platform/upstream/binutils.git
9 years agoconfig/tc-avr.c (md_section_align): Append UL for -1 to avoid the latest gcc's warning
Chen Gang [Sat, 5 Sep 2015 08:02:08 +0000 (16:02 +0800)]
config/tc-avr.c (md_section_align): Append UL for -1 to avoid the latest gcc's warning

The related warning is:

  gcc -DHAVE_CONFIG_H -I. -I../../binutils-gdb/gas  -I. -I../../binutils-gdb/gas -I../bfd -I../../binutils-gdb/gas/config -I../../binutils-gdb/gas/../include -I../../binutils-gdb/gas/.. -I../../binutils-gdb/gas/../bfd -DLOCALEDIR="\"/upstream/release-avr32/share/locale\""  -W -Wall -Wstrict-prototypes -Wmissing-prototypes -Wshadow -Werror -I../../binutils-gdb/gas/../zlib -g -O2 -MT tc-avr.o -MD -MP -MF .deps/tc-avr.Tpo -c -o tc-avr.o `test -f 'config/tc-avr.c' || echo '../../binutils-gdb/gas/'`config/tc-avr.c
../../binutils-gdb/gas/config/tc-avr.c: In function ‘md_section_align’:
../../binutils-gdb/gas/config/tc-avr.c:1233:43: error: left shift of negative value [-Werror=shift-negative-value]
     return ((addr + (1 << align) - 1) & (-1 << align));
                                             ^

2015-09-05  Chen Gang  <gang.chen.5i5j@gmail.com>

* config/tc-avr.c (md_section_align): Append UL for -1 to avoid
the latest gcc's warning.

9 years agoExtend the descriptiopn of how to access linker defined symbols from C.
Nick Clifton [Tue, 8 Sep 2015 08:36:02 +0000 (09:36 +0100)]
Extend the descriptiopn of how to access linker defined symbols from C.

* ld.texinfo (Source Code Reference): Extend description to
include an example using arrays.

9 years agoAutomatic date update in version.in
GDB Administrator [Tue, 8 Sep 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agold: Fix failures in new orphan handling tests.
Andrew Burgess [Mon, 7 Sep 2015 14:39:12 +0000 (15:39 +0100)]
ld: Fix failures in new orphan handling tests.

The new orphan handling tests added in commit c005eb9 fail on a range of
targets.  Some of the failures were fixed in commit e32aa93 but not
all.  This commit should address the remaining failures.

Update results to account for orphan sections being placed in different
orders, and for other, target specific sections, being discarded.

ld/testsuite/ChangeLog:

* ld-elf/orphan-7.map: Allow for other discarded sections.
* ld-elf/orphan-8.map: Updated to allow for different section
ordering on different targets.
* ld-elf/orphan.ld: Place .sbss section.

9 years agoguile: Add as_a_scm_t_subr
Pedro Alves [Mon, 7 Sep 2015 18:34:31 +0000 (19:34 +0100)]
guile: Add as_a_scm_t_subr

Building GDB in C++ mode on Fedora 20, the gdb/guile/ code shows ~280
errors like:

 src/gdb/guile/guile.c:515:1: error: invalid conversion from ‘scm_unused_struct* (*)(SCM, SCM) {aka scm_unused_struct* (*)(scm_unused_struct*, scm_unused_struct*)}’ to ‘scm_t_subr {aka void*}’ [-fpermissive]

This commit fixes them all.

gdb/ChangeLog:
2015-09-07  Pedro Alves  <palves@redhat.com>

* guile/guile-internal.h (as_a_scm_t_subr): New.
* guile/guile.c (misc_guile_functions): Use it.
* guile/scm-arch.c (arch_functions): Use it.
* guile/scm-block.c (block_functions, gdbscm_initialize_blocks):
Use it.
* guile/scm-breakpoint.c (breakpoint_functions): Use it.
* guile/scm-cmd.c (command_functions): Use it.
* guile/scm-disasm.c (disasm_functions): Use it.
* guile/scm-exception.c (exception_functions)
(private_exception_functions): Use it.
* guile/scm-frame.c (frame_functions)
* guile/scm-gsmob.c (gsmob_functions): Use it.
* guile/scm-iterator.c (iterator_functions): Use it.
* guile/scm-lazy-string.c (lazy_string_functions): Use it.
* guile/scm-math.c (math_functions): Use it.
* guile/scm-objfile.c (objfile_functions): Use it.
* guile/scm-param.c (parameter_functions): Use it.
* guile/scm-ports.c (port_functions, private_port_functions): Use
it.
* guile/scm-pretty-print.c (pretty_printer_functions): Use it.
* guile/scm-progspace.c (pspace_functions): Use it.
* guile/scm-string.c (string_functions): Use it.
* guile/scm-symbol.c (symbol_functions): Use it.
* guile/scm-symtab.c (symtab_functions): Use it.
* guile/scm-type.c (type_functions, gdbscm_initialize_types): Use
it.
* guile/scm-value.c (value_functions): Use it.

9 years agoFix internal error caused by IFUNC patch.
Cary Coutant [Mon, 7 Sep 2015 16:44:11 +0000 (09:44 -0700)]
Fix internal error caused by IFUNC patch.

The previous commit to fix PR gold/18886 converted STT_IFUNC
to STT_FUNC when resolving to a symbol defined in a shared library.
This leads to an internal error if the shared library symbol is
seen first, as we do not convert the symbol at all.

We need to override the STT_IFUNC in add_from_dynobj() instead of
in override_base().

gold/
PR gold/18930
PR gold/18886
* resolve.cc (Symbol::override_base): Don't convert IFUNC symbols here.
* symtab.cc (Symbol_table::add_from_dynobj): Convert them here instead.

9 years agoAutomatic date update in version.in
GDB Administrator [Mon, 7 Sep 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoAutomatic date update in version.in
GDB Administrator [Sun, 6 Sep 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoUpdate ld-elf/orphan-8.map to support 32-bit targets
H.J. Lu [Sat, 5 Sep 2015 13:44:53 +0000 (06:44 -0700)]
Update ld-elf/orphan-8.map to support 32-bit targets

* ld-elf/orphan-8.map: Updated to support 32-bit targets.

9 years agoAutomatic date update in version.in
GDB Administrator [Sat, 5 Sep 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agold: Extend options for altering orphan handling behaviour.
Andrew Burgess [Tue, 28 Jul 2015 18:20:37 +0000 (19:20 +0100)]
ld: Extend options for altering orphan handling behaviour.

Replace the options --warn-orphan and --no-warn-orphan with a single
option --orphan-handling=MODE, where mode can be place, warn, error, and
discard.

Mode 'place' is the default, and is the current behaviour, placing the
orphan section into a suitable output section.

Mode 'warn' is the same as '--warn-orphan'.  The orphan is also placed
using the same algorithm as for 'place'.

Mode 'error' is the same as '--warn-orphan' and '--fatal-warnings'.

Mode 'discard' assigns all output sections to the /DISCARD/ section.

ld/ChangeLog:

* ld.h (enum orphan_handling_enum): New.
(ld_config_type): Remove warn_orphan, add orphan_handling.
* ldemul.c (ldemul_place_orphan): Remove warning about orphan
sections.
* ldlang.c (ldlang_place_orphan): New function.
(lang_place_orphans): Call ldlang_place_orphan.
* ldlex.h (enum option_values): Remove OPTION_WARN_ORPHAN and
OPTION_NO_WARN_ORPHAN, add OPTION_ORPHAN_HANDLING.
* lexsup.c (ld_options): Remove 'warn-orphan' and
'no-warn-orphan', add 'orphan-handling'.
(parse_args): Remove handling for OPTION_WARN_ORPHAN and
OPTION_NO_WARN_ORPHAN, add handling for OPTION_ORPHAN_HANDLING.
* NEWS: Replace text about --warn-orphan with --orphan-handling.
* ld.texinfo (Options): Remove --warn-orphan entry and add
entry on --orphan-handling.
(Orphan Sections): Add reference to relevant command line options.

ld/testsuite/ChangeLog:

* ld-elf/elf.exp: Switch to rely on run_dump_test.
* ld-elf/orphan-5.l: Update expected output.
* ld-elf/orphan-5.d: New file.
* ld-elf/orphan-6.d: New file.
* ld-elf/orphan-6.l: New file.
* ld-elf/orphan-7.d: New file.
* ld-elf/orphan-7.map: New file.
* ld-elf/orphan-8.d: New file.
* ld-elf/orphan-8.map: New file.

9 years agogdb/tui: Remove casts of NULL during assignment.
Andrew Burgess [Mon, 31 Aug 2015 20:32:11 +0000 (21:32 +0100)]
gdb/tui: Remove casts of NULL during assignment.

In the following code:

    struct symbol *wsym = (struct symbol *) NULL;

the cast of NULL is redundant, it adds noise, and is just one more thing
to change if the type of wsym ever changes.  There are a relatively
small number of places in gdb where the above code pattern is used.
Usually the cast is removed like this:

    struct symbol *wsym = NULL;

This commit updates all the places within the gdb/tui directory where we
cast NULL during assignment, removing the cast.

gdb/ChangeLog:

* tui/tui-data.c (win_with_focus): Remove cast of NULL pointer.
(tui_next_win): Likewise.
(tui_prev_win): Likewise.
(tui_partial_win_by_name): Likewise.
(tui_init_generic_part): Likewise.
(init_content_element): Likewise.
(tui_del_window): Likewise.
(tui_free_window): Likewise.
(tui_del_data_windows): Likewise.
(tui_free_data_content): Likewise.
* tui/tui-layout.c (make_source_or_disasm_window): Likewise.
* tui/tui-regs.c (tui_show_register_group): Likewise.
* tui/tui-win.c (tui_resize_all): Likewise.
(tui_set_focus): Likewise.
(tui_set_win_height): Likewise.
(make_invisible_and_set_new_height): Likewise.
* tui/tui-windata.c (tui_delete_data_content_windows): Likewise.
* tui/tui-wingeneral.c (make_visible): Likewise.

9 years agogdb/cli: Remove casts of NULL during assignment.
Andrew Burgess [Mon, 31 Aug 2015 20:25:37 +0000 (21:25 +0100)]
gdb/cli: Remove casts of NULL during assignment.

In the following code:

    struct symbol *wsym = (struct symbol *) NULL;

the cast of NULL is redundant, it adds noise, and is just one more thing
to change if the type of wsym ever changes.  There are a relatively
small number of places in gdb where the above code pattern is used.
Usually the cast is removed like this:

    struct symbol *wsym = NULL;

This commit updates all the places within the gdb/cli directory where we
cast NULL during assignment, removing the cast.

gdb/ChangeLog:

* cli/cli-decode.c (find_cmd): Remove cast of NULL pointer.

9 years agogdb: Remove casts of NULL during assignment.
Andrew Burgess [Mon, 31 Aug 2015 20:19:49 +0000 (21:19 +0100)]
gdb: Remove casts of NULL during assignment.

In the following code:

    struct symbol *wsym = (struct symbol *) NULL;

the cast of NULL is redundant, it adds noise, and is just one more thing
to change if the type of wsym ever changes.  There are a relatively
small number of places in gdb where the above code pattern is used.
Usually the cast is removed like this:

    struct symbol *wsym = NULL;

This commit updates all the places within the gdb/ directory where we
cast NULL during assignment, removing the cast.

gdb/ChangeLog:

* c-valprint.c (print_unpacked_pointer): Remove cast of NULL
pointer.
* dbxread.c (dbx_end_psymtab): Likewise.
* gnu-nat.c (gnu_write_inferior): Likewise.
* mdebugread.c (cross_ref): Likewise.
* p-valprint.c (pascal_val_print): Likewise.
* xcoffread.c (xcoff_end_psymtab): Likewise.

9 years agoMention multi-arch debugging support in NEWS
Yao Qi [Fri, 4 Sep 2015 14:26:21 +0000 (15:26 +0100)]
Mention multi-arch debugging support in NEWS

gdb:

2015-09-04  Yao Qi  <yao.qi@linaro.org>

* NEWS: Mention the aarch64 multi-arch debugging support.

