Yao writes:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
GDB gets an internal error when it connects to GDBserver started with
'--disable-packet=qC'.
Sending packet: $QNonStop:0#8c...Packet received: OK
Sending packet: $?#3f...Packet received: T0505:00000000;04:00f0ffbf;08:b0c2e44c;thread:p4255.4255;core:1;
Sending packet: $Hc-1#09...Packet received: E01
Sending packet: $qC#b4...Packet received:
Sending packet: $qAttached:a410#bf...Packet received: E01
Packet qAttached (query-attached) is supported
warning: Remote failure reply: E01
Sending packet: $qOffsets#4b...Packet received:
../../../git/gdb/target.c:3248: internal-error: Can't determine the current address space of thread Thread 16981
When start remote, the call chain is as follows,
remote_start_remote
add_current_inferior_and_thread <--[1]
...
start_remote
wait_for_inferior
remote_wait_as
process_stop_reply
get_thread_arch_regcache <--[2]
remote_notice_new_inferior <--[3]
GDB sends packet "qC" in [1] and adds the thread/inferior if the remote
stubs understands "qC". In [2], GDB looks for the inferior to build a
regcache, and notices a new inferior in [3]. As we can see, GDB assumes
that the inferior can be found in [2]. Once the remote stub doesn't
support "qC", GDB can't look for the inferior in [2], and emits an
internal error.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Right after the initial connection, we query the target for its state,
with the ? packet. We store the resulting wait status / stop reply
aside, and query the target for the current thread, using qC, which
fails, so we fake a ptid for the target's thread. We then later,
after the initial setup, end up consuming that set-aside wait status,
parsing the T stop reply, which contains a "thread" "register" (which
was the thread the target would have replied to qC). We get into
trouble because the ptid in that stop reply doesn't match our faked up
ptid in the initial setup, although the target threads are the same...
So we had the T stop reply handy all along. We might as well extract
the thread's ptid from it, and avoid all the resulting issues.
qC is also used after vRun, in order to discover the new process'es
main thread. But, vRun's reply is also a wait status, just like
'?''s, which is quite convenient.
This means that if we have a "Txx thread: ptid" reply, then we don't
really need qC. The patch makes GDB look in the T reply first, and if
not found, try with qC. The packet handling seems to have been added
in gdb-4.18 (1999), and I see that in that same release, "Txx thread:
ptid" didn't exist yet, which probably explains why nobody though of
doing this before.
Regression tested against a gdbserver with qC disabled (and then
enabled), on x86_64 Fedora 17.
2013-01-25 Pedro Alves <palves@redhat.com>
* remote.c (stop_reply_extract_thread): New.
(add_current_inferior_and_thread): New parameter 'wait_status'.
Handle it.
(remote_start_remote): Pass wait status to
add_current_inferior_and_thread.
(extended_remote_run): Update comment.
(extended_remote_create_inferior_1): Pass wait status to
add_current_inferior_and_thread.
gdb/ChangeLog
* valarith.c (value_vector_widen): New function for replicating a
scalar into a vector.
(value_binop): Use value_vector_widen to widen scalar to vector
rather than casting, this better matches gcc C behaviour.
* valops.c (value_casst): Update logic for casting between vector
types, and for casting from scalar to vector, try to match gcc C
behaviour.
* value.h (value_vector_widen): Declare.
* opencl-lang.c (opencl_value_cast): New opencl specific casting
function, handle special case for casting scalar to vector.
(opencl_relop): Use opencl_value_cast.
(evaluate_subexp_opencl): Use opencl_value_cast instead of
value_cast, and handle BINOP_ASSIGN, UNOP_CAST, and UNOP_CAST_TYPE
in order to use opencl_value_cast.
gdb/testsuite/ChangeLog
* gdb.base/gnu_vector.c: New variable for use in tests.
* gdb.base/gnu_vector.exp: Update and extend tests to reflect
changes in scalar to vector casting and widening.
* gdb.python/py-type.c: New variables for use in tests.
