binutils-gdb.git
9 years ago[Ada] array of variant record subscripting
Joel Brobecker [Thu, 2 Apr 2015 18:09:15 +0000 (11:09 -0700)]
[Ada] array of variant record subscripting

Consider the following (Ada) array...

   A1 : Array_Type := (1 => (I => 0, S => <>),
                       2 => (I => 1, S => "A"),
                       3 => (I => 2, S => "AB"));

... where Array_Type is declared as follow:

   subtype Small_Type is Integer range 0 .. 10;
   type Record_Type (I : Small_Type := 0) is record
      S : String (1 .. I);
   end record;
   type Array_Type is array (Integer range <>) of Record_Type;

Trying to print the value of each element individually does not
always work. Printing the value of the first one does:

(gdb) p a1(1)
    $1 = (i => 0, s => "")

But printing the value of the subsequent ones often does not.
For instance:

    (gdb) p a1(2)
    $2 = (i => 1, s => "")  <<<--- s should be "A"
    (gdb) p a1(3)
    $3 = (i => 2, s => "")  <<<--- s should be "AB"

I traced the problem to ada_value_primitive_packed_val,
which is trying to perform the array subscripting by
extracting the value of the corresponding array element
into a buffer where the contents is now byte-aligned.

The element type that ada_value_primitive_packed_val gets passed
is a dynamic type. As it happens, that dynamic type can get resolved
thanks to:

      v = value_at (type, value_address (obj));
      type = value_type (v);

However, obj represents the array, so the address given in the call
to value_at represents the value of the first element. As a result,
the solution of component S's upper bound always gets resolved based
on the value of component I in the  first element of the array, whose
value is 0, thus leading to GDB mistakely resolving the element type
where S's upper bound is always 0.

The proper fix would be to systematically resolve the element type
first. But, this requires us to extract-and-realign the element's
value so as to be able to pass it as "valaddr" to resolve_dynamic_type.
In the meantime, it's easy to make the situation a little better by
passing "value_address (obj) + offset" as the object address. This
only works when BIT_OFFSET is nul, but that should be the case when
the element type is anything but a scalar, which seems to be the only
situation where it seems important to resolve the type now. And we're
not that worse off otherwise.

But we'll try to find a better solution in a separate patch.

gdb/ChangeLog:

        * ada-lang.c (ada_value_primitive_packed_val): Use a more
        correct address in call to value_at.  Adjust call to
        value_address accordingly.

9 years ago[Ada] Resolve dynamic type before trying to print it.
Joel Brobecker [Wed, 1 Apr 2015 22:46:54 +0000 (15:46 -0700)]
[Ada] Resolve dynamic type before trying to print it.

This is another required step towards trying to print the value of
an array of variant records. For instance:

   A1 : Array_Type := (1 => (I => 0, S => <>),
                       2 => (I => 1, S => "A"),
                       3 => (I => 2, S => "AB"));

... where Array_Type is an array of records whose size is variable:

   subtype Small_Type is Integer range 0 .. 10;
   type Record_Type (I : Small_Type := 0) is record
      S : String (1 .. I);
   end record;
   type Array_Type is array (Integer range <>) of Record_Type;

What happens is that the ada-valprint modules gets passed an array
whose element type is not resolved yet (since each element of the
array needs to be resolved separately). the module then recurses,
and eventually gets called with the first element of the array.
But because the element hasn't been resolved yet, we end up having
trouble printing its value soon after.

This patch fixes the issue by calling resolve_dynamic_type before
trying to print it.

With this patch, GDB is finally able to print the complete value
for variable "A1":

     (gdb) p a1
     $1 = ((i => 0, s => ""), (i => 1, s => "A"), (i => 2, s => "AB"))

gdb/ChangeLog:

        * ada-valprint.c (ada_val_print_1): Resolve TYPE before trying
        to print it.

9 years agoAdd valaddr support in dynamic property resolution.
Joel Brobecker [Wed, 1 Apr 2015 17:00:13 +0000 (10:00 -0700)]
Add valaddr support in dynamic property resolution.

This is the second part of enhancing the debugger to print the value
of arrays of records whose size is variable when only standard DWARF
info is available (no GNAT encoding). For instance:

   subtype Small_Type is Integer range 0 .. 10;
   type Record_Type (I : Small_Type := 0) is record
      S : String (1 .. I);
   end record;
   type Array_Type is array (Integer range <>) of Record_Type;

   A1 : Array_Type := (1 => (I => 0, S => <>),
                       2 => (I => 1, S => "A"),
                       3 => (I => 2, S => "AB"));

Currently, GDB prints the following output:

        (gdb) p a1
        $1 = (

The error happens while the ada-valprint module is trying to print
the value of an element of our array. Because of the fact that
the array's element (type Record_Type) has a variant size, the DWARF
info for our array provide the array's stride:

     <1><749>: Abbrev Number: 10 (DW_TAG_array_type)
        <74a>   DW_AT_name        : (indirect string, offset: 0xb6d): pck__T18s
        <74e>   DW_AT_byte_stride : 16
        <74f>   DW_AT_type        : <0x6ea>

And because our array has a stride, ada-valprint treats it the same
way as packed arrays (see ada-valprint.c::ada_val_print_array):

  if (TYPE_FIELD_BITSIZE (type, 0) > 0)
    val_print_packed_array_elements (type, valaddr, offset_aligned,
                                     0, stream, recurse,
                                     original_value, options);

The first thing that we should notice in the call above is that
the "valaddr" buffer and the associated offset (OFFSET_ALIGNED)
is passed, but that the corresponding array's address is not.
This can be explained by looking inside val_print_packed_array_elements,
where we see that the function unpacks each element of our array from
the buffer alone (ada_value_primitive_packed_val), and then prints
the resulting artificial value instead:

      v0 = ada_value_primitive_packed_val (NULL, valaddr + offset,
                                           (i0 * bitsize) / HOST_CHAR_BIT,
                                           (i0 * bitsize) % HOST_CHAR_BIT,
                                           bitsize, elttype);

      [...]
              val_print (elttype, value_contents_for_printing (v0),
                         value_embedded_offset (v0), 0, stream,
                         recurse + 1, v0, &opts, current_language);

Of particular interest, here, is the fact that we call val_print
with a null address, which is OK, since we're providing a buffer
instead (value_contents_for_printing). Also, providing an address
might not always possible, since packing could place elements at
boundaries that are not byte-aligned.

Things go south when val_print tries to see if there is a pretty-printer
that could be applied. In particular, one of the first things that
the Python pretty-printer does is to create a value using our buffer,
and the given address, which in this case is null (see call to
value_from_contents_and_address in gdbpy_apply_val_pretty_printer).

value_from_contents_and_address, in turn immediately tries to resolve
the type, using the given address, which is null. But, because our
array element is a record containing an array whose bound is the value
of one of its elements (the "s" component), the debugging info for
the array's upper bound is a reference...

 <3><71a>: Abbrev Number: 7 (DW_TAG_subrange_type)
    <71b>   DW_AT_type        : <0x724>
    <71f>   DW_AT_upper_bound : <0x703>

... to component "i" of our record...

 <2><703>: Abbrev Number: 5 (DW_TAG_member)
    <704>   DW_AT_name        : i
    <706>   DW_AT_decl_file   : 2
    <707>   DW_AT_decl_line   : 6
    <708>   DW_AT_type        : <0x6d1>
    <70c>   DW_AT_data_member_location: 0

... where that component is located at offset 0 of the start
of the record. dwarf2_evaluate_property correctly determines
the offset where to load the value of the bound from, but then
tries to read that value from inferior memory using the address
that was given, which is null. See case PROP_ADDR_OFFSET in
dwarf2_evaluate_property:

        val = value_at (baton->offset_info.type,
                        pinfo->addr + baton->offset_info.offset);

This triggers a memory error, which then causes the printing to terminate.

Since there are going to be situations where providing an address
alone is not going to be sufficient (packed arrays where array elements
are not stored at byte boundaries), this patch fixes the issue by
enhancing the type resolution to take both address and data. This
follows the same principle as the val_print module, where both
address and buffer ("valaddr") can be passed as arguments. If the data
has already been fetched from inferior memory (or provided by the
debugging info in some form -- Eg a constant), then use that data
instead of reading it from inferior memory.

Note that this should also be a good step towards being able to handle
dynamic types whose value is stored outside of inferior memory
(Eg: in a register).

With this patch, GDB isn't able to print all of A1, but does perform
a little better:

    (gdb) p a1
    $1 = ((i => 0, s => , (i => 1, s => , (i => 2, s => )

There is another issue which is independent of this one, and will
therefore be patched separately.

gdb/ChangeLog:

        * dwarf2loc.h (struct property_addr_info): Add "valaddr" field.
        * dwarf2loc.c (dwarf2_evaluate_property): Add handling of
        pinfo->valaddr.
        * gdbtypes.h (resolve_dynamic_type): Add "valaddr" parameter.
        * gdbtypes.c (resolve_dynamic_struct): Set pinfo.valaddr.
        (resolve_dynamic_type_internal): Set pinfo.valaddr.
        Add handling of addr_stack->valaddr.
        (resolve_dynamic_type): Add "valaddr" parameter.
        Set pinfo.valaddr field.
        * ada-lang.c (ada_discrete_type_high_bound): Update call to
        resolve_dynamic_type.
        (ada_discrete_type_low_bound): Likewise.
        * findvar.c (default_read_var_value): Likewise.
        * value.c (value_from_contents_and_address): Likewise.

9 years agopreserve the bit stride when resolving an array type.
Joel Brobecker [Tue, 31 Mar 2015 14:59:35 +0000 (07:59 -0700)]
preserve the bit stride when resolving an array type.

Consider the following (Ada) variable...

   A1 : Array_Type := (1 => (I => 0, S => <>),
                       2 => (I => 1, S => "A"),
                       3 => (I => 2, S => "AB"));

... where Array_Type is an array of records whose size is variable:

   subtype Small_Type is Integer range 0 .. 10;
   type Record_Type (I : Small_Type := 0) is record
      S : String (1 .. I);
   end record;
   type Array_Type is array (Integer range <>) of Record_Type;

Trying to print the value of this array currently results in the following
error:

    (gdb) p a1
    Cannot access memory at address 0x61c000

What happens in this case, is that the compiler describes our array
as an array with a specific stride (and bounds being static 1..3):

 <1><749>: Abbrev Number: 10 (DW_TAG_array_type)
    <74a>   DW_AT_name        : (indirect string, offset: 0xb6d): pck__T18s
    <74e>   DW_AT_byte_stride : 16
    <74f>   DW_AT_type        : <0x6ea>
 <2><757>: Abbrev Number: 11 (DW_TAG_subrange_type)
    <758>   DW_AT_type        : <0x75e>
    <75c>   DW_AT_upper_bound : 3

This is because we cannot use, in this case, the size of the record
to determine that stride, since the size of the record depends on
its contents. So the compiler helps us by providing that stride.