9 years agoAutomatic date update in version.in
GDB Administrator [Fri, 4 Sep 2015 00:00:09 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years ago[Ada] Make string_char_type a true TYPE_CODE_CHAR type in Ada
Pierre-Marie de Rodat [Wed, 2 Sep 2015 16:18:15 +0000 (18:18 +0200)]
[Ada] Make string_char_type a true TYPE_CODE_CHAR type in Ada

Before this change, trying to call an overloaded function with at least
one character literal in argument would fail.  For instance, given these
two functions:

   function F (C : Character) return Integer is
   begin
      return Character'Pos (C);
   end F;

   function F (I : Integer) return Integer is
   begin
      return -I;
   end F;

We would get the following GDB session:

    (gdb) p f('A')
    $1 = -65
    (gdb) p f(1)
    $1 = -1

This is wrong because the first call should select the first F function
and thus return 65.

The root problem is that ada-lang.c:ada_language_arch_info stores in
string_char_type a type whose code is TYPE_CODE_INT instead of
TYPE_CODE_CHAR.  As a result, all parsed character literals are turned
into integer values and during overload matching, the TYPE_CODE_CHAR
formal rejects the TYPE_CODE_INT actual.

This change turns string_char_type into a true TYPE_CODE_CHAR type in
ada-lang.c so that we have instead the expected:

    (gdb) p f('A')
    $1 = 65

gdb/ChangeLog:

* ada-lang.c (ada_language_arch_info): Create a TYPE_CODE_CHAR
type instead of a TYPE_CODE_INT one for the string_char_type
and the ada_primitive_type_char types.

gdb/testsuite/ChangeLog:

* gdb.ada/funcall_char.exp: New testcase.
* gdb.ada/funcall_char/foo.adb: New file.

Tested on x86_64-linux, no regression.

9 years agoFix typo in desription of linker script symbols.
Nick Clifton [Thu, 3 Sep 2015 15:49:18 +0000 (16:49 +0100)]
Fix typo in desription of linker script symbols.

(Source Code Reference): Fix off-by-one typo in example of how to
use linker script symbols in C source code.

9 years agoFix seg-fault in readelf when scanniing a corrupt binary.
Nick Clifton [Thu, 3 Sep 2015 15:15:49 +0000 (16:15 +0100)]
Fix seg-fault in readelf when scanniing a corrupt binary.

PR binutils/18879
* readelf.c (get_unwind_section_word): Check for negative offsets
and very small sections.
(dump_arm_unwind): Warn if the table offset is too large.

9 years ago[aarch64] Check region OK for HW watchpoint in GDBserver
Yao Qi [Thu, 3 Sep 2015 13:01:49 +0000 (14:01 +0100)]
[aarch64] Check region OK for HW watchpoint in GDBserver

Nowadays, if user requests HW watchpoint to monitor a large memory area
or unaligned area, aarch64 GDB will split into multiple aligned areas,
and use multiple debugging registers to watch them.  However, the
registers are not updated in a transaction way.  GDBserver doesn't revert
updates in previous iterations if some debugging registers fail to update
due to some reason, like no free debugging registers available, in the
latter iteration.  For example, if we have a char buf[34], and watch buf
in gdb,

(gdb) watch buf
Hardware watchpoint 2: buf
(gdb) c
Continuing.
infrun: clear_proceed_status_thread (Thread 13466)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT)
infrun: step-over queue now empty
infrun: resuming [Thread 13466] for step-over
Sending packet: $m410838,22#35...Packet received: 00000000000000000000000000000000000000000000000000000000000000000000
infrun: skipping breakpoint: stepping past insn at: 0x400524
infrun: skipping breakpoint: stepping past insn at: 0x400524
Sending packet: $Z2,410838,22#80...Packet received: E01 <----- [1]
Packet Z2 (write-watchpoint) is supported
Sending packet: $Z0,7fb7fe0a8c,4#43...Packet received: OK
Warning:
Could not insert hardware watchpoint 2.
Could not insert hardware breakpoints:
You may have requested too many hardware breakpoints/watchpoints.

GDB receives E01 for Z2 packet [1] but GDBserver updates the debugging
register status,

insert_point (addr=0x00410838, len=34, type=hw-write-watchpoint):
BREAKPOINTs:
BP0: addr=0x0, ctrl=0x00000000, ref.count=0
BP1: addr=0x0, ctrl=0x00000000, ref.count=0
BP2: addr=0x0, ctrl=0x00000000, ref.count=0
BP3: addr=0x0, ctrl=0x00000000, ref.count=0
BP4: addr=0x0, ctrl=0x00000000, ref.count=0
BP5: addr=0x0, ctrl=0x00000000, ref.count=0
WATCHPOINTs:
WP0: addr=0x410850, ctrl=0x00001ff5, ref.count=1
WP1: addr=0x410848, ctrl=0x00001ff5, ref.count=1
WP2: addr=0x410840, ctrl=0x00001ff5, ref.count=1
WP3: addr=0x410838, ctrl=0x00001ff5, ref.count=1

four debugging registers can not monitor 34-byte long area, so the last
iteration of updating debugging register state fails but previous
iterations succeed.  This makes GDB think no HW watchpoint is inserted
but some debugging registers are used.

This problem was exposed by "watch buf" gdb.base/watchpoint.exp with
aarch64 GDBserver debugging arm 32-bit program.  The buf is 30-byte long
but 4-byte aligned, and four debugging registers can't cover 34-byte
(extend 4 bytes to be 8-byte aligned) area.  However, this problem
does exist on non-multi-arch debugging scenario as well.

This patch moves code in aarch64_linux_region_ok_for_hw_watchpoint to
aarch64_linux_region_ok_for_watchpoint in nat/aarch64-linux-hw-point.c.
Then, checks with aarch64_linux_region_ok_for_watchpoint, like what we
are doing in GDB.  If the region is OK, call aarch64_handle_watchpoint.

Regression tested on aarch64 with both 64-bit program and 32-bit
program.  Some fails in gdb.base/watchpoint.exp are fixed.

gdb:

2015-09-03  Yao Qi  <yao.qi@linaro.org>

* aarch64-linux-nat.c (aarch64_linux_region_ok_for_hw_watchpoint):
Move code to aarch64_linux_region_ok_for_watchpoint.  Call
aarch64_linux_region_ok_for_watchpoint.
* nat/aarch64-linux-hw-point.c (aarch64_linux_region_ok_for_watchpoint):
New function.
* nat/aarch64-linux-hw-point.h (aarch64_linux_region_ok_for_watchpoint):
Declare it.

gdb/gdbserver:

2015-09-03  Yao Qi  <yao.qi@linaro.org>

* linux-aarch64-low.c (aarch64_insert_point): Call
aarch64_handle_watchpoint if aarch64_linux_region_ok_for_watchpoint
returns true.

9 years agoFix a typo in the linker documentation: The COFF section .build-id should be .buildid.
Nick Clifton [Thu, 3 Sep 2015 11:55:57 +0000 (12:55 +0100)]
Fix a typo in the linker documentation: The COFF section .build-id should be .buildid.

9 years agoRemove convert_mov_to_lea
H.J. Lu [Thu, 3 Sep 2015 11:13:21 +0000 (04:13 -0700)]
Remove convert_mov_to_lea

* elf64-x86-64.c (elf_x86_64_convert_mov_to_lea): Remove
convert_mov_to_lea.

9 years agoAdd ifunc tests for call, jmp, add, test
H.J. Lu [Thu, 3 Sep 2015 11:03:13 +0000 (04:03 -0700)]
Add ifunc tests for call, jmp, add, test

* ld-ifunc/ifunc-21-i386.s: Add tests for call, jmp, add, test.
* ld-ifunc/ifunc-21-x86-64.s: Likewise.
* ld-ifunc/ifunc-22-i386.s: Likewise.
* ld-ifunc/ifunc-22-x86-64.s: Likewise.
* ld-ifunc/ifunc-21-i386.d: Updated.
* ld-ifunc/ifunc-21-x86-64.d: Likewise.
* ld-ifunc/ifunc-22-i386.d: Likewise.
* ld-ifunc/ifunc-22-x86-64.d: Likewise.

9 years agoUse gdbarch obstack to allocate the TYPE_NAME string in arch_type
Patrick Palka [Sat, 29 Aug 2015 22:07:50 +0000 (18:07 -0400)]
Use gdbarch obstack to allocate the TYPE_NAME string in arch_type

Since the type whose name is being set is now being allocated on the
gdbarch obstack, we should allocate its TYPE_NAME on the obstack too.
This reduces the number of individual valgrind warnings for the command
"gdb gdb" from ~300 to ~150.

Tested on x86_64-unknown-linux-gnu.

gdb/ChangeLog:

* gdb_obstack.h (obstack_strdup): Declare.
* gdb_obstack.c (obstack_strdup): Define.
* gdbarch.sh (gdbarch_obstack_strdup): Declare and define.
* gdbarch.c: Regenerate.
* gdbarch.h: Regenerate.
* gdbtypes.c (arch_type): Use gdbarch_obstack_strdup.

9 years agoUpdate documentation of function copy_type_recursive
Patrick Palka [Sat, 29 Aug 2015 22:16:40 +0000 (18:16 -0400)]
Update documentation of function copy_type_recursive

Following commit 8f57eec2fb3 ("Use gdbarch obstack to allocate types in
alloc_type_arch") it is no longer the case that the type returned by
copy_type_recursive is allocated using malloc.  Because the function
uses alloc_type_arch internally, the new type is now allocated on the
gdbarch associated with the type, and is thus owned by that gdbarch.

gdb/ChangeLog:

* gdbtypes.c (copy_type_recursive): Update documentation.

9 years agoAutomatic date update in version.in
GDB Administrator [Thu, 3 Sep 2015 00:00:11 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoAdd 2 ifunc tests for i386 and x86-64
H.J. Lu [Wed, 2 Sep 2015 22:30:32 +0000 (15:30 -0700)]
Add 2 ifunc tests for i386 and x86-64

* ld-ifunc/ifunc-21-i386.d: Likewise.
* ld-ifunc/ifunc-21-i386.s: Likewise.
* ld-ifunc/ifunc-21-x86-64.d: Likewise.
* ld-ifunc/ifunc-21-x86-64.s: Likewise.
* ld-ifunc/ifunc-22-i386.d: Likewise.
* ld-ifunc/ifunc-22-i386.s: Likewise.
* ld-ifunc/ifunc-22-x86-64.d: Likewise.
* ld-ifunc/ifunc-22-x86-64.s: Likewise.

9 years agoRename R_386_irelative to R_386_ext2
H.J. Lu [Wed, 2 Sep 2015 12:13:08 +0000 (05:13 -0700)]
Rename R_386_irelative to R_386_ext2

* elf32-i386.c (R_386_irelative): Renamed to ...
(R_386_ext2): This.
(R_386_vt_offset): Updated.
(elf_i386_rtype_to_howto): Likewise.
(elf_i386_relocate_section): Likewise.

9 years agoTurn IFUNC symbols from shared libraries into normal FUNC symbols
H.J. Lu [Wed, 2 Sep 2015 11:14:21 +0000 (04:14 -0700)]
Turn IFUNC symbols from shared libraries into normal FUNC symbols

Turn IFUNC symbols from shared libraries into normal FUNC symbols when
we are resolving symbol references, instead of when we are writing out
the symbol table.

PR gold/18886
* resolve.cc (Symbol::override_base): Turn IFUNC symbols from
shared libraries into normal FUNC symbols.
* symtab.cc (Symbol_table::sized_write_symbol): Assert IFUNC
symbols aren't from shared libraries.

9 years ago[GOLD] ppc64 out-of-line register save/restore functions
Alan Modra [Wed, 2 Sep 2015 06:21:59 +0000 (15:51 +0930)]
[GOLD] ppc64 out-of-line register save/restore functions

Gold version of a4b6fadd.  Don't emit long branch or plt branch stubs
to save/restore functions.  Copy them instead.

PR 18878
* powerpc.cc (Target_powerpc): Add savres_section_ and accessor.
(Target_powerpc::Branch_info::make_stub): Determine whether long
branch stub is for save/restore function.
(Branch_stub_ent): Add save_res_, and extra parm to constructor.
(Stub_table): Add need_save_res_.
(Stub_table:clear_stubs): Clear need_save_res_.
(Stub_table:set_address_and_size): Add save/restore section size.
(Stub_table::add_long_branch_entry): Add save_res param.  Set
need_save_res_, but don't add space for save/restore stubs.
(Stub_table::find_long_branch_entry): Return offset to local copy
of save/restore func.
(Stub_table::do_write): Don't output save/restore stubs.  Instead
copy the save/restore functions.
(Output_data_save_res:contents): New accessor.
(Target_powerpc::define_save_restore_funcs): Set savres_section_.

