A previous patch taught GDB about a new TARGET_WAITKIND_THREAD_CLONED
event kind, and made the Linux target report clone events.
A following patch will teach Linux GDBserver to do the same thing.
However, for remote debugging, it wouldn't be ideal for GDBserver to
report every clone event to GDB, when GDB only cares about such events
in some specific situations. Reporting clone events all the time
would be potentially chatty. We don't enable thread create/exit
events all the time for the same reason. Instead we have the
QThreadEvents packet. QThreadEvents is target-wide, though.
This patch makes GDB instead explicitly request that the target
reports clone events or not, on a per-thread basis.
In order to be able to do that with GDBserver, we need a new remote
protocol feature. Since a following patch will want to enable thread
exit events on per-thread basis too, the packet introduced here is
more generic than just for clone events. It lets you enable/disable a
set of options at once, modelled on Linux ptrace's PTRACE_SETOPTIONS.
IOW, this commit introduces a new QThreadOptions packet, that lets you
specify a set of per-thread event options you want to enable. The
packet accepts a list of options/thread-id pairs, similarly to vCont,
processed left to right, with the options field being a number
interpreted as a bit mask of options. The only option defined in this
commit is GDB_THREAD_OPTION_CLONE (0x1), which ask the remote target
to report clone events. Another patch later in the series will
introduce another option.
For example, this packet sets option "1" (clone events) on thread
p1000.2345:
QThreadOptions;1:p1000.2345
and this clears options for all threads of process 1000, and then sets
option "1" (clone events) on thread p1000.2345:
QThreadOptions;0:p1000.-1;1:p1000.2345
This clears options of all threads of all processes:
QThreadOptions;0
The target reports the set of supported options by including
"QThreadOptions=<supported options>" in its qSupported response.
infrun is then tweaked to enable GDB_THREAD_OPTION_CLONE when stepping
over a breakpoint.
Unlike PTRACE_SETOPTIONS, fork/vfork/clone children do NOT inherit
their parent's thread options. This is so that GDB can send e.g.,
"QThreadOptions;0;1:TID" without worrying about threads it doesn't
know about yet.
Documentation for this new remote protocol feature is included in a
documentation patch later in the series.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=19675
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27830
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Change-Id: Ie41e5093b2573f14cf6ac41b0b5804eba75be37e
The previous patch taught GDB about a new
TARGET_WAITKIND_THREAD_CLONED event kind, and made the Linux target
report clone events.
A following patch will teach Linux GDBserver to do the same thing.
But before we get there, we need to teach the remote protocol about
TARGET_WAITKIND_THREAD_CLONED. That's what this patch does. Clone is
very similar to vfork and fork, and the new stop reply is likewise
handled similarly. The stub reports "T05clone:...".
GDBserver core is taught to handle TARGET_WAITKIND_THREAD_CLONED and
forward it to GDB in this patch, but no backend actually emits it yet.
That will be done in a following patch.
Documentation for this new remote protocol feature is included in a
documentation patch later in the series.
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Change-Id: If271f20320d864f074d8ac0d531cc1a323da847f
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=19675
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27830
(A good chunk of the problem statement in the commit log below is
Andrew's, adjusted for a different solution, and for covering
displaced stepping too. The testcase is mostly Andrew's too.)
This commit addresses bugs gdb/19675 and gdb/27830, which are about
stepping over a breakpoint set at a clone syscall instruction, one is
about displaced stepping, and the other about in-line stepping.
Currently, when a new thread is created through a clone syscall, GDB
sets the new thread running. With 'continue' this makes sense
(assuming no schedlock):
- all-stop mode, user issues 'continue', all threads are set running,
a newly created thread should also be set running.
- non-stop mode, user issues 'continue', other pre-existing threads
are not affected, but as the new thread is (sort-of) a child of the
thread the user asked to run, it makes sense that the new threads
should be created in the running state.
Similarly, if we are stopped at the clone syscall, and there's no
software breakpoint at this address, then the current behaviour is
fine:
- all-stop mode, user issues 'stepi', stepping will be done in place
(as there's no breakpoint to step over). While stepping the thread
of interest all the other threads will be allowed to continue. A
newly created thread will be set running, and then stopped once the
thread of interest has completed its step.
- non-stop mode, user issues 'stepi', stepping will be done in place
(as there's no breakpoint to step over). Other threads might be
running or stopped, but as with the continue case above, the new
thread will be created running. The only possible issue here is
that the new thread will be left running after the initial thread
has completed its stepi. The user would need to manually select
the thread and interrupt it, this might not be what the user
expects. However, this is not something this commit tries to
change.
The problem then is what happens when we try to step over a clone
syscall if there is a breakpoint at the syscall address.
- For both all-stop and non-stop modes, with in-line stepping:
+ user issues 'stepi',
+ [non-stop mode only] GDB stops all threads. In all-stop mode all
threads are already stopped.
