepoint.h (encode_source_string): Constify.
* tui/tui-data.c (tui_partial_win_by_name): Constify.
* tui/tui-data.h (tui_partial_win_by_name): Constify.
* tui/tui-source.c (tui_set_source_content_nil): Constify.
* tui/tui-source.h (tui_set_source_content_nil): Constify.
* tui/tui-win.c (parse_scrolling_args): Constify.
* tui/tui-windata.c (tui_erase_data_content): Constify.
* tui/tui-windata.h (tui_erase_data_content): Constify.
* tui/tui-winsource.c (tui_erase_source_content): Constify.
* tui/tui.c (tui_enable): Add cast.
* utils.c (defaulted_query): Constify.
(init_page_info): Add cast.
(puts_debug, subset_compare): Constify.
* utils.h (subset_compare): Constify.
* varobj.c (varobj_format_string): Constify.
* varobj.h (varobj_format_string): Constify.
* vax-tdep.c (vax_register_name): Constify.
* windows-nat.c (windows_detach): Constify.
* xcoffread.c (process_linenos, xcoff_next_symbol_text): Constify.
* xml-support.c (gdb_xml_end_element): Constify.
* xml-tdesc.c (tdesc_start_reg): Constify.
* xstormy16-tdep.c (xstormy16_register_name): Constify.
* xtensa-tdep.c (xtensa_find_register_by_name): Constify.
* xtensa-tdep.h (xtensa_register_t::name): Constify.

gdb/gdbserver/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* gdbreplay.c (sync_error): Constify.
* linux-x86-low.c (push_opcode): Constify.

-Wwrite-strings: More Solaris

Some obvious constifications found by attempting to build 64-bit GDB
on Solaris 11.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* proc-api.c (struct trans): Constify.
(procfs_note): Constify.
* proc-events.c (struct trans, syscall_table):
* proc-flags.c (struct trans): Constify.
* proc-utils.h (procfs_note): Constify.
* proc-why.c (struct trans): Constify.
* procfs.c (dead_procinfo, find_syscall, proc_warn, proc_error)
(procfs_detach): Constify.
* sol-thread.c (struct string_map): Constify.
(td_err_string, td_state_string): Constify.

-Wwrite-strings: Fix Solaris "set procfs-file"

Compiling GDB with -Wwrite-strings flags this code in gdb/proc-api.c:

  static char *procfs_filename = "procfs_trace";

as needing a cast.  However, this variable is a command variable, and
as such it's incorrect to initialize it to a literal, since when you
use the corresponding set command, gdb frees the old string...

I didn't manage to fully build Solaris gdb (fails for other reasons),
but I confirmed that the system GDB on Solaris 11 crashes when running
this command:

 (gdb) set procfs-file foo
 Segmentation Fault (core dumped)

So I don't think this commit can make it worse than the status quo.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* proc-api.c (procfs_filename): Don't initialize
        procfs_filename.
(prepare_to_trace): Assume procfs_filename is non-NULL.
(_initialize_proc_api): Give procfs_filename a default value here.

-Wwrite-strings: Some constification in gdb/breakpoint.c

The main motivation here is avoiding having to write a couple casts
like these:

     if (!arg)
  -    arg = "";
  +    arg = (char *) "";

in catch_exception_command_1 and catch_exec_command_1.

That requires making ep_parse_optional_if_clause and
check_for_argument take pointers to const strings.  I then tried
propagating the resulting constification all the way, but that was
spiraling out of control, so instead I settled for keeping const and
non-const overloads.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* break-catch-throw.c (handle_gnu_v3_exceptions): Constify
'cond_string' parameter.
(extract_exception_regexp): Constify 'string' parameter.
(catch_exception_command_1): Constify.
* breakpoint.c (init_catchpoint)
(create_fork_vfork_event_catchpoint): Constify 'cond_string'
parameter.
(ep_parse_optional_if_clause, catch_fork_command_1)
(catch_exec_command_1): Constify.
* breakpoint.h (init_catchpoint): Constify 'cond_string'
parameter.
(ep_parse_optional_if_clause): Constify.
* cli/cli-utils.c (remove_trailing_whitespace)
(check_for_argument): Constify.
* cli/cli-utils.h (remove_trailing_whitespace): Constify and add
non-const overload.
(check_for_argument): Likewise.

-Wwrite-strings: execute_command calls with string literals

This is ugly, but it's just making the uglyness explicit.

All these places would better be calling some API function directly
instead of going through execute_command & friends...

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* event-top.c (command_line_handler): Add cast to execute_command
call.
* record-btrace.c (cmd_record_btrace_bts_start)
(cmd_record_btrace_pt_start, cmd_record_btrace_start)
(cmd_record_btrace_start): Add cast to execute_command call.
* record-full.c (record_full_goto_insn):
* record.c (record_start, record_stop): Add cast to
execute_command_to_string calls.
(cmd_record_start): Add cast to execute_command calls.

-Wwrite-strings: Add a PyArg_ParseTupleAndKeywords "const char *" overload

-Wwrite-strings flags code like:

   static char *keywords[] = {"command", "from_tty", "to_string", NULL };

as needing "(char *)" casts, because string literals are "const char []".

We can get rid of the casts by changing the array type like this:

 -  static char *keywords[] = {"command", "from_tty", "to_string", NULL };
 +  static const char *keywords[] = {"command", "from_tty", "to_string", NULL };

However, passing the such array to PyArg_ParseTupleAndKeywords no longer
works OOTB, because PyArg_ParseTupleAndKeywords expects a "char **":

  PyArg_ParseTupleAndKeywords(PyObject *args, PyObject *kw,
                              const char *format,
      char *keywords[], ...);

and "const char **" is not implicitly convertible to "char **".  C++
is more tolerant that C here WRT aliasing, and a const_cast<char **>
is fine.  However, to avoid having all callers do the cast themselves,
this commit defines a gdb_PyArg_ParseTupleAndKeywords function here
with a corresponding 'keywords' parameter type that does the cast in a
single place.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* python/python-internal.h (gdb_PyArg_ParseTupleAndKeywords): New
static inline function.
* python/py-arch.c (archpy_disassemble): Constify 'keywords'
array and use gdb_PyArg_ParseTupleAndKeywords.
* python/py-cmd.c (cmdpy_init): Likewise.
* python/py-finishbreakpoint.c (bpfinishpy_init): Likewise.
* python/py-inferior.c (infpy_read_memory, infpy_write_memory)
(infpy_search_memory): Likewise.
* python/py-objfile.c (objfpy_add_separate_debug_file)
(gdbpy_lookup_objfile): Likewise.
* python/py-symbol.c (gdbpy_lookup_symbol)
(gdbpy_lookup_global_symbol): Likewise.
* python/py-type.c (gdbpy_lookup_type): Likewise.
* python/py-value.c (valpy_lazy_string, valpy_string): Likewise.
* python/python.c (execute_gdb_command, gdbpy_write, gdbpy_flush):
Likewise.