The problems start when trying to resolve that type. Because the elements
contained in that array type are dynamic, the array itself is considered
dynamic, and thus we end up creating a resolved version of that array.
And during that resolution, we were not handling the case where the array
had a stride. See gdbtypes.c::resolve_dynamic_array...

  return create_array_type (copy_type (type),
                            elt_type,
                            range_type);

As a result, we created an array whose stride was based on the size
of elt_type, which a record whose size isn't static and irrelevant
regardless.

This patch fixes is by calling create_array_type_with_stride instead.

As it happens, there is another issue for us to be able to print
the value of our array, but those are independent of this patch
and will be handled separately. For now, the patch allows us to
get rid of the first error, and the output is now:

     (gdb) p a1
     $1 = (

gdb/ChangeLog:

* gdbtypes.c (resolve_dynamic_array): Use
create_array_type_with_stride instead of create_array_type.

9 years ago[AARCH64] Positively emit symbols for alignment
Renlin Li [Tue, 5 May 2015 16:48:18 +0000 (17:48 +0100)]
[AARCH64] Positively emit symbols for alignment

2015-05-05  Renlin Li  <renlin.li@arm.com>

  gas/
    * config/tc-aarch64.c (aarch64_init_frag): Always generate mapping symbols.

  gas/testsuite/
    * gas/aarch64/mapping_5.d: New.
    * gas/aarch64/mapping_5.s: New.
    * gas/aarch64/mapping_6.d: New.
    * gas/aarch64/mapping_6.s: New.

9 years agoAdd support to the MSP430 linker for the automatic placement of code and data into...
Nick Clifton [Tue, 5 May 2015 12:38:00 +0000 (13:38 +0100)]
Add support to the MSP430 linker for the automatic placement of code and data into either low or high memory regions.

gas * config/tc-msp430.c (MAX_OP_LEN): Increase to 4096.
(msp430_make_init_symbols): New function.
(msp430_section): Call it.
(msp430_frob_section): Likewise.

ld * emulparams/msp430elf.sh (TEMPLATE_NAME): Change to msp430.
* scripttempl/msp430.sc (.text): Add .lower.text and .either.text.
(.data): Add .lower.data and .either.data.
(.bss): Add .lower.bss and .either.bss.
(.rodata): Add .lower.rodata and .either.rodata.
* emultempl/msp430.em: New file.  Implements a new orphan
placement algorithm that divides sections between lower and upper
memory regions.
* Makefile.am (emsp430elf.c): Depend upon msp430.em.
*emsp430X.c): Likewise.
* Makefine.in: Regenerate.

9 years agoxtensa: optimize trampolines relaxation
Max Filippov [Fri, 1 May 2015 08:39:12 +0000 (11:39 +0300)]
xtensa: optimize trampolines relaxation

Currently every fixup in the current segment is checked when relaxing
trampoline frag. This is very expensive. Make a searchable array of
fixups pointing at potentially oversized jumps at the beginning of every
relaxation pass and only check subset of this cache in the reach of
single jump from the trampoline frag currently being relaxed.

Original profile:

% time    self  children    called     name
-----------------------------------------
        370.16  593.38 12283048/12283048     relax_segment
  98.4  370.16  593.38 12283048         xtensa_relax_frag
         58.91  269.26 2691463834/2699602236     xtensa_insnbuf_from_chars
         68.35   68.17 811266668/813338977     S_GET_VALUE
         36.85   29.51 2684369246/2685538060     xtensa_opcode_decode
         28.34    8.84 2684369246/2685538060     xtensa_format_get_slot
         12.39    5.94 2691463834/2699775044     xtensa_format_decode
          0.03    4.60 4101109/4101109     relax_frag_for_align
          0.18    1.76  994617/994617      relax_frag_immed
          0.07    0.09 24556277/24851220     new_logical_line
          0.06    0.00 12283048/14067410     as_where
          0.04    0.00 7094588/15460506     xtensa_format_num_slots
          0.00    0.00       1/712477      xtensa_insnbuf_alloc
-----------------------------------------

Same data, after optimization:

% time    self  children    called     name
-----------------------------------------
          0.51    7.47 12283048/12283048     relax_segment
  58.0    0.51    7.47 12283048         xtensa_relax_frag
          0.02    4.08 4101109/4101109     relax_frag_for_align
          0.18    1.39  994617/994617      relax_frag_immed
          0.01    0.98     555/555         xtensa_cache_relaxable_fixups
          0.21    0.25 7094588/16693271     xtensa_insnbuf_from_chars
          0.06    0.12 24556277/24851220     new_logical_line
          0.06    0.00 7094588/15460506     xtensa_format_num_slots
          0.02    0.04 7094588/16866079     xtensa_format_decode
          0.05    0.00 12283048/14067410     as_where
          0.00    0.00       1/712477      xtensa_insnbuf_alloc
          0.00    0.00   93808/93808       xtensa_find_first_cached_fixup
-----------------------------------------

2015-05-02  Max Filippov  <jcmvbkbc@gmail.com>
gas/
* config/tc-xtensa.c (cached_fixupS, fixup_cacheS): New typedefs.
(struct cached_fixup, struct fixup_cache): New structures.
(fixup_order, xtensa_make_cached_fixup),
(xtensa_realloc_fixup_cache, xtensa_cache_relaxable_fixups),
(xtensa_find_first_cached_fixup, xtensa_delete_cached_fixup),
(xtensa_add_cached_fixup): New functions.
(xtensa_relax_frag): Cache fixups pointing at potentially
oversized jumps at the beginning of every relaxation pass. Only
check subset of this cache in the reach of single jump from the
trampoline frag currently being relaxed.

9 years agoAutomatic date update in version.in
GDB Administrator [Tue, 5 May 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoAutomatic date update in version.in
GDB Administrator [Mon, 4 May 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoAutomatic date update in version.in
GDB Administrator [Sun, 3 May 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoChange Section_id type to use Relobj* instead of Object*.
Cary Coutant [Sat, 2 May 2015 15:40:09 +0000 (08:40 -0700)]
Change Section_id type to use Relobj* instead of Object*.

2015-04-29  Cary Coutant  <cary@google.com>
    Rafael Ávila de Espíndola <rafael.espindola@gmail.com>

gold/
* gc.h (Garbage_collection::is_section_garbage): Change Object*
to Relobj*.
(Garbage_collection::add_reference): Likewise.
(Garbage_collection::gc_process_relocs): Likewise. Don't push
object/shndx pair onto *secvec for dynamic objects. Don't follow
relocations pointing to dynamic objects for GC.
* icf.cc (Icf::find_identical_sections): Change Object* to Relobj*.
(Icf::unfold_section): Likewise.
(Icf::is_section_folded): Likewise.
(Icf::get_folded_section): Likewise.
* icf.h: (Icf::get_folded_section): Likewise.
(Icf::unfold_section): Likewise.
(Icf::is_section_folded): Likewise.
(Icf::section_has_function_pointers): Likewise.
(Icf::set_section_has_function_pointers): Likewise.
* object.h (Section_id): Likewise.
(Const_section_id): Likewise.
* output.cc (Output_section::update_section_layout): Likewise.
* output.h: (Output_section_lookup_maps::find_relaxed_input_section):
Likewise.
* plugin.cc (update_section_order): Likewise.
(unique_segment_for_sections): Likewise.
* powerpc.cc (Powerpc_relobj::add_reference): Likewise.
(Target_powerpc::do_gc_add_reference): Likewise.
(Target_powerpc::gc_process_relocs): Likewise.
(Target_powerpc::do_gc_add_reference): Likewise.
* symtab.cc (Symbol_table::is_section_folded): Likewise.
(Symbol_table::gc_mark_symbol): Likewise.
* symtab.h: (Symbol_table::is_section_folded): Likewise.
* target.h: (Sized_target::gc_add_reference): Likewise.
(Sized_target::do_gc_add_reference): Likewise.

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

9 years agoFix typos in previous patch.
DJ Delorie [Fri, 1 May 2015 19:08:07 +0000 (15:08 -0400)]
Fix typos in previous patch.

* config/rl78-parse.y (MULU): Remove ISA_G14.
(MULH, DIVHU, DIVWU, MACHI, MACH): Update error strings.

9 years agoSync filenames.h with gcc
H.J. Lu [Fri, 1 May 2015 16:10:03 +0000 (09:10 -0700)]
Sync filenames.h with gcc

Merge with gcc:
2014-11-11  Anthony Brandon  <anthony.brandon@gmail.com>
    Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR driver/36312
* filenames.h: Add prototype for canonical_filename_eq.

9 years agoConfigure zlib with --enable-host-shared for shared bfd
H.J. Lu [Fri, 1 May 2015 15:33:55 +0000 (08:33 -0700)]
Configure zlib with --enable-host-shared for shared bfd

When bfd is configured as a shared library, we need to configure zlib
with --enable-host-shared since zlib is used by bfd.

PR ld/18355
* Makefile.def: Add extra_configure_flags to host zlib.
* configure.ac (extra_host_zlib_configure_flags): New.  Set
to --enable-host-shared When bfd is to be built as shared
library.  AC_SUBST.
* Makefile.in: Regenerated.
* configure: Likewise.

9 years agoRemove i386_elf_emit_arch_note
H.J. Lu [Fri, 1 May 2015 15:29:16 +0000 (08:29 -0700)]
Remove i386_elf_emit_arch_note

This x86 assembler patch:

https://sourceware.org/ml/binutils/2001-11/msg00344.html

generates a .note section for .arch directive so that GDB can tell which
architecture an i386 binary belongs:

https://sourceware.org/ml/binutils/2001-11/msg00271.html

However, x86 assembly code can have any instructions.  A .note section
doesn't help.  This patch removes it.

gas/

* config/tc-i386.c (i386_elf_emit_arch_note): Removed.
* config/tc-i386.h (md_end): Likewise.
(i386_elf_emit_arch_note): Likewise.

gas/testsuite/

* gas/i386/i386.exp: Run note.
* gas/i386/note.d: New file.
* gas/i386/note.s: Likewise.