+ GDB removes s/w breakpoint at syscall address,
+ GDB single steps just the thread of interest, all other threads
are left stopped,
+ New thread is created running,
+ Initial thread completes its step,
+ [non-stop mode only] GDB resumes all threads that it previously
stopped.
There are two problems in the in-line stepping scenario above:
1. The new thread might pass through the same code that the initial
thread is in (i.e. the clone syscall code), in which case it will
fail to hit the breakpoint in clone as this was removed so the
first thread can single step,
2. The new thread might trigger some other stop event before the
initial thread reports its step completion. If this happens we
end up triggering an assertion as GDB assumes that only the
thread being stepped should stop. The assert looks like this:
infrun.c:5899: internal-error: int finish_step_over(execution_control_state*): Assertion `ecs->event_thread->control.trap_expected' failed.
- For both all-stop and non-stop modes, with displaced stepping:
+ user issues 'stepi',
+ GDB starts the displaced step, moves thread's PC to the
out-of-line scratch pad, maybe adjusts registers,
+ GDB single steps the thread of interest, [non-stop mode only] all
other threads are left as they were, either running or stopped.
In all-stop, all other threads are left stopped.
+ New thread is created running,
+ Initial thread completes its step, GDB re-adjusts its PC,
restores/releases scratchpad,
+ [non-stop mode only] GDB resumes the thread, now past its
breakpoint.
+ [all-stop mode only] GDB resumes all threads.
There is one problem with the displaced stepping scenario above:
3. When the parent thread completed its step, GDB adjusted its PC,
but did not adjust the child's PC, thus that new child thread
will continue execution in the scratch pad, invoking undefined
behavior. If you're lucky, you see a crash. If unlucky, the
inferior gets silently corrupted.
What is needed is for GDB to have more control over whether the new
thread is created running or not. Issue #1 above requires that the
new thread not be allowed to run until the breakpoint has been
reinserted. The only way to guarantee this is if the new thread is
held in a stopped state until the single step has completed. Issue #3
above requires that GDB is informed of when a thread clones itself,
and of what is the child's ptid, so that GDB can fixup both the parent
and the child.
When looking for solutions to this problem I considered how GDB
handles fork/vfork as these have some of the same issues. The main
difference between fork/vfork and clone is that the clone events are
not reported back to core GDB. Instead, the clone event is handled
automatically in the target code and the child thread is immediately
set running.
Note we have support for requesting thread creation events out of the
target (TARGET_WAITKIND_THREAD_CREATED). However, those are reported
for the new/child thread. That would be sufficient to address in-line
stepping (issue #1), but not for displaced-stepping (issue #3). To
handle displaced-stepping, we need an event that is reported to the
_parent_ of the clone, as the information about the displaced step is
associated with the clone parent. TARGET_WAITKIND_THREAD_CREATED
includes no indication of which thread is the parent that spawned the
new child. In fact, for some targets, like e.g., Windows, it would be
impossible to know which thread that was, as thread creation there
doesn't work by "cloning".
The solution implemented here is to model clone on fork/vfork, and
introduce a new TARGET_WAITKIND_THREAD_CLONED event. This event is
similar to TARGET_WAITKIND_FORKED and TARGET_WAITKIND_VFORKED, except
that we end up with a new thread in the same process, instead of a new
thread of a new process. Like FORKED and VFORKED, THREAD_CLONED
waitstatuses have a child_ptid property, and the child is held stopped
until GDB explicitly resumes it. This addresses the in-line stepping
case (issues #1 and #2).
The infrun code that handles displaced stepping fixup for the child
after a fork/vfork event is thus reused for THREAD_CLONE, with some
minimal conditions added, addressing the displaced stepping case
(issue #3).
The native Linux backend is adjusted to unconditionally report
TARGET_WAITKIND_THREAD_CLONED events to the core.
Following the follow_fork model in core GDB, we introduce a
target_follow_clone target method, which is responsible for making the
new clone child visible to the rest of GDB.
Subsequent patches will add clone events support to the remote
protocol and gdbserver.
displaced_step_in_progress_thread becomes unused with this patch, but
a new use will reappear later in the series. To avoid deleting it and
readding it back, this patch marks it with attribute unused, and the
latter patch removes the attribute again. We need to do this because
the function is static, and with no callers, the compiler would warn,
(error with -Werror), breaking the build.
This adds a new gdb.threads/stepi-over-clone.exp testcase, which
exercises stepping over a clone syscall, with displaced stepping vs
inline stepping, and all-stop vs non-stop. We already test stepping
over clone syscalls with gdb.base/step-over-syscall.exp, but this test
uses pthreads, while the other test uses raw clone, and this one is
more thorough. The testcase passes on native GNU/Linux, but fails
against GDBserver. GDBserver will be fixed by a later patch in the
series.