9 years agoInitialize yet another variable to silence GCC warning from last-but-one commit
Sergio Durigan Junior [Wed, 2 Sep 2015 04:46:43 +0000 (00:46 -0400)]
Initialize yet another variable to silence GCC warning from last-but-one commit

Yet another BuildBot e-mail, yet another breakage on RHEL-7.1 s390x
(which uses an older GCC).  This time,
solib-svr4.c:solib_event_probe_action has the probe_argc variable,
which is now inside a TRY..CATCH and therefore needs to be
initialized.  Pushed as obvious.

gdb/ChangeLog:
2015-09-01  Sergio Durigan Junior  <sergiodj@redhat.com>

* solib-svr4.c (solib_event_probe_action): Initialize 'probe_argc'
as zero.

9 years agoInitialize variable and silence GCC warning from last commit
Sergio Durigan Junior [Wed, 2 Sep 2015 04:34:22 +0000 (00:34 -0400)]
Initialize variable and silence GCC warning from last commit

BuildBot e-mailed me to let me know that my last commit broke GDB on
RHEL-7.1 s390x.  On solib-svr4.c:svr4_handle_solib_event, 'val' now
needs to be initialized as NULL because it is inside a TRY..CATCH
block.  This patch does that.  Pushed as obvious.

gdb/ChangeLog:
2015-09-01  Sergio Durigan Junior  <sergiodj@redhat.com>

* solib-svr4.c (svr4_handle_solib_event): Initialize 'val' as NULL

9 years agoCatching errors on probes-based dynamic linker interface
Sergio Durigan Junior [Fri, 21 Aug 2015 22:28:07 +0000 (18:28 -0400)]
Catching errors on probes-based dynamic linker interface

This patch is intended to make the interaction between the
probes-based dynamic linker interface and the SystemTap SDT probe code
on GDB more robust.  It does that by wrapping the calls to the probe
API with TRY...CATCH'es, so that any exception thrown will be caught
and handled properly.

The idea for this patch came from
<https://bugzilla.redhat.com/show_bug.cgi?id=1196181>, which is a bug
initially filed against Fedora GDB (but now under Fedora GLIBC).  This
bug happens on armhfp (although it could happen on other targets as
well), and is triggered because GCC generates a strange argument for
one of the probes used by GDB in the dynamic linker interface.  As can
be seen in the bug, this argument is "-4@.L1052".

I don't want to discuss the reasons for this argument to be there
(this discussion belongs to the bug, or to another thread), but GDB
could definitely do a better error handling here.  Currently, one sees
the following message when there is an error in the probes-based
dynamic linker interface:

  (gdb) run
  Starting program: /bin/inferior
  warning: Probes-based dynamic linker interface failed.
  Reverting to original interface.

  Cannot parse expression `.L976 4@r4'.
  (gdb)

Which means that one needs to explicitly issue a "continue" command to
make GDB continue running the inferior, even though this error is not
fatal and GDB will fallback to the old interface automatically.

This is where this patch helps: it makes GDB still print the necessary
warnings or error messages, but it *also* does not stop the inferior
unnecessarily.

I have tested this patch on the systems where this error happens, but
I could not come up with a way to create a testcase for it.
Nevertheless, it should be straightforward to see that this patch does
improve the current situation.

gdb/ChangeLog:
2015-09-01  Sergio Durigan Junior  <sergiodj@redhat.com>

* solib-svr4.c (solib_event_probe_action): Call
get_probe_argument_count using TRY...CATCH.
(svr4_handle_solib_event): Likewise, for evaluate_probe_argument.

9 years agoImprove error reporting when handling SystemTap SDT probes
Sergio Durigan Junior [Fri, 21 Aug 2015 22:13:46 +0000 (18:13 -0400)]
Improve error reporting when handling SystemTap SDT probes

This patch improves the error reporting when handling SystemTap SDT
probes.  "Handling", in this case, mostly means "parsing".

On gdb/probe.h, only trivial changes on functions' comments in order
to explicitly mention that some of them can throw exceptions.  This is
just to make the API a bit more clear.

On gdb/stap-probe.c, I have s/internal_error/error/ on two functions
that are responsible for parsing specific bits of the probes'
arguments: stap_get_opcode and stap_get_expected_argument_type.  It is
not correct to call internal_error on such situations because it is
not really GDB's fault if the probes have malformed arguments.  I also
improved the error reported on stap_get_expected_argument_type by also
including the probe name on it.

Aside from that, and perhaps most importantly, I added a check on
stap_get_arg to make sure that we don't try to extract an argument
from a probe that has no arguments.  This check issues an
internal_error, because it really means that GDB is doing something it
shouldn't.

Although it can be considered almost trivial, and despite the fact
that I am the maintainer for this part of the code, I am posting this
patch for review.  I will wait a few days, and if nobody has anything
to say, I will go ahead and push it.

gdb/ChangeLog:
2015-09-01  Sergio Durigan Junior  <sergiodj@redhat.com>

* probe.h (struct probe_ops) <get_probe_argument_count,
evaluate_probe_argument, enable_probe, disable_probe>: Mention in
the comment that the function can throw an exception.
(get_probe_argument_count): Likewise.
(evaluate_probe_argument): Likewise.
* stap-probe.c (stap_get_opcode): Call error instead of
internal_error.
(stap_get_expected_argument_type): Likewise.  Add argument
'probe'.  Improve error message by mentioning the probe's name.
(stap_parse_probe_arguments): Adjust call to
stap_get_expected_argument_type.
(stap_get_arg): Add comment.  Assert that 'probe->args_parsed' is
not zero.  Call internal_error if GDB requests an argument but the
probe has no arguments.

9 years agoAutomatic date update in version.in
GDB Administrator [Wed, 2 Sep 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years ago[Ada] Fix completion for multiple function matches
Pierre-Marie de Rodat [Mon, 31 Aug 2015 14:04:07 +0000 (16:04 +0200)]
[Ada] Fix completion for multiple function matches

Before this change, trying to complete an expression ending with an
ambiguous function name (i.e. for which there are multiple matches)
would display a menu with a prompt for the user to pick one. For
instance:

    (gdb) p func<tab>Multiple matches for func
    [0] cancel
    [1] pack2.func at pack2.adb:5
    [2] pack.func at pack.adb:5
    >

This is not user friendly and actually triggered a segmentation fault
after the user did pick one. It is not clear whether the segmentation
fault needs a separate fix, but this is the only known case which
exhibits it at the moment, and this case must be fixed itself.

The problem lies in ada-lang.c (ada_resolve_function): when we got
multiple matches, we should not display the menu if we are in completion
mode. This patch adjusts the corresponding condition accordingly.

gdb/ChangeLog:

* ada-lang.c (ada_resolve_function): Do not ask the user what
match to use when in completion mode.

gdb/testsuite/ChangeLog:

* gdb.ada/complete.exp: Add "pck.ambiguous_func" to the relevant
expected outputs.  Add two testcases for completing ambiguous
functions.
* gdb.ada/complete/aux_pck.adb: New file.
* gdb.ada/complete/aux_pck.ads: New file.
* gdb.ada/complete/foo.adb: Pull Aux_Pck and call the two
Ambiguous_Func functions.
* gdb.ada/complete/pck.ads: Add an Ambiguous_Func function.
* gdb.ada/complete/pck.adb: Likewise.

Tested on x86_64-linux, no regression.

9 years agoSkip PLT for function pointer initialization
H.J. Lu [Tue, 1 Sep 2015 12:06:16 +0000 (05:06 -0700)]
Skip PLT for function pointer initialization

We use its PLT entry to initialize function pointer at run-time.  If
there is no other usage for the PLT entry, we can generate run-time
function pointer relocations in read-write section, which can be
resolved by dynamic linker, to initialize function pointers.  It avoids
the extra indirect branch overhead in PLT.

bfd/

PR ld/18900
* elf32-i386.c (elf_i386_link_hash_entry): Add
func_pointer_refcount.
(elf_i386_link_hash_newfunc): Clear func_pointer_refcount.
(elf_i386_get_local_sym_hash): Likewise.
(elf_i386_copy_indirect_symbol): Also copy
func_pointer_refcount.
(elf_i386_check_relocs): Increment func_pointer_refcount.
(elf_i386_gc_sweep_hook): Decrement func_pointer_refcount.
(elf_i386_allocate_dynrelocs): Don't create the PLT entry if
there are only function pointer relocations which can be
resolved at run-time.  Keep dynanamic relocations for run-time
function pointer initialization.
(elf_i386_relocate_section): Copy dynamic function pointer
relocations.
* elf64-x86-64.c (elf_x86_64_link_hash_entry): Add
func_pointer_refcount.
(elf_x86_64_link_hash_newfunc): Clear func_pointer_refcount.
(elf_x86_64_get_local_sym_hash): Likewise.
(elf_x86_64_copy_indirect_symbol): Also copy
func_pointer_refcount.
(elf_x86_64_check_relocs): Increment func_pointer_refcount.
(elf_x86_64_gc_sweep_hook): Decrement func_pointer_refcount.
(elf_x86_64_allocate_dynrelocs): Don't create the PLT entry if
there are only function pointer relocations which can be
resolved at run-time.  Keep dynanamic relocations for run-time
function pointer initialization.
(elf_x86_64_relocate_section): Copy dynamic function pointer
relocations.

ld/testsuite/

PR ld/18900
* ld-i386/i386.exp: Run tests for PR ld/18900.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr18900.out: New file.
* ld-i386/pr18900a.c: Likewise.
* ld-i386/pr18900a.c: Likewise.
* ld-i386/pr18900a.rd: Likewise.
* ld-i386/pr18900b.c: Likewise.
* ld-i386/pr18900b.rd: Likewise.
* ld-i386/pr18900c.c: Likewise.
* ld-x86-64/pr18900.out: Likewise.
* ld-x86-64/pr18900a.c: Likewise.
* ld-x86-64/pr18900a.rd: Likewise.
* ld-x86-64/pr18900b.c: Likewise.
* ld-x86-64/pr18900b.rd: Likewise.
* ld-x86-64/pr18900c.c: Likewise.
* ld-x86-64/mpx3.dd: Updated.

9 years agoWrong function reported for linker error message
Alan Modra [Tue, 1 Sep 2015 07:58:13 +0000 (17:28 +0930)]
Wrong function reported for linker error message

OPD lookup goes awry when .opd has been shuffled to remove unused
functions.

* elf64-ppc.c (ppc64_elf_maybe_function_sym): Adjust symbol value
if .opd section has been edited.

9 years agoppc64 out-of-line register save/restore functions
Alan Modra [Mon, 31 Aug 2015 23:26:22 +0000 (08:56 +0930)]
ppc64 out-of-line register save/restore functions

Don't emit long branch or plt branch stubs to save/restore functions.
Copy them instead.  The problem is that plt branch stubs currently
trash r12, one of the parameters to some of the save/restore
functions, and there is no free register available to use instead of
r12.

6f20ed8a is prerequisite for this patch.

PR 18878
* elf64-ppc.c (ARRAY_SIZE): Define.  Use throughout.
(enum ppc_stub_type): Add ppc_stub_save_res.
(struct map_stub): Add "next" and "needs_save_res".
(struct ppc_link_hash_entry): Add "save_res" flag.
(struct ppc_link_hash_table): Add "group".
(sfpr_define): Add stub_sec param.  Define symbol in stub_sec if
stub_sec is non-null.  Set "save_res".
(save_res_funcs): Make file scope, rename from funcs.  Adjust uses.
(ppc64_elf_adjust_dynamic_symbol): Prohibit plt call to save_res syms.
(ppc_build_one_stub): Handle ppc_stub_save_res.
(ppc_size_one_stub): Set stub type to ppc_size_one_stub on finding
stub for linker defined save_res sym.
(group_sections): Init new fields of struct map_stub.
(ppc64_elf_size_stubs): Reserve space for save/restore func copy.
(ppc64_elf_build_stubs): Copy save/restore funcs to groups.  Emit
alias syms too.
(ppc64_elf_relocate_section): Set destination for ppc_stub_save_res.

