While working with pending fork events, I wondered what would happen if
the user detached an inferior while a thread of that inferior had a
pending fork event. What happens with the fork child, which is
ptrace-attached by the GDB process (or by GDBserver), but not known to
the core? Sure enough, neither the core of GDB or the target detach the
child process, so GDB (or GDBserver) just stays ptrace-attached to the
process. The result is that the fork child process is stuck, while you
would expect it to be detached and run.
Make GDBserver detach of fork children it knows about. That is done in
the generic handle_detach function. Since a process_info already exists
for the child, we can simply call detach_inferior on it.
GDB-side, make the linux-nat and remote targets detach of fork children
known because of pending fork events. These pending fork events can be
stored in:
- thread_info::pending_waitstatus, if the core has consumed the event
but then saved it for later (for example, because it got the event
while stopping all threads, to present an all-stop stop on top of a
non-stop target)
- thread_info::pending_follow: if we ran to a "catch fork" and we
detach at that moment
Additionally, pending fork events can be in target-specific fields:
- For linux-nat, they can be in lwp_info::status and
lwp_info::waitstatus.
- For the remote target, they could be stored as pending stop replies,
saved in `remote_state::notif_state::pending_event`, if not
acknowledged yet, or in `remote_state::stop_reply_queue`, if
acknowledged. I followed the model of remove_new_fork_children for
this: call remote_notif_get_pending_events to process /
acknowledge any unacknowledged notification, then look through
stop_reply_queue.
Update the gdb.threads/pending-fork-event.exp test (and rename it to
gdb.threads/pending-fork-event-detach.exp) to try to detach the process
while it is stopped with a pending fork event. In order to verify that
the fork child process is correctly detached and resumes execution
outside of GDB's control, make that process create a file in the test
output directory, and make the test wait $timeout seconds for that file
to appear (it happens instantly if everything goes well).
This test catches a bug in linux-nat.c, also reported as PR 28512
("waitstatus.h:300: internal-error: gdb_signal target_waitstatus::sig()
const: Assertion `m_kind == TARGET_WAITKIND_STOPPED || m_kind ==
TARGET_WAITKIND_SIGNALLED' failed.). When detaching a thread with a
pending event, get_detach_signal unconditionally fetches the signal
stored in the waitstatus (`tp->pending_waitstatus ().sig ()`). However,
that is only valid if the pending event is of type
TARGET_WAITKIND_STOPPED, and this is now enforced using assertions (iit
would also be valid for TARGET_WAITKIND_SIGNALLED, but that would mean
the thread does not exist anymore, so we wouldn't be detaching it). Add
a condition in get_detach_signal to access the signal number only if the
wait status is of kind TARGET_WAITKIND_STOPPED, and use GDB_SIGNAL_0
instead (since the thread was not stopped with a signal to begin with).
Add another test, gdb.threads/pending-fork-event-ns.exp, specifically to
verify that we consider events in pending stop replies in the remote
target. This test has many threads constantly forking, and we detach
from the program while the program is executing. That gives us some
chance that we detach while a fork stop reply is stored in the remote
target. To verify that we correctly detach all fork children, we ask
the parent to exit by sending it a SIGUSR1 signal and have it write a
file to the filesystem before exiting. Because the parent's main thread
joins the forking threads, and the forking threads wait for their fork
children to exit, if some fork child is not detach by GDB, the parent
will not write the file, and the test will time out. If I remove the
new remote_detach_pid calls in remote.c, the test fails eventually if I
run it in a loop.
There is a known limitation: we don't remove breakpoints from the
children before detaching it. So the children, could hit a trap
instruction after being detached and crash. I know this is wrong, and
it should be fixed, but I would like to handle that later. The current
patch doesn't fix everything, but it's a step in the right direction.
Change-Id: I6d811a56f520e3cb92d5ea563ad38976f92e93dd
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28512
A following patch will change targets so that when they detach an
inferior, they also detach any pending fork children this inferior may
have. While doing this, I hit a case where we couldn't differentiate
two cases, where in one we should detach the fork detach but not in the
other.