Co-authored-by: Andrew Burgess <aburgess@redhat.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=19675
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27830
Change-Id: I95c06024736384ae8542a67ed9fdf6534c325c8e
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
I noticed that on an Ubuntu 20.04 system, after a following patch
("Step over clone syscall w/ breakpoint,
TARGET_WAITKIND_THREAD_CLONED"), the gdb.threads/step-over-exec.exp
was passing cleanly, but still, we'd end up with four new unexpected
GDB core dumps:
=== gdb Summary ===
# of unexpected core files 4
# of expected passes 48
That said patch is making the pre-existing
gdb.threads/step-over-exec.exp testcase (almost silently) expose a
latent problem in gdb/linux-nat.c, resulting in a GDB crash when:
#1 - a non-leader thread execs
#2 - the post-exec program stops somewhere
#3 - you kill the inferior
Instead of #3 directly, the testcase just returns, which ends up in
gdb_exit, tearing down GDB, which kills the inferior, and is thus
equivalent to #3 above.
Vis (after said patch is applied):
$ gdb --args ./gdb /home/pedro/gdb/build/gdb/testsuite/outputs/gdb.threads/step-over-exec/step-over-exec-execr-thread-other-diff-text-segs-true
...
(top-gdb) r
...
(gdb) b main
...
(gdb) r
...
Breakpoint 1, main (argc=1, argv=0x7fffffffdb88) at /home/pedro/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/gdb.threads/step-over-exec.c:69
69 argv0 = argv[0];
(gdb) c
Continuing.
[New Thread 0x7ffff7d89700 (LWP 2506975)]
Other going in exec.
Exec-ing /home/pedro/gdb/build/gdb/testsuite/outputs/gdb.threads/step-over-exec/step-over-exec-execr-thread-other-diff-text-segs-true-execd
process 2506769 is executing new program: /home/pedro/gdb/build/gdb/testsuite/outputs/gdb.threads/step-over-exec/step-over-exec-execr-thread-other-diff-text-segs-true-execd
Thread 1 "step-over-exec-" hit Breakpoint 1, main () at /home/pedro/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/gdb.threads/step-over-exec-execd.c:28
28 foo ();
(gdb) k
...
Thread 1 "gdb" received signal SIGSEGV, Segmentation fault.
0x000055555574444c in thread_info::has_pending_waitstatus (this=0x0) at ../../src/gdb/gdbthread.h:393
393 return m_suspend.waitstatus_pending_p;
(top-gdb) bt
#0 0x000055555574444c in thread_info::has_pending_waitstatus (this=0x0) at ../../src/gdb/gdbthread.h:393
#1 0x0000555555a884d1 in get_pending_child_status (lp=0x5555579b8230, ws=0x7fffffffd130) at ../../src/gdb/linux-nat.c:1345
#2 0x0000555555a8e5e6 in kill_unfollowed_child_callback (lp=0x5555579b8230) at ../../src/gdb/linux-nat.c:3564
#3 0x0000555555a92a26 in gdb::function_view<int (lwp_info*)>::bind<int, lwp_info*>(int (*)(lwp_info*))::{lambda(gdb::fv_detail::erased_callable, lwp_info*)#1}::operator()(gdb::fv_detail::erased_callable, lwp_info*) const (this=0x0, ecall=..., args#0=0x5555579b8230) at ../../src/gdb/../gdbsupport/function-view.h:284
#4 0x0000555555a92a51 in gdb::function_view<int (lwp_info*)>::bind<int, lwp_info*>(int (*)(lwp_info*))::{lambda(gdb::fv_detail::erased_callable, lwp_info*)#1}::_FUN(gdb::fv_detail::erased_callable, lwp_info*) () at ../../src/gdb/../gdbsupport/function-view.h:278
#5 0x0000555555a91f84 in gdb::function_view<int (lwp_info*)>::operator()(lwp_info*) const (this=0x7fffffffd210, args#0=0x5555579b8230) at ../../src/gdb/../gdbsupport/function-view.h:247
#6 0x0000555555a87072 in iterate_over_lwps(ptid_t, gdb::function_view<int (lwp_info*)>) (filter=..., callback=...) at ../../src/gdb/linux-nat.c:864
#7 0x0000555555a8e732 in linux_nat_target::kill (this=0x55555653af40 <the_amd64_linux_nat_target>) at ../../src/gdb/linux-nat.c:3590
#8 0x0000555555cfdc11 in target_kill () at ../../src/gdb/target.c:911
...
The root of the problem is that when a non-leader LWP execs, it just
changes its tid to the tgid, replacing the pre-exec leader thread,
becoming the new leader. There's no thread exit event for the execing
thread. It's as if the old pre-exec LWP vanishes without trace. The
ptrace man page says:
"PTRACE_O_TRACEEXEC (since Linux 2.5.46)
Stop the tracee at the next execve(2). A waitpid(2) by the
tracer will return a status value such that
status>>8 == (SIGTRAP | (PTRACE_EVENT_EXEC<<8))
If the execing thread is not a thread group leader, the thread
ID is reset to thread group leader's ID before this stop.
