| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Currently, if
- you're in all-stop mode,
- the inferior last stopped because of a fork catchpoint,
when you next resume the program, gdb checks whether it had last
stopped for a fork/vfork, and if so,
a) if the current thread is the one that forked, gdb follows the
parent/child, depending on "set follow-fork" mode.
b) if the current thread is some other thread (because you switched
threads meanwhile), gdb switches back to that thread, gdb follows
the parent/child, and stops the resumption command.
There's a problem in b), however -- if you have "set schedule-multiple
off", which is the default, or "set scheduler-locking on", gdb will
still switch back to the forking thread, even if you didn't want to
resume it. For example, with:
(gdb) catch fork
(gdb) c
* thread 1 stops for fork
(gdb) thread 2
(gdb) set scheduler-locking on
(gdb) c
gdb switches back to thread 1, and follows the fork.
Or with:
(gdb) add-inferior -exec prog
(gdb) inferior 2
(gdb) start
(gdb) inferior 1
(gdb) catch fork
(gdb) c
* thread 1.1 stops for fork
(gdb) inferior 2
(gdb) set schedule-multiple off # this is the default
(gdb) c
gdb switches back to thread 1.1, and follows the fork.
Another issue is that, because follow_fork relies on
get_last_target_status to find the thread that has a pending fork, it
is possible to confuse it. For example, "run" or "start" call
init_wait_for_inferior, which clears the last target status, so this:
(gdb) catch fork
(gdb) c
* thread 1 stops for fork
(gdb) add-inferior -exec prog
(gdb) inferior 2
(gdb) start
(gdb) set follow-fork child
(gdb) inferior 1
(gdb) n
... does not follow to the fork child of inferior 1, because the
get_last_target_status call in follow_fork doesn't return a
TARGET_WAITKIND_FORKED. Thanks to Simon for this example.
All of the above are fixed by this patch. It changes follow_fork to
not look at get_last_target_status, but to instead iterate over the
set of threads that the user is resuming, and find the one that has a
pending_follow kind of fork/vfork.
gdb.base/foll-fork.exp is augmented to exercise the last "start"
scenario described above. The other cases will be exercised in the
testcase added by the following patch.
Change-Id: Ifcca77e7b2456277387f40660ef06cec2b93b97e
|
|
With gdb.base/catch-follow-exec.exp, we currently see:
~~~~~~~~~~~~~~~
(gdb)
continue
Continuing.
process 693251 is executing new program: /usr/bin/ls
[New inferior 2]
[New process 693251]
[Switching to process 693251]
Thread 2.1 "ls" hit Catchpoint 2 (exec'd /usr/bin/ls), 0x00007ffff7fd0100 in _start () from /lib64/ld-linux-x86-64.so.2
(gdb)
info prog
No selected thread.
~~~~~~~~~~~~~~~
Note the "No selected thread" output. That is totally bogus, because
there _is_ a selected thread. What GDB really means, is that it can't
find the thread that had the latest (user-visible) stop. And that
happens because "info program" gets that info from
get_last_target_status, and the last target status has been cleared.
However, GDB also checks if there is a selected thread, here:
if (ptid == null_ptid || ptid == minus_one_ptid)
error (_("No selected thread."));
.. the null_ptid part. That is also bogus, because what matters is
the thread that last reported a stop, not the current thread:
- in all-stop mode, "info program" displays info about the last stop.
That may have happened on a thread different from the selected
thread.
- in non-stop mode, because all threads are controlled individually,
"info program" shows info about the last stop of the selected
thread.
The current code already behaves this way, though in a poor way. This
patch reimplements it, such that the all-stop version now finds the
thread that last reported an event via the 'previous_thread' strong
reference. Being a strong reference means that if that thread has
exited since the event was reported, 'previous_thread' will still
point to it, so we can say that the thread exited meanwhile.
The patch also extends "info program" output a little, to let the user
know which thread we are printing info for. For example, for the
gdb.base/catch-follow-exec.exp case we shown above, we now get:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Using the running image of child process 710867.
Program stopped at 0x7ffff7fd0100.
It stopped at breakpoint 2.
Type "info stack" or "info registers" for more information.
(gdb)
while in non-stop mode, we get:
(gdb) info prog
Selected thread 2.1 (process 710867).
Using the running image of child process 710867.
Program stopped at 0x7ffff7fd0100.
It stopped at breakpoint 2.
Type "info stack" or "info registers" for more information.
(gdb)
In both cases, the first line of output is new.
The existing code considered these running/exited cases as an error,
but I think that that's incorrect, since this is IMO just plain
execution info as well. So the patch makes those cases regular
prints, not errors.
If the thread is running, we get, in non-stop mode:
(gdb) info prog
Selected thread 2.1 (process 710867).
Selected thread is running.
... and in all-stop:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Thread is now running.
If the thread has exited, we get, in non-stop mode:
(gdb) info prog
Selected thread 2.1 (process 710867).
Selected thread has exited.
... and in all-stop:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Thread has since exited.
The gdb.base/info-program.exp testcase was much extended to test
all-stop/non-stop and single-threaded/multi-threaded.
Change-Id: I51d9d445f772d872af3eead3449ad4aa445781b1
|
|
Change-Id: Ie8ec8870a16d71c5858f5d08958309d23c318302
Reviewed-By: Tom Tromey <tom@tromey.com>
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
|
|
PR rust/30090 points out that a certain "finish" in a Rust program
will cause gdb to crash. This happens due to some confusion about
field indices in rust_language::print_enum. The fix is to use
value_primitive_field so that the correct type can be passed; other
spots in rust-lang.c already do this.
Note that the enclosed test case comes with an xfail. This is needed
because for this function, rustc doesn't follow the platform ABI.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30090
|
|
A range type can usually be treated the same as its underlying integer
type, at least for the purposes of agent expressions. This patch
arranges for range types to be handled this way in ax-gdb.c, letting a
somewhat larger subset of Ada expressions be compiled.
|
|
Ada historically has not implemented agent expressions, and some Ada
constructs probably cannot reasonably be converted to agent
expressions. However, a subset of simple operations can be, and this
patch represents a first step in that direction.
On one internal AdaCore test case, this improves the performance of a
conditional breakpoint from 5 minutes to 5 seconds.
The main tricky part in this patch is ensuring the converted
expressions detect the cases that will not work. This is done by
examining the code in the corresponding evaluation methods.
|
|
In C++ it is possible to use an empty enum as a strong typedef. For
example, a user could write:
enum class my_type : unsigned char {};
Now my_type can be used like 'unsigned char' except the compiler will
not allow implicit conversion too and from the native 'unsigned char'
type.
This is used in the standard library for things like std::byte.
Currently, when GDB prints a value of type my_type, it looks like
this:
(gdb) print my_var
$1 = (unknown: 0x4)
Which isn't great. This gets worse when we consider something like:
std::vector<my_type> vec;
When using a pretty-printer, this could look like this:
std::vector of length 2, capacity 2 = {(unknown: 0x2), (unknown: 0x4)}
Clearly not great. This is described in PR gdb/30148.
The problem here is in dwarf2/read.c, we assume all enums are flag
enums unless we find an enumerator with a non-flag like value.
Clearly an empty enum contains no non-flag values, so we assume the
enum is a flag enum.
I propose adding an extra check here; that is, an empty enum should
never be a flag enum.