9 years agoSupport ix86-*-elf*
H.J. Lu [Fri, 1 May 2015 12:02:30 +0000 (05:02 -0700)]
Support ix86-*-elf*

bfd/

* config.bfd: Support i[3-7]86-*-elf*.

gas/

* configure.tgt: Support i386-*-elf*.

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

9 years agoMake RL78 disassembler and simulator respect ISA for mul/div
DJ Delorie [Thu, 30 Apr 2015 19:25:49 +0000 (15:25 -0400)]
Make RL78 disassembler and simulator respect ISA for mul/div

[gas]
* config/rl78-defs.h (rl78_isa_g10): New.
(rl78_isa_g13): New.
(rl78_isa_g14): New.
* config/rl78-parse.y (ISA_G10): New.
(ISA_G13): New.
(ISA_G14): New.
(MULHU, MULH, MULU, DIVHU, DIVWU, MACHU, MACH): Use them.
* config/tc-rl78.c (rl78_isa_g10): New.
(rl78_isa_g13): New.
(rl78_isa_g14): New.

[gdb]
* rl78-tdep.c (rl78_analyze_prologue): Pass RL78_ISA_DEFAULT to
rl78_decode_opcode

[include]
* dis-asm.h (print_insn_rl78_g10): New.
(print_insn_rl78_g13): New.
(print_insn_rl78_g14): New.
(rl78_get_disassembler): New.
* opcode/rl78.h (RL78_Dis_Isa): New.
(rl78_decode_opcode): Add ISA parameter.

[opcodes]
* disassemble.c (disassembler): Choose suitable disassembler based
on E_ABI.
* rl78-decode.opc (rl78_decode_opcode): Take ISA parameter.  Use
it to decode mul/div insns.
* rl78-decode.c: Regenerate.
* rl78-dis.c (print_insn_rl78): Rename to...
(print_insn_rl78_common): ...this, take ISA parameter.
(print_insn_rl78): New.
(print_insn_rl78_g10): New.
(print_insn_rl78_g13): New.
(print_insn_rl78_g14): New.
(rl78_get_disassembler): New.

[sim]
* rl78/cpu.c (g14_multiply): New.
* rl78/cpu.h (g14_multiply): New.
* rl78/load.c (rl78_load): Decode ISA completely.
* rl78/main.c (main): Expand -M to include other ISAs.
* rl78/rl78.c (decode_opcode): Decode based on ISA.
* rl78/trace.c (rl78_disasm_fn): New.
(sim_disasm_init): Reset it.
(sim_disasm_one): Get correct disassembler for ISA.

9 years agoUse "else if" on cpu_arch_isa
H.J. Lu [Thu, 30 Apr 2015 15:36:17 +0000 (08:36 -0700)]
Use "else if" on cpu_arch_isa

* config/tc-i386.c (i386_target_format): Use "else if" on
cpu_arch_isa.

9 years agoFix handling of relocs for the MeP target.
Nick Clifton [Thu, 30 Apr 2015 14:57:41 +0000 (15:57 +0100)]
Fix handling of relocs for the MeP target.

bfd PR 18317
* elf32-mep.c (MEPREL): Use bfd_elf_generic_reloc instead of
mep_reloc.
(mep_reloc): Delete unused function.

bin * readelf.c (get_machine_flags): Add description of MeP flags.

tests * binutils-all/objdump.exp (cpus_expected): Add MeP CPU names.

9 years agoUndef elf_backend_post_process_headers for Solaris
H.J. Lu [Thu, 30 Apr 2015 13:52:34 +0000 (06:52 -0700)]
Undef elf_backend_post_process_headers for Solaris

* elf32-i386.c (elf_backend_post_process_headers): Undef for
Solaris 2.

9 years agoGAS ARM: Warn if the user creates a symbol with the same name as an instruction.
Nick Clifton [Thu, 30 Apr 2015 10:17:55 +0000 (11:17 +0100)]
GAS ARM: Warn if the user creates a symbol with the same name as an instruction.

PR gas/18347
gas * config/tc-arm.c (md_undefined_symbol): Issue a warning message
(if enabled) when the user creates a symbol with the same name as
an ARM instruction.
(flag_warn_syms): New static variable.
(arm_opts): Add mwarn-syms and mno-warn-syms.
* doc/c-arm.texi (ARM Options): Document the -m[no-]warn-syms
options.

tests * gas/arm/pr18347.s: New file: Test case.
* gas/arm/pr18347.l: New file: Expected assembler output.
* gas/arm/pr18347.d: New file: Test driver.

9 years agoAdds documentation of GAS's .zero directive.
Nick Clifton [Thu, 30 Apr 2015 09:13:53 +0000 (10:13 +0100)]
Adds documentation of GAS's .zero directive.

PR gas/18353
* doc/as.texinfo (Zero): Add documentation of the .zero pseudo-op.

9 years agoSkip setting HW watchpoint if skip_hw_watchpoint_multi_tests in gdb.base/break-idempo...
Yao Qi [Thu, 30 Apr 2015 09:08:10 +0000 (10:08 +0100)]
Skip setting HW watchpoint if skip_hw_watchpoint_multi_tests in gdb.base/break-idempotent.exp

Hi,
I see this fails below on arm linux native testing and remote testing
with "set remote hardware-watchpoint-limit 1",

 rwatch global^M
 There are not enough available hardware resources for this watchpoint.^M
 (gdb) FAIL: gdb.base/break-idempotent.exp: always-inserted off: rwatch: twice: rwatch global

gdb.base/break-idempotent.exp sets two breakpoints/watchpoints on the
same address.  GDB isn't smart enough calculate these two HW
watchpoints can fit in one HW debug register, so the error message
above isn't necessary (there is one HW watchpoint register on arm).
Because target_ops interface can_use_hardware_watchpoint doesn't
pass enough information to the target backend.

Note that if I don't "set remote hardware-watchpoint-limit 1" in
remote testing, this test passes without fails.  However without
"set remote hardware-watchpoint-limit 1", many other watchpoint
tests fail.

This patch is to add a check to skip_hw_watchpoint_multi_tests
for rwatch and awatch.  We can add such check for watch as well,
but GDB is able to switch to software watchpoint if HW resource
isn't available, it doesn't cause any fail, I decide not to skip.

gdb/testsuite:

2015-04-30  Yao Qi  <yao.qi@linaro.org>

* gdb.base/break-idempotent.exp: If
skip_hw_watchpoint_multi_tests returns true, skip the tests
on "rwatch" and "awatch".

9 years agoSkip gdb.base/relativedebug.exp if libc doesn't have debug info
Yao Qi [Thu, 30 Apr 2015 08:55:06 +0000 (09:55 +0100)]
Skip gdb.base/relativedebug.exp if libc doesn't have debug info

Hi,
I see the fail in gdb.base/relativedebug.exp on aarch64 box on which
glibc doesn't have debug info,

 bt^M
 #0 0x0000002000061a88 in raise () from /lib/aarch64-linux-gnu/libc.so.6^M
 #1 0x0000002000064efc in abort () from /lib/aarch64-linux-gnu/libc.so.6^M
 #2 0x0000000000400640 in handler (signo=14) at ../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:25^M
 #3 <signal handler called>^M
 #4 0x00000020000cc478 in ?? () from /lib/aarch64-linux-gnu/libc.so.6^M
 #5 0x0000000000400664 in main () at ../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:32^M
 (gdb) FAIL: gdb.base/relativedebug.exp: pause found in backtrace

if glibc has debug info, this test doesn't fail.

In sysdeps/unix/sysv/linux/generic/pause.c, __libc_pause calls
__syscall_pause,

  static int
  __syscall_pause (void)
  {
    sigset_t set;

    int rc =
      INLINE_SYSCALL (rt_sigprocmask, 4, SIG_BLOCK, NULL, &set, _NSIG / 8);
    if (rc == 0)
      rc = INLINE_SYSCALL (rt_sigsuspend, 2, &set, _NSIG / 8);

    return rc;
  }

  int
  __libc_pause (void)
  {
    if (SINGLE_THREAD_P)
      return __syscall_pause ();     <--- tail call

    int oldtype = LIBC_CANCEL_ASYNC ();

    int result = __syscall_pause ();

    LIBC_CANCEL_RESET (oldtype);

    return result;
  }

and GDB unwinder is confused by the GCC optimized code,

(gdb) disassemble pause
Dump of assembler code for function pause:
   0x0000007fb7f274c4 <+0>:     stp     x29, x30, [sp,#-32]!
   0x0000007fb7f274c8 <+4>:     mov     x29, sp
   0x0000007fb7f274cc <+8>:     adrp    x0, 0x7fb7fd2000
   0x0000007fb7f274d0 <+12>:    ldr     w0, [x0,#364]
   0x0000007fb7f274d4 <+16>:    stp     x19, x20, [sp,#16]
   0x0000007fb7f274d8 <+20>:    cbnz    w0, 0x7fb7f274e8 <pause+36>

   0x0000007fb7f274dc <+24>:    ldp     x19, x20, [sp,#16]
   0x0000007fb7f274e0 <+28>:    ldp     x29, x30, [sp],#32
   0x0000007fb7f274e4 <+32>:    b       0x7fb7f27434    <---- __syscall_pause

   0x0000007fb7f274e8 <+36>:    bl      0x7fb7f5e080

Note that the program stops in __syscall_pause, but its symbol is
stripped in glibc, so GDB doesn't know where the program stops.
__syscall_pause is a tail call in __libc_pause, so it returns to main
instead of __libc_pause.  As a result, the backtrace is like,

 #0  0x0000007fb7ebca88 in raise () from /lib/aarch64-linux-gnu/libc.so.6
 #1  0x0000007fb7ebfefc in abort () from /lib/aarch64-linux-gnu/libc.so.6
 #2  0x0000000000400640 in handler (signo=14) at ../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:25
 #3  <signal handler called>
 #4  0x0000007fb7f27478 in ?? () from /lib/aarch64-linux-gnu/libc.so.6   <-- [in __syscall_pause]
 #5  0x0000000000400664 in main () at ../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:32

looks GDB does nothing wrong here.  I looked back at the test case
gdb.base/relativedebug.exp, which was added
https://sourceware.org/ml/gdb-patches/2006-10/msg00305.html
This test was indented to test the problem that "backtraces no longer
display some libc functions" after separate debug info is installed.
IOW, it makes few sense to test against libc which doesn't have debug
info at all, such as my case.