Since Linux 3.0, the former thread ID can be retrieved with
PTRACE_GETEVENTMSG."
When the core of GDB processes an exec events, it deletes all the
threads of the inferior. But, that is too late -- deleting the thread
does not delete the corresponding LWP, so we end leaving the pre-exec
non-leader LWP stale in the LWP list. That's what leads to the crash
above -- linux_nat_target::kill iterates over all LWPs, and after the
patch in question, that code will look for the corresponding
thread_info for each LWP. For the pre-exec non-leader LWP still
listed, won't find one.
This patch fixes it, by deleting the pre-exec non-leader LWP (and
thread) from the LWP/thread lists as soon as we get an exec event out
of ptrace.
GDBserver does not need an equivalent fix, because it is already doing
this, as side effect of mourning the pre-exec process, in
gdbserver/linux-low.cc:
else if (event == PTRACE_EVENT_EXEC && cs.report_exec_events)
{
...
/* Delete the execing process and all its threads. */
mourn (proc);
switch_to_thread (nullptr);
The crash with gdb.threads/step-over-exec.exp is not observable on
newer systems, which postdate the glibc change to move "libpthread.so"
internals to "libc.so.6", because right after the exec, GDB traps a
load event for "libc.so.6", which leads to GDB trying to open
libthread_db for the post-exec inferior, and, on such systems that
succeeds. When we load libthread_db, we call
linux_stop_and_wait_all_lwps, which, as the name suggests, stops all
lwps, and then waits to see their stops. While doing this, GDB
detects that the pre-exec stale LWP is gone, and deletes it.
If we use "catch exec" to stop right at the exec before the
"libc.so.6" load event ever happens, and issue "kill" right there,
then GDB crashes on newer systems as well. So instead of tweaking
gdb.threads/step-over-exec.exp to cover the fix, add a new
gdb.threads/threads-after-exec.exp testcase that uses "catch exec".
The test also uses the new "maint info linux-lwps" command if testing
on Linux native, which also exposes the stale LWP problem with an
unfixed GDB.
Also tweak a comment in infrun.c:follow_exec referring to how
linux-nat.c used to behave, as it would become stale otherwise.
Reviewed-By: Andrew Burgess <aburgess@redhat.com>
Change-Id: I21ec18072c7750f3a972160ae6b9e46590376643
This adds a maintenance command that lets you list all the LWPs under
control of the linux-nat target.
For example:
(gdb) maint info linux-lwps
LWP Ptid Thread ID
560948.561047.0 None
560948.560948.0 1.1
This shows that "560948.561047.0" LWP doesn't map to any thread_info
object, which is bogus. We'll be using this in a testcase in a
following patch.
Co-Authored-By: Pedro Alves <pedro@palves.net>
Change-Id: Ic4e9e123385976e5cd054391990124b7a20fb3f5
When building gdb with -O2, we run into:
...
gdb/tui/tui-disasm.c: In function ‘CORE_ADDR tui_find_disassembly_address \
(gdbarch*, CORE_ADDR, int)’:
gdb/tui/tui-disasm.c:293:7: warning: ‘last_addr’ may be used uninitialized \
in this function [-Wmaybe-uninitialized]
if (last_addr < pc)
^~
...
The warning triggers since commit 72535eb14b ("[gdb/tui] Fix segfault in
tui_find_disassembly_address").
Fix the warning by ensuring that last_addr is initialized at the point of
use:
...
+ last_addr = asm_lines.back ().addr;
if (last_addr < pc)
...
Tested on x86_64-linux.
In tui_find_disassembly_address we find an assert:
...
if (asm_lines.size () < max_lines)
{
if (!possible_new_low.has_value ())
return new_low;
/* Take the best possible match we have. */
new_low = *possible_new_low;
next_addr = tui_disassemble (gdbarch, asm_lines, new_low, max_lines);
last_addr = asm_lines.back ().addr;
gdb_assert (asm_lines.size () >= max_lines);
}
...
The comment right above:
...
/* If we failed to disassemble the required number of lines then the
following walk forward is not going to work, it assumes that
ASM_LINES contains exactly MAX_LINES entries. Instead we should
consider falling back to a previous possible start address in
POSSIBLE_NEW_LOW. */
...
claims that the more strict asm_lines.size () == max_line is required.
Update the assert to reflect this, and move it to after the if because it's
supposed to hold in general, not just when entering the if.
Tested on x86_64-linux.
defs.h declares re_comp, but it shouldn't. If this is needed, it
should be picked up from xregex.h via gdb_regex.h.
Tested by rebuilding.
