Tom Tromey [Thu, 17 Jan 2019 17:19:44 +0000 (10:19 -0700)]
Replace "the the" with "the"
This replaces "the the" with "the" in various comments.
Tested by rebuilding. This didn't test the solib-dsbt.c change, but
it looks harmless.
gdb/ChangeLog
2019-01-17 Tom Tromey <tromey@bapiya>
* valprint.c: Replace "the the" with "the".
* symtab.c: Replace "the the" with "the".
* solib.c: Replace "the the" with "the".
* solib-dsbt.c: Replace "the the" with "the".
* linespec.c: Replace "the the" with "the".
* dwarf2loc.h: Replace "the the" with "the".
* amd64-windows-tdep.c: Replace "the the" with "the".
* aarch64-tdep.c: Replace "the the" with "the".
Tamar Christina [Thu, 17 Jan 2019 16:05:37 +0000 (16:05 +0000)]
Arm: Clean up PE GAS testsuite correct THUMB tests.
The PE targets don't support mapping symbols and so the disassembler is unable
to correctly output thumb instructions when the input was thumb.
So for testcases that only have thumb output, I have copied them and skipped the
ones for which auto-detection is supposed to work on PE, and added a new one
that will force thumb output. This so that the tests still check the mapping
symbols.
For the tests that switch between thumb and arm in one file I just skip them
entirely on PE targets.
Nick Clifton [Thu, 17 Jan 2019 15:29:43 +0000 (15:29 +0000)]
Update objdump's --disassemble=<symbol> feature so that if <symbol> is a function, the entire function will be disassembled, regardless of the presence of interveening symbols.
* objdump.c (disassemble_section): When disassembling from a
symbol only stop at the next symbol if the original symbol was not
a function symbol. Otherwise continue disassembling until a new
function is reached.
* testsuite/binutils-all/objdump.exp: Add tests of extended
functionality.
* testsuite/binutils-all/disasm.s: New test source file.
Simon Marchi [Wed, 16 Jan 2019 22:01:42 +0000 (17:01 -0500)]
Actually remove definitions of DEFINE_NON_INLINE_P and DEFINE_INLINE_P
I applied the patch "Do not expand macros to 'defined'" by hand because
I couldn't apply it with git-am, and of course forgot to remove the
macro definitions. This patch fixes it, and also makes the ChangeLog
entry a bit cleaner.
Jim Wilson [Wed, 16 Jan 2019 21:37:35 +0000 (13:37 -0800)]
Don't emit vendor attribute section if there is no attribute to emit.
2019-01-16 Kito Cheng <[email protected]>
bfd/
* elf-attrs.c (vendor_obj_attr_size): Return 0 if size is 0 even
for OBJ_ATTR_PROC.
gas/
* testsuite/gas/riscv/attribute-empty.d: New.
Keith Seitz [Wed, 16 Jan 2019 19:38:06 +0000 (11:38 -0800)]
Introduce dwarf2_cu::get_builder
This patch is an attempt to deal with a variety of bugs reported where
GDB segfaults attempting to access a dwarf2_cu's builder. In certain
circumstances, this builder can be NULL. This is especially common
when inheriting DIEs via inlined subroutines in other CUs. The test
case demonstrates one such situation reported by users. See gdb/23773,
rhbz1638798, and dups for other concrete examples.
The approach taken here is to save the ancestor CU into the dwarf2_cu of
all CUs with DIEs that are "imported." This can happen whenever
follow_die_offset and friends are called. This essentially introduces a
chain of CUs that caused the importation of a DIE from a CU. Whenever
a builder is requested of a CU that has none, the ancestors are searched
for the first one with a builder.
A design side effect of this is that the builder can now only be
accessed by getter and setter methods because the builder itself
is private.
The bulk of the patch is relatively mindless text conversion from
"cu->builder" to "cu->get_builder ()". I've included one test which
was derived from one (of the many) bugs reported on the issue in both
sourceware and Fedora bugzillas.
gdb/ChangeLog:
PR gdb/23773
* dwarf2read.c (dwarf2_cu) <ancestor>: New field.
<builder>: Rename to ..
<m_builder>: ... this and make private.
(dwarf2_cu::get_builder): New method. Change all users of
`builder' to use this method.
(dwarf2_start_symtab): Move to ...
(dwarf2_cu::start_symtab): ... here. Update all callers
(setup_type_unit_groups): Move to ...
(dwarf2_cu::setup_type_unit_groups): ... here. Update all
callers.
(dwarf2_cu::reset_builder): New method.
(process_full_compunit, process_full_type_unit): Use
dwarf2_cu::reset_builder.
(follow_die_offset): Record the ancestor CU if it is different
from the followed DIE's CU.
(follow_die_sig_1): Likewise.
gdb/testsuite/ChangeLog:
PR gdb/23773
* gdb.dwarf2/inlined_subroutine-inheritance.exp: New file.
Simon Marchi [Wed, 16 Jan 2019 19:32:13 +0000 (14:32 -0500)]
doc: Add table of MI versions
This patch adds a table summarizing the history or MI versions:
- The version number
- Which GDB version introduced it
- Breaking changes compared to the previous version
The goal of the table is to help writers of front ends know which
version of MI they can use with a given GDB version. It will also help
them update their code to work against a newer MI version.
