Jan Beulich [Tue, 21 Dec 2021 08:31:04 +0000 (09:31 +0100)]
x86: -mfence-as-lock-add=yes doesn't work for 16-bit mode
Rather than trying to fix this (which would require making an assumption
on the upper half of %esp being zero), simply issue an error. While at
it, since the generated code is in conflict with -momit-lock-prefix=yes,
issue an error in that case as well.
Tom Tromey [Sat, 11 Dec 2021 21:57:17 +0000 (14:57 -0700)]
Remove print_spaces
This removes the print_spaces helper function, in favor of using the
"*%s" idiom that's already used in many places in gdb. One spot (in
symmisc.c) is changed to use print_spaces_filtered, because the rest
of that function is using filtered output. (This highlights one way
that the printf idiom is better -- this error is harder to make when
using that.)
Tom Tromey [Sat, 11 Dec 2021 22:10:14 +0000 (15:10 -0700)]
Make n_spaces return a const char *
n_spaces keeps the spaces in a static buffer. If a caller overwrites
these, it may give an incorrect result to a subsequent caller. So,
make the return type const to help avoid this outcome.
Andrew Burgess [Mon, 22 Nov 2021 12:30:36 +0000 (12:30 +0000)]
gdb: add assert in remote_target::wait relating to async being off
While working on another patch I ended up in a situation where I had
async mode disabled (with 'maint set target-async off'), but the async
event token got marked anyway.
In this situation GDB was continually calling into
remote_target::wait, however, the async token would never become
unmarked as the unmarking is guarded by target_is_async_p.
We could just unconditionally unmark the token, but that would feel
like just ignoring a bug, so, instead, lets assert that if
!target_is_async_p, then the async token should not be marked.
This assertion would have caught my earlier mistake.
There should be no user visible changes with this commit.
Andrew Burgess [Wed, 17 Nov 2021 17:17:37 +0000 (17:17 +0000)]
gdb/remote: some fixes for 'maint set target-async off'
While working on another patch relating to remote targets, I wanted to
test with 'maint set target-async off' in place. Unfortunately I ran
into some problems. This commit is an attempt to fix one of the
issues I hit.
In my particular case I was actually running with:
maint set target-async off
maint set target-non-stop off
that is, we're telling GDB to force the targets to operate in
non-async mode, and in all-stop mode. Here's my GDB session showing
the problem:
(gdb) maintenance set target-async off
(gdb) maintenance set target-non-stop off
(gdb) target extended-remote :54321
Remote debugging using :54321
(gdb) attach 2365960
Attaching to process 2365960
No unwaited-for children left.
(gdb)
Notice the 'No unwaited-for children left.' error, this is the
problem. There's no reason why GDB should not be able to attach to
the process.
The problem is this:
1. The user runs 'attach PID' and this sends GDB into attach_command
in infcmd.c. From here we call the ::attach method on the attach
target, which will be the extended_remote_target.
2. In extended_remote_target::attach, we attach to the remote target
and get the first reply (which is a stop packet). We put off
processing the stop packet until the end of ::attach. We setup the
inferior and thread to represent the process we attached to, and
download the target description. Finally, we process the initial
stop packet.
If '!target_is_non_stop_p ()' and '!target_can_async_p ()', which is
the case for us given the maintenance commands we used, we cache the
stop packet within the remote_state::buf for later processing.
3. Back in attach_command, if 'target_is_non_stop_p ()' then we
request that the target stops. This will either process any cached
stop replies, or request that the target stops, and process the stop
replies. However, this code is not what we use due to non-stop mode
being disabled. So, we skip to the next step which is to call
validate_exec_file.
4. Calling validate_exec_file can cause packets to be sent to the
remote target, and replies received, the first path I hit is the
call to target_pid_to_exec_file, which calls
remote_target::pid_to_exec_file, which can then try to read the
executable from the remote. Sending an receiving packets will make
use of the remote_state::buf object.
5. The attempt to attach continues, but the damage is already done...
So, the problem is that, in step #2 we cache a stop reply in the
remote_state::buf, and then in step #4 we reuse the remote_state::buf
object, discarding any cached stop reply. As a result, the initial
stop, which is sent when GDB first attaches to the target, is lost.
This problem can clearly be seen, I feel, by looking at the
remote_state::cached_wait_status flag. This flag tells GDB if there
is a wait status cached in remote_state::buf. However, in
remote_target::putpkt_binary and remote_target::getpkt_or_notif_sane_1
this flag is just set back to 0, doing this immediately discards any
cached data.
I don't know if this scheme ever made sense, looking at commit 2d717e4f8a54, where the cached_wait_status flag was added, it appears
that there was nothing between where the stop was cached, and where
the stop was consumed, so, I suspect, there never was a situation
where we ended up in putpkt_binary or getpkt_or_notif_sane_1 and
needed to clear to the flag, maybe the clearing was added "just in
case". Whatever the history, I claim that this clearing this flag is
no longer a good idea.
So, my first step toward fixing this issue was to replace the two
instances of 'rs->cached_wait_status = 0;' in ::putpkt_binary and
::getpkt_or_notif_sane_1 with 'gdb_assert (rs->cached_wait_status ==
0);', this, at least would show me when GDB was doing something
dangerous, and indeed, this assert is now hit in my test case above.
I did play with using some kind of scoped restore to backup, and
restore the remote_state::buf object in all the places within remote.c
that I was hitting where the ::buf was being corrupted. The first
problem with this is that, where the ::cached_wait_status flag is
reset is _not_ where ::buf is corrupted. For the ::putpkt_binary
case, by the time we get to the method the buffer has already been
corrupted in many cases, so we end up needing to add the scoped
save/restore within the callers, which means we need the save/restore
in _lots_ of places.
