A number of backends want to return bfd_reloc_dangerous messaqes from
relocation special_function, and construct the message using asprintf.
Such messages are not freed anywhere, leading to small memory leaks
inside libbfd. To limit the leaks, I'd implemented a static buffer in
the ppc backends that was freed before use in asprintf output. This
patch extends that scheme to other backends using a shared static
buffer and goes further in freeing the buffer on any bfd_close.
The patch also fixes a few other cases where asprintf output was not
freed after use.
bfd/
* bfd.c (_input_error_msg): Make global and rename to..
(_bfd_error_buf): ..this.
(bfd_asprintf): New function.
(bfd_errmsg): Use bfd_asprintf.
* opncls.c (bfd_close_all_done): Free _buf_error_buf.
* elf32-arm.c (find_thumb_glue, find_arm_glue): Use bfd_asprintf.
* elf32-nios2.c (nios2_elf32_relocate_section): Likewise.
* elf32-ppc.c (ppc_elf_unhandled_reloc): Likewise.
* elf64-ppc.c (ppc64_elf_unhandled_reloc): Likewise.
* elfnn-riscv.c (riscv_resolve_pcrel_lo_relocs): Likewise.
(riscv_elf_relocate_section): Likewise.
* libbfd.h: Regenerate.
gas/
* read.c (read_end): Free current_name and current_label.
(do_s_func): Likewise on error path. strdup label.
ld/
* pe-dll.c (make_head, make_tail, make_one),
(make_singleton_name_thunk, make_import_fixup_entry),
(make_runtime_pseudo_reloc),
(pe_create_runtime_relocator_reference: Free oname after use.
When running test-case gdb.tui/long-prompt.exp with check-read1, we get:
...
(gdb) FAIL: gdb.tui/long-prompt.exp: prompt size == width + 1: \
end of screen: at last line
...
The problem is in these commands:
...
Term::command "echo \\n"
Term::command "echo \\n"
Term::command "echo \\n"
Term::command "echo \\n"
...
The last one makes the terminal scroll, and the scrolling makes the expected
output match on a different line.
Fix this by replacing the sequence with a single command:
...
Term::command "echo \\n\\n\\n\\n\\n\\n"
...
which avoids scrolling.
Tested on x86_64-linux.
There is a test-case that contains a unit test for tuiterm:
gdb.tui/tuiterm.exp.
However, this only excercises the tuiterm itself, and not the functions that
interact with it, like Term::command.
Add a new test-case gdb.tui/tuiterm-2.exp that:
- overrides proc accept_gdb_output (to be able simulate incorrect responses
while avoiding the timeout),
- overrides proc send_gdb (to be able to call Term::command without a gdb
instance, such that all tuiterm input is generated by the test-case).
- issues Term::command calls, and
- checks whether they behave correctly.
This exposes a problem in Term::command. The "prompt before command" regexp
starts with a bit that is supposed to anchor the prompt to the border:
...
set str "(^|\|)$gdb_prompt $str"
...
but that doesn't work due to insufficient escaping. Fix this by adding the
missing escape:
...
set str "(^|\\|)$gdb_prompt $str"
...
Futhermore, the "prompt after command" regexp in Term::wait_for has no
anchoring at all:
...
set prompt_wait_for "$gdb_prompt \$"
...
so add that as well.
Tested on x86_64-linux.
Currently proc with_override does not work with procs with default value args.
Fix this, and add a test-case excercising this scenario.
Tested on x86_64-linux.
On openSUSE Leap 15.4 with system python 3.6, I run into:
...
(gdb) python check_everything()^M
(gdb) FAIL: gdb.dap/type_check.exp: type checker
...
In check_everything, the hasattr test fails silently:
...
def check_everything():
# Older versions of Python can't really implement this.
if hasattr(typing, "get_origin"):
...
and that makes the gdb_test in the test-case fail.
Fix this by emitting UNSUPPORTED instead in check_everything, and detecting
this in the test-case.
Tested on x86_64-linux.
