With a gdb build with --enable-targets=all we run into a KFAIL:
...
KFAIL: gdb.gdb/unittest.exp: executable loaded: maintenance selftest, \
failed none (PRMS: gdb/27891)
...
due to:
...
Running selftest print_one_insn.^M
Self test failed: arch armv8.1-m.main: self-test failed at \
disasm-selftests.c:165^M
...
The test fails because we expect disassembling of one arm insn to consume 4
bytes and produce (using verbose = true in disasm-selftests.c):
...
arm mov r0, #0
...
but instead the disassembler uses thumb mode and only consumes 2
bytes and produces:
...
arm movs r0, r0
...
The failure does not show up in the "no executable loaded" variant because
this code in gdb_print_insn_arm isn't triggered:
...
if (current_program_space->exec_bfd () != NULL)
info->flags |= USER_SPECIFIED_MACHINE_TYPE;
...
and consequently we do this in print_insn:
...
if ((info->flags & USER_SPECIFIED_MACHINE_TYPE) == 0)
info->mach = bfd_mach_arm_unknown;
...
and don't set force_thumb to true in select_arm_features.
The code in gdb_print_insn_arm makes the assumption that the disassembly
architecture matches the exec architecture, which in this case is incorrect,
because the exec architecture is x86_64, and the disassembly architecture is
armv8.1-m.main. Fix that by explicitly checking it:
...
if (current_program_space->exec_bfd () != NULL
&& (current_program_space->exec_bfd ()->arch_info
== gdbarch_bfd_arch_info (gdbarch)))
...
This fixes the print_one_insn failure, so remove the KFAIL.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27891
With a gdb build with --enable-targets=all, we have 2 arch-specific failures
in selftest print_one_insn:
...
$ gdb -q -batch a.out -ex "maint selftest print_one_insn" 2>&1 \
| grep "Self test failed: arch "
Self test failed: arch armv8.1-m.main: self-test failed at \
disasm-selftests.c:165
Self test failed: arch arm_any: self-test failed at disasm-selftests.c:165
$
...
During the first failed test, force_thumb is set to true, and remains so until
and during the second test, which causes the second failure.
Fix this by resetting force_thumb to false in parse_arm_disassembler_options,
such that we get just one failure:
...
$ gdb -q -batch a.out -ex "maint selftest print_one_insn" 2>&1 \
| grep "Self test failed: arch "
Self test failed: arch armv8.1-m.main: self-test failed at \
disasm-selftests.c:165
$
...
Tested on x86_64-linux.
A build without Python will currently fail, because
selftests::test_python uses gdb_python_initialized, which is only
conditionally defined.
This patch fixes the build by making test_python also be conditionally
defined. I chose this approach because the selftest will fail if
Python is not enabled, so it didn't seem useful to leave it defined.
Since the 0x57 is preserved for the vadd.vv instruction in the integration
branch, remove it to make sure the testcase can work.
gas/
* testsuite/gas/riscv/insn.d: Remove 0x57 since it is preserved
for vadd.vv instruction.
* testsuite/gas/riscv/insn.s: Likewise.
With a gdb build with CFLAGS "-O2 -g -flto=auto", I run into:
...
#7 gdb_main (args=0x7fffffffd220) at src/gdb/main.c:1368^M
#8 main (argc=<optimized out>, argv=<optimized out>) at src/gdb/gdb.c:32^M
(gdb) FAIL: gdb.gdb/selftest.exp: backtrace through signal handler
...
which means that this regexp in proc test_with_self fails:
...
-re "#0.*(read|poll).*in main \\(.*\\) at .*gdb\\.c.*$gdb_prompt $" {
...
The problem is that gdb_main has been inlined into main, and consequently the
backtrace uses:
...
#x <fn> ...
...
instead of
...
#x <address> in <fn> ...
...
Fix this by updating the regexp to not require "in" before " main".
Tested on x86_64-linux.
When running gdb.base/batch-exit-status.exp I noticed that the test name
contains a newline:
...
PASS: gdb.base/batch-exit-status.exp: : No such file or directory\.^M
: No such file or directory\.: [lindex $result 2] == 0
...