Reviewed-by: Kevin Buettner <kevinb@redhat.com>
the offset-to-entry mappings are allocated in blocks, which may
become a bit wasteful in case there are extremely many small
input files or sections. This made it so that a large project
(Qt5WebEngine) didn't build anymore on x86 32bit due to address
space limits. It barely fit into address space before the new
string merging, and then got pushed over the limit by this.
So instead of leaving the waste reallocate the maps to their final
size once known. Now the link barely fits again.
bfd/
* merge.c (record_section): Reallocate offset maps to their
final size.
as the bug report shows we had an overflow in the test if
hash table resizing is needed. Reorder the expression to avoid
that. There's still a bug somewhere in gracefully handling
failure in resizing (e.g. out of memory), but this pushes the
boundary for that occurring somewhen into the future and
immediately helps the reporter.
bfd/
PR ld/31009
* merge.c (NEEDS_RESIZE): New macro avoiding overflow.
(sec_merge_maybe_resize): Use it.
(sec_merge_hash_insert): Ditto.
The tests are not compatible with ilp32 abi: the GNU property
note is ABI dependent (size changes) and the disasm is ABI
dependent too. Making the test portable between the ABIs is
not trivial.
For now force lp64 abi.
We decide to emit BTI stubs based on the instruction at the target
location. But PLT code is generated later than the stubs so we always
read 0 which is not a valid BTI.
Fix the logic to special case the PLT section: this is code the linker
generates so we know when it will have BTI.
This avoids BTI stubs in large executables where the PLTs have them
already. An alternative is to never emit BTI stubs for PLTs, instead
use BTI in the PLT if a library gets too big, however that may be
more tricky given the ordering of PLT sizing and stub insertion.
Related to bug 30957.
BTI stub parameters were recomputed even if those were already set up.
This is unnecessary work and leaks the symbol name that is allocated
for the stub.
Input sections are grouped together that can use the same stub area
(within reach) and these groups have a stable id.
Stubs have a name generated from the stub group id and target symbol.
When a relocation requires a stub with a name that already exists, the
stub is reused instead of adding a new one.
For an indirect branch stub another BTI stub may be inserted near the
target to provide a BTI landing pad.
The BTI stub can end up with the same stub group id and thus the same
name as the indirect stub. This happens if the target symbol is within
reach of the indirect branch stub. Then, due to the name collision,
only a single stub was emmitted which branched to itself causing an
infinite loop at runtime.
A possible solution is to just name the BTI stubs differently, but
since in the problematic case the indirect and BTI stub are in the
same stub area, a better solution is to emit a single stub with a
direct branch. The stub is still needed since the caller cannot reach
the target directly and we also want a BTI landing pad in the stub in
case other indirect stubs target the same symbol and thus need a BTI
stub.
In short we convert an indirect branch stub into a BTI stub when the
target is within reach and has no BTI. It is a hassle to change the
symbol of the stub so a BTI stub may end up with *_veneer instead of
*_bti_veneer after the conversion, but this should not matter much.
(Refactoring some of _bfd_aarch64_add_call_stub_entries would be
useful but too much for this bug fix patch.)
The same conversion to direct branch could be done even if the target
did not need a BTI. The stub groups are fixed in the current logic so
linking can fail if too many stubs are inserted and the section layout
is changed too much, but this only happens in extreme cases that can
be reasonably ignored. Because of this the target cannot go out of
reach during stub insertion so the optimization is valid, but not
implemented by this patch for the non-BTI case.
Fixes bug 30930.
The instruction was looked up in the wrong input file (file of branch
source instead of branch target) when optimizing away BTI stubs in
commit 5834f36d93
bfd: aarch64: Optimize BTI stubs PR30076
This can cause adding BTI stubs when they are not necessary or removing
them when they are (the latter is a correctness issue but it is very
unlikely in practice).
Fixes bug 30957.
The erroneous omission of a "reg_value == " in the THE system register
encoding check added in [1] led to an error which was not picked up in
GCC but which was flagged in Clang due to its use of
[-Werror,-Wconstant-logical-operand] check. Together with this fix we
add a new test for the THE registers to pick up their illegal use,
adding an extra and important layer of validation.
Furthermore, in separating system register from instruction
implementation (with which only the former was of concern in the cited
patch), additions made to `aarch64-tbl.h' are rolled back so
that these can be added later when adding THE instructions to the
codebase, a more natural place for these changes.
[1] https://sourceware.org/pipermail/binutils/2023-November/130314.html
opcodes/ChangeLog:
* aarch64-opc.c (aarch64_sys_ins_reg_supported_p): Fix typo.
* aarch64-tbl.h (THE): Remove.
(aarch64_feature_set aarch64_feature_the): Likewise.
gas/ChangeLog:
* testsuite/gas/aarch64/illegal-sysreg-8.l: Add tests for THE
system registers.
* testsuite/gas/aarch64/illegal-sysreg-8.s: Likewise.
