Alan Modra [Wed, 6 Feb 2019 06:56:23 +0000 (17:26 +1030)]
Fix some ldscripts/pr24008 fails
These targets were all failing due to extra symbols.
pdp11-dec-aout +FAIL: ld-scripts/pr24008
powerpc-aix5.1 +FAIL: ld-scripts/pr24008
powerpc-aix5.2 +FAIL: ld-scripts/pr24008
rs6000-aix4.3.3 +FAIL: ld-scripts/pr24008
rs6000-aix5.1 +FAIL: ld-scripts/pr24008
rs6000-aix5.2 +FAIL: ld-scripts/pr24008
Some fails remain, those I saw were segfaults or assertion fails that
indicate target bugs.
PR ld/24008
* testsuite/ld-scripts/pr24008.d: Pass with extra target
defined symbols.
removed check R_X86_64_PC32 relocation against protected symbols in
shared objects. Since elf_x86_64_check_relocs is called after we
have seen all input files, we can check for PC-relative relocations in
elf_x86_64_check_relocs. We should not allow PC-relative relocations
against protected symbols since address of protected function and
location of protected data may not be in the shared object.
bfd/
PR ld/24151
* elf64-x86-64.c (elf_x86_64_need_pic): Check
SYMBOL_DEFINED_NON_SHARED_P instead of def_regular.
(elf_x86_64_relocate_section): Move PIC check for PC-relative
relocations to ...
(elf_x86_64_check_relocs): Here.
(elf_x86_64_finish_dynamic_symbol): Use SYMBOL_DEFINED_NON_SHARED_P
to check if a symbol is defined in a non-shared object.
* elfxx-x86.h (SYMBOL_DEFINED_NON_SHARED_P): New.
ld/
PR ld/24151
* testsuite/ld-x86-64/pr24151a-x32.d: New file.
* testsuite/ld-x86-64/pr24151a.d: Likewise.
* testsuite/ld-x86-64/pr24151a.s: Likewise.
* testsuite/ld-x86-64/x86-64.exp: Run pr24151a and pr24151a-x32.
John Darrington [Fri, 1 Feb 2019 16:42:54 +0000 (17:42 +0100)]
S12Z: GAS: Allow #_symbol operands as mov source
mov.l, mov.p and mov.w (but not mov.b) when called with an immediate source
operand should be accepted a relocatable expression. This change makes that
possible.
gas/
* config/tc-s12z.c (lex_imm): Add new argument exp_o.
(emit_reloc): New function.
(md_apply_fix): [BFD_RELOC_S12Z_OPR] Recognise that it
can be either 2 bytes or 3 bytes long.
* testsuite/gas/s12z/mov-imm-reloc.d: New file.
* testsuite/gas/s12z/mov-imm-reloc.s: New file.
* testsuite/gas/s12z/s12z.exp: Add them.
John Darrington [Fri, 1 Feb 2019 16:42:54 +0000 (17:42 +0100)]
S12Z: GAS: Issue warning if TFR/EXG have identical source and destination.
It is permissible for the source and destination operands of TFR and EXG to be
the same register. However it is a pointless instruction and anyone writing it
has probably made a mistake. This change emits a warning if such an instruction
is encountered.
gas/
* config/tc-s12z.c (tfr): Emit warning if operands are the same.
* testsuite/gas/s12z/exg.d: New test case.
* testsuite/gas/s12z/exg.l: New file.
The assembler permitted instructions which attempted to assign to an immediate
operand. Bizarrely there is a valid machine code for such operations (although
the documentation says it's "inappropriate"). This change causes such attempts
to fail with an error message.
gas/
* config/tc-s12z.c (lex_opr): Add a parameter to indicate whether
immediate mode operands should be permitted.
* testsuite/s12z/imm-dest.d: New file.
* testsuite/s12z/imm-dest.l: New file.
* testsuite/s12z/imm-dest.s: New file.
* testsuite/s12z/s12z.exp: Add them.
Andreas Krebbel [Thu, 31 Jan 2019 16:01:27 +0000 (17:01 +0100)]
S/390: Implement instruction set extensions
opcodes/ChangeLog:
2019-01-31 Andreas Krebbel <krebbel@linux.ibm.com>
* s390-mkopc.c (main): Accept arch13 as cpu string.
* s390-opc.c: Add new instruction formats and instruction opcode
masks.