The mistake is that I passed an output regexp argument to a parameter
interpreted as testname prefix. Fix this by passing a testname prefix
instead.
Add support for checking output, to be able to handle the output regexp
argument.
Tested on x86_64-linux.
When running test-case gdb.base/batch-exit-status.exp for native, it passes.
But with target board cc-with-debug-names, we run into (added missing double
quotes for clarity):
...
builtin_spawn $build/gdb/testsuite/../../gdb/gdb -nw -nx \
-data-directory $build/gdb/testsuite/../data-directory \
-iex "set height 0" -iex "set width 0" -ex "set sysroot" -batch ""^M
: No such file or directory.^M
PASS: gdb.base/batch-exit-status.exp: \
: No such file or directory\.: [lindex $result 2] == 0
FAIL: gdb.base/batch-exit-status.exp: \
: No such file or directory\.: [lindex $result 3] == $expect_status
...
The difference between the passing and failing case is that with native we
have (leaving out set height/width for brevity):
...
$ gdb -batch ""; echo $?
: No such file or directory.
1
...
and with target board cc-with-debug-names:
...
$ gdb -ex "set sysroot" -batch ""; echo $?
: No such file or directory.
0
...
The difference is expected. GDB returns the exit status of the last executed
command. In the former case that's 'file ""', which fails. In the latter case,
that's 'set sysroot', which succeeds.
Fix this by setting sysroot using -iex instead of -ex in local-board.exp, such
that we have the expected:
...
$ gdb -iex "set sysroot" -batch ""; echo $?
: No such file or directory.
1
...
Tested on x86_64-linux.
It's unclear why -H was picked over the more standard -h, but since
-h is still not used, just change -H to -h to match pretty much every
other tool in the sourceware tree.
The test-case gdb.gdb/python-selftest.exp:
- patches the gdb_python_initialized variable in gdb to 0
- checks that the output of a python command is "Python not initialized"
Reimplement gdb.gdb/python-selftest.exp as unittest, using:
- execute_command_to_string to capture the output
- try/catch to catch the "Python not initialized" exception.
Tested on x86_64-linux.
Fix DUPLICATE in gdb.base/skip-solib.exp by using with_test_prefix.
Also fix indentation style and long lines, remove outdated question/answer
bits, and use multi_line.
The documentation of mi_gdb_test states that the command, pattern and message
arguments are mandatory:
...
# mi_gdb_test COMMAND PATTERN MESSAGE [IPATTERN] -- send a command to gdb;
# test the result.
...
However, this is not checked, and when mi_gdb_test is called with less than 3
arguments, it passes or fails silently.
Fix this by using the following semantics:
- if there are 1 or 2 arguments, use the command as the message.
- if there is 1 argument, use ".*" as the pattern.
- if there are no or too much arguments, error out.
Fix a PATH issue in gdb.mi/mi-logging.exp, introduced by using the command as
message. Fix a few other trivial-looking FAILs.
There are 11 less trivial-looking FAILs left in gdb.mi in test-cases:
- mi-nsmoribund.exp
- mi-breakpoint-changed.exp
- mi-break.exp.
Tested on x86_64-linux.
A regexp pattern with escapes like this is hard to read:
...
set re "~\"\[$\]$decimal = 1\\\\n\"\r\n\\^done"
...
We can make it more readable by spacing out parts (which allows us to also use
the curly braces where that's convenient):
...
set re [list "~" {"} {[$]} $decimal " = 1" "\\\\" "n" {"} "\r\n" "\\^" "done"]
set re [join $re ""]
...
or by using string_to_regexp:
...
set re [list \
[string_to_regexp {~"$}] \
$decimal \
[string_to_regexp " = 1\\n\"\r\n^done"]]
set re [join $re ""]
...
Note: we have to avoid applying string_to_list to decimal, which is already a
regexp.
Add a proc string_list_to_regexp to make it easy to do both:
...
set re [list \
[string_list_to_regexp ~ {"} $] \
$decimal \
[string_list_to_regexp " = 1" \\ n {"} \r\n ^ done]]
...
Also add a test-case gdb.testsuite/string_to_regexp.exp.
