binutils-gdb.git
6 years agoReplace objfile field in dwarf2_cu and dwarf2_per_cu_data with dwarf2_per_objfile
Simon Marchi [Sun, 7 Jan 2018 16:22:45 +0000 (11:22 -0500)]
Replace objfile field in dwarf2_cu and dwarf2_per_cu_data with dwarf2_per_objfile

The next patch aims to remove the dwarf2_per_objfile global.  In many
functions, we need to find a way to get a reference to the current
dwarf2_per_objfile through the objects passed in parameters.  Often, we have
access to a dwarf2_cu or a dwarf2_per_cu_data.  These objects have a reference
to the objfile, through which we can get the dwarf2_per_objfile:

  dwarf2_per_objfile = ((struct dwarf2_per_objfile *)
        objfile_data (objfile, dwarf2_objfile_data_key));

However, this is a bit cumbersome to do all over the place.  It would be
more logical if the dwarf2_cu and dwarf2_per_cu_data had a reference to
their dwarf2_per_objfile, which would then have a reference to the
objfile.  It would be more in line with the object hierarchy, where
dwarf2_per_objfile owns the dwarf2_per_cu_data instances.  We could
even remove the reference dwarf2_cu has to dwarf2_per_objfile, since we
can access dwarf2_per_objfile through dwarf2_cu::per_cu.

In a graphical way, references would look like this after the current
patch:

  objfile <--- dwarf2_per_objfile <--- dwarf2_per_cu_data
                                    |      ^
    |      |
                                    `- dwarf2_cu

This patch has been tested on the buildbot.

gdb/ChangeLog:

* dwarf2read.c (struct dwarf2_cu) <objfile>: Remove.
<dwarf2_per_objfile>: New field.
(struct dwarf2_per_cu_data) <objfile>: Remove.
<dwarf2_per_objfile>: New field.
(create_cu_from_index_list): Assign dwarf2_per_objfile instead
of objfile.
(create_signatured_type_table_from_index): Likewise.
(create_debug_type_hash_table): Likewise.
(fill_in_sig_entry_from_dwo_entry): Likewise.
(lookup_dwo_unit): Access objfile through dwarf2_per_objfile.
(create_type_unit_group): Assign dwarf2_per_objfile instead of
objfile.
(create_partial_symtab): Access objfile through
dwarf2_per_objfile.
(process_psymtab_comp_unit_reader): Likewise.
(read_comp_units_from_section): Likewise.
(scan_partial_symbols): Likewise.
(add_partial_symbol): Likewise.
(add_partial_subprogram): Likewise.
(peek_die_abbrev): Likewise.
(fixup_go_packaging): Likewise.
(process_full_comp_unit): Likewise.
(process_full_type_unit): Likewise.
(process_imported_unit_die): Likewise.
(dwarf2_compute_name): Likewise.
(dwarf2_physname): Likewise.
(read_import_statement): Likewise.
(create_cus_hash_table): Assign dwarf2_physname instead of
objfile.
(read_func_scope): Access objfile through dwarf2_per_objfile.
(read_lexical_block_scope): Likewise.
(read_call_site_scope): Likewise.
(read_variable): Likewise.
(dwarf2_rnglists_process): Likewise.
(dwarf2_ranges_process): Likewise.
(dwarf2_ranges_read): Likewise.
(dwarf2_record_block_ranges): Likewise.
(dwarf2_add_field): Likewise.
(dwarf2_add_member_fn): Likewise.
(read_structure_type): Likewise.
(process_structure_scope): Likewise.
(read_enumeration_type): Likewise.
(read_array_type): Likewise.
(read_common_block): Likewise.
(read_namespace_type): Likewise.
(read_namespace): Likewise.
(read_module_type): Likewise.
(read_tag_pointer_type): Likewise.
(read_tag_ptr_to_member_type): Likewise.
(read_tag_string_type): Likewise.
(read_subroutine_type): Likewise.
(read_typedef): Likewise.
(read_base_type): Likewise.
(attr_to_dynamic_prop): Likewise.
(read_subrange_type): Likewise.
(read_unspecified_type): Likewise.
(load_partial_dies): Likewise.
(read_partial_die): Likewise.
(find_partial_die): Likewise.
(guess_partial_die_structure_name): Likewise.
(fixup_partial_die): Likewise.
(read_attribute_value): Likewise.
(read_addr_index_from_leb128): Likewise.
(dwarf2_read_addr_index): Likewise.
(dwarf2_string_attr): Likewise.
(lnp_state_machine::check_line_address): Likewise.
(dwarf_decode_lines_1): Likewise.
(dwarf_decode_lines): Likewise.
(dwarf2_start_symtab): Likewise.
(var_decode_location): Likewise.
(new_symbol_full): Likewise.
(dwarf2_const_value_data): Likewise.
(dwarf2_const_value_attr): Likewise.
(dwarf2_const_value): Likewise.
(die_type): Likewise.
(die_containing_type): Likewise.
(lookup_die_type): Likewise.
(guess_full_die_structure_name): Likewise.
(anonymous_struct_prefix): Likewise.
(dwarf2_name): Likewise.
(follow_die_ref_or_sig): Likewise.
(follow_die_offset): Likewise.
(follow_die_ref): Likewise.
(dwarf2_fetch_die_loc_sect_off): Likewise.
(dwarf2_fetch_constant_bytes): Likewise.
(dwarf2_fetch_die_type_sect_off): Likewise.
(dwarf2_get_die_type): Likewise.
(follow_die_sig): Likewise.
(decode_locdesc): Likewise.
(dwarf2_per_cu_objfile): Likewise.
(dwarf2_per_cu_text_offset): Likewise.
(init_one_comp_unit): Assign dwarf2_per_objfile instead of
objfile.
(set_die_type): Access objfile through
dwarf2_per_objfile.

6 years agoMake parse_debug_format_options return an std::string
Simon Marchi [Sun, 7 Jan 2018 15:50:48 +0000 (10:50 -0500)]
Make parse_debug_format_options return an std::string

This avoid having to manually free the return value.

gdb/gdbserver/ChangeLog:

* server.c (parse_debug_format_options): Return std::string.
(handle_monitor_command, captured_main): Adjust.

6 years agoReplace VEC(converted_character_d) with std::vector
Simon Marchi [Sun, 7 Jan 2018 15:48:21 +0000 (10:48 -0500)]
Replace VEC(converted_character_d) with std::vector

This patch changes the usage of VEC(converted_character_d) to use an
std::vector instead.  This allows getting rid of a cleanup.

gdb/ChangeLog:

* valprint.c (converted_character_d): Remove typedef.
(DEF_VEC_O (converted_character_d)): Remove.
(count_next_character): Use std::vector.
(print_converted_chars_to_obstack): Likewise.
(generic_printstr): Likewise.

6 years agoReplace VEC(gdb_xml_value_s) with std::vector
Simon Marchi [Sun, 7 Jan 2018 14:29:52 +0000 (09:29 -0500)]
Replace VEC(gdb_xml_value_s) with std::vector

This patch replaces VEC(gdb_xml_value_s), which is passed to XML
visitors, with an std::vector.  In order to be able to remove the
cleanup in gdb_xml_parser::start_element, the gdb_xml_parser structure
is made to own the value with a gdb::unique_xmalloc_ptr.

This patch has been tested on the buildbot.

gdb/ChangeLog:

* xml-support.h (struct gdb_xml_value): Add constructor.
<value>: Change type to unique_xmalloc_ptr.
(gdb_xml_value_s): Remove typedef.
(DEF_VEC_O (gdb_xml_value_s)): Remove.
(gdb_xml_element_start_handler): Change parameter type to
std::vector.
(xml_find_attribute): Likewise.
* xml-support.c (xml_find_attribute): Change parameter type to
std::vector and adjust.
(gdb_xml_values_cleanup): Remove.
(gdb_xml_parser::start_element): Adjust to std::vector.
(xinclude_start_include): Change paraeter type to std::vector
and adjust.
* btrace.c (check_xml_btrace_version): Likewise.
(parse_xml_btrace_block): Likewise.
(parse_xml_btrace_pt_config_cpu): Likewise.
(parse_xml_btrace_pt): Likewise.
(parse_xml_btrace_conf_bts): Likewise.
(parse_xml_btrace_conf_pt): Likewise.
* memory-map.c (memory_map_start_memory): Likewise.
(memory_map_start_property): Likewise.
* osdata.c (osdata_start_osdata): Likewise.
(osdata_start_item): Likewise.
(osdata_start_column): Likewise.
* remote.c (start_thread): Likewise.
* solib-aix.c (library_list_start_library): Likewise.
(library_list_start_list): Likewise.
* solib-svr4.c (library_list_start_library): Likewise.
(svr4_library_list_start_list): Likewise.
* solib-target.c (library_list_start_segment): Likewise.
(library_list_start_section): Likewise.
(library_list_start_library): Likewise.
(library_list_start_list): Likewise.
* tracepoint.c (traceframe_info_start_memory): Likewise.
(traceframe_info_start_tvar): Likewise.
* xml-syscall.c (syscall_start_syscall): Likewise.
* xml-tdesc.c (tdesc_start_target): Likewise.
(tdesc_start_feature): Likewise.
(tdesc_start_reg): Likewise.
(tdesc_start_union): Likewise.
(tdesc_start_struct): Likewise.
(tdesc_start_flags): Likewise.
(tdesc_start_enum): Likewise.
(tdesc_start_field): Likewise.
(tdesc_start_enum_value): Likewise.
(tdesc_start_vector): Likewise.

6 years agoRemove xmethod_worker::clone
Simon Marchi [Sun, 7 Jan 2018 14:25:33 +0000 (09:25 -0500)]
Remove xmethod_worker::clone

I think that the clone method of xmethod_worker can be removed.  It is
only used in find_overload_match, to clone an xmethod we want to
keep.  Instead, we can just std::move it out of the vector and into
value_from_xmethod.  value_from_xmethod creates a value that will own
the xmethod_worker from that point.  Other xmethod_workers left in the
vector will get destroyed when the vector gets destroyed, but the chosen
one will keep living inside the value struct.

gdb/ChangeLog:

* extension.h (struct xmethod_worker) <clone>: Remove.
* python/py-xmethods.c (struct python_xmethod_worker) <clone>:
Remove.
(python_xmethod_worker::clone): Remove.
* valops.c (find_overload_match): Use std::move instead of
clone.

6 years agoC++ify xmethod_worker, get rid of VEC(xmethod_worker_ptr)
Simon Marchi [Sun, 7 Jan 2018 14:25:32 +0000 (09:25 -0500)]
C++ify xmethod_worker, get rid of VEC(xmethod_worker_ptr)

The initial goal of this patch was to remove the usage of
VEC(xmethod_worker_ptr) and corresponding cleanups.  I ended up having
to  C++ify the xmethod_worker code, to be able to have xmethod_workers
free their data in destructors, and therefore be able to use vectors of
xmethod_worker unique_ptr.