Suppose we continue past a fork with "follow-fork-mode == child" &&
"detach-on-fork on". follow_fork_inferior calls target_detach to detach
the parent. In that case the target should not detach the fork
child, as we'll continue debugging the child. As of now, the
tp->pending_follow field of the thread who called fork still contains
the details about the fork.
Then, suppose we run to a fork catchpoint and the user types "detach".
In that case, the target should detach the fork child in addition to the
parent. In that case as well, the tp->pending_follow field contains
the details about the fork.
To allow targets to differentiate the two cases, clear
tp->pending_follow a bit earlier, when following a fork. Targets will
then see that tp->pending_follow contains TARGET_WAITKIND_SPURIOUS, and
won't detach the fork child.
As of this patch, no behavior changes are expected.
Change-Id: I537741859ed712cb531baaefc78bb934e2a28153
In preparation for a following patch, refactor a few things that I did
find a bit awkward, and to make them a bit more reusable.
- Pass an inferior to kill_new_fork_children instead of a pid. That
allows iterating on only this inferior's threads and avoid further
filtering on the thread's pid.
- Change thread_pending_fork_status to return a non-nullptr value only
if the thread does have a pending fork status.
- Remove is_pending_fork_parent_thread, as one can just use
thread_pending_fork_status and check for nullptr.
- Replace is_pending_fork_parent with is_fork_status, which just
returns if the given target_waitkind if a fork or a vfork. Push
filtering on the pid to the callers, when it is necessary.
Change-Id: I0764ccc684d40f054e39df6fa5458cc4c5d1cd7b
Move the stop_reply and a few functions up. Some code above them in the
file will need to use them in a following patch. No behavior changes
expected here.
Change-Id: I3ca57d0e3ec253f56e1ba401289d9d167de14ad2
The following patch will add some code paths that need to ptrace-detach
a given PID. Factor out the code that does this and put it in its own
function, so that it can be re-used.
Change-Id: Ie65ca0d89893b41aea0a23d9fc6ffbed042a9705
This patch aims at fixing a bug where an inferior is unexpectedly
created when a fork happens at the same time as another event, and that
other event is reported to GDB first (and the fork event stays pending
in GDBserver). This happens for example when we step a thread and
another thread forks at the same time. The bug looks like (if I
reproduce the included test by hand):
(gdb) show detach-on-fork
Whether gdb will detach the child of a fork is on.
(gdb) show follow-fork-mode
Debugger response to a program call of fork or vfork is "parent".
(gdb) si
[New inferior 2]
Reading /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread from remote target...
Reading /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread from remote target...
Reading symbols from target:/home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.threads/step-while-fork-in-other-thread/step-while-fork-in-other-thread...
[New Thread 965190.965190]
[Switching to Thread 965190.965190]
Remote 'g' packet reply is too long (expected 560 bytes, got 816 bytes): ... <long series of bytes>
The sequence of events leading to the problem is:
- We are using the all-stop user-visible mode as well as the
synchronous / all-stop variant of the remote protocol
- We have two threads, thread A that we single-step and thread B that
calls fork at the same time
- GDBserver's linux_process_target::wait pulls the "single step
complete SIGTRAP" and the "fork" events from the kernel. It
arbitrarily choses one event to report, it happens to be the
single-step SIGTRAP. The fork stays pending in the thread_info.
- GDBserver send that SIGTRAP as a stop reply to GDB
- While in stop_all_threads, GDB calls update_thread_list, which ends
up querying the remote thread list using qXfer:threads:read.
- In the reply, GDBserver includes the fork child created as a result
of thread B's fork.
- GDB-side, the remote target sees the new PID, calls
remote_notice_new_inferior, which ends up unexpectedly creating a new
inferior, and things go downhill from there.
The problem here is that as long as GDB did not process the fork event,
it should pretend the fork child does not exist. Ultimately, this event
will be reported, we'll go through follow_fork, and that process will be
detached.