-Wwrite-strings: Wrap PyGetSetDef for construction with string literals

Unfortunately, PyGetSetDef's 'name' and 'doc' members are 'char *'
instead of 'const char *', meaning that in order to list-initialize
PyGetSetDef arrays using string literals requires writing explicit
'char *' casts.  For example:

    static PyGetSetDef value_object_getset[] = {
   -  { "address", valpy_get_address, NULL, "The address of the value.",
   +  { (char *) "address", valpy_get_address, NULL,
   +    (char *) "The address of the value.",
NULL },
   -  { "is_optimized_out", valpy_get_is_optimized_out, NULL,
   -    "Boolean telling whether the value is optimized "
   +  { (char *) "is_optimized_out", valpy_get_is_optimized_out, NULL,
   +    (char *) "Boolean telling whether the value is optimized "
"out (i.e., not available).",
NULL },
   -  { "type", valpy_get_type, NULL, "Type of the value.", NULL },
   -  { "dynamic_type", valpy_get_dynamic_type, NULL,
   -    "Dynamic type of the value.", NULL },
   -  { "is_lazy", valpy_get_is_lazy, NULL,
   -    "Boolean telling whether the value is lazy (not fetched yet\n\
   +  { (char *) "type", valpy_get_type, NULL,
   +    (char *) "Type of the value.", NULL },
   +  { (char *) "dynamic_type", valpy_get_dynamic_type, NULL,
   +    (char *) "Dynamic type of the value.", NULL },
   +  { (char *) "is_lazy", valpy_get_is_lazy, NULL,
   +    (char *) "Boolean telling whether the value is lazy (not fetched yet\n\
    from the inferior).  A lazy value is fetched when needed, or when\n\
    the \"fetch_lazy()\" method is called.", NULL },
      {NULL}  /* Sentinel */

We have ~20 such arrays, and I first wrote a patch that fixed all of
them like that...  It's not pretty...

One way to make these a bit less ugly would be add a new macro that
hides the casts, like:

  #define GDBPY_GSDEF(NAME, GET, SET, DOC, CLOSURE) \
     { (char *) NAME, GET, SET, (char *) DOC, CLOSURE }

and then use it like:

    static PyGetSetDef value_object_getset[] = {
       GDBPY_GSDEF ("address", valpy_get_address, NULL,
            "The address of the value.", NULL),
       GDBPY_GSDEF ("is_optimized_out", valpy_get_is_optimized_out, NULL,
            "Boolean telling whether the value is optimized ", NULL),
      {NULL}  /* Sentinel */
    };

But since we have C++11, which gives us constexpr and list
initialization, I thought of a way that requires no changes where the
arrays are initialized:

We add a new type that extends PyGetSetDef (called gdb_PyGetSetDef),
and add constexpr constructors that accept const 'name' and 'doc', and
then list/aggregate initialization simply "calls" these matching
constructors instead.

I put "calls" in quotes, because given "constexpr", it's all done at
compile time, and there's no overhead either in binary size or at run
time.  In fact, we get identical binaries, before/after this change.

Unlike the fixes that fix some old Python API to match the API of more
recent Python, this switches to using explicit "gdb_PyGetSetDef"
everywhere, just to be clear that we are using our own version of it.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* python/python-internal.h (gdb_PyGetSetDef): New type.
* python/py-block.c (block_object_getset)
(breakpoint_object_getset): Now a gdb_PyGetSetDef array.
* python/py-event.c (event_object_getset)
(finish_breakpoint_object_getset): Likewise.
* python/py-inferior.c (inferior_object_getset): Likewise.
* python/py-infthread.c (thread_object_getset): Likewise.
* python/py-lazy-string.c (lazy_string_object_getset): Likewise.
* python/py-linetable.c (linetable_entry_object_getset): Likewise.
* python/py-objfile.c (objfile_getset): Likewise.
* python/py-progspace.c (pspace_getset): Likewise.
* python/py-record-btrace.c (btpy_insn_getset, btpy_call_getset):
Likewise.
* python/py-record.c (recpy_record_getset): Likewise.
* python/py-symbol.c (symbol_object_getset): Likewise.
* python/py-symtab.c (symtab_object_getset, sal_object_getset):
Likewise.
* python/py-type.c (type_object_getset, field_object_getset):
Likewise.
* python/py-value.c (value_object_getset): Likewise.

-Wwrite-strings: More fix-old-Python-API wrappers

When building against Python 2.7, -Wwrite-strings flags several cases
of passing a string literal to Python functions that expect a "char
*".  This commit addresses the issue like we already handle several
other similar cases -- wrap the Python API with our own fixed
version that adds the necessary constification.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* python/python-internal.h (gdb_PyObject_CallMethod)
(gdb_PyErr_NewException, gdb_PySys_GetObject, gdb_PySys_SetPath):
New functions.
(GDB_PYSYS_SETPATH_CHAR, PyObject_CallMethod, PyErr_NewException)
(PySys_GetObject, PySys_SetPath): New macros.

-Wwrite-strings: gdbserver/win32-low.c and TARGET_WAITKIND_EXECD

 src/gdb/gdbserver/win32-low.c:1499:39: error: ISO C++ forbids converting a string constant to 'char*' [-Werror=write-strings]
ourstatus->value.execd_pathname = "Main executable";
^

This reporting via TARGET_WAITKIND_EXECD it's totally unnecessary.
get_child_debug_event returns a TARGET_WAITKIND_SPURIOUS by default,
which works just as well here, and is what the equivalent code in
gdb/windows-nat.c does too.

gdb/gdbserver/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* win32-low.c (get_child_debug_event)
<CREATE_PROCESS_DEBUG_EVENT>: Don't report TARGET_WAITKIND_EXECD.
Report TARGET_WAITKIND_SPURIOUS instead.

-Wwrite-strings: gdbserver's 'port' parsing

-Wwrite-strings flags this assignment as requiring a cast:

  port = STDIO_CONNECTION_NAME;

because 'port' is a "char *", and STDIO_CONNECTION_NAME is a string
literal.

gdb/gdbserver/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

        * remote-utils.c (remote_prepare, remote_open): Constify.
        * remote-utils.h (remote_prepare, remote_open): Constify.
        * server.c (captured_main): Constify 'port' handling.

-Wwrite-strings: MI -info-os

-Wwrite-strings flags this attempt to convert a string literal to
 "char *":

      info_osdata_command ("", 0);

info_osdata_command is a command function.  We could address this by
simply passing NULL instead of "".  However, I went a little bit
further and added a new function that is called by both the CLI and
MI.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* mi/mi-cmd-info.c (mi_cmd_info_os): Call info_osdata instead of
info_osdata_command.
* osdata.c (info_osdata_command): Rename to ...
(info_osdata): ... this.  Constify 'type' parameter, and remove
the 'from_tty' parameter.  Accept NULL TYPE.
(info_osdata_command): New function.
* osdata.h (info_osdata_command): Remove declaration.
(info_osdata): New declaration.

-Wwrite-strings: Constify mi_cmd_argv_ftype's 'command' parameter

-Wwrite-strings flags this attempt to pass a literal to a "char *":

mi_cmd_interpreter_exec ("-interpreter-exec", argv, 2);

