Tom Tromey [Wed, 8 Feb 2023 20:59:36 +0000 (13:59 -0700)]
Fix value chain use-after-free
Hannes filed a bug showing a crash, where a pretty-printer written in
Python could cause a use-after-free. He sent a patch, but I thought a
different approach was needed.
In a much earlier patch (see bug #12533), we changed the Python code
to release new values from the value chain when constructing a
gdb.Value. The rationale for this is that if you write a command that
does a lot of computations in a loop, all the values will be kept live
by the value chain, resulting in gdb using a large amount of memory.
However, suppose a value is passed to Python from some code in gdb
that needs to use the value after the call into Python. In this
scenario, value_to_value_object will still release the value -- and
because gdb code doesn't generally keep strong references to values (a
consequence of the ancient decision to use the value chain to avoid
memory management), this will result in a use-after-free.
This scenario can happen, as it turns out, when a value is passed to
Python for pretty-printing. Now, normally this route boxes the value
via value_to_value_object_no_release, avoiding the problematic release
from the value chain. However, if you then call Value.cast, the
underlying value API might return the same value, when is then
released from the chain.
This patch fixes the problem by changing how value boxing is done.
value_to_value_object no longer removes a value from the chain.
Instead, every spot in gdb that might construct new values uses a
scoped_value_mark to ensure that the requirements of bug #12533 are
met. And, because incoming values aren't ever released from the chain
(the Value.cast one comes earlier on the chain than the
scoped_value_mark), the bug can no longer occur. (Note that many
spots in the Python layer already take this approach, so not many
places needed to be touched.)
In the future I think we should replace the use of raw "value *" with
value_ref_ptr pretty much everywhere. This will ensure lifetime
safety throughout gdb.
The test case in this patch comes from Hannes' original patch. I only
made a trivial ("require") change to it. However, while this fails
for him, I can't make it fail on this machine; nevertheless, he tried
my patch and reported the bug as being fixed.
Pedro Alves [Fri, 25 Nov 2022 16:20:22 +0000 (16:20 +0000)]
all-stop "follow-fork parent" and selecting another thread
With:
- catch a fork in thread 1
- select thread 2
- set follow-fork child
- next
... follow_fork notices that thread 1 had last stopped for a fork
which hasn't been followed yet, and because thread 1 is not the
current thread, GDB aborts the execution command, presenting the stop
in thread 1.
That makes sense, as only the forking thread (thread 1) survives in
the child, so better stop and let the user decide how to proceed.
However, with:
- catch a fork in thread 1
- select thread 2
- set follow-fork parent << note difference here
- next
... GDB does the same: follow_fork notices that thread 1 had last
stopped for a fork which hasn't been followed yet, and because thread
1 is not the current thread, GDB aborts the execution command,
presenting the stop in thread 1.
Aborting/stopping in this case doesn't make sense to me. As we're
following the parent, thread 2 will still continue to exist in the
parent. What the child does after we've followed the parent shouldn't
matter -- it can go on running free, be detached, etc., depending on
"set schedule-multiple", "set detach-on-fork", etc. That does not
influence the execution command that the user issued for the parent
thread.
So this patch changes GDB in that direction -- in follow_fork, if
following the parent, and we've switched threads meanwhile, switch
back to the unfollowed thread, follow it (stay with the parent), and
don't abort/stop. If we're following a fork (as opposed to vfork),
then switch back again to the thread that the user was trying to
resume. If following a vfork, however, stay with the vforking-thread
selected, as we will need to see a vfork_done event first, before we
can resume any other thread.
As I was working on this, I managed to end up calling target_resume
for a solo-thread resume (to collect the vfork_done event), with
scope_ptid pointing at the vfork parent thread, and inferior_ptid
pointing to the vfork child. For a solo-thread resume, the scope_ptid
argument to target_resume must the same as inferior_ptid. The mistake
was caught by the assertion in target_resume, like so:
Pedro Alves [Thu, 17 Nov 2022 18:25:36 +0000 (18:25 +0000)]
Make follow_fork not rely on get_last_target_status
Currently, if
- you're in all-stop mode,
- the inferior last stopped because of a fork catchpoint,
when you next resume the program, gdb checks whether it had last
stopped for a fork/vfork, and if so,
a) if the current thread is the one that forked, gdb follows the
parent/child, depending on "set follow-fork" mode.
b) if the current thread is some other thread (because you switched
threads meanwhile), gdb switches back to that thread, gdb follows
the parent/child, and stops the resumption command.
There's a problem in b), however -- if you have "set schedule-multiple
off", which is the default, or "set scheduler-locking on", gdb will
still switch back to the forking thread, even if you didn't want to
resume it. For example, with:
(gdb) catch fork
(gdb) c
* thread 1 stops for fork
(gdb) thread 2
(gdb) set scheduler-locking on
(gdb) c
gdb switches back to thread 1, and follows the fork.
Or with:
(gdb) add-inferior -exec prog
(gdb) inferior 2
(gdb) start
(gdb) inferior 1
(gdb) catch fork
(gdb) c
* thread 1.1 stops for fork
(gdb) inferior 2
(gdb) set schedule-multiple off # this is the default
(gdb) c
gdb switches back to thread 1.1, and follows the fork.
Another issue is that, because follow_fork relies on
get_last_target_status to find the thread that has a pending fork, it
is possible to confuse it. For example, "run" or "start" call
init_wait_for_inferior, which clears the last target status, so this:
(gdb) catch fork
(gdb) c
* thread 1 stops for fork
(gdb) add-inferior -exec prog
(gdb) inferior 2
(gdb) start
(gdb) set follow-fork child
(gdb) inferior 1
(gdb) n
... does not follow to the fork child of inferior 1, because the
get_last_target_status call in follow_fork doesn't return a
TARGET_WAITKIND_FORKED. Thanks to Simon for this example.