Plus, using this save/restore model feels like the wrong solution. I
don't think that it's obvious that the buffer might be holding cached
data, and I think it would be too easy for new corruptions of the
buffer to be introduced, which could easily go unnoticed for a long
time.
So, I really wanted a solution that didn't require us to cache data in
the ::buf object.
Luckily, I think we already have such a solution in place, the
remote_state::stop_reply_queue, it seems like this does exactly the
same task, just in a slightly different way. With the
::stop_reply_queue, the stop packets are processed upon receipt and
the stop_reply object is added to the queue. With the ::buf cache
solution, the unprocessed stop reply is cached in the ::buf, and
processed later.
So, finally, in this commit, I propose to remove the
remote_state::cached_wait_status flag and to stop using the ::buf to
cache stop replies. Instead, stop replies will now always be stored
in the ::stop_reply_queue.
There are two places where we use the ::buf to hold a cached stop
reply, the first is in the ::attach method, and the second is in
remote_target::start_remote, however, the second of these cases is far
less problematic, as after caching the stop reply in ::buf we call the
global start_remote function, which does very little work before
calling normal_stop, which processes the cached stop reply. However,
my plan is to switch both users over to using ::stop_reply_queue so
that the old (unsafe) ::cached_wait_status mechanism can be completely
removed.
The next problem is that the ::stop_reply_queue is currently only used
for async-mode, and so, in remote_target::push_stop_reply, where we
push stop_reply objects into the ::stop_reply_queue, we currently also
mark the async event token. I've modified this so we only mark the
async event token if 'target_is_async_p ()' - note, _is_, not _can_
here. The ::push_stop_reply method is called in places where async
mode has been temporarily disabled, but, when async mode is switched
back on (see remote_target::async) we will mark the event token if
there are events in the queue.
Another change of interest is in remote_target::remote_interrupt_as.
Previously this code checked ::cached_wait_status, but didn't check
for events in the ::stop_reply_queue. Now that ::cached_wait_status
has been removed we now check the queue length instead, which should
have the same result.
Finally, in remote_target::wait_as, I've tried to merge the processing
of the ::stop_reply_queue with how we used to handle the
::cached_wait_status flag.
Currently, when processing the ::stop_reply_queue we call
process_stop_reply and immediately return. However, when handling
::cached_wait_status we run through the whole of ::wait_as, and return
at the end of the function.
If we consider a standard stop packet, the two differences I see are:
1. Resetting of the remote_state::waiting_for_stop_reply, flag; this
is not currently done when processing a stop from the
::stop_reply_queue.
2. The final return value has the possibility of being adjusted at
the end of ::wait_as, as well as there being calls to
record_currthread, non of which are done if we process a stop from
the ::stop_reply_queue.
it was suggested that, when an event is pushed into the
::stop_reply_queue, the ::waiting_for_stop_reply flag is never going
to be set. As a result, we don't need to worry about the first
difference. I have however, added a gdb_assert to validate the
assumption that the flag is never going to be set. If in future the
situation ever changes, then we should find out pretty quickly.
As for the second difference, I have resolved this by having all stop
packets taken from the ::stop_reply_queue, pass through the return
value adjustment code at the end of ::wait_as.
An example of a test that reveals the benefits of this commit is:
make check-gdb \
RUNTESTFLAGS="--target_board=native-extended-gdbserver \
GDBFLAGS='-ex maint\ set\ target-async\ off \
-ex maint\ set\ target-non-stop\ off' \
gdb.base/attach.exp"
For testing I've been running test on x86-64/GNU Linux, and run with
target boards unix, native-gdbserver, and native-extended-gdbserver.
For each board I've run with the default GDBFLAGS, as well as with:
Though running with the above GDBFLAGS is clearly a lot more unstable
both before and after my patch, I'm not seeing any consistent new
failures with my patch, except, with the native-extended-gdbserver
board, where I am seeing new failures, but only because more tests are
now running. For that configuration alone I see the number of
unresolved go down by 49, the number of passes goes up by 446, and the
number of failures also increases by 144. All of the failures are new
tests as far as I can tell.
Simon Marchi [Wed, 15 Dec 2021 20:17:44 +0000 (13:17 -0700)]
Add new gdbarch generator
The new gdbarch generator is a Python program. It reads the
"components.py" that was created in the previous patch, and generates
gdbarch.c and gdbarch-gen.h.
This is a relatively straightforward translation of the existing .sh
code. It doesn't try very hard to be idiomatic Python or to be
especially smart.
Tom Tromey [Wed, 15 Dec 2021 00:19:03 +0000 (17:19 -0700)]
Generate new gdbarch-components.py from gdbarch.sh
The new gdbarch.sh approach will be to edit a Python file, rather than
adding a line to a certain part of gdbarch.sh. We use the existing sh
code, though, to generate the first draft of this .py file.
Documentation on the format will come in a subsequent patch.
Note that some info (like "staticdefault") in the current code is
actually unused, and so is ignored by this new generator.
Tom Tromey [Wed, 15 Dec 2021 20:42:55 +0000 (13:42 -0700)]
Do not sort the fields in gdbarch_dump
This changes gdbarch.sh so that it no longer sorts the fields in
gdbarch_dump. This sorting isn't done anywhere else by gdbarch.sh,
and this simplifies the new generator a little bit.
Tom Tromey [Tue, 14 Dec 2021 23:48:45 +0000 (16:48 -0700)]
Split gdbarch.h into two files
This patch splits gdbarch.h into two files -- gdbarch.h now is
editable and hand-maintained, and the new gdbarch-gen.h file is the
only thing generated by gdbarch.sh. This lets us avoid maintaining
boilerplate in the gdbarch.sh file.
