binutils-gdb.git
2 years agoDon't return the main file as the separate debug info
H.J. Lu [Fri, 10 Dec 2021 21:34:22 +0000 (13:34 -0800)]
Don't return the main file as the separate debug info

On Fedora 35,

$ readelf -d /usr/bin/npc

caused readelf to run out of stack since load_separate_debug_info
returned the input main file as the separate debug info:

(gdb) bt
 #0  load_separate_debug_info (
    main_filename=main_filename@entry=0x510f50 "/export/home/hjl/.cache/debuginfod_client/dcc33c51c49e7dafc178fdb5cf8bd8946f965295/debuginfo",
    xlink=xlink@entry=0x4e5180 <debug_displays+4480>,
    parse_func=parse_func@entry=0x431550 <parse_gnu_debuglink>,
    check_func=check_func@entry=0x432ae0 <check_gnu_debuglink>,
    func_data=func_data@entry=0x7fffffffdb60, file=file@entry=0x51d430)
    at /export/gnu/import/git/sources/binutils-gdb/binutils/dwarf.c:11057
 #1  0x000000000043328d in check_for_and_load_links (file=0x51d430,
    filename=0x510f50 "/export/home/hjl/.cache/debuginfod_client/dcc33c51c49e7dafc178fdb5cf8bd8946f965295/debuginfo")
    at /export/gnu/import/git/sources/binutils-gdb/binutils/dwarf.c:11381
 #2  0x00000000004332ae in check_for_and_load_links (file=0x51b070,
    filename=0x518dd0 "/export/home/hjl/.cache/debuginfod_client/dcc33c51c49e7dafc178fdb5cf8bd8946f965295/debuginfo")

Return NULL if the separate debug info is the same as the input main
file to avoid infinite recursion.

PR binutils/28679
* dwarf.c (load_separate_debug_info): Don't return the input
main file.

2 years agoDon't edit bogus sh_link on reading relocatable objects (Oracle fix)
Alan Modra [Thu, 9 Dec 2021 09:14:07 +0000 (19:44 +1030)]
Don't edit bogus sh_link on reading relocatable objects (Oracle fix)

This reverts a 1995 fix to handle bogus object files.  Presumably such
object files have long gone.

* elf.c (bfd_section_from_shdr): Remove old hack for Oracle
libraries.

2 years agoAutomatic date update in version.in
GDB Administrator [Sat, 11 Dec 2021 00:00:05 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agogdb/testsuite: respect GDBSERVER variable in remote-stdio-gdbserver "board"
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.

2 years agoRevert "gdbsupport: remove unnecessary `#ifndef IN_PROCESS_AGENT`"
Simon Marchi [Fri, 10 Dec 2021 00:02:38 +0000 (19:02 -0500)]
Revert "gdbsupport: remove unnecessary `#ifndef IN_PROCESS_AGENT`"

This reverts commit fe72c32765e1190c8a17d309fc3a7e1882d6a430.

It turns out it was wrong, libinproctrace.so does build its own
gdbsupport/tdesc.cc.  This broke the build:

    make[1]: Entering directory '/home/simark/build/binutils-gdb-one-target/gdbserver'
      CXXLD  libinproctrace.so
    /usr/bin/ld: gdbsupport/tdesc-ipa.o: in function `print_xml_feature::visit_pre(target_desc const*)':
    /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/tdesc.cc:407: undefined reference to `tdesc_architecture_name(target_desc const*)'
    /usr/bin/ld: /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/tdesc.cc:408: undefined reference to `tdesc_architecture_name(target_desc const*)'
    /usr/bin/ld: /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/tdesc.cc:411: undefined reference to `tdesc_osabi_name(target_desc const*)'
    /usr/bin/ld: /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/tdesc.cc:416: undefined reference to `tdesc_compatible_info_list(target_desc const*)'
    /usr/bin/ld: /home/simark/src/binutils-gdb/gdbserver/../gdbsupport/tdesc.cc:418: undefined reference to `tdesc_compatible_info_arch_name(std::unique_ptr<tdesc_compatible_info, std::default_delete<tdesc_compatible_info> > const&)'

2 years agoAutomatic date update in version.in
GDB Administrator [Fri, 10 Dec 2021 00:00:18 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agoPR28674, objdump crash
Alan Modra [Thu, 9 Dec 2021 21:57:58 +0000 (08:27 +1030)]
PR28674, objdump crash

Not returning an error indication here leaves the attribute
uninitialised, which then leads to intemperate behaviour.

PR 28674
* dwarf2.c (read_attribute_value): Return NULL on trying to read
past end of attributes.

2 years agoSet sh_link for reloc sections created as normal sections
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.

2 years agogdbsupport: remove unnecessary `#ifndef IN_PROCESS_AGENT`
Simon Marchi [Thu, 9 Dec 2021 19:12:18 +0000 (14:12 -0500)]
gdbsupport: remove unnecessary `#ifndef IN_PROCESS_AGENT`

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.

Change-Id: I84fc408e330b3a29106df830a09342861cadbaf6

2 years agogdb/microblaze-tdep.c: fix -Wunused-but-set-variable
Simon Marchi [Fri, 3 Dec 2021 21:31:28 +0000 (16:31 -0500)]
gdb/microblaze-tdep.c: fix -Wunused-but-set-variable

Fix this, seen when building with clang 14:

      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;
          ^

Change-Id: I59f726ed33e924912748bc475b6fd9a9394fc0d0

2 years agogdb/csky-tdep.c: fix -Wunused-but-set-variable error
Simon Marchi [Fri, 3 Dec 2021 21:31:27 +0000 (16:31 -0500)]
gdb/csky-tdep.c: fix -Wunused-but-set-variable error

Fix these, seen when building with clang 14:

      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;
                      ^

Change-Id: I6703bcb50e83c50083f716f4084ef6aa30d659c4

2 years agogdb/testsuite: fix default behavior of runto
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.

Change-Id: I2a730d5bc72b6ef0698cd6aad962d9902aa7c3d6

2 years agogdb fix elfv1 Powerpc gdb.dwarf2/frame-inlined-in-outer-frame.exp
Carl Love [Wed, 8 Dec 2021 18:37:13 +0000 (12:37 -0600)]
gdb fix elfv1 Powerpc gdb.dwarf2/frame-inlined-in-outer-frame.exp

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.

2 years ago[gdb/testsuite] Fix gdb.base/maint.exp with -readnow
Tom de Vries [Thu, 9 Dec 2021 10:15:34 +0000 (11:15 +0100)]
[gdb/testsuite] Fix gdb.base/maint.exp with -readnow

With test-case gdb.base/maint.exp and target board -readnow, I run into:
...
FAIL: gdb.base/maint.exp: maint info line-table w/o a file name
...

The problem is that this and other regexps anchored using '^':
...
    -re "^$gdb_prompt $" {
...
don't trigger because other regexps don't consume the entire preceding line.

This is partly due to the addition of the IS-STMT column.

Fix this by making the regexps consume entire lines.

Tested on x86_64-linux with native and target board readnow, as well as
check-read1 and check-readmore.

2 years ago[gdb/testsuite] Fix gdb.base/include-main.exp with -readnow
Tom de Vries [Thu, 9 Dec 2021 10:15:34 +0000 (11:15 +0100)]
[gdb/testsuite] Fix gdb.base/include-main.exp with -readnow

With test-case gdb.base/include-main.exp and target board readnow, I run into:
...
FAIL: gdb.base/include-main.exp: maint info symtab
...

The corresponding check in gdb.base/include-main.exp:
...
gdb_test_no_output "maint info symtab"
...
checks that no CU was expanded, while -readnow ensures that all CUs are
expanded.

Fix this by skipping the check with -readnow.

Tested on x86_64-linux, with native and target board readnow.

2 years agoRISC-V: Clarify the behavior of .option arch directive.
Nelson Chu [Thu, 9 Dec 2021 03:52:16 +0000 (11:52 +0800)]
RISC-V: Clarify the behavior of .option arch directive.

* To be consistent with -march option, removed the "=" operator when
user want to reset the whole architecture string.  So the formats are,

.option arch, +<extension><version>, ...
.option arch, -<extension>
.option arch, <ISA string>

* Don't allow to add or remove the base extensions in the .option arch
directive.  Instead, users should reset the whole architecture string
while they want to change the base extension.

* The operator "+" won't update the version of extension, if the
extension is already in the subset list.

bfd/
* elfxx-riscv.c (riscv_add_subset): Don't update the version
if the extension is already in the subset list.
(riscv_update_subset): To be consistent with -march option,
removed the "=" operator when user want to reset the whole
architecture string.  Besides, Don't allow to add or remove
the base extensions in the .option arch directive.
gas/
* testsuite/gas/riscv/option-arch-01.s: Updated since we cannot
add or remove the base extensions in the .option arch directive.
* testsuite/gas/riscv/option-arch-02.s: Likewise.
* testsuite/gas/riscv/option-arch-fail.l: Likewise.
* testsuite/gas/riscv/option-arch-fail.s: Likewise.
* testsuite/gas/riscv/option-arch-01a.d: Set -misa-spec=2.2.
* testsuite/gas/riscv/option-arch-01b.d: Likewise.
* testsuite/gas/riscv/option-arch-02.d: Updated since the .option
arch, + won't change the version of extension, if the extension is
already in the subset list.
* testsuite/gas/riscv/option-arch-03.s: Removed the "=" operator
when resetting the whole architecture string.

2 years agosim: use ## for automake comments
Mike Frysinger [Thu, 9 Dec 2021 06:40:28 +0000 (01:40 -0500)]
sim: use ## for automake comments

The ## marker tells automake to not include the comment in its
generated output, so use that in most places where the comment
only makes sense in the inputs.