The operations in extension_language_ops that act on one instance of
xmethod_worker (get result type, get args type, invoke) are transformed
to methods of xmethod_worker.  xmethod_worker becomes an abstract base
class with virtual pure methods which python_xmethod_worker implements.
The only xmethod-related operation left in extension_language_ops is
get_matching_xmethod_workers, which returns a list of xmethod_workers.

The changes are relatively straightforward, but here are some notes on
things that may raise eyebrows:

  - I was not really comfortable with the value_of_xmethod function.  At
  first it looks like a simple getter, so I considered making it a
  method of xmethod_worker.  But actually it creates a value and
  transfers the ownership of the xmethod_worker to it.  It would be a
  bit weird and error-prone if calling a method on an object silently
  removed the ownership of the object from the caller.  To reflect the
  behavior more accurately, I renamed it to value_from_xmethod and made
  it accept an rvalue-reference (so the caller knows it gives away the
  ownership).  I noticed the backlink from xmethod_worker to its owning
  value was not used, so I removed it.

  - Some code, like get_matching_xmethod_workers, made each callee fill
  a new vector, which was then merged in the result vector.  I think
  it's safe if we always pass the same vector around, and each
  implementation just appends to it.

  - The clone operation does not seem particularly useful, it is removed
  in the following patch.

gdb/ChangeLog:

* extension-priv.h (enum ext_lang_rc): Remove, move to extension.h.
(struct extension_language_ops) <clone_xmethod_worker_data>: Remove.
<free_xmethod_worker_data>: Remove.
<get_matching_xmethod_workers>: Chance VEC to std::vector.
<get_xmethod_arg_types>: Remove.
<get_xmethod_result_type>: Remove.
<invoke_xmethod>: Remove.
* extension.c (new_xmethod_worker): Remove.
(clone_xmethod_worker): Remove.
(get_matching_xmethod_workers): Return void, pass std::vector by
pointer.
(get_xmethod_arg_types): Rename to...
(xmethod_worker::get_arg_types): ... this, and adjust.
(get_xmethod_result_type): Rename to...
(xmethod_worker::get_result_type): ... this, and adjust.
(invoke_xmethod): Remove.
(free_xmethod_worker): Remove.
(free_xmethod_worker_vec): Remove.
* extension.h (enum ext_lang_rc): Move here from
extension-priv.h.
(struct xmethod_worker): Add constructor and destructor.
<data>: Remove.
<value>: Remove.
<invoke, clone, do_get_result_type, do_get_arg_types>: New
virtual pure methods.
<get_arg_types, get_result_type>: New methods.
(xmethod_worker_ptr): Remove typedef.
(DEF_VEC_P (xmethod_worker_ptr)): Remove.
(xmethod_worker_vec): Remove typedef.
(xmethod_worker_up): New typedef.
(invoke_xmethod): Remove.
(clone_xmethod_worker): Remove.
(free_xmethod_worker): Remove.
(free_xmethod_worker_vec): Remove.
(get_xmethod_arg_types): Remove.
(get_xmethod_result_type): Remove.
* valops.c (find_method_list): Use std::vector, don't use
intermediate vector.
(value_find_oload_method_list): Use std::vector.
(find_overload_match): Use std::vector.
(find_oload_champ): Use std::vector.
* value.c (value_free): Use operator delete.
(value_of_xmethod): Rename to...
(value_from_xmethod): ... this.  Don't assign
xmethod_worker::value, take rvalue-reference.
(result_type_of_xmethod): Adjust.
(call_xmethod): Adjust.
* value.h: Include extension.h.
(struct xmethod_worker): Don't forward-declare.
(value_of_xmethod): Rename to...
(value_from_xmethod): ... this, take rvalue-reference.
* python/py-xmethods.c (struct gdbpy_worker_data): Rename to...
(struct python_xmethod_worker): ... this, add constructor and
destructor.
<invoke, clone, do_get_arg_types, do_get_result_type>: Implement.
(gdbpy_free_xmethod_worker_data): Rename to...
(python_xmethod_worker::~python_xmethod_worker): ... this and
adjust.
(gdbpy_clone_xmethod_worker_data): Rename to...
(python_xmethod_worker::clone): ... this and adjust.
(gdbpy_get_matching_xmethod_workers): Use std::vector, don't use
temporary vector.
(gdbpy_get_xmethod_arg_types): Rename to...
(python_xmethod_worker::do_get_arg_types): ... this and adjust.
(gdbpy_get_xmethod_result_type): Rename to...
(python_xmethod_worker::do_get_result_type): ... this and
adjust.
(gdbpy_invoke_xmethod): Rename to...
(python_xmethod_worker::invoke): ... this and adjust.
(new_python_xmethod_worker): Rename to...
(python_xmethod_worker::python_xmethod_worker): ... this and
adjust.
* python/python-internal.h (gdbpy_clone_xmethod_worker_data):
Remove.
(gdbpy_free_xmethod_worker_data): Remove.
(gdbpy_get_matching_xmethod_workers): Use std::vector.
(gdbpy_get_xmethod_arg_types): Remove.
(gdbpy_get_xmethod_result_type): Remove.
(gdbpy_invoke_xmethod): Remove.
* python/python.c (python_extension_ops): Remove obsolete
callbacks.

6 years agoAutomatic date update in version.in
GDB Administrator [Sun, 7 Jan 2018 00:00:33 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoRISC-V: Print symbol address for jalr w/ zero offset.
Jim Wilson [Sat, 6 Jan 2018 01:51:23 +0000 (17:51 -0800)]
RISC-V: Print symbol address for jalr w/ zero offset.

ld/
* testsuite/ld-riscv-elf/disas-jalr.d: New.
* testsuite/ld-riscv-elf/disas-jalr.s: New.
* testsuite/ld-riscv-elf/ld-riscv-elf.exp: Run new testcase.

opcodes/
* riscv-dis.c (print_insn_args) <'s'>: Call maybe_print_address for a
jalr.

6 years agoAutomatic date update in version.in
GDB Administrator [Sat, 6 Jan 2018 00:00:28 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoRISC-V: Disable shared library support for embedded elf.
Jim Wilson [Fri, 5 Jan 2018 22:29:49 +0000 (14:29 -0800)]
RISC-V: Disable shared library support for embedded elf.

ld/
* emulparams/elf32lriscv-defs.sh (GENERATE_SHLIB_SCRIPT): Move inside
case on $target, and don't set for riscv*-elf targets.
(GENERATE_PIE_SCRIPT): Likewise.

6 years agoFix regression: cannot start with LD_PRELOAD=libSegFault.so (PR gdb/18653#c7)
Pedro Alves [Fri, 5 Jan 2018 18:26:18 +0000 (18:26 +0000)]
Fix regression: cannot start with LD_PRELOAD=libSegFault.so (PR gdb/18653#c7)

At https://sourceware.org/bugzilla/show_bug.cgi?id=18653#c7, Andrew
reports that the fix for PR gdb/18653 made GDB useless if you preload
libSegFault.so, because GDB internal-errors on startup:

 $ LD_PRELOAD=libSegFault.so gdb
 src/gdb/common/signals-state-save-restore.c:64: internal-error: unexpected signal handler
 A problem internal to GDB has been detected,
 further debugging may prove unreliable.
 Aborted (core dumped)
 $

The internal error comes from the code saving the signal dispositions
inherited from gdb's parent:

 (top-gdb) bt
 #0  0x000000000056b001 in internal_error(char const*, int, char const*, ...) (file=0xaf5f38 "src/gdb/common/signals-state-save-restore.c", line=64, fmt=0xaf5f18 "unexpected signal handler") at src/gdb/common/errors.c:54
 #1  0x00000000005752c9 in save_original_signals_state() () at src/gdb/common/signals-state-save-restore.c:64
 #2  0x00000000007425de in captured_main_1(captured_main_args*) (context=0x7fffffffd860)
     at src/gdb/main.c:509
 #3  0x0000000000743622 in captured_main(void*) (data=0x7fffffffd860) at src/gdb/main.c:1145
 During symbol reading, cannot get low and high bounds for subprogram DIE at 24065.
 #4  0x00000000007436f9 in gdb_main(captured_main_args*) (args=0x7fffffffd860) at src/gdb/main.c:1171
 #5  0x0000000000413acd in main(int, char**) (argc=1, argv=0x7fffffffd968) at src/gdb/gdb.c:32

This commit downgrades the internal error to a warning.  You'll get
instead:

~~~
 $ LD_PRELOAD=libSegFault.so gdb
 warning: Found custom handler for signal 11 (Segmentation fault) preinstalled.
 Some signal dispositions inherited from the environment (SIG_DFL/SIG_IGN)
 won't be propagated to spawned programs.
 GNU gdb (GDB) 8.0.50.20171213-git
 Copyright (C) 2017 Free Software Foundation, Inc.
 License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
 This is free software: you are free to change and redistribute it.
 There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
 and "show warranty" for details.
 This GDB was configured as "x86_64-pc-linux-gnu".
 Type "show configuration" for configuration details.
 For bug reporting instructions, please see:
 <http://www.gnu.org/software/gdb/bugs/>.
 Find the GDB manual and other documentation resources online at:
 <http://www.gnu.org/software/gdb/documentation/>.
 For help, type "help".
 Type "apropos word" to search for commands related to "word"...
 (gdb)
~~~

This also moves the location where save_original_signals_state is
called a bit further below (to after option processing), so that "-q"
disables the warning:

~~~
 $ LD_PRELOAD=libSegFault.so gdb -q
 (gdb)
~~~

New testcase included.

gdb/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/18653
* common/signals-state-save-restore.c
(save_original_signals_state): New parameter 'quiet'.  Warn if we
find a custom handler preinstalled, instead of internal erroring.
But only warn if !quiet.
* common/signals-state-save-restore.h
(save_original_signals_state): New parameter 'quiet'.
* main.c (captured_main_1): Move save_original_signals_state call
after option handling, and pass QUIET.

gdb/gdbserver/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/18653
* server.c (captured_main): Pass quiet=false to
save_original_signals_state.

gdb/testsuite/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/18653
* gdb.base/libsegfault.exp: New.

6 years agoFix gdb/spu-tdep.c build breakage
Pedro Alves [Fri, 5 Jan 2018 17:56:33 +0000 (17:56 +0000)]
Fix gdb/spu-tdep.c build breakage

Commit de63c46b549d ("Fix regresssion(internal-error) printing
subprogram argument (PR gdb/22670)") missed updating spu-tdep.c for
the block_lookup_symbol interface change, resulting in:

  ../../binutils-gdb/gdb/spu-tdep.c: In function void spu_catch_start(objfile*):
  ../../binutils-gdb/gdb/spu-tdep.c:1969:59: error: cannot convert domain_enum_tag to symbol_name_match_type for argument 3 to symbol* block_lookup_symbol(const block*, const char*, symbol_name_match_type, domain_enum)
 sym = block_lookup_symbol (block, "main", VAR_DOMAIN);
     ^

gdb/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

* spu-tdep.c (spu_catch_start): Pass
symbol_name_match_type::SEARCH_NAME to block_lookup_symbol.