9 years agoAutomatic date update in version.in
GDB Administrator [Tue, 1 Sep 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoOptimise PowerPC64 r2 adjusting stubs
Alan Modra [Mon, 31 Aug 2015 08:25:14 +0000 (17:55 +0930)]
Optimise PowerPC64 r2 adjusting stubs

Sometimes these stubs don't need to change the low 16-bits of r2, so in
that case omit a useless addi r2,r2,0 insn.  Also, change the get_r2off
error return from 0 to -1 since 0 is a valid return for ELFv2 -R objects.

* elf64-ppc.c (get_r2off): Return -1 on error.
(ppc_build_one_stub): Adjust for get_r2off change.  Don't emit
addi r2,r2,0 on r2off stubs when the low 16-bit delta is zero.
(ppc_size_one_stub): Corresponding size changes for r2off stubs.
Add condition in test for -R objects.

9 years agoppc64 section group handling
Alan Modra [Mon, 31 Aug 2015 08:20:33 +0000 (17:50 +0930)]
ppc64 section group handling

Two organizational changes to the array of additional info kept for
sections.
1) Move group info into a per-group allocated struct, in preparation
for future changes that need per-group accounting.
2) Expand the array to include output sections, which simplifies
sizing and removes the need for a separate output section array.

* section.c (section_id): Make file scope.
(bfd_get_next_section_id): New function.
* elf64-ppc.c (struct map_stub): Remove toc_off field.  Move decl.
(struct ppc_stub_hash_entry): Delete stub_sec and id_sec.  Add
group.  Update all uses.
(struct ppc_link_hash_table): Delete top_id, top_index, and
input_list.  Add sec_info_arr_size.  Rename stub_group to
sec_info, and make group info indirect.  Update stub_group refs
throughout file.
(ppc_add_stub): Don't look for stub_sec on link_sec stub_group
entry.
(ppc_build_one_stub): Delete FIXME.
(ppc64_elf_setup_section_lists): Size htab->sec_info for all
sections, not just input sections.  Don't create htab->input_list.
(ppc64_elf_next_input_section): Update to use sec_info union as
list pointer.
(PREV_SEC): Delete.
(group_sections): Pass "info" param rather than "htab".  Iterate
over output sections rather than input_list.  Use sec_info union
as list pointers.  Alloc atruct map_stub, and return fail status.
* bfd-in2.h: Regenerate.

9 years agogdb/tui: Make local variable const.
Andrew Burgess [Sun, 30 Aug 2015 10:14:58 +0000 (11:14 +0100)]
gdb/tui: Make local variable const.

Small clean up to make a local variable const and remove a cast of NULL.

gdb/ChangeLog:

* tui/tui-data.c (tui_win_name): Make local variable const, remove
cast of NULL.

9 years agoxtensa: initialize call_abi in xtensa_tdep
Max Filippov [Sat, 30 May 2015 10:00:32 +0000 (13:00 +0300)]
xtensa: initialize call_abi in xtensa_tdep

Use XSHAL_ABI value provided by xtensa-config.h to correctly initialize
xtensa_tdep.call_abi
This fixes calls to functions from GDB that otherwise fail with the
following assertion in call0 configuration:

  gdb/regcache.c:602: internal-error: regcache_raw_read: Assertion
  `regnum >= 0 && regnum < regcache->descr->nr_raw_registers' failed.

2015-08-31  Max Filippov  <jcmvbkbc@gmail.com>
gdb/
* xtensa-tdep.h (XTENSA_GDBARCH_TDEP_INSTANTIATE): Initialize
call_abi using XSHAL_ABI macro.

9 years agoMore "Make asection->id and asection->index unsigned"
Alan Modra [Mon, 31 Aug 2015 06:33:42 +0000 (16:03 +0930)]
More "Make asection->id and asection->index unsigned"

Fixes regressions introduced by 7292b3ac.

* elflink.c (elf_sort_symbol): Use correctly sized type for
calculating signed section->id difference.
(elf_link_add_object_symbols): Likewise.
* pef.c (bfd_pef_scan_start_address): Warning fix.
* vms-alpha.c (vms_new_section_hook): Likewise.

9 years agoNDS32 build fixes
Alan Modra [Mon, 31 Aug 2015 03:31:59 +0000 (13:01 +0930)]
NDS32 build fixes

* elf32-nds32.c (nds32_convert_32_to_16_alu1): Warning fix.
(find_relocs_at_address_addr): Make "reloc_type" an enum.
(nds32_elf_relax_delete_blanks): Warning fix.
(nds32_elf_relax_loadstore): Correct loop bound.

9 years agoRegen intl/configure
Alan Modra [Mon, 31 Aug 2015 03:23:36 +0000 (12:53 +0930)]
Regen intl/configure

* configure: Regenerate.

9 years agoMake asection->id and asection->index unsigned
Alan Modra [Sun, 30 Aug 2015 23:02:29 +0000 (08:32 +0930)]
Make asection->id and asection->index unsigned

These int vars are really unsigned, so make them so.

* section.c (struct bfd_section): Make "id" and "index" unsigned.
* coff-rs6000.c (_bfd_xcoff_sizeof_headers): Adjust local var to suit.
* elf.c (elf_map_symbols): Likewise.
* elf64-ppc.c (sym_exists_at): Make "id" param unsigned.
(struct ppc_link_hash_table): Make "top_id" and "top_index" unsigned.
(ppc64_elf_setup_section_lists): Ditto for local vars.
* elf32-arm.c: Similarly to elf64-ppc.c.
* elf32-avr.c: Likewise.
* elf32-hppa.c: Likewise.
* elf32-m68hc1x.c: Likewise.
* elf32-metag.c: Likewise.
* elf32-nios2.c: Likewise.
* elfnn-aarch64.c: Likewise.
* simple.c (struct saved_offsets): Make "section_count" unsigned.
* bfd-in2.h: Regenerate.

9 years agoAutomatic date update in version.in
GDB Administrator [Mon, 31 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agosymtab.h (struct symbol): Tweak comment.
Doug Evans [Sun, 30 Aug 2015 00:10:58 +0000 (17:10 -0700)]
symtab.h (struct symbol): Tweak comment.

gdb/ChangeLog:

* symtab.h (struct symbol): Tweak comment.

9 years agoAutomatic date update in version.in
GDB Administrator [Sun, 30 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoRevert "Use gdbarch obstack to allocate the TYPE_NAME string in arch_type"
Patrick Palka [Sat, 29 Aug 2015 22:03:37 +0000 (18:03 -0400)]
Revert "Use gdbarch obstack to allocate the TYPE_NAME string in arch_type"

This patch manually modified the autogenerated files gdbarch.[ch] instead of
going through gdbarch.sh.

This reverts commit aa78b3b28aeff4bb9977a313f5a8002d920b34c5.

9 years agoUse gdbarch obstack to allocate the TYPE_NAME string in arch_type
Patrick Palka [Mon, 29 Jun 2015 14:45:47 +0000 (10:45 -0400)]
Use gdbarch obstack to allocate the TYPE_NAME string in arch_type

Since the type whose name is being set is now being allocated on the
gdbarch obstack, we should allocate its TYPE_NAME on the obstack too.
This reduces the number of individual valgrind warnings for the command
"gdb gdb" from ~300 to ~150.

Tested on x86_64-unknown-linux-gnu.

gdb/ChangeLog:

* gdbarch.h (gdbarch_obstack_strdup): Declare.
* gdbarch.c (gdbarch_obstack_strdup): Define.
* gdbtypes.c (arch_type): Use it.

9 years agoUse gdbarch obstack to allocate types in alloc_type_arch
Patrick Palka [Mon, 29 Jun 2015 14:39:26 +0000 (10:39 -0400)]
Use gdbarch obstack to allocate types in alloc_type_arch

For the command "gdb gdb" valgrind currently reports 100s of individual
memory leaks, 500 of which originate solely out of the function
alloc_type_arch.  This function allocates a "struct type" associated
with the given gdbarch using malloc but apparently the types allocated
by this function are never freed.

This patch fixes these leaks by making the function alloc_type_arch
allocate these gdbarch-associated types on the gdbarch obstack instead
of on the general heap.  Since, from what I can tell, the types
allocated by this function are all fundamental "wired-in" types, such
types would not benefit from more granular memory management anyway.
They would likely live as long as the gdbarch is alive so allocating
them on the gdbarch obstack makes sense.

With this patch, the number of individual vargrind warnings emitted for
the command "gdb gdb" drops from ~800 to ~300.

Tested on x86_64-unknown-linux-gnu.

gdb/ChangeLog:

* gdbtypes.c (alloc_type_arch): Allocate the type on the given
gdbarch obstack instead of on the heap.  Update commentary
accordingly.

9 years agoAutomatic date update in version.in
GDB Administrator [Sat, 29 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoDocument the GDB 7.10 release in gdb/ChangeLog
Joel Brobecker [Fri, 28 Aug 2015 21:40:54 +0000 (17:40 -0400)]
Document the GDB 7.10 release in gdb/ChangeLog

gdb/ChangeLog:

GDB 7.10 released.

9 years agoUpdate NEWS entry about non-8-bits addressable memory
Simon Marchi [Fri, 28 Aug 2015 19:35:47 +0000 (15:35 -0400)]
Update NEWS entry about non-8-bits addressable memory

9 years agoRevert: solib_global_lookup: Fetch arch from objfile.
Ulrich Weigand [Fri, 28 Aug 2015 17:04:56 +0000 (19:04 +0200)]
Revert: solib_global_lookup: Fetch arch from objfile.

solib_ops are installed as a property of the inferior gdbarch,
so they need to be systematically looked up via that architecture,
not some objfile architecture.

ChangeLog:

Revert:
2014-11-06  Doug Evans  <xdje42@gmail.com>
        * solib.c (solib_global_lookup): Fetch arch from objfile,
        not target_gdbarch.

9 years agoFix incorrect ChangeLog dates.
Ulrich Weigand [Fri, 28 Aug 2015 17:04:12 +0000 (19:04 +0200)]
Fix incorrect ChangeLog dates.

9 years agoAlso install readelf into tooldir/bin directory
H.J. Lu [Fri, 28 Aug 2015 15:34:44 +0000 (08:34 -0700)]
Also install readelf into tooldir/bin directory

* Makefile.am (TOOL_PROGS): Add readelf.
* Makefile.in: Regenerated.

9 years agoAutomatic date update in version.in
GDB Administrator [Fri, 28 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoFix ChangeLog entries starting with "Likewise"
Simon Marchi [Thu, 27 Aug 2015 17:47:44 +0000 (13:47 -0400)]
Fix ChangeLog entries starting with "Likewise"

The first lines of these ChangeLog entries are "Likewise"... it doesn't
make much sense.

9 years ago[Cell/B.E.] Fix wrong relocation for TLS variable offset
Ulrich Weigand [Thu, 27 Aug 2015 17:31:04 +0000 (19:31 +0200)]
[Cell/B.E.] Fix wrong relocation for TLS variable offset

When debugging Cell/B.E. code, the cross-architecture frame unwinding
works by accessing data structures refered to by a thread-local variable
in the inferior.  While accessing this variable via minimal symbol,
code would use BMSYMBOL_VALUE_ADDRESS to determine the offset of the
variable in the thread-local storage block.

This is incorrect, since it adds any relocation offset of the shared
library defining the TLS variable.  That offset would be OK when
accessing the initial copy present in the .tbss section, but it must
not be applied to the offset of the variable in the TLS block at
runtime.  Depending on whether or not the libspe2.so library was
prelinked, access to the variable would fail due to the incorrectly
relocated offset.

ChangeLog:

* ppc-linux-tdep.c (ppc_linux_spe_context_lookup): Do not
attempt to relocate a TLS variable offset.

9 years ago[Cell/B.E.] Make parse_spufs_run more robust
Ulrich Weigand [Thu, 27 Aug 2015 17:27:40 +0000 (19:27 +0200)]
[Cell/B.E.] Make parse_spufs_run more robust

With recent changes to inferior handling, parse_spufs_run needs to be
more careful in assumptions it makes.  In particular, this patch:

- Bails out early if the current inferior has not yet been registered
  (e.g. during fork procession) to avoid assertion failures in register
  cache code.

- Sets inferior_ptid to the current ptid while calling target_read_memory
  to make sure the correct process is accessed if parse_spufs_run is
  called early when inferior_ptid has not yet been switched by the caller.