* gdb.python/py-type.exp: Update vector related tests to reflect
changes in scalar to vector casting and widening.
argument.
(%union): Add new type bfd_vma as vma.
(VMA): New rule.
(opt_base): Use VMA instead of NUMBER rule to evaluate value.
(def_file_print): Use bfd's fprintf_vma to output base-address.
* symtab.c (skip_prologue_using_sal): Consider a file
change the same as an increased line number
gdb/testsuite/
* gdb.base/prologue-include.c: New file.
* gdb.base/prologue-include.exp: New file.
* gdb.base/prologue-include.h: New file.
2013-01-24 Sriraman Tallam <tmsriram@google.com>
* layout.cc (Layout::layout): Check for option text_reorder.
(Layout::make_output_section): Ditto.
* options.h (text_reorder): New option.
* output.cc (Input_section_sort_compare): Remove special ordering
of section names.
(Output_section::
Input_section_sort_section_name_special_ordering_compare::
operator()): New function.
(Output_section::sort_attached_input_sections): Use new sort function
for .text.
* output.h (Input_section_sort_section_name_special_ordering_compare):
New struct.
* testsuite/Makefile.am (text_section_grouping): Test option
--no-text-reorder
* testsuite/Makefile.in: Regenerate.
* testsuite/text_section_grouping.sh: Check order of functions without
default text reordering.
internal representation of architecture via GDB Python API.
* Makefile.in: Add entries corresponding to the new file
python/py-arch.c.
* NEWS (Python Scripting): Add entries for the new class
gdb.Architecture and the new method gdb.Frame.architecture.
* python/py-arch.c: Implement gdb.Architecture class.
* python/py-frame.c (frapy_arch): Implement the method
gdb.Frame.architecture().
(frame_object_methods): Add 'architecture' to the method table.
* python/python-internal.h: Add declarations of new utility
functions.
* python/python.c (_initialize_python): Initialize
gdb.Architecture class.
* doc/gdb.texinfo (Architectures In Python): New sub-sub-section
describing the gdb.Architecture class.
(Frames In Python): Add description about the new method
gdb.Frame.architecture().
* testsuite/gdb.python/frame.exp: Add a test for
gdb.Frame.architecture() method.
* dwarf2read.c (dwarf2_per_cu_data): Split imported_symtabs and
type_unit_group out of union s. All uses updated.
(read_index_from_section): Watch for index version 8.
(follow_die_sig): If using .gdb_index version <= 7, record the TU as
an imported symtab.
(write_psymtabs_to_index): Increment version number to 8.
doc/
* gdb.texinfo (Index Section Format): Document .gdb_index version 8.
bothering the frontend about it... This is the exact same check MI
does.
I also smoke tested Emacs 23 in gud-gdb mode, both annotations=2
and annotations=3. I didn't notice anything break.
gdb/
2013-01-22 Pedro Alves <palves@redhat.com>
* annotate.c (breakpoint_changed): Skip if breakpoint is not
user-visible.
gdb/testsuite/
2013-01-22 Pedro Alves <palves@redhat.com>
* gdb.base/annota1.exp (signal sent): No longer expect
breakpoints-invalid.
* gdb.cp/annota2.exp (continue until exit)
(watch triggered on a.x): Ditto.
observer_notify_breakpoints_changed calls. All, except the
init_raw_breakpoint one. But that one is actually wrong. The
breakpoint is being constructed at that point, and hasn't been placed
on the breakpoint chain yet. It would be better placed in
install_breakpoint, and I actually started out that way. But once the
annotate_breakpoints_changed are parallel to the observer calls, we
can fully move annotations to observers too.
One issue is that this changes the order of annotations a bit.
