Building on an x86_64-linux system with --enable-targets=all fails on the
sparc code with a fortify error:
cc1: warnings being treated as errors
In file included from /usr/include/string.h:640,
from gnulib/string.h:23,
from ../../gdb/gdb_string.h:25,
from ../../gdb/vec.h:25,
from ../../gdb/memattr.h:24,
from ../../gdb/target.h:60,
from ../../gdb/exec.h:23,
from ../../gdb/gdbcore.h:31,
from ../../gdb/sparc-tdep.c:29:
In function 'memcpy',
inlined from 'sparc32_store_return_value' at ../../gdb/sparc-tdep.c:1112,
inlined from 'sparc32_return_value' at ../../gdb/sparc-tdep.c:1170:
/usr/include/bits/string3.h:52: error: call to __builtin___memcpy_chk will
always overflow destination buffer
make: *** [sparc-tdep.o] Error 1
This is due to the gcc optimizer bug PR37060, so tweak the gdb_assert ()
to avoid the issue.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
* cp-valprint.c (cp_print_value_fields): Replaced obstack_base()
method of popping recursion-detection stack with a method based on
obstack_object_size().
* gdb.cp/Makefile.in (EXECUTABLES): Added pr9167.
* gdb.cp/pr9167.cc: New file.
* gdb.cp/pr9167.exp: New file.
Thread support currently does not work with core files. Note that,
in order to thread support to work on tru64, one need to allow GDB
to write in the core file (this is because the thread debug library
needs to write).
An obvious visible symptom of the problem is that "info threads" does
not list the various threads:
(gdb) info threads
* 1 <main task> 0x000003ff805c0918 in __nxm_thread_kill ()
from /usr/shlib/libpthread.so
One other noticeable consequence is that GDB generates some warnings
when using "info tasks":
(gdb) info tasks
warning: Could not find thread id from THREAD = 0x3ffc01b6000
warning: Could not find thread id from THREAD = 0x20000e2b4c0
ID TID P-ID Pri State Name
* 1 140051000 0 30 Runnable main_task
* 2 14005c000 1 30 Accept or Select Term my_t
(notice also how both tasks are marked as being the active task,
which cannot be true).
The problem is that the dec-thread module has not updated its thread list
after the core file got loaded. In fact, the list only gets resync'ed
at the end of each target-wait. The solution was to implement the
find_new_threads target_ops method.
gdb/ChangeLog:
Implement thread support with core files on alpha-tru64.
* dec-thread.c (dec_thread_find_new_threads): New function,
extracted from resync_thread_list.
(resync_thread_list): Add OPS parameter. Replace extracted-out
code by call to dec_thread_find_new_threads.
(dec_thread_wait): Update call to resync_thread_list.
(init_dec_thread_ops): Set dec_thread_ops.to_find_new_threads.
On AVR, the gdb view of an address is different from the machine view of the
same address. We need to use special machinery implemented by value_pointer
to take the pointer of a value.
For instance, considering the following function...
procedure Trace (Unit : T; Message : String);
... where T is an access Integer (a pointer to an integer), call to this
function currently triggers the following warnings:
(gdb) call debug.trace (me, "You")
warning: Value does not fit in 16 bits.
warning: Value does not fit in 16 bits.
Tracing message: You
It could have been worse if Trace actually tried to dereference the Unit
argument...
gdb/ChangeLog (from Tristan Gingold):
* ada-lang.c (value_pointer): New function.
(make_array_descriptor): Call value_pointer to convert addresses to
pointers.
Tested on avr and x86_64-linux.
A long time ago (Oct 2009), I noticed a problem on AIX, where something
failed with an error while the debugger was checking whether an address
was a descriptor or not. Unfortunately, like an idiot, I forgot to write
notes about the scenario where the problem occured - I am usually pretty
meticulous about that because my memory of these things is really bad.
I hope you'll forgive me for not providing a solid testcase - if it's
any consolation, I've searched for a long time before giving up :-(.
Based on the testsuite reports that I have, I think that this happened
while inserting a breakpoint, as follow:
(gdb) break x
Cannot access memory at address 0x200093b4
What happened is that rs6000_convert_from_func_ptr_addr tried to read
the memory at the given address, and fail because of an exception.
It seems pretty clear that, if the address was in fact a descriptor,
GDB would have been able to read the target memory region.
So this patch protects the memory-read against exceptions, and treats
such exceptions as an indication that our address is not a descriptor.
gdb/ChangeLog:
* rs6000-aix-tdep.c: #include exceptions.h.
(rs6000_convert_from_func_ptr_addr): If an exception is thrown
while reading the memory at ADDR, then ADDR cannot be a function
descriptor.