All of the above are fixed by this patch. It changes follow_fork to
not look at get_last_target_status, but to instead iterate over the
set of threads that the user is resuming, and find the one that has a
pending_follow kind of fork/vfork.
gdb.base/foll-fork.exp is augmented to exercise the last "start"
scenario described above. The other cases will be exercised in the
testcase added by the following patch.
Pedro Alves [Thu, 17 Nov 2022 23:21:04 +0000 (23:21 +0000)]
Improve "info program"
With gdb.base/catch-follow-exec.exp, we currently see:
~~~~~~~~~~~~~~~
(gdb)
continue
Continuing.
process 693251 is executing new program: /usr/bin/ls
[New inferior 2]
[New process 693251]
[Switching to process 693251]
Thread 2.1 "ls" hit Catchpoint 2 (exec'd /usr/bin/ls), 0x00007ffff7fd0100 in _start () from /lib64/ld-linux-x86-64.so.2
(gdb)
info prog
No selected thread.
~~~~~~~~~~~~~~~
Note the "No selected thread" output. That is totally bogus, because
there _is_ a selected thread. What GDB really means, is that it can't
find the thread that had the latest (user-visible) stop. And that
happens because "info program" gets that info from
get_last_target_status, and the last target status has been cleared.
However, GDB also checks if there is a selected thread, here:
.. the null_ptid part. That is also bogus, because what matters is
the thread that last reported a stop, not the current thread:
- in all-stop mode, "info program" displays info about the last stop.
That may have happened on a thread different from the selected
thread.
- in non-stop mode, because all threads are controlled individually,
"info program" shows info about the last stop of the selected
thread.
The current code already behaves this way, though in a poor way. This
patch reimplements it, such that the all-stop version now finds the
thread that last reported an event via the 'previous_thread' strong
reference. Being a strong reference means that if that thread has
exited since the event was reported, 'previous_thread' will still
point to it, so we can say that the thread exited meanwhile.
The patch also extends "info program" output a little, to let the user
know which thread we are printing info for. For example, for the
gdb.base/catch-follow-exec.exp case we shown above, we now get:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Using the running image of child process 710867.
Program stopped at 0x7ffff7fd0100.
It stopped at breakpoint 2.
Type "info stack" or "info registers" for more information.
(gdb)
while in non-stop mode, we get:
(gdb) info prog
Selected thread 2.1 (process 710867).
Using the running image of child process 710867.
Program stopped at 0x7ffff7fd0100.
It stopped at breakpoint 2.
Type "info stack" or "info registers" for more information.
(gdb)
In both cases, the first line of output is new.
The existing code considered these running/exited cases as an error,
but I think that that's incorrect, since this is IMO just plain
execution info as well. So the patch makes those cases regular
prints, not errors.
If the thread is running, we get, in non-stop mode:
(gdb) info prog
Selected thread 2.1 (process 710867).
Selected thread is running.
... and in all-stop:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Thread is now running.
If the thread has exited, we get, in non-stop mode:
(gdb) info prog
Selected thread 2.1 (process 710867).
Selected thread has exited.
... and in all-stop:
(gdb) info prog
Last stopped for thread 2.1 (process 710867).
Thread has since exited.
The gdb.base/info-program.exp testcase was much extended to test
all-stop/non-stop and single-threaded/multi-threaded.
Pedro Alves [Thu, 17 Nov 2022 23:08:58 +0000 (23:08 +0000)]
Convert previous_inferior_ptid to strong reference to thread_info
I originally wrote this patch, because while working on some other
patch, I spotted a regression in the
gdb.multi/multi-target-no-resumed.exp.exp testcase. Debugging the
issue, I realized that the problem was related to how I was using
previous_inferior_ptid to look up the thread the user had last
selected. The problem is that previous_inferior_ptid alone doesn't
tell you which target that ptid is from, and I was just always using
the current target, which was incorrect. Two different targets may
have threads with the same ptid.
I decided to fix this by replacing previous_inferior_ptid with a
strong reference to the thread, called previous_thread.
I have since found a new motivation for this change -- I would like to
tweak "info program" to not rely on get_last_target_status returning a
ptid that still exists in the thread list. With both the follow_fork
changes later in this series, and the step-over-thread-exit changes,
that can happen, as we'll delete threads and not clear the last
waitstatus.
A new update_previous_thread function is added that can be used to
update previous_thread from inferior_ptid. This must be called in
several places that really want to get rid of previous_thread thread,
and reset the thread id counter, otherwise we get regressions like
these:
(gdb) info threads -gid
Id GId Target Id Frame
- * 1 1 Thread 2974541.2974541 "tids-gid-reset" main () at src/gdb/testsuite/gdb.multi/tids-gid-reset.c:21
- (gdb) PASS: gdb.multi/tids-gid-reset.exp: single-inferior: after restart: info threads -gid
+ * 1 2 Thread 2958361.2958361 "tids-gid-reset" main () at src/gdb/testsuite/gdb.multi/tids-gid-reset.c:21
+ (gdb) FAIL: gdb.multi/tids-gid-reset.exp: single-inferior: after restart: info threads -gid
and:
Core was generated by `build/gdb/testsuite/outputs/gdb.reverse/sigall-precsave/si'.
Program terminated with signal SIGTRAP, Trace/breakpoint trap.