We recently realized that symbol_needs_eval_fail.exp and
symbol_needs_eval_timeout.exp invalidly dereference an int (4 bytes on
x86_64) by reading 8 bytes (the size of a pointer).
Here how it goes:
In gdb/testsuite/gdb.dwarf2/symbol_needs_eval.c a global variable is
defined:
int exec_mask = 1;
and later both tests build some DWARF using the assembler doing:
set exec_mask_var [gdb_target_symbol exec_mask]
...
DW_TAG_variable {
{DW_AT_name a}
{DW_AT_type :$int_type_label}
{DW_AT_location {
DW_OP_addr $exec_mask_var
DW_OP_deref
...
}
}
The definition of the DW_OP_deref (from Dwarf5 2.5.1.3 Stack Operations)
says that "The size of the data retrieved from the dereferenced address
is the size of an address on the target machine."
On x86_64, the size of an int is 4 while the size of an address is 8.
The result is that when evaluating this expression, the debugger reads
outside of the `a` variable.
Fix this by using `DW_OP_deref_size $int_size` instead. To achieve
this, this patch adds the necessary steps so we can figure out what
`sizeof(int)` evaluates to for the current target.
While at it, also change the definition of the int type in the assembled
DWARF information so we use the actual target's size for an int instead
of the literal 4.
Tested on x86_64 Linux.
Approved-By: Tom Tromey <tom@tromey.com>
Tom Tromey pointed out that the test and call to error() for the
DW_OP_GNU_uninit case in dwarf_expr_context::execute_stack_op (in
gdb/dwarf2/expr.c)...
if (op_ptr != op_end && *op_ptr != DW_OP_piece
&& *op_ptr != DW_OP_bit_piece)
error (_("DWARF-2 expression error: DW_OP_GNU_uninit must always "
"be the very last op in a DWARF expression or "
"DW_OP_piece/DW_OP_bit_piece piece."));
...could be replaced by a call to dwarf_expr_require_composition which
performs a similar check and outputs a suitable error message.
Currently the Fortran test suite does not run with armflang because the
compiler detection fails. This in turn means fortran_runto_main does not
know which main method to use to start a test case.
Fortran compiler detection was added in 44d469c5f85; however, the commit
message notes that it was not tested with armflang.
This commit tests and fixes up a minor issue to get the detection
working.
The goal here is to get the tests running and preventing further
regressions during future work. This change does not do anything to fix
existing failures.
>From what I can understand, the auto detection leverages the
preprocessor to extract the Fortran compiler identity from the defines.
This preprocessor output is then evaluated by the test suite to import
these defines.
In the case of armflang, this evaluation step is disrupted by the
presence of the following warning:
$ armflang -E -fdiagnostics-color=never testsuite/lib/compiler.F90 -o compiler.exp
$ clang-13: warning: argument unused during compilation: '-fdiagnostics-color=never' [-Wunused-command-line-argument]
The evaluation logic is already set up to filter this warning, but the
prefix differs.
This commit fixes the issue by updating the filter to exclude the
armflang flavour of warning.
gdb.fortran regression tests run with GNU, Intel and Intel LLVM. No
regressions detected.
The gdb.fortran test results with ACfL 23.04.1 are as follows.
Before:
# of expected passes 560
# of unexpected failures 113
# of unresolved testcases 2
# of untested testcases 5
# of duplicate test names 2
After:
# of expected passes 5388
# of unexpected failures 628
# of known failures 10
# of untested testcases 8
# of unsupported tests 5
# of duplicate test names 5
As can be seen from the above, there are now considerably more passing
assertions.
Reviewed-By: Luis Machado <luis.machado@arm.com>
Approved-By: Tom Tromey <tom@tromey.com>
Kévin pointed out that gdb claims a minimum Python version of 3.2, but
the DAP code uses f-strings, which were added in 3.6.
This patch removes the uses of f-strings from the DAP code. I can't
test an older version of Python, but I did confirm that this still
works with the version I have.
I realized that I had only implemented DAP breakpoint conditions for
exception breakpoints, and not other kinds of breakpoints. This patch
corrects the oversight.