Right now, we just have 1 and 2, but we expect to add an entry for 3
soon. I did a bit of archelogy and reverse engineering of the code to
come up with the breaking changes for MI 2.
I did some changes to the text around it, some things that I thought
needed to be clarified, seemed a bit dated or seemed just wrong
(especially "Apart from mi0, new versions of @value{GDBN} will not
support old versions of MI").
gdb/doc/ChangeLog:
* gdb.texinfo (GDB/MI Development and Front Ends): Add table of
MI versions. Update text around it.
John Darrington [Mon, 14 Jan 2019 15:55:17 +0000 (16:55 +0100)]
S12Z: gas: Fix bug when a symbol name was the single letter 'c'.
The assembler incorrectly recognised "c" as a register name, and
refused to allow it where it expected a symbol/label.
gas/
* config/tc-s12z.c (lex_reg_name): Compare the length of the strings
before the contents.
* testsuite/gas/s12z/labels.d: New file.
* testsuite/gas/s12z/labels.s: New file.
* testsuite/gas/s12z/s12z.exp: Add them.
John Darrington [Wed, 16 Jan 2019 13:34:50 +0000 (14:34 +0100)]
S12Z: gas: Permit "extend" instructions which don't actually extend.
Other assemblers permit "extending" a register into a register of a
smaller size or the same size. It doesn't make much sense to do this
but would appear to be a valid instruction. So change the error to a
warning.
gas/
* config/tc-s12z.c (tfr): Change as_bad to as_warn.
Also fix message typo and semantics.
John Darrington [Wed, 16 Jan 2019 13:34:50 +0000 (14:34 +0100)]
S12Z: Emit RELOC_S12Z_OPR instead of RELOC_EXT24 where appropriate.
When assembling instructions which involve OPR references, emit
RELOC_S12Z_OPR instead of RELOC_EXT24.
bfd/
* bfd-in2.h [BFD_RELOC_S12Z_OPR]: New reloc.
* libbfd.h: regen.
* elf32-s12z.c (eld_s12z_howto_table): R_S12Z_OPR takes non zero
source field. (md_apply_fix): Apply final fix
to BFD_RELOC_S12Z_OPR.
* reloc.c[BFD_RELOC_S12Z_OPR]: New reloc.
gas/
* config/tc-s12z.c (emit_opr): Emit BFD_RELOC_S12Z_OPR instead of
BFD_RELOC_24.
* testsuite/gas/s12z/opr-indirect-expr.d: Expect R_S12Z_OPR instead
of R_S12Z_EXT24.
Simon Marchi [Tue, 15 Jan 2019 22:45:39 +0000 (17:45 -0500)]
sim: Fix definition of SIM_ARANGE_INLINE
If HAVE_INLINE is false, SIM_ARANGE_INLINE is currently defined as
#define SIM_ARANGE_INLINE EXTERN
However, EXTERN is not defined anywhere, leading to errors such as:
In file included from
/mipt-mips/simulator/../../sim/common/sim-basics.h:147:0,
from /mipt-mips/simulator/export/gdb/sim-main.h:13,
from /mipt-mips/simulator/export/gdb/gdb_interface.cpp:7:
/mipt-mips/simulator/../../sim/common/sim-arange.h:71:27: error: ‘EXTERN’
does not name a type; did you mean ‘EUSERS’?
#define SIM_ARANGE_INLINE EXTERN
^
/mipt-mips/simulator/../../sim/common/sim-arange.h:76:1: note: in expansion
of macro ‘SIM_ARANGE_INLINE’
SIM_ARANGE_INLINE int
^~~~~~~~~~~~~~~~~
I (Simon) have reproduced the problem by simply removing the HAVE_INLINE
definition. This was originally reported by Pavel I. Kryukov
<[email protected]>.
sim/common/ChangeLog:
* sim-arange.h (SIM_ARANGE_INLINE): Change EXTERN to extern.
Tom Tromey [Sun, 6 Jan 2019 18:15:05 +0000 (11:15 -0700)]
Remove cleanup from remote.c
This removes the remaining cleanup from remote.c, by changing
remote_state::buf to be a gdb::char_vector, and then fixing up all the
fallout. The resulting patch is somewhat ugly, but on the other hand,
it eliminates some possibilities for mistakes.
Regression tested using the
Fedora-x86_64-native-extended-gdbserver-m64 builder on the buildbot.
* remote.c (class remote_state) <buf>: Now a char_vector.
<buf_size>: Remove.
(remote_target::getpkt): Change type of buf. Remove sizeof_buf
parameter.
(remote_target::getpkt_or_notif_sane_1)
(remote_target::getpkt_sane)
(remote_target::getpkt_or_notif_sane): Likewise.
(class remote_target) <putpkt>: New overload.
(remote_target::read_frame): Change type of "buf_p". Remove
sizeof_p parameter.
(packet_ok): New overload.
(packet_check_result): New overload.
Update all uses.
Tom Tromey [Sat, 12 Jan 2019 20:35:06 +0000 (13:35 -0700)]
Fix placement of output in TUI mode
The fix for PR tui/28819 regressed gdb command output a bit. In
"nonl" mode, pressing the Enter key will result in a newline not being
echoed properly, so that gdb output for the command will begin on the
same line as the input.