This implements a rudimentary version of the la_print_typedef method
for Ada. Ada usually does not use typedefs, but there is one exception:
pointers to unconstrained arrays. Without this patch, we sometimes
get an error in the "info types" output:
(gdb) info types new_integer_type
All types matching regular expression "new_integer_type":
File foo.adb:
Language not supported.
For now, we treat the typedef as if it did not exist - using the
underlying type instead. This is the right thing to do for most cases,
the only exception being access to array types. Since we already have
a general issue in handling these pointers (we confuse them with fat
pointers), we will enhance ada_print_typedef to handle these pointers
at the same time we address the general issue.
gdb/ChangeLog:
* ada-typeprint.c (ada_print_typedef): New function.
* ada-lang.h (ada_print_typedef): Add declaration.
* ada-lang.c (ada_language_defn): set la_print_typdef field
to ada_print_typedef.
gdb/testsuite/ChangeLog:
* info_types.c, info_types.exp: New files.
Tested on x86_64-linux.
The procfs_address_to_host_pointer function was not used outside of
alpha-tru64, and thus was triggering a compiler warning. Adjusted
accordingly.
gdb/ChangeLog:
* procfs.c (procfs_address_to_host_pointer): Only define when used.
The function proc_find_memory_regions calls iterate_over_mappings as
follow:
> return iterate_over_mappings (pi, func, data,
> find_memory_regions_callback);
The problem is that both func and find_memory_regions_callback
do not match the profile expected by iterate_over_mappings:
> iterate_over_mappings (procinfo *pi, int (*child_func) (), void *data,
> int (*func) (struct prmap *map,
> int (*child_func) (),
> void *data))
We cannot change proc_find_memory_regions such that FUNC is a pointer
to a function that takes no argument (in place of the 6 that it has).
This is because proc_find_memory_regions is used as a target_ops method.
However, it turns out that changing iterate_over_mappings to conform
to the profile imposed by the target_ops vector is possible without
much effort.
gdb/ChangeLog:
* procfs.c (iterate_over_mappings_cb_ftype): New typedef.
(iterate_over_mappings): Adjust function profile. Add declaration.
(insert_dbx_link_bpt_in_region, info_mappings_callback):
Adjust accordingly.
There are currently 2 issues with the placement of this routine:
- It's defined after it is being used (causing an implicit declaration);
- It looks like it's being defined all the time, whereas it is used
only on mips-irix (AFAICT) - shouldn't have this triggered a warning
on sparc-solaris, for instance???
In any case, this patch moves this function up, inside the right region,
just before the function where it is actually used.
gdb/ChangeLog:
* procfs.c (solib_mappings_callback): Move function up to avoid
a compiler warning.
These two functions are only used from procfs_make_note_section, which
itself is only defined if:
#if defined (UNIXWARE) || defined (PIOCOPENLWP) || defined (PCAGENT)
So these two functions are unused on mips-irix, and they get flagged
by -Wunused-function. This patch simply moves these functions closer
to the function that uses them, which also has the effect of putting
them inside the same #if block as procfs_make_note_section. Thus
they are defined only when used.
gdb/ChangeLog:
* procfs.c (find_signalled_thread, find_stop_signal): Move
these functions down to define them only when used.
The problem is printing the wrong value for dynamic local variables
when using the "info locals" command. Consider the following code:
procedure Print (I1 : Positive; I2 : Positive) is
type My_String is array (I1 .. I2) of Character;
I : My_String := (others => 'A');
S : String (1 .. I2 + 3) := (others => ' ');
begin
S (I1 .. I2) := String (I); -- BREAK
Put_Line (S);
end Print;
After the debugger stopped at BREAK, we try printing all local variables.
Here is what we get:
(gdb) info locals
i = "["00"]["00"]"
s = "["00"]["00"]["00"]["00"]["00"]["00"]["00"]["00"]"
Curiously, printing their value using the "print" command works:
(gdb) print i
$1 = "AA"
(gdb) print s
$2 = " "
We traced the problem to trying to get the contents of a variable
(call to value_contents) before "fix'ing" it. For those not familiar
with the Ada language support, "fixing" a value consists of swapping
the value's dynamic type with a static version that is appropriate
for our actual value. As a result, the dynamic type was used to
determine the value size, which is zero, and thus the value contents
was empty.
gdb/ChangeLog:
* valprint.c (common_val_print): Fix the value before extracting
its contents.
* ada-lang.c (ada_to_fixed_value): Make this function extern.
* ada-lang.h (ada_to_fixed_value): New function declaration.
* ada-valprint.c (ada_value_print): Use ada_to_fixed_value
to avoid code duplication and fix a bug in the handling of
fixed types contents.
gdb/testsuite/ChangeLog:
* gdb.ada/dyn_loc: New testcase.