Fix that by constifying mi_cmd_argv_ftype's 'command' parameter and
adjusting all MI commands.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* mi/mi-cmd-break.c (mi_cmd_break_insert_1, mi_cmd_break_insert)
(mi_cmd_dprintf_insert, mi_cmd_break_passcount)
(mi_cmd_break_watch, mi_cmd_break_commands): Constify 'command'
parameter.
* mi/mi-cmd-catch.c (mi_cmd_catch_assert, mi_cmd_catch_exception)
(mi_cmd_catch_load, mi_cmd_catch_unload): Constify cmd' parameter.
* mi/mi-cmd-disas.c (mi_cmd_disassemble): Constify 'command'
parameter.
* mi/mi-cmd-env.c (mi_cmd_env_pwd, mi_cmd_env_cd, mi_cmd_env_path)
(mi_cmd_env_dir, mi_cmd_inferior_tty_set, _cmd_inferior_tty_show)
* mi/mi-cmd-file.c (mi_cmd_file_list_exec_source_file)
(mi_cmd_file_list_exec_source_files)
(mi_cmd_file_list_shared_libraries): Constify 'command' parameter.
* mi/mi-cmd-info.c (mi_cmd_info_ada_exceptions)
(mi_cmd_info_gdb_mi_command, mi_cmd_info_os): Constify 'command'
parameter.
* mi/mi-cmd-stack.c (mi_cmd_enable_frame_filters)
(mi_cmd_stack_list_frames, mi_cmd_stack_info_depth)
(mi_cmd_stack_list_locals, mi_cmd_stack_list_args)
(mi_cmd_stack_list_variables, mi_cmd_stack_select_frame)
(mi_cmd_stack_info_frame): Constify 'command' parameter.
* mi/mi-cmd-target.c (mi_cmd_target_file_get)
(mi_cmd_target_file_put, mi_cmd_target_file_delete): Constify
'command' parameter.
* mi/mi-cmd-var.c (mi_cmd_var_create, mi_cmd_var_delete)
(mi_cmd_var_set_format, mi_cmd_var_set_visualizer)
(mi_cmd_var_set_frozen, mi_cmd_var_show_format)
(mi_cmd_var_info_num_children, mi_cmd_var_list_children)
(mi_cmd_var_info_type, mi_cmd_var_info_path_expression)
(mi_cmd_var_info_expression, mi_cmd_var_show_attributes)
(mi_cmd_var_evaluate_expression, mi_cmd_var_assign)
(mi_cmd_var_update, mi_cmd_enable_pretty_printing)
(mi_cmd_var_set_update_range): Constify 'command' parameter.
* mi/mi-cmds.h (mi_cmd_argv_ftype): Constify 'command' parameter.
* mi/mi-interp.c (mi_cmd_interpreter_exec): Constify 'command'
parameter.
* mi/mi-main.c (mi_cmd_gdb_exit, mi_cmd_exec_next)
(mi_cmd_exec_next_instruction, mi_cmd_exec_step)
(mi_cmd_exec_step_instruction, mi_cmd_exec_finish)
(mi_cmd_exec_return ,mi_cmd_exec_jump, mi_cmd_exec_continue)
(mi_cmd_exec_interrupt, mi_cmd_exec_run, mi_cmd_target_detach)
(mi_cmd_target_flash_erase, mi_cmd_thread_select)
(mi_cmd_thread_list_ids, mi_cmd_thread_info)
(mi_cmd_list_thread_groups, mi_cmd_data_list_register_names)
(mi_cmd_data_list_changed_registers)
(mi_cmd_data_write_register_values)
(mi_cmd_data_evaluate_expression, mi_cmd_data_read_memory)
(mi_cmd_data_read_memory_bytes, mi_cmd_data_write_memory)
(mi_cmd_data_write_memory_bytes, mi_cmd_enable_timings)
(mi_cmd_list_features, mi_cmd_list_target_features)
(mi_cmd_add_inferior, mi_cmd_remove_inferior)
(mi_cmd_trace_define_variable, mi_cmd_trace_list_variables)
(mi_cmd_trace_find, mi_cmd_trace_save, mi_cmd_trace_start)
(mi_cmd_trace_status, mi_cmd_trace_stop, mi_cmd_ada_task_info)
(mi_cmd_trace_frame_collected): Constify 'command'
parameter.
* mi/mi-symbol-cmds.c (mi_cmd_symbol_list_lines): Constify
'command' parameter.

-Wwrite-strings: Constify word break character arrays

-Wwrite-strings flags several cases of missing casts around
initializations like:

   static char *gdb_completer_command_word_break_characters =
    " \t\n!@#$%^&*()+=|~`}{[]\"';:?/>.<,";

Obviously these could/should be const.  However, while at it, there's
no need for these variables to be pointers instead of arrays.  They
are never changed to point to anything else.

Unfortunately, readline's rl_completer_word_break_characters is
"char *", not "const char *".  So we always need a cast somewhere.  The
approach taken here is to add a new
set_rl_completer_word_break_characters function that becomes the only
place that writes to rl_completer_word_break_characters, and thus the
single place that needs the cast.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* ada-lang.c (ada_completer_word_break_characters): Now a const
array.
(ada_get_gdb_completer_word_break_characters): Constify.
* completer.c (gdb_completer_command_word_break_characters)
(gdb_completer_file_name_break_characters)
(gdb_completer_quote_characters): Now const arrays.
(get_gdb_completer_quote_characters): Constify.
(set_rl_completer_word_break_characters): New function.
(set_gdb_completion_word_break_characters)
(complete_line_internal): Use it.
* completer.h (get_gdb_completer_quote_characters): Constify.
(set_rl_completer_word_break_characters): Declare.
* f-lang.c (f_word_break_characters): Constify.
* language.c (default_word_break_characters): Constify.
* language.h (language_defn::la_word_break_characters): Constify.
(default_word_break_characters): Constify.
* top.c (init_main): Use set_rl_completer_word_break_characters.

-Wwrite-strings: Constify target_pid_to_str and target_thread_extra_thread_info

-Wwrite-strings flagged a missing cast for example here:

   static char *
   ravenscar_extra_thread_info (struct target_ops *self, struct thread_info *tp)
   {
     return "Ravenscar task";

Since callers are not supposed to free the string returned by these
methods, change the methods' signature to return const strings.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* aix-thread.c (aix_thread_pid_to_str)
(aix_thread_extra_thread_info): Constify.
* bsd-kvm.c (bsd_kvm_pid_to_str): Constify.
* bsd-uthread.c (bsd_uthread_extra_thread_info)
(bsd_uthread_pid_to_str): Constify.
* corelow.c (core_pid_to_str): Constify.
* darwin-nat.c (darwin_pid_to_str): Constify.
* fbsd-nat.c (fbsd_pid_to_str): Constify.
* fbsd-tdep.c (fbsd_core_pid_to_str, gdbarch_core_pid_to_str):
Constify.
* gnu-nat.c (gnu_pid_to_str): Constify.
* go32-nat.c (go32_pid_to_str): Constify.
* i386-cygwin-tdep.c (i386_windows_core_pid_to_str): Constify.
* inf-ptrace.c (inf_ptrace_pid_to_str): Constify.
* inferior.c (inferior_pid_to_str): Constify.
* linux-nat.c (linux_nat_pid_to_str): Constify.
* linux-tdep.c (linux_core_pid_to_str): Constify.
* linux-thread-db.c (thread_db_pid_to_str)
(thread_db_extra_thread_info): Constify.
* nto-tdep.c (nto_extra_thread_info): Constify.
* nto-tdep.h (nto_extra_thread_info): Constify.
* obsd-nat.c (obsd_pid_to_str): Constify.
* procfs.c (procfs_pid_to_str): Constify.
* ravenscar-thread.c (ravenscar_extra_thread_info)
(ravenscar_pid_to_str): Constify.
* remote-sim.c (gdbsim_pid_to_str): Constify.
* remote.c (remote_threads_extra_info, remote_pid_to_str):
Constify.
* sol-thread.c (solaris_pid_to_str): Constify.
* sol2-tdep.c (sol2_core_pid_to_str): Constify.
* sol2-tdep.h (sol2_core_pid_to_str): Constify.
* target.c (default_pid_to_str, target_pid_to_str)
(normal_pid_to_str, default_pid_to_str): Constify.
* target.h (target_ops::to_pid_to_str)
(target_ops::to_extra_thread_info): Constify.
(target_pid_to_str, normal_pid_to_str): Constify.
* windows-nat.c (windows_pid_to_str): Constify.
* gdbarch.sh (core_pid_to_str): Constify.
* target-delegates.c: Regenerate.
* gdbarch.h, gdbarch.c: Regenerate.

-Wwrite-strings: Constify warning_pre_print

-Wwrite-strings flags a warning here:

  char *warning_pre_print = "\nwarning: ";

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* main.c (captured_main_1): Use gdb::unique_xmalloc_ptr to manage
the memory of the temporary warning_pre_print override.
* utils.c (warning_pre_print): Constify.
* utils.h (warning_pre_print): Constify.