With this the above cases look more like:
(gdb) print my_var
$1 = 4
and:
std::vector of length 2, capacity 2 = {2, 4}
Which look much better.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30148
Reviewed-By: Tom Tromey <tom@tromey.com>
|
|
When looking at some failures of gdb.linespec/cp-completion-aliases.exp,
I noticed that when a completion test will fail, it always fails with a
timeout. This is because most completion tests use gdb_test_multiple
and only add a check for the correct output. This commit adds new
options for both, tab and command completion.
For command completion, the new option will check if the prompt was
printed, and fail in this case. This is enough to know that the test has
failed because the check comes after the PASS path. For tab completion,
we have to check if GDB outputted more than just the input line, because
sometimes GDB would have printed a partial line before finishing with
the correct completion.
Approved-By: Tom Tromey <tom@tromey.com>
|
|
In a recent commit I've added:
...
require {expr [have_compile_flag -fsplit-stack]}
...
but actually the expr bit is unnecessary, and we can just use:
...
require {have_compile_flag -fsplit-stack}
...
Reported-By: Tom Tromey <tom@tromey.com>
|
|
The gdb_assert proc under-quotes the expression that is passed in.
This leads to weird code in a couple of spots that tries to
compensate:
gdb_assert {{$all_regs eq $completed_regs}} ...
The fix is to add a bit of quoting when evaluating the expression.
|
|
Now that Tcl has the {*} operator, we can remove the use of eval from
gdb_breakpoint. Tested on x86-64 Fedora 36.
|
|
Instead of only testing this on systems that have a SYS_time syscall,
test it everywhere using the time(2) C function, and in addition, run
the tests again using the SYS_time syscall.
The C variant ensures that if some platform uses some syscall we are
not aware of yet, we'll still exercise it, and likely fail, at which
point we should teach GDB about the syscall.
The explicit syscall variant is useful on platforms where the C
function does not call a syscall at all by default, e.g., on some
systems the C time function wraps an implementation provided by the
vDSO.
Approved-By: Tom de Vries <tdevries@suse.de>
Change-Id: Id4b755d76577d02c46b8acbfa249d9c31b587633
|
|
A while back I discovered that this does not issue an error:
(gdb) p $x = (void * ) 57
$3 = (void *) 0x39
(gdb) p $x + 7 = 3
$6 = (void *) 0x3
This patch fixes the bug.
Regression tested on x86-64 Fedora 36.
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=19312
|
|
With test-cases gdb.arch/aarch64-mte-core.exp and gdb.arch/aarch64-pauth.exp I
run into compilation errors due to unsupported compilation flags.
Fix this by requiring the compilation flags, such that I have instead:
...
UNSUPPORTED: gdb.arch/aarch64-mte-core.exp: require failed: \
have_compile_flag -march=armv8.5-a+memtag
UNSUPPORTED: gdb.arch/aarch64-pauth.exp: require failed: \
have_compile_flag -mbranch-protection=pac-ret+leaf
...
Tested on aarch64-linux.
|
|
gdb.reverse/step-indirect-call-thunk.exp
On aarch64-linux, I run into:
...
Running gdb.reverse/step-indirect-call-thunk.exp ...
gdb compile failed, gcc: error: unrecognized command line option \
'-mindirect-branch=thunk'; did you mean '-findirect-inlining'?
gcc: error: unrecognized command line option '-mfunction-return=thunk'; \
did you mean '-Wfunction-elimination'?
UNTESTED: gdb.reverse/step-indirect-call-thunk.exp: failed to prepare
...
Fix this by requiring istarget "x86*", similar to what was added in
gdb.base/step-indirect-call-thunk.exp by commit 43127ae5714 ("Fix
gdb.base/step-indirect-call-thunk.exp"), such that we have instead:
...
UNSUPPORTED: gdb.reverse/step-indirect-call-thunk.exp: require failed: \
istarget "x86*
...
Tested on x86_64-linux and aarch64-linux.
|
|
On aarch64-linux, I run into:
...
gdb compile failed, cc1: error: '-fsplit-stack' is not supported by this \
compiler configuration
UNTESTED: gdb.base/morestack.exp: failed to prepare
...
Fix this by requiring -fsplit-stack, such that we have instead:
...
UNSUPPORTED: gdb.base/morestack.exp: require failed: \
expr [have_compile_flag -fsplit-stack]
...
Tested on x86_64-linux and aarch64-linux.
|
|
On aarch64-linux, I run into:
...
Running gdb.reverse/time-reverse.exp ...
gdb compile failed, gdb.reverse/time-reverse.c: In function 'main':
gdb.reverse/time-reverse.c:39:12: error: 'SYS_time' undeclared \
(first use in this function); did you mean 'SYS_times'?
syscall (SYS_time, &time_global);
^~~~~~~~
SYS_times
gdb.reverse/time-reverse.c:39:12: note: each undeclared identifier is \
reported only once for each function it appears in
UNTESTED: gdb.reverse/time-reverse.exp: failed to prepare
...
Fix this by adding a new proc have_syscall, and requiring syscall time, such
that we have instead:
...
UNSUPPORTED: gdb.reverse/time-reverse.exp: require failed: \
expr [have_syscall time]
...
Tested on x86_64-linux and aarch64-linux.
|
|
When running test-case gdb.dap/basic-dap.exp with a gdb without python
support, I run into:
...
builtin_spawn gdb -nw -nx -iex set height 0 -iex set width 0 \
-data-directory data-directory -iex set debug dap-log-file dap.log.1 -q \
-i=dap
>>> {"seq": 1, "type": "request", "command": "initialize"}
Interpreter `dap' unrecognized
ERROR: eof reading json header
...
Fix this by requiring python in the test-case.
Tested on x86_64-linux, both with a gdb without and with python.
|
|
For a -g0 -fasynchronous-unwind-tables exec (without .debug_info but with
.eh_frame section), start using the dwarf2 unwinder instead of the
"amd64 epilogue override" unwinder, by returning true in
compunit_epilogue_unwind_valid for cust == nullptr.
This has effect both on the amd64 and i386 targets, but only add amd64
test-case gdb.base/unwind-on-each-insn-amd64-2.exp.
|
|
I came across:
...
gdb) PASS: gdb.python/py-record-btrace.exp: prepare record: stepi 100
python insn = r.instruction_history^M
warning: Non-contiguous trace at instruction 1 (offset = 0x3e10).^M
(gdb) FAIL: gdb.python/py-record-btrace.exp: prepare record: python insn = r.i\
nstruction_history
...
I'm assuming it's the same root cause as for the already present XFAIL.
Fix this by recognizing above warning in the xfail regexp.
Tested on x86_64-linux, although sofar I was not able to trigger the warning
again.
Approved-By: Markus T. Metzger <markus.t.metzger@intel.com>
|
|
Since commit 9af467b8240 ("[gdb/testsuite] Fix gdb.threads/schedlock.exp on
fast cpu"), the test-case fails for gcc 4.8.5.
The problem is that for gcc 4.8.5, the commit turned a two-line loop:
...
(gdb) next
78 while (*myp > 0)
(gdb) next
81 MAYBE_CALL_SOME_FUNCTION(); (*myp) ++;
(gdb) next
78 while (*myp > 0)
...
into a three-line loop:
...
(gdb) next
83 MAYBE_CALL_SOME_FUNCTION(); (*myp) ++;
(gdb) next
84 cnt++;
(gdb) next
85 }
(gdb) next
83 MAYBE_CALL_SOME_FUNCTION(); (*myp) ++;
(gdb)
...
and the test-case doesn't expect this.
Fix this by reverting back to the original loop shape as much as possible by:
- removing the cnt++ line
- replacing "while (1)" with "while (one)", where one is a volatile variable
set to 1.