#0 gen_ABRT () at src/gdb/testsuite/gdb.reverse/sigall-reverse.c:398
398 kill (getpid (), SIGABRT);
+[Current thread is 1 (LWP 2662066)]
Restored records from core file build/gdb/testsuite/outputs/gdb.reverse/sigall-precsave/sigall.precsave.
#0 gen_ABRT () at src/gdb/testsuite/gdb.reverse/sigall-reverse.c:398
398 kill (getpid (), SIGABRT);
continue
Continuing.
-Program received signal SIGABRT, Aborted.
+Thread 1 received signal SIGABRT, Aborted.
0x00007ffff7dfd55b in kill () at ../sysdeps/unix/syscall-template.S:78
78 ../sysdeps/unix/syscall-template.S: No such file or directory.
-(gdb) PASS: gdb.reverse/sigall-precsave.exp: sig-test-1: get signal ABRT
+(gdb) FAIL: gdb.reverse/sigall-precsave.exp: sig-test-1: get signal ABRT
I.e., GDB was failing to restart the thread counter back to 1, because
the previous_thread thread was being help due to the strong reference.
Tested on GNU/Linux native, gdbserver and gdbserver + "maint set
target-non-stop on".
gdb/ChangeLog:
yyyy-mm-dd Pedro Alves <pedro@palves.net>
Pedro Alves [Sat, 3 Dec 2022 16:20:51 +0000 (16:20 +0000)]
Tweak "Using the running image of ..." output
Currently, "info files" and "info program" on a few native targets
show:
(gdb) info files
Symbols from "/home/pedro/gdb/tests/threads".
Native process:
Using the running image of child Thread 0x7ffff7d89740 (LWP 1097968).
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
(gdb) info program
Using the running image of child Thread 0x7ffff7d89740 (LWP 1097968).
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Program stopped at 0x555555555278.
...
This patch changes them to:
(gdb) info files
Symbols from "/home/pedro/gdb/tests/threads".
Native process:
Using the running image of child process 1097968.
^^^^^^^^^^^^^^^
...
(gdb) info program
Using the running image of child process 1097968.
^^^^^^^^^^^^^^^
Program stopped at 0x555555555278.
...
... which I think makes a lot more sense in this context. The "info
program" manual entry even says:
"Display information about the status of your program: whether it is
running or not, what process it is, and why it stopped."
^^^^^^^^^^^^^
This change affects ptrace targets, procfs targets, and Windows.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Change-Id: I6aab061ff494a84ba3398cf98fd49efd7a6ec1ca
Simon Marchi [Mon, 27 Feb 2023 01:14:02 +0000 (20:14 -0500)]
gdb: make-target-delegates.py: add Entry type
Add the Entry type and use it in the `entries` map, rather than using an
ad-hoc str -> str map that comes from the re.match. This will make it
easier to make typing work in a subsequent patch, but it also helps
readers know what attributes exist for entries, which is not clear
currently.
Change-Id: I5b58dee1ed7ae85987b99bd417e641ede718624c Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:14:00 +0000 (20:14 -0500)]
gdb: gdbarch*.py, copyright.py: add type annotations
Add type annotations to gdbarch*.py to fix all errors shown by pyright.
There is one change in copyright.py too, to fix this one:
/home/simark/src/binutils-gdb/gdb/gdbarch.py
/home/simark/src/binutils-gdb/gdb/gdbarch.py:206:13 - error: Type of "copyright" is partially unknown
Type of "copyright" is "(tool: Unknown, description: Unknown) -> str" (reportUnknownMemberType)
Change-Id: Ia109b53e267f6e2f5bd79a1288d0d5c9508c9ac4 Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:13:59 +0000 (20:13 -0500)]
gdb: split gdbarch component types to gdbarch_types.py
Editing gdbarch-components.py is not an experience in an editor that is
minimally smart about Python. Because gdbarch-components.py is read and
exec'd by gdbarch.py, it doesn't import the Info / Method / Function /
Value types. And because these types are defined in gdbarch.py, it
can't import them, as that would make a cyclic dependency.
Solve this by introducing a third file, gdbarch_types.py, to define
these types. Make gdbarch.py and gdbarch-components.py import it.
Also, replace the read & exec of gdbarch-components.py by a regular
import. For this to work though, gdbarch-components.py needs to be
renamed to gdbarch_components.py.
Change-Id: Ibe994d56ef9efcc0698b3ca9670d4d9bf8bbb853 Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:13:58 +0000 (20:13 -0500)]
gdb: pyproject.toml: set pyright typeCheckingMode = "strict"
While working on other projects, I found the pyright type checker very
helpful when editing Python code. I don't think I have to explain the
advantages of type checking to a crowd used to C/C++.
Setting typeCheckingMode to "strict" makes pyright flag a bit more type
issues than the default of "basic".
Change-Id: I38818ec59f7f73c2ab020cc9226286cdd485abc7 Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:13:57 +0000 (20:13 -0500)]
gdb: gdbarch.py: remove Info.__init__
Info.__init__ currently assigns `self.predicate = None`. This was
helpful to ensure that all component types had a `predicate` attribute.
The generator code could then avoid having code like "if the component
is anything but Info, use predicate". Since the previous commit, all
component types have a predicate attribute which defaults to False. We
can therefore remove the assignment in Info.__init__, and in turn remove
Info.__init__. We however need to make the printer parameter of
_Component.__init__ optional, as Info don't need a printer.