-Wwrite-strings: Constify shell_escape and plug make_command leak

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* cli/cli-cmds.c (shell_escape): Constify 'arg' parameter.
(shell_command): New function.
(make_command): Use std::string.
(init_cli_cmds): Register shell_command instead of shell_escape.

-Wwrite-strings: Don't initialize string command variables to empty string

-Wwrite-strings flags these initializations as requiring a cast.
However, these variables are command variables, and as such point to
heap-allocated memory.  The initial allocation is always done when the
corresponding command is registered.  E.g.,:

    dprintf_function = xstrdup ("printf");
    add_setshow_string_cmd ("dprintf-function", class_support,
    &dprintf_function, _("\
  Set the function to use for dynamic printf"), _("\
  Show the function to use for dynamic printf"), NULL,
    update_dprintf_commands, NULL,
    &setlist, &showlist);

That's why we never reach a bogus attempt to free these string
literals.

So, just drop the incorrect initializations.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* breakpoint.c (dprintf_function, dprintf_channel): Don't initialize.
* tracepoint.c (default_collect): Don't initialize.

-Wwrite-strings: Constify macroexp.c:init_shared_buffer

There's one call in the file that passes a string literal, like:

init_shared_buffer (&va_arg_name, "__VA_ARGS__",
  strlen ("__VA_ARGS__"));

Instead of adding a cast here, make init_shared_buffer take a 'const
char *', and remove the several casts in the file that are made
obsolete.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* macroexp.c (macro_buffer::shared): Now a bool.
(init_buffer): Update.
(init_shared_buffer): Constify 'addr' parameter.
(substitute_args, expand, macro_expand, macro_expand_next): Remove
casts.

-Wwrite-strings: Constify struct disassemble_info's disassembler_options field

The memory disassemble_info::disassembler_options points to is always
owned by the client.  I.e., that field is an non-owning, observing
pointer.  Thus const makes sense.

Are the include/ and opcodes/ bits OK?

Tested on x86_64 Fedora 23, built with --enable-targets=all.

include/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* dis-asm.h (disassemble_info) <disassembler_options>: Now a
"const char *".
(next_disassembler_option): Constify.

opcodes/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* arc-dis.c (parse_option, parse_disassembler_options): Constify.
* arm-dis.c (parse_arm_disassembler_options): Constify.
* ppc-dis.c (powerpc_init_dialect): Constify local.
* vax-dis.c (parse_disassembler_options): Constify.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* arm-tdep.c (show_disassembly_style_sfunc): Constify local.
* disasm.c (set_disassembler_options): Constify local.
* i386-tdep.c (i386_print_insn): Remove cast and FIXME comment.

PR 21352: Add testsuite for "tsave -r" command

This commit adds a test for the fix of PR 21352.

gdb/testsuite/ChangeLog:
2017-04-05  Sergio Durigan Junior  <sergiodj@redhat.com>

PR gdb/21352
* gdb.trace/tsv.exp: Add test for "tsave -r".

Fix PR 21352: Command tsave does not support -r argument

This is an obvious fix for PR 21352.  The problem is that the argument
parsing loop is not using an "else if" where it should, and therefore
the '-r' option ends up unrecognized by GDB.

gdb/ChangeLog:
2017-04-05  Sergio Durigan Junior  <sergiodj@redhat.com>

PR gdb/21352
* tracefile.c (tsave_command): Fix argument parsing for '-r'
option.

Use frame_unwind_register_value in frame_unwind_register_unsigned

gdb:

2017-04-05  Yao Qi  <yao.qi@linaro.org>

* frame.c (frame_unwind_register_unsigned): Call
frame_unwind_register_value.

Fix racy test in gdb.threads/thread-specific-bp.exp

I see the following test fail from time to time, due to the racy test
in gdb.threads/thread-specific-bp.exp.

continue -a^M
Continuing.^M
^M
Thread 1 "thread-specific" hit Breakpoint 4, end () at binutils-gdb/gdb/testsuite/gdb.threads/thread-specific-bp.c:29^M
29      }^M
(gdb) [Thread 0x40322460 (LWP 12950) exited]^M
Thread-specific breakpoint 3 deleted - thread 2 no longer in the thread list.^M
FAIL: gdb.threads/thread-specific-bp.exp: non-stop: continue to end (timeout)

This patch changes gdb_test to gdb_test_multiple to match prompt only
instead of both prompt and anchor.

gdb/testsuite:

2017-04-05  Yao Qi  <yao.qi@linaro.org>

* gdb.threads/thread-specific-bp.exp (check_thread_specific_breakpoint):
Use gdb_test_multiple, and don't match anchor.

PowerPC64le PLT reference counting

A fix for ELFv2 ABI garbage-collection.

* elf64-ppc.c (ppc64_elf_gc_sweep_hook): Support ELFv2 PLT
reference counting.

New global maintainer - Simon Marchi

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* MAINTAINERS (Global Maintainers): Add Simon Marchi.
(Write After Approval): Remove Simon Marchi.

PR ld/21233: xfail failing test-parts for cris*-*-*

Initialize gdb::optional empty payload to quiet false -Wmaybe-uninitialized warnings

Commit ecfb656c37b982 ("dwarf2read.c: Make dir_index and
file_name_index strong typedefs") added a use of gdb::optional that
triggers bogus -Wmaybe-uninitialized warnings:

GCC trunk is complaining like this:

  ../../binutils-gdb/gdb/dwarf2read.c: In function void read_formatted_entries(bfd*, const gdb_byte**, line_header*, const comp_unit_head*, void (*)(line_header*, const char*, dir_index, unsigned int, unsigned int)):
  ../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::length may be used uninitialized in this function [-Werror=maybe-uninitialized]
 callback (lh, fe.name, fe.d_index, fe.mod_time, fe.length);
   ^
  ../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: *((void*)& fe +8) may be used uninitialized in this function [-Werror=maybe-uninitialized]
  ../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::mod_time may be used uninitialized in this function [-Werror=maybe-uninitialized]
  ../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::name may be used uninitialized in this function [-Werror=maybe-uninitialized]

While some older GCCs are complaining like this:

  ../../binutils-gdb/gdb/dwarf2read.c: In function void read_formatted_entries(bfd*, const gdb_byte**, line_header*, const comp_unit_head*, void (*)(line_header*, const char*, dir_index, unsigned int, unsigned int)):
  ../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: uint may be used uninitialized in this function [-Werror=maybe-uninitialized]
 callback (lh, fe.name, fe.d_index, fe.mod_time, fe.length);

Looking around the web, I see that boost's optional implementation
triggers this kind of issue often too.  See:

  http://www.boost.org/doc/libs/1_63_0/libs/optional/doc/html/boost_optional/tutorial/gotchas/false_positive_with__wmaybe_uninitialized.html

I noticed that replacing the gdb::optional uses with real C++17
std::optional uses against GCC 7/trunk makes the warnings go away.
Looking at the implementation, AFAICS, libstdc++ always initializes
its "empty" union payload member (_M_empty, which is defined as an
empty class, like ours).  I.e., all payload types have this ctor:

    struct _Optional_payload.....
    {
      constexpr _Optional_payload()
: _M_empty() {}

The constexpr makes a diference too.  Without it, GCC7 still warns.

So I'm applying the same treatment to our gdb::optional.

gdb/ChangeLog:
2017-04-05  Pedro Alves  <palves@redhat.com>

* common/gdb_optional.h (optional::optional): Make constexpr and
initialize m_dummy.