Note that gdbarch.sh still generates gdbarch.h after this patch. This
makes it easier to re-run when rebasing. This code is removed in a
subsequent patch.
Tom Tromey [Tue, 14 Dec 2021 23:33:56 +0000 (16:33 -0700)]
Move ordinary gdbarch code to arch-utils
While I think it makes sense to generate gdbarch.c, at the same time I
think it is better for ordinary code to be editable in a C file -- not
as a hunk of C code embedded in the generator.
This patch moves this sort of code out of gdbarch.sh and gdbarch.c and
into arch-utils.c, then has arch-utils.c include gdbarch.c.
Avoid redundant operations in `fortran_array_walker'
Move inner dimension's element type determination outside the respective
loops in `fortran_array_walker'. The operation is exactly the same with
each iteration, so there is no point in redoing it for each element and
while a smart compiler might be able to move it outside the loop it is
regardless a bad coding style. No functional change.
Initialize `m_ndimensions' in the member initializer list
Following our coding convention initialize the `m_ndimensions' member in
the member initializer list rather than in the body of the constructor
of the `fortran_array_walker' class. No functional change.
The problem is that at this point rl_instream has the value NULL.
The rl_instream variable is supposed to be initialized during a call to
readline_initialize_everything, which in a normal startup sequence is
called under this call chain:
This function unconditionally installs the SIGWINCH signal handler (this
is done by tui_initialize_win), and then if gdb_stdout is a TTY it
initializes readline. Therefore, if stdout is not a TTY, SIGWINCH is
installed but readline is not initialized. In such situation
rl_instream stays NULL, and when GDB receives a SIGWINCH it calls its
handler and in fine tries to access rl_instream leading to the crash.
This patch proposes to fix this issue by installing the SIGWINCH signal
handler only if GDB is connected to a TTY. Given that this
initialization it the only task of tui_initialize_win, this patch moves
tui_initialize_win just after the call to
tui_ensure_readline_initialized.
Tested on x86_64-linux.
Co-authored-by: Pedro Alves <pedro@palves.net>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=26056
Change-Id: I6458acef7b0d9beda2a10715d0345f02361076d9
Alan Modra [Fri, 17 Dec 2021 05:11:59 +0000 (15:41 +1030)]
asan: NULL dereference in bfd_elf_set_group_contents
* elf-bfd.h (struct output_elf_obj_tdata): Make num_section_syms
unsigned.
* elf.c (bfd_elf_set_group_contents): Bounds check sec->index
and check that entry in elf_section_syms for sec is non-NULL.
(_bfd_elf_symbol_from_bfd_symbol): Adjust.
Alan Modra [Fri, 17 Dec 2021 03:02:36 +0000 (13:32 +1030)]
asan: heap-buffer-overflow in bpf_elf_generic_reloc
The bpf reloc howtos are a bit weird, using bitpos to specify an
offset from r_offset that is outside the size of the reloc as given by
howto.size. That means bfd_get_reloc_size gives the wrong answer for
range checking, and thus bfd_reloc_offset_in_range can't be used.
* elf64-bpf.c (bpf_elf_generic_reloc): Handle bitpos offset reloc
range checking.
This patch adds AArch32 support for -march=armv9.[123]-a.
The behaviour of the new options can be expressed using a
combination of existing feature flags and tables.
The cpu_arch_ver entries for ARM_ARCH_V9_2A and ARM_ARCH_V9_3A
are technically redundant but it seemed less surprising to include
them anyway.
include/
* opcode/arm.h (ARM_ARCH_V9_1A, ARM_ARCH_V9_2A): New macros.
(ARM_ARCH_V9_3A): Likewise.
This patch adds AArch32 support for -march=armv8.[78]-a.
The behaviour of the new options can be expressed using a
combination of existing feature flags and tables.
The cpu_arch_ver entries are technically redundant but
it seemed less surprising to include them anyway.
include/
* opcode/arm.h (ARM_ARCH_V8_7A, ARM_ARCH_V8_8A): New macros.
gas/
* doc/c-arm.texi: Add armv8.7-a and armv8.8-a.
* config/tc-arm.c (armv87a_ext_table, armv88a_ext_table): New macros.
(arm_archs): Add armv8.7-a and armv8.8-a.
(cpu_arch_ver): Add ARM_ARCH_V8_7A and ARM_ARCH_V8_8A.
* NEWS: Mention the above.
* testsuite/gas/arm/attr-march-armv8_7-a.d: New test.
* testsuite/gas/arm/attr-march-armv8_8-a.d: Likewise.
* testsuite/gas/arm/bfloat16-armv8.7-a.d: Likewise.
* testsuite/gas/arm/bfloat16-armv8.8-a.d: Likewise.
* testsuite/gas/arm/i8mm-armv8.7-a.d: Likewise.
* testsuite/gas/arm/i8mm-armv8.8-a.d: Likewise.
This patch adds AArch64 support for -march=armv9.[123]-a.
The behaviour of the new options can be expressed using a
combination of existing feature flags, so we don't need to
eat into the vanishing number of spare AARCH64_FEATURE_* bits.
Hoewver, it was more convenient to separate out the |s of
feature flags so that Armv9.1-A could reuse the set for
Armv8.6-A, and so on.
include/
* opcode/aarch64.h (AARCH64_ARCH_V8_FEATURES): New macro,
split out from...
(AARCH64_ARCH_V8): ...here.
(AARCH64_ARCH_V8_1_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_1): ...here.
(AARCH64_ARCH_V8_2_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_2): ...here.