This patch changes gdb_wgetch to echo the newline. I have only tested
this interactively, as the TUI doesn't have automated tests in
general.
This patch defines pseudo-registers "v0" through "v31" as aliases that
map to the corresponding raw "vr0" through "vr31" vector registers for
Power.
The motivation behind this is that although GDB defines these
registers as "vrX", the disassembler prints them as "vX", e.g. as the
operands in instructions such as "vaddubm v2,v1,v1". This can be
confusing to users trying to print out the values of the operands
while inspecting the disassembled code.
The new aliases are made not to belong to any register group, to avoid
duplicated values in "info register vector" and "info register all".
The arch-specific rs6000_pseudo_register_reggroup_p function had
previously been removed since the other pseudo-registers could have
their groups inferred by their type. It restored with this patch to
handle the aliases. Membership for the other pseudo-registers is
still determined using the default function.
A new tests checks that GDB prints the expected values of vector
registers after they are filled by the inferior, by using both the raw
names and the aliases. Two other existing tests are modified to also
test the aliases.
gdb/ChangeLog:
2019-01-14 Pedro Franco de Carvalho <[email protected]>
* ppc-tdep.h (struct gdbarch_tdep) <ppc_v0_alias_regnum>: New
field.
* rs6000-tdep.c: Include reggroups.h.
(IS_V_ALIAS_PSEUDOREG): Define.
(rs6000_register_name): Return names for the "vX" aliases.
(rs6000_pseudo_register_type): Return type for the "vX" aliases.
(rs6000_pseudo_register_reggroup_p): Restore. Handle "vX"
aliases. Call default_register_reggroup_p for all other
pseudo-registers.
(v_alias_pseudo_register_read, v_alias_pseudo_register_write):
New functions.
(rs6000_pseudo_register_read, rs6000_pseudo_register_write):
Handle "vX" aliases.
(v_alias_pseudo_register_collect): New function.
(rs6000_ax_pseudo_register_collect): Handle "vX" aliases.
(rs6000_gdbarch_init): Initialize "vX" aliases as
pseudo-registers. Restore registration of
rs6000_pseudo_register_reggroup_p with
set_tdesc_pseudo_register_reggroup_p.
gdb/testsuite/ChangeLog:
2019-01-14 Pedro Franco de Carvalho <[email protected]>
* gdb.arch/vsx-regs.exp: Add tests that use the vector register
aliases.
* gdb.arch/altivec-regs.exp: Likewise. Fix indentation of two
tests.
* gdb.arch/powerpc-vector-regs.c: New file.
* gdb.arch/powerpc-vector-regs.exp: New file.
gdb/doc/ChangeLog:
2019-01-14 Pedro Franco de Carvalho <[email protected]>
* gdb.texinfo (PowerPC Features): Document the alias
pseudo-registers for the org.gnu.gdb.power.altivec feature.
Maamoun Tarsha [Mon, 14 Jan 2019 16:00:14 +0000 (16:00 +0000)]
Add support to GNU ld to separate got related plt entries from normal ones in order to be able to switch the non-plt got entries to read-only after startup, conforming to revised Linux for zSeries ABI.
Max Filippov [Sat, 17 Nov 2018 01:31:22 +0000 (17:31 -0800)]
gdb: xtensa: fix register counters for xtensa-linux
Commit 37d9e0623102 ("gdb: xtensa: handle privileged registers") changed
how the tdep->num_regs and tdep->num_pseudo_regs are calculated, but
didn't update these numbers in the gdbarch for the xtensa-linux target.
As a result xtensa-linux-gdb behaves as xtensa-elf-gdb and cannot
communicate with the linux gdbserver.
Fix tdep->num_pseudo_regs calculation and call set_gdbarch_num_regs and
set_gdbarch_num_pseudo_regs in xtensa_linux_init_abi.
Currently, the behaviour is:
(gdb) show style
(gdb) set style
(gdb) show style address
(gdb) set style address
(gdb)
With this patch, the behaviour is:
(gdb) show style
style address background: The "address" background color is: none
style address foreground: The "address" foreground color is: blue
style address intensity: The "address" display intensity is: normal
enabled: CLI output styling is enabled.
style filename background: The "filename" background color is: none
style filename foreground: The "filename" foreground color is: green
style filename intensity: The "filename" display intensity is: normal
style function background: The "function" background color is: none
style function foreground: The "function" foreground color is: yellow
style function intensity: The "function" display intensity is: normal
style variable background: The "variable" background color is: none
style variable foreground: The "variable" foreground color is: cyan
style variable intensity: The "variable" display intensity is: normal
(gdb) set style
"set style" must be followed by an appropriate subcommand.
List of set style subcommands:
set style address -- Address display styling
set style enabled -- Set whether CLI styling is enabled
set style filename -- Filename display styling
set style function -- Function name display styling
set style variable -- Variable name display styling
Type "help set style" followed by set style subcommand name for full documentation.
Type "apropos word" to search for commands related to "word".
Command name abbreviations are allowed if unambiguous.
(gdb) show style address
background: The "filename" background color is: none
foreground: The "filename" foreground color is: green
intensity: The "filename" display intensity is: normal
(gdb) set style address
List of set style address subcommands:
set style address background -- Set the background color for this property
set style address foreground -- Set the foreground color for this property
set style address intensity -- Set the display intensity color for this property
Type "help set style address" followed by set style address subcommand name for full documentation.