2 years agogdb, gdbserver: detach fork child when detaching from fork parent
Simon Marchi [Wed, 1 Dec 2021 14:40:03 +0000 (09:40 -0500)]
gdb, gdbserver: detach fork child when detaching from fork parent

While working with pending fork events, I wondered what would happen if
the user detached an inferior while a thread of that inferior had a
pending fork event.  What happens with the fork child, which is
ptrace-attached by the GDB process (or by GDBserver), but not known to
the core?  Sure enough, neither the core of GDB or the target detach the
child process, so GDB (or GDBserver) just stays ptrace-attached to the
process.  The result is that the fork child process is stuck, while you
would expect it to be detached and run.

Make GDBserver detach of fork children it knows about.  That is done in
the generic handle_detach function.  Since a process_info already exists
for the child, we can simply call detach_inferior on it.

GDB-side, make the linux-nat and remote targets detach of fork children
known because of pending fork events.  These pending fork events can be
stored in:

 - thread_info::pending_waitstatus, if the core has consumed the event
   but then saved it for later (for example, because it got the event
   while stopping all threads, to present an all-stop stop on top of a
   non-stop target)
 - thread_info::pending_follow: if we ran to a "catch fork" and we
   detach at that moment

Additionally, pending fork events can be in target-specific fields:

 - For linux-nat, they can be in lwp_info::status and
   lwp_info::waitstatus.
 - For the remote target, they could be stored as pending stop replies,
   saved in `remote_state::notif_state::pending_event`, if not
   acknowledged yet, or in `remote_state::stop_reply_queue`, if
   acknowledged.  I followed the model of remove_new_fork_children for
   this: call remote_notif_get_pending_events to process /
   acknowledge any unacknowledged notification, then look through
   stop_reply_queue.

Update the gdb.threads/pending-fork-event.exp test (and rename it to
gdb.threads/pending-fork-event-detach.exp) to try to detach the process
while it is stopped with a pending fork event.  In order to verify that
the fork child process is correctly detached and resumes execution
outside of GDB's control, make that process create a file in the test
output directory, and make the test wait $timeout seconds for that file
to appear (it happens instantly if everything goes well).

This test catches a bug in linux-nat.c, also reported as PR 28512
("waitstatus.h:300: internal-error: gdb_signal target_waitstatus::sig()
const: Assertion `m_kind == TARGET_WAITKIND_STOPPED || m_kind ==
TARGET_WAITKIND_SIGNALLED' failed.).  When detaching a thread with a
pending event, get_detach_signal unconditionally fetches the signal
stored in the waitstatus (`tp->pending_waitstatus ().sig ()`).  However,
that is only valid if the pending event is of type
TARGET_WAITKIND_STOPPED, and this is now enforced using assertions (iit
would also be valid for TARGET_WAITKIND_SIGNALLED, but that would mean
the thread does not exist anymore, so we wouldn't be detaching it).  Add
a condition in get_detach_signal to access the signal number only if the
wait status is of kind TARGET_WAITKIND_STOPPED, and use GDB_SIGNAL_0
instead (since the thread was not stopped with a signal to begin with).

Add another test, gdb.threads/pending-fork-event-ns.exp, specifically to
verify that we consider events in pending stop replies in the remote
target.  This test has many threads constantly forking, and we detach
from the program while the program is executing.  That gives us some
chance that we detach while a fork stop reply is stored in the remote
target.  To verify that we correctly detach all fork children, we ask
the parent to exit by sending it a SIGUSR1 signal and have it write a
file to the filesystem before exiting.  Because the parent's main thread
joins the forking threads, and the forking threads wait for their fork
children to exit, if some fork child is not detach by GDB, the parent
will not write the file, and the test will time out.  If I remove the
new remote_detach_pid calls in remote.c, the test fails eventually if I
run it in a loop.

There is a known limitation: we don't remove breakpoints from the
children before detaching it.  So the children, could hit a trap
instruction after being detached and crash.  I know this is wrong, and
it should be fixed, but I would like to handle that later.  The current
patch doesn't fix everything, but it's a step in the right direction.

Change-Id: I6d811a56f520e3cb92d5ea563ad38976f92e93dd
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28512

2 years agogdb: move clearing of tp->pending_follow to follow_fork_inferior
Simon Marchi [Wed, 1 Dec 2021 14:40:03 +0000 (09:40 -0500)]
gdb: move clearing of tp->pending_follow to follow_fork_inferior

A following patch will change targets so that when they detach an
inferior, they also detach any pending fork children this inferior may
have.  While doing this, I hit a case where we couldn't differentiate
two cases, where in one we should detach the fork detach but not in the
other.

Suppose we continue past a fork with "follow-fork-mode == child" &&
"detach-on-fork on".  follow_fork_inferior calls target_detach to detach
the parent.  In that case the target should not detach the fork
child, as we'll continue debugging the child.  As of now, the
tp->pending_follow field of the thread who called fork still contains
the details about the fork.

Then, suppose we run to a fork catchpoint and the user types "detach".
In that case, the target should detach the fork child in addition to the
parent.  In that case as well, the tp->pending_follow field contains
the details about the fork.

To allow targets to differentiate the two cases, clear
tp->pending_follow a bit earlier, when following a fork.  Targets will
then see that tp->pending_follow contains TARGET_WAITKIND_SPURIOUS, and
won't detach the fork child.

As of this patch, no behavior changes are expected.

Change-Id: I537741859ed712cb531baaefc78bb934e2a28153

2 years agogdb/remote.c: refactor pending fork status functions
Simon Marchi [Wed, 1 Dec 2021 14:40:02 +0000 (09:40 -0500)]
gdb/remote.c: refactor pending fork status functions

In preparation for a following patch, refactor a few things that I did
find a bit awkward, and to make them a bit more reusable.

 - Pass an inferior to kill_new_fork_children instead of a pid.  That
   allows iterating on only this inferior's threads and avoid further
   filtering on the thread's pid.
 - Change thread_pending_fork_status to return a non-nullptr value only
   if the thread does have a pending fork status.
 - Remove is_pending_fork_parent_thread, as one can just use
   thread_pending_fork_status and check for nullptr.
 - Replace is_pending_fork_parent with is_fork_status, which just
   returns if the given target_waitkind if a fork or a vfork.  Push
   filtering on the pid to the callers, when it is necessary.

Change-Id: I0764ccc684d40f054e39df6fa5458cc4c5d1cd7b

2 years agogdb/remote.c: move some things up
Simon Marchi [Wed, 1 Dec 2021 14:40:02 +0000 (09:40 -0500)]
gdb/remote.c: move some things up

Move the stop_reply and a few functions up.  Some code above them in the
file will need to use them in a following patch.  No behavior changes
expected here.

Change-Id: I3ca57d0e3ec253f56e1ba401289d9d167de14ad2

2 years agogdb/linux-nat: factor ptrace-detach code to new detach_one_pid function
Simon Marchi [Wed, 1 Dec 2021 14:40:02 +0000 (09:40 -0500)]
gdb/linux-nat: factor ptrace-detach code to new detach_one_pid function

The following patch will add some code paths that need to ptrace-detach
a given PID.  Factor out the code that does this and put it in its own
function, so that it can be re-used.

Change-Id: Ie65ca0d89893b41aea0a23d9fc6ffbed042a9705

2 years agogdbserver: hide fork child threads from GDB
Simon Marchi [Wed, 1 Dec 2021 14:40:02 +0000 (09:40 -0500)]
gdbserver: hide fork child threads from GDB

This patch aims at fixing a bug where an inferior is unexpectedly
created when a fork happens at the same time as another event, and that
other event is reported to GDB first (and the fork event stays pending
in GDBserver).  This happens for example when we step a thread and
another thread forks at the same time.  The bug looks like (if I
reproduce the included test by hand):

    (gdb) show detach-on-fork
    Whether gdb will detach the child of a fork is on.
    (gdb) show follow-fork-mode
    Debugger response to a program call of fork or vfork is "parent".
    (gdb) si
    [New inferior 2]
    Reading /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread from remote target...
    Reading /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread from remote target...
    Reading symbols from target:/home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread...
    [New Thread 965190.965190]
    [Switching to Thread 965190.965190]
    Remote 'g' packet reply is too long (expected 560 bytes, got 816 bytes): ... <long series of bytes>

The sequence of events leading to the problem is:

 - We are using the all-stop user-visible mode as well as the
   synchronous / all-stop variant of the remote protocol
 - We have two threads, thread A that we single-step and thread B that
   calls fork at the same time
 - GDBserver's linux_process_target::wait pulls the "single step
   complete SIGTRAP" and the "fork" events from the kernel.  It
   arbitrarily choses one event to report, it happens to be the
   single-step SIGTRAP.  The fork stays pending in the thread_info.
 - GDBserver send that SIGTRAP as a stop reply to GDB
 - While in stop_all_threads, GDB calls update_thread_list, which ends
   up querying the remote thread list using qXfer:threads:read.
 - In the reply, GDBserver includes the fork child created as a result
   of thread B's fork.
 - GDB-side, the remote target sees the new PID, calls
   remote_notice_new_inferior, which ends up unexpectedly creating a new
   inferior, and things go downhill from there.

The problem here is that as long as GDB did not process the fork event,
it should pretend the fork child does not exist.  Ultimately, this event
will be reported, we'll go through follow_fork, and that process will be
detached.