Automatic date update in version.in

PR ld/21233: Avoid sweeping forced-undefined symbols in section GC

Complement commit 902e9fc76a0e ("PR ld/20828: Move symbol version
processing ahead of GC symbol sweep"), commit b531344c34b0 ("PR
ld/20828: Reorder the symbol sweep stage of section GC") and commit
81ff47b3a546 ("PR ld/20828: Fix linker script symbols wrongly forced
local with section GC"), and prevent symbols forcibly entered in the
output file with the use of the `--undefined=' or `--require-defined='
linker command line options or the EXTERN linker script command from
being swept in section garbage collection and consequently recorded in
the dynamic symbol table as local entries.  This happens in certain
circumstances, where a symbol reference also exists in one of the static
input files, however only in a section which is garbage-collected and
does not make it to the output file, and the symbol is defined in a
dynamic object present in the link.

For example with the `i386-linux' target and the `pr21233.s' and
`pr21233-l.s' sources, and the `pr21233.ld' linker script included with
this change we get:

$ as -o pr21233-l.o pr21233-l.s
$ ld -shared -T pr21233.ld -o libpr21233.so pr21233-l.o
$ as -o pr21233.o pr21233.s
$ ld --gc-sections -e foo --require-defined=bar -T pr21233.ld -o pr21233 pr21233.o libpr21233.so
$ readelf --dyn-syms pr21233

Symbol table '.dynsym' contains 2 entries:
   Num:    Value  Size Type    Bind   Vis      Ndx Name
     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 00000000     0 OBJECT  LOCAL  DEFAULT  UND bar
$

which makes the run-time `bar' dependency of the `pr21233' executable
different from its corresponding link-time dependency, i.e. the presence
of `libpr21233.so' and its `bar' symbol is required at the link time,
however at the run time a copy of `libpr21233.so' without `bar' will do.
Similarly with `--undefined=' and EXTERN which do not actually require
the reference to the symbol requested to be satisfied with a definition
at the link time, however once the definition has been pulled at the
link time, so it should at the dynamic load time.

Additionally with the `mips-linux' target we get:

$ ld --gc-sections -e foo --require-defined=bar -T pr21233.ld -o pr21233 pr21233.o libpr21233.so
ld: BFD (GNU Binutils) 2.28.51.20170324 assertion fail .../bfd/elfxx-mips.c:3861
$

as the target is not prepared to handle such a local dynamic symbol.

With this change in effect we get:

$ readelf --dyn-syms pr21233

Symbol table '.dynsym' contains 2 entries:
   Num:    Value  Size Type    Bind   Vis      Ndx Name
     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 00000000     0 OBJECT  GLOBAL DEFAULT  UND bar
$

instead, for both targets.

ld/
PR ld/21233
* ldlang.c (insert_undefined): Set `mark' for ELF symbols.
* testsuite/ld-elf/pr21233.sd: New test.
* testsuite/ld-elf/pr21233-l.sd: New test.
* testsuite/ld-elf/pr21233.ld: New test linker script.
* testsuite/ld-elf/pr21233-e.ld: New test linker script.
* testsuite/ld-elf/pr21233.s: New test source.
* testsuite/ld-elf/pr21233-l.s: New test source.
* testsuite/ld-elf/shared.exp: Run the new tests.

Remove support for using the bsd-uthread target on FreeBSD.

The bsd-uthread target supports an old thread library ("libc_r") that
was last included in FreeBSD release 6.4 released in 2008.  However,
this library has not been used as the default library since FreeBSD
5.0 (2003) and 4.11 (2005).  Thread support for modern FreeBSD binaries
is provided via "normal" LWP support in core files and the native
FreeBSD target.

gdb/ChangeLog:

* amd64-fbsd-tdep.c: Remove "bsd-uthread.h" include.
(amd64fbsd_jmp_buf_reg_offset): Remove.
(amd64fbsd_supply_uthread): Remove function.
(amd64fbsd_collect_uthread): Remove function.
(amd64fbsd_init_abi): Don't set bsd-uthread callbacks.
* configure.tgt (i[34567]86-*-freebsd*): Remove bsd-uthread.o.
(x86_64-*-freebsd*): Remove bsd-uthread.o.
(fbsd-nat.c): Update comment.
* i386-fbsd-tdep.c: Remove "bsd-uthread.h" include.
(i386fbsd_jmp_buf_reg_offset): Remove.
(i386fbsd_supply_uthread): Remove function.
(i386fbsd_collect_uthread): Remove function.
(i386fbsd_init_abi): Don't set bsd-uthread callbacks.

Remove support for FreeBSD/alpha.

FreeBSD last shipped a release for Alpha (6.3) in 2008.
This also removes support for GNU/kFreeBSD on Alpha.

gdb/ChangeLog:

* Makefile.in (ALL_64_TARGET_OBS): Remove alpha-fbsd-tdep.o.
(ALLDEPFILES): Remove alpha-fbsd-tdep.c
* NEWS: Mention that support for FreeBSD/alpha was removed.
* alpha-fbsd-tdep.c: Delete file.
* config/alpha/fbsd.mh: Delete file.
* configure.host: Delete alpha*-*-freebsd* and
alpha*-*-kfreebsd*-gnu.
* configure.tgt: Delete alpha*-*-freebsd* and
alpha*-*-kfreebsd*-gnu.

Use ptid from regcache instead of inferior_ptid.

gdb/ChangeLog:

* amd64-bsd-nat.c (amd64bsd_fetch_inferior_registers,
amd64bsd_store_inferior_registers): Use ptid from regcache.

dwarf2read.c: C++fy lnp_state_machine

While I was looking at the file, I noticed that this struct could be
nicely converted to a class.  As I was progressing, I ended up moving
all state machine actual internal state manipulation to methods of
lnp_state_machine, essentially decoupling DWARF parsing from state
tracking.  I also noticed that the lnp_reader_state doesn't really
serve any good use, so that's eliminated in the process.

gdb/ChangeLog:
2017-04-04  Pedro Alves  <palves@redhat.com>

* dwarf2read.c (lnp_state_machine): Now a class.  Initialize all
data fields, make them private and add "m_" prefixes.
(lnp_state_machine::lnp_state_machine): New ctor.
(record_line, check_line_address, handle_set_discriminator)
(handle_set_address, handle_advance_pc, handle_special_opcode)
(handle_advance_line, handle_set_file, handle_negate_stmt)
(handle_const_add_pc, handle_fixed_advance_pc, handle_copy)
(end_sequence, advance_line): New methods.
(m_gdbarch, m_record_lines_p): New fields.
(lnp_reader_state): Delete.
(dwarf_record_line): Rename to ...
(lnp_state_machine::record_line): ... adjust.
(init_lnp_state_machine): Delete.
(lnp_state_machine::lnp_state_machine): New.
(check_line_address): Rename to ...
(lnp_state_machine::check_line_address): This.
(dwarf_decode_lines_1): Remove reference to "reader_state".
Adjust lnp_state_machine having a non-default ctor.  Use bool.
State machine internal state manipulation moved to
lnp_state_machine methods.

Make sect_offset and cu_offset strong typedefs instead of structs

A while ago, back when GDB was a C program, the sect_offset and
cu_offset types were made structs in order to prevent incorrect mixing
of those offsets.  Now that we require C++11, we can make them
integers again, while keeping the safety, by exploiting "enum class".
We can add a bit more safety, even, by defining operators that the
types _should_ support, helping making the suspicious uses stand out
more.

Getting at the underlying type is done with the new to_underlying
function added by the previous patch, which also helps better spot
where do we need to step out of the safety net.  Mostly, that's around
parsing the DWARF, and when we print the offset for complaint/debug
purposes.  But there are other occasional uses.

Since we have to define the sect_offset/cu_offset types in a header
anyway, I went ahead and generalized/library-fied the idea of "offset"
types, making it trivial to add more such types if we find a use.  See
common/offset-type.h and the DEFINE_OFFSET_TYPE macro.

I needed a couple generaly-useful preprocessor bits (e.g., yet another
CONCAT implementation), so I started a new common/preprocessor.h file.