When running the gdb testsuite with gnatmake-4.8, I get many fails of the
following form:
...
gcc: error: unrecognized command line option '-fgnat-encodings=all'^M
gnatmake: "gdb.ada/O2_float_param/foo.adb" compilation error^M
compiler exited with status 1
compilation failed: gcc ... gdb.ada/O2_float_param/foo.adb
gcc: error: unrecognized command line option '-fgnat-encodings=all'
gnatmake: "gdb.ada/O2_float_param/foo.adb" compilation error
FAIL: gdb.ada/O2_float_param.exp: scenario=all: compilation foo.adb
...
Fix this by marking the test unsupported instead, such that we have:
...
UNSUPPORTED: gdb.ada/O2_float_param.exp: scenario=all: compilation foo.adb \
(unsupported option '-fgnat-encodings=all')
...
Tested on x86_64-linux.
DWARF5 allows .file 0 to take an optional directory name. Set the entry
0 of the directory table to the directory name in .file 0.
PR gas/28266
* dwarf2dbg.c (get_directory_table_entry): Add an argument for
the directory name in .file 0 and use it, instead of PWD.
(allocate_filenum): Pass NULL to get_directory_table_entry.
(allocate_filename_to_slot): Pass the incoming dirname to
get_directory_table_entry.
* testsuite/gas/elf/dwarf-5-file0-2.d: New file.
* testsuite/gas/elf/dwarf-5-file0-2.s: Likewise.
* testsuite/gas/elf/elf.exp: Run dwarf-5-file0-2.
On openSUSE Leap 42.3 with eu-unstrip 0.158, we run into:
...
(gdb) PASS: gdb.base/coredump-filter-build-id.exp: save corefile
First line of eu-unstrip: \
0x400000+0x202000 f4ae8502bd6a14770182382316bc595e9dc6f08b@0x400284 - - [exe]
FAIL: gdb.base/coredump-filter-build-id.exp: gcore dumped mapping with build-id
...
The test expects an actual file name instead of '[exe]', but that only got
introduced with eu-unstrip 0.161. Before it printed '[exe]' or '[pie]'.
Fix this by updating the regexp.
Tested on x86_64-linux.
I noticed this failure in gdb.mi/mi-sym-info.exp with gcc-4.8:
...
FAIL: gdb.mi/mi-sym-info.exp: -symbol-info-functions --max-results 1 \
(unexpected output)
...
due to function f2 instead of f3 being listed.
AFAICT, this is caused by a difference in debug info:
...
$ readelf -wi outputs/gdb.mi/mi-sym-info/mi-sym-info1.o \
| egrep "DW_AT_name.*: f[1-3]"
<72> DW_AT_name : f1
<a1> DW_AT_name : f2
<d0> DW_AT_name : f3
...
vs:
...
$ readelf -wi outputs/gdb.mi/mi-sym-info/mi-sym-info1.o \
| egrep "DW_AT_name.*: f[1-3]"
<f4> DW_AT_name : f3
<123> DW_AT_name : f2
<152> DW_AT_name : f1
...
and the command documentation does not mention an imposed order, so fix this
by allowing f2 as well.
Doing this fix, it made sense to do a refactoring of adding f2_re and f3_re
variables, in order to write (?:$f2_re|$f3_re), and I applied the same pattern
overall.
Furthermore, I found a silent FAIL due to calling mi_gdb_proc with 2 args, fix
by updating the regexp.
Then I ran with clang and found another FAIL, fix by updating the regexp.
Tested on x86_64-linux with gcc-4.8.5, gcc-7.5.0, gcc-11.2.1, clang-7.0.1 and
clang-12.0.1.
When building gdb with "-Wall -O2 -g -flto=auto", I run into:
...
(gdb) call clear_complaints()^M
No symbol "clear_complaints" in current context.^M
(gdb) FAIL: gdb.gdb/complaints.exp: clear complaints
...
The problem is that lto has optimized away the clear_complaints function
and consequently the selftest doesn't work.
Fix this by reimplementing the selftest as a unit test.
Factor out two new functions:
- void
execute_fn_to_ui_file (struct ui_file *file, std::function<void(void)> fn);
- std::string
execute_fn_to_string (std::function<void(void)> fn, bool term_out);
and use the latter to capture the complaints output.