6 years agoFix regresssion(internal-error) printing subprogram argument (PR gdb/22670)
Pedro Alves [Fri, 5 Jan 2018 14:04:09 +0000 (14:04 +0000)]
Fix regresssion(internal-error) printing subprogram argument (PR gdb/22670)

At <https://sourceware.org/ml/gdb-patches/2017-12/msg00298.html>, Joel
wrote:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider the following code which first declares a tagged type (the
equivalent of a class in Ada), and then a procedure which takes a
pointer (access) to this type's 'Class.

    package Pck is
       type Top_T is tagged record
          N : Integer := 1;
       end record;
       procedure Inspect (Obj: access Top_T'Class);
    end Pck;

Putting a breakpoint in that procedure and then running to it triggers
an internal error:

    (gdb) break inspect
    (gdb) continue
    Breakpoint 1, pck.inspect (obj=0x63e010
    /[...]/gdb/stack.c:621: internal-error: void print_frame_args(symbol*, frame_info*, int, ui_file*): Assertion `nsym != NULL' failed.

What's special about this subprogram is that it takes an access to
what we call a 'Class type, and for implementation reasons, the
compiler adds an extra argument named "objL". If you are curious why,
it allows the compiler for perform dynamic accessibility checks that
are mandated by the language.

If we look at the location where we get the internal error (in
stack.c), we find that we are looping over the symbol of each
parameter, and for each parameter, we do:

    /* We have to look up the symbol because arguments can have
       two entries (one a parameter, one a local) and the one we
       want is the local, which lookup_symbol will find for us.
    [...]
        nsym = lookup_symbol (SYMBOL_LINKAGE_NAME (sym),
                              b, VAR_DOMAIN, NULL).symbol;
        gdb_assert (nsym != NULL);

The lookup_symbol goes through the lookup structure, which means the
symbol's linkage name ("objL") gets transformed into a
lookup_name_info object (in block_lookup_symbol), before it gets fed
to the block symbol dictionary iterators.  This, in turn, triggers the
symbol matching by comparing the "lookup" name which, for Ada, means
among other things, lowercasing the given name to "objl".  It is this
transformation that causes the lookup find no matches, and therefore
trip this assertion.

Going back to the "offending" call to lookup_symbol in stack.c, what
we are trying to do, here, is do a lookup by linkage name.  So, I
think what we mean to be doing is a completely literal symbol lookup,
so maybe not even strcmp_iw, but actually just plain strcmp???

In the past, in practice, you could get that effect by doing a lookup
using the C language. But that doesn't work, because we still end up
somehow using Ada's lookup_name routine which transforms "objL".

So, ideally, as I hinted before, I think what we need is a way to
perform a literal lookup so that searches by linkage names like the
above can be performed.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This commit fixes the problem by implementing something similar to
Joel's literal idea, but with some important differences.

I considered adding a symbol_name_match_type::LINKAGE and supporting
searching by linkage name for any language, but the problem with that
is that the dictionaries only work with SYMBOL_SEARCH_NAME, because
that's what is used for hashing.  We'd need separate dictionaries for
hashed linkage names.

So with the current symbol tables infrastructure, it's not literal
linkage names that we want to pass down, but instead literal _search_
names (SYMBOL_SEARCH_NAME, etc.).

However, psymbols have no overload/function parameter info in C++, so
a straight strcmp doesn't work properly for C++ name matching.

So what we do is be a little less aggressive then and add a new
symbol_name_match_type::SEARCH_SYMBOL instead that takes as input a
non-user-input search symbol, and then we skip any decoding/demangling
steps and make:

 - Ada treat that as a verbatim match,
 - other languages treat it as symbol_name_match_type::FULL.

This also fixes the new '"maint check-psymtabs" for Ada' testcase for
me (gdb.ada/maint_with_ada.exp).  I've not removed the kfail yet
because Joel still sees that testcase failing with this patch.
That'll be fixed in follow up patches.

gdb/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/22670
* ada-lang.c (literal_symbol_name_matcher): New function.
(ada_get_symbol_name_matcher): Use it for
symbol_name_match_type::SEARCH_NAME.
* block.c (block_lookup_symbol): New parameter 'match_type'.  Pass
it down instead of assuming symbol_name_match_type::FULL.
* block.h (block_lookup_symbol): New parameter 'match_type'.
* c-valprint.c (print_unpacked_pointer): Use
lookup_symbol_search_name instead of lookup_symbol.
* compile/compile-object-load.c (get_out_value_type): Pass down
symbol_name_match_type::SEARCH_NAME.
* cp-namespace.c (cp_basic_lookup_symbol): Pass down
symbol_name_match_type::FULL.
* cp-support.c (cp_get_symbol_name_matcher): Handle
symbol_name_match_type::SEARCH_NAME.
* infrun.c (insert_exception_resume_breakpoint): Use
lookup_symbol_search_name.
* p-valprint.c (pascal_val_print): Use lookup_symbol_search_name.
* psymtab.c (maintenance_check_psymtabs): Use
symbol_name_match_type::SEARCH_NAME and SYMBOL_SEARCH_NAME.
* stack.c (print_frame_args): Use lookup_symbol_search_name and
SYMBOL_SEARCH_NAME.
* symtab.c (lookup_local_symbol): Don't demangle the lookup name
if symbol_name_match_type::SEARCH_NAME.
(lookup_symbol_in_language): Pass down
symbol_name_match_type::FULL.
(lookup_symbol_search_name): New.
(lookup_language_this): Pass down
symbol_name_match_type::SEARCH_NAME.
(lookup_symbol_aux, lookup_local_symbol): New parameter
'match_type'.  Pass it down.
* symtab.h (symbol_name_match_type::SEARCH_NAME): New enumerator.
(lookup_symbol_search_name): New declaration.
(lookup_symbol_in_block): New 'match_type' parameter.

gdb/testsuite/ChangeLog:
2018-01-05  Joel Brobecker  <brobecker@adacore.com>

PR gdb/22670
* gdb.ada/access_tagged_param.exp: New file.
* gdb.ada/access_tagged_param/foo.adb: New file.

6 years agoFix gdb.ada/info_addr_mixed_case.exp (PR gdb/22670)
Pedro Alves [Fri, 5 Jan 2018 16:01:57 +0000 (16:01 +0000)]
Fix gdb.ada/info_addr_mixed_case.exp (PR gdb/22670)

The comments about mixed case in the testcase are actually a red
herring.  The problem here is that we'd get to
ada_lookup_encoded_symbol with "my_table", which wraps the looked up
name in "<>"s to force a verbatim match, and that in turn disables
wild matching.

Fix this by swapping around the internals of ada_lookup_encoded_symbol
and ada_lookup_symbol, thus avoiding the encoding and
verbatim-wrapping in the ada_lookup_symbol case, the case that starts
with a user-provided lookup name.

Ada encoding is still done of course, in the ada_lookup_name_info
ctor.  This could be also seen as avoiding the double-encoding problem
in a different way.

gdb/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/22670
* ada-lang.c (ada_lookup_encoded_symbol): Reimplement in terms of
ada_lookup_symbol.
(ada_lookup_symbol): Reimplement in terms of
ada_lookup_symbol_list, bits factored out from
ada_lookup_encoded_symbol.

gdb/testsuite/ChangeLog:
2018-01-05  Pedro Alves  <palves@redhat.com>

PR gdb/22670
* gdb.ada/info_addr_mixed_case.exp: Remove kfail.  Extend test to
exercise lower case too, and to exercise both full matching and
wild matching.

6 years ago(Ada) problem printing renaming which references a subprogram parameter
Joel Brobecker [Fri, 5 Jan 2018 09:29:54 +0000 (04:29 -0500)]
(Ada) problem printing renaming which references a subprogram parameter

Consider the following code, which creates a local variable B
which is a renaming whose expression references a subprogram
parameter:

   procedure Flip (Bits : in out Bits_Type; I : Natural) is
   begin
      declare
         B : Boolean renames Bits (I);
      begin
         B := not B; -- BREAK
      end;
   end Flip;

Trying to print the value of B when at the "BREAK" line currently
does not work:

    (gdb) p b
    Could not find i

What happens is the following: For the renaming, GNAT generates
a variable whose name is encoded as follow:

    b___XR_bits___XEXSi

GDB properly detects that variable, determines that, to compute
the variable's value, we start from the symbol "Bits", which
we then have to subscript (XS) using 'i' as the index. The error
occurs while trying to find 'i'.

This is because we forgot to pass the block in the call to
ada_lookup_encoded_symbol, which this patch fixes.

gdb/ChangeLog:

        * ada-exp.y (write_object_renaming): When subscripting an array
        using a symbol as the index, pass the block in call to
        ada_lookup_encoded_symbol when looking that symbol up.

gdb/testsuite/ChangeLog:

        * gdb.ada/rename_subscript_param: New testcase.

Tested on x86_64-linux.

Note: This requires the following GCC patch:

  | 2017-04-25  Pierre-Marie de Rodat  <derodat@adacore.com>
  |
  | * exp_dbug.adb: In Debug_Renaming_Declaration,
  | when dealing with indexed component, accept to produce a renaming
  | symbol when the index is an IN parameter or when it is a name
  | defined in an outer scope.

6 years ago(Ada) Fix Length attribute on array access
Jerome Guitton [Fri, 5 Jan 2018 08:03:59 +0000 (03:03 -0500)]
(Ada) Fix Length attribute on array access

Consider the following variable "Indexed_By_Enum", declared as
an access to an array whose index type is an enumerated type
whose underlying values have "gaps":

   type Enum_With_Gaps is (LIT0, LIT1, LIT2, LIT3, LIT4);
   for Enum_With_Gaps use (LIT0 => 3,
                           LIT1 => 5,
                           LIT2 => 8,
                           LIT3 => 13,
                           LIT4 => 21);
   for Enum_With_Gaps'size use 16;

   type MyWord is range 0 .. 16#FFFF# ;
   for MyWord'Size use 16;

   type AR is array (Enum_With_Gaps range <>) of MyWord;
   type AR_Access is access AR;

   Indexed_By_Enum : AR_Access :=
     new AR'(LIT1 => 1,  LIT2 => 43, LIT3 => 42, LIT4 => 41);

Trying to print the length (number of elements) of this array using
the 'Length attribute does not work:

    (gdb) print indexed_by_enum'length
    'POS only defined on discrete types

The problem occurs while trying to get the array's index type.
It was using TYPE_INDEX_TYPE for that. It does not work for Ada arrays
in general; use ada_index_type instead.

gdb/ChangeLog:

* ada-lang.c (ada_array_length): Use ada_index_type instead of
TYPE_INDEX_TYPE.

gdb/testsuite/ChangeLog:

        * gdb.ada/arr_acc_idx_w_gap: New testcase.