ChangeLog:

* spu-multiarch.c (parse_spufs_run): Bail out if inferior is not
registered yet.  Set inferior_ptid while calling target_read_memory.

9 years agoDetect SW breakpoints in Cell/B.E. combined debugging
Ulrich Weigand [Thu, 27 Aug 2015 17:20:29 +0000 (19:20 +0200)]
Detect SW breakpoints in Cell/B.E. combined debugging

The Linux target and gdbserver now check the siginfo si_code
reported on a SIGTRAP to detect whether the trap indicates
a software breakpoint was hit.

Unfortunately, on Cell/B.E., the kernel uses an si_code value
of TRAP_BRKPT when a SW breakpoint was hit in PowerPC code,
but a si_code value of SI_KERNEL when a SW breakpoint was
hit in SPU code.

This patch updates Linux target and gdbserver to accept both
si_code values to indicate SW breakpoint on PowerPC.

ChangeLog:

* nat/linux-ptrace.h (GDB_ARCH_TRAP_BRKPT): Replace by ...
(GDB_ARCH_IS_TRAP_BRKPT): ... this.  Add __powerpc__ case.
* linux-nat.c (check_stopped_by_breakpoint): Use
GDB_ARCH_IS_TRAP_BRKPT instead of GDB_ARCH_TRAP_BRKPT.

gdbserver/ChangeLog:

* linux-low.c (check_stopped_by_breakpoint): Use
GDB_ARCH_IS_TRAP_BRKPT instead of GDB_ARCH_TRAP_BRKPT.

9 years agoFix assertion failure in linux-thread-db
Ulrich Weigand [Thu, 27 Aug 2015 17:12:49 +0000 (19:12 +0200)]
Fix assertion failure in linux-thread-db

Since we are no longer using thread events by default in linux-thread-db,
the find_new_threads_once routine contains an assertion that it should
never be called on a live inferior unless using thread events:
  gdb_assert (!target_has_execution || thread_db_use_events ());

However, there is a code path from thread_db_get_thread_local_address
that will in fact call find_new_threads_once in some scenarios.  In
particular, this is currently always triggered when starting up any
Cell/B.E. combined exeuctable.

To fix this, this patch removes the call to thread_db_find_new_threads_1
when the current thread was not yet detected.  In its place, we now just
call thread_from_lwp to detect this one thread if necessary.

ChangeLog:

* linux-thread-db.c (thread_db_get_thread_local_address): If the
thread was not yet discovered, use thread_from_lwp instead of
calling thread_db_find_new_threads_1.

9 years agoReport UNRESOLVED on failures while detecting Cell/B.E.
Ulrich Weigand [Thu, 27 Aug 2015 17:06:03 +0000 (19:06 +0200)]
Report UNRESOLVED on failures while detecting Cell/B.E.

The gdb.cell testcases use the predicate skip_cell_tests defined in
lib/cell.exp to determine whether Cell/B.E. test cases ought to be
run. This tests verifies that we have a toolchain that supports
generating combined Cell/B.E. binaries, and that the target machine
actually is a Cell/B.E.

In order to do so, a small test program is built and run (under the
debugger).  Any failure is taken as a sign that we don't have a
Cell/B.E. machine and the tests are to be skipped.

This has the unfortunate effect that a serious bug in GDB that causes
internal compiler errors even on the trivial test program does not
lead to any failures in the testsuite, since now all gdb.cell test
are simply skipped.

This patch changes skip_cell_tests to at least report UNRESOLVED
in cases where execution of the test program fails in unexpected
ways.

testsuite/ChangeLog:

* lib/cell.exp (skip_cell_tests): Report UNRESOLVED on unexpected
failures to run the test program under GDB.

9 years agoAdd reloc sort test
Alan Modra [Thu, 27 Aug 2015 06:35:54 +0000 (16:05 +0930)]
Add reloc sort test

* ld-powerpc/relocsort.s, * ld-powerpc/relocsort.d: New test.
* ld-powerpc/powerpc.exp: Run it.

9 years agoMore fallout from "Allow symbol and label names to be enclosed in double quotes"
Alan Modra [Thu, 27 Aug 2015 13:20:38 +0000 (22:50 +0930)]
More fallout from "Allow symbol and label names to be enclosed in double quotes"

Some of the TC_START_LABEL implementations need to adjust the end of
the symbol, when a colon doesn't mean a label definition.  That means
they need access to nul_char both the restore the NUL location (it may
be a quote rather than a colon) and to store the new nul_char.  Others
need adjusting to step over a potential trailing quote.

PR gas/18581
* config/tc-aarch64.h (TC_START_LABEL): Redefine.
* config/tc-arm.c (tc_start_label_without_colon): Delete params.
Use input_line_pointer directly.
* config/tc-arm.h (TC_START_LABEL): Redefine.
(TC_START_LABEL_WITHOUT_COLON): Redefine.
(tc_start_label_without_colon): Update prototype.
* config/tc-bfin.c (bfin_start_label): Delete ptr param.  Check
for NUL instead.
* config/tc-bfin.h (bfin_start_label): Update prototype.
(TC_START_LABEL): Redefine.
* config/tc-d30v.h (TC_START_LABEL): Redefine.
* config/tc-fr30.c (restore_colon): Rewrite.
(fr30_is_colon_insn): Add nul_char param.  Return int.  Bump
i_l_p over quote.  Update restore_colon calls.
* config/tc-fr30.h (TC_START_LABEL): Redefine.
(fr30_is_colon_insn): Update prototype.
* config/tc-m32c.c (restore_colon, m32c_is_colon_insn): As above.
* config/tc-m32c.h (TC_START_LABEL): Redefine.
(m32c_is_colon_insn): Update prototype.
* config/tc-m32r.h (TC_START_LABEL): Redefine.
* config/tc-mep.h (TC_START_LABEL): Redefine.
* config/tc-nds32.h (TC_START_LABEL): Redefine.
* config/tc-tic54x.c (tic54x_start_label): Replace params with
nul_char and next_char.  Step over trailing quote.
* config/tc-tic54x.h (TC_START_LABEL_WITHOUT_COLON): Redefine.
(tic54x_start_label): Update prototype.
* read.c (TC_START_LABEL): Redefine.  Update invocation.
(TC_START_LABEL_WITHOUT_COLON): Update invocation.
* config/tc-nios2.c (s_nios2_set): Save initial input_line_pointer
and restore if calling s_set.  Don't restore delim again.

9 years agoFix inverted statements in m88k_analyze_prologue
Simon Marchi [Thu, 27 Aug 2015 13:47:14 +0000 (09:47 -0400)]
Fix inverted statements in m88k_analyze_prologue

These two statements were inverted by mistake in commit "Replace some
xmalloc-family functions with XNEW-family ones".  It obviously doesn't
make sense to have them in this order, which is why I am pushing this as
obvious.

gdb/ChangeLog:

* m88k-tdep.c (m88k_analyze_prologue): Fix inverted allocation
statements.

9 years agoRemove duplicate ChangeLog entry
Simon Marchi [Thu, 27 Aug 2015 13:32:19 +0000 (09:32 -0400)]
Remove duplicate ChangeLog entry

9 years agoDocument non-8-bits addressable memory support in NEWS
Simon Marchi [Thu, 27 Aug 2015 13:30:00 +0000 (09:30 -0400)]
Document non-8-bits addressable memory support in NEWS

I think it would be a good idea to document that gdb now has (basic)
support to read/write memory on architectures with non-8-bits memory.
Hopefully somebody will see it and say "Hey!  We can now (more easily)
port GDB to our strange DSP that has 32-bits-addressable memory!" and do
it.

gdb/ChangeLog:

* NEWS: Document support for non-8-bits addressable memory.

9 years agogdbserver/proc-service.c: bogus return value conversion
Pedro Alves [Thu, 27 Aug 2015 12:37:01 +0000 (13:37 +0100)]
gdbserver/proc-service.c: bogus return value conversion

Building in C++ mode shows:

   int write_inferior_memory (CORE_ADDR memaddr, const unsigned char *myaddr,
       ^
  src/gdb/gdbserver/proc-service.c:93:64: error: invalid conversion from ‘int’ to ‘ps_err_e’ [-fpermissive]
     return write_inferior_memory ((unsigned long) addr, buf, size);
  ^

It only works today by accident, write_inferior_memory does not return
a ps_err_e.

gdb/gdbserver/ChangeLog:
2015-08-27  Pedro Alves  <palves@redhat.com>

* proc-service.c (ps_pdwrite): Return PS_ERR/PS_OK explicily.

9 years agogdb: Drop use of obsolete AC_TYPE_SIGNAL
Pedro Alves [Thu, 27 Aug 2015 12:26:23 +0000 (13:26 +0100)]
gdb: Drop use of obsolete AC_TYPE_SIGNAL

Since we're using sighandler_t, nothing else refers to RETSIGTYPE in
gdb.

(Actually, given gdb/remote.c has been assuming signal handlers return
void for a long time, we could have gotten get rid of this even
without gnulib's sighandler_t.)

gdb/ChangeLog:
2015-08-27  Pedro Alves  <palves@redhat.com>

* configure.ac: Remove AC_TYPE_SIGNAL call.
* configure, config.in: Regenerate.

9 years agogdb: Use sighandler_t everywhere
Pedro Alves [Thu, 27 Aug 2015 12:26:23 +0000 (13:26 +0100)]
gdb: Use sighandler_t everywhere

This fixes 14 build errors like these in C++ mode:

 src/gdb/extension.c: In function ‘void install_sigint_handler(const signal_handler*)’:
 src/gdb/extension.c:698:41: error: invalid conversion from ‘void (*)()’ to ‘__sighandler_t {aka void (*)(int)}’ [-fpermissive]
    signal (SIGINT, handler_state->handler);
  ^
 In file included from build-gnulib/import/signal.h:52:0,
  from ../../src/gdb/extension.c:24:
 /usr/include/signal.h:102:23: error:   initializing argument 2 of ‘void (* signal(int, __sighandler_t))(int)’ [-fpermissive]
  extern __sighandler_t signal (int __sig, __sighandler_t __handler)
^

Instead of this everywhere:

 -  RETSIGTYPE (*handle_sigint_for_compare) () = handle_sigint;
 +  RETSIGTYPE (*handle_sigint_for_compare) (int) = handle_sigint;

Use sighandler_t (a GNU extension).  That's OK to use unconditionaly
because gnulib's signal.h replacement makes sure that it is available.

gdb/ChangeLog:
2015-08-27  Pedro Alves  <palves@redhat.com>

* cp-support.c (gdb_demangle): Use sighandler_t.  Remove cast.
* extension-priv.h: Include signal.h.
(struct signal_handler) <handler>: Change type to sighandler_t.
* extension.c (install_gdb_sigint_handler): Use sighandler_t.
* inflow.c (sigint_ours, sigquit_ours): Change type to
sighandler_t.
(child_terminal_inferior): Remove casts.
(child_terminal_ours_1, new_tty): Use sighandler_t.  Remove casts.
(osig): Change type to sighandler_t.
* nto-procfs.c (ofunc): Change type to sighandler_t.
(procfs_wait): Remove casts.
* remote-m32r-sdi.c (m32r_wait, m32r_load): Use sighandler_t.
* remote-sim.c (gdbsim_wait): Use sighandler_t.
* utils.c (wait_to_die_with_timeout): Use sighandler_t.

9 years agoImport the "signal-h" module from gnulib
Pedro Alves [Thu, 27 Aug 2015 12:26:22 +0000 (13:26 +0100)]
Import the "signal-h" module from gnulib

This gives us a signal.h replacement that makes sure the sighandler_t
typedef (a GNU extension) is always available.  A follow up patch will
make use of this.

gdb/ChangeLog:
2015-08-27  Pedro Alves  <palves@redhat.com>

* gnulib/update-gnulib.sh (IMPORTED_GNULIB_MODULES): Add signal-h.
* gnulib/aclocal.m4: Renegerate.
* gnulib/config.in: Renegerate.
* gnulib/configure: Renegerate.
* gnulib/import/Makefile.am: Update.
* gnulib/import/Makefile.in: Regenerate.
* gnulib/import/m4/gnulib-cache.m4: Update.
* gnulib/import/m4/gnulib-comp.m4: Update.
* gnulib/import/m4/signal_h.m4: New file.
* gnulib/import/signal.in.h: New file.