* s390-opc.txt: Add new arch13 instructions.
include/ChangeLog:
2019-01-31 Andreas Krebbel <krebbel@linux.ibm.com>
2019-01-31 Andreas Krebbel <krebbel@linux.ibm.com>
* config/tc-s390.c (s390_parse_cpu): New entry for arch13.
* doc/c-s390.texi: Document arch13 march option.
* testsuite/gas/s390/s390.exp: Run the arch13 related tests.
* testsuite/gas/s390/zarch-arch13.d: New test.
* testsuite/gas/s390/zarch-arch13.s: New test.
* testsuite/gas/s390/zarch-z13.d: Expect the renamed mnemonics
also for z13.
Alan Modra [Thu, 31 Jan 2019 04:08:45 +0000 (14:38 +1030)]
Assorted warning fixes
gcc-9 flagged warnings at the places I'm patching here, all real bugs.
* config/tc-alpha.c (md_apply_fix): Correct range checks for
BFD_RELOC_ALPHA_NOP, BFD_RELOC_ALPHA_LDA, BFD_RELOC_ALPHA_BSR.
* config/tc-arm.c (md_apply_fix): Use llabs rather than abs.
* config/tc-csky.c (get_macro_reg_vals): Pass s to csky_show_error.
Szabolcs Nagy [Fri, 14 Dec 2018 14:02:54 +0000 (14:02 +0000)]
[PR gdb/23985] Fix libinproctrace.so build
The IPA objects currently may use gnulib replacement apis, which is
wrong: gnulib is not linked into the produced dso and it cannot be
because it is not built with -fPIC -fvisibility=hidden.
The gnulib replacement detection is broken under cross compilation:
for targets other than *-gnu*, replacements are enabled that depend
on execution time detection. This causes unnecessary build failure
when the target has proper support for the replaced api.
This fix tries to undo the replacements, which is tricky because the
gnulib headers are still used for various compile time fixups and
there is no simple knob in gnulib to only turn the replacements off.
Without this workaround gdb fails to cross build to non-gnu targets:
ld: tracepoint-ipa.o: in function `gdb_agent_helper_thread(void*)':
gdb/gdbserver/tracepoint.c:7221: undefined reference to `rpl_strerror'
...
Makefile:434: recipe for target 'libinproctrace.so' failed
Tom Tromey [Sat, 8 Sep 2018 02:02:21 +0000 (20:02 -0600)]
Release the GIL while running a gdb command or expression
PR python/23615 points out that gdb.execute_gdb_command does not
release the Python GIL. This means that, while the gdb command is
running, other Python threads do not run.
This patch solves the problem by introducing a new RAII class that can
be used to temporarily release and then re-acquire the GIL, then puts
this into the appropriate places in execute_gdb_command and
gdbpy_parse_and_eval.
This does not include a test case, because after some research I could
not find a way to write one that was not racy.
gdb/ChangeLog
2019-01-30 Tom Tromey <tom@tromey.com>
PR python/23615:
* python/python.c (execute_gdb_command): Use gdbpy_allow_threads.
(gdbpy_parse_and_eval): Likewise.
* python/python-internal.h (gdbpy_allow_threads): New class.
Max Filippov [Sat, 26 Jan 2019 02:52:32 +0000 (18:52 -0800)]
xtensa: gas: don't keep relocations for constants
xtensa gas chokes on 8/16 bit data entries representing constant symbols
because it leaves BFD_RELOC_8/BFD_RELOC_16 fixups for which xtensa BFD
cannot emit relocations. Resolve fixups for constant symbols in
md_apply_fix.
gas/
2019-01-28 Max Filippov <jcmvbkbc@gmail.com>
* config/tc-xtensa.c (md_apply_fix): Mark fixups for constant
symbols as done in md_apply_fix.
* testsuite/gas/all/forward.d: Don't XFAIL for xtensa.
John Baldwin [Mon, 28 Jan 2019 18:16:58 +0000 (10:16 -0800)]
Use trad_frame_set_reg_addr for FreeBSD arm signal trampoline unwinders.
Replace individual calls to trad_frame_set_reg_addr for the general
purpose and floating point registers in signal trampoline frames used
by FreeBSD/aarch64 and FreeBSD/arm with calls to
trad_frame_set_reg_addr using the register maps for the corresponding
register sets.
gdb/ChangeLog:
* aarch64-fbsd-tdep.c (aarch64_fbsd_gregmap)
(aarch64_fbsd_fpregmap): Move earlier.
