| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
For the same reason explained in the previous patch (which was for the
record-btrace target), move clearing of the async event handler of the
record-full target to the wait method.
I'm not sure if/where that target needs to re-set its async event
handler in the wait method. Since it only supports a single thread,
there probably can't be multiple events to report at the same time.
gdb/ChangeLog:
* record-full.c (record_full_async_inferior_event_handler):
Don't clear async event handler.
(record_full_base_target::wait): Clear async event handler at
beginning.
Change-Id: I146fbdb53d99e3a32766ac7cd337ac5ed7fd9adf
|
|
For the same reason explained in the previous patch (which was for the
remote target), move clearing of the async event handler of the
record-btrace target to the wait method.
The record-btrace target already re-sets its async event handler in its
wait method, so that part doesn't need to be changed:
/* In async mode, we need to announce further events. */
if (target_is_async_p ())
record_btrace_maybe_mark_async_event (moving, no_history);
gdb/ChangeLog:
* record-btrace.c (record_btrace_handle_async_inferior_event):
Don't clear async event handler.
(record_btrace_target::wait): Clear async event handler at
beginning.
Change-Id: Ib32087a81bf94f1b884a938c8167ac8bbe09e362
|
|
The remote target's remote_async_inferior_event_token is a flag that
tells when it wants the infrun loop to call its wait method. The flag
is cleared in the async_event_handler's callback
(remote_async_inferior_event_handler), just before calling
inferior_event_handler. However, since inferior_event_handler may
actually call another target's wait method, there needs to be code that
checks if we need to re-raise the flag.
It would be simpler instead for remote_target::wait to clear the flag
when it returns an event and there are no more to report after that. If
another target's wait method gets called by inferior_event_handler, the
remote target's flag will stay naturally stay marked.
Note that this is already partially implemented in remote_target::wait,
since the remote target may have multiple events to report (and it can
only report one at the time):
if (target_is_async_p ())
{
remote_state *rs = get_remote_state ();
/* If there are are events left in the queue tell the event loop
to return here. */
if (!rs->stop_reply_queue.empty ())
mark_async_event_handler (rs->remote_async_inferior_event_token);
}
The code in remote_async_inferior_event_handler also checks for pending
events as well, in addition to the stop reply queue, so I've made
remote_target::wait check for that as well. I'm not completely sure
this is ok, since I don't understand very well how the pending events
mechanism works. But I figured it was safer to do this, worst case it
just leads to unnecessary calls to remote_target::wait.
gdb/ChangeLog:
* remote.c (remote_target::wait): Clear async event handler at
beginning, mark if needed at the end.
(remote_async_inferior_event_handler): Don't set or clear async
event handler.
Change-Id: I20117f5b5acc8a9972c90f16280249b766c1bf37
|
|
The `ready` flag of async event handlers is cleared by the async event
handler system right before invoking the associated callback, in
check_async_event_handlers.
This is not ideal with how the infrun subsystem consumes events: all
targets' async event handler callbacks essentially just invoke
`inferior_event_handler`, which eventually calls `fetch_inferior_event`
and `do_target_wait`. `do_target_wait` picks an inferior at random,
and thus a target at random (it could be the target whose `ready` flag
was cleared, or not), and pulls one event from it.
So it's possible that:
- the async event handler for a target A is called
- we end up consuming an event for target B
- all threads of target B are stopped, target_async(0) is called on it,
so its async event handler is cleared (e.g.
record_btrace_target::async)
As a result, target A still has events to report while its async event
handler is left unmarked, so these events are not consumed. To counter
this, at the end of their async event handler callbacks, targets check
if they still have something to report and re-mark their async event
handler (e.g. remote_async_inferior_event_handler).
The linux_nat target does not suffer from this because it doesn't use an
async event handler at the moment. It only uses a pipe registered with
the event loop. It is written to in the SIGCHLD handler (and in other
spots that want to get target wait method called) and read from in
the target's wait method. So if linux_nat happened to be target A in
the example above, the pipe would just stay readable, and the event loop
would wake up again, until linux_nat's wait method is finally called and
consumes the contents of the pipe.
I think it would be nicer if targets using async_event_handler worked in
a similar way, where the flag would stay set until the target's wait
method is actually called. As a first step towards that, this patch
moves the responsibility of clearing the ready flags of async event
handlers to the invoked callback.
All async event handler callbacks are modified to clear their ready flag
before doing anything else. So in practice, nothing changes with this
patch. It's only the responsibility of clearing the flag that is
shifted toward the callee.
gdb/ChangeLog:
* async-event.h (async_event_handler_func): Add documentation.
* async-event.c (check_async_event_handlers): Don't clear
async_event_handler ready flag.
* infrun.c (infrun_async_inferior_event_handler): Clear ready
flag.
* record-btrace.c (record_btrace_handle_async_inferior_event):
Likewise.
* record-full.c (record_full_async_inferior_event_handler):
Likewise.
* remote-notif.c (remote_async_get_pending_events_handler):
Likewise.
* remote.c (remote_async_inferior_event_handler): Likewise.
Change-Id: I179ef8e99580eae642d332846fd13664dbddc0c1
|
|
Moving it to an inner scope makes it clearer where it's used (only while
handling the TARGET_WAITKIND_LOADED event).
gdb/ChangeLog:
* infrun.c (handle_inferior_event): Move stop_soon variable to
inner scope.
Change-Id: Ic57685a21714cfbb38f1487ee96cea1d12b44652
|
|
This adds a testcase that exercises detaching while GDB is stepping
over a breakpoint, in all combinations of:
- maint target non-stop off/on
- set non-stop on/off
- displaced stepping on/off
This exercises the bugs fixed in the previous 8 patches.
gdb/testsuite/ChangeLog:
* gdb.threads/detach-step-over.c: New file.
* gdb.threads/detach-step-over.exp: New file.
|
|
A following patch will add a testcase that has a number of threads
constantly stepping over a breakpoint, and then has GDB detach the
process, while threads are running. If we have more than one inferior
running, and we detach from just one of the inferiors, we expect that
the remaining inferior continues running. However, in all-stop, if
GDB needs to pause the target for the detach, nothing is re-resuming
the other inferiors after the detach. "info threads" shows the
threads as running, but they really aren't. This fixes it.
gdb/ChangeLog:
* infcmd.c (detach_command): Hold strong reference to target, and
if all-stop on entry, restart threads on exit.
* infrun.c (switch_back_to_stepped_thread): Factor out bits to ...
(restart_stepped_thread): ... this new function. Also handle
trap_expected.
(restart_after_all_stop_detach): New function.
* infrun.h (restart_after_all_stop_detach): Declare.
|
|
A following patch will add a testcase that has a number of threads
constantly stepping over a breakpoint, and then has GDB detach the
process. That testcase exercises both "set displaced-stepping
on/off". Testing with "set displaced-stepping off" reveals that GDB
does not handle the case of the user typing "detach" just while some
thread is in the middle of an in-line step over. If that thread
belongs to the inferior that is being detached, then the step-over
never finishes, and threads of other inferiors are never re-resumed.
This fixes it.
gdb/ChangeLog:
* infrun.c (struct step_over_info): Initialize fields.
(prepare_for_detach): Handle ongoing in-line step over.
|
|
A following patch will add a testcase that has a number of threads
constantly stepping over a breakpoint, and then has GDB detach the
process. That testcase sometimes fails with the inferior crashing
with SIGTRAP after the detach because of the bug fixed by this patch,
when tested with the native target.
The problem is that target_detach removes breakpoints from the target
immediately, and that does not work with the native GNU/Linux target
(and probably no other native target) currently. The test wouldn't
fail with this issue when testing against gdbserver, because gdbserver
does allow accessing memory while the current thread is running, by
transparently pausing all threads temporarily, without GDB noticing.
Implementing that in gdbserver was a lot of work, so I'm not looking
forward right now to do the same in the native target. Instead, I
came up with a simpler solution -- push the breakpoints removal down
to the targets. The Linux target conveniently already pauses all
threads before detaching them, since PTRACE_DETACH only works with
stopped threads, so we move removing breakpoints to after that. Only
the remote and GNU/Linux targets support support async execution, so
no other target should really need this.
gdb/ChangeLog:
* linux-nat.c (linux_nat_target::detach): Remove breakpoints
here...
* remote.c (remote_target::remote_detach_1): ... and here ...
* target.c (target_detach): ... instead of here.
* target.h (target_ops::detach): Add comment.
|
|
I noticed that "detach" while a program was running sometimes resulted
in the process crashing. I tracked it down to this change to
prepare_for_detach in commit 187b041e ("gdb: move displaced stepping
logic to gdbarch, allow starting concurrent displaced steps"):
/* Is any thread of this process displaced stepping? If not,
there's nothing else to do. */
- if (displaced->step_thread == nullptr)
+ if (displaced_step_in_progress (inf))
return;
The problem above is that the condition was inadvertently flipped. It
should have been:
if (!displaced_step_in_progress (inf))
So I fixed it, and wrote a testcase to exercise it. The testcase has
a number of threads constantly stepping over a breakpoint, and then
GDB detaches the process, while threads are running and stepping over
the breakpoint. And then I was surprised that my testcase would hang
-- GDB would get stuck in an infinite loop in prepare_for_detach,
here:
while (displaced_step_in_progress (inf))
{
...