As we have grown more uses of it, it becomes increasingly more desirable
to replace it by a simpler check. Have i386-gen do at build time what so
far was done at runtime: Deal with templates indicating EVEX-encoding by
other than the EVex attribute, and set that to "dynamic" in such cases.
This then allows simplifying a number of other conditionals as well.
Right now the opcode table has entries with ISA restrictions of the form
FEAT1|FEAT2, the meaning of which depends on context and requires
special treatment in tc-i386.c: Sometimes this means "both features
requires", whereas originally it was intended to solely mean "all of
these features required". Split the field, with the original one
regaining its original meaning. The new field now truly means "any of
these". The combination of both fields is still and &&-type check, i.e.
(all of these) && (any of these). In the opcode table more involved
combinations of features then also need expressing this way: "all"
entities first, follow by "any" entities enclosed in parentheses, e.g.
x64&(AVX|AVX512F). If the "all" part is empty, parentheses may not be
added around the "any" part (unless parsing logic was further relaxed).
Note that this way AVX512VL no longer needs as much special treatment,
and hence templates previously using AVX512F|AVX512VL are switched to
just AVX512VL.
Note further that this requires FMA handling as resulting from
da0784f961 ("x86: fold FMA VEX and EVEX templates") to be slightly
re-done: FMA now becomes more similar to AVX and AVX2.
First of all we want to also accumulate its reverse dependencies, such
that we can use them in cpu_flags_match(). This is in particular in
preparation of APX additions, such that e.g. BMI VEX-encoding templates
can become combined VEX/EVEX ones.
Once we have the reverse dependencies, we can further leverage them to
omit explicit "&x64" from any insn templates dealing with 64-bit-mode-
only ISA extensions. Besides helping readability for several insn
templates we already have, this will also help with what is going to be
added for APX (as all of the new templates would otherwise need to have
"&x64").
Note that rather than leaving a meaningless CPU_64_FLAGS (which is
unused anyway), its emitting is now also suppressed.
The test currently fails for IEEE 128-bit floating point types. PowerPC
supports the IBM double 128-bit floating point format and IEEE 128-bit
format. The IBM double 128-bit floating point format uses two 64-bit
floating point registers to store the 128-bit value. The IEEE 128-bit
floating point format stores the value in a single 128-bit vector-scalar
register (vsr).
The various floating point values, 32-bit float, 64-bit double, IBM double
128-bit float and IEEE 128-bit floating point numbers are all mapped to the
DWARF fpr numbers. The issue is the IEEE 128-bit floating point values are
actually stored in a vsr not the fprs. This patch changes the register
mapping for the vsrs from the fpr to the vsr registers so the value is
properly accessed by GDB. The functions rs6000_linux_register_to_value,
rs6000_linux_value_to_register, rs6000_linux_value_from_register check if
the value is an IEEE 128-bit floating point value and adjust the register
number as needed. The test in function rs6000_convert_register_p is fixed
so it is only true for floating point values.
This patch fixes three regression tests in gdb.base/store.exp.
The patch has been tested on Power 8 LE/BE, Power 9 LE/BE and Power 10 LE
with no regressions.
Overview of issues fixed by the patch.
The primary issue this patch fixes is the DWARF register mapping for
Linux. The changes in ppc-linux-tdep.c fix the DWARF register mapping
issues. The register mapping issue is responsible for two of the
five regression bugs seen in gdb.base/store.exp.
Once the register mapping was fixed, an underlying issue with the unwinding
of the signal trampoline in common-code in ifrun.c was found. This
underlying bug is best described by Ulrich in the following description.
The unwinder bug shows up on platforms where the kernel uses a trampoline
to dispatch "calls to" the signal handler (not just *returns from* the
signal handler). Many platforms use a trampoline for signal return, and
that is working fine, but the only platform I'm (Ulrich) aware of that
uses a trampoline for signal handler calls is (recent kernels for)
PowerPC. I believe the rationale for using a trampoline here
is to improve performance by avoiding unbalancing of the
branch predictor's call/return stack.
However, on PowerPC the bug is dormant as well as it is hidden
by *another* bug that prevents correct unwinding out of the
signal trampoline. This is because the custom CFI for the
trampoline uses a register number (VSCR) that is not ever used
by compiler-generated CFI, and that particular register is
mapped to an invalid number by the current PowerPC DWARF mapper.
The underlying unwinder bug is exposed by the "new" regression failures
in gdb.base/sigstep.exp. These failures were previously masked by
the fact that GDB was not seeing a valid frame when it tried to unwind
the frames. The sigstep.exp test is specifically testing stepping into
a signal handler. With the correct DWARF register mapping in place,
specifically the VSCR mapping, the signal trampoline code now unwinds to a
valid frame exposing the pre-existing bug in how the signal handler on
PowerPC works. The one line change infrun.c fixes the exiting bug in
the common-code for platforms that use a trampoline to dispatch calls
to the signal handler by not stopping in the SIGTRAMP_FRAME.