(AARCH64_MCONTEXT_REG_SIZE, AARCH64_MCONTEXT_FPREG_SIZE): Delete.
(aarch64_fbsd_sigframe_init): Use trad_frame_set_reg_regmap
instead of individual calls to trad_frame_set_reg_addr.
* arm-fbsd-tdep.c (arm_fbsd_gregmap, arm_fbsd_vfpregmap): Move
earlier.
(ARM_MCONTEXT_REG_SIZE, ARM_MCONTEXT_VFP_REG_SIZE): Delete.
(arm_fbsd_sigframe_init): Use trad_frame_set_reg_regmap
instead of individual calls to trad_frame_set_reg_addr.
Fix GCC9 warning on elf32-arm.c:elf32_arm_final_link_relocate
Fedora Rawhide has just switched to GCC9, and now GDB doesn't compile
because of a BFD warning:
BUILDSTDERR: ../../bfd/elf32-arm.c: In function 'elf32_arm_final_link_relocate':
BUILDSTDERR: ../../bfd/elf32-arm.c:10907:10: error: absolute value function 'labs' given an argument of type 'bfd_signed_vma' {aka 'long long int'} but has parameter of type 'long int' which may cause truncation of value [-Werror=absolute-value]
BUILDSTDERR: 10907 | value = labs (relocation);
BUILDSTDERR: | ^~~~
The fix is (apparently) simple: instead of using 'labs', we should use
'llabs', since we're passing a 'bfd_signed_vma' to it, which is at
least a 'long long int', as far as I have checked. This is what this
patch does.
Alan Hayward [Mon, 28 Jan 2019 09:39:55 +0000 (09:39 +0000)]
Replace contribution list in CONTRIBUTE file with link
The GDB wiki page has a much better contribution checklist than
that in the GDB CONTRIBUTE file. In addition, the wiki is easier
to keep up to date with current processes.
Reduce the CONTRIBUTE file down to a short paragraph followed by
a link to the contribution process. This also ensures anyone
reading the CONTRIBUTE file for a given release has access to the
latest processes.
gdb/ChangeLog:
* CONTRIBUTE: Replace contribution list with wiki link.
Alan Hayward [Thu, 24 Jan 2019 18:55:20 +0000 (18:55 +0000)]
gdbserver: When attaching, add process before lwps
The recent BP/WP changes for AArch64 swapping the order in add_lwp()
so that the process was added before the lwp. This was due to the lwp
creation requiring the process data.
This also needs changing in linux_attach().
Fixes gdb.server/ext-attach.exp on Aarch64.
(This regression was hidden due to the racy nature of the gdb.server
tests - now they are no longer racy it'll be easier to spot. Also
checked X86).
gdb/gdbserver/ChangeLog:
2019-01-25 Alan Hayward <alan.hayward@arm.com>
* linux-low.c (linux_attach): Add process before lwp.
Alan Modra [Sun, 27 Jan 2019 23:29:29 +0000 (09:59 +1030)]
PR24008, Wrong value of ternary expression in map file
PR 24008
* ldexp.h (lang_phase_type): Add lang_fixed_phase_enum.
* ldexp.c (fold_name): Move expld.assign_name check later to
avoid an extra lookup.
(exp_fold_tree_1): When lang_fixed_phase_enum, don't change symbol
values, and don't clear expld.assign_name.
* ldlang.c (lang_map): Set expld.phase to lang_fixed_phase_enum.
(print_assignment): Resolve entire assignment expression.
Don't access symbol u.def unless symbol is defined.
Tom Tromey [Sat, 26 Jan 2019 15:51:57 +0000 (08:51 -0700)]
Fix the sim build
Simon pointed out that the "common/" include change in gdb broke the
sim build. The problem was that the sim was using gdb's
create-version.sh, which changed.
This patch copies create-version.sh to the sim, so that it can
generate "version.c" in a way that works in the sim build.
Tested by rebuilding.
sim/common/ChangeLog
2019-01-26 Tom Tromey <tom@tromey.com>
* Make-common.in (version.c): Use sim's create-version.sh.
* create-version.sh: New file.
sim/ppc/ChangeLog
2019-01-26 Tom Tromey <tom@tromey.com>
* Makefile.in (version.c): Use sim's create-version.sh.