(AARCH64_ARCH_V8_3_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_3): ...here.
(AARCH64_ARCH_V8_4_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_4): ...here.
(AARCH64_ARCH_V8_5_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_5): ...here.
(AARCH64_ARCH_V8_6_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_6): ...here.
(AARCH64_ARCH_V8_7_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_7): ...here.
(AARCH64_ARCH_V8_8_FEATURES): New macro, split out from...
(AARCH64_ARCH_V8_8): ...here.
(AARCH64_ARCH_V9_FEATURES): New macro, split out from...
(AARCH64_ARCH_V9): ...here.
(AARCH64_ARCH_V9_1_FEATURES, AARCH64_ARCH_V9_1): New macros.
(AARCH64_ARCH_V9_2_FEATURES, AARCH64_ARCH_V9_2): New macros.
(AARCH64_ARCH_V9_3_FEATURES, AARCH64_ARCH_V9_3): New macros.
Nelson Chu [Thu, 22 Jul 2021 05:47:07 +0000 (13:47 +0800)]
RISC-V: Support svinval extension with frozen version 1.0.
According to the privileged spec, there are five new instructions for
svinval extension. Two of them (HINVAL.VVMA and HINVAL.GVMA) need to
enable the hypervisor extension. But there is no implementation of
hypervisor extension in mainline for now, so let's consider the related
issues later.
This patch is cherry-picked from the riscv integration branch since the
svinval extension is frozen for now. Besides, we fix the funct7 encodings
of hinval.vvma and hinval.gvma, from 0x0011011 and 0x0111011 to 0x0010011
and 0x0110011.
Mike Frysinger [Tue, 30 Nov 2021 07:30:46 +0000 (02:30 -0500)]
sim: mips/or1k: drop redundant arg to bitsize macro
These are just using the default behavior for the 3rd arg, so drop
it to make it more clear. This also makes them match all other
ports that only use the first 2 arguments.
Mike Frysinger [Thu, 16 Dec 2021 07:25:21 +0000 (02:25 -0500)]
sim: fix mingw builds with replacement gnulib open
The header shuffling in here broke the workaround for gnulib defining
"open". Move it back before the sim-specific includes to fix. This
is because the callback struct in the headers has an "open" member and
this file tries to call that.
Sandra Loosemore [Wed, 15 Dec 2021 23:05:41 +0000 (16:05 -0700)]
Adjust compare_link_order for unstable qsort
In a cross toolchain for nios2-elf target and x86_64-w64-mingw32 host
using binutils 2.37, we observed a failure that didn't show up on
x86_64-linux-gnu host: testcase pr25490-5.s was failing with
C:\path\to\nios2-elf-ld.exe: looping in map_segments
FAIL: __patchable_function_entries section 5
* ldelfgen.c (compare_link_order): Don't use section id in
sorting. Keep original ordering instead. Update comments.
Alan Modra [Thu, 16 Dec 2021 00:20:58 +0000 (10:50 +1030)]
Re: Fix an undefined behaviour in the BFD library's DWARF parser
Using an unsigned int cast (to 32 bits) on a pointer difference (of
possibly 64 bits) is wrong. Even though it will work on all real
object files, the fuzzers will eventually find this hole.
PR 28687
* dwarf1.c (parse_die): Cast pointer difference to size_t.
Catch another possible pointer overflow.
Alan Modra [Wed, 15 Dec 2021 04:36:26 +0000 (15:06 +1030)]
PR28691, validate dwarf attribute form
PR28691 is a fuzzing PR that triggers a non-problem of "output changes
per run" with PIEs and/or different compilers. I've closed similar
PRs before as wontfix, but I guess there will be no end of this type
of PR. The trigger is an attribute that usually takes one of the
offset/constant reference DW_FORMs being given an indexed string
DW_FORM. The bfd reader doesn't support indexed strings and returns
an error string instead. The address of the string varies with PIE
runs and/or compiler, and we allow that address to appear in output.
Fix this by validating integer attribute forms, as we do for string
form attributes.
PR 28691
* dwarf2.c (is_str_attr): Rename to..
(is_str_form): ..this. Change param type. Update calls.
(is_int_form): New function.
(read_attribute_value): Handle DW_FORM_addrx2.
(find_abstract_instance): Validate form when using attr.u.val.
(scan_unit_for_symbols, parse_comp_unit): Likewise.
Luis Machado [Fri, 26 Nov 2021 14:31:18 +0000 (11:31 -0300)]
New --enable-threading configure option to control use of threads in GDB/GDBserver
Add the --enable-threading configure option so multithreading can be disabled
at configure time. This is useful for statically-linked builds of
GDB/GDBserver, since the thread library doesn't play well with that setup.
If you try to run a statically-linked GDB built with threading, it will crash
when setting up the number of worker threads.
This new option is also convenient when debugging GDB in a system with lots of
threads, where the thread discovery code in GDB will emit too many messages,
like so:
Alan Modra [Tue, 14 Dec 2021 11:42:17 +0000 (22:12 +1030)]
loongarch32 build failure on 32-bit host
gas/config/tc-loongarch.c: In function ‘assember_macro_helper’:
gas/config/tc-loongarch.c:915:28: error: right shift count >= width of type [-Werror=shift-count-overflow]
915 | hi32 = insn->args[1] >> 32;
| ^~
One possible fix is to make offsetT a 64-bit type for loongarch32.
This also makes bfd/targmatch.h (generated from bfd/config.bfd)
consistent since the loongarch32 match is inside #ifdef BFD64.