The remote target (GDB-side), has some code to remove from the reported
thread list the threads that are the result of forks not processed by
GDB yet.  But that only works for fork events that have made their way
to the remote target (GDB-side), but haven't been consumed by the core
yet, so are still lingering as pending stop replies in the remote target
(see remove_new_fork_children in remote.c).  But in our case, the fork
event hasn't made its way to the GDB-side remote target.  We need to
implement the same kind of logic GDBserver-side: if there exists a
thread / inferior that is the result of a fork event GDBserver hasn't
reported yet, it should exclude that thread / inferior from the reported
thread list.

This was actually discussed a while ago, but not implemented AFAIK:

    https://pi.simark.ca/gdb-patches/1ad9f5a8-d00e-9a26-b0c9-3f4066af5142@redhat.com/#t
    https://sourceware.org/pipermail/gdb-patches/2016-June/133906.html

Implementation details-wise, the fix for this is all in GDBserver.  The
Linux layer of GDBserver already tracks unreported fork parent / child
relationships using the lwp_info::fork_relative, in order to avoid
wildcard actions resuming fork childs unknown to GDB.  This information
needs to be made available to the handle_qxfer_threads_worker function,
so it can filter the reported threads.  Add a new thread_pending_parent
target function that allows the Linux target to return the parent of an
eventual fork child.

Testing-wise, the test replicates pretty-much the sequence of events
shown above.  The setup of the test makes it such that the main thread
is about to fork.  We stepi the other thread, so that the step completes
very quickly, in a single event.  Meanwhile, the main thread is resumed,
so very likely has time to call fork.  This means that the bug may not
reproduce every time (if the main thread does not have time to call
fork), but it will reproduce more often than not.  The test fails
without the fix applied on the native-gdbserver and
native-extended-gdbserver boards.

At some point I suspected that which thread called fork and which thread
did the step influenced the order in which the events were reported, and
therefore the reproducibility of the bug.  So I made the test try  both
combinations: main thread forks while other thread steps, and vice
versa.  I'm not sure this is still necessary, but I left it there
anyway.  It doesn't hurt to test a few more combinations.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28288
Change-Id: I2158d5732fc7d7ca06b0eb01f88cf27bf527b990

2 years agoAutomatic date update in version.in
GDB Administrator [Thu, 9 Dec 2021 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agoUse for-each more in gdb
Tom Tromey [Fri, 3 Dec 2021 21:45:37 +0000 (14:45 -0700)]
Use for-each more in gdb

There are some loops in gdb that use ARRAY_SIZE (or a wordier
equivalent) to loop over a static array.  This patch changes some of
these to use foreach instead.

Regression tested on x86-64 Fedora 34.

2 years agoFix error in file_and_directory patch
Tom Tromey [Wed, 8 Dec 2021 19:52:34 +0000 (12:52 -0700)]
Fix error in file_and_directory patch

In my earlier C++-ization patch for file_and_directory, I introduced
an error:

-  if (strcmp (fnd.name, "<unknown>") != 0)
+  if (fnd.is_unknown ())

This change inverted the sense of the test, which causes failures with
.debug_names.

This patch fixes the bug.  Regression tested on x86-64 Fedora 34.  I
also tested it using the AdaCore internal test suite, with
.debug_names -- this was failing before, and now it works.

2 years agogdb/python: Use tp_init instead of tp_new to setup gdb.Value
Andrew Burgess [Wed, 17 Nov 2021 11:55:37 +0000 (11:55 +0000)]
gdb/python: Use tp_init instead of tp_new to setup gdb.Value

The documentation suggests that we implement gdb.Value.__init__,
however, this is not currently true, we really implement
gdb.Value.__new__.  This will cause confusion if a user tries to
sub-class gdb.Value.  They might write:

  class MyVal (gdb.Value):
      def __init__ (self, val):
          gdb.Value.__init__(self, val)

  obj = MyVal(123)
  print ("Got: %s" % obj)

But, when they source this code they'll see:

  (gdb) source ~/tmp/value-test.py
  Traceback (most recent call last):
    File "/home/andrew/tmp/value-test.py", line 7, in <module>
      obj = MyVal(123)
    File "/home/andrew/tmp/value-test.py", line 5, in __init__
      gdb.Value.__init__(self, val)
  TypeError: object.__init__() takes exactly one argument (the instance to initialize)
  (gdb)

The reason for this is that, as we don't implement __init__ for
gdb.Value, Python ends up calling object.__init__ instead, which
doesn't expect any arguments.

The Python docs suggest that the reason why we might take this
approach is because we want gdb.Value to be immutable:

   https://docs.python.org/3/c-api/typeobj.html#c.PyTypeObject.tp_new

But I don't see any reason why we should require gdb.Value to be
immutable when other types defined in GDB are not.  This current
immutability can be seen in this code:

  obj = gdb.Value(1234)
  print("Got: %s" % obj)
  obj.__init__ (5678)
  print("Got: %s" % obj)

Which currently runs without error, but prints:

  Got: 1234
  Got: 1234

In this commit I propose that we switch to using __init__ to
initialize gdb.Value objects.

This does introduce some additional complexity, during the __init__
call a gdb.Value might already be associated with a gdb value object,
in which case we need to cleanly break that association before
installing the new gdb value object.  However, the cost of doing this
is not great, and the benefit - being able to easily sub-class
gdb.Value seems worth it.

After this commit the first example above works without error, while
the second example now prints:

  Got: 1234
  Got: 5678

In order to make it easier to override the gdb.Value.__init__ method,
I have tweaked the definition of gdb.Value.__init__.  The second,
optional argument to __init__ is a gdb.Type, if this argument is not
present then GDB figures out a suitable type.

However, if we want to override the __init__ method in a sub-class,
and still support the default argument, it is easier to write:

  class MyVal (gdb.Value):
      def __init__ (self, val, type=None):
          gdb.Value.__init__(self, val, type)

Currently, passing None for the Type will result in an error:

  TypeError: type argument must be a gdb.Type.

After this commit I now allow the type argument to be None, in which
case GDB figures out a suitable type just as if the type had not been
passed at all.

Unless a user is trying to reinitialize a value, or create sub-classes
of gdb.Value, there should be no user visible changes after this
commit.

2 years agogdb: use try/catch around a gdb_disassembler::print_insn call
Andrew Burgess [Sat, 6 Nov 2021 11:25:12 +0000 (11:25 +0000)]
gdb: use try/catch around a gdb_disassembler::print_insn call

While investigating some disassembler problems I ran into this case;
GDB compiled on a 32-bit arm target, with --enable-targets=all.  Then
in GDB:

  (gdb) set architecture i386
  (gdb) disassemble 0x0,+4
  unknown disassembler error (error = -1)

This is interesting because it shows a case where the libopcodes
disassembler is returning -1 without first calling the
memory_error_func callback.  Indeed, the return from libopcodes
happens from this code snippet in i386-dis.c in the print_insn
function:

  if (address_mode == mode_64bit && sizeof (bfd_vma) < 8)
    {
      (*info->fprintf_func) (info->stream,
     _("64-bit address is disabled"));
      return -1;
    }

Notice how, prior to the return the disassembler tries to print a
helpful message out, but GDB doesn't print this message.

The reason this message goes missing is the call stack, it looks like
this:

  gdb_pretty_print_disassembler::pretty_print_insn
    gdb_disassembler::print_insn
      gdbarch_print_insn
        ...
          i386-dis.c:print_insn

When i386-dis.c:print_insn returns -1 this is handled in
gdb_disassembler::print_insn, where an exception is thrown.  However,
the actual printing of the disassembler output is done in
gdb_pretty_print_disassembler::pretty_print_insn, and is only done if
an exception is not thrown.

In this commit I change this.  The pretty_print_insn now uses
try/catch around the call to gdb_disassembler::print_insn, if we catch
an error then we first print any pending output in the instruction
buffer, before rethrowing the exception.  As a result, even if an
exception is thrown we still print any pending disassembler output to
the screen; in the above case the helpful message will now be shown.

Before my patch we might expect to see this output:

  (gdb) disassemble 0x0,+4
  Dump of assembler code from 0x0 to 0x4:
     0x0000000000000000: unknown disassembler error (error = -1)
  (gdb)

But now we see this:

  (gdb) disassemble 0x0,+4
  Dump of assembler code from 0x0 to 0x4:
     0x0000000000000000: 64-bit address is disabled
  unknown disassembler error (error = -1)

If the disassembler returns -1 without printing a helpful message then
we would still expect a change in output, something like:

  (gdb) disassemble 0x0,+4
  Dump of assembler code from 0x0 to 0x4:
     0x0000000000000000:
  unknown disassembler error (error = -1)

Which I think is still acceptable, though at this point I think a
strong case can be made that this is a disassembler bug (not printing
anything, but still returning -1).

Notice however, that the error message is always printed on a new line
now.  This is also true for the memory error case, where before we
might see this:

  (gdb) disassemble 0x0,+4
  Dump of assembler code from 0x0 to 0x4:
     0x00000000: Cannot access memory at address 0x0

We now get this:

  (gdb) disassemble 0x0,+4
  Dump of assembler code from 0x0 to 0x4:
     0x00000000:
  Cannot access memory at address 0x0

For me, I'm happy to accept this change, having the error on a line by
itself, rather than just appended to the end of the previous line,
seems like an improvement, but I'm aware others might feel
differently, so I'd appreciate any feedback.

2 years agoppc: recognize all program traps
Jan Vrany [Wed, 8 Dec 2021 10:46:49 +0000 (10:46 +0000)]
ppc: recognize all program traps

Permanent program breakpoints (ones inserted into the code) other than
the one GDB uses for POWER (0x7fe00008) did not result in stop but
caused GDB to loop infinitely.

This was because GDB did not recognize trap instructions other than
"trap". For example, "tw 12, 4, 4" was not be recognized, causing GDB
to loop forever.