Type "apropos word" to search for commands related to "word".
Command name abbreviations are allowed if unambiguous.
(gdb)
gdb/ChangeLog
* cli/cli-style.h (class cli_style_option): <add_setshow_commands>
Remove arg prefixname, add do_set and do_show.
Add member functions set_list and show_list.
* cli/cli-style.c (class cli_style_option): Update accordingly.
(style_set_list): Move to file scope.
(style_show_list): Likewise.
(set_style): Call help_list.
(show_style): Call cmd_show_list.
(_initialize_cli_style): New macro STYLE_ADD_SETSHOW_COMMANDS.
Update to use the new macro.
Joel Brobecker [Sat, 12 Jan 2019 14:47:58 +0000 (09:47 -0500)]
expand a bit the in-GDB help for the "catch exception" help text
One of our users remarked that the help doesn't mention the fact that
the "catch exception" supports the special argument "unhandled" to catch
exceptions which do not have a handler. This patch changes the output
of...
| (gdb) help catch exception
| Catch Ada exceptions, when raised.
| With an argument, catch only exceptions with the given name.
... to ...
| (gdb) help catch exception
| Catch Ada exceptions, when raised.
| Usage: catch exception [ ARG ]
|
| Without any argument, stop when any Ada exception is raised.
| If ARG is "unhandled" (without the quotes), only stop when the exception
| being raised does not have a handler (and will therefore lead to the task's
| termination).
| Otherwise, the catchpoint only stops when the name of the exception being
| raised is the same as ARG.
gdb/ChangeLog:
* ada-lang.c (_initialize_ada_language): Expand the help text
for the "catch exception" command.
Andrew Burgess [Fri, 11 Jan 2019 23:27:33 +0000 (23:27 +0000)]
gdb/testsuite: Don't allow paths to appear in test name
Having paths in the test names makes it harder to compare results
between two runs in different directories. Give the test a name so
that the path doesn't appear.
gdb/ChangeLog:
* gdb.base/style.exp: Don't include path in testname.
Fix 'obj' may be used uninitialized warning in symtab.c:matching_obj_sections.
Fix warning:
gdb/symtab.c: In function ‘int matching_obj_sections(obj_section*, obj_section*)’:
gdb/symtab.c:1024:12: warning: ‘obj’ may be used uninitialized in this function [-Wmaybe-uninitialized]
if (obj->separate_debug_objfile_backlink != NULL
Keith Seitz [Thu, 10 Jan 2019 21:57:09 +0000 (13:57 -0800)]
gdb/23712: Test case for multidictionary
This is a test derived from one of the reproducers in symtab/23010.
The DIE tree used here is typical of compilations with LTO, where an
artificial parent DIE of language C99 imports DIEs of other languages.
gdb/testsuite/ChangeLog:
PR gdb/23712
PR symtab/23010
* gdb.dwarf2/multidictionary.exp: New file.
Keith Seitz [Thu, 10 Jan 2019 21:57:08 +0000 (13:57 -0800)]
gdb/23712: Remove dw2_add_symbol_to_list
Finally, we can remove dw2_add_symbol_to_list since the wrapper function
originally introduced to catch this multi-language scenario is no longer
needed. With multi-language dictionaries, we can now support adding
symbols of multiple languages, negating the need for the assertion
entirely.
This patch should now fix gdb/23712 (and symtab/23010). At least it will
if the NULL buildsym_compunit problem doesn't strike first (see gdb/23773).
Now that multidictionary's are being used, there is no longer any need
to retain the four temporary functions introduced in the beginning of
this series.
This patch removes them.
As an additional cleanup, since the single-language dictionaries are
no longer used outside dictionary.c, make all of those functions
static.
gdb/ChangeLog:
PR gdb/23712
PR symtab/23010
* dictionary.c (pending_to_vector): Remove.
(dict_create_hashed_1, dict_create_linear_1, dict_add_pending_1):
Remove _1 suffix, replacing functions of the same name. Update
all callers.
(dict_create_hashed, dict_create_hashed_expandable)
(dict_create_linear, dict_create_linear_expandable, dict_free)
(dict_add_symbol, dict_add_pending, dict_size, dict_empty):
Make functions static.
Keith Seitz [Thu, 10 Jan 2019 21:57:08 +0000 (13:57 -0800)]
gdb/23712: Use new multidictionary API
This patch builds on the previous by enabling the `new' multidictionary
API. A lot of the hunks are simply textual replacements of "dict_"
with "mdict_" and similar transformations.
A word of warning, even with the use of multidictionaries, the code
still does not satisfactorily fix the reported problems with gdb/23712
(or gdb/23010). We still have additional changes to make before that
happens.
gdb/ChangeLog:
PR gdb/23712
PR symtab/23010
* dictionary.h (struct dictionary): Replace declaration with
multidictionary.
(dict_create_hashed, dict_create_hashed_expandable)
(dict_create_linear, dict_create_linear_expandable)
(dict_free, dict_add_symbol, dict_add_pending, dict_empty)
(dict_iterator_first, dict_iterator_next, dict_iter_match_first)
(dict_iter_match_next, dict_size): Rename to "mdict_" versions
taking multidictionary argument.
[ALL_DICT_SYMBOLS]: Update for multidictionary.