Tested on x86_64-linux.

6 years agomemory error printing component of record from convenience variable
Joel Brobecker [Fri, 5 Jan 2018 07:37:38 +0000 (02:37 -0500)]
memory error printing component of record from convenience variable

Consider the following situation Ada code:

   type Kind_T is (One, Two, Three);
   type Time_Set_T is array (Kind_T) of Integer;

   type T  is record
      Started    : Time_Set_T;
   end record;

   Null_T : constant T := (Started => (others => 0));

   My_Item : Pck.T := Pck.Null_T;

Trying to print the value of My_Item.Started is no problem:

    (gdb) p item.started
    $1 = (0, 0, 0)

However, if you save My_Item into a convenience variable first,
and then try to print a component of that record, you get
an unexpected memory error, instead of getting the same result.
For instance:

    (gdb) set variable $item := item
    (gdb) p $item.started
    Cannot access memory at address 0x0

The issue occurs when, after we extracted the component from
the convenience variable, we next try to "fix" it (which is
ada-lang speak for resolving the type into a static type).
This is done in ada_to_fixed_value, which delegates to
ada_to_fixed_value_create via:

    val = ada_to_fixed_value_create (value_type (val),
                                     value_address (val), val);

And looking at ada_to_fixed_value_create, we see that:

  struct type *type = ada_to_fixed_type (type0, 0, address, NULL, 1);

  if (type == type0 && val0 != NULL)
    return val0;
  else
    return value_from_contents_and_address (type, 0, address);

The part that interests us, in this case, is the "else" branch,
where we obviously make the implicit assumption that our object
has an address, which is not true, in this case, because we are
actually dealing with a convenience variable.

This patch plugs that hole by adding special handing for situations
where val does not live in memory. In that case, we just create
a not_lval value using val's contents.

gdb/ChangeLog:

        * ada-lang.c (ada_to_fixed_value_create): Add handling of
        the case where VALUE_LVAL (val0) is not lval_memory.

gdb/testsuite/ChangeLog:

        * gdb.ada/convvar_comp: New testcase.

Tested on x86_64-linux.

6 years agofix date of latest gdb/ChangeLog entry
Joel Brobecker [Fri, 5 Jan 2018 07:39:10 +0000 (02:39 -0500)]
fix date of latest gdb/ChangeLog entry

6 years agoFix email address in latest gdb/testsuite/ChangeLog entry
Joel Brobecker [Fri, 5 Jan 2018 04:51:55 +0000 (23:51 -0500)]
Fix email address in latest gdb/testsuite/ChangeLog entry

6 years ago(Ada) Remove printing of array's first index when unneeded
Xavier Roirand [Fri, 5 Jan 2018 04:47:05 +0000 (23:47 -0500)]
(Ada) Remove printing of array's first index when unneeded

Consider the following code:

  type Table is array (Character) of Natural;
  My_Table : Table := (others => 4874);

Printing this table in gdb leads to:

  (gdb) p my_table
  $1 = ('["00"]' => 4874 <repeats 256 times>)

In this case, the index of the first element in this array is also
the first element of the index type (character type). Similar to what
we do we enumeration types, we do not need to print the index of the
first element when printing the array.

This patch fixes this issue and changes the output as follow:

  (gdb) p my_table
  $1 = (4874 <repeats 256 times>)

gdb/ChangeLog:

        * ada-valprint.c (print_optional_low_bound): Handle
        character-indexed array printing like boolean-indexed array
        printing.

gdb/testuite/ChangeLog:

        * testsuite/gdb.ada/array_char_idx/pck.ads (Table): New type.
        (My_Table): New global variable.
        * testsuite/gdb.ada/array_char_idx.exp: Add test.

Tested on x86_64-linux.

6 years agoUpdate NEWS post GDB 8.1 branch creation.
Joel Brobecker [Fri, 5 Jan 2018 04:18:05 +0000 (08:18 +0400)]
Update NEWS post GDB 8.1 branch creation.

gdb/ChangeLog:

* NEWS: Create a new section for the next release branch.
Rename the section of the current branch, now that it has
been cut.

6 years agoBump version to 8.1.50.DATE-git.
Joel Brobecker [Fri, 5 Jan 2018 04:08:09 +0000 (08:08 +0400)]
Bump version to 8.1.50.DATE-git.

Now that the GDB 8.1 branch has been created, we can
bump the version number.

gdb/ChangeLog:

GDB 8.1 branch created (5219ac6237c272b938c28517bf371429260c71e7):
* version.in: Bump version to 8.1.50.DATE-git.

6 years agoAutomatic date update in version.in
GDB Administrator [Fri, 5 Jan 2018 00:00:35 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoRISC-V: Add 2 missing privileged registers.
Jim Wilson [Thu, 4 Jan 2018 22:17:53 +0000 (14:17 -0800)]
RISC-V: Add 2 missing privileged registers.

gas/
* testsuite/gas/riscv/priv-reg.s: Add missing stval and mtval.
* testsuite/gas/riscv/priv-reg.d: Likewise.

include/
* opcode/riscv-opc.h (CSR_SBADADDR): Rename to CSR_STVAL.  Rename
DECLARE_CSR entry.  Add alias to map sbadaddr to CSR_STVAL.
(CSR_MBADADDR): Rename to CSR_MTVAL.  Rename DECLARE_CSR entry.
Add alias to map mbadaddr to CSR_MTVAL.

6 years agogdb.ada/maint_with_ada.exp: New testcase
Joel Brobecker [Thu, 4 Jan 2018 09:40:17 +0000 (04:40 -0500)]
gdb.ada/maint_with_ada.exp: New testcase

This commit adds a new testcase testing the "check-psymtabs" and
"check-symtabs" maintenance commands.

The "maintenance check-psymtabs" commands is currently known to
produce some errors. While the situation was admetedly made worse
by the following patch...

    commit b5ec771e60c1a0863e51eb491c85c674097e9e13
    Date:   Wed Nov 8 14:22:32 2017 +0000
    Subject: Introduce lookup_name_info and generalize Ada's FULL/WILD name matching

... hence the reference to PR gdb/22670, the command was already
spotting one inconsistency prior to it:

    (gdb) maintenance check-psymtabs
    Global symbol `interfaces__cS' only found in /[...]/b~var_arr_typedef.adb psymtab

For now, the "check-psymtab" test is KFAIL-ed.

gdb/testsuite/ChangeLog:

        PR gdb/22670
        * gdb.ada/maint_with_ada: New testcase.

Tested on x86_64-linux.

6 years agoAdd new gdb.ada/bp_c_mixed_case testcase for PR gdb/22670
Joel Brobecker [Thu, 4 Jan 2018 08:10:34 +0000 (03:10 -0500)]
Add new gdb.ada/bp_c_mixed_case testcase for PR gdb/22670

This patch adds a new testcase to demonstrate a regression introduced by:

    commit b5ec771e60c1a0863e51eb491c85c674097e9e13
    Date:   Wed Nov 8 14:22:32 2017 +0000
    Subject: Introduce lookup_name_info and generalize Ada's FULL/WILD name matching

The purpose of the testcase is to verify that a user can insert
a breakpoint on a C function while debugging Ada, even if the name
of the function includes uppercase letters, requiring us to use
Ada's "<...>" notation to tell the GDB that the symbol name should
be looked up verbatim.

As of the commit above, GDB is no longer finding the function:

    (gdb) break <MixedCaseFunc>
    Function "<MixedCaseFunc>" not defined.
    Make breakpoint pending on future shared library load? (y or [n])

Before the patch, the breakpoint was inserted without problem.

gdb/testsuite/ChangeLog:

        PR gdb/22670
        * gdb.ada/bp_c_mixed_case: New testcase.

Tested on x86_64-linux; generates a KPASS before the regression
was introduced, and now generates a KFAIL.

6 years agoAdd "complete break ada" test to gdb.ada/complete.exp
Joel Brobecker [Thu, 4 Jan 2018 07:39:58 +0000 (02:39 -0500)]
Add "complete break ada" test to gdb.ada/complete.exp

This patch adds a new test to demonstrate a regression introduced by:

    commit b5ec771e60c1a0863e51eb491c85c674097e9e13
    Date:   Wed Nov 8 14:22:32 2017 +0000
    Subject: Introduce lookup_name_info and generalize Ada's FULL/WILD name matching

The original purpose of the new test is to exercise the "complete"
command with an expression for which a large number of matches are
returned and to verify that each match returned is a plausible match.
In this particular case, the commit above causes GDB to generate
additional matches which should in fact not appear in the list
(internally generated symbols, or symbols that should be enclosed
between "<...>"). These extraneous entries are easy to spot, because
they have uppercase characters, such as:

    break ada__stringsS
    break ada__strings__R11s
    [etc]

For now, the new test is KFAIL'ed, to avoid generating a new FAIL
while we work on fixing that regression.

gdb/testsuite/ChangeLog:

        PR gdb/22670
        * gdb.ada/complete.exp: Add "complete break ada" test.

Tested on x86_64-linux with GDB built before and after the patch
that caused the regression (b5ec771e60c1a0863e51eb491c85c674097e9e13).
The test passes before the regression, and generates a KFAIL after.

6 years agoAdd gdb.ada/info_addr_mixed_case new testcase
Joel Brobecker [Thu, 4 Jan 2018 04:28:14 +0000 (23:28 -0500)]
Add gdb.ada/info_addr_mixed_case new testcase

This patch adds a new testcase to demonstrate a regression introduced by:

    commit b5ec771e60c1a0863e51eb491c85c674097e9e13
    Date:   Wed Nov 8 14:22:32 2017 +0000
    Subject: Introduce lookup_name_info and generalize Ada's FULL/WILD name matching

The purpose of the testcase is to verify that a user can use any
casing for an Ada symbol name passed to the "info address" command.
After the patch above was applied, GDB was no longer able to find
the symbol:

    (gdb) info address My_Table
    No symbol "My_Table" in current context.

gdb/testsuite/ChangeLog:

PR gdb/22670
* gdb.ada/info_addr_mixed_case: New testcase.

Tested on x86_64-linux, both before and after the patch.