Tested on x86_64-linux, using compilers:
- gcc 4.8.5, 7.5.0, 12.2.1
- clang 4.0.1, 13.0.1
|
|
The new DWARF indexer broke "start" for some languages.
For D, it is broken because, while the code in cooked_index_shard::add
specifically excludes Ada, it fails to exclude D. This means that the
C "main" will be detected as "main" here -- whereas what is intended
is for the code in find_main_name to use d_main_name to find the name.
The Rust compiler, on the other hand, uses DW_AT_main_subprogram.
However, the code in dwarf2_build_psymtabs_hard fails to create a
fully-qualified name, so the name always ends up as plain "main".
For D and Ada, a very simple approach suffices: remove the check
against "main" from cooked_index_shard::add. This also has the
benefit of slightly speeding up DWARF indexing. I assume this
approach will work for Pascal and Modula-2 as well, but I don't have a
way to test those at present.
For Rust, though, this is not sufficient. And, computing the
fully-qualified name in dwarf2_build_psymtabs_hard will crash, because
cooked_index_entry::full_name uses the canonical name -- and that is
not computed until after canonicalization.
However, we don't want to wait for canonicalization to be done before
computing the main name. That would remove any benefit from doing
canonicalization is the background.
This patch solves this dilemma by noticing that languages using
DW_AT_main_subprogram are, currently, disjoint from languages
requiring canonicalization. Because of this, we can add a parameter
to full_name to let us avoid crashes, slowdowns, and races here.
This is kind of tricky and ugly, so I've tried to comment it
sufficiently.
While doing this, I had to change gdb.dwarf2/main-subprogram.exp. A
different possibility here would be to ignore the canonicalization
needs of C in this situation, because those only affect certain types.
However, I chose this approach because the test case is artificial
anyhow.
A long time ago, in an earlier threading attempt, I changed the global
current_language to be a function (hidden behind a macro) to let us
attempt lazily computing the current language. Perhaps this approach
could still be made to work. However, that also seemed rather tricky,
more so than this patch.
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30116
|
|
go_symbol_package_name package name asserts that it is only passed a
Go symbol, but this is not enforced by one caller. It seems simplest
to just check and return early in this case.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=17876
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
|
|
A regression in gdb.xml/maint_print_struct.exp was introduced with
commit:
commit 81b86eced24f905545b58aa6c27478104c364976
Date: Fri Jan 6 09:30:40 2023 -0700
Do not record a rejected target description
The test relied on an invalid target description being stored within
the tdesc_info of the current inferior, the above commit stopped this
behaviour.
Update the test to check that the invalid architecture is NOT stored,
and then check printing the target description directly from the
file.
Approved-By: Tom Tromey <tromey@adacore.com>
|
|
On SLE-11, with glibc 2.11.3, I run into:
...
(gdb) PASS: gdb.arch/amd64-disp-step-avx.exp: vex3: \
var128 has expected value after
continue^M
Continuing.^M
^M
Program received signal SIGSEGV, Segmentation fault.^M
0x0000000000400283 in _exit (status=0) at \
../sysdeps/unix/sysv/linux/_exit.c:33^M
33 ../sysdeps/unix/sysv/linux/_exit.c: No such file or directory.^M
(gdb) FAIL: gdb.arch/amd64-disp-step-avx.exp: \
continue until exit at amd64-disp-step-avx
...
This is not related to gdb, we get the same result by just running the exec.
The problem is that the test-case:
- calls glibc's _exit, and
- uses -nostartfiles -static, putting the burden for any necessary
initialization for calling glibc's _exit on the test-case itself.
So, when we get to the second insn in _exit:
...
000000000040acb0 <_exit>:
40acb0: 48 63 d7 movslq %edi,%rdx
40acb3: 64 4c 8b 14 25 00 00 mov %fs:0x0,%r10
...
no glibc-related initialization is done, and we run into the segfault.
Adding this (borrowed from __libc_start_main) in _start in the .S file is
sufficient to fix it:
...
.rept 200
nop
+ call __pthread_initialize_minimal
.endr
...
But that already doesn't compile with say glibc 2.31, and regardless I think
this sort of fix is too fragile.
We could of course fix this by simply not running to exit. But ideally we'd
have an exec that doesn't segfault when you just run it.
Alternatively, we could hand-code an _exit syscall and bypass glibc
all together. But I'd rather fix this in a way that simplifies the test-case.
Taking a step back, the -nostartfiles -static was added to address that the
xmm registers were not zero at main (which AFAICT is a valid thing to happen).
[ The change itself silently broke the test-case, needing further fixing by
commit 40310f30a51 ("gdb: make gdb.arch/amd64-disp-step-avx.exp actually test
displaced stepping"). ]
Instead, simplify things by reverting to the original situation:
- no -nostartfiles -static compilation flags,
- no _start in the .S file,
- use exit instead of _exit in the .S file,
and fix the original problem by setting the xmm registers to zero rather than
checking that they're zero.
Now that we're no longer forcing -static, add nopie to the flags to prevent
compilation failure with target board unix/-fPIE/-pie.
Tested on x86_64-linux.
PR testsuite/30132
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30132
|
|
I noticed that if Ctrl-C was typed just while GDB is evaluating a
breakpoint condition in the background, and GDB ends up reaching out
to the Python interpreter, then the breakpoint condition would still
fail, like:
c&
Continuing.