Tom Tromey [Tue, 22 Jan 2019 07:26:45 +0000 (00:26 -0700)]
Don't use -I for common subdirectory
This changes the Makefiles to remove the -I for the common/
subdirectory. This will enforce the rule that includes must use the
'common/filename.h' form.
gdb/ChangeLog
2019-01-25 Tom Tromey <tom@tromey.com>
* Makefile.in (GDB_CFLAGS): Don't add -I for common.
gdb/gdbserver/ChangeLog
2019-01-25 Tom Tromey <tom@tromey.com>
* Makefile.in (INCLUDE_CFLAGS): Don't add -I for common.
Sudi Das [Fri, 25 Jan 2019 14:28:07 +0000 (14:28 +0000)]
AArch64: Update encodings for stg, st2g, stzg and st2zg.
This patch is part of a series of patches to introduce a few changes to the
Armv8.5-A Memory Tagging Extension. This patch updates the st*g instructions
to use a previously reserved field for a new register operand. Thus the
new versions of the instructions are as follows:
Sudi Das [Fri, 25 Jan 2019 14:15:45 +0000 (14:15 +0000)]
AArch64: Add new STZGM instruction for Armv8.5-A Memory Tagging Extension.
This patch is part of a series of patches to introduce a few changes to the
Armv8.5-A Memory Tagging Extension. This patch adds the new STZGM instruction.
STGZM Xt, [<Xn|SP>]
Committed on behalf of Sudakshina Das.
*** gas/ChangeLog ***
* testsuite/gas/aarch64/armv8_5-a-memtag.d: New tests for stzgm.
* testsuite/gas/aarch64/armv8_5-a-memtag.s: Likewise.
* testsuite/gas/aarch64/illegal-memtag.l: Likewise.
* testsuite/gas/aarch64/illegal-memtag.s: Likewise.
Sudi Das [Fri, 25 Jan 2019 13:57:14 +0000 (13:57 +0000)]
AArch64: Remove ldgv and stgv instructions from Armv8.5-A Memory Tagging Extension.
This patch is part of a series of patches to introduce a few changes to the
Armv8.5-A Memory Tagging Extension. This patch removes the LDGV and STGV
instructions. These instructions needed special infrastructure to support
[base]! style for addressing mode. That is also removed now.
Committed on behalf of Sudakshina Das.
*** gas/ChangeLog ***
* config/tc-aarch64.c (parse_address_main): Remove support for
[base]! address expression.
(parse_operands): Remove support for AARCH64_OPND_ADDR_SIMPLE_2.
(warn_unpredictable_ldst): Remove support for ldstgv_indexed.
* testsuite/gas/aarch64/armv8_5-a-memtag.d: Remove tests for ldgv
and stgv.
* testsuite/gas/aarch64/armv8_5-a-memtag.s: Likewise.
* testsuite/gas/aarch64/illegal-memtag.l: Likewise.
* testsuite/gas/aarch64/illegal-memtag.s: Likewise.
Nick Clifton [Fri, 25 Jan 2019 13:16:06 +0000 (13:16 +0000)]
Prevent a potential illegal memory access in readelf when parsing a note with a zero name size.
PR 24131
* readelf.c (process_notes_at): Prevent an illegal memory access
when the note's namesize is zero.
(decode_tic6x_unwind_bytecode): Add code to handle the case where
no registers are specified in a frame pop instruction.
Pedro Alves [Thu, 24 Jan 2019 18:25:06 +0000 (18:25 +0000)]
target_pass_signals/target_program_signals: Use gdb::array_view
This replaces the pointer and length parameters of target_pass_signals
and target_program_signals with a gdb::array_view parameter, and fixes
the fallout.
In infrun.c, the signal_stop, signal_print, signal_program,
signal_catch, signal_pass globals are currently pointers to
heap-allocated memory. I see no point in that, so I converted them to
arrays. This allows simplifying the calls to
target_pass_signals/target_program_signals, since we can pass the
array directly, which can implicitly convert to gdb::array_view.
gdb/ChangeLog:
2019-01-24 Pedro Alves <palves@redhat.com>
Pedro Alves [Thu, 24 Jan 2019 18:01:49 +0000 (18:01 +0000)]
Fix clang/libc++ build
This fixes the following build error with clang/libc++, reported at
<https://sourceware.org/ml/gdb-patches/2019-01/msg00537.html>:
(...)