This patch is to tweak the test case to catch the output of
"info shared", if "(*)" is found for libc.so, which means libc doesn't
have debug info, then skip the test.

gdb/testsuite:

2015-04-30  Yao Qi  <yao.qi@linaro.org>

* gdb.base/relativedebug.exp: Invoke gdb command
"info sharedlibrary", and if libc.so doesn't have debug info,
skip the test.

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

9 years agoPR python/18285
Doug Evans [Wed, 29 Apr 2015 20:24:21 +0000 (13:24 -0700)]
PR python/18285

gdb/ChangeLog:

PR python/18285
* NEWS: Document new gdb.XMethodWorker.get_result_type method.
* eval.c (evaluate_subexp_standard) <OP_FUNCALL>: Handle
EVAL_AVOID_SIDE_EFFECTS for xmethods.
* extension-priv.h (struct extension_language_ops)
<get_xmethod_result_type>: New member.
* extension.c (get_xmethod_result_type): New function.
* extension.h (get_xmethod_result_type): Declare.
* python/py-xmethods.c (get_result_type_method_name): New static
global.
(py_get_result_type_method_name): Ditto.
(gdbpy_get_xmethod_result_type): New function.
(gdbpy_initialize_xmethods): Initialize py_get_result_type_method_name.
* python/python-internal.h (gdbpy_get_xmethod_result_type): Declare.
* python/python.c (python_extension_ops): Add
gdbpy_get_xmethod_result_type.
* python/lib/gdb/xmethod.py (XMethodWorker): Add get_result_type.
* valarith.c (value_x_binop): Handle EVAL_AVOID_SIDE_EFFECTS for
xmethods.
(value_x_unop): Ditto.
* value.c (result_type_of_xmethod): New function.
* value.h (result_type_of_xmethod): Declare.

gdb/testsuite/ChangeLog:

* gdb.python/py-xmethods.exp: Add ptype tests.
* gdb.python/py-xmethods.py (E_method_char_worker): Add
get_result_type method.

gdb/doc/ChangeLog:

* python.texi (Xmethod API) <gdb.XMethodWorker.get_result_type>:
Document.
(Writing an Xmethod): Add get_result_type to example.

9 years agoFix an internal error in GAS when assembling a bogus piece of source code.
Nick Clifton [Wed, 29 Apr 2015 16:09:05 +0000 (17:09 +0100)]
Fix an internal error in GAS when assembling a bogus piece of source code.

gas PR 18256
* config/tc-arm.c (encode_arm_cp_address): Issue an error message
if the operand is neither a register nor a vector.

tests * gas/arm/pr18256.s: New file: Test case.
* gas/arm/pr18256.l: New file: Expected assembler output.
* gas/arm/pr18256.d: New file: Test driver.

9 years agoLink the last *normal against libfoozlib.so
H.J. Lu [Wed, 29 Apr 2015 15:43:22 +0000 (08:43 -0700)]
Link the last *normal against libfoozlib.so

Link the last zlibnormal gnunormal and gabinormal against libfoozlib.so
so that their only differences are DWARF debug sections.

PR ld/18354
* ld-elf/compress.exp (run_tests): Link the last zlibnormal,
gnunormal and gabinormal against libfoozlib.so.

9 years agoFix 18354
H.J. Lu [Wed, 29 Apr 2015 15:04:53 +0000 (08:04 -0700)]
Fix 18354

9 years agoUpdated translations for various binutils components.
Nick Clifton [Wed, 29 Apr 2015 15:24:52 +0000 (16:24 +0100)]
Updated translations for various binutils components.

gold * po/fi.po: Updated Finnish translation.

opcodes * po/fr.po: Updated French translation.

gprof * po/da.po: Update Danish translation.

9 years agoUse software watchpoints if hardware watchpoints are not available when testing gdb...
Luis Machado [Wed, 29 Apr 2015 15:22:24 +0000 (12:22 -0300)]
Use software watchpoints if hardware watchpoints are not available when testing gdb.base/watch-bitfields.exp

There are targets GDB thinks support hardware watchpoints, but in reality they
don't.  Though it may seem that hardware watchpoint creation was successful,
the actual insertion of such watchpoint will fail when GDB moves the inferior.

(gdb) watch -location q.a^M
Hardware watchpoint 2: -location q.a^M
(gdb) PASS: gdb.base/watch-bitfields.exp: -location watch against bitfields: watch -location q.a
watch -location q.e^M
Hardware watchpoint 3: -location q.e^M
(gdb) PASS: gdb.base/watch-bitfields.exp: -location watch against bitfields: watch -location q.e
print q.a^M
$1 = 0^M
(gdb) PASS: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 0->1: print expression before
continue^M
Continuing.^M
Warning:^M
Could not insert hardware watchpoint 2.^M
Could not insert hardware watchpoint 3.^M
Could not insert hardware breakpoints:^M
You may have requested too many hardware breakpoints/watchpoints.^M
^M
(gdb) FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 0->1: continue

This leads to a number of FAILs:

FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 0->1: continue
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 0->1: print expression after
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.e: 0->5: continue
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.e: 0->5: print expression after
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 1->0: print expression before
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.a: 1->0: continue
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.e: 5->4: print expression before
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.e: 5->4: continue
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: q.e: 5->4: print expression after
FAIL: gdb.base/watch-bitfields.exp: -location watch against bitfields: continue until exit
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 0->4: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 0->4: print expression after
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 4->10: print expression before
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 4->10: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 4->10: print expression after
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 10->3: print expression before
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 10->3: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 10->3: print expression after
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 3->2: print expression before
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 3->2: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 3->2: print expression after
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 2->1: print expression before
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 2->1: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 2->1: print expression after
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 1->0: print expression before
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: q.d + q.f + q.g: 1->0: continue
FAIL: gdb.base/watch-bitfields.exp: regular watch against bitfields: continue until exit

We can avoid these errors/FAILs by checking the board data and switching to
software watchpoints if the board does not support hardware watchpoints.

gdb/testsuite/ChangeLog:

2015-04-29  Luis Machado  <lgustavo@codesourcery.com>

* gdb.base/watch-bitfields.exp: Switch to software watchpoints if
the target does not support hardware watchpoints.

9 years agoHandle memory write errors on gdb.base/break-always.exp
Luis Machado [Wed, 29 Apr 2015 15:09:40 +0000 (12:09 -0300)]
Handle memory write errors on gdb.base/break-always.exp

This is another case of the testcase not handling memory write errors that
happen on some targets (QEMU) when GDB attempts to modify an address that
should contain a breakpoint, for example.

The following patch handles this and prevents spurious failures from
happening. It also adds a foreach loop to avoid duplication of code
and hardcoded patterns.

gdb/testsuite/ChangeLog:

2015-04-29  Luis Machado  <lgustavo@codesourcery.com>

* gdb.base/break-always.exp: Abort testing if writing to memory
causes an error.

9 years agoFix problems in the sim sources discovered by running the cppcheck static analysis...
Nick Clifton [Wed, 29 Apr 2015 15:02:02 +0000 (16:02 +0100)]
Fix problems in the sim sources discovered by running the cppcheck static analysis tool.

erc32 PR 18273
* sis.c (main): Remove unreachable code.

m68hc11 * gencode.c (gen_fetch_operands): Remove unreachable code.

ppc * hw_htab.c (htab_map_binary): Fix overlap check.

common * sim-fpu.c (INLINE_SIM_FPU): Fix static analysis warning by
increasing parenthesis around casts to signed values.

9 years agoAllow passing fd == NULL to exec_file_find and solib_find
Gary Benson [Wed, 29 Apr 2015 14:20:22 +0000 (15:20 +0100)]
Allow passing fd == NULL to exec_file_find and solib_find

This commit allows NULL to be passed as the int *fd argument
to exec_file_find and solib_find to simplify use cases where
the caller does not require the file to be opened.

gdb/ChangeLog:

* solib.c (solib_find_1): Allow fd argument to be NULL.
(exec_file_find): Update comment.
(solib_find): Likewise.
* exec.c (exec_file_locate_attach): Use NULL as fd
argument to exec_file_find to avoid having to close
the opened file.
* infrun.c (follow_exec): Likewise.

9 years agoAdd support for absolute PE/x86 relocations.
Nick Clifton [Wed, 29 Apr 2015 11:26:46 +0000 (12:26 +0100)]
Add support for absolute PE/x86 relocations.

PR 17099
* coff-i386.c (coff_i386_rtype_to_howto): Allow absolute PCRLONG
relocs.

9 years agoUpdates the description of GAS's .set directive, to note that for some targets a...
Nick Clifton [Wed, 29 Apr 2015 10:10:45 +0000 (11:10 +0100)]
Updates the description of GAS's .set directive, to note that for some targets a symbolic value can only be set once.

* doc/as.texinfo (Set): Note that a symbol cannot be set multiple
times if the expression is not constant and the target uses linker
relaxation.

9 years ago[ARM] Update ld testcases
Jiong Wang [Wed, 29 Apr 2015 10:04:17 +0000 (11:04 +0100)]
[ARM] Update ld testcases

2015-04-29  Renlin Li  <renlin.li@arm.com>
ld/testsuite/
  * ld-arm/ifunc-10.dd: Adjust expected output.
  * ld-arm/ifunc-2.dd: Likewise.

9 years agoPR python/18299
Doug Evans [Wed, 29 Apr 2015 05:14:23 +0000 (22:14 -0700)]
PR python/18299

gdb/ChangeLog:

PR python/18299
* python/lib/gdb/printing.py (register_pretty_printer): Handle
name or __name__ attributes.  Handle gdb module as first argument.

gdb/testsuite/ChangeLog:

* gdb.python/py-pp-maint.py: Move "replace" testing to ...
* gdb.python/py-pp-registration.exp: ... here.  New file.
* gdb.python/py-pp-registration.c: New file.
* gdb.python/py-pp-registration.py: New file.