Alan Modra [Tue, 14 Dec 2021 11:36:29 +0000 (22:06 +1030)]
loongarch64 build failure on 32-bit host
gas/config/tc-loongarch.c: In function ‘loongarch_args_parser_can_match_arg_helper’:
gas/config/tc-loongarch.c:661:13: error: cast from pointer to integer of different size [-Werror=pointer
-to-int-cast]
661 | imm = (offsetT) str_hash_find (r_htab, arg);
| ^
Cast it to the correct size int, relying on normal integer promotions
if offsetT is larger than a pointer.
* config/tc-loongarch.c (loongarch_args_parser_can_match_arg_helper):
Cast return from str_hash_find to intptr_t, not offsetT.
Alan Modra [Tue, 14 Dec 2021 10:43:21 +0000 (21:13 +1030)]
XCOFF C_STSYM test failure on 32-bit host
This test was failing here and on another similar symbol:
[ 4](sec 1)(fl 0x00)(ty 0)(scl 143) (nx 0) 0x05d1745d11745d21 .bs
where correct output is
[ 4](sec 1)(fl 0x00)(ty 0)(scl 143) (nx 0) 0x000000000000000a .bs
The problem is caused by a 32-bit host pointer being sign-extended
when stored into a 64-bit bfd_vma, and then that value not being
trimmed back to 32 bits when used. The following belt-and-braces
patch fixes both the store and subsequent reads.
* coffcode.h (coff_slurp_symbol_table): Do not sign extend
when storing a host pointer to syment.n_value.
* coffgen.c (coff_get_symbol_info): Cast syment.n_value to a
bfd_hostptr_t before using in arithmetic.
(coff_print_symbol): Likewise.
... trying to start an application with libinproctrace.so loaded
crashes:
$ LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libasan.so.6:./libinproctrace.so /bin/ls
/home/smarchi/src/binutils-gdb/gdbserver/../gdbsupport/common-utils.cc:69: A problem internal to GDBserver in-process agent has been detected.
xsnprintf: Assertion `ret < size' failed.
Looking at the rest of the socket initialization code, the intent seems
to be that if something goes wrong, we warn but let the program
execute. So crashing on this failed assertions seems against the intent.
Commit 6cebaf6e1ae4 ("use xsnprintf instead of snprintf.") changed this
code to use xsnprintf instead of snprintf, introducing this assertion.
Before that, snprintf would return a value bigger that UNIX_PATH_MAX and
the "if" after would catch it and emit a warning, which is exactly what
we want. That change was done because LynxOS didn't have snprintf.
Since LynxOS isn't supported anymore, we can simply revert to use
snprintf there.
With this patch, we get a warning (printed by the caller of
gdb_agent_socket_init), but the program keeps executing:
$ LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libasan.so.6:./libinproctrace.so /bin/ls
ipa: could not create sync socket
...
Simon Marchi [Tue, 14 Dec 2021 19:35:03 +0000 (14:35 -0500)]
gdbserver/tracepoint.cc: work around -Wstringop-truncation error
When building gdb with on AArch64 with g++ 11.1.0 (and some preceding
versions too), -O2 and -fsanitize=address, I get this error.
CXX tracepoint-ipa.o
cc1plus: warning: command-line option ‘-Wmissing-prototypes’ is valid for C/ObjC but not for C++
In file included from /usr/include/string.h:519,
from ../gnulib/import/string.h:41,
from /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/common-defs.h:95,
from /home/simark/src/binutils-gdb/gdbserver/server.h:22,
from /home/simark/src/binutils-gdb/gdbserver/tracepoint.cc:19:
In function ‘char* strncpy(char*, const char*, size_t)’,
inlined from ‘int init_named_socket(const char*)’ at /home/simark/src/binutils-gdb/gdbserver/tracepoint.cc:6902:11,
inlined from ‘int gdb_agent_socket_init()’ at /home/simark/src/binutils-gdb/gdbserver/tracepoint.cc:6953:26,
inlined from ‘void* gdb_agent_helper_thread(void*)’ at /home/simark/src/binutils-gdb/gdbserver/tracepoint.cc:7204:41:
/usr/include/bits/string_fortified.h:95:34: error: ‘char* __builtin_strncpy(char*, const char*, long unsigned int)’ output may be truncated copying 107 bytes from a string of length 107 [-Werror=stringop-truncation]
95 | return __builtin___strncpy_chk (__dest, __src, __len,
| ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~
96 | __glibc_objsize (__dest));
| ~~~~~~~~~~~~~~~~~~~~~~~~~
Note that _FORTIFY_SOURCE changes the message a bit, but I get a similar
error if I use -D_FORTIFY_SOURCE=0.
I am pretty sure it's spurious and might be related to this GCC bug:
From what I can see, we are copying from a static 108 bytes long buffer
(the global array agent_socket_name) to a 108 bytes long array,
sockaddr_un::sun_path. I don't see anything wrong.
Still, it would make things easier if we didn't see this error. Change
the code to check that the source string length is smaller than the
destination buffer (including space for the NULL byte) and use strcpy.
For anybody who would like to try to reproduce, the full command line
is:
Simon Marchi [Fri, 3 Dec 2021 21:31:09 +0000 (16:31 -0500)]
bfd: fix -Wunused errors with clang 13+
Clang 13 and 14 produce some -Wunused-but-set-{variable,parameter} for
situations where gcc doesn't. In particular, when a variable is set and
then used in a way to update its own value. For example, if `i` is only
used in this way:
int i = 2;
i++;
i = i + 1;
gcc won't warn, but clang will.
Fix all such errors found in an --enable-targets=all build. It would be
important for somebody who knows what they're doing to just make sure
that these variables can indeed be deleted, and that there a no cases
where it's a bug, and the variable should actually be used.