Tested on x86_64-linux.
Python 2 has a bit flag Py_TPFLAGS_HAVE_ITER which can be passed as
part of the tp_flags field when defining a new object type. This flag
is not defined in Python 3 and so we define it to 0 in
python-internal.h (when IS_PY3K is defined).
The meaning of this flag is that the object has the fields tp_iter and
tp_iternext. Note the use of "has" here, the flag says nothing about
the values in those fields, just that the type object has the fields.
In early versions of Python 2 these fields were no part of the
PyTypeObject struct, they were added in version 2.2 (see
https://docs.python.org/release/2.3/api/type-structs.html). And so,
there could be a some code compiled out there which has a PyTypeObject
structure within it that doesn't even have the tp_iter and tp_iternext
fields, attempting to access these fields would be undefined
behaviour.
And so Python added the Py_TPFLAGS_HAVE_ITER flag. If the flag is
present then Python is free to access the tp_iter and tp_iternext
fields.
If we consider GDB then we always assume that the tp_iter and
tp_iternext fields are part of PyTypeObject. If someone was crazy
enough to try and compile GDB against Python 2.1 then we'd get lots of
build errors saying that we were passing too many fields when
initializing PyTypeObject structures. And so, I claim, we can be sure
that GDB will always be compiled with a version of Python that has the
tp_iter and tp_iternext fields in PyTypeObject.
Next we can look at the Py_TPFLAGS_DEFAULT flag. In Python 2, each
time additional fields are added to PyTypeObject a new Py_TPFLAGS_*
flag would be defined to indicate whether those flags are present or
not. And, those new flags would be added to Py_TPFLAGS_DEFAULT. And
so, in the latest version of Python 2 the Py_TPFLAGS_DEFAULT flag
includes Py_TPFLAGS_HAVE_ITER (see
https://docs.python.org/2.7/c-api/typeobj.html).
In GDB we pass Py_TPFLAGS_DEFAULT as part of the tp_flags for all
objects we define.
And so, in this commit, I propose to remove all uses of
Py_TPFLAGS_HAVE_ITER from GDB, it's simply not needed.
There should be no user visible changes after this commit.
Many GNU tools accept -EB/-EL as short options for selecting big &
little endian modes. While the sim has an -E option, it requires
spelling out "big" and "little". Adding support for -EB & -EL is
thus quite trivial, so lets round it out to be less annoying.
Only the ppc arch supports this kind of source file override logic.
All the others expose knobs via configure flags, and for some of
these, the ppc code does as well. For others, it doesn't make sense
to ever change them. Since it's unlikely anyone is using this, drop
it all to simplify the code (and to get us a little closer to the
common sim code).
We use these older names inconsistently in the sim codebase, and time
has moved on long ago, so drop support for these non-standard names.
POSIX provides O_NONBLOCK for us, so use it everywhere.
We have enough functionality from gnulib now to build sockser on
all platforms.
Non-blocking I/O is supported when F_GETFL/F_SETFL are unavailable,
but we can address that in a follow up commit. This mirrors what
is done in other places in the sim already.
The cgen framework provides a "VOID" type for code to use, but this
defines ends up conflicting with the standard Windows VOID define.
Since they actually define to the same thing ("void"), undef it here
to fix the Windows build.
We might want to reconsider the need for "VOID" in cgen, but that
will take larger discussion & coordination with the cgen project.