Change-Id: I611edeca9cc9837eb49dddfe038595e1ff3b7239 Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:13:56 +0000 (20:13 -0500)]
gdb: gdbarch.py: spell out parameters of _Component.__init__
The way _Component uses kwargs is handy to save a few characters, but it
doesn't play well with static analysis. When editing gdbarch.py, my
editor (which uses pylance under the hood) knows nothing about the
properties of components. So it's full of squiggly lines, and typing
analysis (which I find really helpful) doesn't work. I therefore think
it would be better to spell out the parameters.
Change-Id: Iaf561beb0d0fbe170ce1c79252a291e0945e1830 Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Simon Marchi [Mon, 27 Feb 2023 01:13:54 +0000 (20:13 -0500)]
gdb: remove invalid / dead code from gdbarch.py
My editor flagged that the variable `c` (in the lines removed by this
patch) was unknown. I guess it ends up working because there is a `c`
variable in the global scope. I tried putting `assert False` inside
that if, and it is not hit, showing that we never enter this if. So,
remove it. There is no change in the generated files.
Change-Id: Id3b9f67719e88cada7c6fde673c8d7842ab13617 Reviewed-By: Tom Tromey <tom@tromey.com> Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Tom Tromey [Thu, 9 Feb 2023 19:12:42 +0000 (12:12 -0700)]
Fix crash with "finish" in Rust
PR rust/30090 points out that a certain "finish" in a Rust program
will cause gdb to crash. This happens due to some confusion about
field indices in rust_language::print_enum. The fix is to use
value_primitive_field so that the correct type can be passed; other
spots in rust-lang.c already do this.
Note that the enclosed test case comes with an xfail. This is needed
because for this function, rustc doesn't follow the platform ABI.
Tom Tromey [Wed, 4 Jan 2023 16:36:59 +0000 (09:36 -0700)]
Handle range types in ax-gdb.c
A range type can usually be treated the same as its underlying integer
type, at least for the purposes of agent expressions. This patch
arranges for range types to be handled this way in ax-gdb.c, letting a
somewhat larger subset of Ada expressions be compiled.
Tom Tromey [Wed, 21 Dec 2022 18:52:37 +0000 (11:52 -0700)]
Implement some agent expressions for Ada
Ada historically has not implemented agent expressions, and some Ada
constructs probably cannot reasonably be converted to agent
expressions. However, a subset of simple operations can be, and this
patch represents a first step in that direction.
On one internal AdaCore test case, this improves the performance of a
conditional breakpoint from 5 minutes to 5 seconds.
The main tricky part in this patch is ensuring the converted
expressions detect the cases that will not work. This is done by
examining the code in the corresponding evaluation methods.
$ cd gdb/syscalls/
$ ./update-linux-defaults.sh ~/strace.git/
Generating linux-defaults.xml.in
$ make
for f in aarch64-linux.xml amd64-linux.xml arm-linux.xml bfin-linux.xml \
i386-linux.xml mips-n32-linux.xml mips-n64-linux.xml \
mips-o32-linux.xml ppc64-linux.xml ppc-linux.xml s390-linux.xml \
s390x-linux.xml sparc64-linux.xml sparc-linux.xml; do \
xsltproc --output $f apply-defaults.xsl $f.in; \
done
The result is that a lot more syscalls end up assigned to groups.
Some lose their group info, but that just mirrors what strace does.
The gdb/syscalls/linux-defaults.xml.in file shows a large diff because
the new version is ASCII sorted, while the current version was
somewhat (but not consistently) sorted by "family" of syscalls.
If I sort the old file and diff against the new, the difference is
like this:
Pedro Alves [Fri, 24 Feb 2023 18:58:31 +0000 (18:58 +0000)]
Autogenerate gdb/syscalls/linux-defaults.xml.in (groups) from strace sources
I noticed that "catch syscall group:process" doesn't catch clone3,
while it does catch clone.
The catch syscall group information is recorded in the
gdb/syscalls/linux-defaults.xml.in file, which says:
<!-- The group field information was based on strace. -->
So I looked at the strace sources, to confirm that clone3 is in fact
recorded in the "process" group there too, and to check what other
syscalls might be missing groups.
After some digging, I found that strace records the group info in C
arrays, with entries like:
...
[ 61] = { 4, TP, SEN(wait4), "wait4" },
[ 62] = { 2, TS|TP, SEN(kill), "kill" },
[ 63] = { 1, 0, SEN(uname), "uname" },
...
You can see the current master's table for Linux x86-64 here:
The column with TS|TP above is what defines each syscall's groups. So
I wrote a script that extracts this information and generates
linux-defaults.xml.in.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Change-Id: I679d59d42fb2a914bf7a99e4c558e9696e5adff1
Clément Chigot [Mon, 27 Feb 2023 12:57:12 +0000 (13:57 +0100)]
gas/testsuite: adjust another test for case insensitive file systems
As 1fafeaac8503eea2f61c3a35f0eef183b7e7cc65, "line.s" and "Line.s" are
identical in case insensitive file systems. Thus, gas doesn't trigger
an input file switch.
gas/ChangeLog:
* testsuite/gas/elf/dwarf-5-macro.s: Change Line.s to Line2.s.
Clearly not great. This is described in PR gdb/30148.
The problem here is in dwarf2/read.c, we assume all enums are flag
enums unless we find an enumerator with a non-flag like value.
Clearly an empty enum contains no non-flag values, so we assume the
enum is a flag enum.
I propose adding an extra check here; that is, an empty enum should
never be a flag enum.
Bruno Larsen [Tue, 21 Feb 2023 16:10:31 +0000 (17:10 +0100)]
gdb/testsuite: Improve testing of GDB's completion functions
When looking at some failures of gdb.linespec/cp-completion-aliases.exp,
I noticed that when a completion test will fail, it always fails with a
timeout. This is because most completion tests use gdb_test_multiple
and only add a check for the correct output. This commit adds new
options for both, tab and command completion.