The first instance of this error fix by this patch is:
CC elf32-score.lo
/home/simark/src/binutils-gdb/bfd/elf32-score.c:450:11: error: variable 'relocation' set but not used [-Werror,-Wunused-but-set-variable]
bfd_vma relocation;
^
Jan Vrany [Mon, 8 Nov 2021 14:00:13 +0000 (14:00 +0000)]
gdb/mi: rename mi_cmd to mi_command
Just give this class a new name, more inline with the name of the
sub-classes. I've also updated mi_cmd_up to mi_command_up in
mi-cmds.c inline with this new naming scheme.
There should be no user visible changes after this commit.
Jan Vrany [Tue, 23 Jun 2020 13:45:38 +0000 (14:45 +0100)]
gdb/mi: use separate classes for different types of MI command
This commit changes the infrastructure in mi-cmds.{c,h} to add new
sub-classes for the different types of MI command. Instances of these
sub-classes are then created and added into mi_cmd_table.
The existing mi_cmd class becomes the abstract base class, this has an
invoke method and takes care of the suppress notifications handling,
before calling a do_invoke virtual method which is implemented by all
of the sub-classes.
There's currently two different sub-classes, one of pure MI commands,
and a second for MI commands that delegate to CLI commands.
There should be no user visible changes after this commit.
Andrew Burgess [Mon, 8 Nov 2021 12:25:56 +0000 (12:25 +0000)]
gdb/mi: int to bool conversion in mi_execute_cli_command
Change an argument of mi_execute_cli_command from int to bool. Update
the callers to take this into account. Within mi_execute_cli_command,
update a comparison of a pointer to 0 with a comparison to nullptr,
and add an assert, if we are not using the argument string then the
string should be nullptr. Also removed a cryptic 'gdb_????' comment,
which isn't really helpful.
There should be no user visible changes after this commit.
Jan Vrany [Tue, 23 Jun 2020 13:45:38 +0000 (14:45 +0100)]
gdb/mi: use std::map for MI commands in mi-cmds.c
This changes the hashmap used in mi-cmds.c from a custom structure to
std::map. Not only is replacing a custom container with a standard
one an improvement, but using std::map will make it easier to
dynamically add commands; which is something that is planned for a
later series, where we will allow MI commands to be implemented in
Python.
There should be no user visible changes after this commit.
Nelson Chu [Tue, 14 Dec 2021 04:53:48 +0000 (12:53 +0800)]
RISC-V: Added ld testcases for the medlow and medany code models.
There are two linker scripts, code-model-01.ld and code-model-02.ld,
which are corresponding to the two different memory layouts,
* code-model-01.ld: the text section is in the 32-bit address range, but
the data section is far away from the text section, which means the data
section is over the 32-bit address range.
* code-model-02.ld: the text section is over the 32-bit address range, but
the data section is placed nearly zero address.
We use the two linker scripts, to test the current medlow and medany behaviors
of GNU ld, including the weak symbol references and the relaxations behaviors.
Besides, these testcases also show the limits of the current medlow and medany
code models, that is - we may get the truncated to fit errors when linking
with the above two linker scripts.
ld/
* testsuite/ld-riscv-elf/code-model-01.ld: New testcases to test the
behaviors of the current medlow and medany code models.
* testsuite/ld-riscv-elf/code-model-02.ld: Likewise.
* testsuite/ld-riscv-elf/code-model-medany-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medany-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medany-weakref-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medany-weakref-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medlow-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medlow-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medlow-weakref-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-medlow-weakref-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medany-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medany-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medany-weakref-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medany-weakref-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medlow-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medlow-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medlow-weakref-01.d: Likewise.
* testsuite/ld-riscv-elf/code-model-relax-medlow-weakref-02.d: Likewise.
* testsuite/ld-riscv-elf/code-model.s: Likewise.
* testsuite/ld-riscv-elf/ld-riscv-elf.exp: Updated.
Andrew Burgess [Fri, 15 Oct 2021 14:16:20 +0000 (15:16 +0100)]
gdb: improve reuse of value contents when fetching array elements
While working on a Python script, which was interacting with a remote
target, I noticed some weird slowness in GDB. In my program I had a
structure something like this:
struct foo_t
{
int array[5];
};
struct foo_t global_foo;
Then in the Python script I was fetching a complete copy of global
foo, like:
val = gdb.parse_and_eval('global_foo')
val.fetch_lazy()
Then I would work with items in foo_t.array, like:
print(val['array'][1])
I called the fetch_lazy method specifically because I knew I was going
to end up accessing almost all of the contents of val, and so I wanted
GDB to do a single remote protocol call to fetch all the contents in
one go, rather than trying to do lazy fetches for a couple of bytes at
a time.
What I observed was that, after the fetch_lazy call, GDB does,
correctly, fetch the entire contents of global_foo, including all of
the contents of array, however, when I access val.array[1], GDB still
goes and fetches the value of this element from the remote target.
What's going on is that in valarith.c, in value_subscript, for C like
languages, we always end up treating the array value as a pointer, and
then doing value_ptradd, and value_ind, the second of these calls
always returns a lazy value.
My guess is that this approach allows us to handle indexing off the
end of an array, when working with zero element arrays, or when
indexing a raw pointer as an array. And, I agree, that in these
cases, where, even when the original value is non-lazy, we still will
not have the content of the array loaded, we should be using the
value_ind approach.
However, for cases where we do have the array contents loaded, and we
do know the bounds of the array, I think we should be using
value_subscripted_rvalue, which is what we use for non C like
languages.