Detailed description of the DWARF register mapping fix.
The PowerPC DWARF register mapping is the same for the .eh_frame and
.debug_frame on Linux. PowerPC uses different mapping for .eh_frame and
.debug_frame on other operating systems. The current GDB support for
mapping the DWARF registers in rs6000_linux_dwarf2_reg_to_regnum and
rs6000_adjust_frame_regnum file gdb/rs6000-tdep.c is not correct for Linux.
The files have some legacy mappings for spe_acc, spefscr, EV which was
removed from GCC in 2017.
This patch adds a two new functions rs6000_linux_dwarf2_reg_to_regnum,
and rs6000_linux_adjust_frame_regnum in file gdb/ppc-linux-tdep.c to handle
the DWARF register mappings on Linux. Function
rs6000_linux_dwarf2_reg_to_regnum is installed for both gdb_dwarf_to_regnum
and gdbarch_stab_reg_to_regnum since the mappings are the same.
The ppc_linux_init_abi function in gdb/ppc-linux-tdep.c is updated to
call set_gdbarch_dwarf2_reg_to_regnum map the new function
rs6000_linux_dwarf2_reg_to_regnum for the architecture. Similarly,
dwarf2_frame_set_adjust_regnum is called to map
rs6000_linux_adjust_frame_regnum into the architecture.
Additional detail on the signal handling fix.
The specific sequence of events for handling a signal on most
architectures is as follows:
1) Some code is running when a signal arrives.
2) The kernel handles the signal and dispatches to the handler.
...
However on PowerPC the sequence of events is:
1) Some code is running when a signal arrives.
2) The kernel handles the signal and dispatches to the trampoline.
3) The trampoline performs a normal function call to the handler.
...
We want the "nexti" to step into, not over, signal handlers invoked by
the kernel. This is the case for most platforms as the kernel puts a
signal trampoline frame onto the stack to handle proper return after the
handler. However, on some platforms such as PowerPC, the kernel actually
uses a trampoline to handle *invocation* of the handler. We do not
want GDB to stop in the SIGTRAMP_FRAME. The issue is fixed in function
process_event_stop_test by adding a check that the frame is not a
SIGTRAMP_FRAME to the if statement to stop in a subroutine call. This
prevents GDB from erroneously detecting the trampoline invocation as a
subroutine call.
This patch fixes two regression test failures in gdb.base/store.exp.
The patch then fixes an exposed, dormant, signal handling issue that
is exposed in the signal handling test gdb.base/sigstep.exp.
The patch has been tested on Power 8 LE/BE, Power 9 LE/BE, Power 10 with
no new regressions. Note, only two of the five failures in store.exp
are fixed. The remaining three failures are fixed in a following
patch.
I noticed that if GDB is using a remote or extended-remote target,
then, if an inferior call caused a new thread to appear, or for an
existing thread to exit, then these events are not reported to the
user.
The problem is that for these targets GDB relies on a call to
update_thread_list to learn about changes to the inferior's thread
list.
If GDB doesn't pass through the normal stop code then GDB will not
call update_thread_list, and so will not report changes in the thread
list.
This commit adds an additional update_thread_list call, after which
thread events are correctly reported.
I noticed that sometimes the value returned by $_inferior_thread_count
can become out of sync with the actual thread count of the inferior,
and will disagree with the number of threads reported by 'info
threads'. This commit fixes this issue.
The cause of the problem is that 'info threads' includes a call to
update_thread_list, this can be seen in print_thread_info_1 in
thread.c, while $_inferior_thread_count doesn't include a similar
call, see the function inferior_thread_count_make_value also in
thread.c.
Of course, this is only a problem when GDB is running on a target that
relies on update_thread_list calls to learn about new threads,
e.g. remote or extended-remote targets. Native targets generally
learn about new threads as soon as they appear and will not have this
problem.
I ran into this issue when writing a test for the next commit which
uses inferior function calls to add an remove threads from an
inferior. But for testing I've made use of non-stop mode and
asynchronous inferior execution; by reading the inferior state I can
know when a new thread has been created, at which point I can print
$_inferior_thread_count while the inferior is still running. This is
important, if I stop the inferior then GDB will pass through an
update_thread_list call in the normal stop code, which will
synchronise the thread list, after which $_inferior_thread_count will
report the correct value.
With this change in place $_inferior_thread_count is now correct.
Add a new command completer function for the disassemble command.
There are two things that this completion function changes. First,
after the previous commit, the new function calls skip_over_slash_fmt,
which means that hitting tab after entering a /OPT flag now inserts a
space ready to start typing the address to disassemble at:
(gdb) disassemble /r<TAB>
(gdb) disassemble /r <CURSOR>
But also, we now get symbol completion after a /OPT option set,
previously this would do nothing:
(gdb) disassemble /r mai<TAB>
But now:
(gdb) disassemble /r mai<TAB>
(gdb) disassemble /r main <CURSOR>
Which was my main motivation for working on this commit.