(gdb) Error in testing breakpoint condition:
Quit
That happens because while evaluating the breakpoint condition, we
enter Python, and end up calling PyErr_SetInterrupt (it's called by
gdbpy_set_quit_flag, in frame #0):
(top-gdb) bt
#0 gdbpy_set_quit_flag (extlang=0x558c68f81900 <extension_language_python>) at ../../src/gdb/python/python.c:288
#1 0x0000558c6845f049 in set_quit_flag () at ../../src/gdb/extension.c:785
#2 0x0000558c6845ef98 in set_active_ext_lang (now_active=0x558c68f81900 <extension_language_python>) at ../../src/gdb/extension.c:743
#3 0x0000558c686d3e56 in gdbpy_enter::gdbpy_enter (this=0x7fff2b70bb90, gdbarch=0x558c6ab9eac0, language=0x0) at ../../src/gdb/python/python.c:212
#4 0x0000558c68695d49 in python_on_memory_change (inferior=0x558c6a830b00, addr=0x555555558014, len=4, data=0x558c6af8a610 "") at ../../src/gdb/python/py-inferior.c:146
#5 0x0000558c6823a071 in std::__invoke_impl<void, void (*&)(inferior*, unsigned long, long, unsigned char const*), inferior*, unsigned long, long, unsigned char const*> (__f=@0x558c6a8ecd98: 0x558c68695d01 <python_on_memory_change(inferior*, CORE_ADDR, ssize_t, bfd_byte const*)>) at /usr/include/c++/11/bits/invoke.h:61
#6 0x0000558c68237591 in std::__invoke_r<void, void (*&)(inferior*, unsigned long, long, unsigned char const*), inferior*, unsigned long, long, unsigned char const*> (__fn=@0x558c6a8ecd98: 0x558c68695d01 <python_on_memory_change(inferior*, CORE_ADDR, ssize_t, bfd_byte const*)>) at /usr/include/c++/11/bits/invoke.h:111
#7 0x0000558c68233e64 in std::_Function_handler<void (inferior*, unsigned long, long, unsigned char const*), void (*)(inferior*, unsigned long, long, unsigned char const*)>::_M_invoke(std::_Any_data const&, inferior*&&, unsigned long&&, long&&, unsigned char const*&&) (__functor=..., __args#0=@0x7fff2b70bd40: 0x558c6a830b00, __args#1=@0x7fff2b70bd38: 93824992247828, __args#2=@0x7fff2b70bd30: 4, __args#3=@0x7fff2b70bd28: 0x558c6af8a610 "") at /usr/include/c++/11/bits/std_function.h:290
#8 0x0000558c6830a96e in std::function<void (inferior*, unsigned long, long, unsigned char const*)>::operator()(inferior*, unsigned long, long, unsigned char const*) const (this=0x558c6a8ecd98, __args#0=0x558c6a830b00, __args#1=93824992247828, __args#2=4, __args#3=0x558c6af8a610 "") at /usr/include/c++/11/bits/std_function.h:590
#9 0x0000558c6830a620 in gdb::observers::observable<inferior*, unsigned long, long, unsigned char const*>::notify (this=0x558c690828c0 <gdb::observers::memory_changed>, args#0=0x558c6a830b00, args#1=93824992247828, args#2=4, args#3=0x558c6af8a610 "") at ../../src/gdb/../gdbsupport/observable.h:166
#10 0x0000558c68309d95 in write_memory_with_notification (memaddr=0x555555558014, myaddr=0x558c6af8a610 "", len=4) at ../../src/gdb/corefile.c:363
#11 0x0000558c68904224 in value_assign (toval=0x558c6afce910, fromval=0x558c6afba6c0) at ../../src/gdb/valops.c:1190
#12 0x0000558c681e3869 in expr::assign_operation::evaluate (this=0x558c6af8e150, expect_type=0x0, exp=0x558c6afcfe60, noside=EVAL_NORMAL) at ../../src/gdb/expop.h:1902
#13 0x0000558c68450c89 in expr::logical_or_operation::evaluate (this=0x558c6afab060, expect_type=0x0, exp=0x558c6afcfe60, noside=EVAL_NORMAL) at ../../src/gdb/eval.c:2330
#14 0x0000558c6844a896 in expression::evaluate (this=0x558c6afcfe60, expect_type=0x0, noside=EVAL_NORMAL) at ../../src/gdb/eval.c:110
#15 0x0000558c6844a95e in evaluate_expression (exp=0x558c6afcfe60, expect_type=0x0) at ../../src/gdb/eval.c:124
#16 0x0000558c682061ef in breakpoint_cond_eval (exp=0x558c6afcfe60) at ../../src/gdb/breakpoint.c:4971
...
The fix is to disable cooperative SIGINT handling while handling
inferior events, so that SIGINT is saved in the global quit flag, and
not in the extension language, while handling an event.
This commit augments the testcase added by the previous commit to test
this scenario as well.
Approved-By: Tom Tromey <tom@tromey.com>
Change-Id: Idf8ab815774ee6f4b45ca2d0caaf30c9b9f127bb
|
|
This implements what I suggested here:
https://inbox.sourceware.org/gdb-patches/ab97c553-f406-b094-cdf3-ba031fdea925@palves.net/
Here is the current default quit_handler, a function that ends up
called by the QUIT macro:
void
default_quit_handler (void)
{
if (check_quit_flag ())
{
if (target_terminal::is_ours ())
quit ();
else
target_pass_ctrlc ();
}
}
As we can see above, when the inferior is running in the foreground,
then a Ctrl-C is translated into a call to target_pass_ctrlc().
The target_terminal::is_ours() case above is there to handle the
scenario where GDB has the terminal, meaning it is handling some
command the user typed, like "list", or "p a + b" or some such.
However, when the inferior is running on the background, say with
"c&", GDB also has the terminal. Run control handling is now done in
the "background". The CLI is responsive to user commands. If users
type Ctrl-C, they're expecting it to interrupt whatever command they
next type in the CLI, which again, could be "list", "p a + b", etc.
It's as if background run control was handled by a separate thread,
and the Ctrl-C is meant to go to the main thread, handling the CLI.
However, when handling an event, inside fetch_inferior_event &
friends, a Ctrl-C _also_ results in a Quit exception, from the same
default_quit_handler function shown above. This quit aborts run
control handling, breakpoint condition evaluation, etc., and may even
leave run control in an inconsistent state.
The testcase added by this patch illustrates this. The test program
just loops a number of times calling the "foo" function.
The idea is to set a breakpoint in the "foo" function with a condition
that sends SIGINT to GDB, and then evaluates to false, which results
in the program being re-resumed in the background. The SIGINT-sending
emulates pressing Ctrl-C just while GDB was evaluating the breakpoint
condition, except, it's more deterministic.
It looks like this:
(gdb) p $counter = 0
$1 = 0
(gdb) b foo if $counter++ == 10 || $_shell("kill -SIGINT `pidof gdb`") != 0
Breakpoint 2 at 0x555555555131: file gdb.base/bg-exec-sigint-bp-cond.c, line 21.
(gdb) c&
Continuing.
(gdb)
After that background continue, the breakpoint should be hit 10 times,
and we should see 10 "Quit" being printed on the screen. As if the
user typed Ctrl-C on the prompt a number of times with no inferior
running:
(gdb) <<< Ctrl-C
(gdb) Quit <<< Ctrl-C
(gdb) Quit <<< Ctrl-C
(gdb)
However, here's what you see instead:
(gdb) c&
Continuing.
(gdb) Quit
(gdb)
Just one Quit, and nothing else. If we look at the thread's state, we see:
(gdb) info threads
Id Target Id Frame
* 1 Thread 0x7ffff7d6f740 (LWP 112192) "bg-exec-sigint-" foo () at gdb.base/bg-exec-sigint-bp-cond.c:21
So the thread stopped, but we didn't report a stop...
Issuing another continue shows the same immediate-and-silent-stop:
(gdb) c&
Continuing.
(gdb) Quit
(gdb) p $counter
$2 = 2
As mentioned, since the run control handling, and breakpoint and
watchpoint evaluation, etc. are running in the background from the
perspective of the CLI, when users type Ctrl-C in this situation,
they're thinking of aborting whatever other command they were typing
or running at the prompt, not the run control side, not the previous
"c&" command.
So I think that we should install a custom quit_handler while inside
fetch_inferior_event, where we already disable pagination and other
things for a similar reason. This custom quit handler does nothing if
GDB has the terminal, and forwards Ctrl-C to the inferior otherwise.
With the patch implementing that, and the same testcase, here's what
you see instead:
(gdb) p $counter = 0
$1 = 0
(gdb) b foo if $counter++ == 10 || $_shell("kill -SIGINT `pidof gdb`") != 0
Breakpoint 2 at 0x555555555131: file gdb.base/bg-exec-sigint-bp-cond.c, line 21.
(gdb) c&
Continuing.
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb) Quit
(gdb)
Breakpoint 2, foo () at gdb.base/bg-exec-sigint-bp-cond.c:21
21 return 0;
Approved-By: Tom Tromey <tom@tromey.com>
Change-Id: I1f10d99496a7d67c94b258e45963e83e439e1778
|
|
For testing a following patch, I wanted a way to send a SIGINT to GDB
from a breakpoint condition. And I didn't want to do it from a Python
breakpoint or Python function, as I wanted to exercise non-Python code
paths. So I thought I'd add a new $_shell internal function, that
runs a command under the shell, and returns the exit code. With this,
I could write:
(gdb) b foo if $_shell("kill -SIGINT $gdb_pid") != 0 || <other condition>
I think this is generally useful, hence I'm proposing it here.