The sim-main.h header is a bit of a dumping ground. Every arch can
(and many do) define all sorts of weird & common names that end up
conflicting with system headers. So including it before the system
headers sets us up for pain. v850 is a good example of this -- when
building for mingw, we see weird failures:
$ i686-w64-mingw32-gcc ... -c -o dv-sockser.o ../../../../sim/v850/../common/dv-sockser.c
In file included from ../../../../sim/v850/sim-main.h:11,
from ../../../../sim/v850/../common/dv-sockser.c:24:
../../../../sim/v850/../common/sim-base.h:97:32: error: expected ')' before '->' token
97 | # define STATE_CPU(sd, n) ((sd)->cpu[0])
| ^~
While gcc is unhelpful at first, running it through the preprocessor
by hand shows more details:
$ i686-w64-mingw32-gcc ... -E -dD -o dv-sockser.i ../../../../sim/v850/../common/dv-sockser.c
$ i686-w64-mingw32-gcc -c dv-sockser.i
In file included from /usr/i686-w64-mingw32/usr/include/minwindef.h:163,
from /usr/i686-w64-mingw32/usr/include/windef.h:9,
from /usr/i686-w64-mingw32/usr/include/windows.h:69,
from /usr/i686-w64-mingw32/usr/include/winsock2.h:23,
from ../../gnulib/import/sys/socket.h:684,
from ../../gnulib/import/netinet/in.h:43,
from ../../../../sim/v850/../common/dv-sockser.c:39:
/usr/i686-w64-mingw32/usr/include/winnt.h:4803:25: error: expected ')' before '->' token
4803 | DWORD State;
| ^
| )
This is because v850 sets up this common name:
All of this needs cleaning up someday, but since the dv-sockser code
definitely should be fixed in this way, lets do that now and unblock
the v850 code.
It's unclear what this define is for as it appears to be unused, and
has never been used in the history of the mips sim. Delete it to tidy
up, and to fix build errors for Windows targets that have a standard
"PSIZE" struct in their system headers. This doesn't show up yet as
most sim files don't include many system headers, but enabling sockser
code for mingw uncovers the conflict.
Unfortunately the error produced by gcc is inscrutable, but running
it through the preprocessor manually manages to provide a pointer to
the underlying issue.
$ i686-w64-mingw32-gcc ... -c -o dv-sockser.o ../../../../sim/mips/../common/dv-sockser.c
<command-line>: error: expected identifier or '(' before numeric constant
In file included from /usr/i686-w64-mingw32/usr/include/windows.h:71,
from /usr/i686-w64-mingw32/usr/include/winsock2.h:23,
from ../../gnulib/import/sys/socket.h:684,
from ../../gnulib/import/netinet/in.h:43,
from ../../../../sim/mips/../common/dv-sockser.c:39:
/usr/i686-w64-mingw32/usr/include/wingdi.h:2934:59: error: unknown type name 'LPSIZE'; did you mean 'LPSIZEL'?
2934 | WINGDIAPI WINBOOL WINAPI GetAspectRatioFilterEx(HDC hdc,LPSIZE lpsize);
| ^~~~~~
| LPSIZEL
...
$ i686-w64-mingw32-gcc ... -E -dD -o dv-sockser.i ../../../../sim/mips/../common/dv-sockser.c
$ i686-w64-mingw32-gcc -c dv-sockser.i
In file included from /usr/i686-w64-mingw32/usr/include/windows.h:69,
from /usr/i686-w64-mingw32/usr/include/winsock2.h:23,
from ../../gnulib/import/sys/socket.h:684,
from ../../gnulib/import/netinet/in.h:43,
from ../../../../sim/mips/../common/dv-sockser.c:39:
/usr/i686-w64-mingw32/usr/include/windef.h:104:9: error: expected identifier or '(' before numeric constant
104 | } SIZE,*PSIZE,*LPSIZE;
| ^~
Now that the ppc code has been cleaned up enough to use the same set
of warning flags as the common code, delete the ppc-specific configure
logic so we can leverage what the common code already defined for us.
When compiling with --enable-werror and CFLAGS="-O0 -g -Wall", we run into:
...
src/sim/ppc/hw_memory.c: In function 'hw_memory_init_address':
src/sim/ppc/hw_memory.c:204:7: error: pointer targets in passing argument 4 \
of 'device_find_integer_array_property' differ in signedness \
[-Werror=pointer-sign]
&new_chunk->size);
^
...
Fix these by adding an explicit pointer cast. It's a bit ugly to use APIs
based on signed integers to read out unsigned values, but in practice, this
is par for the course in the ppc code.
We already use signed APIs and assign the result to unsigned values a lot:
see how device_find_integer_property returns a signed integer (cell), but
then assign it to unsigned types. The array APIs are not used that often
which is why we don't see many warnings, and we disable warnings when we
assign signed integers to unsigned integers in general.