9 years agoFix ChangeLog entry date and mention gdb/18804
Pedro Alves [Thu, 27 Aug 2015 12:09:39 +0000 (13:09 +0100)]
Fix ChangeLog entry date and mention gdb/18804

9 years agogdb: remove packet size limit
Pedro Alves [Thu, 27 Aug 2015 12:03:15 +0000 (13:03 +0100)]
gdb: remove packet size limit

The remote packet buffer size is currently capped to 16384 mostly for
historical reasons, related to use of alloca.  Stop using alloca and
remove the limitation.

Tested on x86_64 Fedora 20.

gdb/ChangeLog:
2015-08-27  Pedro Alves  <palves@redhat.com>

* remote.c (DEFAULT_MAX_MEMORY_PACKET_SIZE)
(MIN_MEMORY_PACKET_SIZE): New.
(MAX_REMOTE_PACKET_SIZE, MIN_REMOTE_PACKET_SIZE): Delete.
(get_memory_packet_size): Adjust.  No longer limit the max packet
size.
(set_memory_packet_size): Adjust, and remove dead code.
(remote_check_symbols): Use xmalloc and a cleanup instead of
alloca.
(remote_packet_size): No longer cap the packet size.
(putpkt_binary): Use xmalloc and a cleanup instead of alloca.

9 years agoMention language in compile error message
Luis Machado [Thu, 27 Aug 2015 05:00:16 +0000 (02:00 -0300)]
Mention language in compile error message

This patch improves one of the compile error messages by mentioning the
language.

Before - No compiler support for this language.
After  - No compiler support for language <language>.

gdb/ChangeLog:

2015-08-26  Luis Machado  <lgustavo@codesourcery.com>

* compile/compile.c (compile_to_object): Mention language in
error message.

9 years agoForce language to C in gdb.compile/compile-ifunc.exp
Luis Machado [Thu, 27 Aug 2015 04:49:05 +0000 (01:49 -0300)]
Force language to C in gdb.compile/compile-ifunc.exp

Due to the lack of debug information in the binary, GDB is unable to figure
out what language is being used.  This may be a problem when doing remote
debugging and the binary stops at the entry point containing asm code.

In this case GDB will switch to asm as current language and will not switch
back to C when it reaches main, which in turn causes the compile feature check
to malfunction.

This is solved by forcing the language to C after reaching main.

gdb/testsuite/ChangeLog:

2015-08-26  Luis Machado  <lgustavo@codesourcery.com>

* gdb.compile/compile-ifunc.exp (with_test_prefix): Force language
to C.

9 years agoAdd missing pr reference
Alan Modra [Thu, 27 Aug 2015 01:29:36 +0000 (10:59 +0930)]
Add missing pr reference

9 years agoDon't sort ld -r output relocs on alpha
Alan Modra [Thu, 27 Aug 2015 01:24:28 +0000 (10:54 +0930)]
Don't sort ld -r output relocs on alpha

LITERAL/LITUSE relocs must be kept together.

* elf64-alpha.c (elf64_alpha_sort_relocs_p): New function.
(elf_backend_sort_relocs_p): Define.

9 years agoUnset attach_flag when running a new process
Patrick Palka [Thu, 30 Jul 2015 00:03:43 +0000 (20:03 -0400)]
Unset attach_flag when running a new process

We currently set attach_flag when attaching to a process, so we should
make sure to unset it when forking a new process.  Otherwise attach_flag
would remain set after forking, if the previous process associated with
the inferior was attached to.

gdb/ChangeLog:

* target.c (target_pre_inferior): Unset attach_flag.

gdb/testsuite/ChangeLog:

* gdb.base/run-after-attach.exp: New test file.
* gdb.base/run-after-attach.c: New test file.

9 years agoAutomatic date update in version.in
GDB Administrator [Thu, 27 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoFix gdbarch.c build error
Simon Marchi [Wed, 26 Aug 2015 21:31:47 +0000 (17:31 -0400)]
Fix gdbarch.c build error

I made a mistake while handling my previous patch.  A change in
gdbarch causes a build failure.

gdb/ChangeLog:

* gdbarch.sh (append_name): Fix type in XRESIZEVEC.
* gdbarch.c: Re-generate.

9 years agoReplace some xmalloc-family functions with XNEW-family ones
Simon Marchi [Wed, 26 Aug 2015 21:16:07 +0000 (17:16 -0400)]
Replace some xmalloc-family functions with XNEW-family ones

This patch is part of the make-gdb-buildable-in-C++ effort.  The idea is
to change some calls to the xmalloc family of functions to calls to the
equivalents in the XNEW family.  This avoids adding an explicit cast, so
it keeps the code a bit more readable.  Some of them also map relatively
well to a C++ equivalent (XNEW (struct foo) -> new foo), so it will be
possible to do scripted replacements if needed.

I only changed calls that were obviously allocating memory for one or
multiple "objects".  Allocation of variable sizes (such as strings or
buffer handling) will be for later (and won't use XNEW).

  - xmalloc (sizeof (struct foo)) -> XNEW (struct foo)
  - xmalloc (num * sizeof (struct foo)) -> XNEWVEC (struct foo, num)
  - xcalloc (1, sizeof (struct foo)) -> XCNEW (struct foo)
  - xcalloc (num, sizeof (struct foo)) -> XCNEWVEC (struct foo, num)
  - xrealloc (p, num * sizeof (struct foo) -> XRESIZEVEC (struct foo, p, num)
  - obstack_alloc (ob, sizeof (struct foo)) -> XOBNEW (ob, struct foo)
  - obstack_alloc (ob, num * sizeof (struct foo)) -> XOBNEWVEC (ob, struct foo, num)
  - alloca (sizeof (struct foo)) -> XALLOCA (struct foo)
  - alloca (num * sizeof (struct foo)) -> XALLOCAVEC (struct foo, num)

Some instances of xmalloc followed by memset to zero the buffer were
replaced by XCNEW or XCNEWVEC.

I regtested on x86-64, Ubuntu 14.04, but the patch touches many
architecture-specific files.  For those I'll have to rely on the
buildbot or people complaining that I broke their gdb.

gdb/ChangeLog:

* aarch64-linux-nat.c (aarch64_add_process): Likewise.
* aarch64-tdep.c (aarch64_gdbarch_init): Likewise.
* ada-exp.y (write_ambiguous_var): Likewise.
* ada-lang.c (resolve_subexp): Likewise.
(user_select_syms): Likewise.
(assign_aggregate): Likewise.
(ada_evaluate_subexp): Likewise.
(cache_symbol): Likewise.
* addrmap.c (allocate_key): Likewise.
(addrmap_create_mutable): Likewise.
* aix-thread.c (sync_threadlists): Likewise.
* alpha-tdep.c (alpha_push_dummy_call): Likewise.
(alpha_gdbarch_init): Likewise.
* amd64-windows-tdep.c (amd64_windows_push_arguments): Likewise.
* arm-linux-nat.c (arm_linux_add_process): Likewise.
* arm-linux-tdep.c (arm_linux_displaced_step_copy_insn): Likewise.
* arm-tdep.c (push_stack_item): Likewise.
(arm_displaced_step_copy_insn): Likewise.
(arm_gdbarch_init): Likewise.
(_initialize_arm_tdep): Likewise.
* avr-tdep.c (push_stack_item): Likewise.
* ax-general.c (new_agent_expr): Likewise.
* block.c (block_initialize_namespace): Likewise.
* breakpoint.c (alloc_counted_command_line): Likewise.
(update_dprintf_command_list): Likewise.
(parse_breakpoint_sals): Likewise.
(decode_static_tracepoint_spec): Likewise.
(until_break_command): Likewise.
(clear_command): Likewise.
(update_global_location_list): Likewise.
(get_breakpoint_objfile_data) Likewise.
* btrace.c (ftrace_new_function): Likewise.
(btrace_set_insn_history): Likewise.
(btrace_set_call_history): Likewise.
* buildsym.c (add_symbol_to_list): Likewise.
(record_pending_block): Likewise.
(start_subfile): Likewise.
(start_buildsym_compunit): Likewise.
(push_subfile): Likewise.
(end_symtab_get_static_block): Likewise.
(buildsym_init): Likewise.
* cli/cli-cmds.c (source_command): Likewise.
* cli/cli-decode.c (add_cmd): Likewise.
* cli/cli-script.c (build_command_line): Likewise.
(setup_user_args): Likewise.
(realloc_body_list): Likewise.
(process_next_line): Likewise.
(copy_command_lines): Likewise.
* cli/cli-setshow.c (do_set_command): Likewise.
* coff-pe-read.c (read_pe_exported_syms): Likewise.
* coffread.c (coff_locate_sections): Likewise.
(coff_symtab_read): Likewise.
(coff_read_struct_type): Likewise.
* common/cleanups.c (make_my_cleanup2): Likewise.
* common/common-exceptions.c (throw_it): Likewise.
* common/filestuff.c (make_cleanup_close): Likewise.
* common/format.c (parse_format_string): Likewise.
* common/queue.h (DEFINE_QUEUE_P): Likewise.
* compile/compile-object-load.c (munmap_list_add): Likewise.
(compile_object_load): Likewise.
* compile/compile-object-run.c (compile_object_run): Likewise.
* compile/compile.c (append_args): Likewise.
* corefile.c (specify_exec_file_hook): Likewise.
* cp-support.c (make_symbol_overload_list): Likewise.
* cris-tdep.c (push_stack_item): Likewise.
(cris_gdbarch_init): Likewise.
* ctf.c (ctf_trace_file_writer_new): Likewise.
* dbxread.c (init_header_files): Likewise.
(add_new_header_file): Likewise.
(init_bincl_list): Likewise.
(dbx_end_psymtab): Likewise.
(start_psymtab): Likewise.
(dbx_end_psymtab): Likewise.
* dcache.c (dcache_init): Likewise.
* dictionary.c (dict_create_hashed): Likewise.
(dict_create_hashed_expandable): Likewise.
(dict_create_linear): Likewise.
(dict_create_linear_expandable): Likewise.
* dtrace-probe.c (dtrace_process_dof_probe): Likewise.
* dummy-frame.c (register_dummy_frame_dtor): Likewise.
* dwarf2-frame-tailcall.c (cache_new_ref1): Likewise.
* dwarf2-frame.c (dwarf2_build_frame_info): Likewise.
(decode_frame_entry_1): Likewise.
* dwarf2expr.c (new_dwarf_expr_context): Likewise.
* dwarf2loc.c (dwarf2_compile_expr_to_ax): Likewise.
* dwarf2read.c (dwarf2_has_info): Likewise.
(create_signatured_type_table_from_index): Likewise.
(dwarf2_read_index): Likewise.
(dw2_get_file_names_reader): Likewise.
(create_all_type_units): Likewise.
(read_cutu_die_from_dwo): Likewise.
(init_tu_and_read_dwo_dies): Likewise.
(init_cutu_and_read_dies): Likewise.
(create_all_comp_units): Likewise.
(queue_comp_unit): Likewise.
(inherit_abstract_dies): Likewise.
(read_call_site_scope): Likewise.
(dwarf2_add_field): Likewise.
(dwarf2_add_typedef): Likewise.
(dwarf2_add_member_fn): Likewise.
(attr_to_dynamic_prop): Likewise.
(abbrev_table_alloc_abbrev): Likewise.
(abbrev_table_read_table): Likewise.
(add_include_dir): Likewise.
(add_file_name): Likewise.
(dwarf_decode_line_header): Likewise.
(dwarf2_const_value_attr): Likewise.
(dwarf_alloc_block): Likewise.
(parse_macro_definition): Likewise.
(set_die_type): Likewise.
(write_psymtabs_to_index): Likewise.
(create_cus_from_index): Likewise.
(dwarf2_create_include_psymtab): Likewise.
(process_psymtab_comp_unit_reader): Likewise.
(build_type_psymtab_dependencies): Likewise.
(read_comp_units_from_section): Likewise.
(compute_compunit_symtab_includes): Likewise.
(create_dwo_unit_in_dwp_v1): Likewise.
(create_dwo_unit_in_dwp_v2): Likewise.
(read_func_scope): Likewise.
(process_structure_scope): Likewise.
(mark_common_block_symbol_computed): Likewise.
(load_partial_dies): Likewise.
(dwarf2_symbol_mark_computed): Likewise.
* elfread.c (elf_symfile_segments): Likewise.
(elf_read_minimal_symbols): Likewise.
* environ.c (make_environ): Likewise.
* eval.c (evaluate_subexp_standard): Likewise.
* event-loop.c (create_file_handler): Likewise.
(create_async_signal_handler): Likewise.
(create_async_event_handler): Likewise.
(create_timer): Likewise.
* exec.c (build_section_table): Likewise.
* fbsd-nat.c (fbsd_remember_child): Likewise.
* fork-child.c (fork_inferior): Likewise.
* frv-tdep.c (new_variant): Likewise.
* gdbarch.sh (gdbarch_alloc): Likewise.
(append_name): Likewise.
* gdbtypes.c (rank_function): Likewise.
(copy_type_recursive): Likewise.
(add_dyn_prop): Likewise.
* gnu-nat.c (make_proc): Likewise.
(make_inf): Likewise.
(gnu_write_inferior): Likewise.
* gnu-v3-abi.c (build_gdb_vtable_type): Likewise.
(build_std_type_info_type): Likewise.
* guile/scm-param.c (compute_enum_list): Likewise.
* guile/scm-utils.c (gdbscm_parse_function_args): Likewise.
* guile/scm-value.c (gdbscm_value_call): Likewise.
* h8300-tdep.c (h8300_gdbarch_init): Likewise.
* hppa-tdep.c (hppa_init_objfile_priv_data): Likewise.
(read_unwind_info): Likewise.
* ia64-tdep.c (ia64_gdbarch_init): Likewise.
* infcall.c (dummy_frame_context_saver_setup): Likewise.
(call_function_by_hand_dummy): Likewise.
* infcmd.c (step_once): Likewise.
(finish_forward): Likewise.
(attach_command): Likewise.
(notice_new_inferior): Likewise.
* inferior.c (add_inferior_silent): Likewise.
* infrun.c (add_displaced_stepping_state): Likewise.
(save_infcall_control_state): Likewise.
(save_inferior_ptid): Likewise.
(_initialize_infrun): Likewise.
* jit.c (bfd_open_from_target_memory): Likewise.
(jit_gdbarch_data_init): Likewise.
* language.c (add_language): Likewise.
* linespec.c (decode_line_2): Likewise.
* linux-nat.c (add_to_pid_list): Likewise.
(add_initial_lwp): Likewise.
* linux-thread-db.c (add_thread_db_info): Likewise.
(record_thread): Likewise.
(info_auto_load_libthread_db): Likewise.
* m32c-tdep.c (m32c_gdbarch_init): Likewise.
* m68hc11-tdep.c (m68hc11_gdbarch_init): Likewise.
* m68k-tdep.c (m68k_gdbarch_init): Likewise.
* m88k-tdep.c (m88k_analyze_prologue): Likewise.
* macrocmd.c (macro_define_command): Likewise.
* macroexp.c (gather_arguments): Likewise.
* macroscope.c (sal_macro_scope): Likewise.
* macrotab.c (new_macro_table): Likewise.
* mdebugread.c (push_parse_stack): Likewise.
(parse_partial_symbols): Likewise.
(parse_symbol): Likewise.
(psymtab_to_symtab_1): Likewise.
(new_block): Likewise.
(new_psymtab): Likewise.
(mdebug_build_psymtabs): Likewise.
(add_pending): Likewise.
(elfmdebug_build_psymtabs): Likewise.
* mep-tdep.c (mep_gdbarch_init): Likewise.
* mi/mi-main.c (mi_execute_command): Likewise.
* mi/mi-parse.c (mi_parse_argv): Likewise.
* minidebug.c (lzma_open): Likewise.
* minsyms.c (terminate_minimal_symbol_table): Likewise.
* mips-linux-nat.c (mips_linux_insert_watchpoint): Likewise.
* mips-tdep.c (mips_gdbarch_init): Likewise.
* mn10300-tdep.c (mn10300_gdbarch_init): Likewise.
* msp430-tdep.c (msp430_gdbarch_init): Likewise.
* mt-tdep.c (mt_registers_info): Likewise.
* nat/aarch64-linux.c (aarch64_linux_new_thread): Likewise.
* nat/linux-btrace.c (linux_enable_bts): Likewise.
(linux_enable_pt): Likewise.
* nat/linux-osdata.c (linux_xfer_osdata_processes): Likewise.
(linux_xfer_osdata_processgroups): Likewise.
* nios2-tdep.c (nios2_gdbarch_init): Likewise.
* nto-procfs.c (procfs_meminfo): Likewise.
* objc-lang.c (start_msglist): Likewise.
(selectors_info): Likewise.
(classes_info): Likewise.
(find_methods): Likewise.
* objfiles.c (allocate_objfile): Likewise.
(update_section_map): Likewise.
* osabi.c (gdbarch_register_osabi): Likewise.
(gdbarch_register_osabi_sniffer): Likewise.
* parse.c (start_arglist): Likewise.
* ppc-linux-nat.c (hwdebug_find_thread_points_by_tid): Likewise.
(hwdebug_insert_point): Likewise.
* printcmd.c (display_command): Likewise.
(ui_printf): Likewise.
* procfs.c (create_procinfo): Likewise.
(load_syscalls): Likewise.
(proc_get_LDT_entry): Likewise.
(proc_update_threads): Likewise.
* prologue-value.c (make_pv_area): Likewise.
(pv_area_store): Likewise.
* psymtab.c (extend_psymbol_list): Likewise.
(init_psymbol_list): Likewise.
(allocate_psymtab): Likewise.
* python/py-inferior.c (add_thread_object): Likewise.
* python/py-param.c (compute_enum_values): Likewise.
* python/py-value.c (valpy_call): Likewise.
* python/py-varobj.c (py_varobj_iter_next): Likewise.
* python/python.c (ensure_python_env): Likewise.
* record-btrace.c (record_btrace_start_replaying): Likewise.
* record-full.c (record_full_reg_alloc): Likewise.
(record_full_mem_alloc): Likewise.
(record_full_end_alloc): Likewise.
(record_full_core_xfer_partial): Likewise.
* regcache.c (get_thread_arch_aspace_regcache): Likewise.
* remote-fileio.c (remote_fileio_init_fd_map): Likewise.
* remote-notif.c (remote_notif_state_allocate): Likewise.
* remote.c (demand_private_info): Likewise.
(remote_notif_stop_alloc_reply): Likewise.
(remote_enable_btrace): Likewise.
* reverse.c (save_bookmark_command): Likewise.
* rl78-tdep.c (rl78_gdbarch_init): Likewise.
* rx-tdep.c (rx_gdbarch_init): Likewise.
* s390-linux-nat.c (s390_insert_watchpoint): Likewise.
* ser-go32.c (dos_get_tty_state): Likewise.
(dos_copy_tty_state): Likewise.
* ser-mingw.c (ser_windows_open): Likewise.
(ser_console_wait_handle): Likewise.
(ser_console_get_tty_state): Likewise.
(make_pipe_state): Likewise.
(net_windows_open): Likewise.
* ser-unix.c (hardwire_get_tty_state): Likewise.
(hardwire_copy_tty_state): Likewise.
* solib-aix.c (solib_aix_new_lm_info): Likewise.
* solib-dsbt.c (dsbt_current_sos): Likewise.
(dsbt_relocate_main_executable): Likewise.
* solib-frv.c (frv_current_sos): Likewise.
(frv_relocate_main_executable): Likewise.
* solib-spu.c (spu_bfd_fopen): Likewise.
* solib-svr4.c (lm_info_read): Likewise.
(svr4_copy_library_list): Likewise.
(svr4_default_sos): Likewise.
* source.c (find_source_lines): Likewise.
(line_info): Likewise.
(add_substitute_path_rule): Likewise.
* spu-linux-nat.c (spu_bfd_open): Likewise.
* spu-tdep.c (info_spu_dma_cmdlist): Likewise.
* stabsread.c (dbx_lookup_type): Likewise.
(read_type): Likewise.
(read_member_functions): Likewise.
(read_struct_fields): Likewise.
(read_baseclasses): Likewise.
(read_args): Likewise.
(_initialize_stabsread): Likewise.
* stack.c (func_command): Likewise.
* stap-probe.c (handle_stap_probe): Likewise.
* symfile.c (addrs_section_sort): Likewise.
(addr_info_make_relative): Likewise.
(load_section_callback): Likewise.
(add_symbol_file_command): Likewise.
(init_filename_language_table): Likewise.
* symtab.c (create_filename_seen_cache): Likewise.
(sort_search_symbols_remove_dups): Likewise.
(search_symbols): Likewise.
* target.c (make_cleanup_restore_target_terminal): Likewise.
* thread.c (new_thread): Likewise.
(enable_thread_stack_temporaries): Likewise.
(make_cleanup_restore_current_thread): Likewise.
(thread_apply_all_command): Likewise.
* tic6x-tdep.c (tic6x_gdbarch_init): Likewise.
* top.c (gdb_readline_wrapper): Likewise.
* tracefile-tfile.c (tfile_trace_file_writer_new): Likewise.
* tracepoint.c (trace_find_line_command): Likewise.
(all_tracepoint_actions_and_cleanup): Likewise.
(make_cleanup_restore_current_traceframe): Likewise.
(get_uploaded_tp): Likewise.
(get_uploaded_tsv): Likewise.
* tui/tui-data.c (tui_alloc_generic_win_info): Likewise.
(tui_alloc_win_info): Likewise.
(tui_alloc_content): Likewise.
(tui_add_content_elements): Likewise.
* tui/tui-disasm.c (tui_find_disassembly_address): Likewise.
(tui_set_disassem_content): Likewise.
* ui-file.c (ui_file_new): Likewise.
(stdio_file_new): Likewise.
(tee_file_new): Likewise.
* utils.c (make_cleanup_restore_integer): Likewise.
(add_internal_problem_command): Likewise.
* v850-tdep.c (v850_gdbarch_init): Likewise.
* valops.c (find_oload_champ): Likewise.
* value.c (allocate_value_lazy): Likewise.
(record_latest_value): Likewise.
(create_internalvar): Likewise.
* varobj.c (install_variable): Likewise.
(new_variable): Likewise.
(new_root_variable): Likewise.
(cppush): Likewise.
(_initialize_varobj): Likewise.
* windows-nat.c (windows_make_so): Likewise.
* x86-nat.c (x86_add_process): Likewise.
* xcoffread.c (arrange_linetable): Likewise.
(allocate_include_entry): Likewise.
(process_linenos): Likewise.
(SYMBOL_DUP): Likewise.
(xcoff_start_psymtab): Likewise.
(xcoff_end_psymtab): Likewise.
* xml-support.c (gdb_xml_parse_attr_ulongest): Likewise.
* xtensa-tdep.c (xtensa_register_type): Likewise.
* gdbarch.c: Regenerate.
* gdbarch.h: Regenerate.

gdb/gdbserver/ChangeLog:

* ax.c (gdb_parse_agent_expr): Likewise.
(compile_bytecodes): Likewise.
* dll.c (loaded_dll): Likewise.
* event-loop.c (append_callback_event): Likewise.
(create_file_handler): Likewise.
(create_file_event): Likewise.
* hostio.c (handle_open): Likewise.
* inferiors.c (add_thread): Likewise.
(add_process): Likewise.
* linux-aarch64-low.c (aarch64_linux_new_process): Likewise.
* linux-arm-low.c (arm_new_process): Likewise.
(arm_new_thread): Likewise.
* linux-low.c (add_to_pid_list): Likewise.
(linux_add_process): Likewise.
(handle_extended_wait): Likewise.
(add_lwp): Likewise.
(enqueue_one_deferred_signal): Likewise.
(enqueue_pending_signal): Likewise.
(linux_resume_one_lwp_throw): Likewise.
(linux_resume_one_thread): Likewise.
(linux_read_memory): Likewise.
(linux_write_memory): Likewise.
* linux-mips-low.c (mips_linux_new_process): Likewise.
(mips_linux_new_thread): Likewise.
(mips_add_watchpoint): Likewise.
* linux-x86-low.c (initialize_low_arch): Likewise.
* lynx-low.c (lynx_add_process): Likewise.
* mem-break.c (set_raw_breakpoint_at): Likewise.
(set_breakpoint): Likewise.
(add_condition_to_breakpoint): Likewise.
(add_commands_to_breakpoint): Likewise.
(clone_agent_expr): Likewise.
(clone_one_breakpoint): Likewise.
* regcache.c (new_register_cache): Likewise.
* remote-utils.c (look_up_one_symbol): Likewise.
* server.c (queue_stop_reply): Likewise.
(start_inferior): Likewise.
(queue_stop_reply_callback): Likewise.
(handle_target_event): Likewise.
* spu-low.c (fetch_ppc_memory): Likewise.
(store_ppc_memory): Likewise.
* target.c (set_target_ops): Likewise.
* thread-db.c (thread_db_load_search): Likewise.
(try_thread_db_load_1): Likewise.
* tracepoint.c (add_tracepoint): Likewise.
(add_tracepoint_action): Likewise.
(create_trace_state_variable): Likewise.
(cmd_qtdpsrc): Likewise.
(cmd_qtro): Likewise.
(add_while_stepping_state): Likewise.
* win32-low.c (child_add_thread): Likewise.
(get_image_name): Likewise.