Here's the new function in action:
(gdb) p $_shell("true")
$1 = 0
(gdb) p $_shell("false")
$2 = 1
(gdb) p $_shell("echo hello")
hello
$3 = 0
(gdb) p $_shell("foobar")
bash: line 1: foobar: command not found
$4 = 127
(gdb) help function _shell
$_shell - execute a shell command and returns the result.
Usage: $_shell (command)
Returns the command's exit code: zero on success, non-zero otherwise.
(gdb)
NEWS and manual changes included.
Approved-By: Andrew Burgess <aburgess@redhat.com>
Approved-By: Tom Tromey <tom@tromey.com>
Approved-By: Eli Zaretskii <eliz@gnu.org>
Change-Id: I7e36d451ee6b428cbf41fded415ae2d6b4efaa4e
|
|
Currently, printing the type of an internal function in Ada shows
double <>s, like:
(gdb) with language ada -- ptype $_isvoid
type = <<internal function>>
while all other languages print it with a single <>, like:
(gdb) with language c -- ptype $_isvoid
type = <internal function>
I don't think there's a reason that Ada needs to be different. We
currently print the double <>s because we take this path in
ada_print_type:
switch (type->code ())
{
default:
gdb_printf (stream, "<");
c_print_type (type, "", stream, show, level, language_ada, flags);
gdb_printf (stream, ">");
break;
... and the type's name already has the <>s.
Fix this by simply adding an early check for
TYPE_CODE_INTERNAL_FUNCTION.
Approved-By: Andrew Burgess <aburgess@redhat.com>
Approved-By: Tom Tromey <tom@tromey.com>
Change-Id: Ic2b6527b9240a367471431023f6e27e6daed5501
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30105
|
|
Currently, looking at the type of an internal function, like below,
hits an odd error:
(gdb) ptype $_isvoid
type = <internal function>type not handled in c_type_print_varspec_prefix()
That is an error thrown from
c-typeprint.c:c_type_print_varspec_prefix, where it reads:
...
case TYPE_CODE_DECFLOAT:
case TYPE_CODE_FIXED_POINT:
/* These types need no prefix. They are listed here so that
gcc -Wall will reveal any types that haven't been handled. */
break;
default:
error (_("type not handled in c_type_print_varspec_prefix()"));
break;
Internal function types have type code TYPE_CODE_INTERNAL_FUNCTION,
which is not explicitly handled by that switch.
That comment quoted above says that gcc -Wall will reveal any types
that haven't been handled, but that's not actually true, at least with
modern GCCs. You would need to enable -Wswitch-enum for that, which
we don't. If I do enable that warning, then I see that we're missing
handling for the following type codes:
TYPE_CODE_INTERNAL_FUNCTION,
TYPE_CODE_MODULE,
TYPE_CODE_NAMELIST,
TYPE_CODE_XMETHOD
TYPE_CODE_MODULE and TYPE_CODE_NAMELIST and Fortran-specific, so it'd
be a little weird to handle them here.
I tried to reach this code with TYPE_CODE_XMETHOD, but couldn't figure
out how to. ptype on an xmethod isn't treated specially, it just
complains that the method doesn't exist. I've extended the
gdb.python/py-xmethods.exp testcase to make sure of that.
My thinking is that whatever type code we add next, the most likely
scenario is that it won't need any special handling, so we'd just be
adding another case to that "do nothing" list. If we do need special
casing for whatever type code, I think that tests added at the same
time as the feature would uncover it anyhow. If we do miss adding the
special casing, then it still looks better to me to print the type
somewhat incompletely than to error out and make it harder for users
to debug whatever they need. So I think that the best thing to do
here is to just remove all those explicit "do nothing" cases, along
with the error default case.
After doing that, I decided to write a testcase that iterates over all
supported languages doing "ptype INTERNAL_FUNC". That revealed that
Pascal has a similar problem, except the default case hits a
gdb_assert instead of an error:
(gdb) with language pascal -- ptype $_isvoid
type =
../../src/gdb/p-typeprint.c:268: internal-error: type_print_varspec_prefix: unexpected type
A problem internal to GDB has been detected,
further debugging may prove unreliable.
That is fixed by this patch in the same way.
You'll notice that the new testcase special-cases the Ada expected
output:
} elseif {$lang == "ada"} {
gdb_test "ptype \$_isvoid" "<<internal function>>"
} else {
gdb_test "ptype \$_isvoid" "<internal function>"
}
That will be subject of the following patch.
Approved-By: Andrew Burgess <aburgess@redhat.com>
Change-Id: I81aec03523cceb338b5180a0b4c2e4ad26b4c4db
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30105
|
|
There's a HW bug affecting Processor Trace on some Intel processors
(Ice Lake to Raptor Lake microarchitectures).
The bug was exposed by linux kernel commit 670638477aed
("perf/x86/intel/pt: Opportunistically use single range output mode"),
added in version v5.5.0, and was worked around by commit ce0d998be927
("perf/x86/intel/pt: Fix sampling using single range output") in version
6.1.0.
The bug manifests (on a Performance-core of an i7-1250U, an Alder Lake cpu) in
a single test-case:
...
(gdb) python insn = r.instruction_history^M
warning: Decode error (-20) at instruction 33 (offset = 0x3d6a, \
pc = 0x400501): compressed return without call.^M
(gdb) FAIL: gdb.python/py-record-btrace.exp: prepare record: \
python insn = r.instruction_history
...
Add a corresponding XFAIL.
Note that the i7-1250U has both Performance-cores and Efficient-cores, and on
an Efficient-Core the test-case runs without any problems, so if the testsuite
run is not pinned to a specific cpu, the test may either PASS or XFAIL.
Tested on x86_64-linux:
- openSUSE Leap 15.4 with linux kernel version 5.14.21
- openSUSE Tumbleweed with linux kernel version 6.1.8
PR testsuite/30075
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30075
|
|
Factor out new proc linux_kernel_version from test-case
gdb.arch/i386-pkru.exp.
Tested on x86_64-linux.
|
|
This renames all the fields of struct value, in preparation for the
coming changes.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
The previous commit relied on spotting when a Python defined TUI
window factory was deleted. I spotted that the window factories are
not deleted when GDB shuts down its Python environment, they are only
deleted when one window factory replaces another. Consider this
example Python script:
class TestWindowFactory:
def __init__(self, msg):
self.msg = msg
print("Entering TestWindowFactory.__init__: %s" % self.msg)
def __call__(self, tui_win):
print("Entering TestWindowFactory.__call__: %s" % self.msg)
return TestWindow(tui_win, self.msg)
def __del__(self):
print("Entering TestWindowFactory.__del__: %s" % self.msg)
gdb.register_window_type("test_window", TestWindowFactory("A"))
gdb.register_window_type("test_window", TestWindowFactory("B"))
And this GDB session:
(gdb) source tui.py
Entering TestWindowFactory.__init__: A
Entering TestWindowFactory.__init__: B
Entering TestWindowFactory.__del__: B
(gdb) quit
Notice that when the 'B' window replaces the 'A' window we see the 'A'
object being deleted. But, when Python is shut down (after the
'quit') the 'B' object is never deleted.
Instead, GDB retains a reference to the window factory object, which
forces the Python object to remain live even after the Python
interpreter itself has been shut down.
The references themselves are held in a dynamically allocated
std::unordered_map (in tui/tui-layout.c) which is never deallocated,
thus the underlying Python references are never decremented to zero,
and so GDB never tries to delete these Python objects.