In file included from breakpoint.c:34:
In file included from ./inferior.h:54:
./common/forward-scope-exit.h:98:7: error: no matching constructor for
initialization of 'decltype(std::bind(&delete_longjmp_breakpoint,
std::declval<int>()))' (aka '__bind<void (*)(int), int>')
: m_bind_function (std::bind (function, args...))
^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
./common/gdb_optional.h:155:19: note: in instantiation of member
function 'detail::forward_scope_exit<void (int),
&delete_longjmp_breakpoint, void (int)>::forward_scope_exit' requested
here
new (&m_item) T (std::forward<Args>(args)...);
^
breakpoint.c:11127:18: note: in instantiation of function template
specialization 'gdb::optional<detail::forward_scope_exit<void (int),
&delete_longjmp_breakpoint, void (int)> >::emplace<int &>' requested
here
lj_deleter.emplace (thread);
^
/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/include/c++/v1/functional:2220:7:
note: candidate constructor (the implicit copy constructor) not
viable: no known conversion from '__bind<[...], int &>' to 'const
__bind<[...], int>' for 1st argument
class __bind
^
(...)
I don't really know why I ended up with a copy here. We can just pass
the arguments directly to the being-constructed bind.
gdb/ChangeLog:
2019-01-24 Pedro Alves <palves@redhat.com>
* common/forward-scope-exit.h
(forward_scope_exit::forward_scope_exit): Pass arguments to
m_bind_function directly, instead of creating a std::bind and
copying that.
Alan Hayward [Thu, 24 Jan 2019 08:17:39 +0000 (08:17 +0000)]
AArch64 AAPCS: Ignore static members
Static members in C++ structs are global data and therefore not part of the
list of struct members considered for passing in registers.
Note the corresponding code in GCC (from which the GDB AAPCS code is based)
does not have any static member checks due to the static members not being
part of the struct type at that point.
Extend gdb.base/infcall-nested-structs.exp to test structs with static
members when compiled for C++. XFAIL more cases for x86_64 (see gdb/24104).
For completeness, ensure some test cases have both empty structures and
static members.
Also fixes gdb.dwarf2/dw2-cp-infcall-ref-static.exp.
gdb/ChangeLog:
* aarch64-tdep.c (aapcs_is_vfp_call_or_return_candidate_1): Check
for static members.
(pass_in_v_vfp_candidate): Likewise.
Tom Tromey [Wed, 23 Jan 2019 18:58:37 +0000 (18:58 +0000)]
Use scope_exit in regcache.c
This removes the regcache_invalidator class in favor of a scope_exit.
This seems like an improvement (albeit a minor one) because
regcache_invalidator is only used in a single spot.
gdb/ChangeLog:
2019-01-23 Tom Tromey <tom@tromey.com>
Pedro Alves <palves@redhat.com>
* regcache.c (class regcache_invalidator): Remove.
(regcache::raw_write): Use make_scope_exit.
Tom Tromey [Wed, 23 Jan 2019 18:58:34 +0000 (18:58 +0000)]
Remove clear_symtab_users_cleanup
This removes clear_symtab_users_cleanup, replacing it with uses of
forward_scope_exit.
gdb/ChangeLog:
2019-01-23 Tom Tromey <tom@tromey.com>
Pedro Alves <palves@redhat.com>
* symfile.c: Include forward-scope-exit.h.
(clear_symtab_users_cleanup): Replace forward declaration with
a FORWARD_SCOPE_EXIT.
(syms_from_objfile_1): Use the forward_scope_exit and
gdb::optional instead of cleanup_function.
(reread_symbols): Use the forward_scope_exit instead of
cleanup_function.
(clear_symtab_users_cleanup): Remove function.
v3: sorted earlier in the series, and replaces the cleanup in
fetch_inferior_event here too instead of in some other patch.
This removes delete_just_stopped_threads_infrun_breakpoints_cleanup,
replacing it with uses of scope_exit.
gdb/ChangeLog:
2019-01-23 Tom Tromey <tom@tromey.com>
Pedro Alves <palves@redhat.com>
* infrun.c: Include "common/scope-exit.h"
(delete_just_stopped_threads_infrun_breakpoints_cleanup): Remove.
(wait_for_inferior): Use SCOPE_EXIT.