2010-04-20 Jan Kratochvil <jan.kratochvil@redhat.com>
* f-exp.y: Add new production to recognize the `logical*8' type.
(LOGICAL_S8_KEYWORD): New token.
* f-lang.c (enum f_primitive_types)
<f_primitive_type_logical_s8>: New field.
(f_language_arch_info): Handling `logical*8' type.
(build_fortran_types): Building `logical*8' type.
* f-lang.h (struct builtin_f_type) <builtin_logical_s8>: New field.
gdb/testsuite/ChangeLog:
2010-04-20 Jan Kratochvil <jan.kratochvil@redhat.com>
Sergio Durigan Junior <sergiodj@redhat.com>
* gdb.fortran/logical.exp: New testcase.
* gdb.fortran/logical.f90: New file.
(CONTEXT_EXTENDED_REGISTERS): Set macro to zero if not exisiting.
(init_registers_amd64): Declare.
(mappings): Add 64-bit version of array.
(init_windows_x86): New function.
(the_low_target): Change init_arch field to init_windows_x86.
(child_xfer_memory): Use uintptr_t type for local variable `addr'.
(get_image_name): Use SIZE_T type for local variable `done'.
(psapi_get_dll_name): Use LPVOID type for parameter `BaseAddress'.
(toolhelp_get_dll_name): Idem.
(handle_load_dll): Use CORE_ADDR type for local variable `load_addr'.
Use uintptr_t typecast to avoid warning.
(handle_unload_dll): Use uintptr_t typecast to avoid warning.
(handle_exception): Use phex_nz to avoid warning.
(win32_wait): Remove unused local variable `process'.
Vladimir Prus <vladimir@codesourcery.com>
* tracepoint.c (tfind_1): Add missing newline, report exit from
tfind mode as such.
* target.c (update_current_target): Make default
to_trace_find return -1.
The current ObjC logic will check both the symbol name and the target
address space when trying to locate an appropriate selector. The problem
is that first the target address space is checked before the symbol name.
This may lead to a lot of unnecessary host<->target transactions when
dealing with a non-OjbC target that does use function descriptors to
describe functions as every symbol will have its FD read just to have the
result thrown away with non-matching symbol names.
It also may lead to problems when a non-FD symbol is found that points near
the end of the address space as the target will throw up a memory_error().
One such example are symbols that are not functions, smaller than a FD,
and are the last valid location. Obviously treating it as a larger data
struct can cause memory overflows.
So to speed things up and not screw over such targets, check the symbol
name (which we already have locally) first before attempting to read the
function's descriptor. This fixes breakpoints with Blackfin Linux FDPIC
ELFs, and seems to cause no native regressions on my x86_64/Linux system.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
* gdb.base/solib-display.exp: Replace gdb_exit, gdb_start,
gdb_reinitialize_dir and gdb_load by clean_restart. Remove trailing
gdb_exit and return.
(executable): New variable.
(binfile): Use it.
Implement `save-breakpoints'.
gdb/
* breakpoint.c (save_cmdlist): New.
(breakpoint_set_cmdlist, breakpoint_show_cmdlist): Moved up close
to save_cmdlist.
(print_recreate_catch_fork): New.
(catch_fork_breakpoint_ops): Install it.
(print_recreate_catch_vfork): New.
(catch_vfork_breakpoint_ops): Install it.
(print_recreate_catch_syscall): New.
(catch_syscall_breakpoint_ops): Install it.
(print_recreate_catch_exec): New.
(catch_exec_breakpoint_ops): Install it.
(print_recreate_exception_catchpoint): New.
(gnu_v3_exception_catchpoint_ops): Install it.
(save_breakpoints): New, based on tracepoint_save_command, but
handle all breakpoint types.
(save_breakpoints_command): New.
(tracepoint_save_command): Rename to...
(save_tracepoints_command): ... this, and reimplement using
save_breakpoints.
(save_command): New.
(_initialize_breakpoints): Install the "save" command prefix.
Install the "save breakpoints" command. Make "save-tracepoints" a
deprecated alias for "save tracepoints".
* breakpoint.h (struct breakpoint_ops): New field `print_recreate'.
* ada-lang.c (print_recreate_exception): New.
(print_recreate_catch_exception): New.
(catch_exception_breakpoint_ops): Install it.
(print_recreate_catch_exception_unhandled): New.
(catch_exception_unhandled_breakpoint_ops): Install it.
(print_recreate_catch_assert): New.
(catch_assert_breakpoint_ops): Install it.
* NEWS: Mention the new `save breakpoints' command. Mention the
new `save tracepoints' alias and that `save-tracepoints' is now
deprecated.
gdb/doc/
* gdb.texinfo (Save Breakpoints): New node.
(save-tracepoints): Rename to ...