This commit is the first half of the work to clean up this edge case.
All gdbpy_tui_window_maker objects (the objects that implement the
TUI window factory callback for Python defined TUI windows), are now
linked together into a global list using the intrusive list mechanism.
When GDB shuts down the Python interpreter we can now walk this global
list and release the reference that is held to the underlying Python
object. By releasing this reference the Python object will now be
deleted.
I've added a new assert in gdbpy_tui_window_maker::operator(), this
will catch the case where we somehow end up in here after having
reset the reference to the underlying Python object. I don't think
this should ever happen though as we only clear the references when
shutting down the Python interpreter, and the ::operator() function is
only called when trying to apply a new TUI layout - something that
shouldn't happen while GDB itself is shutting down.
This commit does not update the std::unordered_map in tui-layout.c,
that will be done in the next commit.
Reviewed-By: Tom Tromey <tom@tromey.com>
|
|
The documentation for gdb.register_window_type says:
"... It's an error to try to replace one of the built-in windows,
but other window types can be replaced. ..."
I take this to mean that if I imported a Python script like this:
gdb.register_window_type('my_window', FactoryFunction)
Then GDB would have a new TUI window 'my_window', which could be
created by calling FactoryFunction(). If I then, in the same GDB
session imported a script which included:
gdb.register_window_type('my_window', UpdatedFactoryFunction)
Then GDB would replace the old 'my_window' factory with my new one,
GDB would now call UpdatedFactoryFunction().
This is pretty useful in practice, as it allows users to iterate on
their window implementation within a single GDB session.
However, right now, this is not how GDB operates. The second call to
register_window_type is basically ignored and the old window factory
is retained.
This is because in tui_register_window (tui/tui-layout.c) we use
std::unordered_map::emplace to insert the new factory function, and
emplace doesn't replace an existing element in an unordered_map.
In this commit, before the emplace call, I now search for an already
existing element, and delete any matching element from the map, the
emplace call will then add the new factory function.
Reviewed-By: Tom Tromey <tom@tromey.com>
|
|
A new guile test added in commit:
commit 0a9ccb9dd79384f3ba3f8cd75940e8868f3b526f
Date: Mon Feb 6 13:04:16 2023 +0000
gdb: only allow one of thread or task on breakpoints or watchpoints
fails for some versions of guile. It turns out that some versions of
guile emit an error like this:
(gdb) guile (set-breakpoint-thread! bp 1)
ERROR: In procedure set-breakpoint-thread!:
In procedure gdbscm_set_breakpoint_thread_x: cannot set both task and thread attributes
Error while executing Scheme code.
while other versions of guile emit the error like this:
(gdb) guile (set-breakpoint-thread! bp 1)
ERROR: In procedure set-breakpoint-thread!:
ERROR: In procedure gdbscm_set_breakpoint_thread_x: cannot set both task and thread attributes
Error while executing Scheme code.
notice the extra 'ERROR: ' on the second line of output. This commit
updates the test regexp to handle this optional 'ERROR: ' string.
|
|
If the hipcc compiler cannot be found in dejagnu's tool_root_dir, look
for it in $::env(ROCM_PATH) (if set). If hipcc is still not found,
fallback to "hipcc" so the compiler will be searched in the PATH. This
removes the fallback to the hard-coded "/opt/rocm/bin" prefix.
This change is done so ROCM tools are searched in a uniform manner.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
Update allow_hipcc_tests so all gdb.rocm tests are skipped if we do not
have a working hipcc compiler available.
To achieve this, adjust gdb_simple_compile to ensure that the hip
program is saved in a ".cpp" file before calling hipcc otherwise
compilation will fail.
One thing to note is that it is possible to have a hipcc installed with
a CUDA backend. Compiling with this back-end will successfully result
in an application, but GDB cannot debug it (at least for the offload
part). In the context of the gdb.rocm tests, we want to detect such
situation where gdb_simple_compile would give a false positive.
To achieve this, this patch checks that there is at least one AMDGPU
device available and that hipcc can compile for this or those targets.
Detecting the device is done using the rocm_agent_enumerator tool which
is installed with the all ROCm installations (it is used by hipcc to
detect identify targets if this is not specified on the comand line).
This patch also makes the allow_hipcc_tests proc a cached proc.
Co-Authored-By: Pedro Alves <pedro@palves.net>
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
Update allow_hipcc_tests to check that GDB has the amd-dbgapi support
built-in. Without this support, all tests using hipcc and the rocm
stack will fail.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
Rename skip_hipcc_tests to allow_hipcc_tests so it can be used as a
"require" predicate in tests.
Use require in gdb.rocm/simple.exp.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
This commit should fix PR gdb/20091, PR gdb/17201, and PR gdb/17071.
Additionally, PR gdb/17199 relates to this area of code, but is more
of a request to refactor some parts of GDB, this commit does not
address that request, but it is probably worth reading that PR when
looking at this commit.
When the current language is C++, and the user places a breakpoint on
a function in a shared library, GDB will currently find two locations
for the breakpoint, one location will be within the function itself as
we would expect, but the other location will be within the PLT table
for the call to the named function. Consider this session:
$ gdb -q /tmp/breakpoint-shlib-func
Reading symbols from /tmp/breakpoint-shlib-func...
(gdb) start
Temporary breakpoint 1 at 0x40112e: file /tmp/breakpoint-shlib-func.cc, line 20.
Starting program: /tmp/breakpoint-shlib-func
Temporary breakpoint 1, main () at /tmp/breakpoint-shlib-func.cc:20
20 int answer = foo ();
(gdb) break foo
Breakpoint 2 at 0x401030 (2 locations)
(gdb) info breakpoints
Num Type Disp Enb Address What
2 breakpoint keep y <MULTIPLE>
2.1 y 0x0000000000401030 <foo()@plt>
2.2 y 0x00007ffff7fc50fd in foo() at /tmp/breakpoint-shlib-func-lib.cc:20
This is not the expected behaviour. If we compile the same test using
a C compiler then we see this:
(gdb) break foo
Breakpoint 2 at 0x7ffff7fc50fd: file /tmp/breakpoint-shlib-func-c-lib.c, line 20.
(gdb) info breakpoints
Num Type Disp Enb Address What
2 breakpoint keep y 0x00007ffff7fc50fd in foo at /tmp/breakpoint-shlib-func-c-lib.c:20
Here's what's happening. When GDB parses the symbols in the main
executable and the shared library we see a number of different symbols
for foo, and use these to create entries in GDB's msymbol table:
- In the main executable we see a symbol 'foo@plt' that points at
the plt entry for foo, from this we add two entries into GDB's
msymbol table, one called 'foo@plt' which points at the plt entry
and has type mst_text, then we create a second symbol, this time
called 'foo' with type mst_solib_trampoline which also points at
the plt entry,
- Then, when the shared library is loaded we see another symbol
called 'foo', this one points at the actual implementation in the
shared library. This time GDB creates a msymbol called 'foo' with
type mst_text that points at the implementation.
This means that GDB creates 3 msymbols to represent the 2 symbols
found in the executable and shared library.
When the user creates a breakpoint on 'foo' GDB eventually ends up in
search_minsyms_for_name (linespec.c), this function then calls
iterate_over_minimal_symbols passing in the name we are looking for
wrapped in a lookup_name_info object.
In iterate_over_minimal_symbols we iterate over two hash tables (using
the name we're looking for as the hash key), first we walk the hash
table of symbol linkage names, then we walk the hash table of
demangled symbol names.