What is going on is that since we now have two displaced stepping
buffers, as one displaced step finishes, GDB starts another, and
there's another one already in progress, and on and on, so the
displaced_step_in_progress condition never turns false. This happens
because we go via the whole handle_inferior_event, which tries to
start new step overs when one finishes. And also because while we
remove breakpoints from the target before prepare_for_detach is
called, handle_inferior_event ends up calling insert_breakpoints via
e.g. keep_going.
Thinking through all this, I came to the conclusion that going through
the whole handle_inferior_event isn't ideal. A _lot_ is done by that
function, e.g., some thread may get a signal which is passed to the
inferior, and gdb decides to try to get over the signal handler, which
reinstalls breakpoints. Or some process may exit. We can end up
reporting these events via normal_stop while detaching, maybe end up
running some breakpoint commands, or maybe even something runs an
inferior function call. Etc. All this after the user has already
declared they don't want to debug the process anymore, by asking to
detach.
I came to the conclusion that it's better to do the minimal amount of
work possible, in a more controlled fashion, without going through
handle_inferior_event. So in the new approach implemented by this
patch, if there are threads of the inferior that we're detaching in
the middle of a displaced step, stop them, and cancel the displaced
step. This is basically what stop_all_threads already does, via
wait_one and (the now factored out) handle_one, so I'm reusing those.
gdb/ChangeLog:
* infrun.c (struct wait_one_event): Move higher up.
(prepare_for_detach): Abort in-progress displaced steps instead of
letting them complete.
(handle_one): If the inferior is detaching, don't add the thread
back to the global step-over chain.
(restart_threads): Don't restart threads if detaching.
(handle_signal_stop): Remove inferior::detaching reference.
|
|
After detaching from a process, the inf->detaching flag is
inadvertently left set to true. If you afterwards reuse the same
inferior to start a new process, GDB will mishave...
The problem is that prepare_for_detach discards the scoped_restore at
the end, while the intention is for the flag to be set only for the
duration of prepare_for_detach.
This was already a bug in the original commit that added
prepare_for_detach, commit 24291992dac3 ("PR gdb/11321"), by yours
truly. Back then, we still used cleanups, and the function called
discard_cleanups instead of do_cleanups, by mistake.
gdb/ChangeLog:
* infrun.c (prepare_for_detach): Don't release scoped_restore
before returning.
|
|
This moves the code handling an event out of wait_one to a separate
function, to be used in another context in a following patch.
gdb/ChangeLog:
* infrun.c (handle_one): New function, factored out from ...
(stop_all_threads): ... here.
|
|
A following patch will add a testcase that has two processes with
threads stepping over a breakpoint continuously, and then detaches
from one of the processes while threads are running. The other
process continues stepping over its breakpoint. And then the testcase
sends a SIGUSR1, expecting that GDB reports it. That would sometimes
hang against gdbserver, due to the bugs fixed here. Both bugs are
related, in that they're about remote protocol asynchronous Stop
notifications. There's a bug in GDB, and another in GDBserver.
The GDB bug:
- when we detach from a process, the remote target discards any
pending RSP notification related to that process, including the
in-flight, yet-unacked notification. Discarding the in-flight
notification is the problem. Until the in-flight notification is
acked with a vStopped packet, the server won't send another %Stop
notification. As a result, the debug session gets messed up. In
the new testcase's case, GDB would hang inside stop_all_threads,
waiting for a stop for one of the process'es threads, which never
arrived -- its stop reply was permanently stuck in the stop reply
queue, waiting for a vStopped packet that never arrived.
In summary:
1. GDBserver sends stop notification about thread X, the remote
target receives it and stores it
2. At the same time, GDB detaches thread X's inferior
3. The remote target discards the received stop notification
4. GDBserver waits forever for the ack
The GDBserver bug:
GDBserver has the opposite bug. It also discards notifications for
the process being detached. If that discards the head of the
notification queue, when gdb sends an ack, it ends up acking the
_next_ notification. Meaning, gdb loses one notification. In the
testcase, this results in a similar hang in stop_all_threads.
So we have two very similar bugs in GDB and GDBserver, both resulting
in a similar symptom. That's why I'm fixing them both at the same
time.
gdb/ChangeLog:
* remote.c (remote_notif_stop_ack): Don't error out on
TARGET_WAITKIND_IGNORE; instead, just ignore the notification.
(remote_target::discard_pending_stop_replies): Don't delete
in-flight notification; instead, clear its contents.
gdbserver/ChangeLog:
* server.cc (discard_queued_stop_replies): Don't ever discard the
notification at the head of the list.
|
|
This adds a testcase exercising attaching to a multi-threaded process,
in all combinations of:
- set non-stop on/off
- maint target non-stop off/on
- "attach" vs "attach &"
This exercises the bugs fixed in the two previous patches.
gdb/testsuite/ChangeLog:
* gdb.threads/attach-non-stop.c: New file.
* gdb.threads/attach-non-stop.exp: New file.
|
|
With "target extended-remote" + "maint set target-non-stop", attaching
hangs like so:
(gdb) attach 1244450
Attaching to process 1244450
[New Thread 1244450.1244450]
[New Thread 1244450.1244453]
[New Thread 1244450.1244454]
[New Thread 1244450.1244455]
[New Thread 1244450.1244456]
[New Thread 1244450.1244457]
[New Thread 1244450.1244458]
[New Thread 1244450.1244459]
[New Thread 1244450.1244461]
[New Thread 1244450.1244462]
[New Thread 1244450.1244463]
* hang *
Attaching to the hung GDB shows that GDB is busy in an infinite loop
in stop_all_threads:
(top-gdb) bt
#0 stop_all_threads () at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:4755
#1 0x000055555597b424 in stop_waiting (ecs=0x7fffffffd930) at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:7738
#2 0x0000555555976fba in handle_signal_stop (ecs=0x7fffffffd930) at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:5868
#3 0x0000555555975f6a in handle_inferior_event (ecs=0x7fffffffd930) at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:5527
#4 0x0000555555971da4 in fetch_inferior_event () at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:3910
#5 0x00005555559540b2 in inferior_event_handler (event_type=INF_REG_EVENT) at /home/pedro/gdb/binutils-gdb/src/gdb/inf-loop.c:42
#6 0x000055555597e825 in infrun_async_inferior_event_handler (data=0x0) at /home/pedro/gdb/binutils-gdb/src/gdb/infrun.c:9162
#7 0x0000555555687d1d in check_async_event_handlers () at /home/pedro/gdb/binutils-gdb/src/gdb/async-event.c:328
#8 0x0000555555e48284 in gdb_do_one_event () at /home/pedro/gdb/binutils-gdb/src/gdbsupport/event-loop.cc:216
#9 0x00005555559e7512 in start_event_loop () at /home/pedro/gdb/binutils-gdb/src/gdb/main.c:347
#10 0x00005555559e765d in captured_command_loop () at /home/pedro/gdb/binutils-gdb/src/gdb/main.c:407
#11 0x00005555559e8f80 in captured_main (data=0x7fffffffdb70) at /home/pedro/gdb/binutils-gdb/src/gdb/main.c:1239
#12 0x00005555559e8ff2 in gdb_main (args=0x7fffffffdb70) at /home/pedro/gdb/binutils-gdb/src/gdb/main.c:1254
#13 0x0000555555627c86 in main (argc=12, argv=0x7fffffffdc88) at /home/pedro/gdb/binutils-gdb/src/gdb/gdb.c:32
The problem is that the remote sends stops for all the threads:
Packet received: l/home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.threads/attach-non-stop/attach-non-stop
Sending packet: $vStopped#55...Packet received: T0006:f06e25edec7f0000;07:f06e25edec7f0000;10:f14190ccf4550000;thread:p12fd22.12fd2f;core:15;
Sending packet: $vStopped#55...Packet received: T0006:f0dea5f0ec7f0000;07:f0dea5f0ec7f0000;10:e84190ccf4550000;thread:p12fd22.12fd27;core:4;
Sending packet: $vStopped#55...Packet received: T0006:f0ee25f1ec7f0000;07:f0ee25f1ec7f0000;10:f14190ccf4550000;thread:p12fd22.12fd26;core:5;
Sending packet: $vStopped#55...Packet received: T0006:f0bea5efec7f0000;07:f0bea5efec7f0000;10:f14190ccf4550000;thread:p12fd22.12fd29;core:1;
Sending packet: $vStopped#55...Packet received: T0006:f0ce25f0ec7f0000;07:f0ce25f0ec7f0000;10:e84190ccf4550000;thread:p12fd22.12fd28;core:a;
Sending packet: $vStopped#55...Packet received: T0006:f07ea5edec7f0000;07:f07ea5edec7f0000;10:e84190ccf4550000;thread:p12fd22.12fd2e;core:f;
Sending packet: $vStopped#55...Packet received: T0006:f0ae25efec7f0000;07:f0ae25efec7f0000;10:df4190ccf4550000;thread:p12fd22.12fd2a;core:6;
Sending packet: $vStopped#55...Packet received: T0006:0000000000000000;07:c0e8a381fe7f0000;10:bf43b4f1ec7f0000;thread:p12fd22.12fd22;core:2;
Sending packet: $vStopped#55...Packet received: T0006:f0fea5f1ec7f0000;07:f0fea5f1ec7f0000;10:df4190ccf4550000;thread:p12fd22.12fd25;core:8;
Sending packet: $vStopped#55...Packet received: T0006:f09ea5eeec7f0000;07:f09ea5eeec7f0000;10:e84190ccf4550000;thread:p12fd22.12fd2b;core:b;
Sending packet: $vStopped#55...Packet received: OK
But then wait_one never consumes them, always hitting this path:
4473 if (nfds == 0)
4474 {
4475 /* No waitable targets left. All must be stopped. */
4476 return {NULL, minus_one_ptid, {TARGET_WAITKIND_NO_RESUMED}};
4477 }
Resulting in GDB constanly calling target_stop to stop threads, but
the remote target never reporting back the stops to infrun.