The remote target (GDB-side), has some code to remove from the reported
thread list the threads that are the result of forks not processed by
GDB yet. But that only works for fork events that have made their way
to the remote target (GDB-side), but haven't been consumed by the core
yet, so are still lingering as pending stop replies in the remote target
(see remove_new_fork_children in remote.c). But in our case, the fork
event hasn't made its way to the GDB-side remote target. We need to
implement the same kind of logic GDBserver-side: if there exists a
thread / inferior that is the result of a fork event GDBserver hasn't
reported yet, it should exclude that thread / inferior from the reported
thread list.
This was actually discussed a while ago, but not implemented AFAIK:
https://pi.simark.ca/gdb-patches/1ad9f5a8-d00e-9a26-b0c9-3f4066af5142@redhat.com/#thttps://sourceware.org/pipermail/gdb-patches/2016-June/133906.html
Implementation details-wise, the fix for this is all in GDBserver. The
Linux layer of GDBserver already tracks unreported fork parent / child
relationships using the lwp_info::fork_relative, in order to avoid
wildcard actions resuming fork childs unknown to GDB. This information
needs to be made available to the handle_qxfer_threads_worker function,
so it can filter the reported threads. Add a new thread_pending_parent
target function that allows the Linux target to return the parent of an
eventual fork child.
Testing-wise, the test replicates pretty-much the sequence of events
shown above. The setup of the test makes it such that the main thread
is about to fork. We stepi the other thread, so that the step completes
very quickly, in a single event. Meanwhile, the main thread is resumed,
so very likely has time to call fork. This means that the bug may not
reproduce every time (if the main thread does not have time to call
fork), but it will reproduce more often than not. The test fails
without the fix applied on the native-gdbserver and
native-extended-gdbserver boards.
At some point I suspected that which thread called fork and which thread
did the step influenced the order in which the events were reported, and
therefore the reproducibility of the bug. So I made the test try both
combinations: main thread forks while other thread steps, and vice
versa. I'm not sure this is still necessary, but I left it there
anyway. It doesn't hurt to test a few more combinations.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28288
Change-Id: I2158d5732fc7d7ca06b0eb01f88cf27bf527b990
There are some loops in gdb that use ARRAY_SIZE (or a wordier
equivalent) to loop over a static array. This patch changes some of
these to use foreach instead.
Regression tested on x86-64 Fedora 34.
In my earlier C++-ization patch for file_and_directory, I introduced
an error:
- if (strcmp (fnd.name, "<unknown>") != 0)
+ if (fnd.is_unknown ())
This change inverted the sense of the test, which causes failures with
.debug_names.
This patch fixes the bug. Regression tested on x86-64 Fedora 34. I
also tested it using the AdaCore internal test suite, with
.debug_names -- this was failing before, and now it works.
The documentation suggests that we implement gdb.Value.__init__,
however, this is not currently true, we really implement
gdb.Value.__new__. This will cause confusion if a user tries to
sub-class gdb.Value. They might write:
class MyVal (gdb.Value):
def __init__ (self, val):
gdb.Value.__init__(self, val)
obj = MyVal(123)
print ("Got: %s" % obj)
But, when they source this code they'll see:
(gdb) source ~/tmp/value-test.py
Traceback (most recent call last):
File "/home/andrew/tmp/value-test.py", line 7, in <module>
obj = MyVal(123)
File "/home/andrew/tmp/value-test.py", line 5, in __init__
gdb.Value.__init__(self, val)
TypeError: object.__init__() takes exactly one argument (the instance to initialize)
(gdb)
The reason for this is that, as we don't implement __init__ for
gdb.Value, Python ends up calling object.__init__ instead, which
doesn't expect any arguments.
The Python docs suggest that the reason why we might take this
approach is because we want gdb.Value to be immutable:
https://docs.python.org/3/c-api/typeobj.html#c.PyTypeObject.tp_new
But I don't see any reason why we should require gdb.Value to be
immutable when other types defined in GDB are not. This current
immutability can be seen in this code:
obj = gdb.Value(1234)
print("Got: %s" % obj)
obj.__init__ (5678)
print("Got: %s" % obj)
Which currently runs without error, but prints:
Got: 1234
Got: 1234
In this commit I propose that we switch to using __init__ to
initialize gdb.Value objects.