When the language is C++ the symbols for 'foo' will all have been
mangled, as a result, in this case, the iteration of the linkage name
hash table will find no matching results.
However, when we walk the demangled hash table we do find some
results. In order to match symbol names, GDB obtains a symbol name
matching function by calling the get_symbol_name_matcher method on the
language_defn class. For C++, in this case, the matching function we
use is cp_fq_symbol_name_matches, which delegates the work to
strncmp_iw_with_mode with mode strncmp_iw_mode::MATCH_PARAMS and
language set to language_cplus.
The strncmp_iw_mode::MATCH_PARAMS mode means that strncmp_iw_mode will
skip any parameters in the demangled symbol name when checking for a
match, e.g. 'foo' will match the demangled name 'foo()'. The way this
is done is that the strings are matched character by character, but,
once the string we are looking for ('foo' here) is exhausted, if we
are looking at '(' then we consider the match a success.
Lets consider the 3 symbols GDB created. If the function declaration
is 'void foo ()' then from the main executable we added symbols
'_Z3foov@plt' and '_Z3foov', while from the shared library we added
another symbol call '_Z3foov'. When these are demangled they become
'foo()@plt', 'foo()', and 'foo()' respectively.
Now, the '_Z3foov' symbol from the main executable has the type
mst_solib_trampoline, and in search_minsyms_for_name, we search for
any symbols of type mst_solib_trampoline and filter these out of the
results.
However, the '_Z3foov@plt' symbol (from the main executable), and the
'_Z3foov' symbol (from the shared library) both have type mst_text.
During the demangled name matching, due to the use of MATCH_PARAMS
mode, we stop the comparison as soon as we hit a '(' in the demangled
name. And so, '_Z3foov@plt', which demangles to 'foo()@plt' matches
'foo', and '_Z3foov', which demangles to 'foo()' also matches 'foo'.
By contrast, for C, there are no demangled hash table entries to be
iterated over (in iterate_over_minimal_symbols), we only consider the
linkage name symbols which are 'foo@plt' and 'foo'. The plain 'foo'
symbol obviously matches when we are looking for 'foo', but in this
case the 'foo@plt' will not match due to the '@plt' suffix.
And so, when the user asks for a breakpoint in 'foo', and the language
is C, search_minsyms_for_name, returns a single msymbol, the mst_text
symbol for foo in the shared library, while, when the language is C++,
we get two results, '_Z3foov' for the shared library function, and
'_Z3foov@plt' for the plt entry in the main executable.
I propose to fix this in strncmp_iw_with_mode. When the mode is
MATCH_PARAMS, instead of stopping at a '(' and assuming the match is a
success, GDB will instead search forward for the matching, closing,
')', effectively skipping the parameter list, and then resume
matching. Thus, when comparing 'foo' to 'foo()@plt' GDB will
effectively compare against 'foo@plt' (skipping the parameter list),
and the match will fail, just as it does when the language is C.
There is one slight complication, which is revealed by the test
gdb.linespec/cpcompletion.exp, when searching for the symbol of a
const member function, the demangled symbol will have 'const' at the
end of its name, e.g.:
struct_with_const_overload::const_overload_fn() const
Previously, the matching would stop at the '(' character, but after my
change the whole '()' is skipped, and the match resumes. As a result,
the 'const' modifier results in a failure to match, when previously
GDB would have found a match.
To work around this issue, in strncmp_iw_with_mode, when mode is
MATCH_PARAMS, after skipping the parameter list, if the next character
is '@' then we assume we are looking at something like '@plt' and
return a value indicating the match failed, otherwise, we return a
value indicating the match succeeded, this allows things like 'const'
to be skipped.
With these changes in place I now see GDB correctly setting a
breakpoint only at the implementation of 'foo' in the shared library.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=20091
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=17201
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=17071
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=17199
Tested-By: Bruno Larsen <blarsen@redhat.com>
Approved-By: Simon Marchi <simon.marchi@efficios.com>
|
|
After this mailing list posting:
https://sourceware.org/pipermail/gdb-patches/2023-February/196607.html
it seems to me that in practice an Ada task maps 1:1 with a GDB
thread, and so it doesn't really make sense to allow uses to give both
a thread and a task within a single breakpoint or watchpoint
condition.
This commit updates GDB so that the user will get an error if both
are specified.
I've added new tests to cover the CLI as well as the Python and Guile
APIs. For the Python and Guile testing, as far as I can tell, this
was the first testing for this corner of the APIs, so I ended up
adding more than just a single test.
For documentation I've added a NEWS entry, but I've not added anything
to the docs themselves. Currently we document the commands with a
thread-id or task-id as distinct command, e.g.:
'break LOCSPEC task TASKNO'
'break LOCSPEC task TASKNO if ...'
'break LOCSPEC thread THREAD-ID'
'break LOCSPEC thread THREAD-ID if ...'
As such, I don't believe there is any indication that combining 'task'
and 'thread' would be expected to work; it seems clear to me in the
above that those four options are all distinct commands.
I think the NEWS entry is enough that if someone is combining these
keywords (it's not clear what the expected behaviour would be in this
case) then they can figure out that this was a deliberate change in
GDB, but for a new user, the manual doesn't suggest combining them is
OK, and any future attempt to combine them will give an error.
Approved-By: Pedro Alves <pedro@palves.net>
|
|
I noticed that describe_other_breakpoints doesn't show the task
number, but does show the thread-id. I can't see any reason why we'd
want to not show the task number in this situation, so this commit
adds this missing information, and extends gdb.ada/tasks.exp to check
this case.
Approved-By: Pedro Alves <pedro@palves.net>
|
|
I noticed that describe_other_breakpoints was printing the global
thread-id to the CLI. For CLI output we should be printing the
inferior local thread-id (e.g. "2.1"). This can be seen in the
following GDB session:
(gdb) info threads
Id Target Id Frame
1.1 Thread 4065742.4065742 "bp-thread-speci" main () at /tmp/bp-thread-specific.c:27
* 2.1 Thread 4065743.4065743 "bp-thread-speci" main () at /tmp/bp-thread-specific.c:27
(gdb) break foo thread 2.1
Breakpoint 3 at 0x40110a: foo. (2 locations)
(gdb) break foo thread 1.1
Note: breakpoint 3 (thread 2) also set at pc 0x40110a.
Note: breakpoint 3 (thread 2) also set at pc 0x40110a.
Breakpoint 4 at 0x40110a: foo. (2 locations)
Notice that GDB says:
Note: breakpoint 3 (thread 2) also set at pc 0x40110a.
The 'thread 2' in here is using the global thread-id, we should
instead say 'thread 2.1' which corresponds to how the user specified
the breakpoint.
This commit fixes this issue and adds a test.
Approved-By: Pedro Alves <pedro@palves.net>
|
|
The test added in this commit tests for a long fixed readline issue
relating to long command lines. A similar patch has existed in the
Fedora GDB tree for several years, but I don't see any reason why this
test would not be suitable for inclusion in upstream GDB. I've
updated the patch to current testsuite standards.
The test is checking for an issue that was fixed by this readline
patch:
https://lists.gnu.org/archive/html/bug-readline/2006-11/msg00002.html
Which was merged into readline 6.0 (released ~2010). The issue was
triggered when the user enters a long command line, which wrapped over
multiple terminal lines. The crash looks like this:
free(): invalid pointer
Fatal signal: Aborted
----- Backtrace -----
0x4fb583 gdb_internal_backtrace_1
../../src/gdb/bt-utils.c:122
0x4fb583 _Z22gdb_internal_backtracev
../../src/gdb/bt-utils.c:168
0x6047b9 handle_fatal_signal
../../src/gdb/event-top.c:964
0x7f26e0cc56af ???