That TARGET_WAITKIND_NO_RESUMED path shown above is always taken
because here, in wait_one too, just above:
4428 for (inferior *inf : all_inferiors ())
4429 {
4430 process_stratum_target *target = inf->process_target ();
4431 if (target == NULL
4432 || !target->is_async_p ()
^^^^^^^^^^^^^^^^^^^^^
4433 || !target->threads_executing)
4434 continue;
... the remote target is not async.
And in turn that happened because extended_remote_target::attach
misses enabling async in the target-non-stop path.
A testcase exercising this will be added in a following patch.
gdb/ChangeLog:
* remote.c (extended_remote_target::attach): Set target async in
the target-non-stop path too.
|
|
Attaching in non-stop mode currently misbehaves, like so:
(gdb) attach 1244450
Attaching to process 1244450
[New LWP 1244453]
[New LWP 1244454]
[New LWP 1244455]
[New LWP 1244456]
[New LWP 1244457]
[New LWP 1244458]
[New LWP 1244459]
[New LWP 1244461]
[New LWP 1244462]
[New LWP 1244463]
No unwaited-for children left.
At this point, GDB's stopped/running thread state is out of sync with
the inferior:
(gdb) info threads
Id Target Id Frame
* 1 LWP 1244450 "attach-non-stop" 0xf1b443bf in ?? ()
2 LWP 1244453 "attach-non-stop" (running)
3 LWP 1244454 "attach-non-stop" (running)
4 LWP 1244455 "attach-non-stop" (running)
5 LWP 1244456 "attach-non-stop" (running)
6 LWP 1244457 "attach-non-stop" (running)
7 LWP 1244458 "attach-non-stop" (running)
8 LWP 1244459 "attach-non-stop" (running)
9 LWP 1244461 "attach-non-stop" (running)
10 LWP 1244462 "attach-non-stop" (running)
11 LWP 1244463 "attach-non-stop" (running)
(gdb)
(gdb) interrupt -a
(gdb)
*nothing*
The problem is that attaching installs an inferior continuation,
called when the target reports the initial attach stop, here, in
inf-loop.c:inferior_event_handler:
/* Do all continuations associated with the whole inferior (not
a particular thread). */
if (inferior_ptid != null_ptid)
do_all_inferior_continuations (0);
However, currently in non-stop mode, inferior_ptid is still null_ptid
when we get here.
If you try to do "set debug infrun 1" to debug the problem, however,
then the attach completes correctly, with GDB reporting a stop for
each thread.
The bug is that we're missing a switch_to_thread/context_switch call
when handling the initial stop, here:
if (stop_soon == STOP_QUIETLY_NO_SIGSTOP
&& (ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_STOP
|| ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP
|| ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_0))
{
stop_print_frame = true;
stop_waiting (ecs);
ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
return;
}
Note how the STOP_QUIETLY / STOP_QUIETLY_REMOTE case above that does
call context_switch.
And the reason "set debug infrun 1" "fixes" it, is that the debug path
has a switch_to_thread call.
This patch fixes it by moving the main context_switch call earlier.
It also removes the:
if (ecs->ptid != inferior_ptid)
check at the same time because:
#1 - that is half of what context_switch already does
#2 - deprecated_context_hook is only used in Insight, and all it does
is set an int. It won't care if we call it when the current
thread hasn't actually changed.
A testcase exercising this will be added in a following patch.
gdb/ChangeLog:
PR gdb/27055
* infrun.c (handle_signal_stop): Move main context_switch call
earlier, before STOP_QUIETLY_NO_SIGSTOP.
|
|
This patch makes the inferior command display information about the
current inferior when called with no argument. This behavior is similar
to the one of the thread command.
Before patch:
(gdb) info inferior
Num Description Connection Executable
* 1 process 19221 1 (native) /home/lsix/tmp/a.out
2 process 19239 1 (native) /home/lsix/tmp/a.out
(gdb) inferior 2
[Switching to inferior 2 [process 19239] (/home/lsix/tmp/a.out)]
[Switching to thread 2.1 (process 19239)]
#0 0x0000000000401146 in main ()
(gdb) inferior
Argument required (expression to compute).
After patch:
(gdb) info inferior
Num Description Connection Executable
* 1 process 18699 1 (native) /home/lsix/tmp/a.out
2 process 18705 1 (native) /home/lsix/tmp/a.out
(gdb) inferior 2
[Switching to inferior 2 [process 18705] (/home/lsix/tmp/a.out)]
[Switching to thread 2.1 (process 18705)]
#0 0x0000000000401146 in main ()
(gdb) inferior
[Current inferior is 2 [process 18705] (/home/lsix/tmp/a.out)]
gdb/doc/ChangeLog:
* gdb.texinfo (Inferiors Connections and Programs): Document the
inferior command when used without argument.
gdb/ChangeLog:
* NEWS: Add entry for the behavior change of the inferior command.
* inferior.c (inferior_command): When no argument is given to the
inferior command, display info about the currently selected
inferior.
gdb/testsuite/ChangeLog:
* gdb.base/inferior-noarg.c: New test.
* gdb.base/inferior-noarg.exp: New test.
|
|
Fixes:
Running /home/simark/src/binutils-gdb/gdb/testsuite/gdb.base/scope.exp ...
DUPLICATE: gdb.base/scope.exp: print 'scope0.c'::filelocal_ro
DUPLICATE: gdb.base/scope.exp: print 'scope0.c'::filelocal_ro
DUPLICATE: gdb.base/scope.exp: next
gdb/testsuite/ChangeLog:
* gdb.base/scope.exp: Use proc_with_prefix.
Change-Id: Ic40e24a603da6f2a4f8003b74b2ff3040d2b098d
|
|
It is possible for the tables in the .debug_{rng,loc}lists sections to
not have an array of offsets. In that case, the offset_entry_count
field of the header is 0. The forms DW_FORM_{rng,loc}listx (reference
by index) can't be used with that table. Instead, the
DW_FORM_sec_offset form, which references a {rng,loc}list by direct
offset in the section, must be used. From what I saw, this is what GCC
currently produces.
Add tests for this case. I didn't see any bug related to this, I just
think that it would be nice to have coverage for this. A new
`-with-offset-array` option is added to the `table` procs, used when
generating {rng,loc}lists, to decide whether to generate the offset
array.
gdb/testsuite/ChangeLog:
* lib/dwarf.exp (rnglists): Add -no-offset-array option to
table proc.
* gdb.dwarf2/rnglists-sec-offset.exp: Add test for
.debug_rnglists table without offset array.
* gdb.dwarf2/loclists-sec-offset.exp: Add test for
.debug_loclists table without offset array.
Change-Id: I8e34a7bf68c9682215ffbbf66600da5b7db91ef7
|
|
I think it's wrong that read_loclist_index and read_rnglist_index return
a CORE_ADDR. A CORE_ADDR is an address in the program. These functions
return offset in sections (.debug_loclists and .debug_rnglists). I
think sect_offset is more appropriate.
I'm wondering if struct attribute should have a "set_sect_offset"
method, that takes a sect_offset parameter, or if it's better to be
left as a simple "unsigned".
gdb/ChangeLog:
* dwarf2/read.c (read_loclist_index, read_rnglist_index): Return
a sect_offset.
(read_attribute_reprocess): Adjust.
Change-Id: I0e22e0864130fb490072b41ae099762918b8ad4d
|
|
Consider the test case added in this patch. It defines a compilation
unit with a DW_AT_rnglists_base attribute (used for attributes of form
DW_FORM_rnglistx), but also uses DW_AT_ranges of form
DW_FORM_sec_offset:
0x00000027: DW_TAG_compile_unit
DW_AT_ranges [DW_FORM_sec_offset] (0x0000004c
[0x0000000000005000, 0x0000000000006000))
DW_AT_rnglists_base [DW_FORM_sec_offset] (0x00000044)
The DW_AT_rnglists_base does not play a role in reading the DW_AT_ranges of
form DW_FORM_sec_offset, but it should also not do any harm.