(fetch_inferior_event): Use scope_exit.
Pedro Alves [Wed, 23 Jan 2019 18:58:30 +0000 (18:58 +0000)]
Use forward_scope_exit for scoped_finish_thread_state
This reimplements the manually-written scoped_finish_thread_state
class as a forward_scope_exit instantiation. forward_scope_exit has
the same interface as scoped_finish_thread_state, so nothing else has
to change.
A forward_scope_exit is preferred over make_scope_exit here because
infrun.c:normal_stop needs to wrap scoped_finish_thread_state in a
gdb::optional. Since we need the type there, might as well use it
everywhere.
gdb/ChangeLog:
2019-01-23 Pedro Alves <palves@redhat.com>
Andrew Burgess <andrew.burgess@embecosm.com>
* gdbthread.h: Include "common/forward-scope-exit.h".
(scoped_finish_thread_state): Redefine custom class in terms of
forward_scope_exit.
Pedro Alves [Wed, 23 Jan 2019 18:58:29 +0000 (18:58 +0000)]
Introduce forward_scope_exit
This adds a template that can be used to automatically instantiate
scope_exit-like types that wrap some cleanup function. The
instantiated type has a ctor that has the same interface as the
wrapped function. While the "magic" is just straight C++11, the
intended use is via the FORWARD_SCOPE_EXIT macro, which is a minimal
macro that avoids spelling out the wrapped function name more than
once:
The above runs:
some_function (some_int, some_obj_ptr);
at scope exit.
This is mainly useful as opposed to a simpler SCOPE_EXIT when you need
to:
- cancel the scope_exit, in which case you need the object's name
- wrap the scope_exit in a gdb::optional, in which case you need the
scope_exit's type in advance.
More details in the code comments.
gdb/ChangeLog:
2019-01-23 Pedro Alves <palves@redhat.com>
Andrew Burgess <andrew.burgess@embecosm.com>
Pedro Alves [Wed, 23 Jan 2019 18:58:29 +0000 (18:58 +0000)]
Introduce scope_exit
This add a new template class scope_exit. scope_exit is a
general-purpose scope guard that calls its exit function at the end of
the current scope. A scope_exit may be canceled by calling the
"release" method. The API is modeled on P0052R5 - Generic Scope Guard
and RAII Wrapper for the Standard Library, which is itself based on
Andrej Alexandrescu's ScopeGuard/SCOPE_EXIT.
The main advantage of scope_exit is avoiding writing single-use RAII
classes and its boilerplate. Following patches will remove a few of
such classes.
There are two forms available:
- The "make_scope_exit" form allows canceling the scope guard. Use
it like this:
auto cleanup = make_scope_exit ( <function, function object, lambda> );
...
cleanup.release (); // cancel
- If you don't need to cancel the guard, you can use the SCOPE_EXIT
macro, like this:
SCOPE_EXIT { /* any code you like here. */ }
Note: scope_exit instances do not allocate anything on the heap.
gdb/ChangeLog:
2019-01-23 Pedro Alves <palves@redhat.com>
Andrew Burgess <andrew.burgess@embecosm.com>
Tom Tromey <tom@tromey.com>
Pedro Alves [Wed, 23 Jan 2019 18:58:28 +0000 (18:58 +0000)]
Rename ESC -> ESC_PARENS
A following patch will include common/preprocessor.h in some .c file
that also includes readline.h, and that revealed a conflict -- ESC is
defined by readline.h as well (actually readline's chardefs.h) with a
completely unrelated meaning:
#define ESC CTRL('[')
Rename our version to avoid the conflict.
gdb/ChangeLog:
2019-01-23 Pedro Alves <palves@redhat.com>
Tom Tromey [Fri, 18 Jan 2019 14:45:01 +0000 (07:45 -0700)]
Remove a warning from symtab.c
When building symtab.c, I get:
../../binutils-gdb/gdb/language.h: In function ‘void print_symbol_info(search_domain, symbol*, int, const char*)’:
../../binutils-gdb/gdb/language.h:738:20: warning: ‘*((void*)& l +4)’ may be used uninitialized in this function [-Wmaybe-uninitialized]
set_language (m_lang);
~~~~~~~~~~~~~^~~~~~~~
../../binutils-gdb/gdb/symtab.c:4613:41: note: ‘*((void*)& l +4)’ was declared here
scoped_switch_to_sym_language_if_auto l (sym);
^
This is another instance of the std::optional problem, see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80635.