9 years agoPR python/18089
Doug Evans [Wed, 29 Apr 2015 04:53:54 +0000 (21:53 -0700)]
PR python/18089

gdb/ChangeLog:

PR python/18089
* python/py-prettyprint.c (print_children): Verify result of children
iterator.  Provide better error message.
* python/python-internal..h (gdbpy_print_python_errors_p): Declare.
* python/python.c (gdbpy_print_python_errors_p): New function.

gdb/testsuite/ChangeLog:

* gdb.python/py-bad-printers.c: New file.
* gdb.python/py-bad-printers.py: New file.
* gdb.python/py-bad-printers.exp: New file.

9 years ago* gdbtypes.h (struct cplus_struct_type) <n_baseclasses>: Fix comment.
Doug Evans [Wed, 29 Apr 2015 04:23:24 +0000 (21:23 -0700)]
* gdbtypes.h (struct cplus_struct_type) <n_baseclasses>: Fix comment.

gdb/ChangeLog:

* gdbtypes.h (struct cplus_struct_type) <n_baseclasses>: Fix comment.

9 years agoAdd gdb.Type.optimized_out method.
Sasha Smundak [Wed, 29 Apr 2015 00:41:09 +0000 (17:41 -0700)]
Add gdb.Type.optimized_out method.

gdb/ChangeLog:

* NEWS: Mention gdb.Type.optimized_out method.
* python/py-type.c (typy_optimized_out):  New function.

gdb/doc/ChangeLog:

* python.texi: New method documented.

gdb/testsuite/ChangeLog:

* gdb.python/py-type.exp: New test.

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

9 years agoEat newlines inside INPUT statements in linker scripts
Andreas Schwab [Tue, 28 Apr 2015 18:32:06 +0000 (20:32 +0200)]
Eat newlines inside INPUT statements in linker scripts

ld/
PR ld/18344
* ldlex.l (INPUTLIST): Increment lineno on newline.

9 years agoUse "gdb_wait.h" instead of <sys/wait.h>.
John Baldwin [Tue, 28 Apr 2015 15:30:23 +0000 (11:30 -0400)]
Use "gdb_wait.h" instead of <sys/wait.h>.

gdb/ChangeLog:
* fbsd-nat.c: Include "gdb_wait.h" instead of <sys/wait.h>.

9 years ago[ARM]Positively emit symbols for alignment
Renlin Li [Tue, 28 Apr 2015 16:10:26 +0000 (17:10 +0100)]
[ARM]Positively emit symbols for alignment

2015-04-28  Renlin Li  <renlin.li@arm.com>
  gas/
    * config/tc-arm.c (arm_init_frag): Always emit mapping symbols.

  gas/testsuite/
    * gas/arm/thumb2_vpool_be.d: Adjust the desired output.
    * gas/arm/vldconst_be.d: Ditto.

9 years agoDisable readline's SIGWINCH handler
Patrick Palka [Thu, 23 Apr 2015 23:28:32 +0000 (19:28 -0400)]
Disable readline's SIGWINCH handler

We no longer need it as we handle SIGWINCH ourselves.  Also move the
call to init_page_info() from initialize_utils() to the latter
function's only caller, gdb_init().

gdb/ChangeLog:

* utils.c (init_page_info): Set rl_catch_sigwinch to zero.
(initialize_utils): Move call of init_page_info() to ...
* top.c (gdb_init): ... here.

9 years agoUpdate our idea of our terminal's dimensions even outside of TUI
Patrick Palka [Thu, 23 Apr 2015 22:36:58 +0000 (18:36 -0400)]
Update our idea of our terminal's dimensions even outside of TUI

When in the CLI, GDB's "width" and "height" variables are not kept in sync
when the underlying terminal gets resized.

This patch fixes this issue by making sure sure to update GDB's "width"
and "height" variables in the !tui_active case of our SIGWINCH handler.

gdb/ChangeLog:

* tui/tui-win.c (tui_sigwinch_handler): Remove now-stale comment.
(tui_sigwinch_handler): Still update our idea of
the terminal's width and height even when TUI is not active.

9 years agoIntroduce function for directly updating GDB's screen dimensions
Patrick Palka [Thu, 23 Apr 2015 22:31:38 +0000 (18:31 -0400)]
Introduce function for directly updating GDB's screen dimensions

... to replace the roundabout pattern of

  execute_command ("set width %d");
  execute_command ("set height %d");

for doing the same thing.

gdb/ChangeLog:

* utils.h (set_screen_width_and_height): Declare.
* utils.c (set_screen_width_and_height): Define.
* tui/tui-win.c (tui_update_gdb_sizes): Use it.

9 years agoUse exec_file_find to prepend gdb_sysroot in follow_exec
Gary Benson [Tue, 28 Apr 2015 11:21:32 +0000 (12:21 +0100)]
Use exec_file_find to prepend gdb_sysroot in follow_exec

This commit updates follow_exec to use exec_file_find to prefix
the new executable's filename with gdb_sysroot rather than doing
it longhand.

gdb/ChangeLog:

* infrun.c (solist.h): New include.
(follow_exec): Use exec_file_find to prefix execd_pathname
with gdb_sysroot.

9 years agoFix compile time warnings about a local variable being used before it is set.
Nick Clifton [Tue, 28 Apr 2015 10:22:57 +0000 (11:22 +0100)]
Fix compile time warnings about a local variable being used before it is set.

PR 18313
* cond.c (s_if): Stop compile time warning about stopc being used
before it is set.
(s_ifc): Likewise.

9 years agoFix py-parameter.exp and scm-parameter.exp path matching
Andy Wingo [Tue, 28 Apr 2015 09:15:47 +0000 (11:15 +0200)]
Fix py-parameter.exp and scm-parameter.exp path matching

gdb/testsuite/ChangeLog:

* gdb.python/py-parameter.exp:
* gdb.guile/scm-parameter.exp: Escape the path that we are
matching against, as it might contain characters that are special
to regular expressions.

9 years agoTidy PowerPC gold find_global_entry uses
Alan Modra [Tue, 28 Apr 2015 07:15:34 +0000 (16:45 +0930)]
Tidy PowerPC gold find_global_entry uses

Completely removing the assert probably wasn't the best idea, so
reinstate it for allocated sections.  Also cope with debug info
potentially referring to a missing plt call stub.

And a tidy.  find_global_entry now returns an Address, so make temps
holding the return value of type Address, and compare against
invalid_address.

* powerpc.cc (Target_powerpc::do_dynsym_value): Use Address rather
than unsigned int for find_global_entry result temp.  Compare
against invalid_address.
(Target_powerpc::do_plt_address_for_global): Likewise.
(Target_powerpc::Relocate::relocate): Likewise.  Don't assert
on plt call stub existence for debug info.  Do assert for plt
and global entry stub existence if an alloc section.

9 years agoPowerPC gold assertion on missing global entry stub
Alan Modra [Tue, 28 Apr 2015 03:58:29 +0000 (13:28 +0930)]
PowerPC gold assertion on missing global entry stub

Global entry stubs are used on ELFv2 to provide addresses for
functions not defined in a non-PIC executable but whose address is
taken, in much the same way as PLT stub code is used on other
targets to provide function addresses.  We don't want to insert a
global entry stub just because (bogus) debug info refers to the
address of a non-local function, but we also don't want gold to die.

* powerpc.cc (Target_powerpc::Relocate::relocate): Don't assert
on missing global entry stub due to bogus debug info.

9 years agoTUI: avoid calling strcpy() on identical string objects
Patrick Palka [Sat, 25 Apr 2015 14:29:29 +0000 (10:29 -0400)]
TUI: avoid calling strcpy() on identical string objects

In tui_set_source_content(), when offset == 0 the source and destination
pointers of the call to strcpy() are actually the same.  In this case
not only is strcpy() unnecessary but it is also UB when the two strings
overlap.

gdb/ChangeLog:

* tui/tui-source.c (tui_set_source_content): Avoid calling
strcpy() when offset is 0.

9 years agoFix PR gdb/18155
Patrick Palka [Sun, 26 Apr 2015 01:59:02 +0000 (21:59 -0400)]
Fix PR gdb/18155

For no good reason the function tui_free_window() is freeing the locator
window when we pass it an SRC_WIN or a DISASSEM_WIN.  This behavior
doesn't make much sense because the locator window is always visible and
its contents do not change when the main window changes.

This behavior triggers the above PR because when we switch from one TUI
window to another (in the PR, from the src window to the asm window) we
call tui_free_window() on the previously active window (in the PR, the
src window).  The function then frees the src window along with the
locator window and later we segfault when the now-active asm window
tries to query the locator window about the inferior's PC.

This patch fixes this apparently wrong behavior by changing
tui_free_window() to not free the locator window when we pass it an
SRC_WIN or a DISASSEM_WIN.

gdb/ChangeLog:

PR gdb/18155
* tui/tui-data.c (tui_free_window): Don't free the locator
window when passed an SRC_WIN or a DISASSEM_WIN.

9 years agoMake type-safe the 'content' field of struct tui_gen_win_info
Patrick Palka [Fri, 24 Apr 2015 12:26:50 +0000 (08:26 -0400)]
Make type-safe the 'content' field of struct tui_gen_win_info

The 'content' field of struct tui_gen_win_info currently has type
void ** but the field always stores an object of type tui_win_content.
Instead of unnecessarily casting to and from void ** we should just give
the field the type tui_win_content in the first place.

This patch does this and also eliminates all now-redundant casts
involving the 'content' struct field that I could find.

gdb/ChangeLog:

* tui/tui-data.h (struct tui_win_element): Forward-declare.
(tui_win_content): Move declaration.
(struct tui_gen_win_info): Give 'content' field the
type tui_win_content.
* tui/tui-data.c (init_content_element): Remove redundant and
erroneous casts.
(tui_add_content_elements): Remove erroneous cast.
* tui/tui-disasm.c (tui_set_disassem_content): Remove redundant
casts.
(tui_get_begin_asm_address): Likewise.
* tui/tui-regs.c (tui_show_registers): Likewise.
(tui_show_register_group): Likewise.
(tui_display_registers_from): Likewise.
(tui_check_register_values): Likewise.
* tui/tui-source.c (tui_set_source_content): Likewise.
(tui_set_source_content_nil): Likewise.
(tui_source_is_displayed): Likewise.
* tui/tui-stack.c (tui_show_locator_content): Likewise.
(tui_set_locator_fullname): Likewise.
(tui_set_locator_info): Likewise.
(tui_show_frame_info): Likewise.
* tui/tui-winsource.c (tui_clear_source_content): Likewise.
(tui_show_source_line): Likewise.
(tui_horizontal_source_scroll): Likewise.
(tui_update_breakpoint_info): Likewise.
(tui_set_exec_info_content): Likewise.
(tui_show_exec_info_content): Likewise.
(tui_alloc_source_buffer): Likewise.
(tui_line_is_displayed): Likewise.
(tui_addr_is_displayed): Likewise.