This case is currently not handled correctly by GDB. This is not
something that a compiler is likely to emit, but in my opinion there's
no reason why GDB should fail reading it.
The problem is that in partial_die_info::read and a few other places
where the same logic is replicated, the cu->ranges_base value,
containing the DW_AT_rnglists_base value, is wrongfully added to the
DW_AT_ranges value.
It is quite messy how to decide whether cu->ranges_base should be added
to the attribute's value or not. But to summarize, the only time we
want to add it is when the attribute comes from a pre-DWARF 5 split unit
file (a .dwo) [1]. In this case, the DW_AT_ranges attribute from the
split unit file will have form DW_FORM_sec_offset, pointing somewhere in
the linked file's .debug_ranges section. *But* it's not a "true"
DW_FORM_sec_offset, in that it's an offset relative to the beginning of
that CU's contribution in the section, not relative to the beginning of
the section. So in that case, and only that case, do we want to add the
ranges base value, which we found from the DW_AT_GNU_ranges_base
attribute on the skeleton unit.
Almost all instances of the DW_AT_ranges attribute will be found in the
split unit (on DW_TAG_subprogram, for example), and therefore need to
have the ranges base added. However, the DW_TAG_compile_unit DIE in the
skeleton may also have a DW_AT_ranges attribute. For that one, the
ranges base must not be added. Once the DIEs have been loaded in GDB,
however, the distinction between what's coming from the skeleton and
what's coming from the split unit is not clear. It is all merged in one
big happy tree. So how do we know if a given attribute comes from the
split unit or not?
We use the fact that in pre-DWARF 5 split DWARF, DW_AT_ranges is found
on the skeleton's DW_TAG_compile_unit (in the linked file) and never in
the split unit's DW_TAG_compile_unit. This is why you have this in
partial_die_info::read:
int need_ranges_base = (tag != DW_TAG_compile_unit
&& attr.form != DW_FORM_rnglistx);
However, with the corner case described above (where we have a
DW_AT_rnglists_base attribute and a DW_AT_ranges attribute of form
DW_FORM_sec_offset) the condition gets it wrong when it encounters an
attribute like DW_TAG_subprogram with a DW_AT_ranges attribute of
DW_FORM_sec_offset form: it thinks that it is necessary to add the base,
when it reality it is not.
The problem boils down to failing to differentiate these cases:
- a DW_AT_ranges attribute of form DW_FORM_sec_offset in a
pre-DWARF 5 split unit (in which case we need to add the base)
- a DW_AT_ranges attribute of form DW_FORM_sec_offset in a DWARF 5
non-split unit (in which case we must not add the base)
What makes it unnecessarily complex is that the cu->ranges_base field is
overloaded, used to hold the pre-DWARF 5, non-standard
DW_AT_GNU_ranges_base and the DWARF 5 DW_AT_rnglists_base. In reality,
these two are called "bases" but are not the same thing. The result is
that we need twisted conditions to try to determine whether or not we
should add the base to the attribute's value.
To fix it, split the field in two distinct fields. I renamed everything
related to the "old" ranges base to "gnu_ranges_base", to make it clear
that it's about the non-standard, pre-DWARF 5 thing. And everything
related to the DWARF 5 thing gets renamed "rnglists". I think it
becomes much easier to reason this way.
The issue described above gets fixed by the fact that the
DW_AT_rnglists_base value does not end up in cu->gnu_ranges_base, so
cu->gnu_ranges_base stays 0. The condition to determine whether
gnu_ranges_base should be added can therefore be simplified back to:
tag != DW_TAG_compile_unit
... as it was before rnglistx support was added.
Extend the gdb.dwarf2/rnglists-sec-offset.exp to cover this case. I
also extended the test case for loclists similarly, just to see if there
would be some similar problem. There wasn't, but I think it's not a bad
idea to test that case for loclists as well, so I left it in the patch.
[1] https://gcc.gnu.org/wiki/DebugFission
gdb/ChangeLog:
* dwarf2/die.h (struct die_info) <ranges_base>: Split in...
<gnu_ranges_base>: ... this...
<rnglists_base>: ... and this.
* dwarf2/read.c (struct dwarf2_cu) <ranges_base>: Split in...
<gnu_ranges_base>: ... this...
<rnglists_base>: ... and this.
(read_cutu_die_from_dwo): Adjust
(dwarf2_get_pc_bounds): Adjust
(dwarf2_record_block_ranges): Adjust.
(read_full_die_1): Adjust
(partial_die_info::read): Adjust.
(read_rnglist_index): Adjust.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/rnglists-sec-offset.exp: Add test for DW_AT_ranges
of DW_FORM_sec_offset form plus DW_AT_rnglists_base attribute.
* gdb.dwarf2/loclists-sec-offset.exp: Add test for
DW_AT_location of DW_FORM_sec_offset plus DW_AT_loclists_base
attribute
Change-Id: Icd109038634b75d0e6e9d7d1dcb62fb9eb951d83
|
|
Add tests for the various issues fixed in the previous patches.
Add a new "loclists" procedure to the DWARF assembler, to allow
generating .debug_loclists sections.
gdb/testsuite/ChangeLog:
PR gdb/26813
* lib/dwarf.exp (_handle_DW_FORM): Handle DW_FORM_loclistx.
(loclists): New proc.
* gdb.dwarf2/loclists-multiple-cus.c: New.
* gdb.dwarf2/loclists-multiple-cus.exp: New.
* gdb.dwarf2/loclists-sec-offset.c: New.
* gdb.dwarf2/loclists-sec-offset.exp: New.
Change-Id: I209bcb2a9482762ae943e518998d1f7761f76928
|
|
The _location proc is used to assemble a location description. It needs
to know some contextual information:
- size of an address
- size of an offset (into another DWARF section)
- DWARF version
It currently get all this directly from global variables holding the
compilation unit information. This is fine because as of now, all
location descriptions are generated in the context of creating a
compilation unit. However, a subsequent patch will generate location
descriptions while generating a .debug_loclists section. _location
should therefore no longer rely on the current compilation unit's
properties.
Change it to accept these values as parameters instead of accessing the
values for the CU.
No functional changes intended.
gdb/testsuite/ChangeLog:
* lib/dwarf.exp (_location): Add parameters.
(_handle_DW_FORM): Adjust.
Change-Id: Ib94981979c83ffbebac838081d645ad71c221637
|
|
Add tests for the various issues fixed in the previous patches.
Add a new "rnglists" procedure to the DWARF assembler, to allow
generating .debug_rnglists sections. A trivial change is required to
support the DWARF 5 CU header layout.
gdb/testsuite/ChangeLog:
PR gdb/26813
* lib/dwarf.exp (_handle_DW_FORM): Handle DW_FORM_rnglistx.
(cu): Generate header for DWARF 5.
(rnglists): New proc.
* gdb.dwarf2/rnglists-multiple-cus.exp: New.
* gdb.dwarf2/rnglists-sec-offset.exp: New.
Change-Id: I5b297e59c370c60cf671dec19796a6c3b9a9f632
|
|
When loading the binary from PR 26813 in GDB, we get:
DW_FORM_rnglistx index pointing outside of .debug_rnglists offset array [in module /home/simark/build/binutils-gdb/gdb/MagicPurse]
... and the symbols fail to load.
In read_rnglist_index and read_loclist_index, we read the header
(documented in sections 7.28 and 7.29 of DWARF 5) of the CU's
contribution to the .debug_rnglists / .debug_loclists sections to
validate that the index we want to read makes sense. However, we always
read the header at the beginning of the section, rather than the header
for the contribution from which we want to read the index.
To illustrate, here's what the binary from PR 26813 contains. There are
two compile units:
0x0000000c: DW_TAG_compile_unit 1
DW_AT_ranges [DW_FORM_rnglistx]: 0x0
DW_AT_rnglists_base [DW_FORM_sec_offset]: 0xC
0x00003ec9: DW_TAG_compile_unit 2
DW_AT_ranges [DW_FORM_rnglistx]: 0xB
DW_AT_rnglists_base [DW_FORM_sec_offset]: 0x85
The layout of the .debug_rnglists is the following:
[0x00, 0x0B]: header for CU 1's contribution
[0x0C, 0x0F]: list of offsets for CU 1 (1 element)
[0x10, 0x78]: range lists data for CU 1
[0x79, 0x84]: header for CU 2's contribution
[0x85, 0xB4]: list of offsets for CU 2 (12 elements)
[0xB5, 0xBD7]: range lists data for CU 2
The DW_AT_rnglists_base attrbute points to the beginning of the list of
offsets for that CU, relative to the start of the .debug_rnglists
section. That's right after the header for that contribution.