However, it seemed straightforward and inexpensive to me to silence
this one, which is what this patch does.
gdb/ChangeLog
2019-01-23 Tom Tromey <tom@tromey.com>
* language.h (class scoped_switch_to_sym_language_if_auto):
Initialize m_lang in both cases.
Tom Tromey [Mon, 21 Jan 2019 22:18:50 +0000 (15:18 -0700)]
corelow.c does not need sys/file.h
I did not see any reason that corelow.c should include <sys/file.h>.
The provided explanatory comment seems to be wrong. This patch
removes the include.
Tom Tromey [Mon, 21 Jan 2019 21:50:24 +0000 (14:50 -0700)]
Include coff/sym.h from coff/ecoff.h
coff/ecoff.h refers to some names defined in coff/sym.h. Include the
latter from the former, so that users of the header don't need to know
this detail.
Tom Tromey [Mon, 21 Jan 2019 21:17:32 +0000 (14:17 -0700)]
Rename "lines" parameter in source-cache.h
A compile in the TUI somehow had "lines" defined as a macro. This
caused a compile error when including source-cache.h after whatever
header did that. I tracked this down to a #define in
/usr/include/term.h, so I just changed source-cache.h to avoid the
clash.
gdb/ChangeLog
2019-01-22 Tom Tromey <tom@tromey.com>
Tom Tromey [Mon, 21 Jan 2019 20:49:06 +0000 (13:49 -0700)]
Do not include py-ref.h in most files
py-ref.h can really only be included from a specific spot in
python-internal.h. The other includes are not useful, and cause
compilation errors if the includes are ever sorted. So, remove these
includes.
Arguably, py-ref.h should simply not be a separate header.
gdb/ChangeLog
2019-01-22 Tom Tromey <tom@tromey.com>
* python/py-arch.c: Do not include py-ref.h.
* python/py-bpevent.c: Do not include py-ref.h.
* python/py-cmd.c: Do not include py-ref.h.
* python/py-continueevent.c: Do not include py-ref.h.
* python/py-event.h: Do not include py-ref.h.
* python/py-evtregistry.c: Do not include py-ref.h.
* python/py-finishbreakpoint.c: Do not include py-ref.h.
* python/py-frame.c: Do not include py-ref.h.
* python/py-framefilter.c: Do not include py-ref.h.
* python/py-function.c: Do not include py-ref.h.
* python/py-infevents.c: Do not include py-ref.h.
* python/py-linetable.c: Do not include py-ref.h.
* python/py-objfile.c: Do not include py-ref.h.
* python/py-param.c: Do not include py-ref.h.
* python/py-prettyprint.c: Do not include py-ref.h.
* python/py-progspace.c: Do not include py-ref.h.
* python/py-symbol.c: Do not include py-ref.h.
* python/py-symtab.c: Do not include py-ref.h.
* python/py-type.c: Do not include py-ref.h.
* python/py-unwind.c: Do not include py-ref.h.
* python/py-utils.c: Do not include py-ref.h.
* python/py-value.c: Do not include py-ref.h.
* python/py-varobj.c: Do not include py-ref.h.
* python/py-xmethods.c: Do not include py-ref.h.
* python/python.c: Do not include py-ref.h.
* varobj.c: Do not include py-ref.h.
Tom Tromey [Mon, 21 Jan 2019 20:45:00 +0000 (13:45 -0700)]
Use "struct bcache" in objfiles.h
If objfiles.h is included after bcache.h, then the "bcache" function
will cause a compiler error because "bcache" will be seen as a
function, not a type. Fix this error by using the "struct" keyword.
gdb/ChangeLog
2019-01-22 Tom Tromey <tom@tromey.com>
* objfiles.h (struct objfile_per_bfd_storage): Use "struct"
keyword for bcache.
Tom Tromey [Mon, 21 Jan 2019 20:43:16 +0000 (13:43 -0700)]
Remove a comment in compile/compile-cplus-types.c
The include sorter can't handle multi-line comments on the same line
as a #include. This patch removes the only such comment.
In general I think these sorts of comments do not provide much value:
more often than not, I find that the comment is obsolete in one way or
another, and so the include sorter removes them in most cases.
gdb/ChangeLog
2019-01-22 Tom Tromey <tom@tromey.com>
* compile/compile-cplus-types.c: Remove a comment by #include.