This commit fixes this by providing POWER specific hook
(gdbarch_program_breakpoint_here_p) recognizing all tw, twi, td and tdi
instructions.

Tested on Linux on PowerPC e500 and on QEMU PPC64le.

2 years agoppc: use "trap" ("tw, 31, 0, 0") as breakpoint instruction
Jan Vrany [Wed, 8 Dec 2021 10:46:49 +0000 (10:46 +0000)]
ppc: use "trap" ("tw, 31, 0, 0") as breakpoint instruction

Power ISA 3.0 B spec [1], sections 3.3.11 "Fixed-Point Trap Instructions"
and section C.6 "Trap Mnemonics" specify "tw, 31, 0, 0" (encoded as
0x7fe00008) as canonical unconditional trap instruction.

This commit changes the breakpoint instruction used by GDB from
"tw 12, r2, r2" to unconditional "trap".

[1]: https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0

2 years agobfd_section_from_shdr: Support SHT_RELR sections
Fangrui Song [Wed, 8 Dec 2021 02:24:14 +0000 (18:24 -0800)]
bfd_section_from_shdr: Support SHT_RELR sections

If a.so contains an SHT_RELR section, objcopy a.so will fail with:

    a.so: unknown type [0x13] section `.relr.dyn'

This change allows objcopy to work.

bfd/
    * elf.c (bfd_section_from_shdr): Support SHT_RELR.

2 years agoPR28673, input file 'gcov' is the same as output file
Alan Modra [Wed, 8 Dec 2021 09:23:54 +0000 (19:53 +1030)]
PR28673, input file 'gcov' is the same as output file

PR 28673
* ldlang.c (open_output): Use local_sym_name when checking
output against input files rather than filename.

2 years agoFix bug in source.c change
Tom Tromey [Wed, 8 Dec 2021 04:51:03 +0000 (21:51 -0700)]
Fix bug in source.c change

My earlier change to source.c ("Remove an xfree from add_path")
introduced a regression.  This patch fixes the problem.

2 years agogdb: make struct linespect contain vectors, not pointers to vectors
Simon Marchi [Fri, 3 Dec 2021 21:35:38 +0000 (16:35 -0500)]
gdb: make struct linespect contain vectors, not pointers to vectors

struct linespec contains pointers to vectors, instead of containing
vectors directly.  This is probably historical, when linespec_parser
(which contains a struct linespec field) was not C++-ified yet.  But it
seems easy to change the pointers to vectors to just vectors today.
This simplifies the code, we don't need to manually allocate and delete
the vectors and there's no pointer that can be NULL.

As far as I understand, there was not meaningful distinction between a
NULL pointer to vector and an empty vector.  So all NULL checks are
changed for !empty checks.

Change-Id: Ie759707da14d9d984169b93233343a86e2de9ee6

2 years agoAutomatic date update in version.in
GDB Administrator [Wed, 8 Dec 2021 00:00:15 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agoRemove an xfree from add_path
Tom Tromey [Sun, 24 Oct 2021 21:58:24 +0000 (15:58 -0600)]
Remove an xfree from add_path

This removes a temporary \0 assignment and an xfree from add_path,
replacing it with a simpler use of std::string.

2 years agogdb/linespec.c: simplify condition
Simon Marchi [Fri, 3 Dec 2021 21:35:37 +0000 (16:35 -0500)]
gdb/linespec.c: simplify condition

We can remove the empty check: if the vector has size 1, it is obviously
not empty.  This code ended up like this because the empty check used to
be a NULL check.

Change-Id: I1571bd0228818ca93f6a6b444e9b010dc2da4c08

2 years agogdb: rename "maint agent" functions
Simon Marchi [Tue, 7 Dec 2021 20:56:25 +0000 (15:56 -0500)]
gdb: rename "maint agent" functions

Functions agent_eval_command and agent_command are used to implement
maintenance commands, rename them accordingly (with the maint_ prefix),
as well as the agent_command_1 helper function.

Change-Id: Iacf96d4a0a26298e8dd4648a0f38da649ea5ef61

2 years agogdb: make set_raw_breakpoint static
Simon Marchi [Tue, 7 Dec 2021 03:15:37 +0000 (22:15 -0500)]
gdb: make set_raw_breakpoint static

set_raw_breakpoint is only used in breakpoint.c, make it static.

Change-Id: I7fbeda067685309a30b88aceaf957eff7a28e310

2 years agoSupport AT_FXRNG and AT_KPRELOAD on FreeBSD.
John Baldwin [Tue, 7 Dec 2021 18:29:01 +0000 (10:29 -0800)]
Support AT_FXRNG and AT_KPRELOAD on FreeBSD.

FreeBSD's kernel has recently added two new ELF auxiliary vector
entries.  AT_FXRNG points to a root seed version for the kernel's
PRNG.  Userland can use this to reseed a userland PRNG after the
kernel's PRNG has reseeded.  AT_KPRELOAD is the base address of a
kernel-provided vDSO.

This change displays the proper name and description of these entries
in 'info auxv'.

include/ChangeLog:

* elf/common.h (AT_FREEBSD_FXRNG, AT_FREEBSD_KPRELOAD): Define.

2 years agoAvoid extra work in global_symbol_searcher::expand_symtabs
Tom Tromey [Mon, 6 Dec 2021 19:20:28 +0000 (12:20 -0700)]
Avoid extra work in global_symbol_searcher::expand_symtabs

I noticed that global_symbol_searcher::expand_symtabs always passes a
file matcher to expand_symtabs_matching.  However, if 'filenames' is
empty, then this always returns true.  It's slightly more efficient to
pass a null file matcher in this case, because that lets the "quick"
symbol implementations skip any filename checks.

Regression tested on x86-64 Fedora 34.

2 years ago[gdb/testsuite] Fix options arg handling in compile_jit_elf_main_as_so
Tom de Vries [Tue, 7 Dec 2021 08:44:36 +0000 (09:44 +0100)]
[gdb/testsuite] Fix options arg handling in compile_jit_elf_main_as_so

In commit 80ad340c902 ("[gdb/testsuite] use -Ttext-segment for jit-elf tests")
the following change was made:
...
 proc compile_jit_elf_main_as_so {main_solib_srcfile main_solib_binfile options} {
-    set options [concat $options debug]
+    global jit_load_address jit_load_increment
+
+    set options [list \
+       additional_flags="-DMAIN=jit_dl_main" \
+       additional_flags=-DLOAD_ADDRESS=$jit_load_address \
+       additional_flags=-DLOAD_INCREMENT=$jit_load_increment \
+       debug]
...

Before the change, the options argument was used, but after the change not
anymore.

Fix this by reverting back to using "set options [concat $options ...]".

Fixing this gets us twice the -DMAIN=jit_dl_main bit, once from a caller, and
once from compile_jit_elf_main_as_so.  Fix this by removing the bit from
compile_jit_elf_main_as_so, which makes the code similar to compile_jit_main.

Tested on x86_64-linux.

2 years ago[gdb/testsuite] Fix FAIL in gdb.tui/basic.exp
Tom de Vries [Tue, 7 Dec 2021 07:16:42 +0000 (08:16 +0100)]
[gdb/testsuite] Fix FAIL in gdb.tui/basic.exp

On openSUSE Leap 15.2 aarch64 I ran into:
...
FAIL: gdb.tui/basic.exp: check main is where we expect on the screen
...
while this is passing on x86_64.

On x86_64-linux we have at the initial screen dump for "list -q main":
...
 0 +-/home/vries/gdb_versions/devel/src/gdb/testsuite/gdb.tui/tui-layout.c--+
 1 |       15     You should have received a copy of the GNU General Public |
 2 |       16     along with this program.  If not, see <http://www.gnu.org/|
 3 |       17                                                               |
 4 |       18  int                                                          |
 5 |       19  main ()                                                      |
 6 |       20  {                                                            |
 7 |       21    return 0;                                                  |
 8 |       22  }                                                            |
 9 |       23                                                               |
...
but on aarch64:
...
 0 +-/home/tdevries/gdb/src/gdb/testsuite/gdb.tui/tui-layout.c--------------+
 1 |       16     along with this program.  If not, see <http://www.gnu.org/|
 2 |       17                                                               |
 3 |       18  int                                                          |
 4 |       19  main ()                                                      |
 5 |       20  {                                                            |
 6 |       21    return 0;                                                  |
 7 |       22  }                                                            |
 8 |       23                                                               |
 9 |       24                                                               |
...

The cause of the diffferent placement is that we have as line number for main
on x86_64:
...
$ gdb -q -batch outputs/gdb.tui/basic/basic -ex "info line main"
Line 20 of "tui-layout.c" starts at address 0x4004a7 <main> \
  and ends at 0x4004ab <main+4>.
...
and on aarch64 instead:
...
$ gdb -q -batch outputs/gdb.tui/basic/basic -ex "info line main"
Line 21 of "tui-layout.c" starts at address 0x4005f4 <main> \
  and ends at 0x4005f8 <main+4>.
...

Fix this by using a new source file main-one-line.c, that implements the
entire main function on a single line, in order to force the compiler to use
that line number.

Also try to do less hard-coding in the test-case.

Tested on x86_64-linux and aarch64-linux.

2 years ago[gdb/tdep] Fix inferior plt calls in PIE for i386
Tom de Vries [Tue, 7 Dec 2021 07:07:18 +0000 (08:07 +0100)]
[gdb/tdep] Fix inferior plt calls in PIE for i386

Consider test-case test.c:
...
int main (void) {
  void *p = malloc (10);
  return 0;
}
...