One problem I did run into, exposed by gdb.base/charset.exp, was that
value_subscripted_rvalue stripped typedefs from the element type of
the array, which means the value returned will not have the same type
as an element of the array, but would be the raw, non-typedefed,
type. In charset.exp we got back an 'int' instead of a
'wchar_t' (which is a typedef of 'int'), and this impacts how we print
the value. Removing typedefs from the resulting value just seems
wrong, so I got rid of that, and I don't see any test regressions.
With this change in place, my original Python script is now doing no
additional memory accesses, and its performance increases about 10x!
gdbserver: replace direct assignments to current_thread
Replace the direct assignments to current_thread with
switch_to_thread. Use scoped_restore_current_thread when appropriate.
There is one instance remaining in linux-low.cc's wait_for_sigstop.
This will be handled in a separate patch.
Regression-tested on X86-64 Linux using the native-gdbserver and
native-extended-gdbserver board files.
gdbserver: introduce scoped_restore_current_thread and switch_to_thread
Introduce a class for restoring the current thread and a function to
switch to the given thread. This is a preparation for a refactoring
that aims to remove direct assignments to 'current_thread'.
Andrew Burgess [Fri, 12 Nov 2021 11:18:58 +0000 (11:18 +0000)]
gdb: make post_startup_inferior a virtual method on inf_ptrace_target
While working on a later patch that required me to understand how GDB
starts up inferiors, I was confused by the
target_ops::post_startup_inferior method.
The post_startup_inferior target function is only called from
inf_ptrace_target::create_inferior.
Part of the target class hierarchy looks like this:
Every sub-class of inf_ptrace_target, except rs6000_nat_target,
implements ::post_startup_inferior. The rs6000_nat_target picks up
the implementation of ::post_startup_inferior not from
inf_ptrace_target, but from inf_child_target.
No descendent of inf_child_target, outside the inf_ptrace_target
sub-tree, implements ::post_startup_inferior, which isn't really
surprising, as they would never see the method called (remember, the
method is only called from inf_ptrace_target::create_inferior).
What I find confusing is the role inf_child_target plays in
implementing, what is really a helper function for just one of its
descendents.
In this commit I propose that we formally make ::post_startup_inferior
a helper function of inf_ptrace_target. To do this I will remove the
::post_startup_inferior from the target_ops API, and instead make this
a protected, pure virtual function on inf_ptrace_target.
I'll remove the empty implementation of ::post_startup_inferior from
the inf_child_target class, and add a new empty implementation to the
rs6000_nat_target class.
All the other descendents of inf_ptrace_target already provide an
implementation of this method and so don't need to change beyond
making the method protected within their class declarations.
To me, this makes much more sense now. The helper function, which is
only called from within the inf_ptrace_target class, is now a part of
the inf_ptrace_target class.
The only way in which this change is visible to a user is if the user
turns on 'set debug target 1'. With this debug flag on, prior to this
patch the user would see something like:
After this patch these lines are no longer present, as the
post_startup_inferior is no longer a top level target method. For me,
this is an acceptable change.
Andrew Burgess [Fri, 12 Nov 2021 12:04:08 +0000 (12:04 +0000)]
gdb: have mips_nbsd_nat_target inherit from nbsd_nat_target
While working on another patch I had reason to look at
mips-netbsd-nat.c, and noticed that the class mips_nbsd_nat_target
inherits directly from inf_ptrace_target.
This is a little strange as alpha_bsd_nat_target,
arm_netbsd_nat_target, hppa_nbsd_nat_target, m68k_bsd_nat_target,
ppc_nbsd_nat_target, sh_nbsd_nat_target, and vax_bsd_nat_target all
inherit from nbsd_nat_target.
When the target tree was converted to C++, all of the above classes
inherited from inf_ptrace_target except for hppa_nbsd_nat_target,
which was the only class that originally inherited from
nbsd_nat_target.
Later on all the remaining targets (except mips) were converted to
inherit from nbsd_nat_target, these are the commits:
I could only find mailing list threads for ppc and sh in the archive ,
and unfortunately, none of the commits has any real detail that might
explain why mips was missed out, the only extra context I could find
was this message:
Which says that "proper" OS support is going to be added to
nbsd_nat_target, hence the need to inherit from that class.
My guess is that leaving mips_nbsd_nat_target unchanged was an
oversight, so, in this commit, I propose changing mips_nbsd_nat_target
to inherit from nbsd_nat_target just like all the other nbsd targets.
My motivation for this patch relates to the post_startup_inferior
target method. In a future commit I want to change how this method is
handled. Currently the mips_nbsd_nat_target will pick up the empty
implementation of inf_child_target::post_startup_inferior rather than
the version in netbsd-nat.c. This feels like a bug to me, as surely,
enabling of proc events is something that would need to be done for
all netbsd targets, regardless of architecture.
In my future patch I have a choice then, either (a) add a new, empty
implementation of post_startup_inferior to mips_nbsd_nat_target,
or (b) this commit, have mips_nbsd_nat_target inherit from
nbsd_nat_target. Option (b) seems like the right way to go, hence,
this commit.
I've done absolutely no testing for this change, not even building it,
as that would require at least an environment in which I can x-build
mips-netbsd applications, which I have no idea how to set up.
Andrew Burgess [Thu, 9 Dec 2021 17:51:10 +0000 (17:51 +0000)]
gdb: only include mips and riscv targets if building with 64-bit bfd
While testing another patch I was trying to build different
configurations of GDB, and, during one test build I ran into a
problem, I configured with `--enable-targets=all
--host=i686-w64-mingw32` and saw this error while linking GDB:
.../i686-w64-mingw32/bin/ld: mips-tdep.o: in function `mips_gdbarch_init':
.../src/gdb/mips-tdep.c:8730: undefined reference to `disassembler_options_mips'
.../i686-w64-mingw32/bin/ld: riscv-tdep.o: in function `riscv_gdbarch_init':
.../src/gdb/riscv-tdep.c:3851: undefined reference to `disassembler_options_riscv'
So the `disassembler_options_mips` and `disassembler_options_riscv`
symbols are missing.