Before, we'd emit the annotation, and after call "mention()" on the
breakpoint (which prints the breakpoint number, etc.). But, we call
the observers _after_ mention is called, so the annotation output will
change a little:
void
install_breakpoint (int internal, struct breakpoint *b, int update_gll)
{
add_to_breakpoint_chain (b);
set_breakpoint_number (internal, b);
if (is_tracepoint (b))
set_tracepoint_count (breakpoint_count);
if (!internal)
mention (b);
observer_notify_breakpoint_created (b);
if (update_gll)
update_global_location_list (1);
}
I believe this order doesn't really matter (the frontend needs to wait
for the prompt anyway), so I just adjust the expected output in the
tests. Emacs in annotations mode doesn't seem to complain. Couple
that with the previous patch that suppressed duplicated annotations,
and, the fact that some annotations calls were actually missing (were
we do have observer calls), more changes to the tests are needed
anyway.
Tested on x86_64 Fedora 17.
gdb/
2013-01-22 Pedro Alves <palves@redhat.com>
* annotate.c (annotate_breakpoints_changed): Rename to ...
(annotate_breakpoints_invalid): ... this. Make static.
(breakpoint_changed): Adjust.
(_initialize_annotate): Always install the observers. Install a
"breakpoint_created" observer.
* annotate.h (annotate_breakpoints_changed): Delete declaration.
* breakpoint.c (set_breakpoint_condition)
(breakpoint_set_commands, do_map_commands_command)
(init_raw_breakpoint, clear_command, set_ignore_count)
(enable_breakpoint_disp): No longer call
annotate_breakpoints_changed.
gdb/testsuite/
2013-01-22 Pedro Alves <palves@redhat.com>
* gdb.base/annota1.exp (breakpoints_invalid): New variable.
Adjust tests to breakpoints-invalid changes.
* gdb.cp/annota2.exp (breakpoints_invalid, frames_invalid): New
variables.
Adjust tests to breakpoints-invalid changes.
handled, one of those being to place SSS breakpoints on the breakpoint
chain as all other breakpoints, annota1.exp times out with lots and
lots of breakpoint-invalid and frame-changed annotations. All those
extra annotations are actually unnecessary. For one, SSS breakpoints
are internal breakpoints, so the frontend shouldn't care if they were
added, removed or changed. Then, there's really no point in emitting
"breakpoints-invalid" or "frames-invalid" more than once between times
the frontend/user can actually issues GDB commands; the frontend will
have to wait for the GDB prompt to refresh its state, so emitting
those annotations at most once between prompts is enough. Non-stop or
async would complicate this, but no frontend will be using annotations
in those modes (one of goes of emacs switching to MI was non-stop mode
support, AFAIK). The previous patch reveals there has been an
intention in the past to suppress multiple breakpoints-invalid
annotations caused by ignore count changes. As the previous patch
shows, that's always been broken, but in any case, this patch actually
makes it work. The next patch will remove several annotation-specific
calls in breakpoint.c in favor of always using the breakpoint modified
& friends observers, and that causes yet more of these annotations,
because several calls to the corresponding annotate_* functions in
breakpoint.c are missing, particularly in newer code.
So all in all, here's a simple mechanism that avoids sending the same
annotation to the frontend more than once until gdb is ready to accept
further commands.
Tested on x86_64 Fedora 17.
2013-01-22 Pedro Alves <palves@redhat.com>
* annotate.c: Include "inferior.h".
(frames_invalid_emitted)
(breakpoints_invalid_emitted): New globals.
(async_background_execution_p): New function.
(annotate_breakpoints_changed, annotate_frames_invalid): Skip
emitting the annotation if it has already been emitted.
(annotate_display_prompt): New function.
* annotate.h (annotate_display_prompt): New declaration.
* event-top.c: Include annotate.h.
(display_gdb_prompt): Call annotate_display_prompt.
suppress multiple breakpoints-invalid annotations when the ignore
count of a breakpoint changes, up until the target actually stops.