This does introduce some additional complexity, during the __init__
call a gdb.Value might already be associated with a gdb value object,
in which case we need to cleanly break that association before
installing the new gdb value object. However, the cost of doing this
is not great, and the benefit - being able to easily sub-class
gdb.Value seems worth it.
After this commit the first example above works without error, while
the second example now prints:
Got: 1234
Got: 5678
In order to make it easier to override the gdb.Value.__init__ method,
I have tweaked the definition of gdb.Value.__init__. The second,
optional argument to __init__ is a gdb.Type, if this argument is not
present then GDB figures out a suitable type.
However, if we want to override the __init__ method in a sub-class,
and still support the default argument, it is easier to write:
class MyVal (gdb.Value):
def __init__ (self, val, type=None):
gdb.Value.__init__(self, val, type)
Currently, passing None for the Type will result in an error:
TypeError: type argument must be a gdb.Type.
After this commit I now allow the type argument to be None, in which
case GDB figures out a suitable type just as if the type had not been
passed at all.
Unless a user is trying to reinitialize a value, or create sub-classes
of gdb.Value, there should be no user visible changes after this
commit.
While investigating some disassembler problems I ran into this case;
GDB compiled on a 32-bit arm target, with --enable-targets=all. Then
in GDB:
(gdb) set architecture i386
(gdb) disassemble 0x0,+4
unknown disassembler error (error = -1)
This is interesting because it shows a case where the libopcodes
disassembler is returning -1 without first calling the
memory_error_func callback. Indeed, the return from libopcodes
happens from this code snippet in i386-dis.c in the print_insn
function:
if (address_mode == mode_64bit && sizeof (bfd_vma) < 8)
{
(*info->fprintf_func) (info->stream,
_("64-bit address is disabled"));
return -1;
}
Notice how, prior to the return the disassembler tries to print a
helpful message out, but GDB doesn't print this message.
The reason this message goes missing is the call stack, it looks like
this:
gdb_pretty_print_disassembler::pretty_print_insn
gdb_disassembler::print_insn
gdbarch_print_insn
...
i386-dis.c:print_insn
When i386-dis.c:print_insn returns -1 this is handled in
gdb_disassembler::print_insn, where an exception is thrown. However,
the actual printing of the disassembler output is done in
gdb_pretty_print_disassembler::pretty_print_insn, and is only done if
an exception is not thrown.
In this commit I change this. The pretty_print_insn now uses
try/catch around the call to gdb_disassembler::print_insn, if we catch
an error then we first print any pending output in the instruction
buffer, before rethrowing the exception. As a result, even if an
exception is thrown we still print any pending disassembler output to
the screen; in the above case the helpful message will now be shown.
Before my patch we might expect to see this output:
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x0000000000000000: unknown disassembler error (error = -1)
(gdb)
But now we see this:
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x0000000000000000: 64-bit address is disabled
unknown disassembler error (error = -1)
If the disassembler returns -1 without printing a helpful message then
we would still expect a change in output, something like:
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x0000000000000000:
unknown disassembler error (error = -1)
Which I think is still acceptable, though at this point I think a
strong case can be made that this is a disassembler bug (not printing
anything, but still returning -1).
Notice however, that the error message is always printed on a new line
now. This is also true for the memory error case, where before we
might see this:
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x00000000: Cannot access memory at address 0x0
We now get this:
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x00000000:
Cannot access memory at address 0x0
For me, I'm happy to accept this change, having the error on a line by
itself, rather than just appended to the end of the previous line,
seems like an improvement, but I'm aware others might feel
differently, so I'd appreciate any feedback.
Permanent program breakpoints (ones inserted into the code) other than
the one GDB uses for POWER (0x7fe00008) did not result in stop but
caused GDB to loop infinitely.
This was because GDB did not recognize trap instructions other than
"trap". For example, "tw 12, 4, 4" was not be recognized, causing GDB
to loop forever.
This commit fixes this by providing POWER specific hook
(gdbarch_program_breakpoint_here_p) recognizing all tw, twi, td and tdi
instructions.