When compiled to a non-PIE exec:
...
$ gcc -m32 test.c
...
the call sequence looks like:
...
 8048447:       83 ec 0c                sub    $0xc,%esp
 804844a:       6a 0a                   push   $0xa
 804844c:       e8 bf fe ff ff          call   8048310 <malloc@plt>
...
which calls to:
...
08048310 <malloc@plt>:
 8048310:       ff 25 0c a0 04 08       jmp    *0x804a00c
 8048316:       68 00 00 00 00          push   $0x0
 804831b:       e9 e0 ff ff ff          jmp    8048300 <.plt>
...
where the first insn at 0x8048310 initially jumps to the following address
0x8048316, read from the .got.plt @ 0x804a00c:
...
 804a000 0c9f0408 00000000 00000000 16830408  ................
 804a010 26830408                             &...
...

Likewise, when compiled as a PIE:
...
$ gcc -m32 -fPIE -pie test.c
...
we have this call sequence (with %ebx setup to point to the .got.plt):
...
0000055d <main>:
 579:   83 ec 0c                sub    $0xc,%esp
 57c:   6a 0a                   push   $0xa
 57e:   89 c3                   mov    %eax,%ebx
 580:   e8 6b fe ff ff          call   3f0 <malloc@plt>
...
which calls to:
...
000003f0 <malloc@plt>:
 3f0:   ff a3 0c 00 00 00       jmp    *0xc(%ebx)
 3f6:   68 00 00 00 00          push   $0x0
 3fb:   e9 e0 ff ff ff          jmp    3e0 <.plt>
...
where the insn at 0x3f0 initially jumps to following address 0x3f6, read from
the .got.plt at offset 0xc:
...
 2000 f41e0000 00000000 00000000 f6030000  ................
 2010 06040000                             ....
...

When instead doing an inferior call to malloc (with nosharedlib to force
malloc to resolve to malloc@plt rather than the functions in ld.so or libc.so)
with the non-PIE exec, we have the expected:
...
$ gdb -q -batch a.out -ex start -ex nosharedlib -ex "p /x (void *)malloc (10)"
Temporary breakpoint 1 at 0x8048444

Temporary breakpoint 1, 0x08048444 in main ()
$1 = 0x804b160
...

But with the PIE exec, we run into:
...
$ gdb -q -batch a.out -ex start -ex nosharedlib -ex "p /x (void *)malloc (10)"
Temporary breakpoint 1 at 0x56c

Temporary breakpoint 1, 0x5655556c in main ()

Program received signal SIGSEGV, Segmentation fault.
0x565553f0 in malloc@plt ()
...

The segfault happens because:
- the inferior call mechanism doesn't setup %ebx
- %ebx instead is 0
- the jump to "*0xc(%ebx)" reads from memory at 0xc

Fix this by setting up %ebx properly in i386_thiscall_push_dummy_call.

Fixes this failure with target board unix/-m32/-pie/-fPIE reported in
PR28467:
...
FAIL: gdb.base/nodebug.exp: p/c (int) array_index("abcdef",2)
...

Tested on x86_64-linux, with target board unix/-m32 and unix/-m32/-fPIE/-pie.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28467

2 years ago[gdb/symtab] Support -readnow during reread
Tom de Vries [Tue, 7 Dec 2021 06:51:25 +0000 (07:51 +0100)]
[gdb/symtab] Support -readnow during reread

When running test-case gdb.base/cached-source-file.exp with target board
readnow, we run into:
...
FAIL: gdb.base/cached-source-file.exp: rerun program (the program exited)
...

The problem is that when rereading, the readnow is ignored.

Fix this by copying the readnow handling code from symbol_file_add_with_addrs
to reread_symbols.

Tested on x86_64-linux.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=26800

2 years ago[gdb/ada] Fix assert in ada_is_unconstrained_packed_array_type
Tom de Vries [Tue, 7 Dec 2021 06:35:10 +0000 (07:35 +0100)]
[gdb/ada] Fix assert in ada_is_unconstrained_packed_array_type

On openSUSE Leap 42.3, with system compiler gcc 4.8.5 I run into:
...
(gdb) print u_one_two_three^M
src/gdb/gdbtypes.h:1050: internal-error: field: \
 Assertion `idx >= 0 && idx < num_fields ()' failed.^M
...

We run into trouble while doing this in
ada_is_unconstrained_packed_array_type:
...
1953          return TYPE_FIELD_BITSIZE (type, 0) > 0;
...
which tries to get field 0 from a type without fields:
...
(gdb) p type->num_fields ()
$6 = 0
...
which is the case because the type is a typedef:
...
(gdb) p type->code ()
$7 = TYPE_CODE_TYPEDEF
...

Fix this by using the type referenced by the typedef instead.

Tested on x86_64-linux.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28323

2 years agoRe: Add support for AArch64 EFI (efi-*-aarch64)
Alan Modra [Tue, 7 Dec 2021 02:06:31 +0000 (12:36 +1030)]
Re: Add support for AArch64 EFI (efi-*-aarch64)

Commit b69c9d41e8 was broken in multiple ways regarding the realloc
of the target string, most notably in that "-little" wasn't actually
appended to the input_target or output_target.  This caused asan
errors and "FAIL: Check if efi app format is recognized".  I also
noticed that the input_target string wasn't being copied but rather
the output_target when dealing with the input target.  Fix that too.

PR 26206
* objcopy.c (convert_efi_target): Rewrite.  Allocate modified
target strings here..
(copy_main): ..rather than here.  Do handle input_target,
not output_target for input.

2 years agoError on ld output file name matching input file name
Alan Modra [Tue, 7 Dec 2021 01:56:55 +0000 (12:26 +1030)]
Error on ld output file name matching input file name

It's not foolproof, for example we don't catch output to a linker
script, to a library specified with -l, or to an element of a thin
archive.

* ldlang.c (open_output): Exit with error on output file matching
an input file.
* testsuite/ld-misc/just-symbols.exp: Adjust ld -r test to suit.

2 years agoAutomatic date update in version.in
GDB Administrator [Tue, 7 Dec 2021 00:00:10 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agogdb: Add PowerPC support to gdb.dwarf2/frame-inlined-in-outer-frame
Carl Love [Fri, 19 Nov 2021 18:33:51 +0000 (18:33 +0000)]
gdb: Add PowerPC support to gdb.dwarf2/frame-inlined-in-outer-frame

This patch adds an #elif defined for PowerPC to setup the exit_0 macro.
This patch addes the needed macro definitionald logic to handle both elfV1
and elfV2.

The patch has been successfully tested on both PowerPC BE, Powerpc LE and
X86_64 with no regressions.

2 years ago[gdb/testsuite] Use precise align in gdb.arch/i386-{avx,sse}.exp
Tom de Vries [Mon, 6 Dec 2021 15:01:47 +0000 (16:01 +0100)]
[gdb/testsuite] Use precise align in gdb.arch/i386-{avx,sse}.exp

Test-cases gdb.arch/i386-{avx,sse}.exp use assembly instructions that require
the memory operands to be aligned to a certain boundary, and the test-cases
use C11's _Alignas to make that happen.

The draw-back of using _Alignas is that while it does enforce a minimum
alignment, the actual alignment may be bigger, which makes the following
scenario possible:
- copy say, gdb.arch/i386-avx.c as basis for a new test-case
- run the test-case and observe a PASS
- commit the new test-case in the supposition that the test-case is correct
  and well-tested
- run later into a failure on a different test setup (which may be a setup
  where reproduction and investigation is more difficult and time-consuming),
  and find out that the specified alignment was incorrect and should have been
  updated to say, 64 bytes.  The initial PASS occurred only because the actual
  alignment happened to be greater than required.

The idea of having precise alignment as a means of having more predictable
execution which allows flushing out bugs earlier, has been filed as PR
gcc/103095.

Add a new file lib/precise-aligned-alloc.c with functions
precise_aligned_alloc and precise_aligned_dup, to support precise alignment.

Use precise_aligned_dup in aforementioned test-cases to:
- verify that the specified alignment is indeed sufficient, rather
  than too little but accidentally over-aligned.
- prevent the same type of problems in any new test-cases based on these

Tested on x86_64-linux, with both gcc and clang.

2 years ago[gdb/testsuite] Fix data alignment in gdb.arch/i386-{avx,sse}.exp
Tom de Vries [Mon, 6 Dec 2021 15:01:47 +0000 (16:01 +0100)]
[gdb/testsuite] Fix data alignment in gdb.arch/i386-{avx,sse}.exp

When running test-case gdb.arch/i386-avx.exp with clang I ran into:
...
(gdb) PASS: gdb.arch/i386-avx.exp: set first breakpoint in main
continue^M
Continuing.^M
^M
Program received signal SIGSEGV, Segmentation fault.^M
0x000000000040052b in main (argc=1, argv=0x7fffffffd3c8) at i386-avx.c:54^M
54        asm ("vmovaps 0(%0), %%ymm0\n\t"^M
(gdb) FAIL: gdb.arch/i386-avx.exp: continue to breakpoint: \
  continue to first breakpoint in main
...

The problem is that the vmovaps insn requires an 256-bit (or 32-byte) aligned
address, and it's only 16-byte aligned:
...
(gdb) p /x $rax
$1 = 0x601030
...

Fix this by using a sufficiently aligned address, using _Alignas.

Compile using -std=gnu11 to support _Alignas.

Likewise in gdb.arch/i386-sse.exp.

Tested on x86_64-linux, with both gcc and clang.