* block.h (struct block) <dict>: Change to multidictionary
and rename `multidict'.
* block.c, buildsym.c, jit.c, mdebugread.c, objfiles.c,
symmisc.c: Update all dictionary references to multidictionary.
Keith Seitz [Thu, 10 Jan 2019 21:57:08 +0000 (13:57 -0800)]
gdb/23712: Introduce multidictionary's
gdb/23712 is a new manifestation of the now-infamous (at least to me)
symtab/23010 assertion failure (DICT_LANGUAGE == SYMBOL_LANGAUGE).
An example of the problem (using test case from symtab/23010):
Reading symbols from /home/rdiez/rdiez/arduino/JtagDue/BuildOutput/JtagDue-obj-release/firmware.elf...done.
(gdb) p SysTick_Handler
dwarf2read.c:9715: internal-error: void dw2_add_symbol_to_list(symbol*, pending**): Assertion `(*listhead) == NULL || (SYMBOL_LANGUAGE ((*listhead)->symbol[0]) == SYMBOL_LANGUAGE (symbol))' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n)
This assertion was added specifically to catch this condition (of adding
symbols of different languages to a single pending list).
The problems we're now seeing on systems utilizing DWARF debugging seem to
be caused by the use of LTO, which adds a CU with an artificial DIE of
language C99 which references DIEs in other CUs of language C++.
Thus, we create a dictionary containing symbols of C99 but end up
stuffing C++ symbols into it, and the dw2_add_symbol_to_list triggers.
The approach taken here to fix this is to introduce multi-language
dictionaries to "replace" the standard, single-language dictionaries
used today.
Note to reviewers: This patch introduces some temporary functions to
aide with review. This and other artifacts (such as "See dictionary.h"
which appear incorrect) will all be valid at the end of the series.
This first patch introduces the new multidictionary and its API (which
is, by design, identical to the old dictionary interface). It also
mutates dict_create_hashed and dict_create_linear so that they take
a std::vector instead of the usual struct pending linked list. This will
be needed later on.
This patch does /not/ actually enable multidictionary's. That is left
for a subsequent patch in the series.
I've done exhaustive performance testing with this approach, and I've
attempted to minimize the overhead for the (overwhelmingly) most common
one-language scenario.
On average, a -g3 -O0 GDB (the one we developers use) will see
approximately a 4% slowdown when initially reading symbols. [I've
tested only GDB and firefox with -readnow.] When using -O2, this
difference shrinks to ~0.5%. Since a number of runs with these
patches actually run /faster/ than unpatched GDB, I conclude that
these tests have at least a 0.5% error margin.
On our own gdb.perf test suite, again, results appear to be pretty
negligible. Differences to unpatched GDB range from -7.8% (yes,
patched version is again faster than unpatched) to 27%. All tests
lying outside "negligible," such as the 27% slowdown, involve a total
run time of 0.0007 (or less) with smaller numbers of CUs/DSOs (usually 10
or 100). In all cases, the follow-up tests with more CUs/DSOs is never
more than 3% difference to the baseline, unpatched GDB.
In my opinion, these results are satisfactory.
gdb/ChangeLog:
PR gdb/23712
PR symtab/23010
* dictionary.c: Include unordered_map.
(pending_to_vector): New function.
(dict_create_hashed_1, dict_create_linear_1, dict_add_pending_1):
Rewrite the non-"_1" functions to take vector instead
of linked list.
(dict_create_hashed, dict_create_linear, dict_add_pending): Use the
"new" _1 versions of the same name.
(multidictionary): Define.
(std::hash<enum language): New definition.
(collate_pending_symbols_by_language, mdict_create_hashed)
(mdict_create_hashed_expandable, mdict_create_linear)
(mdict_create_linear_expandable, mdict_free)
(find_language_dictionary, create_new_language_dictionary)
(mdict_add_symbol, mdict_add_pending, mdict_iterator_first)
(mdict_iterator_next, mdict_iter_match_first, mdict_iter_match_next)
(mdict_size, mdict_empty): New functions.
* dictionary.h (mdict_iterator): Define.
Pedro Alves [Thu, 10 Jan 2019 17:52:39 +0000 (17:52 +0000)]
Fix tracepoint.c:parse_tracepoint_definition leak (and one more)
Coverity points out that gdb/tracepoint.c:parse_tracepoint_definition
can leak 'cond' in this line:
cond = (char *) xmalloc (2 * xlen + 1);
That can leak because we're in a loop and 'cond' may have already been
xmalloc'ed into in a previous iteration. That won't normally happen,
because we don't expect to see a tracepoint definition with multiple
conditions listed, but, it doesn't hurt to be pedantically correct,
in case some stub manages to send something odd back to GDB.
At first I thought I'd just replace the xmalloc call with:
cond = (char *) xrealloc (cond, 2 * xlen + 1);
and be done with it. However, my pedantic self realizes that
warning() can throw as well (due to pagination + Ctrl-C), so I fixed
it using gdb::unique_xmalloc_ptr instead.
While doing this, I noticed that these vectors in struct uploaded_tp:
hold heap-allocated strings, but nothing is freeing the strings,
AFAICS.
So I ended up switching all the heap-allocated strings in uploaded_tp
to unique pointers. This patch is the result of that.