When we try to read the DW_AT_ranges attribute for CU 2,
read_rnglist_index reads the header for CU 1 instead of the one for CU
2. Since there's only one element in CU 1's offset list, it believes
(wrongfully) that the index 0xB is out of range.
Fix it by reading the header just before where DW_AT_rnglists_base
points to. With this patch, I am able to load GDB built with clang-11
and -gdwarf-5 in itself, with and without -readnow.
gdb/ChangeLog:
PR gdb/26813
* dwarf2/read.c (read_loclists_rnglists_header): Add
header_offset parameter and use it.
(read_loclist_index): Read header of the current contribution,
not the one at the beginning of the section.
(read_rnglist_index): Likewise.
Change-Id: Ie53ff8251af8c1556f0a83a31aa8572044b79e3d
|
|
We hit an assertion when loading the binary from PR 26813. When fixing
it, execution goes a up bit further but then hits another assert, and
another, and another. With these fours fixes, I am able to load the
binary and get to the prompt. An error is shown (index pointing outside
of the section), because the DW_FORM_rnglistx attribute is not read
correctly, but that one is taken care of by the next patch.
The four fixes are:
- attribute::form_requires_reprocessing needs to handle forms
DW_FORM_rnglistx and DW_FORM_loclistx, because set_unsigned_reprocess
is called for them in read_attribute_value.
- read_attribute_reprocess must call set_unsigned for them, not
set_address. The parameter of set_address is a CORE_ADDR, meaning
it's for program addresses. Post-reprocess, DW_FORM_rnglistx and
DW_FORM_loclistx are offsets into their respective sections
(.debug_rnglists and .debug_loclists). set_unsigned is the current
attribute value setter that fits the best. But perhaps we should have
a setter that takes a sect_offset?
- read_attribute_process must call as_unsigned_reprocess instead of
as_unsigned to get the pre-reprocess value, otherwise we hit the
assert inside as_unsigned that makes sure the attribute doesn't need
reprocessing.
- attribute::set_unsigned needs to clear the requires_reprocessing flag,
otherwise it stays set when reprocessing DW_FORM_rnglistx and
DW_FORM_loclistx attributes.
There's another assert that we hit once the next patch is applied, but
since it's in the same vein as the changes in this patch, I included it
in this patch:
- attribute::form_is_unsigned must handle form DW_FORM_loclistx,
otherwise we hit the assert when trying to call set_unsigned for an
attribute of this form. DW_FORM_rnglistx is already handled.
gdb/ChangeLog:
PR gdb/26813
* dwarf2/attribute.h (struct attribute) <set_unsigned>: Clear
requires_reprocessing flag.
* dwarf2/attribute.c (attribute::form_is_unsigned): Handle
DW_FORM_loclistx.
(attribute::form_requires_reprocessing): Handle DW_FORM_rnglistx
and DW_FORM_loclistx.
* dwarf2/read.c (read_attribute_reprocess): Use set_unsigned
instead of set_address for DW_FORM_loclistx and
DW_FORM_rnglistx.
Change-Id: I06c156fa3913ca98e4e39085f4ef171645b4bc1e
|
|
In read_rnglist_index and read_loclist_index, we check that both the
start and end of the offset that we read from the offset table are
within the section. I think it's unecessary to do both: if the end of
the offset is within the section, then surely the start of the offset is
within it.
Remove the check for the start of the offset in both functions.
gdb/ChangeLog:
* dwarf2/read.c (read_loclist_index): Remove bound check for
start of offset.
(read_rnglist_index): Likewise.
Change-Id: I7b57ddf4f8a8a28971738f0e3f3af62108f9e19a
|
|
read_rnglist_index has a bound check to make sure that we don't go past
the end of the section while reading the offset, but read_loclist_index
doesn't. Add it to read_loclist_index.
gdb/ChangeLog:
* dwarf2/read.c (read_loclist_index): Add bound check for the end
of the offset.
Change-Id: Ic4b55c88860fdc3e007740949c78ec84cdb4da60
|
|
I think this check in read_rnglist_index is wrong:
/* Validate that reading won't go beyond the end of the section. */
if (start_offset + cu->header.offset_size > rnglist_base + section->size)
error (_("Reading DW_FORM_rnglistx index beyond end of"
".debug_rnglists section [in module %s]"),
objfile_name (objfile));
The addition `rnglist_base + section->size` doesn't make sense.
rnglist_base is an offset into `section`, so it doesn't make sense to
add it to `section`'s size. `start_offset` also is an offset into
`section`, so we should just compare it to just `section->size`.
gdb/ChangeLog:
* dwarf2/read.c (read_rnglist_index): Fix bound check.
Change-Id: If0ff7c73f4f80f79aac447518f4e8f131f2db8f2
|
|
Unlike read_rnglists_index, read_loclist_index uses complaints when it
detects an inconsistency (a DW_FORM_loclistx value without a
.debug_loclists section or an offset outside of the section). I really
think they should be errors, since there's no point in continuing if
this situation happens, we will likely segfault or read garbage.
gdb/ChangeLog:
* dwarf2/read.c (read_loclist_index): Change complaints into
errors.
Change-Id: Ic3a1cf6e682d47cb6e739dd76fd7ca5be2637e10
|
|
When running test-case gdb.dwarf2/fission-reread.exp with target board
cc-with-gdb-index, we run into an abort during the generation of the gdb-index
by cc-with-tweaks.sh:
...
build/gdb/testsuite/cache/gdb.sh: line 1: 27275 Aborted (core dumped)
...
This can be reproduced on the command line like this:
...
$ gdb -batch ./outputs/gdb.dwarf2/fission-reread/fission-reread \
-ex 'save gdb-index ./outputs/gdb.dwarf2/fission-reread'
warning: Could not find DWO TU fission-reread.dwo(0x9022f1ceac7e8b19) \
referenced by TU at offset 0x0 [in module fission-reread]
warning: Could not find DWO CU fission-reread.dwo(0x807060504030201) \
referenced by CU at offset 0x561 [in module fission-reread]
Aborted (core dumped)
...
The abort is a segfault due to a using a nullptr psymtab in
write_one_signatured_type.
The problem is that we're trying to write index entries for the type unit
with signature:
...
(gdb) p /x entry->signature
$2 = 0x9022f1ceac7e8b19
...
which is a skeleton type unit:
...
Contents of the .debug_types section:
Compilation Unit @ offset 0x0:
Length: 0x4a (32-bit)
Version: 4
Abbrev Offset: 0x165
Pointer Size: 4
Signature: 0x9022f1ceac7e8b19
Type Offset: 0x0
<0><17>: Abbrev Number: 2 (DW_TAG_type_unit)
<18> DW_AT_comp_dir : /tmp/src/gdb/testsuite
<2f> DW_AT_GNU_dwo_name: fission-reread.dwo
<42> DW_AT_GNU_pubnames: 0x0
<46> DW_AT_GNU_pubtypes: 0x0
<4a> DW_AT_GNU_addr_base: 0x0
...
referring to a .dwo file, but as the warnings show, the .dwo file is not
found.
Fix this by skipping the type unit in write_one_signatured_type if
psymtab == nullptr.
Tested on x86_64-linux.
gdb/ChangeLog:
2021-02-02 Tom de Vries <tdevries@suse.de>
PR symtab/24620
* dwarf2/index-write.c (write_one_signatured_type): Skip if
psymtab == nullptr.
gdb/testsuite/ChangeLog:
2021-02-02 Tom de Vries <tdevries@suse.de>
PR symtab/24620
* gdb.dwarf2/fission-reread.exp: Add test-case.
|
|
When running test-case gdb.dwarf2/fission-reread.exp with target board
cc-with-gdb-index, we run into:
...
gdb compile failed, warning: Could not find DWO TU \
fission-reread.dwo(0x9022f1ceac7e8b19) referenced by TU at offset 0x0 \
[in module outputs/gdb.dwarf2/fission-reread/fission-reread]
...
The problem is that the .dwo file is not found.
There's code added in the .exp file to make sure the .dwo can be found:
...
# Make sure we can find the .dwo file, regardless of whether we're
# running in parallel mode.
gdb_test_no_output "set debug-file-directory [file dirname $binfile]" \
"set debug-file-directory"
...
This works normally, but not for the gdb invocation done by cc-with-tweaks.sh
for target board cc-with-gdb-index.
Fix this by finding the full path to the .dwo file and passing it
to the compilation.
Tested on x86_64-linux with native and target boards cc-with-gdb-index,
cc-with-debug-names and readnow.
gdb/testsuite/ChangeLog:
2021-02-01 Tom de Vries <tdevries@suse.de>
* gdb.dwarf2/fission-base.S: Pass -DDWO=$dwo.
* gdb.dwarf2/fission-loclists-pie.S: Same.
* gdb.dwarf2/fission-loclists.S: Same.
* gdb.dwarf2/fission-multi-cu.S: Same.
* gdb.dwarf2/fission-reread.S: Same.
* gdb.dwarf2/fission-base.exp: Use DWO.
* gdb.dwarf2/fission-loclists-pie.exp: Same.