For command completion, the new option will check if the prompt was
printed, and fail in this case. This is enough to know that the test has
failed because the check comes after the PASS path. For tab completion,
we have to check if GDB outputted more than just the input line, because
sometimes GDB would have printed a partial line before finishing with
the correct completion.
Tom Tromey [Fri, 16 Dec 2022 19:50:29 +0000 (12:50 -0700)]
Remove struct buffer
I've long wanted to remove 'struct buffer', and thanks to Simon's
earlier patch, I was finally able to do so. My feeling has been that
gdb already has several decent structures available for growing
strings: std::string of course, but also obstack and even objalloc
from BFD and dyn-string from libiberty. The previous patches in this
series removed all the uses of struct buffer, so this one can remove
the code and the remaining #includes.
Tom Tromey [Fri, 16 Dec 2022 19:53:20 +0000 (12:53 -0700)]
Don't use struct buffer in top.c
This changes top.c to use std::string rather than struct buffer. Like
the event-top.c change, this is not completely ideal in that it
requires a copy of the string.
Tom Tromey [Fri, 16 Dec 2022 19:45:40 +0000 (12:45 -0700)]
Don't use struct buffer in event-top.c
This changes event-top.c to use std::string rather than struct buffer.
This isn't completely ideal, in that it requires a copy of the string
to be made.
Tom Tromey [Wed, 13 Apr 2022 17:25:53 +0000 (11:25 -0600)]
Write the DWARF index in the background
The new DWARF cooked indexer interacts poorly with the DWARF index
cache. In particular, the cache will require gdb to wait for the
cooked index to be finalized. As this happens in the foreground, it
means that users with this setting enabled will see a slowdown.
This patch changes gdb to write the cache entry a worker thread. (As
usual, in the absence of threads, this work is simply done immediately
in the main thread.)
Some care is taken to ensure that this can't crash, and that gdb will
not exit before the task is complete.
To avoid use-after-free problems, the DWARF per-BFD object explicitly
waits for the index cache task to complete.
To avoid gdb exiting early, an exit observer is used to wait for all
such pending tasks.
In normal use, neither of these waits will be very visible. For users
using "-batch" to pre-generate the index, though, it would be.
However I don't think there is much to be done about this, as it was
the status quo ante.
Tom Tromey [Wed, 13 Apr 2022 17:21:20 +0000 (11:21 -0600)]
Only use the per-BFD object to write a DWARF index
The DWARF index does not need access to the objfile or per-objfile
objects when writing -- it's entirely based on the objfile-independent
per-BFD data.
This patch implements this idea by changing the entire API to only be
passed the per-BFD object. This simplifies some lifetime reasoning
for the next patch.
This patch removes some code that ensures that the BFD came from a
file. It seems to me that checking for the existence of a build-id is
good enough for the index cache.
Clément Chigot [Tue, 21 Feb 2023 14:07:58 +0000 (15:07 +0100)]
testsuite: prune DOS drive letter in test outputs
On DOS systems, absolute paths start with the drive letter. This can
trigger failures in the regexp from dump tests, especially for those
checking for warnings or errors. They are usually skipping everything
before the first ":" as it has to be the file path.
| [^:]*: warning: ...
In order to avoid modifying many regexps to allow such drive letters,
prune them from all the outputs if they are found at the beginning of
a line.
binutils/ChangeLog:
* testsuite/lib/binutils-common.exp (prune_dump_output): New
(run_dump_test): Use it.
ld/ChangeLog:
* testsuite/ld-elf/noinit-sections-2.l: Remove DOS drive letter
handler.
Jan Beulich [Fri, 24 Feb 2023 13:00:11 +0000 (14:00 +0100)]
x86: allow to request ModR/M encoding
Several insns have a (typically shorter) non-ModR/M and a (typically
longer) ModR/M encoding. In most cases the former is used by default.
This isn't too dissimilar from register-only insns sometimes having two
encoding forms. In those cases {load} or {store} can be used to control
the encoding used. Extend this to ModR/M-less encodings which have a
ModR/M counterpart (note that BSWAP hasn't). For insn reading and
writing their (explicit) memory operand, both prefixes are honored;
otherwise only the applicable one is.
Note that for some forms of XCHG, {store} has already been performing
this function, apparently as an unnoticed side effect of adding D to
the template.
Jan Beulich [Fri, 24 Feb 2023 12:58:35 +0000 (13:58 +0100)]
x86-64: don't permit LAHF/SAHF with "generic64"
The feature isn't universally available on 64-bit CPUs.
Note that in i386-gen.c:isa_dependencies[] I'm only adding it to models
where I'm certain the functionality exists. For Nocona and Core I'm
uncertain in particular.
Jan Beulich [Fri, 24 Feb 2023 12:57:31 +0000 (13:57 +0100)]
x86: have insns acting on segment selector values allow for consistent operands
While MOV to/from segment register as well as selector storing insns
already permit 32- and 64-bit GPR operands, selector loading insns and
ARPL do not. Split templates accordingly.
For shifts (but not ordinary rotates) and other cases where an immediate
describes e.g. a bit count or position, allowing negative operands is at
best confusing. An extreme example would be the two rotate-through-carry
insns, where a negative value would _not_ mean rotating the
corresponding number of bits in the other direction. To refuse such,
give meaning to the combination of Imm8 and Imm8S in templates (so far
these weren't used together anywhere). The issue was with
smallest_imm_type() blindly setting .imm8 for signed numbers determined
to fit in a byte.