I also wrote an alternative, but similar patch that uses std::string
throughout instead of gdb::unique_xmalloc_ptr, but in the end reverted
it because the code didn't look that much better, and I kind of
dislike replacing pointers with fat std::string's (3 or 4 times the
size of a pointer) in structures.
Pedro Alves [Thu, 10 Jan 2019 17:52:38 +0000 (17:52 +0000)]
Fix leak in mdebugread.c
Coverity points out that all the "continue;" statements in the switch
case in parse_partial_symbols leak STABSTRING. This is because we
only release STABSTRING at the end of the scope, with:
Tom Tromey [Thu, 10 May 2018 22:23:57 +0000 (16:23 -0600)]
Move psymtabs to their own obstack
Previously, the psymtab obstack was just a pointer to the objfile
obstack. This patch changes psymtabs to use their own obstack,
instead. A gdb::optional is used to avoid unnecessary allocation when
the obstack is not needed.
After this patch, the psymtab code lifetime model is that, in the core
psymtab code, objects allocated on the psymtab obstack may point to
other such objects, or to objects on the per-BFD obstack -- but never
to the objfile obstack.
Note however that this invariant is only obeyed the core psymtab code,
and even there not quite fully: there is still a link from the psymtab
to the full symtab.
Symbol readers are free to work however they like; and in particular,
even after this patch, in practice all symbol readers violate this
invariant via the read_symtab_private field.
Tom Tromey [Thu, 10 May 2018 22:23:56 +0000 (16:23 -0600)]
Make psymtab_storage::free_psymtabs private
This adds a new psymtab allocation method to psymtab_storage and
changes the free_psymtabs member to be private. While not strictly
necessary, this seems like a decent cleanup, and also makes it simpler
to move psymtabs off of obstacks entirely, should that prove
desirable.
* psymtab.h (psymtab_storage::allocate_psymtab): New method.
<free_psymtabs>: Now private.
* psymtab.c (psymtab_storage::allocate_psymtab): Implement.
(allocate_psymtab): Use new method.
Tom Tromey [Thu, 10 May 2018 22:23:55 +0000 (16:23 -0600)]
Add psymtab_storage::allocate_dependencies
This adds a new method to psymtab_storage to allocate storage for
psymtab dependencies, then changes the symbol readers to use it. This
has the effect of moving the storage to the psymtab storage obstack.
Tom Tromey [Thu, 10 May 2018 22:23:53 +0000 (16:23 -0600)]
Allocate the address map on the psymtab obstack
After this patch, the psymtab address map will now be allocated on the
psymtab obstack rather than the objfile obstack. This also changes
the psymtab storage object to make the obstack private; this will be
used later.
Tom Tromey [Thu, 10 May 2018 22:23:52 +0000 (16:23 -0600)]
Introduce objfile::reset_psymtabs
This introduces a new method, objfile::reset_psymtabs, and changes
reread_symbols to use it. This method simply destroys the existing
partial symbols and recreates the psymtab_storage object.
This patch fixes a latent bug -- namely, that reread_symbols should
clear objfile::psymbol_map, but does not. I can submit that
separately if you'd prefer.
Tom Tromey [Thu, 10 May 2018 22:23:51 +0000 (16:23 -0600)]
Introduce class psymtab_storage
This introduces a new psymtab_storage class, which holds all
psymbol-related objects that are independent of the objfile. (This
latter contraint explains why psymbol_map was not moved; though this
could still be done with some work.)
This patch does not yet change where psymtab allocation is done --
that comes later. This just wraps everything in a single object to
make further transformations simpler.
Note that a shared_ptr is used to link from the objfile to the
psymtab_storage object. The end goal here is to allow a given symbol
reader to simply attach to the psymtab_storage object to the BFD, then
reuse it in later invocations; shared_ptr makes this simple to reason
about.
Tom Tromey [Thu, 10 May 2018 22:23:49 +0000 (16:23 -0600)]
Change symbol_set_names to take an objfile_per_bfd_storage
This changes symbol_set_names to take an objfile_per_bfd_storage
argument, and updates the users. It also changes PSYMBOL_SET_NAMES to
take this argument directly; I feel this clarifies the storage
location of objects created in psymtab.c.
Tom Tromey [Thu, 10 May 2018 22:23:48 +0000 (16:23 -0600)]
Change create_demangled_names_hash to take an objfile_per_bfd_storage
This changes create_demangled_names_hash to take an
objfile_per_bfd_storage parameter. This makes it clearer where it is
storing the objects it allocates.
Tom Tromey [Thu, 10 May 2018 22:23:47 +0000 (16:23 -0600)]
Simplify calls to init_psymbol_list
Existing callers to init_psymbol_list were checking to see if psymbols
had already been initialized. It seemed better to me to do this check
directly in init_psymbol_list, simplifying the callers.
Tom Tromey [Thu, 10 May 2018 22:23:46 +0000 (16:23 -0600)]
Change add_psymbol_to_list to use an enum
This changes add_psymbol_to_list to use an enum, rather than a pointer
to a vector, to decide where to put the new symbol. This reduces the
number of direct references to the static_psymbols and global_psymbols
members of the objfile, which is handy in a later patch.
