Mike Frysinger [Wed, 7 Jul 2021 03:37:56 +0000 (23:37 -0400)]
sim: nltvals: pull target open flags out into a dedicated source file
Like we just did for pulling out the errno & signal maps, pull out the
open flag map into a dedicated common file. All newlib ports are using
the same map which makes it easy.
Mike Frysinger [Wed, 7 Jul 2021 03:21:20 +0000 (23:21 -0400)]
sim: nltvals: localize TARGET_<open> defines
Code should not be using these directly, instead they should be
resolving these dynamically via the open_map. Rework the common
callback code that was using the defines to use symbolic names
instead, and localize some of the defines in the ARM code (since
it's a bit unclear how many different APIs it supports currently),
then remove the defines out of the header so no new code can rely on
them.
Mike Frysinger [Wed, 7 Jul 2021 02:43:10 +0000 (22:43 -0400)]
sim: nltvals: pull target signal out into a dedicated source file
Like we just did for pulling out the errno map, pull out the signal
map into a dedicated common file. All newlib ports are using the
same signal map which makes it easy.
Mike Frysinger [Wed, 7 Jul 2021 02:10:53 +0000 (22:10 -0400)]
sim: nltvals: pull target errno out into a dedicated source file
The current system maintains a list of target errno constants in the
nltvals.def file, then runs a build-time tool to turn that into a C
file. This list of errno values is the same for all arches, so we
don't need the arch-specific flexibility. Further, these are only
for newlib/libgloss environments, which makes it confusing to support
other userland runtimes (like Linux). Let's simplify to make this
easier to understand & build. We don't namespace the variables yet,
but sets up the framework for it.
Create a new target-newlib-errno.c template file. The template file
is hand written, but the inline map is still automatically generated.
This allows us to move it to the common set of objects so it's only
built once in a multi-target build.
Now we can remove the output from the gentmap build-time tool since
it's checked into the tree.
Then we stop including the errno lists in nltvals.def since nothing
uses it.
Mike Frysinger [Sun, 31 Oct 2021 04:55:50 +0000 (00:55 -0400)]
sim: igen: tighten up build output
Add a new stamp helper for quiet builds, and don't dump the command
line options when it runs. That isn't standard tool behavior, and
doesn't really seem necessary in any way.
Orgad Shaneh [Thu, 28 Oct 2021 09:07:46 +0000 (09:07 +0000)]
sim: fix compilation on mingw64 [PR sim/28476]
...by reordering includes.
1. sim-utils.c
sim/mips/sim-main.h defines UserMode, while there is a struct in winnt.h
which has UserMode as a member. So if sim-main.h is included before winnt.h,
compilation fails.
2. ppc
registers.h defines CR, which is used as a member in winnt.h.
winsock2.h is included by sys/time.h, so sys/time.h has to be included
before registers.h.
Alan Modra [Sat, 30 Oct 2021 23:38:15 +0000 (10:08 +1030)]
Don't include coff/pe.h in coff-x86_64.c
This (and other) code from coffcode.h is broken for x86_64_coff_vec,
and has been ever since support was added in 2006 commit 99ad839030c1
Here, bfd_coff_aoutsz must match coff_swap_aouthdr_out otherwise we
end up writing garbage.
/* Note that peicode.h fills in a PEAOUTHDR, not an AOUTHDR.
include/coff/pe.h sets AOUTSZ == sizeof (PEAOUTHDR)). */
char * buff;
bfd_size_type amount = bfd_coff_aoutsz (abfd);
We have removed support for --target=x86_64-coff, likely because it
never worked properly, but still produce coff-x86_64.o with
--enable-targets=all. This means objcopy can recognize x86_64 COFF
files but will write garbage to the output file, a fact found by
fuzzers. I suspect x86_64 COFF is still broken after this fix, and
mention of coff-x86_64.* should be removed from bfd/Makefile.am.
* coff-x86_64.c: Don't include coff/pe.h.
(COFF_WITH_pex64): Don't define here.
* pe-x86_64.c: Include coff/pe.h and other headers.
(PEI_HEADERS): Define.
That's because when compiling with C++ >= 17, we use the standard
version of string_view, and don't have a selftest for it. So the list
of selftests shown by the tab completion when completing "string_v"
differs.