VPROT{B,W,D,Q} is a little special: The rotate count there is a signed
quantity, so Imm8 is replaced by Imm8S. Adjust affected testcases
accordingly as well.
Another small adjustment to the testsuite is necessary: AAM and AAD were
never sensible to use with 0xffffff90 operands. This should have been an
error.
Tom de Vries [Fri, 24 Feb 2023 12:52:12 +0000 (13:52 +0100)]
[gdb/testsuite] Cleanup unnecessary expr from require line
In a recent commit I've added:
...
require {expr [have_compile_flag -fsplit-stack]}
...
but actually the expr bit is unnecessary, and we can just use:
...
require {have_compile_flag -fsplit-stack}
...
GDB: Fix out of bounds accesses with limited-length values
Fix accesses to limited-length values in `contents_copy_raw' and
`contents_copy_raw_bitwise' so that they observe the limit of the
original allocation.
Reported by Simon Marchi as a heap-buffer-overflow AddressSanitizer
issue triggered with gdb.ada/limited-length.exp.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Alan Modra [Thu, 23 Feb 2023 07:53:12 +0000 (18:23 +1030)]
PR30155, ld segfault in _bfd_nearby_section
The segfault was a symptom of messing with the absolute section next
field, confusing bfd_section_removed_from_list in linker.c:fix_syms.
That's not all that was going wrong. The INSERT list of output
sections was being inserted into itself, ie. lost from the main
list of linker statements.
PR 30155
* ldlang.c (process_insert_statements): Handle pathological
case of the insert script being inserted before the first
output section statement in the default script.
(output_prev_sec_find): Don't test section owner here.
(insert_os_after): Change parameter to a list union pointer.
(lang_insert_orphan): Test section owner here and adjust
insert_os_after call.
Palmer Dabbelt [Thu, 26 Jan 2023 23:09:29 +0000 (15:09 -0800)]
gdb/doc: The RISC-V vector registers didn't change
When we merged the GDB vector register support we did it a bit early,
just eating the risk in the very unlikely case that the vector register
names changed. They didn't, so we can now remove the caveat in the docs
that they might.
Simon Marchi [Tue, 14 Feb 2023 16:13:43 +0000 (11:13 -0500)]
gdb: remove --disable-gdbmi configure option
I noticed that the --disable-gdbmi option was broken for almost a year
(since 740b42ceb7c "gdb/python/mi: create MI commands using python").
The problem today is the python/py-cmd.c file. It is included in the
build if Python support is enabled, and it calls into some MI functions
(e.g. insert_mi_cmd_entry). If MI support is disabled, we get some
undefined symbols like:
The python/py-cmd.c file should be included in the build if both Python
and MI support are enabled. It is not a case we support today, but it
could be done with a bit more configure code. However, I think we
should just remove the --disable-gdbmi option, and just include MI
support unconditionally.
Tom Tromey proposed a while ago to remove this option, but it ended
staying:
However, there was no strong opposition to remove it. The argument was
just "bah, it doesn't hurt anybody".
But given today's case, I would rather remove complexity rather than add
some. I couldn't find anybody caring deeply for that option, and it's
not like MI adds any external dependency. It's just a bit more code.
Removing the option will not break anybody using --disable-gdbmi (it can
be found in many build scripts [1]), since we don't flag invalid
configure flags.
So, remove the option from configure.ac, and adjust Makefile.in
accordingly to always include the MI objects in the build.
Simon Marchi [Thu, 23 Feb 2023 16:27:23 +0000 (11:27 -0500)]
gdb: add AMDGPU header files to HFILES_NO_SRCDIR
Commit 18b4d0736bc5 ("gdb: initial support for ROCm platform (AMDGPU)
debugging") missed adding these header files to the HFILES_NO_SRCDIR
list in the Makefile. Fix that now.
Hui Li [Thu, 16 Feb 2023 00:59:10 +0000 (08:59 +0800)]
gdb: LoongArch: Modify the result of the info reg command
The "info register" command should only display general registers,
but it shows the information of all registers in the current code,
add loongarch_register_reggroup_p() so that we can get the expected
result.
Signed-off-by: Hui Li <lihui@loongson.cn> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Nick Clifton [Thu, 23 Feb 2023 09:44:50 +0000 (09:44 +0000)]
Fix the BFD library's find_nearest_line feature to produce consistent results.
PR 30150
* dwarf2.c (comp_unit_contains_address): Renamed to ... (comp_unit_may_contain_address): this,
and added code to return true if the CU's ranges have not yet been computed.
(_bfd_dwarf2_find_nearest_line_with_alt): Use the renamed function, simplifying code in the process.
Alan Modra [Tue, 21 Feb 2023 21:17:36 +0000 (07:47 +1030)]
Test SEC_HAS_CONTENTS before reading section contents
bfd_malloc_and_get_section does size sanity checking before allocating
memory and reading contents. These size checks are not done for bss
style sections, because they typically don't occupy file space and
thus can't be compared against file size. However, if you are
expecting to look at something other than a whole lot of zeros, don't
allow fuzzers to avoid the size checking.
Pedro Alves [Wed, 22 Feb 2023 15:40:58 +0000 (15:40 +0000)]
gdb.reverse/time-reverse.exp: test both time syscall and C time function
Instead of only testing this on systems that have a SYS_time syscall,
test it everywhere using the time(2) C function, and in addition, run
the tests again using the SYS_time syscall.