Tom Tromey [Thu, 10 May 2018 22:23:45 +0000 (16:23 -0600)]
Remove parameters from start_psymtab_common
start_psymtab_common takes references to the global_psymbols and
static_psymbols vectors, but it also has an objfile parameter. This
is redundant, so this patch simplifies the function by removing those
reference parameters.
Tom Tromey [Thu, 10 May 2018 22:23:44 +0000 (16:23 -0600)]
Remove some unneeded psymtab initializations
allocate_psymtab has long cleared the new psymtab that is returned.
This patch documents this behavior and then removes some redundant
initializations.
Remove support for demangling GCC 2.x era mangling schemes.
* cplus-dem.c: Remove cplus_mangle_opname, cplus_demangle_opname,
internal_cplus_demangle, and all subroutines.
(libiberty_demanglers): Remove entries for ancient GNU (pre-3.0),
Lucid, ARM, HP, and EDG demangling styles.
(cplus_demangle): Remove 'work' variable. Don't call
internal_cplus_demangle.
include * Merge from GCC:
2018-12-22 Jason Merrill <[email protected]>
* demangle.h: Remove support for ancient GNU (pre-3.0), Lucid,
ARM, HP, and EDG demangling styles.
Tom Tromey [Sat, 24 Nov 2018 18:54:26 +0000 (11:54 -0700)]
Remove ALL_OBJSECTIONS
This removes the ALL_OBJSECTIONS macro, replacing its uses with ranged
for loops.
The special code in this macro for noticing a "break" from the inner
loop was only needed in a single place; so rather than try to
replicate this, I've simply replaced that use with a "goto".
Tom Tromey [Sat, 24 Nov 2018 16:00:42 +0000 (09:00 -0700)]
Remove ALL_OBJFILE_COMPUNITS
This removes ALL_OBJFILE_COMPUNITS, replacing its uses with ranged for
loops. Because ALL_COMPUNITS is also updated, in some places a
declaration must be deleted to avoid shadowing.
Tom Tromey [Fri, 23 Nov 2018 19:20:05 +0000 (12:20 -0700)]
Remove most uses of ALL_OBJFILES
This removes most uses of ALL_OBJFILES, replacing them with ranged for
loops. The remaining uses are all in macros, and will be removed in
subsequent patches.
Tom Tromey [Fri, 15 Jun 2018 05:12:16 +0000 (22:12 -0700)]
Introduce all_objfiles and next_iterator
This introduces an iterable object which can be used to iterate over
objfiles. It also introduces a generic "next_iterator", which can be
used to iterate over types that have a "next" field.
John Darrington [Fri, 4 Jan 2019 08:44:58 +0000 (09:44 +0100)]
S12Z: Don't crash when disassembling invalid instructions.
Check for null before dereferencing an operand pointer. Normally
this situation should never arise, but could happen if a "partial"
instruction is encountered at the end of a file or section.
opcodes/
* s12z-dis.c (print_insn_s12z): Do not dereference an
operand if it is null.
John Darrington [Mon, 31 Dec 2018 07:48:10 +0000 (07:48 +0000)]
S12Z: Fix disassembly of indexed OPR operands with zero index.
gas/
* testsuite/gas/s12z/jsr.s: New case.
* testsuite/gas/s12z/jsr.d: New case.
opcodes/
* s12z-dis.c (opr_emit_disassembly): Do not omit an index if it is
zero.
Simon Marchi [Wed, 9 Jan 2019 17:57:16 +0000 (12:57 -0500)]
gdb: Remove support for old mangling schemes
An upcoming sync with gcc's libiberty [1] will remove support for old
mangling schemes (GNU v2, Lucid, ARM, HP and EDG). It will remove the
cplus_demangle_opname function, so we need to get rid of its usages in
GDB (it's a GNU v2 specific function).
I think the changes are mostly relatively obvious, some hacks that were
necessary to support overloaded operators with GNU v2 mangling are not
needed anymore.
The change in stabsread.c is perhaps less obvious. I think we could get
rid of more code in that region that is specific to old mangling
schemes, but I chose to do only the minimal changes required to remove
the cplus_demangle_opname uses. There is also a detailed comment just
above that explaining how GNU v2 and v3 mangled symbols are handled, I
decided to leave it as-is, since I wasn't sure which part to remove,
change or leave there.
[1] The commit "Remove support for demangling GCC 2.x era mangling
schemes.", specifically.
gdb/ChangeLog:
* gdbtypes.c (check_stub_method_group): Remove handling of old
mangling schemes.
* linespec.c (find_methods): Likewise.
* stabsread.c (read_member_functions): Likewise.
* valops.c (search_struct_method): Likewise.
(value_struct_elt_for_reference): Likewise.
* NEWS: Mention this change.
gdb/testsuite/ChangeLog:
* gdb.cp/demangle.exp (test_gnu_style_demangling): Rename to...
(test_gnuv3_style_demangling): ... this.
(test_lucid_style_demangling): Remove.
(test_arm_style_demangling): Remove.
(test_hp_style_demangling): Remove.
(do_tests): Remove calls to the above.
gdb/doc/ChangeLog:
* gdb.texinfo (Print Settings): Remove mention of specific
demangle-style values, just refer to the in-process help.