Tested on Linux on PowerPC e500 and on QEMU PPC64le.
Power ISA 3.0 B spec [1], sections 3.3.11 "Fixed-Point Trap Instructions"
and section C.6 "Trap Mnemonics" specify "tw, 31, 0, 0" (encoded as
0x7fe00008) as canonical unconditional trap instruction.
This commit changes the breakpoint instruction used by GDB from
"tw 12, r2, r2" to unconditional "trap".
[1]: https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0
If a.so contains an SHT_RELR section, objcopy a.so will fail with:
a.so: unknown type [0x13] section `.relr.dyn'
This change allows objcopy to work.
bfd/
* elf.c (bfd_section_from_shdr): Support SHT_RELR.
struct linespec contains pointers to vectors, instead of containing
vectors directly. This is probably historical, when linespec_parser
(which contains a struct linespec field) was not C++-ified yet. But it
seems easy to change the pointers to vectors to just vectors today.
This simplifies the code, we don't need to manually allocate and delete
the vectors and there's no pointer that can be NULL.
As far as I understand, there was not meaningful distinction between a
NULL pointer to vector and an empty vector. So all NULL checks are
changed for !empty checks.
Change-Id: Ie759707da14d9d984169b93233343a86e2de9ee6
We can remove the empty check: if the vector has size 1, it is obviously
not empty. This code ended up like this because the empty check used to
be a NULL check.
Change-Id: I1571bd0228818ca93f6a6b444e9b010dc2da4c08
Functions agent_eval_command and agent_command are used to implement
maintenance commands, rename them accordingly (with the maint_ prefix),
as well as the agent_command_1 helper function.
Change-Id: Iacf96d4a0a26298e8dd4648a0f38da649ea5ef61
FreeBSD's kernel has recently added two new ELF auxiliary vector
entries. AT_FXRNG points to a root seed version for the kernel's
PRNG. Userland can use this to reseed a userland PRNG after the
kernel's PRNG has reseeded. AT_KPRELOAD is the base address of a
kernel-provided vDSO.
This change displays the proper name and description of these entries
in 'info auxv'.
include/ChangeLog:
* elf/common.h (AT_FREEBSD_FXRNG, AT_FREEBSD_KPRELOAD): Define.
I noticed that global_symbol_searcher::expand_symtabs always passes a
file matcher to expand_symtabs_matching. However, if 'filenames' is
empty, then this always returns true. It's slightly more efficient to
pass a null file matcher in this case, because that lets the "quick"
symbol implementations skip any filename checks.
Regression tested on x86-64 Fedora 34.
In commit 80ad340c90 ("[gdb/testsuite] use -Ttext-segment for jit-elf tests")
the following change was made:
...
proc compile_jit_elf_main_as_so {main_solib_srcfile main_solib_binfile options} {
- set options [concat $options debug]
+ global jit_load_address jit_load_increment
+
+ set options [list \
+ additional_flags="-DMAIN=jit_dl_main" \
+ additional_flags=-DLOAD_ADDRESS=$jit_load_address \
+ additional_flags=-DLOAD_INCREMENT=$jit_load_increment \
+ debug]
...
Before the change, the options argument was used, but after the change not
anymore.
Fix this by reverting back to using "set options [concat $options ...]".
Fixing this gets us twice the -DMAIN=jit_dl_main bit, once from a caller, and
once from compile_jit_elf_main_as_so. Fix this by removing the bit from
compile_jit_elf_main_as_so, which makes the code similar to compile_jit_main.
Tested on x86_64-linux.
On openSUSE Leap 15.2 aarch64 I ran into:
...
FAIL: gdb.tui/basic.exp: check main is where we expect on the screen
...
while this is passing on x86_64.
On x86_64-linux we have at the initial screen dump for "list -q main":
...