(save tracepoints): ... this. Mention that `save-tracepoints' is
a deprecated alias to `save tracepoints'.
gdb/testsuite/
* gdb.trace/save-trace.exp: Adjust.
* tui/tui-out.c: Include cli-out.h.
(tui_table_begin, tui_table_body, tui_table_end, tui_table_header)
(tui_begin, tui_end, tui_field_int, tui_field_skip)
(tui_field_string, tui_field_fmt, tui_spaces, tui_text)
(tui_message, tui_wrap_hint, tui_flush): Delete forward
declarations.
(struct ui_out_data): Rename to...
(struct tui_ui_out_data): ... this. Remove `stream' and
`suppress_output' fields, and inherit cli_ui_out_data.
(tui_out_data): New typedef.
(tui_ui_out_impl): Don't initialize fields staticaly.
(tui_table_begin, tui_table_body, tui_table_end, tui_table_header)
(tui_begin, tui_end): Delete.
(tui_field_int): Adjust to delegate most work to the base type.
(tui_field_skip): Delete.
(tui_field_string, tui_field_fmt): Adjust comment. Adjust to
delegate most work to the base type.
(tui_spaces): Delete.
(tui_text): Adjust to delegate most work to the base type.
(tui_message): Delete.
(tui_wrap_hint): Delete.
(tui_flush): Delete.
(out_field_fmt): Delete.
(field_separator): Delete.
(tui_out_new): Adjust to initialize the base type.
(_initialize_tui_out): Initialize tui_ui_out_impl.
* cli-out.c (struct ui_out_data): Moved out to cli-out.h, renamed
cli_ui_out_data.
(cli_out_data): Adjust.
(cli_ui_out_impl): Make extern.
(cli_table_header, cli_field_int, cli_field_skip): Use
uo_field_string instead of cli_field_string.
(cli_redirect): Adjust to use cli_out_data.
(cli_out_data_ctor): New.
(cli_out_new): Use it.
* cli-out.h (struct ui_file): Remove forward declaration.
(struct cli_ui_out_data): New, moved from cli-out.c, and renamed.
(cli_ui_out_impl): Declare.
(cli_out_data_ctor): Declare.
* ui-out.c (struct ui_out) <data>: Change type to void pointer.
(uo_field_string): No longer static.
(ui_out_data): Change return type to void pointer.
(ui_out_new): Change `data' parameter type to void pointer.
* ui-out.h (struct ui_out_data): Don't forward declare.
(ui_out_data): Change return type to void pointer.
(ui_out_new): Change `data' parameter type to void pointer.
(uo_field_string): Declare.
is_hardware_watchpoint.
(watchpoints_triggered): Ditto.
(bpstat_check_location): Use is_watchpoint and
is_hardware_watchpoint.
(bpstat_check_watchpoint): Use is_watchpoint and
is_hardware_watchpoint.
(bpstat_stop_status): Fix comment.
(user_settable_breakpoint): Use is_watchpoint.
(hw_watchpoint_used_count): Use is_hardware_watchpoint.
(disable_watchpoints_before_interactive_call_start): Use
is_watchpoint.
(enable_watchpoints_after_interactive_call_stop): Use
is_watchpoint.
(clear_command): Use is_watchpoint.
(do_enable_breakpoint): Use is_watchpoint.
The recent change to reload_shared_libraries() broke FDPIC shared libraries as
the solib-frv.c code was implicitly relying on the initial order of calls
(first solib_addr() and then solib_create_inferior_hook()). It was
maintaining internal state via enable_break{1,2}_done to handle this.
While I could tweak these values a bit more, the original code wasn't terribly
bullet proof -- if during the initial debug you attempted to view shared
libraries, the enable2_break() code would whine about the ldso internal debug
addresses being unfetchable (and would actually attempt to read address 0x8 on
the target). So I've dropped this implicit dependency on order (i.e.
enable_break1_done) and updated the ldso poking code (i.e. enable_break2) to
silently return when the internal debug address is still set to 0. It will
remain this way until the ldso gets a chance to initialize at which point the
code will act the same as before.
While I have no way of testing the FRV, the Blackfin FDPIC code is using this
same base in a 100% copy & paste method since we implemented FDPIC the same
way as the FRV guys (I'll address this in the future). This fix was required
in order to handle shared libraries with Blackfin FDPIC properly, and I see no
reason why it wouldn't also work for FRV (since the uClibc ldso FDPIC code is
the same too and that's really what this is poking).
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Mike Frysinger
Chris Moller
Aleksandar Ristovski
Pierre Muller
Doug Evans
gdbadmin
Joel Brobecker
Tom Tromey
Mark Kettenis
Stan Shebs
Sergio Durigan Junior
Pedro Alves
Jan Kratochvil
H.J. Lu