This turns out to be because mips-dis.c and riscv-dis.c, in which
these symbols are defined, are in the TARGET64_LIBOPCODES_CFILES list
in opcodes/Makefile.am, these files are only built when we are
building with a 64-bit bfd.
If we look further, into bfd/Makefile.am, we can see that all the
files matching elf*-riscv.lo are in the BFD64_BACKENDS list, as are
the elf*-mips.lo files, and (I know because I tried), the two
disassemblers do, not surprisingly, depend on features supplied from
libbfd.
So, though we can build most of GDB just fine for riscv and mips with
a 32-bit bfd, if I understand correctly, the final GDB
executable (assuming we could get it to link) would not understand
these architectures at the bfd level, nor would there be any
disassembler available. This sounds like a pretty useless GDB to me.
So, in this commit, I move the riscv and mips targets into GDB's list
of targets that require a 64-bit bfd. After this I can build GDB just
fine with the configure options given above.
This was discussed on the mailing list in a couple of threads:
and it is agreed, that it is unfortunate that the 32-bit riscv and
32-bit mips targets require a 64-bit bfd. If in the future this
situation ever changes then it would be expected that some (or all) of
this patch would be reverted. Until then though, this patch allows
GDB to build when configured with --enable-targets=all, and when using
a 32-bit libbfd.
Tom Tromey [Tue, 26 Oct 2021 00:20:24 +0000 (18:20 -0600)]
C++-ify path substitution code
I found some uses of xfree in the path substitution code in source.c.
C++-ifying struct substitute_path_rule both simplifies the code and
removes manual memory management.
Jan Vrany [Fri, 10 Dec 2021 11:43:20 +0000 (11:43 +0000)]
gdb/testsuite: respect GDBSERVER variable in remote-stdio-gdbserver "board"
The comment on top of gdb/testsuite/boards/remote-stdio-gdbserver.exp says
that user can specify path to gdbserver on remote system by setting
GDBSERVER variable. However, this variable was ignored and /usr/bin/gdbserver
was used unconditionally.
This commit updates the code to respect GDBSERVER if set and fall back to
/usr/bin/gdbserver if not.
Alan Modra [Wed, 8 Dec 2021 09:34:25 +0000 (20:04 +1030)]
Set sh_link for reloc sections created as normal sections
binutils-all/strip-13 and binutils-all/strip-14 tests create
SHT_REL/SHT_RELA sections by hand. These don't have sh_link set to
the .symtab section as they should, leading to readelf warnings if you
happen to be looking at the object files.
* elf.c (assign_section_numbers): Formatting. Set sh_link for
reloc sections created as normal sections in relocatable
objects.
I suppose this code was copied from GDBserver and this ifndef was left
there. As far as I know, IN_PROCESS_AGENT will never be defined when
building this file, so we can remove this.
CXX microblaze-tdep.o
/home/simark/src/binutils-gdb/gdb/microblaze-tdep.c:207:7: error: variable 'flags' set but not used [-Werror,-Wunused-but-set-variable]
int flags = 0;
^
CXX csky-tdep.o
/home/simark/src/binutils-gdb/gdb/csky-tdep.c:332:7: error: variable 'need_dummy_stack' set but not used [-Werror,-Wunused-but-set-variable]
int need_dummy_stack = 0;
^
/home/simark/src/binutils-gdb/gdb/csky-tdep.c:805:12: error: variable 'offset' set but not used [-Werror,-Wunused-but-set-variable]
int offset = 0;
^
Simon Marchi [Wed, 8 Dec 2021 02:56:48 +0000 (21:56 -0500)]
gdb/testsuite: fix default behavior of runto
The documented behavior of proc runto is to not emit a PASS when
succeeding to to run to the specified location, but emit a FAIL when
failing to do so. I suppose the intent is that it won't pollute the
results normally passing tests (although I don't see why we would care),
but make visible any problems.
However, it seems like the implementation makes it default to never
print anything. "no-message" is prependend to "args", so if "message"
is not passed, we will always take the path that sets print_fail to 0,
which will silence any failure.
This unfortunately means that tests relying on runto_main won't emit a
FAIL if failing to run to main. And since commit 4dfef5be6812
("gdb/testsuite: make runto_main not pass no-message to runto"), tests
don't emit a FAIL themselves when failing to run to main. This means
that tests failing to run to main just fail silently, and that's bad.
This can be reproduced by hacking gdb.base/template.exp like so:
diff --git a/gdb/testsuite/gdb.base/template.c b/gdb/testsuite/gdb.base/template.c
index bcf39c377d92..052be5b79d73 100644
--- a/gdb/testsuite/gdb.base/template.c
+++ b/gdb/testsuite/gdb.base/template.c
@@ -15,6 +15,14 @@
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>. */
+#include <stdlib.h>
+
+__attribute__((constructor))
+static void c (void)
+{
+ exit (1);
+}
+
int
main (void)
{
Running the modified gdb.base/template.exp shows that it exits without
printing any result.
Remove the line that prepends no-message to args, that should make
runto's behavior match its documentation.
This patch will appear to add many failures, but in reality they already
existed, they were just silenced.
On ELFv1, the _start symbol must point to the *function descriptor* (in
the .opd section), not to the function code (in the .text section) like
with ELFv2 and other architectures.