0 +-/home/vries/gdb_versions/devel/src/gdb/testsuite/gdb.tui/tui-layout.c--+
1 | 15 You should have received a copy of the GNU General Public |
2 | 16 along with this program. If not, see <http://www.gnu.org/|
3 | 17 |
4 | 18 int |
5 | 19 main () |
6 | 20 { |
7 | 21 return 0; |
8 | 22 } |
9 | 23 |
...
but on aarch64:
...
0 +-/home/tdevries/gdb/src/gdb/testsuite/gdb.tui/tui-layout.c--------------+
1 | 16 along with this program. If not, see <http://www.gnu.org/|
2 | 17 |
3 | 18 int |
4 | 19 main () |
5 | 20 { |
6 | 21 return 0; |
7 | 22 } |
8 | 23 |
9 | 24 |
...
The cause of the diffferent placement is that we have as line number for main
on x86_64:
...
$ gdb -q -batch outputs/gdb.tui/basic/basic -ex "info line main"
Line 20 of "tui-layout.c" starts at address 0x4004a7 <main> \
and ends at 0x4004ab <main+4>.
...
and on aarch64 instead:
...
$ gdb -q -batch outputs/gdb.tui/basic/basic -ex "info line main"
Line 21 of "tui-layout.c" starts at address 0x4005f4 <main> \
and ends at 0x4005f8 <main+4>.
...
Fix this by using a new source file main-one-line.c, that implements the
entire main function on a single line, in order to force the compiler to use
that line number.
Also try to do less hard-coding in the test-case.
Tested on x86_64-linux and aarch64-linux.
Consider test-case test.c:
...
int main (void) {
void *p = malloc (10);
return 0;
}
...
When compiled to a non-PIE exec:
...
$ gcc -m32 test.c
...
the call sequence looks like:
...
8048447: 83 ec 0c sub $0xc,%esp
804844a: 6a 0a push $0xa
804844c: e8 bf fe ff ff call 8048310 <malloc@plt>
...
which calls to:
...
08048310 <malloc@plt>:
8048310: ff 25 0c a0 04 08 jmp *0x804a00c
8048316: 68 00 00 00 00 push $0x0
804831b: e9 e0 ff ff ff jmp 8048300 <.plt>
...
where the first insn at 0x8048310 initially jumps to the following address
0x8048316, read from the .got.plt @ 0x804a00c:
...
804a000 0c9f0408 00000000 00000000 16830408 ................
804a010 26830408 &...
...
Likewise, when compiled as a PIE:
...
$ gcc -m32 -fPIE -pie test.c
...
we have this call sequence (with %ebx setup to point to the .got.plt):
...
0000055d <main>:
579: 83 ec 0c sub $0xc,%esp
57c: 6a 0a push $0xa
57e: 89 c3 mov %eax,%ebx
580: e8 6b fe ff ff call 3f0 <malloc@plt>
...
which calls to:
...
000003f0 <malloc@plt>:
3f0: ff a3 0c 00 00 00 jmp *0xc(%ebx)
3f6: 68 00 00 00 00 push $0x0
3fb: e9 e0 ff ff ff jmp 3e0 <.plt>
...
where the insn at 0x3f0 initially jumps to following address 0x3f6, read from
the .got.plt at offset 0xc:
...
2000 f41e0000 00000000 00000000 f6030000 ................
2010 06040000 ....
...
When instead doing an inferior call to malloc (with nosharedlib to force
malloc to resolve to malloc@plt rather than the functions in ld.so or libc.so)
with the non-PIE exec, we have the expected:
...
$ gdb -q -batch a.out -ex start -ex nosharedlib -ex "p /x (void *)malloc (10)"
Temporary breakpoint 1 at 0x8048444
Temporary breakpoint 1, 0x08048444 in main ()
$1 = 0x804b160
...
But with the PIE exec, we run into:
...
$ gdb -q -batch a.out -ex start -ex nosharedlib -ex "p /x (void *)malloc (10)"
Temporary breakpoint 1 at 0x56c
Temporary breakpoint 1, 0x5655556c in main ()
Program received signal SIGSEGV, Segmentation fault.
0x565553f0 in malloc@plt ()
...