I included units tests covering the "offset" types API.  These are
mostly compile-time tests, using SFINAE to check that expressions that
shouldn't compile (e.g., comparing unrelated offset types) really are
invalid and would fail to compile.  This same idea appeared in my
pending enum-flags revamp from a few months ago (though this version
is a bit further modernized compared to what I had posted), and I plan
on reusing the "check valid expression" bits added here in that
series, so I went ahead and defined the CHECK_VALID_EXPR macro in its
own header -- common/valid-expr.h.  I think that's nicer regardless.

I was borderline between calling the new types "offset" types, or
"index" types, BTW.  I stuck with "offset" simply because that's what
we're already calling them, mostly.

gdb/ChangeLog:
2017-04-04  Pedro Alves  <palves@redhat.com>

* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
unittests/offset-type-selftests.c.
(SUBDIR_UNITTESTS_OBS): Add offset-type-selftests.o.
* common/offset-type.h: New file.
* common/preprocessor.h: New file.
* common/traits.h: New file.
* common/valid-expr.h: New file.
* dwarf2expr.c: Include "common/underlying.h".  Adjust to use
sect_offset and cu_offset strong typedefs throughout.
* dwarf2expr.h: Adjust to use sect_offset and cu_offset strong
typedefs throughout.
* dwarf2loc.c: Include "common/underlying.h".  Adjust to use
sect_offset and cu_offset strong typedefs throughout.
* dwarf2read.c: Adjust to use sect_offset and cu_offset strong
typedefs throughout.
* gdbtypes.h: Include "common/offset-type.h".
(cu_offset): Now an offset type (strong typedef) instead of a
struct.
(sect_offset): Likewise.
(union call_site_parameter_u): Rename "param_offset" field to
"param_cu_off".
* unittests/offset-type-selftests.c: New file.

dwarf2read.c: Make dir_index and file_name_index strong typedefs

This should help catch mistakes related to mixing the 1-based DWARF
indexes with 0-based std::vector indexes, since the new types do not
implicitly convert to anything.

The change in read_formatted_entries relates to the fact that doing
the seemingly simpler:

 -       uintp = &fe.dir_index;
 +       uintp = (unsigned int *) &fe.dir_index;

would be undefined C/C++.  So to address that, I made the function
extract the form before assigning to the file_entry.  It felt natural
to use gdb::optional for "do I have this value", and this is what
motivated the previous patch that added the missing observer methods
to gdb::optional.

gdb/ChangeLog:
2017-04-04  Pedro Alves  <palves@redhat.com>

* common/underlying.h: New file.
* dwarf2read.c: Include "common/gdb_optional.h" and
"common/underlying.h".
(dir_index, file_name_index): New types.
(file_entry): Use them.
(file_entry::include): Use to_underlying.
(line_header::add_file_name): Use dir_index.
(read_formatted_entries): Use gdb::optional.  Read form before
writting to file_entry.
(dwarf_decode_line_header): Use dir_index.
(lnp_state_machine::current_file): Use to_underlying.
(lnp_state_machine::file): Change type to file_name_index.
(dwarf_record_line): Use to_underlying.
(init_lnp_state_machine): Use file_name_index.
(dwarf_decode_lines_1): Use dir_index and file_name_index.

gdb::optional: Add observers

Currently, gdb::optional is really minimal and can only be used for
lazy initialization.  There's no way to get at the value contained
inside the optinal.  This commit corrects that, by adding observer
methods, mostly copied from libstdc++'s implementation of C++17
std::optional.

This will be used in the following patch.

gdb/ChangeLog:
2017-04-04  Pedro Alves  <palves@redhat.com>

* common/gdb_optional.h (gdb::optiona): Add operator->, operator*,
operator bool, has_value and get methods.

dwarf2read.c: Some C++fycation, use std::vector, std::unique_ptr

This starts off as replacing a couple custom open coded vector
implementations in the file with std::vector, and then the rest falls
off of that.  I.e., use new/delete instead of XCNEW/xfree, add
ctors/dtors/initializers where appropriate.  And then use
std::unique_ptr instead of cleanups.  Some functions became methods,
and in a couple spots, some single-use callback functions that would
have to be tweaked anyway are converted to lambdas instead.

gdb/ChangeLog:
2017-04-04  Pedro Alves  <palves@redhat.com>

* dwarf2read.c (struct file_entry): Add ctors, and initialize all
fields.
(line_header): Initialize all data fields.  Change type of
standard_opcode_lengths to std::unique_ptr<unsigned char[]>.
Change type of include_dirs to std::vector<const char *>.  Remove
num_include_dirs, include_dirs_size.  Change type of file_names to
std::vector<file_entry>.  Remove num_file_names, file_names_size.
(line_header::line_header): New.
(line_header::add_include_dir, line_header::add_file_name): New
methods.
(line_header::include_dir_at): Remove NULL check.
(line_header::file_name_at): Add const overload.
(line_header_up): New unique_ptr typedef.
(dw2_get_file_names_reader): Use line_header_up.  Adjust to use
std::vector.  Remove free_line_header call.
(dwarf2_build_include_psymtabs): Use line_header_up.  Remove
free_line_header call.
(free_cu_line_header): Delete.
(handle_DW_AT_stmt_list, handle_DW_AT_stmt_list)
(setup_type_unit_groups): Use line_header_up instead of cleanups.
Adjust to use std::vector.
(free_line_header): Delete.
(free_line_header_voidp): Use delete.
(add_include_dir): Replace with ...
(line_header::add_include_dir): ... this method.  Use std::vector.
(add_file_name): Replace with ...
(line_header::add_file_name): ... this method.  Use std::vector.
(add_include_dir_stub): Delete.
(read_formatted_entries): Remove memset.
(dwarf_decode_line_header): Return a line_header_up instead of a
raw pointer.  Remove cleanup handling.  Pass lambdas to
read_formatted_entries.  Adjust to use line_header methods.
(dwarf_decode_lines_1): Adjust to use line_header methods.
(dwarf_decode_lines, file_file_name, file_full_name): Adjust to
use std::vector.

Use symbolic constants instead of magic numbers with add_data_entry()

Use symbolic constants from include/coff/internal.h instead of magic numbers
with add_data_entry()

bfd/ChangeLog:

2017-04-02  Jon Turney  <jon.turney@dronecode.org.uk>

(_bfd_XXi_swap_aouthdr_out): For clarity, use defines rather than
numbers for DataDirectory entry indicies passed to
add_data_entry().

Support ELF SHF_GNU_MBIND and PT_GNU_MBIND_XXX

Mark an ALLOC section, which should be placed in special memory area,
with SHF_GNU_MBIND.  Its sh_info field indicates the special memory
type.  GNU_MBIND section names start with ".mbind" so that they are
placed as orphan sections by linker.  All input GNU_MBIND sections
with the same sh_type, sh_flags and sh_info are placed in one output
GNU_MBIND section.  In executable and shared object, create a
GNU_MBIND segment for each GNU_MBIND section and its segment type is
PT_GNU_MBIND_LO plus the sh_info value.  Each GNU_MBIND segment is
aligned at page boundary.

The assembler syntax:

    .section .mbind.foo,"adx",%progbits
                          ^             0: Special memory type.
                          |
                         'd' for SHF_GNU_MBIND.

    .section .mbind.foo,"adx",%progbits,0x1
                          ^             1: Special memory type.
                          |
                         'd' for SHF_GNU_MBIND.