* gdb.dwarf2/fission-loclists.exp: Same.
* gdb.dwarf2/fission-multi-cu.exp: Same.
* gdb.dwarf2/fission-reread.exp: Same.
|
|
While reviewing the Linux and FreeBSD core dumping code within GDB for
another patch series, I noticed that the code that collects the
registers for each thread and writes these into ELF note format is
basically identical between Linux and FreeBSD.
This commit merges this code and moves it into the gcore.c file,
which seemed like the right place for generic writing a core file
code.
The function find_signalled_thread is moved from linux-tdep.c despite
not being shared. A later commit will make use of this function.
There are a couple of minor changes to the FreeBSD target after this
commit, but I believe that these are changes for the better:
(1) For FreeBSD we always used to record the thread-id in the core file by
using ptid_t.lwp (). In contrast the Linux code did this:
/* For remote targets the LWP may not be available, so use the TID. */
long lwp = ptid.lwp ();
if (lwp == 0)
lwp = ptid.tid ();
Both target now do this:
/* The LWP is often not available for bare metal target, in which case
use the tid instead. */
if (ptid.lwp_p ())
lwp = ptid.lwp ();
else
lwp = ptid.tid ();
Which is equivalent for Linux, but is a change for FreeBSD. I think
that all this means is that in some cases where GDB might have
previously recorded a thread-id of 0 for each thread, we might now get
something more useful.
(2) When collecting the registers for Linux we collected into a zero
initialised buffer. By contrast on FreeBSD the buffer is left
uninitialised. In the new code the buffer is always zero initialised.
I suspect once the registers are copied into the buffer there's
probably no gaps left so this makes no difference, but if it does then
using zeros rather than random bits of GDB's memory is probably a good
thing.
Otherwise, there should be no other user visible changes after this
commit.
Tested this on x86-64/GNU-Linux and x86-64/FreeBSD-12.2 with no
regressions.
gdb/ChangeLog:
* Makefile.in (HFILES_NO_SRCDIR): Add corefile.h.
* gcore.c (struct gcore_collect_regset_section_cb_data): Moved
here from linux-tdep.c and given a new name. Minor cleanups.
(gcore_collect_regset_section_cb): Likewise.
(gcore_collect_thread_registers): Likewise.
(gcore_build_thread_register_notes): Likewise.
(gcore_find_signalled_thread): Likewise.
* gcore.h (gcore_build_thread_register_notes): Declare.
(gcore_find_signalled_thread): Declare.
* fbsd-tdep.c: Add 'gcore.h' include.
(struct fbsd_collect_regset_section_cb_data): Delete.
(fbsd_collect_regset_section_cb): Delete.
(fbsd_collect_thread_registers): Delete.
(struct fbsd_corefile_thread_data): Delete.
(fbsd_corefile_thread): Delete.
(fbsd_make_corefile_notes): Call
gcore_build_thread_register_notes instead of the now deleted
FreeBSD code.
* linux-tdep.c: Add 'gcore.h' include.
(struct linux_collect_regset_section_cb_data): Delete.
(linux_collect_regset_section_cb): Delete.
(linux_collect_thread_registers): Delete.
(linux_corefile_thread): Call
gcore_build_thread_register_notes.
(find_signalled_thread): Delete.
(linux_make_corefile_notes): Call gcore_find_signalled_thread.
|
|
Consider the test-case small.c:
...
$ cat -n small.c
1 __attribute__ ((noinline, noclone))
2 int foo (char *c)
3 {
4 asm volatile ("" : : "r" (c) : "memory");
5 return 1;
6 }
7
8 int main ()
9 {
10 char tpl1[20] = "/tmp/test.XXX";
11 char tpl2[20] = "/tmp/test.XXX";
12 int fd1 = foo (tpl1);
13 int fd2 = foo (tpl2);
14 if (fd1 == -1) {
15 return 1;
16 }
17
18 return 0;
19 }
...
Compiled with gcc-8 and optimization:
...
$ gcc-8 -O2 -g small.c
...
We step through the calls to foo, but fail to visit line 13:
...
12 int fd1 = foo (tpl1);
(gdb) step
foo (c=c@entry=0x7fffffffdea0 "/tmp/test.XXX") at small.c:5
5 return 1;
(gdb) step
foo (c=c@entry=0x7fffffffdec0 "/tmp/test.XXX") at small.c:5
5 return 1;
(gdb) step
main () at small.c:14
14 if (fd1 == -1) {
(gdb)
...
This is caused by the following. The calls to foo are implemented by these
insns:
....
4003df: 0f 29 04 24 movaps %xmm0,(%rsp)
4003e3: 0f 29 44 24 20 movaps %xmm0,0x20(%rsp)
4003e8: e8 03 01 00 00 callq 4004f0 <foo>
4003ed: 48 8d 7c 24 20 lea 0x20(%rsp),%rdi
4003f2: 89 c2 mov %eax,%edx
4003f4: e8 f7 00 00 00 callq 4004f0 <foo>
4003f9: 31 c0 xor %eax,%eax
...
with corresponding line table entries:
...
INDEX LINE ADDRESS IS-STMT
8 12 0x00000000004003df Y
9 10 0x00000000004003df
10 11 0x00000000004003e3
11 12 0x00000000004003e8
12 13 0x00000000004003ed
13 12 0x00000000004003f2
14 13 0x00000000004003f4 Y
15 13 0x00000000004003f4
16 14 0x00000000004003f9 Y
17 14 0x00000000004003f9
...
Once we step out of the call to foo at 4003e8, we land at 4003ed, and gdb
enters process_event_stop_test to figure out what to do.
That entry has is-stmt=n, so it's not the start of a line, so we don't stop
there. However, we do update ecs->event_thread->current_line to line 13,
because the frame has changed (because we stepped out of the function).
Next we land at 4003f2. Again the entry has is-stmt=n, so it's not the start
of a line, so we don't stop there. However, because the frame hasn't changed,
we don't update update ecs->event_thread->current_line, so it stays 13.
Next we land at 4003f4. Now is-stmt=y, so it's the start of a line, and we'd
like to stop here.
But we don't stop because this test fails:
...
if ((ecs->event_thread->suspend.stop_pc == stop_pc_sal.pc)
&& (ecs->event_thread->current_line != stop_pc_sal.line
|| ecs->event_thread->current_symtab != stop_pc_sal.symtab))
{
...
because ecs->event_thread->current_line == 13 and stop_pc_sal.line == 13.
Fix this by resetting ecs->event_thread->current_line to 0 if is-stmt=n and
the frame has changed, such that we have:
...
12 int fd1 = foo (tpl1);
(gdb) step
foo (c=c@entry=0x7fffffffdbc0 "/tmp/test.XXX") at small.c:5
5 return 1;
(gdb) step
main () at small.c:13
13 int fd2 = foo (tpl2);
(gdb)
...
Tested on x86_64-linux, with gcc-7 and gcc-8.
gdb/ChangeLog:
2021-01-29 Tom de Vries <tdevries@suse.de>
PR breakpoints/26063
* infrun.c (process_event_stop_test): Reset
ecs->event_thread->current_line to 0 if is-stmt=n and frame has
changed.
gdb/testsuite/ChangeLog:
2021-01-29 Tom de Vries <tdevries@suse.de>
PR breakpoints/26063
* gdb.dwarf2/dw2-step-out-of-function-no-stmt.c: New test.
* gdb.dwarf2/dw2-step-out-of-function-no-stmt.exp: New file.
|
|
When running test-case gdb.opt/solib-intra-step.exp with target board
unix/-m32 and gcc-10, I run into:
...
(gdb) step^M
__x86.get_pc_thunk.bx () at ../sysdeps/i386/crti.S:68^M
68 ../sysdeps/i386/crti.S: No such file or directory.^M
(gdb) step^M
shlib_second (dummy=0) at solib-intra-step-lib.c:23^M
23 abort (); /* second-hit */^M
(gdb) FAIL: gdb.opt/solib-intra-step.exp: second-hit
...
The problem is that the test-case expects to step past the retry line,
which is optional.
Fix this by removing the state tracking logic from the gdb_test_multiples. It
makes the test more difficult to understand, and doesn't specifically test for
faulty gdb behaviour.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2021-01-29 Tom de Vries <tdevries@suse.de>
* gdb.opt/solib-intra-step.exp: Remove state tracking logic.
|
|
I spotted that every time thr_try_catch_cmd is called GDB has already
switched to the required thread. The call to switch_to_thread at the
head of thr_try_catch_cmd is therefore redundant.
This commit replaces the call to switch_to_thread with an assertion
that we already have the required thread selected.
I also extended the header comment on thr_try_catch_cmd to make it
clearer when this function could throw an exception.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* thread.c (thr_try_catch_cmd): Replace swith_to_thread with an
assert. Extend the header comment.
|
|
The locator window, or status window as it is sometimes called is
handled differently to all the other windows.