Aaron Merey [Fri, 29 Oct 2021 23:55:57 +0000 (19:55 -0400)]
gdb: add set/show commands for managing debuginfod
Add 'set debuginfod' command. Accepts 'on', 'off' or 'ask' as an
argument. 'on' enables debuginfod for the current session. 'off'
disables debuginfod for the current session. 'ask' will prompt
the user to either enable or disable debuginfod when the next query
is about to be performed:
This GDB supports auto-downloading debuginfo from the following URLs:
<URL1> <URL2> ...
Enable debuginfod for this session? (y or [n]) y
Debuginfod has been enabled.
To make this setting permanent, add 'set debuginfod on' to .gdbinit.
For interactive sessions, 'ask' is the default. For non-interactive
sessions, 'off' is the default.
Add 'show debuginfod status' command. Displays whether debuginfod
is set to 'on', 'off' or 'ask'.
Add 'set/show debuginfod urls' commands. Accepts a string of
space-separated debuginfod server URLs to be queried. The default
value is copied from the DEBUGINFOD_URLS environment variable.
Finally add 'set/show debuginfod verbose' commands to control whether
debuginfod-related output is displayed. Verbose output is enabled
by default.
(gdb) run
Starting program: /bin/sleep 5
Download failed: No route to host. Continuing without debug info for /lib64/libc.so.6.
If GDB is not built with debuginfod then these commands will just display
Tom Tromey [Fri, 22 Oct 2021 16:49:19 +0000 (10:49 -0600)]
Add gdb.Architecture.integer_type Python function
This adds a new Python function, gdb.Architecture.integer_type, which
can be used to look up an integer type of a given size and
signed-ness. This is useful to avoid dependency on debuginfo when a
particular integer type would be useful.
v2 moves this to be a method on gdb.Architecture and addresses other
review comments.
Tom Tromey [Wed, 20 Oct 2021 20:25:21 +0000 (14:25 -0600)]
Remove ada_value_print_inner
I noticed that the only caller of ada_value_print_inner is
valprint.c:do_val_print (via ada_language::value_print_inner), meaning
that the try/catch logic in this function is redundant. This patch
removes the wrapper function.
Tom Tromey [Tue, 19 Oct 2021 20:42:04 +0000 (14:42 -0600)]
Document resolve_dynamic_type oddity
Today I re-learned that resolve_dynamic_type can return a type for
which is_dynamic_type returns true. This can happen for an array
whose elements have dynamic type -- the array is reported as dynamic,
but resolving the elements would be incorrect, because each element
might have a different type after resolution.
You can see the special case in resolve_dynamic_array_or_string:
I looked into having the TYPE_CODE_ARRAY case in
is_dynamic_type_internal follow this same logic, but that breaks down
on the gdb.fortran/dynamic-ptype-whatis.exp test case. In particular
this code in fortran_undetermined::evaluate:
... relies on is_dynamic_type returning true for such an array.
I wasn't really sure of the best way to fix this, so in the meantime I
wrote this patch, which documents the oddity so that I might have a
chance of remembering this in the future.
Tom Tromey [Wed, 29 Sep 2021 16:14:52 +0000 (10:14 -0600)]
Avoid self-test failures on x86-linux
The disassembly tests in "maint selftest" will fail on x86-linux.
This happens because opcodes rejects an attempt to disassemble for an
arch with a 64-bit address size when bfd_vma is 32-bit.
This patch avoids this problem by avoiding the test in this case. I
chose to do it this way because this seems to be the only situation
where opcodes checks the size of bfd_vma.
For v2 of this patch, I've also updated memory_error_test to do the
same thing. This is needed due to the "improve error reporting from
the disassembler" patch.
Tom de Vries [Fri, 29 Oct 2021 12:11:08 +0000 (14:11 +0200)]
[gdb/build] Fix build with --disable-unit-tests
A build with --disable-unit-tests currently run into:
...
ld: maint.o: in function \
`maintenance_selftest_completer(cmd_list_element*, completion_tracker&,
char const*, char const*)':
src/gdb/maint.c:1183: undefined reference to \
`selftests::for_each_selftest(
gdb::function_view<
void (std::__cxx11::basic_string<char,std::char_traits<char>,
std::allocator<char> > const&)>)'
...
Fix this by guarding the call to selftests::for_each_selftest in
maintenance_selftest_completer with GDB_SELF_TEST, such that the "-verbose"
completion still works.
Rebuild on x86_64-linux and ran gdb.gdb/unittest.exp.
Tejas Belagod [Tue, 5 Oct 2021 08:43:12 +0000 (08:43 +0000)]
Support for a new pacbti unwind opcode.