    .section .mbind.bar,"adG",%progbits,.foo_group,comdat,0x2
                          ^                               2: Special memory type.
                          |
                         'd' for SHF_GNU_MBIND.

bfd/

* elf.c (get_program_header_size): Add a GNU_MBIND segment for
each GNU_MBIND section and align GNU_MBIND section to page size.
(_bfd_elf_map_sections_to_segments): Create a GNU_MBIND
segment for each GNU_MBIND section.
(_bfd_elf_init_private_section_data): Copy sh_info from input
for GNU_MBIND section.

binutils/

* NEWS: Mention support for ELF SHF_GNU_MBIND and
PT_GNU_MBIND_XXX.
* readelf.c (get_segment_type): Handle PT_GNU_MBIND_XXX.
(get_elf_section_flags): Handle SHF_GNU_MBIND.
(process_section_headers): Likewise.
* testsuite/binutils-all/mbind1.s: New file.
* testsuite/binutils-all/objcopy.exp: Run readelf test on
mbind1.s.

gas/

* NEWS: Mention support for ELF SHF_GNU_MBIND.
* config/obj-elf.c (section_match): New.
(get_section): Match both sh_info and group name.
(obj_elf_change_section): Add argument for sh_info.  Pass both
sh_info and group name to get_section. Issue an error for
SHF_GNU_MBIND section without SHF_ALLOC.  Set sh_info.
(obj_elf_parse_section_letters): Set SHF_GNU_MBIND for 'd'.
(obj_elf_section): Support SHF_GNU_MBIND section info.
* config/obj-elf.h (obj_elf_change_section): Add argument for
sh_info.
* config/tc-arm.c (start_unwind_section): Pass 0 as sh_info to
obj_elf_change_section.
* config/tc-ia64.c (obj_elf_vms_common): Likewise.
* config/tc-microblaze.c (microblaze_s_data): Likewise.
(microblaze_s_sdata): Likewise.
(microblaze_s_rdata): Likewise.
(microblaze_s_bss): Likewise.
* config/tc-mips.c (s_change_section): Likewise.
* config/tc-msp430.c (msp430_profiler): Likewise.
* config/tc-rx.c (parse_rx_section): Likewise.
* config/tc-tic6x.c (tic6x_start_unwind_section): Likewise.
* doc/as.texinfo: Document 'd' for SHF_GNU_MBIND.
* testsuite/gas/elf/elf.exp: Run section12a, section12b and
section13.
* testsuite/gas/elf/section10.d: Updated.
* testsuite/gas/elf/section10.s: Likewise.
* testsuite/gas/elf/section12.s: New file.
* testsuite/gas/elf/section12a.d: Likewise.
* testsuite/gas/elf/section12b.d: Likewise.
* testsuite/gas/elf/section13.l: Likewise.
* testsuite/gas/elf/section13.d: Likewise.
* testsuite/gas/elf/section13.s: Likewise.

include/

* elf/common.h (PT_GNU_MBIND_NUM): New.
(PT_GNU_MBIND_LO): Likewise.
(PT_GNU_MBIND_HI): Likewise.
(SHF_GNU_MBIND): Likewise.

ld/

* NEWS: Mention support for ELF SHF_GNU_MBIND and
PT_GNU_MBIND_XXX.
* emultempl/elf32.em (gld${EMULATION_NAME}_place_orphan): Place
input GNU_MBIND sections with the same type, attributes and
sh_info field into a single output GNU_MBIND section.
* testsuite/ld-elf/elf.exp: Run mbind2a and mbind2b.
* testsuite/ld-elf/mbind1.s: New file.
* testsuite/ld-elf/mbind1a.d: Likewise.
* testsuite/ld-elf/mbind1b.d: Likewise.
* testsuite/ld-elf/mbind1c.d: Likewise.
* testsuite/ld-elf/mbind2a.s: Likewise.
* testsuite/ld-elf/mbind2b.c: Likewise.

RISC-V: Resurrect GP-relative disassembly hints

We missed a "_gp" when changing the GP symbol.  To make sure that
doesn't happen again, we now use the same definition everywhere (thanks,
Nick).

include/ChangeLog:

2017-04-03  Palmer Dabbelt  <palmer@dabbelt.com>

        * elf/riscv.h (RISCV_GP_SYMBOL): New define.

bfd/ChangeLog:

2017-04-03  Palmer Dabbelt  <palmer@dabbelt.com>

        * elfnn-riscv.c (GP_NAME): Delete.
        (riscv_global_pointer_value): Change GP_NAME to RISCV_GP_SYMBOL.
        (_bfd_riscv_relax_lui): Likewise.

opcodes/ChangeLog:

2017-04-03  Palmer Dabbelt  <palmer@dabbelt.com>

        * riscv-dis.c (riscv_disassemble_insn): Change "_gp" to
        RISCV_GP_SYMBOL.

gdbserver: Clear .deps on clean

In some situations, the dependency tracking files in .deps can refer to
source files that were removed or renamed, leading to errors like:

  make: *** No rule to make target `version.c', needed by `version.o'. Stop.

This patch makes the clean target clear the .deps directory, which gives
the user a chance to recover from the error wihtout knowing about the
internals of the build system.

It is already done for GDB.  See here for more details:

  https://sourceware.org/ml/gdb-patches/2009-03/msg00000.html

gdb/gdbserver/ChangeLog:

* Makefile.in (clean): Clear .deps.

remote.c: Use ptid_t instead of struct ptid

It's the only place in the codebase that uses "struct ptid", so change
it to ptid_t for consistency.

gdb/ChangeLog:

* remote.c (set_general_thread, set_continue_thread): Use ptid_t
instead of struct ptid.

Fix null pointer dereferences when using a link built with clang.

PR binutils/21342
* elflink.c (_bfd_elf_define_linkage_sym): Prevent null pointer
dereference.
(bfd_elf_final_link): Only initialize the extended symbol index
section if there are extended symbol tables to list.

Remove MAX_REGISTER_SIZE from frame.c

gdb/
* frame.c (get_frame_register_bytes): Unwind using value.
(put_frame_register_bytes): Likewise.

Automatic date update in version.in

RISC-V: Avoid a const warning

2017-04-03  Palmer Dabbelt  <palmer@dabbelt.com>

       * config/tc-riscv.c (riscv_clear_subsets): Cast argument to free to
       avoid const warnings.

IA16 support

       * config.sub: Handle ia16 in $basic_machine.

       bfd/
       * config.bfd: Handle ia16.

       gas/
       * configure.tgt: Handle ia16.

       ld/
       * configure.tgt: Handle ia16.

Correct the ChangeLog entry

ld: Support ELF GNU program properties

From .note.gnu.property section in each ELF input, we build a list of
GNU properties if .note.gnu.property section isn't corrupt.  The unknown
properties are ignored.  All property lists in relocatable inputs are
merged into an output property list.  When -z stack-size=N is used and
N isn't 0, the GNU_PROPERTY_STACK_SIZE property will be merged with or
added to the output property list.  .note.gnu.property section is
generated in output from the output property list.

bfd/

* Makefile.am (BFD32_BACKENDS): Add elf-properties.lo.
(BFD32_BACKENDS_CFILES): Add elf-properties.c.
* configure.ac (elf): Add elf-properties.lo.
* Makefile.in: Regenerated.
* configure: Likewise.
* elf-bfd.h (elf_property_kind): New.
(elf_property): Likewise.
(elf_property_list): Likewise.
(elf_properties): Likewise.
(_bfd_elf_parse_gnu_properties): Likewise.
(_bfd_elf_get_property): Likewise.
(_bfd_elf_link_setup_gnu_properties): Likewise.
(elf_backend_data): Add parse_gnu_properties, merge_gnu_properties
and setup_gnu_properties.
(elf_obj_tdata): Add properties.
* elf-properties.c: New file.
* elf32-i386.c (elf_i386_parse_gnu_properties): New.
(elf_i386_merge_gnu_properties): Likewise.
(elf_backend_parse_gnu_properties): Likewise.
(elf_backend_merge_gnu_properties): Likewise.
* elf64-x86-64.c (elf_x86_64_parse_gnu_properties): Likewise.
(elf_x86_64_merge_gnu_properties): Likewise.
(elf_backend_parse_gnu_properties): Likewise.
(elf_backend_merge_gnu_properties): Likewise.
* elfxx-target.h (elf_backend_merge_gnu_properties): Likewise.
(elf_backend_parse_gnu_properties): Likewise.
(elf_backend_setup_gnu_properties): Likewise.
(elfNN_bed): Add elf_backend_parse_gnu_properties,
elf_backend_merge_gnu_properties and
elf_backend_setup_gnu_properties.