6 years agoAutomatic date update in version.in
GDB Administrator [Thu, 4 Jan 2018 00:00:20 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoCreate pseudo sections for FreeBSD NT_PROCSTAT_(PROC|FILES|VMMAP) notes.
John Baldwin [Wed, 3 Jan 2018 20:06:26 +0000 (12:06 -0800)]
Create pseudo sections for FreeBSD NT_PROCSTAT_(PROC|FILES|VMMAP) notes.

bfd/ChangeLog:

* elf.c (elfcore_grok_freebsd_note): Handle
NT_FREEBSD_PROCSTAT_PROC, NT_FREEBSD_PROCSTAT_FILES, and
NT_FREEBSD_PROCSTAT_VMMAP.

6 years agoFix compile time warning introduced by previous delta.
Nick Clifton [Wed, 3 Jan 2018 17:32:59 +0000 (17:32 +0000)]
Fix compile time warning introduced by previous delta.

* objcopy.c (merge_gnu_build_notes): Add cast to maximum address
constants.

6 years agoAdd support for v3 binary annotation notes.
Nick Clifton [Wed, 3 Jan 2018 10:28:33 +0000 (10:28 +0000)]
Add support for v3 binary annotation notes.

  I am checking in the attached patch which updates the binutils
  support for version 3 binary annotation notes.  (Version 3 adds
  an end address to the ranges covered by the notes, so that it
  is possible to detect gaps in the coverage).

  This patch also stops the note merging feature of objcopy from
  executing if the notes have relocations against them.  This makes the
  code simpler, and prevents the problems with architectures which have
  unusual relocation management issues.

* objcopy.c (objcopy_internal_note): New structure.
(gap_exists): New function.
(is_open_note): New function.
(is_func_note): New function.
(is_64bit): New function.
(merge_gnu_build_notes): Handle v3 notes.  Do not merge
if there are relocations against the notes.
* readelf.c (get_note_type): Use short names for build attribute
notes.
(print_symbol_for_build_attribute): Rename to
get_symbol_for_build_attribute.  Returns the found symbol rather
than printing it.
(print_gnu_build_attribute_description): Maintain address ranges
for function notes as well as global notes.  Handle v3 notes.
(print_gnu_build_attribute_name): Use more space for printing the
name in wide mode.
* testsuite/binutils-all/note-2-32.s: Use .dc.l instead of .word.
Eliminate symbol references in order to remove the need for
relocations.
* testsuite/binutils-all/note-2-64.s: Likewise.
* testsuite/binutils-all/note-3-32.s: Add a size to the note_1
symbol.
* testsuite/binutils-all/note-3-64.s: Likewise.
* testsuite/binutils-all/mips/mips-note-2r-n32.d: Update expected
output.
* testsuite/binutils-all/mips/mips-note-2r-n64.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2r.d: Likewise.
* testsuite/binutils-all/note-2-32.d: Likewise.
* testsuite/binutils-all/note-2-64.d: Likewise.
* testsuite/binutils-all/note-3-32.d: Likewise.
* testsuite/binutils-all/note-3-64.d: Likewise.
* testsuite/binutils-all/note-4-64.s: New test.  Checks v3 notes.
* testsuite/binutils-all/note-4-32.s: New test.
* testsuite/binutils-all/note-4-64.d: New test result file.
* testsuite/binutils-all/note-4-32.d: New test result file.

6 years ago(Ada) New command to stop at start of exception handler.
Xavier Roirand [Wed, 22 Nov 2017 09:40:39 +0000 (10:40 +0100)]
(Ada) New command to stop at start of exception handler.

When using gdb for debugging Ada source code, there are several catchpoint
types you can define in order to stop upon certain conditions.  Let's
use this small example:

procedure Foo is
begin
   begin
      raise Constraint_Error;
   exception
      when Program_Error =>
         null;
      when Constraint_Error =>
         null;
      when others =>
         null;
   end;
end Foo;

One can stop when the exception is being raised by using the exception
catchpoint like below:

(gdb) catch exception
Catchpoint 1: all Ada exceptions
(gdb)

In that case, when running Foo, gdb will stop at the line where the exception
was raised:

   begin
>>>   raise Constraint_Error;
   exception

This patch introduces new type of catchpoint, when the user wants to stop
at the location of the exception handling.
Imagine we want to stop on any exception handled by the program, we can do:

(gdb) catch handlers
Catchpoint 1: all Ada exceptions handlers
(gdb) r
Starting program: /tmp/foo

By doing so, when running Foo, gdb will stop here:

Catchpoint 1, exception at 0x000000000040255a in foo () at foo.adb:25
25          when Constraint_Error =>
(gdb)

It is also possible to stop when the Constraint_Error exception is being
handled in this program.  With this patch, we can use:

(gdb) catch handlers Constraint_Error
Catchpoint 1: `Constraint_Error' Ada exception handlers
(gdb)

Like for other catchpoint, you can set a condition when adding a catchpoint
on exception handlers.
Here the handlers catchpoint checks Global_Var:

(gdb) catch handlers Constraint_Error if Global_Var /= 0

gdb/ChangeLog:

        * ada-lang.h (ada_exception_catchpoint_kind) <ada_catch_handlers>:
        Add field.
        * ada-lang.c (struct exception_support_info) <catch_handlers_sym>:
        Add field.
        (default_exception_support_info) <catch_handlers_sym>: Add field.
        (exception_support_info_fallback) <catch_handlers_sym>: Add field.
        (ada_exception_name_addr_1): Add "catch handlers" handling.
        (ada_exception_catchpoint_cond_string) <ex>: New parameter.
        Update all callers.
        (create_excep_cond_exprs) <ex>: Add parameter.
        (re_set_exception): Update create_excep_cond_exprs call.
        (print_it_exception, print_one_exception, print_mention_exception)
        (print_recreate_exception): Add "catch handler" handling.
        (allocate_location_catch_handlers, re_set_catch_handlers)
        (check_status_catch_handlers, print_it_catch_handlers)
        (print_one_catch_handlers, print_mention_catch_handlers)
        (print_recreate_catch_handlers): New function.
        (catch_handlers_breakpoint_ops): New variable.
        (catch_ada_exception_command_split) <is_catch_handlers_cmd>:
        Add parameter.  Add "catch handler" handling.
        (ada_exception_sym_name, ada_exception_breakpoint_ops):
        Add "catch handler" handling.
        (ada_exception_catchpoint_cond_string): Add "catch handler"
        handling.
        (create_ada_exception_catchpoint): Update create_excep_cond_exprs
        call.
        (catch_ada_handlers_command): New function.
        (initialize_ada_catchpoint_ops): Initialize "catch handlers"
        operations structure.
        (_initialize_ada_language): Add "catch handlers" command entry.
        * NEWS: Document "catch handlers" feature.

gdb/doc/ChangeLog:

        * gdb.texinfo (Set Catchpoints): Add documentation for new
        "catch handlers" action.

gdb/testsuite/ChangeLog:

        * gdb.ada/excep_handle.exp: New testcase.
        * gdb.ada/excep_handle/foo.adb: New file.
        * gdb.ada/excep_handle/pck.ads: New file.

Tested on x86_64-linux.

6 years agoUpdate year range in copyright notice of binutils files
Alan Modra [Wed, 3 Jan 2018 05:17:27 +0000 (15:47 +1030)]
Update year range in copyright notice of binutils files

6 years agoChangeLog rotation
Alan Modra [Wed, 3 Jan 2018 05:15:17 +0000 (15:45 +1030)]
ChangeLog rotation

6 years agoUnbreak spurious fails in gdb.base/step-line.exp
Joel Brobecker [Wed, 3 Jan 2018 03:50:38 +0000 (07:50 +0400)]
Unbreak spurious fails in gdb.base/step-line.exp

My copyright year update in the copyright headers of the step-line
testcase caused it to start failing, due to shift in line numbers.

    FAIL: gdb.base/step-line.exp: continue to f1
    FAIL: gdb.base/step-line.exp: next to dummy 2
    FAIL: gdb.base/step-line.exp: next over dummy 2
    FAIL: gdb.base/step-line.exp: step into f2
    FAIL: gdb.base/step-line.exp: next over dummy 4
    FAIL: gdb.base/step-line.exp: next to dummy 5
    FAIL: gdb.base/step-line.exp: next to dummy 6
    FAIL: gdb.base/step-line.exp: next over dummy 6
    FAIL: gdb.base/step-line.exp: next to dummy 7
    FAIL: gdb.base/step-line.exp: next to dummy 8
    FAIL: gdb.base/step-line.exp: next over dummy 8
    FAIL: gdb.base/step-line.exp: next to dummy 9
    FAIL: gdb.base/step-line.exp: next to dummy 10
    FAIL: gdb.base/step-line.exp: next over dummy 10

This was caused by the copyright.py script after I changed it to
also update this testcase, which made the following transformation:

    -   Copyright (C) 2001-2017
    -   Free Software Foundation, Inc.
    +   Copyright (C) 2001-2018 Free Software Foundation, Inc.

We can easily unbreak the testcase without having to go back to
manual edits each year by adding an extra line in the header.

gdb/testsuite/ChangeLog:

* gdb.base/step-line.c: Add extra empty line in copyright header.
        * gdb.base/step-line.inp: Likewise.

Tested on x86_64-linux.

6 years agoAutomatic date update in version.in
GDB Administrator [Wed, 3 Jan 2018 00:00:31 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoFix compile time warning (in the ARM simulator) about a print statement with insuffic...
Nick Clifton [Tue, 2 Jan 2018 17:15:16 +0000 (17:15 +0000)]
Fix compile time warning (in the ARM simulator) about a print statement with insufficient arguments.

PR 22663
* maverick.c (DSPCDP4): Add missing parameter to debug print
statement.

6 years agoFix typo in do_mrs function in ARM assembler.
Nick Clifton [Tue, 2 Jan 2018 12:13:17 +0000 (12:13 +0000)]
Fix typo in do_mrs function in ARM assembler.

PR 18119
* config/tc-arm.c (do_mrs): Fix test of bits 16-19 in non-banked
version of ARM MRS instruction.

6 years agox86: partial revert of 10c17abdd0
Jan Beulich [Tue, 2 Jan 2018 10:44:04 +0000 (11:44 +0100)]
x86: partial revert of 10c17abdd0

Other than the variables in tc-i386.c using them, OPERAND_TYPE_REGYMM
and OPERAND_TYPE_REGZMM they aren't entirely unused. No need to update
i386-init.h though, as it mistakenly wasn't updated by the original
commit.