Currently, gdb will unwind the entire stack in response to the
stackTrace request. I had erroneously thought that the totalFrames
attribute was required in the response. However, the spec says:
If omitted or if `totalFrames` is larger than the available
frames, a client is expected to request frames until a request
returns less frames than requested (which indicates the end of the
stack).
This patch removes this from the response in order to improve
performance when the stack trace is very long.
Co-workers who work on a program that uses DAP asked for the ability
to have gdb stop at the main subprogram when launching. This patch
implements this extension.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
This adds a new "target" to the DAP attach request. This is passed to
"target remote". I thought "attach" made the most sense for this,
because in some sense gdb is attaching to a running process. It's
worth noting that all DAP "attach" parameters are defined by the
implementation.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
A DAP client can report the supportsVariableType capability in the
initialize request. In this case, gdb can include the type of a
variable or expression in various results.
This adds an 'assign' method to gdb.Value. This allows for assignment
without requiring the use of parse_and_eval.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
It occurred to me recently that gdb's DAP implementation should
probably check the types of objects coming from the client. This
patch implements this idea by reusing Python's existing type
annotations, and supplying a decorator that verifies these at runtime.
Python doesn't make it very easy to do runtime type-checking, so the
core of the checker is written by hand. I haven't tried to make a
fully generic runtime type checker. Instead, this only checks the
subset that is needed by DAP. For example, only keyword-only
functions are handled.
Furthermore, in a few spots, it wasn't convenient to spell out the
type that is accepted. I've added a couple of comments to this effect
in breakpoint.py.
I've tried to make this code compatible with older versions of Python,
but I've only been able to try it with 3.9 and 3.10.
My co-worker Kévin taught me that using a mutable object as a default
argument in Python is somewhat dangerous, because the object is
created a single time (when the function is defined), and so if it is
mutated in the body of the function, the changes will stick around.
This patch changes the cases like this in DAP to use () rather than []
as the default. This patch is merely preventative, as no bugs like
this are in the code.
The 'request' decorator is intended to also ensure that the request
function runs in the DAP thread. However, the unwrapped function is
installed in the global request map, so the wrapped version is never
called. This patch fixes the bug.
When I first started implementing DAP, I had some vague plan of having
the implementation functions use the same name as the request. I
abandoned this idea, but one vestige remained. This patch renames the
one remaining function to be gdb-ish.
This implements the DAP "attach" request.
Note that the copyright dates on the new test source file are not
incorrect -- this was copied verbatim from another directory.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
This implements the DAP setExceptionBreakpoints request for Ada. This
is a somewhat minimal implementation, in that "exceptionOptions" are
not implemented (or advertised) -- I wasn't completely sure how this
feature is supposed to work.
I haven't added C++ exception handling here, but it's easy to do if
needed.
This patch relies on the new MI command execution support to do its
work.
Currently, Ada catchpoints require that the inferior be running.
However, there's no deep reason for this -- for example, C++ exception
catchpoints do not have this requirement. Instead, those work like
ordinary breakpoints: they are pending until the needed runtime
locations are seen.
This patch changes Ada catchpoints to work the same way.
This patch merges create_excep_cond_exprs into ada_catchpoint::re_set.
This is less verbose and is also a step toward making ada_catchpoint
work more like the other code_breakpoint-based exception catchpoints.
gnat_runtime_has_debug_info starts a new gdb to do its work. However,
it also leaves this gdb running, which can potentially confuse the
calling test -- I encountered this when writing a new DAP test. This
patch changes the proc to shut down gdb.
With a gdb 13.2 based package on SLE-15 aarch64, I run into:
...
(gdb) PASS: gdb.python/py-rbreak.exp: nosharedlibrary
py sl = gdb.rbreak("^[^_]",minsyms=False)^M
Breakpoint 2 at 0x4004ac: file ../sysdeps/aarch64/crti.S, line 63.^M
...
(gdb) py print(len(sl))^M
12^M
(gdb) FAIL: gdb.python/py-rbreak.exp: check number of returned breakpoints is 11
...