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

9 years agoAdd support for catching exec events on FreeBSD.
John Baldwin [Sat, 18 Apr 2015 05:00:06 +0000 (01:00 -0400)]
Add support for catching exec events on FreeBSD.

FreeBSD kernels that support fork tracing always stop a process to
report events for exec.  Such a process will have the PL_FLAG_EXEC
flag set in the pl_flags field of the ptrace_lwpinfo struct returned
by PT_LWPINFO.  The structure does not include the pathname passed to
exec, so use fbsd_pid_to_exec_file to query the pathname of the
process' executable.

gdb/ChangeLog:

* fbsd-nat.c: (fbsd_wait) [PL_FLAG_EXEC]: Report TARGET_WAITKIND_EXECD
event if PL_FLAG_EXEC is set.
[PL_FLAG_EXEC] (fbsd_insert_exec_catchpoint): New function.
[PL_FLAG_EXEC] (fbsd_remove_exec_catchpoint): New function.
(fbsd_nat_add_target) [PL_FLAG_EXEC]: Set
"to_insert_exec_catchpoint" to "fbsd_insert_exec_catchpoint".
Set "to_remove_exec_catchpoint" to "fbsd_remove_exec_catchpoint".

9 years agoEnable fork tracing for native FreeBSD targets.
John Baldwin [Fri, 17 Apr 2015 20:20:47 +0000 (16:20 -0400)]
Enable fork tracing for native FreeBSD targets.

Enable PT_FOLLOW_FORK on all processes.  When this is enabled, both
the parent and child process stop when a fork or vfork occurs.

A target operation for wait uses PT_LWPINFO to fetch more detailed
information about the state of a stopped process.  A parent process
sets the PL_FLAG_FORKED flag in the pl_flags field of the structure
returned by PT_LWPINFO as well as the pid of the new child process.
The child process sets the PL_FLAG_CHILD flag in the pl_flags field.

When a fork is detected, the wait hook waits for both processes to
report their respective events.  It then reports the fork to GDB as
a single TARGET_WAITKIND_FORKED or TARGET_WAITKIND_VFORKED event.
The kernel does not guarantee the order the events are reported in.
If the parent process' event is reported first, then the wait hook
explicitly waits for the child process.  If the child process' event
is reported first, the event is recorded on an internal list of
pending child events and the wait hook waits for another event.
Later when the parent process' event is reported, the parent will
use the previously-recorded child process event instead of explicitly
waiting on the child process.

To distinguish vfork events from fork events, the external process
structure for the child process is extracted from the kernel.  The
P_PPWAIT flag is set in the ki_flags field of this structure if the
process was created via vfork, but it is not set for a regular fork.

gdb/ChangeLog:

* fbsd-nat.c: [PT_LWPINFO] New variable super_wait.
[TDP_RFPPWAIT] New variable fbsd_pending_children.
[TDP_RFPPWAIT] (fbsd_remember_child): New function.
[TDP_RFPPWAIT] (fbsd_is_child_pending): New function.
[TDP_RFPPWAIT] (fbsd_fetch_kinfo_proc): New function.
[PT_LWPINFO] (fbsd_wait): New function.
[TDP_RFPPWAIT] (fbsd_follow_fork): New function.
[TDP_RFPPWAIT] (fbsd_insert_fork_catchpoint): New function.
[TDP_RFPPWAIT] (fbsd_remove_fork_catchpoint): New function.
[TDP_RFPPWAIT] (fbsd_insert_vfork_catchpoint): New function.
[TDP_RFPPWAIT] (fbsd_remove_vfork_catchpoint): New function.
[TDP_RFPPWAIT] (fbsd_enable_follow_fork): New function.
[TDP_RFPPWAIT] (fbsd_post_startup_inferior): New function.
[TDP_RFPPWAIT] (fbsd_post_attach): New function.
(fbsd_nat_add_target) [PT_LWPINFO] Set "to_wait" to
"fbsd_wait".
[TDP_RFPPWAIT] Set "to_follow_fork" to "fbsd_follow_fork".
Set "to_insert_fork_catchpoint" to "fbsd_insert_fork_catchpoint".
Set "to_remove_fork_catchpoint" to "fbsd_remove_fork_catchpoint".
Set "to_insert_vfork_catchpoint" to "fbsd_insert_vfork_catchpoint".
Set "to_remove_vfork_catchpoint" to "fbsd_remove_vfork_catchpoint".
Set "to_post_startup_inferior" to "fbsd_post_startup_inferior".
Set "to_post_attach" to "fbsd_post_attach".

9 years agoAdd fbsd_nat_add_target.
John Baldwin [Fri, 17 Apr 2015 18:02:03 +0000 (14:02 -0400)]
Add fbsd_nat_add_target.

Add a wrapper for add_target in fbsd-nat.c to override target operations
common to all native FreeBSD targets.

gdb/ChangeLog:

* fbsd-nat.c (fbsd_pid_to_exec_file): Mark static.
(fbsd_find_memory_regions): Mark static.
(fbsd_nat_add_target): New function.
* fbsd-nat.h: Export fbsd_nat_add_target and remove prototypes for
fbsd_pid_to_exec_file and fbsd_find_memory_regions.
* amd64fbsd-nat.c (_initialize_amd64fbsd_nat): Use fbsd_nat_add_target.
* i386fbsd-nat.c (_initialize_i386fbsd_nat): Likewise.
* ppcfbsd-nat.c (_initialize_ppcfbsd_nat): Likewise.
* sparc64fbsd-nat.c (_initialize_sparc64fbsd_nat): Likewise.

9 years ago[gold] Rename '--fix-cortex-a53' to '--fix-cortex-a53-843419'.
Han Shen [Mon, 27 Apr 2015 22:01:44 +0000 (15:01 -0700)]
[gold] Rename '--fix-cortex-a53' to '--fix-cortex-a53-843419'.

Keep gold consistent with bfd erratum-fixing option names, so as to
ease life in Makefile/scripts.

gold/

* options.h (--fix-cortex-a53-843419): Rename option.
* aarch64.cc (AArch64_relobj::do_count_local_symbols): Use renamed
option.
(AArch64_relobj::scan_sections_for_stubs): Use renamed option.

9 years agoIf a range is missing, assume the input address is mapped.
Rafael Ávila de Espíndola [Mon, 27 Apr 2015 18:33:02 +0000 (14:33 -0400)]
If a range is missing, assume the input address is mapped.

When Output_section::is_input_address_mapped is called we have entries for
all dropped ranges, but not for all ranges.

9 years agoopcodes/
Peter Bergner [Mon, 27 Apr 2015 16:06:54 +0000 (11:06 -0500)]
opcodes/

* ppc-opc.c (DCBT_EO): New define.
(powerpc_opcodes) <lbarx>: Enable for POWER8 and later.
<lharx>: Likewise.
<stbcx.>: Likewise.
<sthcx.>: Likewise.
<waitrsv>: Do not enable for POWER7 and later.
<waitimpl>: Likewise.
<dcbt>: Default to the two operand form of the instruction for all
"old" cpus.  For "new" cpus, use the operand ordering that matches
whether the cpu is server or embedded.
<dcbtst>: Likewise.

gas/testsuite/

* gas/ppc/a2.s: Fixup test case due to dcbt/dcbtst embedded operand
ordering change.
* gas/ppc/a2.d: Likewise.
* gas/ppc/476.d: Likewise.
* gas/ppc/booke.s: Remove invalid 3 operand dcbt tests.
* gas/ppc/booke.d: Likewise.
* gas/ppc/power7.s: Remove lbarx, lharx, stbcx., sthcx., waitrsv
and waitimpl tests.
* gas/ppc/power7.d: Likewise.

9 years agoDo not manipulate "target:" filenames as local paths
Gary Benson [Mon, 27 Apr 2015 10:46:13 +0000 (11:46 +0100)]
Do not manipulate "target:" filenames as local paths

This commit alters two places that manipulate object file filenames
to detect "target:" filenames and to not attempt to manipulate them
as paths on the local filesystem:

 - allocate_objfile is updated to not attempt to expand "target:"
   filenames with gdb_abspath.

 - load_auto_scripts_for_objfile is updated to not attempt to load
   auto-load scripts for object files with "target:" filenames.

gdb/ChangeLog:

* objfiles.c (allocate_objfile): Do not attempt to expand name
if name is a "target:" filename.
* auto-load.c (load_auto_scripts_for_objfile): Do not attempt
to load auto-load scripts for objfiles with "target:" filenames.

9 years agosim: avr: Fix 'multiple definition of sim_{read,write}'
Senthil Kumar Selvaraj [Mon, 27 Apr 2015 08:56:04 +0000 (14:26 +0530)]
sim: avr: Fix 'multiple definition of sim_{read,write}'

This patch does whatever was done in
https://sourceware.org/ml/gdb-patches/2015-04/msg00437.html to fix
broken gdb build for the AVR target.

9 years ago[AArch64] Don't try to align insn in non-executale section
Renlin Li [Mon, 27 Apr 2015 10:36:12 +0000 (11:36 +0100)]
[AArch64] Don't try to align insn in non-executale section

2015-04-27  Renlin Li  <renlin.li@arm.com>

  gas/
    * config/tc-aarch64.c (s_aarch64_inst): Don't align code for non-text
    section.
    (md_assemble): Likewise, move the align code outside the loop.

9 years agoS390: Vector ABI support
Andreas Arnez [Mon, 27 Apr 2015 09:38:47 +0000 (11:38 +0200)]
S390: Vector ABI support

With the S390 vector ABI, vector registers are used for passing vector
arguments and for returning a vector.  Support this ABI in inferior
function calls and when setting or retrieving a function's return
value.

gdb/ChangeLog:

* s390-linux-tdep.c: Include "elf/s390.h" and "elf-bfd.h".
(enum s390_vector_abi_kind): New enum.
(struct gdbarch_tdep)<vector_abi>: New field.
(s390_effective_inner_type): Add parameter min_size.  Stop
unwrapping if the inner type is smaller than min_size.
(s390_function_arg_float): Adjust call to
s390_effective_inner_type.
(s390_function_arg_vector): New function.
(s390_function_arg_integer): Adjust comment.
(struct s390_arg_state)<vr>: New field.
(s390_handle_arg): Add parameter 'is_unnamed'.  Pass vector
arguments according to vector ABI when appropriate.
(s390_push_dummy_call): Initialize the argument state's field
'vr'.  Adjust calls to s390_handle_arg.
(s390_register_return_value): Handle vector return values.
(s390_return_value): Apply the "register" return value convention
to a vector when appropriate.
(s390_gdbarch_init): Initialize tdep->vector_abi.
* NEWS: Announce S390 vector ABI support.