The segfault happens because:
- the inferior call mechanism doesn't setup %ebx
- %ebx instead is 0
- the jump to "*0xc(%ebx)" reads from memory at 0xc
Fix this by setting up %ebx properly in i386_thiscall_push_dummy_call.
Fixes this failure with target board unix/-m32/-pie/-fPIE reported in
PR28467:
...
FAIL: gdb.base/nodebug.exp: p/c (int) array_index("abcdef",2)
...
Tested on x86_64-linux, with target board unix/-m32 and unix/-m32/-fPIE/-pie.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28467
When running test-case gdb.base/cached-source-file.exp with target board
readnow, we run into:
...
FAIL: gdb.base/cached-source-file.exp: rerun program (the program exited)
...
The problem is that when rereading, the readnow is ignored.
Fix this by copying the readnow handling code from symbol_file_add_with_addrs
to reread_symbols.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=26800
On openSUSE Leap 42.3, with system compiler gcc 4.8.5 I run into:
...
(gdb) print u_one_two_three^M
src/gdb/gdbtypes.h:1050: internal-error: field: \
Assertion `idx >= 0 && idx < num_fields ()' failed.^M
...
We run into trouble while doing this in
ada_is_unconstrained_packed_array_type:
...
1953 return TYPE_FIELD_BITSIZE (type, 0) > 0;
...
which tries to get field 0 from a type without fields:
...
(gdb) p type->num_fields ()
$6 = 0
...
which is the case because the type is a typedef:
...
(gdb) p type->code ()
$7 = TYPE_CODE_TYPEDEF
...
Fix this by using the type referenced by the typedef instead.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28323
Commit b69c9d41e8 was broken in multiple ways regarding the realloc
of the target string, most notably in that "-little" wasn't actually
appended to the input_target or output_target. This caused asan
errors and "FAIL: Check if efi app format is recognized". I also
noticed that the input_target string wasn't being copied but rather
the output_target when dealing with the input target. Fix that too.
PR 26206
* objcopy.c (convert_efi_target): Rewrite. Allocate modified
target strings here..
(copy_main): ..rather than here. Do handle input_target,
not output_target for input.
It's not foolproof, for example we don't catch output to a linker
script, to a library specified with -l, or to an element of a thin
archive.
* ldlang.c (open_output): Exit with error on output file matching
an input file.
* testsuite/ld-misc/just-symbols.exp: Adjust ld -r test to suit.
This patch adds an #elif defined for PowerPC to setup the exit_0 macro.
This patch addes the needed macro definitionald logic to handle both elfV1
and elfV2.
The patch has been successfully tested on both PowerPC BE, Powerpc LE and
X86_64 with no regressions.
Test-cases gdb.arch/i386-{avx,sse}.exp use assembly instructions that require
the memory operands to be aligned to a certain boundary, and the test-cases
use C11's _Alignas to make that happen.
The draw-back of using _Alignas is that while it does enforce a minimum
alignment, the actual alignment may be bigger, which makes the following
scenario possible:
- copy say, gdb.arch/i386-avx.c as basis for a new test-case
- run the test-case and observe a PASS
- commit the new test-case in the supposition that the test-case is correct
and well-tested
- run later into a failure on a different test setup (which may be a setup
where reproduction and investigation is more difficult and time-consuming),
and find out that the specified alignment was incorrect and should have been
updated to say, 64 bytes. The initial PASS occurred only because the actual
alignment happened to be greater than required.
The idea of having precise alignment as a means of having more predictable
execution which allows flushing out bugs earlier, has been filed as PR
gcc/103095.
Add a new file lib/precise-aligned-alloc.c with functions
precise_aligned_alloc and precise_aligned_dup, to support precise alignment.
Use precise_aligned_dup in aforementioned test-cases to:
- verify that the specified alignment is indeed sufficient, rather
than too little but accidentally over-aligned.
- prevent the same type of problems in any new test-cases based on these
Tested on x86_64-linux, with both gcc and clang.
When running test-case gdb.arch/i386-avx.exp with clang I ran into:
...