Andrew Burgess [Mon, 7 Jan 2019 07:26:35 +0000 (07:26 +0000)]
gdb: Avoid signed integer overflow when printing source lines
When printing source lines with calls to print_source_lines we need to
pass a start line number and an end line number. The end line number
is calculated by calling get_lines_to_list and adding this value to
the start line number. For example this code from list_command:
print_source_lines (cursal.symtab, first,
first + get_lines_to_list (), 0);
The problem is that get_lines_to_list returns a value based on the
GDB setting `set listsize LISTSIZE`. By default LISTSIZE is 10,
however, its also possible to set LISTSIZE to unlimited, in which
case get_lines_to_list will return INT_MAX.
As the parameter signature for print_source_lines is:
and `first` in the above code is an `int`, then when LISTSIZE is
`unlimited` the above code will result in signed integer overflow,
which is undefined.
The solution in this patch is a new class source_lines_range that can
be constructed from a single line number and a direction (forward or
backward). The range is then constructed from the line number and the
value of get_lines_to_list.
gdb/ChangeLog:
* cli/cli-cmds.c (list_command): Pass a source_lines_range to
print_source_lines.
* source.c (print_source_lines_base): Update line number check.
(print_source_lines): New function.
(source_lines_range::source_lines_range): New function.
* source.h (class source_lines_range): New class.
(print_source_lines): New declaration.
Andrew Burgess [Thu, 3 Jan 2019 21:13:52 +0000 (21:13 +0000)]
gdb/testsuite: Remove interactive prompt case from mi_gdb_test
I noticed that when running this test:
make check-gdb RUNTESTFLAGS="--target_board=native-gdbserver gdb.mi/mi-break.exp"
I would occasionally see some UNRESOLVED test results like this:
(gdb)
PASS: gdb.mi/mi-break.exp: mi-mode=separate: breakpoint at main
Expecting: ^(kill[
]+)?(.*[
]+[(]gdb[)]
[ ]*)
kill
&"kill\n"
~"Kill the program being debugged? (y or n) [answered Y; input not from terminal]\n"
=thread-group-exited,id="i1"
ERROR: Got interactive prompt.
UNRESOLVED: gdb.mi/mi-break.exp: mi-mode=separate:
The problem appears to be that the expect buffer fills up to include
the '(y or n)' prompt without including the following lines.
The pattern supplied by the outer test script is looking for the
following lines. As the following lines are not present then expect
matches on the interactive prompt case rather than the case for the
user supplied pattern.
The problem with this is that we are not really at an interactive
prompt, GDB is providing an answer for us and then moving on. When I
examine a successful run of the test the output from GDB is identical,
the only difference is where expect happens to buffer the output from
GDB.
This patch remove all special handling of the interactive prompt
case. This means that if we ever break GDB and start seeing an
unexpected interactive prompt then tests will rely on a timeout to
fail, instead of having dedicated interactive prompt detection, but
this solves the problem that an auto-answered prompt looks very
similar to an interactive prompt.
With this patch in place I can now leave the following loop running
indefinitely, where before it would fail usually after ~10
iterations.
while make check-gdb RUNTESTFLAGS="--target_board=native-gdbserver gdb.mi/mi-break.exp"; \
do /bin/true; \
done
Valgrind reports a leak in many tests, such as:
==9382== 16 bytes in 1 blocks are definitely lost in loss record 236 of 3,282
==9382== at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==9382== by 0x4197AF: xrealloc (common-utils.c:64)
==9382== by 0x51D16A: xresizevec<linespec_canonical_name> (poison.h:170)
==9382== by 0x51D16A: add_sal_to_sals(linespec_state*, std::vector<symtab_and_line, std::allocator<symtab_and_line> >*, symtab_and_line*, char const*, int) (linespec.c:1041)
==9382== by 0x51E2BF: create_sals_line_offset (linespec.c:2215)
==9382== by 0x51E2BF: convert_linespec_to_sals(linespec_state*, linespec*) (linespec.c:2358)
==9382== by 0x521B5D: convert_explicit_location_to_sals (linespec.c:2473)
Fix leak by xfree-ing self->canonical_names in linespec_state_destructor.
The leak probably appeared with the patch 'Remove cleanup from linespec.c',
as there was a cleanup to xfree canonical_names before the patch.
Tested on Debian/amd64, native and under valgrind.
Tom Tromey [Sun, 6 Jan 2019 16:49:11 +0000 (09:49 -0700)]
Fix build failure with macOS bison
PR gdb/24060 points out a compilation failure of the C, Fortran and Pascal
parsers when they are built using the macOS system bison. The bug is a name
clash between the VARIABLE token name and the VARIABLE enumerator in ui-out.h.
This patch renames VARIABLE in c-exp.y, f-exp.y and p-exp.y to DOLLAR_VARIABLE
to avoid the clash. It also renames similar variables in other .y files so
that all languages use the same name.
Nick Clifton [Tue, 8 Jan 2019 16:53:02 +0000 (16:53 +0000)]
Remove support for old gnu v2 name mangling.
PR 24044
* cxxfilt.c (hp_symbol_characters): Delete.
(main): Remove depcreated demangling styles.
* stabs.c (parse_stab_argtypes): Remove support for old gnu v2
demangling opnames.
* testsuite/binutils-all/cxxfilt.exp: Use the
--no-strip-underscore option for targets that do prefix their
symbols with underscores.
Update tests to elimiate those that use gnu v2 encoding.
projects / binutils.git / log