This patch adds readelf support for decoding the exception table
opcode for restoring the RA_AUTH_CODE pseudo register defined by the
EHABI
(https://github.com/ARM-software/abi-aa/releases/download/2021Q1/ehabi32.pdf
Section 10.3).
* readelf.c (decode_arm_unwind_bytecode): Add support to decode
restoring RA_AUTH_CODE pseudo register.
Stafford Horne [Fri, 23 Oct 2020 04:39:53 +0000 (13:39 +0900)]
gdb: or1k: add single step for linux native debugging
Needed for single stepping in Linux, this adds the or1k implementation
of or1k_software_single_step. Most of the implementation is borrowed
from the bare metal single step code from or1k_single_step_through_delay
which has been extracted and shared in helper function
or1k_delay_slot_p.
Stafford Horne [Thu, 8 Oct 2020 21:26:30 +0000 (06:26 +0900)]
gdb: or1k: add native linux support
This patch adds support for running gdb natively on OpenRISC linux.
Debugging support is provided via the linux PTRACE interface which is
mostly handled by GDB genric code. This patch provides the logic of how
to read and write the ptrace registers between linux and GDB.
Markus Klein [Thu, 28 Oct 2021 16:17:25 +0000 (17:17 +0100)]
ARM assembler: Allow up to 32 single precision registers in the VPUSH and VPOP instructions.
PR 28436
* config/tc-arm.c (do_vfp_nsyn_push_pop_check): New function.
(do_vfp_nsyn_pop): Use the new function.
(do_vfp_nsyn_push): Use the new function.
* testsuite/gas/arm/v8_1m-mve.s: Add new instructions.
* testsuite/gas/arm/v8_1m-mve.d: Updated expected disassembly.
Simon Marchi [Thu, 28 Oct 2021 15:25:45 +0000 (11:25 -0400)]
gdb: use ptid_t::to_string in infrun debug messages
In debug messages, I think it would be more helpful to print ptid using
the simple "pid.lwp.tid" notation in infrun debug messages. I am
currently debugging some fork issues, and find the pid_to_str output not
so useful, as it doesn't tell which process a thread belongs to.
It currently shows up like this:
[infrun] resume_1: step=1, signal=GDB_SIGNAL_0, trap_expected=0, current thread [Thread 0x7ffff7d95740 (LWP 892942)] at 0x55555555521f
With the patch, it shows up like this:
[infrun] resume_1: step=1, signal=GDB_SIGNAL_0, trap_expected=1, current thread [894072.894077.0] at 0x5555555551d9
Simon Marchi [Thu, 23 Sep 2021 13:56:51 +0000 (09:56 -0400)]
gdb: add selftest name completion
After the previous commit, it is easy to add completion for selftest
names. Again, this is not particularly high value, but I rarely touched
completion, so it served as a simple example to get some practice.
Change the for_each_selftest_ftype parameter to gdb::function_view, so
that we can pass a lambda that captures things.
Tejas Belagod [Thu, 28 Oct 2021 14:44:42 +0000 (15:44 +0100)]
arm: add unwinder encoding support for PACBTI
This patch adds support for encoding the Return Address Authentication pseudo
register - '.save {ra_auth_code}' as defined by the DWARF ABI - in the
exception tables where the opcode is defined by the EHABI
gas/Changelog:
* config/tc-arm.c (arm_reg_type): Add new type REG_TYPE_PSEUDO.
(reg_expected_msgs): Add message for pseudo reg type.
(reg_list_els): Add new reg list type REGLIST_PSEUDO.
(parse_reg_list): Handle new REGLIST_PSEUDO type.
(s_arm_unwind_save_pseudo): Encode pseudo reg list save in exception
tables.
(s_arm_unwind_save): Handle new REG_TYPE_PSEUDO.
(reg_names): Add ra_auth_code pseudo register.
* testsuite/gas/arm/unwind-pacbti-m.s: New test.
* testsuite/gas/arm/unwind-pacbti-m.d: New test.
* testsuite/gas/arm/unwind-pacbti-m-readelf.d: New test.
Simon Marchi [Wed, 22 Sep 2021 14:38:16 +0000 (10:38 -0400)]
gdb: add "maint set/show selftest verbose" commands and use process_options
I saw the new -verbose switch to "maint selftests" and thought it would
be nice for it to use the option framework. For example, that makes
having completion easy. It's not that high value, given this is a
maintenance command, but I had never used the framework myself, so it
was a good way to practice.