(gdb) PASS: gdb.arch/i386-avx.exp: set first breakpoint in main
continue^M
Continuing.^M
^M
Program received signal SIGSEGV, Segmentation fault.^M
0x000000000040052b in main (argc=1, argv=0x7fffffffd3c8) at i386-avx.c:54^M
54 asm ("vmovaps 0(%0), %%ymm0\n\t"^M
(gdb) FAIL: gdb.arch/i386-avx.exp: continue to breakpoint: \
continue to first breakpoint in main
...
The problem is that the vmovaps insn requires an 256-bit (or 32-byte) aligned
address, and it's only 16-byte aligned:
...
(gdb) p /x $rax
$1 = 0x601030
...
Fix this by using a sufficiently aligned address, using _Alignas.
Compile using -std=gnu11 to support _Alignas.
Likewise in gdb.arch/i386-sse.exp.
Tested on x86_64-linux, with both gcc and clang.
The fixes gold failures to handle inline PLT sequences properly.
PowerPC gold was always turning these back into direct calls due to
gsym->use_plt_offset() returning false. This is fixed for dynamic
linking by correcting get_reference_flags, and for static linking by
overriding use_plt_offset() in relocate(). The rest of the patch
revolves around needing to create PLT entries for inline PLT calls
when statically linking (for gcc -mlongcall). The lplt section
handled that for local symbols, now it does globals too.
* powerpc.cc (Target_powerpc::plt_off): Return proper section
for static link.
(Target_powerpc::symval_for_branch): Make public.
(Target_powerpc::make_lplt_section): Add Symbol_table* param.
Adjust all calls.
(Target_powerpc::make_local_plt_entry): Likewise.
(Target_powerpc::make_local_plt_entry): New variant for global syms.
(Powerpc_relobj::do_relocate_sections): Don't write lplt contents.
(Output_data_plt_powerpc::do_write): Write lplt contents here.
(Output_data_plt_powerpc::Output_data_plt_powerpc): Save
symbol table pointer. Adjust all uses.
(Output_data_plt_powerpc::add_entry): Add stash parameter. Don't
do dynamic reloc handling when no reloc section. Save symbol
for local plt entries.
(Output_data_plt_powerpc::add_local_entry): Save symbol.
(Output_data_plt_powerpc::Local_plt_ent): New class.
(Output_data_plt_powerpc::sym_ents_): New vector.
(Target_powerpc::Scan::get_reference_flags): Return
FUNCTION_CALL|RELATIVE_REF for inline plt relocs.
(Target_powerpc::Scan::global): Make entries in lplt for inline
plt call relocation symbols.
(Target_powerpc::Relocate::relocate): Rename has_plt_offset to
use_plt_offset. Set use_plt_offset for inline plt relocs.
It's now possible to refer symbols in the main program from the
shared library. However, it still impossible to have the same
overriden features between shared objects and mains than ELF,
without using the runtime linking feature which isn't yet fully
available.
ld/ChangeLog:
* testsuite/ld-shared/shared.exp: Improve XCOFF support
* testsuite/ld-shared/main.c: Likewise.
* testsuite/ld-shared/sh1.c: Likewise.
* testsuite/ld-shared/xcoff.dat: Likewise.
This fixes a use-after-free that Simon pointed out.
process_psymtab_comp_unit_reader was allocating an artificial name for
a CU, and then discarding it. However, this name was preserved in the
cached file_and_directory. This patch arranges for the allocated name
to be preserved there.
We're including system headers after local headers in a bunch of
places, but this leads to conflicts when our local headers happen
to define symbols that show up in the system headers.
Use the more standard order of:
* config.h (via defs.h)
* system headers
* local library headers (e.g. bfd & libiberty)
* sim specific headers
ASan made me notice a memory leak, where the memory tied to the file
handle name string wasn't freed. When register a file handler with an
fd that is already registered, we re-use the file_handler object, so we
ended up creating a new std::string object and overwriting the
file_handler::name pointer, without free-ing the old std::string.
Fix this by allocating file_handler with new, deleting it with
delete, and making file_handler::name not a pointer.
Change-Id: Ie304cc78ab5ae5dfad9a1366e9890c09de651f43