The reason for this is that the class representing this
window (tui_locator_window) does two jobs, first this class represents
a window just like any other that has space on the screen and fills
the space with content. The second job is that this class serves as a
storage area to hold information about the current location that the
TUI windows represent, so the class has members like 'addr' and
'line_no', for example which are used within this class, and others
when they want to know which line/address the TUI windows should be
showing to the user.
Because of this dual purpose we must always have an instance of the
tui_locator_window so that there is somewhere to store this location
information.
The result of this is that the locator window must never be deleted
like other windows, which results in some special case code.
In this patch I propose splitting the two roles of the
tui_locator_window class. The tui_locator_window class will retain
just its window drawing parts, and will be treated just like any other
window. This should allow all special case code for this window to be
deleted.
The other role, that of tracking the current tui location will be
moved into a new class (tui_location_tracker), of which there will be
a single global instance. All of the places where we previously use
the locator window to get location information will now be updated to
get this from the tui_location_tracker.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* Makefile.in (SUBDIR_TUI_SRCS): Add tui/tui-location.c.
(HFILES_NO_SRCDIR): Add tui/tui-location.h.
* tui/tui-data.h (TUI_STATUS_WIN): Define.
(tui_locator_win_info_ptr): Delete declaration.
* tui/tui-disasm.c: Add 'tui/tui-location.h' include.
(tui_disasm_window::set_contents): Fetch state from tui_location
global.
(tui_get_begin_asm_address): Likewise.
* tui/tui-layout.c (tui_apply_current_layout): Remove special case
for locator window.
(get_locator_window): Delete.
(initialize_known_windows): Treat locator window just like all the
rest.
* tui/tui-source.c: Add 'tui/tui-location.h' include.
(tui_source_window::set_contents): Fetch state from tui_location
global.
(tui_source_window::showing_source_p): Likewise.
* tui/tui-stack.c: Add 'tui/tui-location.h' include.
(_locator): Delete.
(tui_locator_win_info_ptr): Delete.
(tui_locator_window::make_status_line): Fetch state from
tui_location global.
(tui_locator_window::rerender): Remove check of 'handle',
reindent function body.
(tui_locator_window::set_locator_fullname): Delete.
(tui_locator_window::set_locator_info): Delete.
(tui_update_locator_fullname): Delete.
(tui_show_frame_info): Likewise.
(tui_show_locator_content): Access window through TUI_STATUS_WIN.
* tui/tui-stack.h (tui_locator_window::set_locator_info): Moved to
tui/tui-location.h and renamed to
tui_location_tracker::set_location.
(tui_locator_window::set_locator_fullname): Moved to
tui/tui-location.h and renamed to
tui_location_tracker::set_fullname.
(tui_locator_window::full_name): Delete.
(tui_locator_window::proc_name): Delete.
(tui_locator_window::line_no): Delete.
(tui_locator_window::addr): Delete.
(tui_locator_window::gdbarch): Delete.
(tui_update_locator_fullname): Delete declaration.
* tui/tui-wingeneral.c (tui_refresh_all): Removed special handling
for locator window.
* tui/tui-winsource.c: Add 'tui/tui-location.h' include.
(tui_display_main): Call function on tui_location directly.
* tui/tui.h (enum tui_win_type): Add STATUS_WIN.
* tui/tui-location.c: New file.
* tui/tui-location.h: New file.
|
|
When running test-case gdb.arch/i386-gnu-cfi.exp with target board unix/-m32, I get:
...
(gdb) up 3^M
79 abort.c: No such file or directory.^M
(gdb) FAIL: gdb.arch/i386-gnu-cfi.exp: shift up to the modified frame
...
The preceding backtrace looks like this:
...
(gdb) bt^M
#0 0xf7fcf549 in __kernel_vsyscall ()^M
#1 0xf7ce8896 in __libc_signal_restore_set (set=0xffffc3bc) at \
../sysdeps/unix/sysv/linux/internal-signals.h:104^M
#2 __GI_raise (sig=6) at ../sysdeps/unix/sysv/linux/raise.c:47^M
#3 0xf7cd0314 in __GI_abort () at abort.c:79^M
#4 0x0804919f in gate (gate=0x8049040 <abort@plt>, data=0x0) at gate.c:3^M
#5 0x08049176 in main () at i386-gnu-cfi.c:27^M
...
with function gate at position #4, while on another system where the test passes,
I see instead function gate at position #3.
Fix this by capturing the position of function gate in the backtrace, and
using that in the rest of the test instead of hardcoded constant 3.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2021-01-28 Tom de Vries <tdevries@suse.de>
* gdb.arch/i386-gnu-cfi.exp: Capture the position of function gate
in the backtrace, and use that in the rest of the test instead of
hardcoded constant 3. Use "frame" instead of "up" for robustness.
|
|
When running test-case gdb.arch/i386-sse-stack-align.exp on target board
unix/-m32, I run into:
...
(gdb) print (int) g0 ()^M
Invalid data type for function to be called.^M
(gdb) FAIL: gdb.arch/i386-sse-stack-align.exp: print (int) g0 ()
...
Gdb is supposed to use minimal symbol g0:
...
$ nm i386-sse-stack-align | grep g0
08049194 t g0
...
but instead it finds a g0 symbol in the debug info of libm, specifically in
./sysdeps/ieee754/ldbl-96/e_lgammal_r.c.
Fix this by renaming g[0-4] to test_g[0-4].
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2021-01-28 Tom de Vries <tdevries@suse.de>
* gdb.arch/i386-sse-stack-align.S: Rename g[0-4] to test_g[0-4].
* gdb.arch/i386-sse-stack-align.c: Same.
* gdb.arch/i386-sse-stack-align.exp: Same.
|
|
... and update all users.
gdb/ChangeLog:
* gdbtypes.h (get_type_arch): Rename to...
(struct type) <arch>: ... this, update all users.
Change-Id: I0e3ef938a0afe798ac0da74a9976bbd1d082fc6f
|
|
I think this makes the names of the methods clearer, especially for the
arch. The type::arch method (which gets the arch owner, or NULL if the
type is not arch owned) is easily confused with the get_type_arch method
(which returns an arch no matter what). The name "arch_owner" will make
it intuitive that the method returns NULL if the type is not arch-owned.
Also, this frees the type::arch name, so we will be able to morph the
get_type_arch function into the type::arch method.
gdb/ChangeLog:
* gdbtypes.h (struct type) <arch>: Rename to...
<arch_owner>: ... this, update all users.
<objfile>: Rename to...
<objfile_owner>: ... this, update all users.
Change-Id: Ie7c28684c7b565adec05a7619c418c69429bd8c0
|
|
Since this commit:
commit 64aaad6349d2b2c45063a5383f877ce9a3a0c354
Date: Fri Sep 25 14:50:56 2020 +0100
gdb: use get_standard_config_dir when looking for .gdbinit
GDB has been checking for ${XDG_CONFIG_HOME}/gdb/gdbinit on startup.
Most tests pass -nx to GDB to block loading of gdbinit files, but
there are a few tests (e.g. gdb.base/gdbinit-history.exp) that don't
use -nx and instead setup a fake HOME directory containing a gdbinit
file.
However, since the above commit, if XDG_CONFIG_HOME is set then once
-nx is no longer being passed GDB will load any gdbinit file it finds
in that directory, which could cause the test to fail.
As a concrete example:
$ mkdir -p fake_xdg_config_home/gdb/
$ cat <<EOF >fake_xdg_config_home/gdb/gdbinit
echo goodbye\n
quit
EOF
$ export XDG_CONFIG_HOME=$PWD/fake_xdg_config_home
$ make check-gdb TESTS="gdb.base/gdbinit-history.exp"
Should result in the test failing.
The solution I propose is to unset XDG_CONFIG_HOME in
default_gdb_init, we already unset a bunch of environment variables in
this proc.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (default_gdb_init): Unset XDG_CONFIG_HOME.
|
|
The function execute_command_to_string had two header comments, one in
gdbcmd.h and one in top.c.
This commit merges the two comments into one and places this comment
in gdbcmd.h. The comment in top.c is updated to just reference
gdbcmd.h.
gdb/ChangeLog:
* gdbcmd.h (execute_command_to_string): Update comment.
* top.c (execute_command_to_string): Update header comment.
|
|
When running test-case gdb.base/longjmp.exp with target board unix/-m32, we
run into:
...
(gdb) next^M
Warning:^M
Cannot insert breakpoint 0.^M
Cannot access memory at address 0x7dbf7353^M
^M
__libc_siglongjmp (env=0x804a040 <env>, val=1) at longjmp.c:28^M
28 longjmps++;^M
(gdb) FAIL: gdb.base/longjmp.exp: next over longjmp(1)
...
The failure to access memory happens in i386_get_longjmp_target and is due to
glibc having pointer encryption (aka "pointer mangling" or "pointer guard") of
the long jump buffer. This is a known problem.
In create_longjmp_master_breakpoint (which attempts to install a master
longjmp breakpoint) a preference scheme is present, which installs a
probe breakpoint if a libc:longjmp probe is present, and otherwise falls back
to setting breakpoints at the names in the longjmp_names array.