6 years ago[gdb/Ada] slices of arrays with dynamic strides
Joel Brobecker [Tue, 2 Jan 2018 03:53:55 +0000 (22:53 -0500)]
[gdb/Ada] slices of arrays with dynamic strides

    Consider the following Ada code:

       procedure Nested (L, U : Integer) is
          subtype Small_Type is Integer range L .. U;
          type Record_Type (I : Small_Type := L) 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"));

          procedure Discard (R : Record_Type) is
          begin
             null;
          end Discard;

       begin
          Discard (A1 (1));  -- STOP
       end;

Trying to print a slice of that array currently yields:

    (gdb) p a1(1..3)
    $1 = ((i => 0, s => ""), (i => 0, s => ""), (i => 0, s => ""))

We expected instead:

    (gdb) p a1(1..3)
    $1 = ((i => 0, s => ""), (i => 1, s => "A"), (i => 2, s => "AB"))

This is because the functions we use in ada-lang.c to create the type
of the array slice (ada_value_slice and ada_value_slice_from_ptr) was
not taking into account the stride of the array. This patch fixes this.

gdb/ChangeLog:

        * ada-lang.c (ada_value_slice_from_ptr): Take array stride into
        account when creating the array type of the slice.
        (ada_value_slice): Likewise.

gdb/testsuite/ChangeLog:

        * gdb.ada/dyn_stride.exp: Add slice test.

Note that, with the current use of ada_value_slice, the enhancement
to handle dynamic array strides seems unnecessary, because I do not
see how an array with a dynamic stride can be referenced by either
by reference or pointer. Since references are coerced to array pointers,
in both cases, the slice is performed by ada_value_slice_from_ptr.
But ada_value_slice is enhanced nonetheless, in the spirit of making
the code more robust, in case we missed something, and also as similar
as possible with its from_ptr counterpart.

tested on x86_64-linux.

6 years agoAdd support for dynamic DW_AT_byte_stride.
Joel Brobecker [Tue, 2 Jan 2018 03:47:18 +0000 (22:47 -0500)]
Add support for dynamic DW_AT_byte_stride.

This patch adds support for DW_AT_byte_stride, using Ada as one
example of where this would be useful. However, the implementation
is language-agnostic.

Consider the following Ada code:

   procedure Nested (L, U : Integer) is
      subtype Small_Type is Integer range L .. U;
      type Record_Type (I : Small_Type := L) 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"));

      procedure Discard (R : Record_Type) is
      begin
         null;
      end Discard;

   begin
      Discard (A1 (1));  -- STOP
   end;

It defines an array A1 of Record_Type, which is a variant record
type whose maximum size actually depends on the value of the
parameters passed when calling Nested. As a result, the stride
of the array A1 cannot be known statically, which leads the compiler
to generate a dynamic DW_AT_byte_stride attribute for our type.
Here is what the debugging info looks like with GNAT:

        .uleb128 0x10   # (DIE (0x14e) DW_TAG_array_type)
        .long   .LASF17 # DW_AT_name: "foo__nested__T18b"
        .long   0x141   # DW_AT_byte_stride
        .long   0xdc    # DW_AT_type
        .uleb128 0x11   # (DIE (0x15f) DW_TAG_subrange_type)
        .long   0x166   # DW_AT_type
        .byte   0x3     # DW_AT_upper_bound
        .byte   0       # end of children of DIE 0x14e

There DW_AT_byte_stride is a reference to a local (internal)
variable:

        .uleb128 0x9    # (DIE (0x141) DW_TAG_variable)
        .long   .LASF6  # DW_AT_name: "foo__nested__T18b___PAD___XVZ"

This patch enhances GDB to handle this dynamic byte stride attribute
by first adding a new dynamic_prop_node_kind (DYN_PROP_BYTE_STRIDE)
to store the array dynamic stride info (when dynamic). It then enhances
the dynamic type resolver to handle this dynamic property.

Before applying this patch, trying to print the value of some of
A1's elements after having stopped at the "STOP" comment does not
work. For instance:

    (gdb) p a1(2)
    Cannot access memory at address 0x80000268dec0

With this patch applied, GDB now prints the value of all 3 elements
correctly:

    (gdb) print A1(1)
    $1 = (i => 0, s => "")
    (gdb) print A1(2)
    $2 = (i => 1, s => "A")
    (gdb) print A1(3)
    $3 = (i => 2, s => "AB")

gdb/ChangeLog:

        * gdbtypes.h (enum dynamic_prop_node_kind) <DYN_PROP_BYTE_STRIDE>:
        New enum value.
        (create_array_type_with_stride): Add byte_stride_prop parameter.
        * gdbtypes.c (create_array_type_with_stride) <byte_stride_prop>:
        New parameter.  Update all callers in this file.
        (array_type_has_dynamic_stride): New function.
        (is_dynamic_type_internal, resolve_dynamic_array): Add handling
        of arrays with dynamic byte strides.
        * dwarf2read.c (read_array_type): Add support for dynamic
        DW_AT_byte_stride attributes.

gdb/testsuite/ChangeLog:

        * gdb.ada/dyn_stride: New testcase.

Tested on x86_64-linux.

6 years agotreat Ada DW_TAG_unspecified_type DIEs as stub types
Joel Brobecker [Tue, 2 Jan 2018 03:41:36 +0000 (22:41 -0500)]
treat Ada DW_TAG_unspecified_type DIEs as stub types

Consider the gdb.ada/taft_type.exp testcase, which exercises
the situation where a variable is defined using a type which
is a pointer to an incomplete type, with the actual type
definition being provided by another unit. Up to now, the
strategy used by GNAT when generating the DWARF debugging info
was to produce a incomplete DW_TAG_enumeration_type DIE with
a DW_AT_declaration flag attached to it:

        .uleb128 0x4    # (DIE (0x3e) DW_TAG_enumeration_type)
        .long   .LASF4  # DW_AT_name: "pck__empty"
                        # DW_AT_declaration

However, a more standard way for the compiler to describe
this kind of type is to use the DW_TAG_unspecified_type tag.
When the compiler is enhanced to do so, we'll need to treat
such types as stubs -- we only do so with types from Ada units,
however, as the meaning of this TAG is intentionally left
permissive and language-specific by the DWARF standard.

Without this patch, running the testcase above with an enhanced
compiler now yields:

    (gdb) print w.e.all
    Attempt to dereference a generic pointer.
    FAIL: gdb.ada/taft_type.exp: print w.e.all

gdb/ChangeLog:

* dwarf2read.c (read_unspecified_type): Treat
        DW_TAG_enumeration_type DIEs from Ada units as stubs.

Tested on x86_64-linux, fixes the FAIL in gdb.ada/taft_type.exp above.

6 years agoUpdate copyright year range in all GDB files
Joel Brobecker [Mon, 1 Jan 2018 04:43:02 +0000 (08:43 +0400)]
Update copyright year range in all GDB files

gdb/ChangeLog:

        Update copyright year range in all GDB files

6 years agogdb/copyright.py: Remove testsuite/gdb.base/step-line.{c,inp} special handling
Joel Brobecker [Mon, 1 Jan 2018 04:18:16 +0000 (08:18 +0400)]
gdb/copyright.py: Remove testsuite/gdb.base/step-line.{c,inp} special handling

In the past, these files needed to be handled by hand, because
the testcase was sensitive to the length of the header, which was
potentially changing when new copyright years were added to
the copyright header. Now that we simply maintain and update
a range, the length of the copyright header should not change
as a consequence of the update performed by this script, so
special handling of those files is no longer necessary.

gdb/ChangeLog:

        * copyright.py (BY_HAND): Remove gdb/testsuite/gdb.base/step-line.inp
        and gdb/testsuite/gdb.base/step-line.c.

6 years agogdb/copyright.py: Do not forget to remind about MULTIPLE_COPYRIGHT_HEADERS
Joel Brobecker [Mon, 1 Jan 2018 04:30:28 +0000 (08:30 +0400)]
gdb/copyright.py: Do not forget to remind about MULTIPLE_COPYRIGHT_HEADERS

There is a small logical error in the part of the script that
dumps the list of files in BY_HAND + MULTIPLE_COPYRIGHT_HEADERS
but only checkis the contents of BY_HAND. The issue becomes
apparent as soon as BY_HAND is empty. Prevent this from happening
by treating the two lists separately, as this allows us to provide
a more informative message in the case of MULTIPLE_COPYRIGHT_HEADERS.

gdb/ChangeLog:

        * copyright.py (main): Dump the contents of
        MULTIPLE_COPYRIGHT_HEADERS (separately) from BY_HAND,
        even if BY_HAND is empty.

6 years agoUpdate copyright year in version message of GDB, GDBserver and GDBreplay
Joel Brobecker [Mon, 1 Jan 2018 04:12:21 +0000 (08:12 +0400)]
Update copyright year in version message of GDB, GDBserver and GDBreplay

gdb/ChangeLog:

* top.c (print_gdb_version): Update Copyright year in version
message.

gdb/gdbserver/ChangeLog:

* gdbreplay.c (gdbreplay_version): Update copyright year in
version message.
* server.c (gdbserver_version): Likewise.

6 years agoYearly rotation of the gdb/ChangeLog file
Joel Brobecker [Mon, 1 Jan 2018 04:07:25 +0000 (08:07 +0400)]
Yearly rotation of the gdb/ChangeLog file

gdb/ChangeLog

        * config/djgpp/fnchange.lst: Add entry for gdb/ChangeLog-2017.

6 years agoAutomatic date update in version.in
GDB Administrator [Tue, 2 Jan 2018 00:00:32 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Mon, 1 Jan 2018 00:00:38 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAvoid indexing std::vector past the end
Ruslan Kabatsayev [Sat, 30 Dec 2017 19:14:41 +0000 (22:14 +0300)]
Avoid indexing std::vector past the end

The code here wants to find address of an element, and often this
element is one past the end of std::vector. Dereferencing that element
leads to undefined behavior, so it's better to simply use pointer
arithmetic instead of taking address of invalid dereference.

gdb/ChangeLog:

* psymtab.c (recursively_search_psymtabs): Use pointer arithmetic
instead of dereferencing std::vector past the end.

6 years agoOnly ignore -Wenum-compare-switch if it exists
Simon Marchi [Sun, 31 Dec 2017 04:33:26 +0000 (23:33 -0500)]
Only ignore -Wenum-compare-switch if it exists

My patch

  dwarf2read: Silence -Wenum-compare-switch warning
  132448f8359a268f34f074b0908b5255b568da06

made some parts of dwarf2read.c ignore warnings about switch using enums
of different kinds.  What I did not realize was that older Clang
versions (prior to 6) did not have that warning, and therefore give this
error:

/home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:24187:7: error: unknown warning group '-Wenum-compare-switch', ignored [-Werror,-Wunknown-pragmas]
      DIAGNOSTIC_IGNORE_SWITCH_DIFFERENT_ENUM_TYPES
      ^
/home/emaisin/src/binutils-gdb/gdb/common/diagnostics.h:42:3: note: expanded from macro 'DIAGNOSTIC_IGNORE_SWITCH_DIFFERENT_ENUM_TYPES'
  DIAGNOSTIC_IGNORE ("-Wenum-compare-switch")
  ^
/home/emaisin/src/binutils-gdb/gdb/common/diagnostics.h:27:3: note: expanded from macro 'DIAGNOSTIC_IGNORE'
  _Pragma (STRINGIFY (GCC diagnostic ignored option))
  ^
<scratch space>:10:25: note: expanded from here
 GCC diagnostic ignored "-Wenum-compare-switch"
                        ^

Clang has a way to test if it knows about a particular warning.  This
patch uses that feature to only define
DIAGNOSTIC_IGNORE_SWITCH_DIFFERENT_ENUM_TYPES to something if the
warning is recognized by the Clang version being used.  I tested
building dwarf2read.c with clang 4, 5, 6, as well as gcc.

gdb/ChangeLog:

* common/diagnostics.h
(DIAGNOSTIC_IGNORE_SWITCH_DIFFERENT_ENUM_TYPES): Only
define if the compiler knows about -Wenum-compare-switch.