9 years agoFix native follow-exec-mode "new"
Don Breazeal [Wed, 26 Aug 2015 20:31:59 +0000 (13:31 -0700)]
Fix native follow-exec-mode "new"

This patch fixes a segmentation fault in native GDB when
handling an exec event with follow-exec-mode set to "new".

The stack trace from the segfault was this:

 0  0x0000000000669594 in gdbarch_data (gdbarch=0x0, data=0x20da7a0)
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/gdbarch.c:4847
 1  0x00000000004d430e in get_remote_arch_state ()
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/remote.c:603
 2  0x00000000004d431e in get_remote_state ()
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/remote.c:616
 3  0x00000000004dda8b in discard_pending_stop_replies (inf=0x217c710)
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/remote.c:5775
 4  0x00000000006a5928 in observer_inferior_exit_notification_stub (
    data=0x4dda7a <discard_pending_stop_replies>, args_data=0x7fff12c258f0)
    at ./observer.inc:1137
 5  0x00000000006a419a in generic_observer_notify (subject=0x21dfbe0,
    args=0x7fff12c258f0)
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/observer.c:167
 6  0x00000000006a59ba in observer_notify_inferior_exit (inf=0x217c710)
    at ./observer.inc:1162
 7  0x00000000007981d5 in exit_inferior_1 (inftoex=0x217c710, silent=1)
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/inferior.c:244
 8  0x00000000007982f2 in exit_inferior_num_silent (num=1)
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/inferior.c:286
 9  0x000000000062f93d in follow_exec (ptid=...,
    execd_pathname=0x7fff12c259a0 "/scratch/dbreazea/sandbox/exec-nat/build/gdb/testsuite/gdb.base/execd-prog")
    at /scratch/dbreazea/sandbox/exec-nat/binutils-gdb/gdb/infrun.c:1195

In follow_exec we were creating a new inferior for the execd program,
as required by the exec mode, but we were doing it before calling
exit_inferior_num_silent on the original inferior.  So on entry to
exit_inferior_num_silent we had two inferiors with the same ptid.

In the calls made by exit_inferior_num_silent, the current inferior
is temporarily saved and replaced in order to make use of functions
that only operate on the current inferior (for example, in
do_all_continuations, called while deleting the threads of the original
inferior).  When we restored the original inferior, we just took the
first inferior that matched the ptid of the original and got the new
(wrong) one.  It hadn't been initialized yet and had no gdbarch
pointer, and GDB segfaulted.

The fix for that is to call exit_inferior_num_silent before adding the new
inferior, so that we never have two inferiors with the same ptid.  Then
exit_inferior_num_silent uses the original inferior as the current inferior
throughout, and can find a valid gdbarch pointer.

Once we have finished with the exit of the old inferior and added the
new one, we need to create a new thread for the new inferior.  In the
function that called follow_exec, handle_inferior_event_1,
ecs->event_thread now points to the thread that was deleted with the
exit of the original inferior.  To remedy this we create the new thread,
and once we return from follow_exec we reset ecs->event_thread.

Note that we are guaranteed that we can reset ecs->event_thread
safely using inferior_thread because we have set the current
inferior in follow_exec, and inferior_ptid was set by the call
to context_switch at the beginning of exec event handling.

gdb/ChangeLog:

* infrun.c (follow_exec): Re-order operations for
handling follow-exec-mode "new".
(handle_inferior_event_1): Assign ecs->event_thread
to the current thread.
* remote.c (get_remote_arch_state): Add an assertion.

9 years agoNew test for follow-exec-mode
Don Breazeal [Wed, 26 Aug 2015 20:26:36 +0000 (13:26 -0700)]
New test for follow-exec-mode

This patch implements a new GDB test for follow-exec-mode.  Although
there is a GDB test for debugging across an exec, there is no test for
follow-exec-mode.  This test is derived from gdb.base/foll-exec.exp,
and re-uses execd-prog.c as the program to exec.

The following behavior is tested:

follow-exec-mode == "same"
 - 'next' over the exec, check for one inferior
 - 'continue' past the exec to a breakpoint, check for one inferior
 - after the exec, use a 'run' command to run the current binary
follow-exec-mode == "new"
 - 'next' over the exec, check for two inferiors
 - 'continue' past the exec to a breakpoint, check for two inferiors
 - after the exec, use a 'run' command to run the current binary
 - after the exec, use the 'inferior' command to switch inferiors,
   then use a 'run' command to run the current binary

Note that single-step breakpoints do not survive across an exec.
There has to be a breakpoint in the execed program in order for
it to stop right after the exec.

gdb/testsuite/ChangeLog:

* gdb.base/foll-exec-2.c: New test program.
* gdb.base/foll-exec-2.exp: New test.

9 years agogdb: Add Markus Metzger as btrace maintainer
Pedro Alves [Wed, 26 Aug 2015 16:50:02 +0000 (17:50 +0100)]
gdb: Add Markus Metzger as btrace maintainer

gdb/ChangeLog:
2015-08-26  Pedro Alves  <palves@redhat.com>

* MAINTAINERS: Add Markus Metzger as btrace maintainer.

9 years agoUse stable sort for ld -r relocs
Alan Modra [Wed, 26 Aug 2015 08:02:08 +0000 (17:32 +0930)]
Use stable sort for ld -r relocs

A number of targets emit multiple relocs at a given r_offset, and
depend on those relocs staying in their original order.

PR 18867
* elflink.c (cmp_ext32l_r_offset, cmp_ext32b_r_offset): Delete.
(cmp_ext64l_r_offset, cmp_ext64b_r_offset): Delete.
(ext32l_r_offset, ext32b_r_offset, ext64l_r_offset, ext64b_r_offset):
New functions.
(elf_link_adjust_relocs): Use an insertion sort to sort relocs.

9 years agoSelect MIPS ABi based upon abiflags rather than e_flags.
Matthew Fortune [Wed, 26 Aug 2015 13:42:56 +0000 (14:42 +0100)]
Select MIPS ABi based upon abiflags rather than e_flags.

bfd PR ld/18401
* elfxx-mips.c (bfd_mips_isa_ext_mach): New function: Converts an
ISA value to a bfd machine number.
(bfd_mips_isa_ext): Update the ISA level and revision if
necessary.  Use mips_mach_extends_p to decide if the ISA extension
needs to be set.
(_bfd_mips_elf_merge_private_bfd_data): Allow the ISA extension to
be inferred from the ABI setting.  Set the ISA level from the
maximum of the incoming and outgoing ISA levels.

tests
* ld-mips-elf/mips-elf-flags.exp: Add more ISA conflict tests.

9 years agoFallout from "Allow symbol and label names to be enclosed in double quotes"
Alan Modra [Wed, 26 Aug 2015 13:25:41 +0000 (22:55 +0930)]
Fallout from "Allow symbol and label names to be enclosed in double quotes"

PR gas/18581
* config/tc-mn10200.c (md_assemble <mdr>): Move restore_line_pointer
call to where input line used to be restored.
* config/tc-mn10300.c (md_assemble <usp>): Remove redundant input
line restore.
* config/tc-tilepro.c (parse_reg_expression): Add regname var.

9 years agoAdd asprintf to the list of functions whose decls are checked at configure time.
Nick Clifton [Wed, 26 Aug 2015 09:56:26 +0000 (10:56 +0100)]
Add asprintf to the list of functions whose decls are checked at configure time.

PR gas/18842
* configure.ac (AS_CHECK_DECLS): Add asprintf.
* configure: Regenerate.
* config.in: Regenerate.

9 years agoFix problem where script specified both address and region for a section.
Cary Coutant [Wed, 26 Aug 2015 07:03:04 +0000 (00:03 -0700)]
Fix problem where script specified both address and region for a section.

If a script specifies both address and region for an output section
declaration, gold ignores the region specification. This can lead to
bogus "moves backward" errors. This patch fixes gold so that if a
section specifies both address and region, it will place the section
at the specified address in the region, and update the location counter
within the region.

gold/
PR gold/18847
* script-sections.cc (Memory_region::set_address): New method.
(Script_sections::find_memory_region): Add explicit_only parameter.
(Output_section_definition::set_section_addresses): Handle case where
script specifies both address and vma region.
* script-sections.h (Script_sections::find_memory_region): Add
explicit_only parameter.

9 years agoFix --no-as-needed when shared library is listed twice on the command line.
Cary Coutant [Fri, 21 Aug 2015 19:32:33 +0000 (12:32 -0700)]
Fix --no-as-needed when shared library is listed twice on the command line.

When a shared library is listed twice on the command line, the linker
ignores the second mention. If the first mention is in the scope of
an --as-needed option, and the second one is under the scope of a
--no-as-needed option, the --no-as-needed should take effect, but
doesn't.  This patch keeps track of the objects we've already seen,
and updates the --as-needed flag so that if a shared object is ever
seen with --no-as-needed, it will be marked as such.

gold/
PR gold/18859
* object.cc (Input_objects::add_object): Store objects in a map,
indexed by soname; update as-needed flag when necessary.
* object.h (Object::clear_as_needed): New method.
(Input_objects::so_names_): Change from set to map.

9 years agoFix internal error in gold when script uses section address in assignment.
Cary Coutant [Wed, 26 Aug 2015 00:41:19 +0000 (17:41 -0700)]
Fix internal error in gold when script uses section address in assignment.

When processing assignment expressions in a linker script, gold processes
absolute assignments early, but when one of those assignments involves the
address of a section that has not yet been finalized, we get an internal
error in address. This patch fixes the problem by gracefully returning
from expression evaluation even if the address is not yet valid, and
deferring the assignment in such a case.

gold/
PR gold/14746
* expression.cc (Expression::Expression_eval_info): Add
is_valid_pointer field.
(Expression::eval_maybe_dot): Add is_valid_pointer parameter.
Adjust all callers.
(Addr_expression::value_from_output_section): Check whether address
is valid.
* script.cc (Symbol_assignment::set_if_absolute): Defer assignment
if evaluation failed due to address that is not yet valid.
* script.h: (Expression::eval_maybe_dot): Add is_valid_pointer
parameter.

9 years agoFix regression introduced by fix for PR 18703.
Cary Coutant [Wed, 26 Aug 2015 00:53:53 +0000 (17:53 -0700)]
Fix regression introduced by fix for PR 18703.

When a user object overrides a versioned definition in a shared object,
it's possible to trigger an internal error because the versioned
definition we saw didn't override the unversioned definition. This
was introduced by the fix for PR 18703. This patch fixes the problem
by reordering the tests so that we do not check ret->is_default()
until we're confident that the symbol is in fact versioned.

gold/
PR gold/18866
PR gold/18703
* symtab.cc (Symbol_table): Reorder conditions to avoid internal error.

9 years agoAutomatic date update in version.in
GDB Administrator [Wed, 26 Aug 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoReplace assertions with warning messages about ELF binaries containing multiple symbo...
Nick Clifton [Tue, 25 Aug 2015 16:48:20 +0000 (17:48 +0100)]
Replace assertions with warning messages about ELF binaries containing multiple symbol tables.

PR binutils/18854
* elf.c (bfd_section_from_shdr): Replace assertions with warnings
about multiple symbol tables.

9 years agoUpdate MIPS tests for test failures on some configurations.
Simon Dardis [Tue, 25 Aug 2015 16:04:51 +0000 (17:04 +0100)]
Update MIPS tests for test failures on some configurations.

* gas/mips/micromips32-move.d: Update expected disassembly.
* gas/mips/move.d: Likewise.
* gas/mips/move.s: Fix for some MIPS configurations.