The C variant ensures that if some platform uses some syscall we are
not aware of yet, we'll still exercise it, and likely fail, at which
point we should teach GDB about the syscall.
The explicit syscall variant is useful on platforms where the C
function does not call a syscall at all by default, e.g., on some
systems the C time function wraps an implementation provided by the
vDSO.
Approved-By: Tom de Vries <tdevries@suse.de>
Change-Id: Id4b755d76577d02c46b8acbfa249d9c31b587633
Jan Beulich [Wed, 22 Feb 2023 13:12:52 +0000 (14:12 +0100)]
x86-64: LAR and LSL don't need REX.W
Just like we suppress emitting REX.W for e.g. MOV from/to segment
register, there's also no need for it for LAR and LSL - these can only
ever return 32-bit values and hence always zero-extend their results
anyway.
While there also drop the redundant Word from the first operand of
the second template each - this is already implied by Reg16.
Jan Beulich [Wed, 22 Feb 2023 13:12:24 +0000 (14:12 +0100)]
x86: optimize BT{,C,R,S} $imm,%reg
In 64-bit mode BT can have REX.W or a data size prefix dropped in
certain cases. Outside of 64-bit mode all 4 insns can have the data
size prefix dropped in certain cases.
Alan Modra [Tue, 21 Feb 2023 22:34:57 +0000 (09:04 +1030)]
Re: objdump read_section_stabs
Commit f9c36cc99518 changed (and renamed) read_section_stabs with one
difference in overall behaviour. Previously read_section_stabs would
return a NULL for an empty section, which was then treated the same as
a missing section. Now an empty section is recognized and dumped.
This leads to NULL stabp and stabs_end in print_section_stabs. Since
stabs_end - STABSIZE is then a pointer to a very large address, the
test "stabp < stabs_end - STABSIZE" succeeds.
Philippe Blain [Sun, 19 Feb 2023 22:37:35 +0000 (17:37 -0500)]
gdb: add --with-curses to --configuration output
'gdb --configuration' does not mention if GDB was built with curses.
Since b5075fb68d4 (Rename to allow_tui_tests, 2023-01-08) it does show
--enable-tui (or --disable-tui), but one might want to know if GDB was
built with curses independently of the availability of the TUI.
Since configure.ac uses AC_SEARCH_LIBS to check for the curses library,
we do not get an automatically defined HAVE_LIBCURSES symbol in
config.in. We do have symbols defined by AC_CHECK_HEADERS
(HAVE_CURSES_H, etc.) but it would be cumbersome to use those in
print_gdb_configuration because we would have to check for all 6 symbols
corresponding the 6 headers listed. This would also increase the
maintenance burden if support for other variations of curses are added.
Instead, define 'HAVE_LIBCURSES' ourselves by adding an
'action-if-found' argument to AC_SEARCH_LIBS, and use it in
print_gdb_configuration.
While at it, remove the condition on 'ac_cv_search_waddstr' and set
'curses_found' directly in 'action-if-found'.
Change-Id: Id90e3d73990e169cee51bcc3e1d52072cfacd5b8 Approved-By: Simon Marchi <simon.marchi@efficios.com>
Tom de Vries [Tue, 21 Feb 2023 14:26:24 +0000 (15:26 +0100)]
[gdb/testsuite] Require compilation flags in two gdb.arch/aarch64 test-cases
With test-cases gdb.arch/aarch64-mte-core.exp and gdb.arch/aarch64-pauth.exp I
run into compilation errors due to unsupported compilation flags.
Fix this by requiring the compilation flags, such that I have instead:
...
UNSUPPORTED: gdb.arch/aarch64-mte-core.exp: require failed: \
have_compile_flag -march=armv8.5-a+memtag
UNSUPPORTED: gdb.arch/aarch64-pauth.exp: require failed: \
have_compile_flag -mbranch-protection=pac-ret+leaf
...
Tom de Vries [Tue, 21 Feb 2023 14:06:50 +0000 (15:06 +0100)]
[gdb/testsuite] Require istarget x86* in gdb.reverse/step-indirect-call-thunk.exp
On aarch64-linux, I run into:
...
Running gdb.reverse/step-indirect-call-thunk.exp ...
gdb compile failed, gcc: error: unrecognized command line option \
'-mindirect-branch=thunk'; did you mean '-findirect-inlining'?
gcc: error: unrecognized command line option '-mfunction-return=thunk'; \
did you mean '-Wfunction-elimination'?
UNTESTED: gdb.reverse/step-indirect-call-thunk.exp: failed to prepare
...
Fix this by requiring istarget "x86*", similar to what was added in
gdb.base/step-indirect-call-thunk.exp by commit 43127ae5714 ("Fix
gdb.base/step-indirect-call-thunk.exp"), such that we have instead:
...
UNSUPPORTED: gdb.reverse/step-indirect-call-thunk.exp: require failed: \
istarget "x86*
...
Tom de Vries [Tue, 21 Feb 2023 13:41:14 +0000 (14:41 +0100)]
[gdb/testsuite] Require -fsplit-stack in gdb.base/morestack.exp
On aarch64-linux, I run into:
...
gdb compile failed, cc1: error: '-fsplit-stack' is not supported by this \
compiler configuration
UNTESTED: gdb.base/morestack.exp: failed to prepare
...
Fix this by requiring -fsplit-stack, such that we have instead:
...
UNSUPPORTED: gdb.base/morestack.exp: require failed: \
expr [have_compile_flag -fsplit-stack]
...
Tom de Vries [Tue, 21 Feb 2023 13:10:12 +0000 (14:10 +0100)]
[gdb/testsuite] Require syscall time in gdb.reverse/time-reverse.exp
On aarch64-linux, I run into:
...