9 years agoS390: Re-arrange implementation of s390_return_value
Andreas Arnez [Mon, 27 Apr 2015 09:38:47 +0000 (11:38 +0200)]
S390: Re-arrange implementation of s390_return_value

Move related logic in the implementation of s390_return_value closer
together.  This makes it easier to read and extend.

gdb/ChangeLog:

* s390-linux-tdep.c (s390_return_value_convention): Remove
function.  Inline its logic...
(s390_return_value): ...here.  Instead, move the handling of the
"register" return value convention...
(s390_register_return_value): ...here.  New function.

9 years agoS390: Restructure s390_push_dummy_call
Andreas Arnez [Mon, 27 Apr 2015 09:38:46 +0000 (11:38 +0200)]
S390: Restructure s390_push_dummy_call

Simplify the structure of s390_push_dummy_call and its various helper
functions.  This reduces the code and makes it easier to extend.  The
new code should be functionally equivalent to the old one, except that
copies created by the caller are now always aligned on an 8-byte
boundary.

gdb/ChangeLog:

* s390-linux-tdep.c
(is_float_singleton): Remove function.  Move the "singleton" part
of the logic...
(s390_effective_inner_type): ...here.  New function.
(is_float_like): Remove function.  Inline its logic...
(s390_function_arg_float): ...here.
(is_pointer_like, is_integer_like, is_struct_like): Remove
functions.  Inline their logic...
(s390_function_arg_integer): ...here.
(s390_function_arg_pass_by_reference): Remove function.
(extend_simple_arg): Remove function.
(alignment_of): Remove function.
(struct s390_arg_state): New structure.
(s390_handle_arg): New function.
(s390_push_dummy_call): Move parameter placement logic to the new
function s390_handle_arg.  Call it for calculating the stack area
sizes first, and again for actually writing the parameters.

9 years agoS390: For zero, let is_power_of_two() return false
Andreas Arnez [Mon, 27 Apr 2015 09:38:46 +0000 (11:38 +0200)]
S390: For zero, let is_power_of_two() return false

This fixes a minor issue with the helper function is_power_of_two(),
which returned non-zero ("true") if the argument was zero.  This led
to a wrong decision when passing a zero-sized struct in an inferior
function call.

gdb/ChangeLog:

* s390-linux-tdep.c (is_power_of_two): Add comment.  Return
  false if the argument is zero.

9 years ago[Ada] Cache all static structures and reset cache during resolution
Pierre-Marie de Rodat [Mon, 27 Apr 2015 09:06:07 +0000 (11:06 +0200)]
[Ada] Cache all static structures and reset cache during resolution

Currently, ada-lang.c:template_to_static_fixed_type (working on
structure types only) caches its result into the unused TYPE_TARGET_TYPE
field. This introduces inconsistencies when the input type is
specialized, for instance during type resolution: the cached static
fixed type is copied along with the original type, but it's no longer
adapted to the copy once the copy is modified:
template_to_static_fixed_type has to compute another static fixed type
for it.

This change first introduces a cache reset during type resolution for
structure types so that this inconsistency does not happen anymore. It
also makes template_to_static_fixed_type smarter with respect to types
that do not need static fixed copies so that less computations is done
in general.

This inconsistency was spotted thanks to code reading, not because of
any sort of failure and we did not manage to exhibit a failure yet, so
no testcase for this.

gdb/ChangeLog:

* ada-lang.c (template_to_static_fixed_type): Return input type
when it is already fixed. Cache the input type itself when not
creating a static fixed copy. Make it explicit that we never
molestate the input type.
* gdbtypes.c (resolve_dynamic_struct): Reset the
TYPE_TARGET_TYPE field for resolved copies.

9 years ago[Ada] Preserve typedef layer when getting struct element
Joel Brobecker [Mon, 27 Apr 2015 09:04:47 +0000 (11:04 +0200)]
[Ada] Preserve typedef layer when getting struct element

Consider the following declarations:

   type Int_Access is access Integer;
   type Record_Type is record
      IA : Int_Access;
   end record;

   R : Record_Type;

Printing the type name of "R.IA" yields:

    (gdb) whatis r.ia
    type = access integer

It should be:

    (gdb) whatis r.ia
    type = bar.int_access

Looking at the debugging info, field "r.ia" is defined as
a typedef which has the name of the field type:

        .uleb128 0x3    # (DIE (0x4e) DW_TAG_typedef)
        .long   .LASF4  # DW_AT_name: "bar__int_access"
        .long   0x8b    # DW_AT_type

... with the typedef's target type being an anonymous pointer
type:

        .uleb128 0x7    # (DIE (0x8b) DW_TAG_pointer_type)
        .byte   0x8     # DW_AT_byte_size
        .long   0x91    # DW_AT_type

What happens here is that a couple of function in ada-lang.c
always start by stripping all typedef layers when handling
struct fields, with the effect of making us lose the type name
in this case.

We did not understand this at the time the code was written,
but typedefs should be stripped only when we know we do not
need them. So this patch, adjust the code to avoid the stripping
while handling the fields, and adds it back in the lone place
which handles the result of processing and didn't know how to
handle typedefs struct fields yet.

gdb/ChangeLog:

        * ada-lang.c (ada_is_tagged_type): Add call to ada_check_typedef.
        (ada_lookup_struct_elt_type): Remove calls to ada_check_typedef.
        (template_to_static_fixed_type): Call ada_check_typedef only
        when necessary.

gdb/testsuite/ChangeLog:

        * gdb.ada/rec_comp: New testcase.

9 years agoS/390: z13 use GNU attribute to indicate vector ABI
Andreas Krebbel [Mon, 27 Apr 2015 08:32:23 +0000 (10:32 +0200)]
S/390: z13 use GNU attribute to indicate vector ABI

bfd/
* elf-s390-common.c (elf_s390_merge_obj_attributes): New function.
* elf32-s390.c (elf32_s390_merge_private_bfd_data): Call
elf_s390_merge_obj_attributes.
* elf64-s390.c (elf64_s390_merge_private_bfd_data): New function.

binutils/
* readelf.c (display_s390_gnu_attribute): New function.
(process_s390_specific): New function.
(process_arch_specific): Call process_s390_specific.

gas/
* doc/as.texinfo: Document Tag_GNU_S390_ABI_Vector.

include/elf/
* s390.h: Define Tag_GNU_S390_ABI_Vector.

9 years agoS/390: Fixes for z13 instructions.
Andreas Krebbel [Mon, 27 Apr 2015 08:29:16 +0000 (10:29 +0200)]
S/390: Fixes for z13 instructions.

opcodes/
* s390-opc.c: New instruction type VV0UU2.
* s390-opc.txt: Fix instruction types for VFCE, VLDE, VFSQ, WFK,
and WFC.

gas/testsuite/
* gas/s390/zarch-z13.d: Fix tests for VFCE, VLDE, VFSQ, WFK, and
WFC.
   * gas/s390/zarch-z13.s: Likewise.

9 years agoS/390: Fix gotreloc_31-1 testcase.
Andreas Krebbel [Mon, 27 Apr 2015 08:24:24 +0000 (10:24 +0200)]
S/390: Fix gotreloc_31-1 testcase.

Since we changed the default arch for objdump to zarch the following
testcase needs to check for the real instruction mnemonics instead of
just bytes.

This fixes the following testsuite fail on s390x:

FAIL: GOT: symbol address load from got to larl

9 years agoAutomatic date update in version.in
GDB Administrator [Mon, 27 Apr 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoClear variable "coredump_var_addr" before using it on gdb.base/coredump-filter.exp
Sergio Durigan Junior [Sun, 26 Apr 2015 19:34:29 +0000 (15:34 -0400)]
Clear variable "coredump_var_addr" before using it on gdb.base/coredump-filter.exp

This commit is a continuation of the fix committed on:

  commit 8cd8f2f8ac49276437b7da37f275706ea1c1c925
  Author: Sergio Durigan Junior <sergiodj@redhat.com>
  Date:   Mon Apr 13 02:40:08 2015 -0400

      Rename variable "addr" to "coredump_var_addr" in gdb.base/coredump-filter.exp

Pedro pointed out that this fix was not complete, because the
testsuite could be run several times in a row (for example), which
means that it is not enough to just make the variable name unique: it
also needs to be cleared out if it is global.

This commit does that.  It is actually just a commit made to make
things totally correct; this specific test does not fail if you run it
several times in a row.

gdb/testsuite/ChangeLog:
2015-04-26  Sergio Durigan Junior  <sergiodj@redhat.com>

* gdb.base/coredump-filter.exp: Clear variable "coredump_var_addr"
before using it.

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

9 years agoAutomatic date update in version.in
GDB Administrator [Sat, 25 Apr 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoNon-alloc sections don't belong in PT_LOAD segments
Alan Modra [Fri, 24 Apr 2015 09:49:37 +0000 (19:19 +0930)]
Non-alloc sections don't belong in PT_LOAD segments

Taking them out showed a bug in the powerpc64 backend with .branch_lt
being removed from output_bfd but not from previously set up segment
section maps.  Removing the bfd sections meant their sh_flags (and
practically everything else) remaining zero, ie. not SHF_ALLOC,
triggering complaints about "`.branch_lt' can't be allocated in
segment".

include/elf/
* internal.h (ELF_SECTION_IN_SEGMENT_1): Ensure PT_LOAD and
similar segments only contain alloc sections.
ld/
* emultempl/ppc64elf.em (gld${EMULATION_NAME}_after_allocation):
Call gld${EMULATION_NAME}_map_segments regardless of need_laying_out.
ld/testsuite/
* ld-powerpc/tocnovar.d: Revert last change.