6 years agodwarf2read: Silence -Wenum-compare-switch warning
Simon Marchi [Sun, 31 Dec 2017 02:24:29 +0000 (21:24 -0500)]
dwarf2read: Silence -Wenum-compare-switch warning

Compiling with Clang 6 gives:

/home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:24385:14: error: comparison of two values with different enumeration types in switch statement ('enum dwarf_macro_record_type' and 'dwarf_macinfo_record_type') [-Werror,-Wenum-compare-switch]
        case DW_MACINFO_vendor_ext:
             ^~~~~~~~~~~~~~~~~~~~~
/home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:24561:7: error: comparison of two values with different enumeration types in switch statement ('enum dwarf_macro_record_type' and 'dwarf_macinfo_record_type') [-Werror,-Wenum-compare-switch]
        case DW_MACINFO_vendor_ext:
             ^~~~~~~~~~~~~~~~~~~~~

This code uses the two enum types on purpose, because it handles both
.debug_macro and .debug_macinfo sections.  Add some pragmas to disable
the warning in these specific cases.

gdb/ChangeLog:

* dwarf2read.c (dwarf_decode_macro_bytes): Ignore
-Wenum-compare-switch warning.
(dwarf_decode_macros): Likewise.

6 years agoC++-ify parser_state
Tom Tromey [Thu, 23 Nov 2017 04:45:53 +0000 (21:45 -0700)]
C++-ify parser_state

This mildly C++-ifies parser_state and stap_parse_info -- just enough
to remove some cleanups.

This version includes the changes implemented by Simon.

Regression tested by the buildbot.

gdb/ChangeLog
2017-12-30  Tom Tromey  <tom@tromey.com>
    Simon Marchi  <simon.marchi@ericsson.com>

* stap-probe.h (struct stap_parse_info): Add constructor,
destructor.
* stap-probe.c (stap_parse_argument): Update.
* rust-exp.y (rust_lex_tests): Update.
* parser-defs.h (struct parser_state): Add constructor,
destructor, release method.
<expout>: Change type to expression_up.
(null_post_parser): Change type.
(initialize_expout, reallocate_expout): Remove.
* parse.c (parser_state::parser_state): Rename from
initialize_expout.
(parser_state::release): Rename from reallocate_expout.
(write_exp_elt, parse_exp_in_context_1, increase_expout_size):
Update.
(null_post_parser): Change type of "exp".
* dtrace-probe.c (dtrace_probe::build_arg_exprs): Update.
* ada-lang.c (resolve, resolve_subexp)
(replace_operator_with_call): Change type of "expp".
* language.h (struct language_defn) <la_post_parser>: Change type
of "expp".

6 years agoAutomatic date update in version.in
GDB Administrator [Sun, 31 Dec 2017 00:00:27 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoMake mapped_debug_names and mapped_index final
Simon Marchi [Sat, 30 Dec 2017 05:10:05 +0000 (00:10 -0500)]
Make mapped_debug_names and mapped_index final

When compiling with Clang 6, I see these warnings:

/home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:25421:5: error: destructor called on non-final 'mapped_index' that has virtual functions but non-virtual destructor [-Werror,-Wdelete-non-virtual-dtor]
    data->index_table->~mapped_index ();
    ^
In file included from /home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:31:
In file included from /home/emaisin/src/binutils-gdb/gdb/defs.h:28:
In file included from /home/emaisin/src/binutils-gdb/gdb/common/common-defs.h:92:
In file included from /home/emaisin/src/binutils-gdb/gdb/common/gdb_unique_ptr.h:23:
In file included from /usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/memory:81:
/usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/unique_ptr.h:76:2: error: delete called on non-final 'mapped_debug_names' that has virtual functions but non-virtual destructor [-Werror,-Wdelete-non-virtual-dtor]
        delete __ptr;
        ^
/usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/unique_ptr.h:236:4: note: in instantiation of member function 'std::default_delete<mapped_debug_names>::operator()' requested here
          get_deleter()(__ptr);
          ^
/home/emaisin/src/binutils-gdb/gdb/dwarf2read.c:2374:21: note: in instantiation of member function 'std::unique_ptr<mapped_debug_names, std::default_delete<mapped_debug_names> >::~unique_ptr' requested here
dwarf2_per_objfile::dwarf2_per_objfile (struct objfile *objfile_,
                    ^

This patch silences these warnings by making the classes final.

gdb/ChangeLog:

* dwarf2read.c (struct mapped_debug_names): Make final.
(struct mapped_index): Make final.

6 years agoIgnore warning about using different types of enums in switch
Simon Marchi [Sat, 30 Dec 2017 04:32:13 +0000 (23:32 -0500)]
Ignore warning about using different types of enums in switch

When compiling with clang 6, I see a bunch of warnings like this:

/home/emaisin/src/binutils-gdb/gdb/amd64-linux-tdep.c:1427:8: error: comparison of two values with different enumeration types in switch statement ('enum amd64_syscall' and 'amd
64_x32_syscall') [-Werror,-Wenum-compare-switch]
  case amd64_x32_sys_move_pages:
       ^~~~~~~~~~~~~~~~~~~~~~~~

In this switch, we indeed use enumerators of both types
amd64_x32_syscall and amd64_syscall.  This is done on purpose, and the
enum values are chosen so that they are complementary.

I think it's still a useful warning, so I chose to ignore just that
particular case.

gdb/ChangeLog:

* common/diagnostics.h
(DIAGNOSTIC_IGNORE_SWITCH_DIFFERENT_ENUM_TYPES): New macro.
* amd64-linux-tdep.c (amd64_canonicalize_syscall): Use it.

6 years agoRemove unnecessary call to get_thread_db_info
Simon Marchi [Sat, 30 Dec 2017 02:56:58 +0000 (21:56 -0500)]
Remove unnecessary call to get_thread_db_info

In thread_db_detach, we call get_thread_db_info to first check if there
exists a thread_db_info entry for the pid to detach.  If there is, then
we call delete_thread_db_info.  It's unnecessary to call
get_thread_db_info in the first place, since delete_thread_db_info
handles the case where no thread_db_info entry exist for the given pid.

gdb/ChangeLog:

* linux-thread-db.c (thread_db_detach): Remove call to
delete_thread_db_info.

6 years agoAutomatic date update in version.in
GDB Administrator [Sat, 30 Dec 2017 00:00:30 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Fri, 29 Dec 2017 00:00:34 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoRISC-V: Add missing privileged spec registers.
Jim Wilson [Thu, 28 Dec 2017 21:21:46 +0000 (13:21 -0800)]
RISC-V: Add missing privileged spec registers.

gas/
* testsuite/gas/riscv/priv-reg.d, testsuite/gas/riscv/priv-reg.s: New.

include/
* opcode/riscv-opc.h (DECLARE_CSR): Add missing privileged registers.
Sort to match privileged spec documentation order.
(DECLARE_CSR_ALIAS): Add ubadaddr, and comments.

6 years agoRemove unused HP-UX TARGET_OBJECT_ enums
Simon Marchi [Thu, 28 Dec 2017 17:29:33 +0000 (12:29 -0500)]
Remove unused HP-UX TARGET_OBJECT_ enums

These two enumerators are unused, remove them.

gdb/ChangeLog:

* target.h (enum target_object) <TARGET_OBJECT_HPUX_UREGS,
TARGET_OBJECT_HPUX_SOLIB_GOT>: Remove.

6 years agoAutomatic date update in version.in
GDB Administrator [Thu, 28 Dec 2017 00:00:21 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agotdesc: handle arbitrary strings in tdesc_register_in_reggroup_p
Stafford Horne [Wed, 27 Dec 2017 14:36:31 +0000 (23:36 +0900)]
tdesc: handle arbitrary strings in tdesc_register_in_reggroup_p

tdesc_register_in_reggroup_p in now able to handle arbitrary
groups. This is useful when groups are created while the
target descriptor file is received from the remote.

This can be the case of a soft core target processor where
registers/groups can change.

gdb/ChangeLog:

yyyy-mm-dd  Franck Jullien  <franck.jullien@gmail.com>
    Stafford Horne  <shorne@gmail.com>

* target-descriptions.c (tdesc_register_in_reggroup_p): Support
arbitrary strings.
(tdesc_use_registers): Add logic to register new reggroups.
(tdesc_reg::group): Update comment to indicate we allow
arbitrary strings.
* NEWS (Changes since GDB 8.0): Announce that GDB supports
arbitrary reggroups.

gdb/testsuite/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* gdb.xml/extra-regs.xml: Add example foo reggroup.
* gdb.xml/tdesc-regs.exp: Add test to check for foo reggroup.

gdb/doc/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* gdb.texinfo (Target Description Format): Explain that arbitrary
strings are now allowed for register groups.

6 years agoreggroups: Add reggroup_gdbarch_new, reggroup_find for dynamic reggroups
Stafford Horne [Wed, 27 Dec 2017 14:36:31 +0000 (23:36 +0900)]
reggroups: Add reggroup_gdbarch_new, reggroup_find for dynamic reggroups

Traditionally reggroups have been created via reggroup_new() during
initialization code and never freed.  Now, if we want to initialize
reggroups dynamically (i.e. in target description) we should be able to
free them.  Create this function reggroup_gdbarch_new() which will
allocate the reggroup memory onto the passed gdbarch obstack.

Also creating reggroup_find() as a utility to find a gdbarch registered
reggroup object by name.

gdb/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* reggroups.c (reggroup_gdbarch_new): New function.
(reggroup_find): New function.
* reggroups.h (reggroup_gdbarch_new): New function.
(reggroup_find): New function.

6 years agoreggroups: Convert reggroups from post_init to pre_init
Stafford Horne [Wed, 27 Dec 2017 14:36:30 +0000 (23:36 +0900)]
reggroups: Convert reggroups from post_init to pre_init

Currently the reggroups gdbarch_data cannot be manipulated until after
the gdbarch is completely initialized.  This is usually done when the
object init depends on architecture specific fields.  In the case of
reggroups it only depends on the obstack being available.