Tom Tromey [Tue, 22 Jan 2019 08:58:59 +0000 (01:58 -0700)]
Remove the EXTERN define from stabsread.h
This removes the EXTERN define from stabsread.h. This is the only
spot that still uses this approach, and it interfered with sorting the
includes in stabsread.c.
Tested by rebuilding.
gdb/ChangeLog
2019-01-22 Tom Tromey <tom@tromey.com>
* stabsread.c (EXTERN): Do not define.
(symnum, next_symbol_text_func, processing_gcc_compilation)
(within_function, global_sym_chain, global_stabs)
(previous_stab_code, this_object_header_files)
(n_this_object_header_files)
(n_allocated_this_object_header_files): Define.
* stabsread.h (EXTERN): Never define. Use "extern".
E.g. in gdb.base/default.exp, valgrind detects leaks such as
==17663== 1,438 bytes in 101 blocks are definitely lost in loss record 2,804 of 2,884
==17663== at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==17663== by 0x418A17: xmalloc (common-utils.c:44)
==17663== by 0x4E6F19C: history_expand (histexpand.c:1061)
==17663== by 0x4B4490: handle_line_of_input(buffer*, char const*, int, char const*) (event-top.c:685)
==17663== by 0x4B4562: command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) (event-top.c:753)
...
Fix the leak by using an unique_xmalloc_ptr for history_value.
gdb/ChangeLog
2019-01-22 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* event-top.c (handle_line_of_input): use unique_xmalloc_ptr for
history_value.
Tom Tromey [Mon, 21 Jan 2019 15:41:28 +0000 (08:41 -0700)]
Fix leak in splay-tree
Philippe Waroquiers noticed a memory leak in gdb, which he tracked
down to a bug in splay-tree. splay_tree_remove does not call the
`delete_key' function when it removes the old node; but it should.
I looked at every splay tree in GCC and there is only one that passes
a non-NULL delete function -- the one in lto.c. That file does not
call splay_tree_remove. So, I think this is safe to check in.
I re-ran the LTO tests to double check.
libiberty/
* splay-tree.c (splay_tree_remove): Delete the key if necessary.
Alan Hayward [Mon, 21 Jan 2019 15:51:49 +0000 (15:51 +0000)]
AArch64 AAPCS: Empty structs have non zero size in C++
When gdb.base/infcall-nested-structs.c is complied as C++, the compiler
will not pass structs containing empty structs via float arguments.
This is because structs in C++ have a minimum size of 1, causing padding
in the struct once compiled. The AAPCS does not allow structs with
padding to be passed in float arguments.
Add padding checks to AArch64 and add C++ compile variant to the test.
Some of the tests fail on X86_64. This has been raised as bug gdb/24104.
gdb/ChangeLog:
* aarch64-tdep.c (aapcs_is_vfp_call_or_return_candidate_1): Check
for padding.
gdb/testsuite/ChangeLog:
* gdb.base/infcall-nested-structs.exp: Test C++ in addition to C.
Alan Hayward [Mon, 21 Jan 2019 10:43:53 +0000 (10:43 +0000)]
Testsuite: Ensure stack protection is off for GCC
Using -fstack-protector-strong will cause GDB to break on the wrong line
when placing a breakpoint on a function. This is due to inadequate dwarf
line numbering, and is being tracked by the GCC bug
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88432
GCC (and Clang) provided by Debian/Ubuntu default to stack protector
being enabled.
Ensure that when running the GDB testsuite, stack protector is always
turned off for GCC 4.1.0 (when stack protector was added) and above.
Ensure that this does not cause infinite recursion due to
test_compiler_info having to compile a file itself.
Add a test to explicitly test breakpoints with various levels of stack
protection on both GCC and Clang, with xfail for the known errors.
Restore change in ovldbreak.exp which worked around the issue.
gdb/testsuite/ChangeLog:
2019-01-18 Alan Hayward <alan.hayward@arm.com>
* gdb.base/stack-protector.c: New test.
* gdb.base/stack-protector.exp: New file.
* gdb.cp/ovldbreak.exp: Only allow a single break line.
* lib/gdb.exp (get_compiler_info): Use getting_compiler_info
option.
(gdb_compile): Remove stack protector for GCC and prevent
recursion.