9 years ago[AArch64] PR18270, fix handling of GOT entry for local symbol
Jiong Wang [Fri, 24 Apr 2015 22:25:28 +0000 (23:25 +0100)]
[AArch64] PR18270, fix handling of GOT entry for local symbol

2015-04-24  Jiong. Wang  <jiong.wang@arm.com>

bfd/
  PR ld/18270
  * elfnn-aarch64.c (elfNN_aarch64_size_dynamic): Count local symbol for
  GOT_NORMAL for both sgot/srelgot section.
  (elfNN_aarch64_final_link_relocate): Relocate against GOT entry address
  and generate necessary runtime relocation for GOT entry.

9 years agoCopy is_linker_input to archive member
H.J. Lu [Fri, 24 Apr 2015 22:02:56 +0000 (15:02 -0700)]
Copy is_linker_input to archive member

We must copy is_linker_input to archive member.

PR binutils/18209
* archive.c (_bfd_get_elt_at_filepos): Also copy is_linker_input.

9 years agogdb: Add internationalization support to a few strings.
Andrew Burgess [Thu, 23 Apr 2015 21:18:55 +0000 (22:18 +0100)]
gdb: Add internationalization support to a few strings.

Spotted a few strings that were missing internationalization support.

gdb/ChangeLog:

* cli/cli-dump.c (srec_dump_command): Add internationalization
mark ups.
(ihex_dump_command): Likewise.
(tekhex_dump_command): Likewise.
(binary_dump_command): Likewise.
(binary_append_command): Likewise.

9 years agogdb: Add support for dumping to verilog hex format.
Andrew Burgess [Wed, 22 Apr 2015 21:52:36 +0000 (22:52 +0100)]
gdb: Add support for dumping to verilog hex format.

Extend the gdb 'dump' command to allow creating output in verilog hex
format.  Add some tests to cover new functionality.  As bfd does not
currently support reading in verilog hex formats the tests only cover
the 'dump' command, not the 'restore' command.

gdb/ChangeLog:

* cli/cli-dump.c (verilog_cmdlist): New variable.
(dump_verilog_memory): New function.
(dump_verilog_value): New function.
(verilog_dump_command): New function.
(_initialize_cli_dump): Add new commands to support verilog dump
format.
* NEWS: Add entry for "dump verilog".

gdb/doc/ChangeLog:

* gdb.texinfo (Dump/Restore Files): Add detail about verilog dump
format.

gdb/testsuite/ChangeLog:

* gdb.base/dump.exp: Add *.verilog files to all_files list.  Add
new tests for verilog output.

9 years ago[AArch64] Improve PC-relative relocation check for shared library
Jiong Wang [Fri, 24 Apr 2015 21:35:04 +0000 (22:35 +0100)]
[AArch64] Improve PC-relative relocation check for shared library

2015-04-24  Jiong. Wang  <jiong.wang@arm.com>

bfd/
  * elfnn-aarch64.c (elfNN_aarch64_final_link_relocate): Reject
  PC-relative relocation for external symbol.

ld/testsuite/
  * ld-aarch64/pcrel.s: New testcase.
  * ld-aarch64/pcrel_pic_defiend_local.d: New expect file.
  * ld-aarch64/pcrel_pic_undefined.d: Ditto.
  * ld-aarch64/aarch64-elf.exp: Run them.

9 years agoFix typo
DJ Delorie [Fri, 24 Apr 2015 21:17:02 +0000 (17:17 -0400)]
Fix typo

9 years agoChange msp430 emulation to msp430elf
DJ Delorie [Fri, 24 Apr 2015 21:05:13 +0000 (17:05 -0400)]
Change msp430 emulation to msp430elf

* Makefile.am (msp430): Rename primary emulation to msp430elf.
(emsp430.c): Rename to emsp430elf.c, update dependencies
(emsp430X.c): Update dependencies.
* Makefile.in: Likewise.
* configure.tgt (msp430-*-*): Rename primary emulation to msp430elf.
* emulparame/msp430.sh: Rename to msp430elf.sh.
* emulparams/msp430X.sh: Update.

9 years agogas thunderx support
Jim Wilson [Fri, 24 Apr 2015 20:38:25 +0000 (13:38 -0700)]
gas thunderx support

gas/
* config/tc-aarch64.c (aarch64_cpus): Add CRC and CRYPTO features
for thunderx.

9 years ago* python.texi (Xmethods In Python): Fix name of method to call the xmethod.
Doug Evans [Fri, 24 Apr 2015 18:16:12 +0000 (11:16 -0700)]
* python.texi (Xmethods In Python): Fix name of method to call the xmethod.

gdb/doc/ChangeLog:

* python.texi (Xmethods In Python): Fix name of method to call the
xmethod.

9 years agoFix compile time warning messages about variables being used before they are initialised.
Nick Clifton [Fri, 24 Apr 2015 16:13:22 +0000 (17:13 +0100)]
Fix compile time warning messages about variables being used before they are initialised.

PR 18313
bin * ieee.c (ieee_read_cxx_class): Initialise the varargs variable.
* readelf.c (uncompress_section_contents): Zero initialise the
zstream structure.

bfd * compress.c (decompress_contents): Zero initialse the z_stream
structure.

9 years ago[ARM]: Don't tail-pad over-aligned functions to the alignment boundary.
Richard Earnshaw [Fri, 24 Apr 2015 14:51:49 +0000 (15:51 +0100)]
[ARM]: Don't tail-pad over-aligned functions to the alignment boundary.

2015-04/24  Richard Earnshaw  <rearnsha@arm.com>

gas/
* config/tc-arm.h (arm_min): New function.
(SUB_SEGMENT_ALIGN): Define.

gas/testsuite/
* gas/arm/align64.d: Delete trailing padding NOPs.

ld/testsuite/
* ld-arm/armthumb-lib.d: Regenerate expected output.
* ld-arm/armthumb-lib.d: Likewise.
* ld-arm/armthumb-lib.sym: Likewise.
* ld-arm/cortex-a8-fix-b-rel-arm.d: Likewise.
* ld-arm/cortex-a8-fix-b-rel-thumb.d: Likewise.
* ld-arm/cortex-a8-fix-b.d: Likewise.
* ld-arm/cortex-a8-fix-bcc-rel-thumb.d: Likewise.
* ld-arm/cortex-a8-fix-bcc.d: Likewise.
* ld-arm/cortex-a8-fix-bl-rel-arm.d: Likewise.
* ld-arm/cortex-a8-fix-bl-rel-plt.d: Likewise.
* ld-arm/cortex-a8-fix-bl-rel-thumb.d: Likewise.
* ld-arm/cortex-a8-fix-bl.d: Likewise.
* ld-arm/cortex-a8-fix-blx-bcond.d: Likewise.
* ld-arm/cortex-a8-fix-blx-rel-arm.d: Likewise.
* ld-arm/cortex-a8-fix-blx-rel-thumb.d: Likewise.
* ld-arm/cortex-a8-fix-blx.d: Likewise.
* ld-arm/cortex-a8-fix-hdr.d: Likewise.
* ld-arm/farcall-mixed-app-v5.d: Likewise.
* ld-arm/farcall-mixed-app.d: Likewise.
* ld-arm/farcall-mixed-lib-v4t.d: Likewise.
* ld-arm/farcall-mixed-lib.d: Likewise.
* ld-arm/mixed-app-v5.d: Likewise.
* ld-arm/mixed-app.d: Likewise.
* ld-arm/mixed-lib.d: Likewise.

9 years agoFix typos in sim sources exposed by static analysis.
Nick Clifton [Fri, 24 Apr 2015 14:43:21 +0000 (15:43 +0100)]
Fix typos in sim sources exposed by static analysis.

bfin PR 18273
* bfin-sim.c (decode_dsp32alu_0): Remove spurious check for
s == 1.

erc32 PR 18273
* exec.c (add32): Fix typo in check for overflow.

igen PR 18273
* misc.c (a2i): Fix typos checking for uppercase letters.

9 years agoFix typo in check for valid register number in RX sim.
Nick Clifton [Fri, 24 Apr 2015 14:31:36 +0000 (15:31 +0100)]
Fix typo in check for valid register number in RX sim.

PR sim/18273
* reg.c (put_reg): Fix check for valid register number.

9 years agoExtend test for local labels to include fake symbols and local labels with a numeric...
Nick Clifton [Fri, 24 Apr 2015 14:17:13 +0000 (15:17 +0100)]
Extend test for local labels to include fake symbols and local labels with a numeric suffix.

* elf.c (_bfd_elf_is_local_label_name): Extend test for assembler
local labels to include local labels with a numeric suffix and
fake symbols.

9 years agoFix printing for GNAT stuff for types that do not have descr. types
Pierre-Marie de Rodat [Fri, 24 Apr 2015 14:13:19 +0000 (16:13 +0200)]
Fix printing for GNAT stuff for types that do not have descr. types

gdb/ChangeLog:
2015-04-24  Pierre-Marie de Rodat  <derodat@adacore.com>

* gdbtypes.c (print_gnat_stuff): Do not recurse on the
descriptive type when there is none.

9 years agoDefine SEC_MEP_VLIW in bfd/section.c
H.J. Lu [Fri, 24 Apr 2015 12:40:23 +0000 (05:40 -0700)]
Define SEC_MEP_VLIW in bfd/section.c

Bits in section flags should only be defined in bfd/section.c.  This
patch moves SEC_MEP_VLIW to bfd/section.c.

bfd/

PR binutils/18316
* section.c (SEC_MEP_VLIW): New.
* bfd-in2.h: Regenerated.

include/elf/

PR binutils/18316
* mep.h (SEC_MEP_VLIW): Removed.

9 years agoA new board file remote-gdbserver-on-localhost.exp
Yao Qi [Fri, 24 Apr 2015 10:00:14 +0000 (11:00 +0100)]
A new board file remote-gdbserver-on-localhost.exp

This patch is to add a new board file that does real remote gdbserver
testing on localhost.  This board file can be used to reproduce PR 18208.

gdb/testsuite

2015-04-24  Yao Qi  <yao.qi@linaro.org>

* boards/remote-gdbserver-on-localhost.exp: New file.

9 years agoSkip the compressed1d test for targets which do not use the elf32.em linker emulation...
Nick Clifton [Fri, 24 Apr 2015 09:15:41 +0000 (10:15 +0100)]
Skip the compressed1d test for targets which do not use the elf32.em linker emulation file.

* ld-elf/compressed1d.d: Add notarget for ELF based targets that
do not use elf.em.