0x7f26e0cc5625 ???
0x7f26e0cae8d8 ???
0x7f26e0d094be ???
0x7f26e0d10aab ???
0x7f26e0d124ab ???
0x7f26e1d32e12 rl_free_undo_list
../../readline-5.2/undo.c:119
0x7f26e1d229eb readline_internal_teardown
../../readline-5.2/readline.c:405
0x7f26e1d3425f rl_callback_read_char
../../readline-5.2/callback.c:197
0x604c0d gdb_rl_callback_read_char_wrapper_noexcept
../../src/gdb/event-top.c:192
0x60581d gdb_rl_callback_read_char_wrapper
../../src/gdb/event-top.c:225
0x60492f stdin_event_handler
../../src/gdb/event-top.c:545
0xa60015 gdb_wait_for_event
../../src/gdbsupport/event-loop.cc:694
0xa6078d gdb_wait_for_event
../../src/gdbsupport/event-loop.cc:593
0xa6078d _Z16gdb_do_one_eventi
../../src/gdbsupport/event-loop.cc:264
0x6fc459 start_event_loop
../../src/gdb/main.c:411
0x6fc459 captured_command_loop
../../src/gdb/main.c:471
0x6fdce4 captured_main
../../src/gdb/main.c:1310
0x6fdce4 _Z8gdb_mainP18captured_main_args
../../src/gdb/main.c:1325
0x44f694 main
../../src/gdb/gdb.c:32
---------------------
I recreated the above crash by a little light hacking on GDB, and then
linking GDB against readline 5.2. The above stack trace was generated
from the test included in this patch, and matches the trace that was
included in the original bug report.
It is worth acknowledging that without hacking things GDB has a
minimum requirement of readline 7.0. This test is not about checking
whether GDB has been built against an older version of readline, it is
about checking that readline doesn't regress in this area.
Reviewed-By: Tom Tromey <tom@tromey.com>
|
|
This commit introduces the idea of loading only part of an array in
order to print it, what I call "limited length" arrays.
The motivation behind this work is to make it possible to print slices
of very large arrays, where very large means bigger than
`max-value-size'.
Consider this GDB session with the current GDB:
(gdb) set max-value-size 100
(gdb) p large_1d_array
value requires 400 bytes, which is more than max-value-size
(gdb) p -elements 10 -- large_1d_array
value requires 400 bytes, which is more than max-value-size
notice that the request to print 10 elements still fails, even though 10
elements should be less than the max-value-size. With a patched version
of GDB:
(gdb) p -elements 10 -- large_1d_array
$1 = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9...}
So now the print has succeeded. It also has loaded `max-value-size'
worth of data into value history, so the recorded value can be accessed
consistently:
(gdb) p -elements 10 -- $1
$2 = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9...}
(gdb) p $1
$3 = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
20, 21, 22, 23, 24, <unavailable> <repeats 75 times>}
(gdb)
Accesses with other languages work similarly, although for Ada only
C-style [] array element/dimension accesses use history. For both Ada
and Fortran () array element/dimension accesses go straight to the
inferior, bypassing the value history just as with C pointers.
Co-Authored-By: Maciej W. Rozycki <macro@embecosm.com>
|
|
Add a `-nonl' option to `gdb_test' making it possible to match output
from commands such as `output' that do not produce a new line sequence
at the end, e.g.:
(gdb) output 0
0(gdb)
|
|
While it makes sense to allow accessing out-of-bounds elements in the
debuggee and see whatever there might happen to be there in memory (we
are a debugger and not a programming rules enforcement facility and we
want to make people's life easier in chasing bugs), e.g.:
(gdb) print one_hundred[-1]
$1 = 0
(gdb) print one_hundred[100]
$2 = 0
(gdb)
we shouldn't really pretend that we have any meaningful data around
values recorded in history (what these commands really retrieve are
current debuggee memory contents outside the original data accessed,
really confusing in my opinion). Mark values recorded in history as
such then and verify accesses to be in-range for them:
(gdb) print one_hundred[-1]
$1 = <unavailable>
(gdb) print one_hundred[100]
$2 = <unavailable>
Add a suitable test case, which also covers integer overflows in data
location calculation.
Approved-By: Tom Tromey <tom@tromey.com>
|
|
We have an inconsistency in value history accesses where array element
accesses cause an error for entries exceeding the currently selected
`max-value-size' setting even where such accesses successfully complete
for elements located in the inferior, e.g.:
(gdb) p/d one
$1 = 0
(gdb) p/d one_hundred
$2 = {0 <repeats 100 times>}
(gdb) p/d one_hundred[99]
$3 = 0
(gdb) set max-value-size 25
(gdb) p/d one_hundred
value requires 100 bytes, which is more than max-value-size
(gdb) p/d one_hundred[99]
$7 = 0
(gdb) p/d $2
value requires 100 bytes, which is more than max-value-size
(gdb) p/d $2[99]
value requires 100 bytes, which is more than max-value-size
(gdb)
According to our documentation the `max-value-size' setting is a safety
guard against allocating an overly large amount of memory. Moreover a
statement in documentation says, concerning this setting, that: "Setting
this variable does not affect values that have already been allocated
within GDB, only future allocations." While in the implementer-speak
the sentence may be unambiguous I think the outside user may well infer
that the setting does not apply to values previously printed.
Therefore rather than just fixing this inconsistency it seems reasonable
to lift the setting for value history accesses, under an implication
that by having been retrieved from the debuggee they have already passed
the safety check. Do it then, by suppressing the value size check in
`value_copy' -- under an observation that if the original value has been
already loaded (i.e. it's not lazy), then it must have previously passed
said check -- making the last two commands succeed:
(gdb) p/d $2
$8 = {0 <repeats 100 times>}
(gdb) p/d $2 [99]
$9 = 0
(gdb)
Expand the testsuite accordingly, covering both value history handling
and the use of `value_copy' by `make_cv_value', used by Python code.
|
|
Following commit 4e2a80ba606 ("gdb/testsuite: expect SIGSEGV from top
GDB spawn id"), the next failure I get in gdb.gdb/selftest.exp, using
the native-extended-gdbserver, is:
(gdb) PASS: gdb.gdb/selftest.exp: send ^C to child process
signal SIGINT
Continuing with signal SIGINT.
FAIL: gdb.gdb/selftest.exp: send SIGINT signal to child process (timeout)
The problem is that in this gdb_test_multiple:
set description "send SIGINT signal to child process"
gdb_test_multiple "signal SIGINT" "$description" {
-re "^signal SIGINT\r\nContinuing with signal SIGINT.\r\nQuit\r\n.* $" {
pass "$description"
}
}
The "Continuing with signal SIGINT" portion is printed by the top GDB,
while the Quit portion is printed by the bottom GDB. As the
gdb_test_multiple is written, it expects both the the top GDB's spawn
id.
Fix this by splitting the gdb_test_multiple in two. The first one
expects the "Continuing with signal SIGINT" from the top GDB. The
second one expect "Quit" and the "(xgdb)" prompt from
$inferior_spawn_id. When debugging natively, this spawn id will be the
same as the top GDB's spawn id, but it's different when debugging with
GDBserver.
Change-Id: I689bd369a041b48f4dc9858d38bf977d09600da2
|