This patch also adds the "maint set/show selftest verbose" setting. It
would be possible to use option framework without adding the setting,
but using the framework makes adding the option almost trivial, so I
thought why not.
Simon Marchi [Wed, 22 Sep 2021 17:43:25 +0000 (13:43 -0400)]
gdb: add add_setshow_prefix_cmd
There's a common pattern to call add_basic_prefix_cmd and
add_show_prefix_cmd to add matching set and show commands. Add the
add_setshow_prefix_cmd function to factor that out and use it at a few
places.
Tom de Vries [Thu, 28 Oct 2021 12:18:15 +0000 (14:18 +0200)]
[gdb/testsuite] Fix assembly comments in gdb.dwarf2/clang-debug-names.exp.tcl
On openSUSE Leap 15.2 aarch64 I ran into:
...
clang-debug-names-debug.S:72: \
Error: junk at end of line, first unrecognized character is `#'
...
due to:
...
71 .Ldebug_names_start:
72 .short 5 # Header: version
...
Fix this by using the /* ... */ comment style instead:
...
$ sed -i 's% #\([^"]*\)%/*\1 */%' clang-debug-names.exp.tcl
...
Tom de Vries [Thu, 28 Oct 2021 08:43:34 +0000 (10:43 +0200)]
[gdb/symtab] Handle DW_AT_string_length with location list
Consider a fortran routine where a string variable s is modified:
...
subroutine f(s)
character*(*) s
print *, s
s(1:3) = 'oof'
print *, s
end subroutine f
...
When compiling with optimization level -O1 and printing the type of
variable s we get:
...
$ gdb -q -batch outputs/gdb.opt/fortran-string/fortran-string \
-ex "b f" \
-ex run \
-ex "ptype s"
Breakpoint 1 at 0x4006f7: file fortran-string.f90, line 21.
Breakpoint 1, f (s=..., _s=_s@entry=3) at fortran-string.f90:21
21 subroutine f(s)
type = character*1
...
while with -O0 we have instead:
...
type = character (3)
...
The problem is that the type of s is:
...
<1><2d6>: Abbrev Number: 21 (DW_TAG_string_type)
<2d7> DW_AT_string_length: 0xbf (location list)
<2db> DW_AT_byte_size : 4
...
where the DW_AT_string_length is a location list, a case that is not handled
by attr_to_dynamic_prop.
Fix this by handling attr->form_is_section_offset () in attr_to_dynamic_prop.
Tested on x86_64-linux.
The test-case is based on gdb.opt/fortran-string.exp from
https://src.fedoraproject.org/rpms/gdb/raw/f32/f/gdb-archer-vla-tests.patch .
I've updated the copyrights to stretch to 2021.
[ I've tried to create a dwarf assembly test-case for this, but didn't
manage. ]
Co-Authored-By: Jan Kratochvil <jan.kratochvil@redhat.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=26910
[gdb/testsuite] Initialize anonymous union in gdb.cp/koenig.cc
GDB test fails while running the test case gdb.cp/koenig.exp using
clang compiler:
[...]
p foo (p_union)
No symbol "p_union" in current context.
(gdb) FAIL: gdb.cp/koenig.exp: p foo (p_union)
[...]
In the testcase, "p_union" is an unused/uninitialized variable of
anonymous union type. Clang does not emit symbol for unused anonymous
union/struct variables at any optimization level. Since the compiler
itself is not emitting the symbol for "p_union", debug info is also
not emitted when built with debug option. If the anonymous union is
initialized (or used), then clang emits the symbol "p_union" which
enables emitting debug info for "p_union".
[...]
p foo (p_union)
Cannot resolve function foo to any overloaded instance
(gdb) PASS: gdb.cp/koenig.exp: p foo (p_union)
[...]
Alan Modra [Thu, 28 Oct 2021 02:46:04 +0000 (13:16 +1030)]
asan: mmo: NULL dereferenc in mmo_xore_32
mmo_get_loc can return NULL. It's commented even, and that the caller
then must handle a split field. mmo_xore_* don't handle split fields,
instead just segfault. Stop that happening, and refuse to recognise
fuzzed mmo files that trigger this problem.