2 years ago[GOLD] PowerPC64 inline plt sequences
Alan Modra [Wed, 24 Nov 2021 13:30:50 +0000 (00:00 +1030)]
[GOLD] PowerPC64 inline plt sequences

The fixes gold failures to handle inline PLT sequences properly.
PowerPC gold was always turning these back into direct calls due to
gsym->use_plt_offset() returning false.  This is fixed for dynamic
linking by correcting get_reference_flags, and for static linking by
overriding use_plt_offset() in relocate().  The rest of the patch
revolves around needing to create PLT entries for inline PLT calls
when statically linking (for gcc -mlongcall).  The lplt section
handled that for local symbols, now it does globals too.

* powerpc.cc (Target_powerpc::plt_off): Return proper section
for static link.
(Target_powerpc::symval_for_branch): Make public.
(Target_powerpc::make_lplt_section): Add Symbol_table* param.
Adjust all calls.
(Target_powerpc::make_local_plt_entry): Likewise.
(Target_powerpc::make_local_plt_entry): New variant for global syms.
(Powerpc_relobj::do_relocate_sections): Don't write lplt contents.
(Output_data_plt_powerpc::do_write): Write lplt contents here.
(Output_data_plt_powerpc::Output_data_plt_powerpc): Save
symbol table pointer.  Adjust all uses.
(Output_data_plt_powerpc::add_entry): Add stash parameter.  Don't
do dynamic reloc handling when no reloc section.  Save symbol
for local plt entries.
(Output_data_plt_powerpc::add_local_entry): Save symbol.
(Output_data_plt_powerpc::Local_plt_ent): New class.
(Output_data_plt_powerpc::sym_ents_): New vector.
(Target_powerpc::Scan::get_reference_flags): Return
FUNCTION_CALL|RELATIVE_REF for inline plt relocs.
(Target_powerpc::Scan::global): Make entries in lplt for inline
plt call relocation symbols.
(Target_powerpc::Relocate::relocate): Rename has_plt_offset to
use_plt_offset.  Set use_plt_offset for inline plt relocs.

2 years agold: improve shared tests for AIX
Clément Chigot [Thu, 18 Nov 2021 15:12:14 +0000 (16:12 +0100)]
ld: improve shared tests for AIX

It's now possible to refer symbols in the main program from the
shared library. However, it still impossible to have the same
overriden features between shared objects and mains than ELF,
without using the runtime linking feature which isn't yet fully
available.

ld/ChangeLog:

* testsuite/ld-shared/shared.exp: Improve XCOFF support
* testsuite/ld-shared/main.c: Likewise.
* testsuite/ld-shared/sh1.c: Likewise.
* testsuite/ld-shared/xcoff.dat: Likewise.

2 years agoAutomatic date update in version.in
GDB Administrator [Mon, 6 Dec 2021 00:00:13 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agoPreserve artificial CU name in process_psymtab_comp_unit_reader
Tom Tromey [Sun, 5 Dec 2021 20:13:33 +0000 (13:13 -0700)]
Preserve artificial CU name in process_psymtab_comp_unit_reader

This fixes a use-after-free that Simon pointed out.
process_psymtab_comp_unit_reader was allocating an artificial name for
a CU, and then discarding it.  However, this name was preserved in the
cached file_and_directory.  This patch arranges for the allocated name
to be preserved there.

2 years agosim: include ansidecl.h when needed
Mike Frysinger [Sun, 5 Dec 2021 02:09:23 +0000 (21:09 -0500)]
sim: include ansidecl.h when needed

Avoid implicit include deps with this to help untangle sim headers
so we can get rid of arch-specific sim-main.h.

2 years agosim: include stdint.h when needed
Mike Frysinger [Sun, 5 Dec 2021 02:06:54 +0000 (21:06 -0500)]
sim: include stdint.h when needed

Avoid implicit include deps with this to help untangle sim headers
so we can get rid of arch-specific sim-main.h.

2 years agosim: include stdarg.h when used
Mike Frysinger [Sun, 5 Dec 2021 02:04:41 +0000 (21:04 -0500)]
sim: include stdarg.h when used

Avoid implicit include deps with this to help untangle sim headers
so we can get rid of arch-specific sim-main.h.

2 years agosim: reorder header includes
Mike Frysinger [Sun, 5 Dec 2021 01:24:55 +0000 (20:24 -0500)]
sim: reorder header includes

We're including system headers after local headers in a bunch of
places, but this leads to conflicts when our local headers happen
to define symbols that show up in the system headers.

Use the more standard order of:
* config.h (via defs.h)
* system headers
* local library headers (e.g. bfd & libiberty)
* sim specific headers

2 years agogdbsupport: fix memory leak in create_file_handler when re-using file handler
Simon Marchi [Thu, 2 Dec 2021 19:04:18 +0000 (14:04 -0500)]
gdbsupport: fix memory leak in create_file_handler when re-using file handler

ASan made me notice a memory leak, where the memory tied to the file
handle name string wasn't freed.  When register a file handler with an
fd that is already registered, we re-use the file_handler object, so we
ended up creating a new std::string object and overwriting the
file_handler::name pointer, without free-ing the old std::string.

Fix this by allocating file_handler with new, deleting it with
delete, and making file_handler::name not a pointer.

Change-Id: Ie304cc78ab5ae5dfad9a1366e9890c09de651f43

2 years agoAutomatic date update in version.in
GDB Administrator [Sun, 5 Dec 2021 00:00:14 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agosim: moxie: hoist dtb rules up to common builds
Mike Frysinger [Sat, 4 Dec 2021 18:35:25 +0000 (13:35 -0500)]
sim: moxie: hoist dtb rules up to common builds

These rules don't depend on the target compiler settings, so hoist
the build logic up to the common builds for better parallelization.

2 years agosim: m68hc11: delete unused profile flags
Mike Frysinger [Sat, 4 Dec 2021 18:27:23 +0000 (13:27 -0500)]
sim: m68hc11: delete unused profile flags

These were moved to the common configure script a while ago and have
the same default as these, so just delete it.

2 years agosim: msp430: delete redundant comments & settings
Mike Frysinger [Sat, 4 Dec 2021 18:26:24 +0000 (13:26 -0500)]
sim: msp430: delete redundant comments & settings

These were copied from the example docs, so aren't adding any value.

2 years agosim: erc32: drop old configure target
Mike Frysinger [Sat, 4 Dec 2021 18:25:00 +0000 (13:25 -0500)]
sim: erc32: drop old configure target

There is no configure script in here anymore to regenerate.

2 years agosim: m32c/rl78: drop redundant -Wall settings
Mike Frysinger [Sat, 4 Dec 2021 18:23:54 +0000 (13:23 -0500)]
sim: m32c/rl78: drop redundant -Wall settings

We already turn these on in the configure script.

2 years agoCache the result of find_file_and_directory
Tom Tromey [Mon, 29 Nov 2021 18:34:17 +0000 (11:34 -0700)]
Cache the result of find_file_and_directory

This changes the DWARF reader to cache the result of
find_file_and_directory.  This is not especially important now, but it
will help the new DWARF indexer.

2 years agoMove file_and_directory to new file and C++-ize
Tom Tromey [Sun, 28 Nov 2021 17:48:53 +0000 (10:48 -0700)]
Move file_and_directory to new file and C++-ize

This moves file_and_directory to a new file, and then C++-izes it --
replacing direct assignments with methods, and arranging for it to own
any string that must be computed.  Finally, the CU's objfile will only
be used on demand; this is an important property for the new DWARF
indexer's parallel mode.

2 years agoRemove Irix case from find_file_and_directory
Tom Tromey [Mon, 29 Nov 2021 20:50:24 +0000 (13:50 -0700)]
Remove Irix case from find_file_and_directory

find_file_and_directory has a special case for the Irix 6.2 compiler.
Since this is long obsolete, this patch removes it.

2 years agosim: frv: split up testsuite a bit
Mike Frysinger [Sat, 27 Nov 2021 07:01:03 +0000 (02:01 -0500)]
sim: frv: split up testsuite a bit

Running frv's allinsn in serial is quite slow due to the sheer number
of tests it contains.  By splitting it up and running in parallel, the
execution time on my system goes from ~100sec to ~60sec.

2 years agogdb: don't show deprecated aliases
Simon Marchi [Mon, 1 Nov 2021 04:30:25 +0000 (00:30 -0400)]
gdb: don't show deprecated aliases

I don't think it's very useful to show deprecated aliases to the
user.  It encourages the user to use them, when the goal is the
opposite.

For example, before:

    (gdb) help set index-cache enabled
    set index-cache enabled, set index-cache off, set index-cache on
      alias set index-cache off = set index-cache enabled off
      alias set index-cache on = set index-cache enabled on
    Enable the index cache.
    When on, enable the use of the index cache.

    (gdb) help set index-cache on
    Warning: 'set index-cache on', an alias for the command 'set index-cache enabled', is deprecated.
    Use 'set index-cache enabled on'.

    set index-cache enabled, set index-cache off, set index-cache on
      alias set index-cache off = set index-cache enabled off
      alias set index-cache on = set index-cache enabled on
    Enable the index cache.
    When on, enable the use of the index cache.

After:

    (gdb) help set index-cache enabled
    Enable the index cache.
    When on, enable the use of the index cache.
    (gdb) help set index-cache on
    Warning: 'set index-cache on', an alias for the command 'set index-cache enabled', is deprecated.
    Use 'set index-cache enabled on'.

    Enable the index cache.
    When on, enable the use of the index cache.