The dtc/libfdt project (which is the standard in other projects) processes
the fdt blob as a series of bytes without any type information. Typing is
left to the caller. They have core APIs that read/write bytes, and a few
helper functions to cast/convert those bytes to the right value (e.g. u32).
In this ppc sim code, the core APIs use signed integers, and the callers
convert to unsigned, usually implicitly.
We could add some core APIs to the ppc sim that deal with raw bytes and then
add some helpers to convert to the right type, but that seems like a lot of
lifting for what boils down to a cast, and is effectively equivalent to all
the implicit assignments we use elsewhere. Long term, a lot of the ppc code
should either get converted to existing sim common code, or we should stand
up proper APIs in the common code first, or use standard libraries to do all
the processing (e.g. libfdt). Either way, this device.c code would all get
deleted, and callers (like these hw_*.c files) would get converted. Which
is also why we go with a cast rather new (but largely unused) APIs.
This aligns with common code which already uses this flag. We have
to add another local prototype to fix the failure, and add another
local decl for the SIM_DESC type. Unwinding these will require a
lot more work & conversions in the process, so going with the decl
for now unblocks the warning unification.
The basic "byte" type conflicts with Windows headers, and we already
have common types that provide the right sizes. So replace these with
the common ones to avoid issues.
CC dv-sockser.o
In file included from /usr/i686-w64-mingw32/usr/include/wtypes.h:8,
from /usr/i686-w64-mingw32/usr/include/winscard.h:10,
from /usr/i686-w64-mingw32/usr/include/windows.h:97,
from /usr/i686-w64-mingw32/usr/include/winsock2.h:23,
from ../../gnulib/import/sys/socket.h:684,
from ../../gnulib/import/netinet/in.h:43,
from .../build/sim/../../../sim/microblaze/../common/dv-sockser.c:39:
/usr/i686-w64-mingw32/usr/include/rpcndr.h:63:25: error: conflicting types for 'byte'; have 'unsigned char'
63 | typedef unsigned char byte;
| ^~~~
In file included from .../buildsim/../../../sim/microblaze/sim-main.h:21,
from .../buildsim/../../../sim/microblaze/../common/dv-sockser.c:24:
.../buildsim/../../../sim/microblaze/microblaze.h:94:25: note: previous declaration of 'byte' with type 'byte' {aka 'char'}
94 | typedef char byte;
| ^~~~
make: *** [Makefile:513: dv-sockser.o] Error 1
The disassembler has support to pretty print values created by an lui/addi
pair, but there is no support for addiw. There is also no support for
c.addi and c.addiw. This patch extends the pretty printing support to
handle these 3 instructions in addition to addi. Existing testcases serve
as tests for the new feature.
opcodes/
* riscv-dis.c (maybe_print_address): New arg wide. Sign extend when
wide is true.
(print_insn_args): Fix calls to maybe_print_address. Add checks for
c.addi, c.addiw, and addiw, and call maybe_print_address for them.
gas/
* testsuite/gas/riscv/insn.d: Update for disassembler change.
* testsuite/gas/li32.d, testsuite/gas/li64.d: Likwise.
* testsuite/gas/lla64.d: Likewise.
This copies logic used in the common sim warning configure code to fix
build errors for mingw targets. Turning format warnings on triggers
a failure in the debug.c file, so apply a minor fix at the same time.
This file is compiled for the --host & --build system which leads to
including the configure generated config.h in both environments.
This obviously doesn't work when the two targets don't look alike at
all and can cause build failures here (e.g. a mingw host & a linux
build). Since we don't actually need any config settings in this
very simple file, drop the includes entirely.
Some sim ports use these to provide networking functionality via the
dv-sockser module or via direct emulation for a few ports.
Gdb seems to build just fine still too.
Like Tom de Vries' earlier patch to fix the no-CXX_STD_THREAD case in
maint.c, this patch fixes a similar problem in
parallel-for-selftests.c. This fixes a build failure on Windows.
This hardens the powerpc64 linker code transformations.
* elf64-ppc.c (is_8byte_reloc, offset_in_range): New functions.
(ppc64_elf_relocate_section): Sanity check r_offset before
accessing section contents for various code transformations.