Tom Tromey [Mon, 4 Oct 2021 18:38:23 +0000 (14:38 -0400)]
Fix watchpoints with multiple threads on Windows
A recent internal change pointed out that watchpoints were not working
on Windows when the inferior was multi-threaded. This happened
because the debug registers were only updated for certain threads --
in particular, those that were being resumed and that were not marked
as suspended. In the case of single-stepping, the need to update the
debug registers in other threads could also be "forgotten".
This patch changes windows-nat.c to mark all threads needing a debug
register update. This brings the code closer to what gdbserver does
(though, unfortunately, it still seems more complicated than needed).
Tom de Vries [Wed, 27 Oct 2021 16:57:15 +0000 (18:57 +0200)]
[gdb/testsuite] Fix port detection in gdb.debuginfod/fetch_src_and_symbols.exp
On OBS I ran into this failure with test-case
gdb.debuginfod/fetch_src_and_symbols.exp:
...
Failed to listen for connections: Address already in use^M
[Thu Oct 21 11:48:49 2021] (559/559): started http server on IPv6 port=8000^M
...
FAIL: gdb.debuginfod/fetch_src_and_symbols.exp: local_url: find port timeout
...
The test-case is trying to start debuginfod on a port to see if it's
available, and it handles either this message:
"started http server on IPv4 IPv6 port=$port"
meaning success, or:
"failed to bind to port"
meaning failure, in which case the debuginfod instance is killed, and we try
the next port.
The test-case only uses the v4 address 127.0.0.1, so fix this by:
- accepting "started http server on IPv4 port=$port"
- rejecting "started http server on IPv6 port=$port"
Nelson Chu [Wed, 27 Oct 2021 10:54:41 +0000 (18:54 +0800)]
RISC-V: Tidy riscv assembler and disassembler.
Tidy the gas/config/tc-riscv.c and opcodes/riscv-dis.c, to prepare for
moving the released extensions (including released vendor extensions)
from integration branch back to mainline.
* Added parts of missing comments.
* Updated md_show_usage.
* For validate_riscv_insn, riscv_ip and print_insn_args, unify the
following pointer names,
- oparg: pointed to the parsed operand defined in the riscv_opcodes.
- asarg: pointed to the parsed operand from assembly.
- opargStart: recorded the parsed operand name from riscv_opcodes.
- asargStart: recorded the parsed operand name from assembly.
gas/
* config/tc-riscv.c: Added parts of missind comments and updated
the md_show_usage.
(riscv_multi_subset_supports): Tidy codes.
(validate_riscv_insn): Unify the pointer names, oparg, asarg,
opargStart and asargStart, to prepare for moving the released
extensions from integration branch back to mainline.
(riscv_ip): Likewise.
(macro_build): Added fmtStart, also used to prepare for moving
released extensions.
(md_show_usage): Added missing descriptions for new options.
opcodes/
* riscv-dis.c (print_insn_args): Unify the pointer names,
oparg and opargStart, to prepare for moving the released
extensions from integration branch back to mainline.
opcodes: Fix RPATH not being set for dynamic libbfd dependency
If built as a shared library, libopcodes has a load-time dependency on
libbfd, which is recorded in the dynamic section, however without a
corresponding RPATH entry for the directory to find libbfd in. This
causes loading to fail whenever libbfd is only pulled by libopcodes
indirectly and libbfd has been installed in a directory that is not in
the dynamic loader's search path.
It does not happen with the programs included with binutils or GDB,
because they all also pull libbfd when using libopcodes, but it can
happen with external software, e.g.:
$ gdbserver --help
gdbserver: error while loading shared libraries: libbfd-[...].so: cannot open shared object file: No such file or directory
$
(not our `gdbserver').
Indirect dynamic dependencies are handled by libtool automatically by
adding RPATH entries as required, however our setup for libopcodes
prevents this from happening by linking in libbfd with an explicit file
reference sneaked through to the linker directly behind libtool's back
via the `-Wl' linker command-line option rather than via `-l' combined
with a suitable library search path specified via `-L', as it would be
usually the case, or just referring to the relevant .la file in a fully
libtool-enabled configuration such as ours.
According to an observation in the discussion back in 2007[1][2][3] that
has led to the current arrangement it is to prevent libtool from picking
up the wrong version of libbfd. It does not appear to be needed though,
not at least with our current libtool incarnation, as directly referring
`libbfd.la' does exactly what it should, as previously suggested[4], and
with no link-time reference to the installation directory other than to
set RPATH. Uninstalled version of libopcodes has libbfd's build-time
location prepended to RPATH too, as also expected.