Change-Id: I989b618a5ad96ba975367e9d16db95523cd57a4c

2 years agogdb/testsuite: fix two "maint info line-table"-related tests
Simon Marchi [Sat, 4 Dec 2021 02:01:42 +0000 (21:01 -0500)]
gdb/testsuite: fix two "maint info line-table"-related tests

Commit 92228a334ba2 ("gdb: small "maintenance info line-table"
readability improvements") change the output format of "maint info
line-table" slightly, adding some empty lines between each
line-table.  This causes two tests to start failing, update them to
account for those empty lines.

Change-Id: I9d33a58fce3e860ba0554b25f5582e8066a5c519

2 years agogdb: revert one array_view copy change in ada-lang.c
Simon Marchi [Sat, 4 Dec 2021 01:39:57 +0000 (20:39 -0500)]
gdb: revert one array_view copy change in ada-lang.c

Commit 4bce7cdaf481 ("gdbsupport: add array_view copy function") caused
an internal error when running gdb.ada/packed_array_assign.exp:

    print pra(1) := pr^M
    /home/smarchi/src/binutils-gdb/gdb/../gdbsupport/array-view.h:217: internal-error: copy: Assertion `dest.size () == src.size ()' failed.^M

I am not sure what's the root cause of this, whether it is a GDB bug
exposed by using the array_view copy function or not.  Back out the
change that triggers the internal error for now, while we investigate
it.

Change-Id: I055ab14143e4cfd3ca7ce8f4855c6c3c05db52a7

2 years agobfd: unify header generation rules
Mike Frysinger [Fri, 3 Dec 2021 05:31:02 +0000 (00:31 -0500)]
bfd: unify header generation rules

The logic between these rules are extremely similar, so unify them
into a single variable.

2 years agobfd: move header updates up a directory
Mike Frysinger [Fri, 3 Dec 2021 05:23:20 +0000 (00:23 -0500)]
bfd: move header updates up a directory

The rules for rebuilding the bfd headers live in the doc/ subdir
(most likely) because they rely on the chew & related tools.  But
we can collapse them into the main Makefile while keeping the tools
in the doc subdir easily enough.  This makes the code simpler and
allows for rebuilding them in parallel.

Also add automake silent rule support while we're here.

2 years agobfd: convert bfdver.h to silent automake rules
Mike Frysinger [Fri, 3 Dec 2021 05:42:47 +0000 (00:42 -0500)]
bfd: convert bfdver.h to silent automake rules

2 years agoAutomatic date update in version.in
GDB Administrator [Sat, 4 Dec 2021 00:00:15 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agogdb: small "maintenance info line-table" readability improvements
Simon Marchi [Thu, 2 Dec 2021 01:40:13 +0000 (20:40 -0500)]
gdb: small "maintenance info line-table" readability improvements

 - separate each entry with a newline, to visually separate them
 - style filenames with the filename style
 - print the name of the compunit_symtab

A header now looks like this, with the compunit_symtab name added (and
the coloring, but you can't really see it here):

    objfile: /home/simark/build/babeltrace/src/cli/.libs/babeltrace2 ((struct objfile *) 0x613000005980)
    compunit_symtab: babeltrace2-cfg-cli-args.c ((struct compunit_symtab *) 0x62100da1ed10)
    symtab: /usr/include/glib-2.0/glib/gdatetime.h ((struct symtab *) 0x62100d9ee530)
    linetable: ((struct linetable *) 0x0):

Change-Id: Idc23e10aaa66e2e692adb0a6a74144f72c4fa1c7

2 years agogdb: change some alias functions parameters to const-reference
Simon Marchi [Wed, 1 Dec 2021 16:41:32 +0000 (11:41 -0500)]
gdb: change some alias functions parameters to const-reference

Now that we use intrusive list to link aliases, it becomes easier to
pass cmd_list_element arguments by const-reference rather than by
pointer to some functions, change a few.

Change-Id: Id0df648ed26e9447da0671fc2c858981cda31df8

2 years agogdb: use intrusive_list for cmd_list_element aliases list
Simon Marchi [Wed, 1 Dec 2021 16:41:31 +0000 (11:41 -0500)]
gdb: use intrusive_list for cmd_list_element aliases list

Change the manually-implemented linked list to use intrusive_list.  This
is not strictly necessary, but it makes the code much simpler.

Change-Id: Idd08090ebf2db8bdcf68e85ef72a9635f1584ccc

2 years agogdb: trivial changes to use array_view
Simon Marchi [Tue, 26 Oct 2021 03:50:22 +0000 (23:50 -0400)]
gdb: trivial changes to use array_view

Change a few relatively obvious spots using value contents to propagate
the use array_view a bit more.

Change-Id: I5338a60986f06d5969fec803d04f8423c9288a15

2 years agogdb: make extract_integer take an array_view
Simon Marchi [Tue, 26 Oct 2021 03:29:34 +0000 (23:29 -0400)]
gdb: make extract_integer take an array_view

I think it would make sense for extract_integer, extract_signed_integer
and extract_unsigned_integer to take an array_view.  This way, when we
extract an integer, we can validate that we don't overflow the buffer
passed by the caller (e.g. ask to extract a 4-byte integer but pass a
2-byte buffer).

 - Change extract_integer to take an array_view
 - Add overloads of extract_signed_integer and extract_unsigned_integer
   that take array_views.  Keep the existing versions so we don't
   need to change all callers, but make them call the array_view
   versions.

This shortens some places like:

  result = extract_unsigned_integer (value_contents (result_val).data (),
     TYPE_LENGTH (value_type (result_val)),
     byte_order);

into

  result = extract_unsigned_integer (value_contents (result_val), byte_order);

value_contents returns an array view that is of length
`TYPE_LENGTH (value_type (result_val))` already, so the length is
implicitly communicated through the array view.

Change-Id: Ic1c1f98c88d5c17a8486393af316f982604d6c95

2 years agogdbsupport: add array_view copy function
Simon Marchi [Mon, 8 Nov 2021 21:06:07 +0000 (16:06 -0500)]
gdbsupport: add array_view copy function

An assertion was recently added to array_view::operator[] to ensure we
don't do out of bounds accesses.  However, when the array_view is copied
to or from using memcpy, it bypasses that safety.

To address this, add a `copy` free function that copies data from an
array view to another, ensuring that the destination and source array
views have the same size.  When copying to or from parts of an
array_view, we are expected to use gdb::array_view::slice, which does
its own bounds check.  With all that, any copy operation that goes out
of bounds should be caught by an assertion at runtime.

copy is implemented using std::copy and std::copy_backward, which, at
least on libstdc++, appears to pick memmove when copying trivial data.
So in the end there shouldn't be much difference vs using a bare memcpy,
as we do right now.  When copying non-trivial data, std::copy and
std::copy_backward assigns each element in a loop.

To properly support overlapping ranges, we must use std::copy or
std::copy_backward, depending on whether the destination is before the
source or vice-versa.  std::copy and std::copy_backward don't support
copying exactly overlapping ranges (where the source range is equal to
the destination range).  But in this case, no copy is needed anyway, so
we do nothing.

The order of parameters of the new copy function is based on std::copy
and std::copy_backward, where the source comes before the destination.

Change a few randomly selected spots to use the new function, to show
how it can be used.

Add a test for the new function, testing both with arrays of a trivial
type (int) and of a non-trivial type (foo).  Test non-overlapping
ranges as well as three kinds of overlapping ranges: source before dest,
dest before source, and dest == source.

Change-Id: Ibeaca04e0028410fd44ce82f72e60058d6230a03

2 years agogdb: change store_waitstatus to return a target_waitstatus by value
Simon Marchi [Wed, 1 Dec 2021 18:09:56 +0000 (13:09 -0500)]
gdb: change store_waitstatus to return a target_waitstatus by value

store_waitstatus is basically a translation function between a status
integer and an equivalent target_waitstatus object.  It would make sense
for it to take the integer as a parameter and return the
target_waitstatus by value.  Do that, and rename to
host_status_to_waitstatus.  Users can then do:

  ws = host_status_to_waitstatus (status)

which does the right thing, given the move constructor of
target_waitstatus.

Change-Id: I7a07d59d3dc19d3ed66929642f82f44f3e85d61b

2 years agogdb: return *this in target_waitstatus setters
Simon Marchi [Thu, 2 Dec 2021 01:40:01 +0000 (20:40 -0500)]
gdb: return *this in target_waitstatus setters

While playing with some code creating target_waitstatus objects, I was
mildly annoyed by the fact that we can't just return a new
target_waitstatus object.  We have to do:

    target_waitstatus ws;
    ws.set_exited (123);
    return ws;

Make the setters return the "this" object as a reference, such that it's
possible to do:

    return target_waitstatus ().set_exited (123);

I initially thought of adding static creation functions, which you would
use like:

    return target_waitstatus::make_exited (123);

However, making the setters return a reference to the object achieves
pretty much the same thing, with less new code.

Change-Id: I45159b7f9fcd9db5b20603480e323020b14ed147

2 years agogdb: make saved_filename an std::string
Simon Marchi [Thu, 2 Dec 2021 19:53:06 +0000 (14:53 -0500)]
gdb: make saved_filename an std::string

Make this variable an std::string, avoiding manual memory management.

Change-Id: Ie7a8d7381449ab9c4dfc4cb8b99e63b9ffa8f947

2 years agoaarch64: Fix uninitialised memory
Richard Sandiford [Fri, 3 Dec 2021 11:57:17 +0000 (11:57 +0000)]
aarch64: Fix uninitialised memory

AARCH64_OPDE_EXPECTED_A_AFTER_B and AARCH64_OPDE_A_SHOULD_FOLLOW_B
are not paired with an error string, but we had an assert that the
error was nonnull.  Previously this assert was testing uninitialised
memory and so could pass or fail arbitrarily.

opcodes/
* aarch64-opc.c (verify_mops_pme_sequence): Initialize the error
field to null for AARCH64_OPDE_EXPECTED_A_AFTER_B and
AARCH64_OPDE_A_SHOULD_FOLLOW_B.
* aarch64-dis.c (print_verifier_notes): Move assert.