Running gdb.reverse/time-reverse.exp ...
gdb compile failed, gdb.reverse/time-reverse.c: In function 'main':
gdb.reverse/time-reverse.c:39:12: error: 'SYS_time' undeclared \
(first use in this function); did you mean 'SYS_times'?
syscall (SYS_time, &time_global);
^~~~~~~~
SYS_times
gdb.reverse/time-reverse.c:39:12: note: each undeclared identifier is \
reported only once for each function it appears in
UNTESTED: gdb.reverse/time-reverse.exp: failed to prepare
...
Fix this by adding a new proc have_syscall, and requiring syscall time, such
that we have instead:
...
UNSUPPORTED: gdb.reverse/time-reverse.exp: require failed: \
expr [have_syscall time]
...
Clément Chigot [Wed, 11 Jan 2023 14:12:47 +0000 (15:12 +0100)]
gas/testsuite: adjust a test for case insensitive file systems
When dealing with case insensitive file systems, ".file line.s" and
".file Line.s" are identical and thus gas won't change the current
input file.
However, in line.l test, it's expecting to trigger an input file switch.
As the second filename doesn't matter in it, change it to fit for those
file systems.
gas/ChangeLog:
* testsuite/gas/elf/line.l: Change Line.s to Line2.s.
* testsuite/gas/elf/line.s: Adjust output.
Luis Machado [Sun, 11 Sep 2022 19:47:18 +0000 (20:47 +0100)]
[aarch64] Enable pointer authentication support for aarch64 bare metal/kernel mode addresses
At the moment GDB only handles pointer authentication (pauth) for userspace
addresses and if we're debugging a Linux-hosted program.
The Linux Kernel can be configured to use pauth instructions for some
additional security hardening, but GDB doesn't handle this well.
To overcome this limitation, GDB needs a couple things:
1 - The target needs to advertise pauth support.
2 - The hook to remove non-address bits from a pointer needs to be registered
in aarch64-tdep.c as opposed to aarch64-linux-tdep.c.
There is a patch for QEMU that addresses the first point, and it makes
QEMU's gdbstub expose a couple more pauth mask registers, so overall we will
have up to 4 pauth masks (2 masks or 4 masks):
pauth_dmask and pauth_cmask are the masks used to remove pauth signatures
from userspace addresses. pauth_dmask_high and pauth_cmask_high masks are used
to remove pauth signatures from kernel addresses.
The second point is easily addressed by moving code around.
When debugging a Linux Kernel built with pauth with an unpatched GDB, we get
the following backtrace:
#0 __fput (file=0xffff0000c17a6400) at /repos/linux/fs/file_table.c:296
#1 0xffff8000082bd1f0 in ____fput (work=<optimized out>) at /repos/linux/fs/file_table.c:348
#2 0x30008000080ade30 [PAC] in ?? ()
#3 0x30d48000080ade30 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
With a patched GDB, we get something a lot more meaningful:
#0 __fput (file=0xffff0000c1bcfa00) at /repos/linux/fs/file_table.c:296
#1 0xffff8000082bd1f0 in ____fput (work=<optimized out>) at /repos/linux/fs/file_table.c:348
#2 0xffff8000080ade30 [PAC] in task_work_run () at /repos/linux/kernel/task_work.c:179
#3 0xffff80000801db90 [PAC] in resume_user_mode_work (regs=0xffff80000a96beb0) at /repos/linux/include/linux/resume_user_mode.h:49
#4 do_notify_resume (regs=regs@entry=0xffff80000a96beb0, thread_flags=4) at /repos/linux/arch/arm64/kernel/signal.c:1127
#5 0xffff800008fb9974 [PAC] in prepare_exit_to_user_mode (regs=0xffff80000a96beb0) at /repos/linux/arch/arm64/kernel/entry-common.c:137
#6 exit_to_user_mode (regs=0xffff80000a96beb0) at /repos/linux/arch/arm64/kernel/entry-common.c:142
#7 el0_svc (regs=0xffff80000a96beb0) at /repos/linux/arch/arm64/kernel/entry-common.c:638
#8 0xffff800008fb9d34 [PAC] in el0t_64_sync_handler (regs=<optimized out>) at /repos/linux/arch/arm64/kernel/entry-common.c:655
#9 0xffff800008011548 [PAC] in el0t_64_sync () at /repos/linux/arch/arm64/kernel/entry.S:586
Backtrace stopped: Cannot access memory at address 0xffff80000a96c0c8
Clément Chigot [Fri, 13 Jan 2023 09:21:07 +0000 (10:21 +0100)]
ld/testsuite: don't output to /dev/null
Mingw doesn't have /dev/null and thus "-o /dev/null" will fail.
Currently, all the options are checked using this "-o /dev/null",
resulting in them being disabled on mingw hosts.
Fix that by outputting to a real file for all targets.
ld/ChangeLog:
* testsuite/config/default.exp: Replace "-o /dev/null" by a
file.
Alan Modra [Mon, 20 Feb 2023 23:27:49 +0000 (09:57 +1030)]
alpha-*-vms missing libraries
For this:
./ld-new: cannot find -limagelib: No such file or directory
./ld-new: cannot find -lstarlet: No such file or directory
./ld-new: cannot find -lsys$public_vectors: No such file or directory
the logs showed
creating dummy tmpdir/libimagelib:
creating dummy No
creating dummy such
etc.
So rubbish instead of tmpdir/libimagelib.a and the other required libs.
projects / thirdparty / binutils-gdb.git / log