But in fact, both the probe breakpoint and the longjmp_names breakpoints are
set. The latter ones are set when processing libc.so.debug, and the former
one when processing libc.so. In other words, this is the longjmp variant of
PR26881, which describes the same problem for master exception breakpoints.
This problem only triggers when the glibc debug info package is installed,
which is not due to the debug info itself in libc.so.debug, but due to the
minimal symbols (because create_longjmp_master_breakpoint uses minimal symbols
to translate the longjmp_names to addresses).
The problem doesn't trigger for -m64, because there tdep->jb_pc_offset is not
set.
Fix this similar to commit 1940319c0ef (the fix for PR26881): only install
longjmp_names breakpoints in libc.so/libc.so.debug if installing the
libc:longjmp probe in libc.so failed.
Tested on x86_64-linux.
gdb/ChangeLog:
2021-01-28 Tom de Vries <tdevries@suse.de>
PR breakpoints/27205
* breakpoint.c (create_longjmp_master_breakpoint_probe)
(create_longjmp_master_breakpoint_names): New function, factored out
of ...
(create_longjmp_master_breakpoint): ... here. Only try to install
longjmp_names breakpoints in libc.so/libc.so.debug if installing probe
breakpoint in libc.so failed.
|
|
When running test-case gdb.ada/out_of_line_in_inlined.exp with target board
unix/-m32 on a system with gcc-10 default compiler, we run into:
...
(gdb) break foo_o224_021.child1.child2^M
Breakpoint 1 at 0x804ba59: foo_o224_021.child1.child2. (3 locations)^M
(gdb) FAIL: gdb.ada/out_of_line_in_inlined.exp: scenario=all: \
break foo_o224_021.child1.child2
...
The test does not expect the "3 locations" part.
Fix this by using gdb_breakpoint instead of gdb_test.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2021-01-28 Tom de Vries <tdevries@suse.de>
* gdb.ada/out_of_line_in_inlined.exp: Use gdb_breakpoint.
|
|
When running test-case gdb.dwarf2/dw2-out-of-range-end-of-seq.exp on a
system with debug packages installed, I run into:
...
(gdb) maint info line-table^M
... <lots of output> ...
ERROR: internal buffer is full.
UNRESOLVED: gdb.dwarf2/dw2-out-of-range-end-of-seq.exp: \
END with address 1 eliminated
...
Fix this by limiting the output of the command using a regexp.
I also noticed that when making the regexp match nothing, meaning
the command has no output, the test didn't FAIL. Fixed this by adding a
PASS pattern.
I also noticed that the FAIL pattern didn't work with -m32, fixed that as
well.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2021-01-28 Tom de Vries <tdevries@suse.de>
* gdb.dwarf2/dw2-out-of-range-end-of-seq.exp: Add regexp to
"maint info line-table". Make PASS pattern more specific. Make
FAIL pattern work for -m32.
|
|
This patch addresses PR gdb/27133. Before it, the following succession
of commands would cause gdb to crash:
set logging redirect on
set logging debugredirect on
set logging on
The problem eventually comes down to a use after free. The function
cli_interp_base::set_logging is called with a unique_ptr argument that
holds a pointer to the redirection file. In the problematic use case,
no-one ever took ownership of that pointer (as far as unique_ptr is
concerned), so the call to its dtor at the end of the function causes
the file object to be deleted. Any later use of the pointer to the
redirection file is therefore an error.
This patch ensures that the unique_ptr is released when required (so it
does not assume ownership anymore). The internal logic of
cli_interp_base::set_logging takes care of freeing the ui_file when it
is not necessary anymore using the saved_output.file_to_delete field.
gdb/ChangeLog:
PR gdb/27133
* cli/cli-interp.c (cli_interp_base::set_logging): Ensure the
unique_ptr is released when the wrapped pointer is kept for later
use.
gdb/testsuite/ChangeLog:
PR gdb/27133
* gdb.base/ui-redirect.exp: Add test case that ensures that
redirecting both logging and debug does not cause gdb to crash.
|
|
Enable displaced stepping over a BR/BLR instruction
Displaced stepping over an instruction executes a instruction in a
scratch area and then manually fixes up the PC address to leave
execution where it would have been if the instruction were in its
original location.
The BR instruction does not need modification in order to run correctly
at a different address, but the displaced step fixup method should not
manually adjust the PC since the BR instruction sets that value already.
The BLR instruction should also avoid such a fixup, but must also have
the link register modified to point to just after the original code
location rather than back to the scratch location.
This patch adds the above functionality.
We add this functionality by modifying aarch64_displaced_step_others
rather than by adding a new visitor method to aarch64_insn_visitor.
We choose this since it seems that visitor approach is designed
specifically for PC relative instructions (which must always be modified
when executed in a different location).
It seems that the BR and BLR instructions are more like the RET
instruction which is already handled specially in
aarch64_displaced_step_others.
This also means the gdbserver code to relocate an instruction when
creating a fast tracepoint does not need to be modified, since nothing
special is needed for the BR and BLR instructions there.
Regression tests showed nothing untoward on native aarch64 (though it
took a while for me to get the testcase to account for PIE).
------#####
Original observed (mis)behaviour before was that displaced stepping over
a BR or BLR instruction would not execute the function they called.
Most easily seen by putting a breakpoint with a condition on such an
instruction and a print statement in the functions they called.
When run with the breakpoint enabled the function is not called and
"numargs called" is not printed.
When run with the breakpoint disabled the function is called and the
message is printed.
--- GDB Session
~ [15:57:14] % gdb ../using-blr
Reading symbols from ../using-blr...done.
(gdb) disassemble blr_call_value
Dump of assembler code for function blr_call_value:
...
0x0000000000400560 <+28>: blr x2
...
0x00000000004005b8 <+116>: ret
End of assembler dump.
(gdb) break *0x0000000000400560
Breakpoint 1 at 0x400560: file ../using-blr.c, line 22.
(gdb) condition 1 10 == 0
(gdb) run
Starting program: /home/matmal01/using-blr
[Inferior 1 (process 33279) exited with code 012]
(gdb) disable 1
(gdb) run
Starting program: /home/matmal01/using-blr
numargs called
[Inferior 1 (process 33289) exited with code 012]
(gdb)
Test program:
---- using-blr ----
\#include <stdio.h>
typedef int (foo) (int, int);
typedef void (bar) (int, int);
struct sls_testclass {
foo *x;
bar *y;
int left;
int right;
};
__attribute__ ((noinline))
int blr_call_value (struct sls_testclass x)
{
int retval = x.x(x.left, x.right);
if (retval % 10)
return 100;
return 9;
}
__attribute__ ((noinline))
int blr_call (struct sls_testclass x)
{
x.y(x.left, x.right);
if (x.left % 10)
return 100;
return 9;
}
int
numargs (__attribute__ ((unused)) int left, __attribute__ ((unused)) int right)
{
printf("numargs called\n");
return 10;
}
void
altfunc (__attribute__ ((unused)) int left, __attribute__ ((unused)) int right)
{
printf("altfunc called\n");
}
int main(int argc, char **argv)
{
struct sls_testclass x = { .x = numargs, .y = altfunc, .left = 1, .right = 2 };
if (argc > 2)
{
blr_call (x);
}
else
blr_call_value (x);
return 10;
}
|
|
Internally at AdaCore, documentation is still built with Texinfo 4.13.
This version gave an error when building gdb.texinfo:
../../../binutils-gdb/gdb/doc/gdb.texinfo:27672: @pxref expected braces.
../../../binutils-gdb/gdb/doc/gdb.texinfo:27672: ` {dotdebug_gdb_scripts section,,The @cod...' is too long for expansion; not expanded.
... followed by many more spurious errors that were caused by this
one.
This patch fix the problem by removing the extra space.
I don't know whether it's advisable to try to support this ancient
version of Texinfo (released in 2008 apparently); but in this
particular case the fix is trivial, so I'm checking it in.
gdb/doc/ChangeLog
2021-01-27 Tom Tromey <tromey@adacore.com>
* gdb.texinfo (Auto-loading extensions): Remove extraneous space.
|
|
This changes windows-nat.c and nat/windows-nat.c to use the new
debug_prefixed_printf_cond facility. I tried this out on a Windows
build and I think it makes the output look a little nicer.
2021-01-26 Tom Tromey <tromey@adacore.com>
* windows-nat.c (DEBUG_EXEC, DEBUG_EVENTS, DEBUG_MEM)
(DEBUG_EXCEPT): Use debug_prefixed_printf_cond.
(windows_init_thread_list, windows_nat::handle_load_dll)
(windows_nat::handle_unload_dll, windows_nat_target::resume)
(windows_nat_target::resume)
(windows_nat_target::get_windows_debug_event)
(windows_nat_target::interrupt, windows_xfer_memory)
(windows_nat_target::close): Update.
* nat/windows-nat.c (DEBUG_EVENTS): Use
debug_prefixed_printf_cond.
(matching_pending_stop, fetch_pending_stop)
(continue_last_debug_event): Update.
|