2 years agogdb: make value_subscripted_rvalue static
Andrew Burgess [Fri, 15 Oct 2021 14:14:58 +0000 (15:14 +0100)]
gdb: make value_subscripted_rvalue static

The function value_subscripted_rvalue is only used in valarith.c, so
lets make it a static function.

There should be no user visible change after this commit.

2 years agogdb/testsuite: give a test a real name
Andrew Burgess [Fri, 3 Dec 2021 09:57:16 +0000 (09:57 +0000)]
gdb/testsuite: give a test a real name

A test in gdb.python/py-send-packet.exp added in this commit:

  commit 24b2de7b776f8f23788d855b1eec290c6e208821
  Date:   Tue Aug 31 14:04:36 2021 +0100

      gdb/python: add gdb.RemoteTargetConnection.send_packet

included a large amount of binary data in the command sent to GDB.  As
this test didn't have a real test name the binary data was included in
the gdb.sum file.  The contents of the binary data could change
between different runs of GDB, and this makes comparing results
harder.

This commit gives the test a real test name.

2 years agogdb/remote: fix use after free bug
Andrew Burgess [Thu, 2 Dec 2021 11:05:17 +0000 (11:05 +0000)]
gdb/remote: fix use after free bug

This commit:

  commit 288712bbaca36bff6578bc839ebcdc3707662f81
  Date:   Mon Nov 22 15:16:27 2021 +0000

      gdb/remote: use scoped_restore to control starting_up flag

introduced a use after free bug.  The scoped restore added in the
above commit resets a flag within a remote_target's remote_state
object.

However, in some situations, the remote_target can be unpushed before
the error is thrown.  If the only reference to the target is the one
in the target stack, then unpushing the target will cause the
remote_target to be deleted, which, in turn, will delete the
remote_state object.  The scoped restore will then try to reset the
flag within a deleted object.

This problem was caught in the gdb.server/server-connect.exp test,
which, when run with the address sanitizer enabled, highlights the
write after free bug described above.

This commit resolves this issue by adding a new class specifically for
the purpose of managing the starting_up flag.  As well as setting, and
then clearing the starting_up flag, this new class increments, and
then decrements the reference count on the remote_target object.  This
prevents the remote_target from being deleted until after the flag has
been reset.

The gdb.server/server-connect.exp now runs cleanly with the address
sanitizer enabled.

2 years agolibctf: workaround automake bug with conditional info pages
Mike Frysinger [Fri, 3 Dec 2021 04:13:15 +0000 (23:13 -0500)]
libctf: workaround automake bug with conditional info pages

It looks like automake makes assumptions about its ability to build info
pages based on the GNU standard behavior of shipping info pages with the
distributions.  So even though the info pages were conditionalized, and
automake disabled some of the targets, it was still creeping in by way
of unconditional INFO_DEPS settings.

We can workaround this by adding a stub target for the info page when
building info pages are disabled.  This tricks automake into disabling
its own extended generation target.  I'll follow up with the automake
folks to see what they think.

2 years agoAdd myself and Zhensong Liu as the LoongArch port maintainer.
Chenghua Xu [Fri, 3 Dec 2021 02:09:30 +0000 (10:09 +0800)]
Add myself and Zhensong Liu as the LoongArch port maintainer.

2 years agoRevert "Re: Don't compile some opcodes files when bfd is 32-bit only"
Alan Modra [Thu, 2 Dec 2021 09:02:57 +0000 (19:32 +1030)]
Revert "Re: Don't compile some opcodes files when bfd is 32-bit only"

This reverts commit 7a53275579e7cec9389ccb924f5ecf69e8d89d41.
The bpf sim doesn't work with a 32-bit bfd after all.

2 years agoAutomatic date update in version.in
GDB Administrator [Fri, 3 Dec 2021 00:00:15 +0000 (00:00 +0000)]
Automatic date update in version.in

2 years agogdb: remove unexpected xstrdup in _initialize_maint_test_settings
Simon Marchi [Thu, 2 Dec 2021 19:12:44 +0000 (14:12 -0500)]
gdb: remove unexpected xstrdup in _initialize_maint_test_settings

That xstrdup is not correct, since we are assigning an std::string.  The
result of xstrdup is used to initialize the string, and then lost
forever.  Remove it.

Change-Id: Ief7771055e4bfd643ef3b285ec9fb7b1bfd14335

2 years agoFix illegal memory access whilst parsing corrupt DWARF debug information.
Nick Clifton [Thu, 2 Dec 2021 17:48:20 +0000 (17:48 +0000)]
Fix illegal memory access whilst parsing corrupt DWARF debug information.

PR 28645
* dwarf.c (process_cu_tu_index): Add test for overruning section
whilst processing slots.

2 years ago[gdb/tdep] Fix avx512 -m32 support in gdbserver
Tom de Vries [Thu, 2 Dec 2021 17:20:13 +0000 (18:20 +0100)]
[gdb/tdep] Fix avx512 -m32 support in gdbserver

PR27257 reports a problem that can be reproduced as follows:
- use x86_64 machine with avx512 support
- compile a hello world with -m32 to a.out
- start a gdbserver session with a.out
- use gdb to connect to the gdbserver session

This makes us run into:
...
Listening on port 2346
Remote debugging from host ::1, port 34940
src/gdbserver/regcache.cc:257: \
  A problem internal to GDBserver has been detected.
Unknown register zmm16h requested
...

The problem is that i387_xsave_to_cache in gdbserver/i387-fp.cc can't find a
register zmm16h in the register cache.

To understand how this happens, first some background.

SSE has 16 128-bit wide xmm registers.

AVX extends the SSE registers set as follows:
- it extends the 16 existing 128-bit wide xmm registers to 256-bit wide ymm
  registers.

AVX512 extends the AVX register set as follows:
- it extends the 16 existing 256-bit wide ymm registers to 512-bit wide zmm
  registers.
- it adds 16 additional 512-bit wide zmm registers (with corresponding ymm and
  xmm subregisters added as well)

However, in 32-bit mode, there are only 8 xmm/ymm/zmm registers.

The problem we're running into is that gdbserver/i387-fp.cc uses these
constants to describe the size of the register file:
...
static const int num_avx512_zmmh_low_registers = 16;
static const int num_avx512_zmmh_high_registers = 16;
static const int num_avx512_ymmh_registers = 16;
static const int num_avx512_xmm_registers = 16;
...
which are all incorrect for the 32-bit case.

Fix this by replacing the constants with variables that have the appropriate
values in 64-bit and 32-bit mode.

Tested on x86_64-linux with native and unix/-m32.

2 years agogdb/testsuite: update tests looking for "DWARF 2" debug format
Simon Marchi [Thu, 2 Dec 2021 15:28:40 +0000 (10:28 -0500)]
gdb/testsuite: update tests looking for "DWARF 2" debug format

Commit ab557072b8ec ("gdb: use actual DWARF version in compunit's
debugformat field") changes the debug format string in "info source" to
show the actual DWARF version, rather than always show "DWARF 2".

However, it failed to consider that some tests checked for the "DWARF 2"
string to see if the test program is compiled with DWARF debug
information.  Since everything is compiled with DWARF 4 or 5 nowadays,
that changed the behavior of those tests.  Notably, it prevent the
tests using skip_inline_var_tests to run.

Grep through the testsuite for "DWARF 2" and change all occurrences I
could find to use "DWARF [0-9]" instead (that string is passed to TCL's
string match).

Change-Id: Ic7fb0217fb9623880c6f155da6becba0f567a885

2 years ago(PPC64) fix handling of fixed-point values when using "return" command
Joel Brobecker [Thu, 21 Oct 2021 19:22:40 +0000 (13:22 -0600)]
(PPC64) fix handling of fixed-point values when using "return" command

In the gdb.ada/fixed_points_function.exp testcase, we have the following
Ada code...

   type FP1_Type is delta 0.1 range -1.0 .. +1.0; --  Ordinary
   function Call_FP1 (F : FP1_Type) return FP1_Type is
   begin
      FP1_Arg := F;
      return FP1_Arg;
   end Call_FP1;

... used as follow:

   F1 : FP1_Type := 1.0;
   F1 := Call_FP1 (F1);

The testcase, among other things, verifies that "return" works
properly as follow:

    | (gdb) return 1.0
    | Make pck.call_fp1 return now? (y or n) y
    | [...]
    | 9          F1 := Call_FP1 (F1);
    | (gdb) next
    | (gdb) print f1
    | $1 = 0.0625

The output of the last command shows that we returned the wrong
value. The value printed gives a clue about the problem, since
it is 1/16th of the value we expected, where 1/16 is FP1_Type's
scaling factor.

The problem, here, comes from the fact that the function
handling return values for base types (ppc64_sysv_abi_return_value_base)
writes the return value using unpack_long which, upon seeing that
the value being unpacked is a fixed point type, applies the scaling
factor, to get the integer-representation of our fixed-point value
(similar to what it does with floats, for instance).

So, the fix consists in teaching ppc64_sysv_abi_return_value_base
about fixed-point types, and to avoid the unwanted application
of the scaling factor.

Note that the "finish" function, on the other hand, does not
suffer from this issue, simply becaue the value returned by
the function is read from register without the use of a type,
thus avoiding an unwanted application of a scaling factor.

No test added, as this change is already tested by
gdb.ada/fixed_points_function.exp.

Co-Authored-By: Tristan Gingold <gingold@adacore.com>