But, the code is bogus:
void
annotate_breakpoints_changed (void)
{
if (annotation_level == 2)
{
target_terminal_ours ();
printf_unfiltered (("\n\032\032breakpoints-invalid\n"));
if (ignore_count_changed)
ignore_count_changed = 0; /* Avoid multiple break annotations. */
}
}
The "ignore_count_changed" flag isn't actually guarding the output of
the annotation at all. It would have been better written something
like:
void
annotate_breakpoints_changed (void)
{
if (annotation_level == 2 && !ignore_count_changed)
{
target_terminal_ours ();
printf_unfiltered (("\n\032\032breakpoints-invalid\n"));
ignore_count_changed = 0; /* Avoid multiple break annotations. */
}
}
but, it wasn't. AFAICS, that goes all the way back to the original
patch'es submission and check in, at
<http://sourceware.org/ml/gdb-patches/1999-q4/msg00106.html>. I
looked a tar of HP's wdb from 1999, and even though that contains
local changes in the annotate code, this suppression seems borked
there too to me.
The original patch added a test to supposedly exercise this
suppression, but, it actually doesn't. It merely tests that
"breakpoints-invalid" is output after "stopped", but doesn't check
whether the duplicates supression actually works (IOW, check that only
_one_ annotation is seen). I was going to simply delete the tests
too, but a following patch will eliminate the duplicates in a
different way (which I needed for a different reason), so instead, I'm
making the tests actually fail if a duplicate annotation is seen.
Worry not, the test doesn't actually fail! The reason is that
breakpoint.c does:
else if (b->ignore_count > 0)
{
b->ignore_count--;
annotate_ignore_count_change ();
bs->stop = 0;
/* Increase the hit count even though we don't stop. */
++(b->hit_count);
observer_notify_breakpoint_modified (b);
}
where the annotate_ignore_count_change call is meant to inform the
"breakpoint_modified" annotation observer to ignore the notification.
All sounds good. But, the trouble is that nowadays annotate.c only
installs the observers if GDB is started with annotations enabled with
a command line option (gdb --annotate=2):
void
_initialize_annotate (void)
{
if (annotation_level == 2)
{
observer_attach_breakpoint_deleted (breakpoint_changed);
observer_attach_breakpoint_modified (breakpoint_changed);
}
}
and annota1.exp, to enable annotations, starts GDB normally, and
afterwards does "set annotate 2", so the observers aren't installed
when annota1.exp is run, and therefore changing the ignore count isn't
triggering any annotation at all...
gdb/
2013-01-22 Pedro Alves <palves@redhat.com>
* annotate.c (ignore_count_changed): Delete.
(annotate_breakpoints_changed): Don't clear ignore_count_changed.
(annotate_ignore_count_change): Delete.
(annotate_stopped): Don't emit a delayed breakpoints-changed
annotation.
* annotate.h (annotate_ignore_count_change): Delete.
* breakpoint.c (bpstat_check_breakpoint_conditions): Don't call
annotate_ignore_count_change.
gdb/testsuite/
2013-01-22 Pedro Alves <palves@redhat.com>
* gdb.base/annota1.exp (annotate ignore count change): Add
expected output for failure case.
* breakpoint.c (print_one_breakpoint_location): Add MI
field 'thread-groups' when printing a breakpoint.
(output_thread_groups): New function.
2013-01-21 Marc Khouzam <marc.khouzam@ericsson.com>
* gdb.texinfo (GDB/MI Breakpoint Commands): Document new
'thread-groups' field when printing a breakpoint in MI.
2013-01-21 Marc Khouzam <marc.khouzam@ericsson.com>
* gdb.mi/mi-break.exp: Expect new 'thread-groups' field.
* gdb.mi/mi-nsmoribund.exp: Expect new 'thread-groups' field.
Also handle 'thread' field.
* gdb.mi/mi-simplerun.exp: Expect new 'thread-groups' field.
* gdb.mi/mi-watch.exp: Ditto.
* lib/mi-support.exp: Ditto.
(CompoundExplorer.explore_expr): Correct the name of a method
being invoked.
(ExploreTypeCommand.invoke): Add a missing 'return'.
* testsuite/gdb.python/py-explore.exp: Improve a test