Use a direct reference to `libbfd.la' then, making the load error quoted
above go away. Alternatively `-L' and `-l' could be used to the same
effect, but it seems an unnecessary complication and just another way to
circumvent rather than making use of libtool.
References:
[1] "compile failure due to undefined symbol",
<https://sourceware.org/ml/binutils/2007-08/msg00476.html>
H.J. Lu [Mon, 25 Oct 2021 15:42:24 +0000 (08:42 -0700)]
gold: Place .note.gnu.property section before other note sections
Place the .note.gnu.property section before all other note sections to
avoid being placed between other note sections with different alignments.
PR gold/28494
* layout.cc (Layout::create_note): Set order to ORDER_PROPERTY_NOTE
for the .note.gnu.property section.
* layout.h (Output_section_order): Add ORDER_PROPERTY_NOTE.
Tom de Vries [Tue, 26 Oct 2021 08:45:08 +0000 (10:45 +0200)]
[gdb/doc] Fix print inferior-events default
In the docs about print inferior-events we read:
...
By default, these messages will not be printed.
...
That used to be the case, but is no longer so since commit f67c0c91715 "Enable
'set print inferior-events' and improve detach/fork/kill/exit messages".
Simon Marchi [Tue, 5 Oct 2021 00:47:06 +0000 (20:47 -0400)]
gdb: change functions returning value contents to use gdb::array_view
The bug fixed by this [1] patch was caused by an out-of-bounds access to
a value's content. The code gets the value's content (just a pointer)
and then indexes it with a non-sensical index.
This made me think of changing functions that return value contents to
return array_views instead of a plain pointer. This has the advantage
that when GDB is built with _GLIBCXX_DEBUG, accesses to the array_view
are checked, making bugs more apparent / easier to find.
This patch changes the return types of these functions, and updates
callers to call .data() on the result, meaning it's not changing
anything in practice. Additional work will be needed (which can be done
little by little) to make callers propagate the use of array_view and
reap the benefits.
Simon Marchi [Tue, 19 Oct 2021 19:32:08 +0000 (15:32 -0400)]
gdbsupport: add assertions in array_view
Add assertions to ensure we don't access an array_view out of bounds.
Enable these assertions only when _GLIBCXX_DEBUG is set, as we did for
gdb::optional.
This doesn't work, because target_pid_to_str stores its result in a
static buffer, so my message would show twice the same ptid. Change
target_pid_to_str to return an std::string to avoid this. I don't think
we save much by using a static buffer, but it is more error-prone.
Tom de Vries [Mon, 25 Oct 2021 16:15:15 +0000 (18:15 +0200)]
[gdb/testsuite] Fix duplicate in gdb.mi/mi-var-cp.exp
With test-case gdb.mi/mi-var-cp.exp I run into this duplicate:
...
PASS: gdb.mi/mi-var-cp.exp: run to mi-var-cp.cc:104 (set breakpoint)
PASS: gdb.mi/mi-var-cp.exp: create varobj for s
PASS: gdb.mi/mi-var-cp.exp: create varobj for s
DUPLICATE: gdb.mi/mi-var-cp.exp: create varobj for s
...
This is due to a duplicate test name here:
...
$ cat -n gdb/testsuite/gdb.mi/mi-var-cp.cc
...
100 int reference_to_struct ()
101 {
102 /*: BEGIN: reference_to_struct :*/
103 S s = {7, 8};
104 S& r = s;
105 /*:
106 mi_create_varobj S s "create varobj for s"
107 mi_create_varobj R r "create varobj for s"
...
Nick Alcock [Mon, 25 Oct 2021 10:17:02 +0000 (11:17 +0100)]
libctf, ld: handle nonrepresentable types better
ctf_type_visit (used, among other things, by the type dumping code) was
aborting when it saw a nonrepresentable type anywhere: even a single
structure member with a nonrepresentable type caused an abort with
ECTF_NONREPRESENTABLE. This is not useful behaviour, given that the
abort comes from a type-resolution we are only doing in order to
determine whether the type is a structure or union. We know
nonrepresentable types can't be either, so handle that case and
pass the nonrepresentable type down.