The FAIL is due to:
- the glibc object crti.o containing debug information for function
call_weak_fn, and
- the test-case not expecting this.
The debug information is there due to compiling glibc using a binutils which
contains commit 591cc9fbbf ("gas/Dwarf: record functions").
I've run into a similar issue before, see commit 3fbbcf473a ("[gdb/testsuite]
Fix regexp in py-rbreak.exp").
The fix I applied there was to use a regexp "^[^_]" to filter out
__libc_csu_fini and __libc_csu_init, but that doesn't work for call_weak_fn.
Fix this by:
- reverting the regexp to "", and
- rewriting the check to require at least 11 functions, rather than a precise
match.
Tested on x86_64-linux.
PR testsuite/30538
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30538
With a gdb 13.2 based package on openSUSE Tumbleweed i586, I ran into:
...
(gdb) run ^M
Starting program: out_of_line_in_inlined/foo_o224_021-all ^M
[Thread debugging using libthread_db enabled]^M
Using host libthread_db library "/lib/libthread_db.so.1".^M
^M
Breakpoint 1.1, foo_o224_021.child1.child2 (s=...) at foo_o224_021.adb:26^M
26 for C of S loop^M
(gdb) FAIL: gdb.ada/out_of_line_in_inlined.exp: scenario=all: \
run to foo_o224_021.child1.child2
...
I can reproduce the same issue with gdb trunk on x86_64, by using optimize=-O3
instead of optimize=-O2.
Fix this by using $bkptno_num_re.
Tested on x86_64-linux.
PR testsuite/30539
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30539
Replace macro HELP_ATTRIBUTE_MODE with a std::string.
Tested on x86_64-linux.
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
Reviewed-By: Tom Tromey <tom@tromey.com>
This commit implements a fix for a bug reported against GDB on
Fedora bugzilla...
https://bugzilla.redhat.com/show_bug.cgi?id=2166796
The test case in that bug report involved running gdb against the 'jq'
program (which is a command-line JSON processor) on Fedora 37. Since
the debug info is compiler (and compile-time option) dependent, it
won't necessarily show up in other distributions or even past or
future versions of Fedora. (E.g. when trying the example shown below
on Fedora 38, GDB says that the value of 'value' has been optimized
out. I.e. it does not demonstrate the same DWARF error that can be
see when using Fedora 37.)
That said, on Fedora 37, the bug could be reproduced as follows:
[kev@f37-1 ~]$ gdb jq -q -ex 'b src/util.c:415' -ex 'r </dev/null'
Reading symbols from jq...
This GDB supports auto-downloading debuginfo from the following URLs:
<https://debuginfod.fedoraproject.org/>
Enable debuginfod for this session? (y or [n]) y
Debuginfod has been enabled.
To make this setting permanent, add 'set debuginfod enabled on' to .gdbinit.
Reading symbols from /home/kev/.cache/debuginfod_client/9d3c8b4197350a190a74972d481de32abf641aa4/debuginfo...
No source file named src/util.c.
Make breakpoint pending on future shared library load? (y or [n]) y
Breakpoint 1 (src/util.c:415) pending.
Starting program: /usr/bin/jq </dev/null
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
Breakpoint 1, jq_util_input_next_input (state=0x55555555d7f0) at src/util.c:416
416 if (state->parser == NULL) {
(gdb) p value
DWARF-2 expression error: DW_OP_GNU_uninit must always be the very last op.
This is undesirable - rather than output an error about the DWARF
info, we'd prefer to see a value, even if it is uninitialized.
Examination of the debuginfo showed the following:
<1><468f1>: Abbrev Number: 112 (DW_TAG_subprogram)
<468f2> DW_AT_external : 1
<468f2> DW_AT_name : (indirect string, offset: 0x4781): jq_util_input_next_input
<468f6> DW_AT_decl_file : 10
<468f6> DW_AT_decl_line : 411
<468f8> DW_AT_decl_column : 4
<468f9> DW_AT_prototyped : 1
<468f9> DW_AT_type : <0x3f2>
<468fd> DW_AT_sibling : <0x4692e>
...