ld/

* ld/NEWS: Mention support for ELF GNU program properties.
* emultempl/elf32.em (gld${EMULATION_NAME}_after_open): Call
ELF setup_gnu_properties.
* testsuite/ld-i386/i386.exp: Run property tests for Linux/i386.
* testsuite/ld-i386/pass.c: New file.
* testsuite/ld-i386/property-1.r: Likewise.
* testsuite/ld-i386/property-2.r: Likewise.
* testsuite/ld-i386/property-3.r: Likewise.
* testsuite/ld-i386/property-4.r: Likewise.
* testsuite/ld-i386/property-5.r: Likewise.
* testsuite/ld-i386/property-6.r: Likewise.
* testsuite/ld-i386/property-6a.c: Likewise.
* testsuite/ld-i386/property-6b.c: Likewise.
* testsuite/ld-i386/property-6c.S: Likewise.
* testsuite/ld-i386/property-7.r: Likewise.
* testsuite/ld-i386/property-no-copy.S: Likewise.
* testsuite/ld-i386/property-stack.S: Likewise.
* testsuite/ld-i386/property-unsorted-1.S: Likewise.
* testsuite/ld-i386/property-unsorted-2.S: Likewise.
* testsuite/ld-i386/property-x86-1.S: Likewise.
* testsuite/ld-i386/property-x86-2.S: Likewise.
* testsuite/ld-x86-64/pass.c: Likewise.
* testsuite/ld-x86-64/property-1.r: Likewise.
* testsuite/ld-x86-64/property-2.r: Likewise.
* testsuite/ld-x86-64/property-3.r: Likewise.
* testsuite/ld-x86-64/property-4.r: Likewise.
* testsuite/ld-x86-64/property-5.r: Likewise.
* testsuite/ld-x86-64/property-6.r: Likewise.
* testsuite/ld-x86-64/property-6a.c: Likewise.
* testsuite/ld-x86-64/property-6b.c: Likewise.
* testsuite/ld-x86-64/property-6c.S: Likewise.
* testsuite/ld-x86-64/property-7.r: Likewise.
* testsuite/ld-x86-64/property-no-copy.S: Likewise.
* testsuite/ld-x86-64/property-stack.S: Likewise.
* testsuite/ld-x86-64/property-unsorted-1.S: Likewise.
* testsuite/ld-x86-64/property-unsorted-2.S: Likewise.
* testsuite/ld-x86-64/property-x86-1.S: Likewise.
* testsuite/ld-x86-64/property-x86-2.S: Likewise.
* testsuite/ld-x86-64/x86-64.exp: Run property tests for
Linux/x86-64.

readelf: Fix overlarge memory allocation when reading a binary with an excessive number of program headers.

PR binutils/21345
* readelf.c (get_program_headers): Check for there being too many
program headers before attempting to allocate space for them.

Fix runtime seg-fault in readelf when parsing a corrupt MIPS binary.

PR binutils/21344
* readelf.c (process_mips_specific): Check for an out of range GOT
entry before reading the module pointer.

readelf: Update check for invalid word offsets in ARM unwind information.

PR binutils/21343
* readelf.c (get_unwind_section_word): Fix snafu checking for
invalid word offsets in ARM unwind information.

Automatic date update in version.in

Automatic date update in version.in

Automatic date update in version.in

RISC-V: Allow ISA subsets to be disabled

Without this patch, passing "-march=rv64ic -march=rv64i" results in
you getting a "RV64IC" toolchain, which isn't expected.

gas/ChangeLog:

2017-03-30  Palmer Dabbelt  <palmer@dabbelt.com>

       * config/tc-riscv.c (riscv_clear_subsets): New function.
       (riscv_add_subset): Call riscv_clear_subsets and riscv_set_rvc to
       clear RVC when it's been previously set.

RISC-V: Add physical memory protection CSRs

2017-03-27  Andrew Waterman  <andrew@sifive.com>

       * opcode/riscv-opc.h (CSR_PMPCFG0): New define.
       (CSR_PMPCFG1): Likewise.
       (CSR_PMPCFG2): Likewise.
       (CSR_PMPCFG3): Likewise.
       (CSR_PMPADDR0): Likewise.
       (CSR_PMPADDR1): Likewise.
       (CSR_PMPADDR2): Likewise.
       (CSR_PMPADDR3): Likewise.
       (CSR_PMPADDR4): Likewise.
       (CSR_PMPADDR5): Likewise.
       (CSR_PMPADDR6): Likewise.
       (CSR_PMPADDR7): Likewise.
       (CSR_PMPADDR8): Likewise.
       (CSR_PMPADDR9): Likewise.
       (CSR_PMPADDR10): Likewise.
       (CSR_PMPADDR11): Likewise.
       (CSR_PMPADDR12): Likewise.
       (CSR_PMPADDR13): Likewise.
       (CSR_PMPADDR14): Likewise.
       (CSR_PMPADDR15): Likewise.
       (pmpcfg0): Declare register.
       (pmpcfg1): Likewise.
       (pmpcfg2): Likewise.
       (pmpcfg3): Likewise.
       (pmpaddr0): Likewise.
       (pmpaddr1): Likewise.
       (pmpaddr2): Likewise.
       (pmpaddr3): Likewise.
       (pmpaddr4): Likewise.
       (pmpaddr5): Likewise.
       (pmpaddr6): Likewise.
       (pmpaddr7): Likewise.
       (pmpaddr8): Likewise.
       (pmpaddr9): Likewise.
       (pmpaddr10): Likewise.
       (pmpaddr11): Likewise.
       (pmpaddr12): Likewise.
       (pmpaddr13): Likewise.
       (pmpaddr14): Likewise.
       (pmpaddr15): Likewise.

gdbserver: Suffix generated C files with -generated

I noticed that there were some missing files in gdbserver's gitignore
(some generated register format .c files).  Of course the easy fix would
be to add those files to .gitignore, but I think we can do a better job,
so that we don't have to worry about adding generated files to
.gitignore or the clean Makefile target.

I suggest naming all generated source files -generated.c.  This way, we
can use a single rule in .gitignore and do a "rm -f *-generated.c" to
clean them up.

New in v2:

  - Don't rename version.o and xml-builtin.o

gdb/gdbserver/ChangeLog:

* .gitignore: Remove generated files, replace with wildcard.
* (clean): Replace removal of generated files with wildcard.
(version.c): Replace with...
(version-generated.c): ...this.
(xml-builtin.c): Replace with...
(xml-builtin-generated.c): ...this.
(%-ipa.o: %-generated.c, %.o: %-generated.c): New rules.
(%.c: *regformats*): Replace with...
(%-generated.c: *regformats*): ...this.

Fix minor problems with previous wasm32 binutils commit.

binutils * NEWS: Use "WebAssembly" consistently.
* testsuite/binutils-all/wasm32/wasm32.exp: Fix copyright notice.

Reduce the size of s390 symbol tables by allowing relocations in mergeable string sections (eg .debug_str) to be made section relative rather than symbol relative.

PR gas/21333
* config/tc-s390.c (tc_s390_fix_adjustable): Allow non pc-relative
fixups in mergeable sections to be adjusted.