(The added test verifies that the dumper now handles this case and
prints nonrepresentable structure members as it already does
nonrepresentable top-level types, rather than skipping the whole
structure -- or, without the previous commit, skipping the whole types
section.)
ld/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
* testsuite/ld-ctf/nonrepresentable-member.*: New test.
libctf/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
Nick Alcock [Mon, 25 Oct 2021 10:17:02 +0000 (11:17 +0100)]
libctf: dump: do not stop dumping types on error
If dumping of a single type fails, we obviously can't dump it; but just
as obviously this doesn't make the other types in the types section
invalid or undumpable. So we should not propagate errors seen when
type-dumping, but rather ignore them and carry on, so we dump as many
types as we can (leaving out the ones we can't grok).
libctf/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
* ctf-dump.c (ctf_dump_type): Do not abort on error.
Nick Alcock [Mon, 25 Oct 2021 10:17:02 +0000 (11:17 +0100)]
binutils, ld: make objdump --ctf's parameter optional
ld by default (and always, unless adjusted with a hand-rolled linker
script) emits deduplicated CTF into the .ctf section. But viewing
it needs you to explicitly tell objdump this: it doesn't default
its argument, even though what you always end up typing is
--ctf=.ctf.
This is annoying, so make the argument optional.
binutils/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
* objdump.c (usage): --ctf now has an optional argument.
(main): Adjust accordingly.
(dump_ctf): Default it.
* doc/ctf.options.texi: Adjust.
ld/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
Nick Alcock [Mon, 25 Oct 2021 10:17:02 +0000 (11:17 +0100)]
binutils: make objdump/readelf --ctf-parent actually useful
This option has been present since the very early days of the
development of libctf as part of binutils, and it shows. Back in the
earliest days, I thought we might handle ambiguous types by introducing
new ELF sections on the fly named things like .ctf.foo.c for ambiguous
types found only in foo.c, etc. This turned out to be a terrible idea,
so we moved to using a CTF archive in the .ctf section which contained
all the CTF dictionaries -- but the --ctf-parent option in objdump and
readelf was never adjusted, and lingered as a mechanism to specify CTF
parent dictionaries in sections other than .ctf, even though the linker
has no way to produce parent dictionaries in different sections from
their children, libctf's ctf_open can't handle such split-up
parent/child dicts, and they are never found in the wild, emitted by GNU
ld or by any known third-party linking tool.
Meanwhile, the actually-useful ctf_link feature (albeit not used by ld)
which lets you remap the names of CTF archive members (so you can end up
with a parent archive member named something other than ".ctf", still
contained with all its children in a single .ctf section) had no support
in objdump or readelf: there was no way to tell them that these members
were parents, so all the types in the associated child dicts always
appeared corrupted, referencing nonexistent types from a parent objdump
couldn't find.
So adjust --ctf-parent so that rather than taking a section name it
takes a member name instead (if not specified, the name is ".ctf", which
is what GNU ld emits). Because the option was always useless before
now, this is expected to have no backward-compatibility implications.
As part of this, we have to slightly adjust the code which skips the
archive member name if redundant: right now it skips it if it's ".ctf",
on the assumption that this name will almost always be at the start
of the objdump output and thus we'll end up with a shared dump
and then smaller, headed dumps for the per-TU child dicts; but if
the parent name has been changed, that won't be true any more.
So change the rules to "members named .ctf which appear first in the
first have their member name skipped". Since we now need to count
members, move from ctf_archive_iter (for which passing in extra
parameters requires defining a new struct and is clumsy) to
ctf_archive_next, allowing us to just *call* dump_ctf_archive_member and
maintain a member count in the obvious way. In the process we fix a
tiny difference between readelf and objdump: if a ctf_dump ever failed,
readelf skipped every later member, while objdump tried to keep going as
much as it could. For a dumping tool the former is clearly preferable.
binutils/ChangeLog
2021-10-25 Nick Alcock <nick.alcock@oracle.com>
* objdump.c (usage): --ctf-parent now takes a name, not a section.
(dump_ctf): Don't open a separate section; use the parent_name in
ctf_dict_open instead. Use ctf_archive_next, not ctf_archive_iter,
so we can pass down a member count.
(dump_ctf_archive_member): Add the member count; don't return
anything. Import parents into children no matter what the
parent's name, while still avoiding displaying the header for the
common parent name of ".ctf".
* readelf.c (usage): Adjust similarly.
(dump_section_as_ctf): Likewise.
(dump_ctf_archive_member): Likewise. Never stop iterating over
archive members, even if ctf_dump of one member fails.
* doc/ctf.options.texi: Adjust.
projects / thirdparty / binutils-gdb.git / log