<2><46921>: Abbrev Number: 102 (DW_TAG_variable)
<46922> DW_AT_name : (indirect string, offset: 0x8cb): value
<46926> DW_AT_decl_file : 10
<46926> DW_AT_decl_line : 414
<46928> DW_AT_decl_column : 6
<46929> DW_AT_type : <0x3f2>
Note that there's no DW_AT_location, so I looked for an abstract origin entry:
<2><2dfa0>: Abbrev Number: 90 (DW_TAG_variable)
<2dfa1> DW_AT_abstract_origin: <0x46921>
<2dfa5> DW_AT_location : 0x27cf1 (location list)
<2dfa9> DW_AT_GNU_locviews: 0x27ce1
(Note that the DW_AT_abstract_origin attribute's value is 0x46921 which
is the DIE for the local variable "value".)
Looking at the location list, I see:
00027cf1 v000000000000000 v000000000000000 views at 00027ce1 for:
000000000002f8fe 000000000002f92e (DW_OP_reg13 (r13); DW_OP_GNU_uninit; DW_OP_piece: 8; DW_OP_reg12 (r12); DW_OP_GNU_uninit; DW_OP_piece: 8)
While DW_OP_GNU_uninit is not the very last op, it is the last op
prior to DW_OP_piece. The fix involved changing the DW_OP_GNU_uninit
case in dwarf_expr_context::execute_stack_op in gdb/dwarf2/expr.c so
that DW_OP_GNU_uninit may appear just before DW_OP_piece.
With the fix in place, attempting to print 'value' now looks like
this:
(gdb) p value
$1 = [uninitialized] {kind_flags = 0 '\000', pad_ = 0 '\000', offset = 0,
size = 0, u = {ptr = 0x0, number = 0}}
Note that "[uninitialized]" is part of the output. (But also note
that there's an extra space character.)
I've made a new test case,
gdb.dwarf2/DW_OP_piece_with_DW_OP_GNU_uninit.exp, by adapting an
existing one, gdb.dwarf2/opt-out-not-implptr.exp. Since it uses the
DWARF assembler, the test case does not depend on a specific compiler
version or compiler options.
Tested on Fedora 37 and Fedora 38.
Add a new test plt-findfre-1 to ensure lookup of SFrame stack trace
information for pltN entries is correct.
In this test, a dummy SFrame FDE of type SFRAME_FDE_TYPE_PCMASK is
created. The size of the 'function code block' covered by the SFrame
FDE is equivalent to 5 pltN entries of 16 bytes each.
The test first looks up SFrame FREs for some addresses in the first pltN
entry, followed by lookups for some addresses in the fourth pltN entry.
libsframe/
* Makefile.in: Regenerated.
* testsuite/libsframe.find/find.exp: Add new test.
* testsuite/libsframe.find/local.mk: Likewise.
* testsuite/libsframe.find/plt-findfre-1.c: New test.
To find SFrame stack trace information from an FDE of type
SFRAME_FDE_TYPE_PCMASK, sframe_find_fre () was doing an operation
like,
(start_ip_offset & 0xff) >= (pc & 0xff), etc.
This is buggy and needs correction. The mask 0xff should be 0xf (to
work for a pltN entry of size say, 16 bytes).
At this time, the size of the pltN entry is implicitly assumed to be 16
bytes by libsframe. In next version of the SFrame format, we can encode
this information explicitly in the SFrame FDE.
For now, we should fix the code to at least behave correctly for the
generated code and the generated SFrame stack trace information for the
pltN entries on x86_64.
libsframe/
* sframe.c (sframe_find_fre): Correct the bitmask used for
SFrame FDEs of type SFRAME_FDE_TYPE_PCMASK.
As noted by Tom Tromey, there are some formatting issues with the ternary
operator in the aarch64/arm codebase. This patch fixes those.
Reviewed-By: Tom Tromey <tom@tromey.com>