However, I have made a second change in the completion function.
Currently, the disassemble command calls the generic
location_completer function, however, the disassemble docs say:
Note that the 'disassemble' command's address arguments are specified
using expressions in your programming language (*note Expressions:
Expressions.), not location specs (*note Location Specifications::).
So, for example, if you want to disassemble function 'bar' in file
'foo.c', you must type 'disassemble 'foo.c'::bar' and not 'disassemble
foo.c:bar'.
And indeed, if I try:
(gdb) disassemble hello.c:main
No symbol "hello" in current context.
(gdb) disassemble hello.c::main
No symbol "hello" in current context.
(gdb) disassemble 'hello.c'::main
Dump of assembler code for function main:
... snip ...
But, if I do this:
(gdb) disassemble hell<TAB>
(gdb) disassemble hello.c:<CURSOR>
which is a consequence of using the location_completer function. So
in this commit, after calling skip_over_slash_fmt, I forward the bulk
of the disassemble command completion to expression_completer. Now
when I try this:
(gdb) disassemble hell<TAB>
gives nothing, which I think is an improvement. There is one slight
disappointment, if I do:
(gdb) disassemble 'hell<TAB>
I still get nothing. I had hoped that this would expand to:
'hello.c':: but I guess this is a limitation of the current
expression_completer implementation, however, I don't think this is a
regression, the previous expansion was just wrong. Fixing
expression_completer is out of scope for this commit.
I've added some disassembler command completion tests, and also a test
that disassembling using 'FILE'::FUNC syntax works, as I don't think
that is tested anywhere.
Move the function skip_over_slash_fmt into completer.c, and make it
extern, with a declaration in completer.h.
This is a refactor in order to support the next commit. I've not
changed any of the code in skip_over_slash_fmt.
There should be no user visible changes after this commit.
The disassembler gained a new /b flag in this commit:
commit d4ce49b7ac
Date: Tue Jun 21 20:23:35 2022 +0100
gdb: disassembler opcode display formatting
The /b and /r flags result in the instruction opcodes displayed in
different formats, so it's not possible to have both at the same
time. Currently the /b flag overrides the /r flag.
We have a similar situation with the /m and /s flags, but here, if the
user tries to use both flags then they will get an error.
I think the error is clearer, so in this commit I propose that we add
an error if /r and /b are both used.
Obviously this change breaks backwards compatibility. I don't have a
compelling argument for why we should make the change beyond my
feeling that it was a mistake not to add this error from the start,
and that the new behaviour is better.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
In a safety context, it could interesting to track the trampolines being
generated, ensuring there are expected or not.
bfd/ChangeLog:
* elf32-ppc.c (ppc_elf_relax_section): Log branch fixups.
ld/ChangeLog:
* ld.texi (--print-map): Add new item about fixups.
This patch provides some minimal thread-safety to BFD.
The BFD client can request thread-safety by providing a lock and
unlock function. The globals used during BFD creation (e.g.,
bfd_id_counter) are then locked, and the file descriptor cache is also
locked. A function to clean up any thread-local data is now provided
for BFD clients.
* bfd-in2.h: Regenerate.
* bfd.c (lock_fn, unlock_fn): New globals.
(bfd_thread_init, bfd_thread_cleanup, bfd_lock, bfd_unlock): New
functions.
* cache.c (bfd_cache_lookup_worker): Use _bfd_open_file_unlocked.
(cache_btell, cache_bseek, cache_bread, cache_bwrite): Lock
and unlock.
(cache_bclose): Add comment.
(cache_bflush, cache_bstat, cache_bmmap): Lock and unlock.
(_bfd_cache_init_unlocked): New function.
(bfd_cache_init): Use it. Lock and unlock.
(_bfd_cache_close_unlocked): New function.
(bfd_cache_close, bfd_cache_close_all): Use it. Lock and unlock.
(_bfd_open_file_unlocked): New function.
(bfd_open_file): Use it. Lock and unlock.
* doc/bfd.texi (BFD front end): Add Threading menu item.
* libbfd.h: Regenerate.
* opncls.c (_bfd_new_bfd): Lock and unlock.
* po/bfd.pot: Regenerate.
This makes _bfd_error_buf static and adds a way to clear it. I felt
that this made the subsequent patches a little cleaner.
* bfd.c (_bfd_error_buf): Now static.
(bfd_set_input_error): Use _bfd_clear_error_data.
(_bfd_clear_error_data): New function.
(bfd_init): Use _bfd_clear_error_data.
* libbfd.h: Regenerate.
* opncls.c (bfd_close_all_done): Use _bfd_clear_error_data.
* po/bfd.pot: Regenerate.