Coverting this to pre_init allows using reggroups during gdbarch
initialization.  This is needed to allow registering arbitrary reggroups
during gdbarch initializations.

gdb/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* reggroups.c (reggroups_init): Change to depend only on
obstack rather than gdbarch.
(reggroup_add): Remove logic for forcing premature init.
(_initialize_reggroup): Set `reggroups_data` with
gdbarch_data_register_pre_init() rather than
gdbarch_data_register_post_init().

6 years agoreggroups: Add test and docs for `info reg $reggroup` feature
Stafford Horne [Wed, 27 Dec 2017 14:36:30 +0000 (23:36 +0900)]
reggroups: Add test and docs for `info reg $reggroup` feature

Until now this feature has existed but was not documented.  Adding docs
and tests.

gdb/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* infcmd.c (_initialize_infcmd): Add help for info reg $reggroup
and info all-registers $reggroup feature.

gdb/doc/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* gdb.texinfo (Registers): Document info reg $reggroup feature.

gdb/testsuite/ChangeLog:

yyyy-mm-dd  Stafford Horne  <shorne@gmail.com>

* gdb.base/reggroups.c: New file.
* gdb.base/reggroups.exp: New file.

6 years agoAutomatic date update in version.in
GDB Administrator [Wed, 27 Dec 2017 00:00:21 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Tue, 26 Dec 2017 00:00:28 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Mon, 25 Dec 2017 00:00:30 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Sun, 24 Dec 2017 00:00:39 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAdd ATTRIBUTE_PRINTF to printf_field_type_assignment
Simon Marchi [Sat, 23 Dec 2017 21:25:06 +0000 (16:25 -0500)]
Add ATTRIBUTE_PRINTF to printf_field_type_assignment

Get rid of:

/home/simark/src/binutils-gdb/gdb/target-descriptions.c:2026:25: error: format string is not a string literal [-Werror,-Wformat-nonliteral]
    vprintf_unfiltered (fmt, args);

when building with clang.

gdb/ChangeLog:

* target-descriptions.c (print_c_tdesc)
<printf_field_type_assignment>: Add ATTRIBUTE_PRINTF.

6 years agoAutomatic date update in version.in
GDB Administrator [Sat, 23 Dec 2017 00:00:31 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Fri, 22 Dec 2017 00:00:31 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoDo not emit "field_type" var if not needed on "maint print c-tdesc"
Simon Marchi [Thu, 21 Dec 2017 23:19:17 +0000 (18:19 -0500)]
Do not emit "field_type" var if not needed on "maint print c-tdesc"

While fiddling a bit with -Wunused-variable, Sergio noticed that "maint
print c-tdesc" was always generating code for the "tdesc_type
*field_type" variable, even when it wasn't used.  This is caught by GCC
when using -Wunused-variable, of course.

This patch changes the print_c_tdesc class to only output the field
declaration when we actually need it.

It shouldn't be necessary to do the same with the other variable
declarations (type_with_fields and element_type), because they are
always if they are declared.

The C files in features/ are regenerated, some declarations of
field_type are removed, as expected, while some others move to where
they are used for the first time.

gdb/ChangeLog:

* target-descriptions.c (print_c_tdesc) <visit>: Don't output
field_type declaration, use printf_field_type_assignment
instead.
<printf_field_type_assignment>: New method.
* features/aarch64-core.c, features/aarch64-fpu.c
features/arc-arcompact.c, features/arc-v2.c,
features/arm/arm-with-iwmmxt.c, features/i386/32bit-core.c,
features/i386/32bit-mpx.c, features/i386/32bit-sse.c,
features/i386/64bit-avx512.c, features/i386/64bit-core.c,
features/i386/64bit-mpx.c, features/i386/64bit-sse.c,
features/i386/x32-core.c, features/or1k.c,
features/rs6000/powerpc-7400.c,
features/rs6000/powerpc-altivec32.c,
features/rs6000/powerpc-altivec32l.c,
features/rs6000/powerpc-altivec64.c,
features/rs6000/powerpc-altivec64l.c,
features/rs6000/powerpc-cell32l.c,
features/rs6000/powerpc-cell64l.c,
features/rs6000/powerpc-isa205-altivec32l.c,
features/rs6000/powerpc-isa205-altivec64l.c,
features/rs6000/powerpc-isa205-vsx32l.c,
features/rs6000/powerpc-isa205-vsx64l.c,
features/rs6000/powerpc-vsx32.c,
features/rs6000/powerpc-vsx32l.c,
features/rs6000/powerpc-vsx64.c,
features/rs6000/powerpc-vsx64l.c, features/s390-gs-linux64.c,
features/s390-tevx-linux64.c, features/s390-vx-linux64.c,
features/s390x-gs-linux64.c, features/s390x-tevx-linux64.c,
features/s390x-vx-linux64.c: Re-generate.

6 years agoRemove write-only assignment in rs6000-tdep.c
Uros Bizjak [Thu, 21 Dec 2017 01:42:05 +0000 (20:42 -0500)]
Remove write-only assignment in rs6000-tdep.c

The result of this memory read is never used, so it can be removed.

gdb/ChangeLog:

* rs6000-tdep.c (ppc_deal_with_atomic_sequence): Remove
write-only assignment to "insn" variable.

6 years agoAutomatic date update in version.in
GDB Administrator [Thu, 21 Dec 2017 00:00:39 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoRISC-V: Add compressed instruction hints, and a few misc cleanups.
Jim Wilson [Wed, 20 Dec 2017 21:37:44 +0000 (13:37 -0800)]
RISC-V: Add compressed instruction hints, and a few misc cleanups.

gas/
* config/tc-riscv.c (risc_ip) <o>: Add comment.
* testsuite/gas/riscv/c-nonzero-imm.d,
* testsuite/gas/riscv/c-nonzero-imm.l,
* testsuite/gas/riscv/c-nonzero-imm.s,
* testsuite/gas/riscv/c-nonzero-reg.d,
* testsuite/gas/riscv/c-nonzero-reg.l,
* testsuite/gas/riscv/c-nonzero-reg.s,
* testsuite/gas/riscv/c-zero-imm-64.d,
* testsuite/gas/riscv/c-zero-imm-64.s,
* testsuite/gas/riscv/c-zero-imm.d, testsuite/gas/riscv/c-zero-imm.s,
* testsuite/gas/riscv/c-zero-reg.d,
* testsuite/gas/riscv/c-zero-reg.s: New.

opcodes/
* riscv-opc.c (match_c_add_with_hint, match_c_lui_with_hint): New.
(riscv_opcodes) <li>: Delete "d,0" line.  Change Cj to Co.
<andi, and, add, addiw, addw, c.addi>: Change Cj to Co.
<add>: Add explanatory comment for 4-operand add instruction.
<c.nop>: Add support for immediate operand.
<c.mv, c.add>: Use match_c_add_with_hint instead of match_c_add.
<c.lui>: Use match_c_lui_with_hint instead of match_c_lui.
<c.li, c.slli>: Use match_opcode instead of match_rd_nonzero.

6 years ago[Cell/B.E.] Fix regression due to gdbarch_significant_addr_bit
Ulrich Weigand [Wed, 20 Dec 2017 12:59:14 +0000 (13:59 +0100)]
[Cell/B.E.] Fix regression due to gdbarch_significant_addr_bit

On Cell/B.E. multi-architecture debugging we use a "merged" address space
that encodes both the main PowerPC address space and the local store address
spaces of all active SPUs.  This will always occupy 64 bits.

However, gdbarch_addr_bit is set to 32 on SPU, and may be set to 32 as well
on PowerPC.  Since the new gdbarch_significant_addr_bit defaults to the
value of gdbarch_addr_bit, this means addresses may be improperly truncated.

Work around this problem by explicitly setting gdbarch_significant_addr_bit
to 64 both for the SPU target and also for PowerPC target that support
Cell/B.E. execution.

gdb/ChangeLog:
2017-12-20  Ulrich Weigand  <uweigand@de.ibm.com>

* spu-tdep.c (spu_gdbarch_init): Set set_gdbarch_significant_addr_bit
to 64 bits.
(ppc_linux_init_abi): Likewise, if Cell/B.E. is supported.

6 years agoFix ChangeLog formatting
Simon Marchi [Wed, 20 Dec 2017 03:09:40 +0000 (22:09 -0500)]
Fix ChangeLog formatting

Doing some unrelated grepping found that there were some missing
spaces, fix it.

6 years agoAutomatic date update in version.in
GDB Administrator [Wed, 20 Dec 2017 00:00:21 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoCorrect disassembly of dot product instructions.
Tamar Christina [Tue, 19 Dec 2017 12:05:20 +0000 (12:05 +0000)]
Correct disassembly of dot product instructions.

Dot products deviate from the normal disassembly rules for lane indexed
instruction. Their canonical representation is in the form of:

v0.2s, v0.8b, v0.4b[0] instead of v0.2s, v0.8b, v0.b[0] to try to denote
that these instructions select 4x 1 byte elements instead of a single 1 byte
element.

Previously we were disassembling them following the normal rules, this patch
corrects the disassembly.

gas/

PR gas/22559
* config/tc-aarch64.c (vectype_to_qualifier): Support AARCH64_OPND_QLF_S_4B.
* gas/testsuite/gas/aarch64/dotproduct.d: Update disassembly.

include/

PR gas/22559
* aarch64.h (aarch64_opnd_qualifier): Add AARCH64_OPND_QLF_S_4B.

opcodes/

PR gas/22559
* aarch64-asm.c (aarch64_ins_reglane): Change AARCH64_OPND_QLF_S_B to
AARCH64_OPND_QLF_S_4B
* aarch64-dis.c (aarch64_ext_reglane): Change AARCH64_OPND_QLF_S_B to
AARCH64_OPND_QLF_S_4B
* aarch64-opc.c (aarch64_opnd_qualifiers): Add 4b variant.
* aarch64-tbl.h (QL_V2DOT): Change S_B to S_4B.

6 years agoAdd support for V_4B so we can properly reject it.
Tamar Christina [Tue, 19 Dec 2017 12:04:13 +0000 (12:04 +0000)]
Add support for V_4B so we can properly reject it.

Previously parse_vector_type_for_operand was changed to allow the use of 4b
register size for indexed lane instructions. However this had the unintended
side effect of also allowing 4b for normal vector registers.

Because this support was only partial the rest of the tool silently treated
4b as 8b and continued. This patch adds full support for 4b so it can be
properly distinguished from 8b and the correct errors are generated.

With this patch you still can't encode any instruction which actually requires
v<num>.4b but such instructions don't exist so to prevent needing a workaround
in get_vreg_qualifier_from_value this was just omitted.

gas/

PR gas/22529
* config/tc-aarch64.c (vectype_to_qualifier): Support AARCH64_OPND_QLF_V_4B.
* gas/testsuite/gas/aarch64/pr22529.s: New.
* gas/testsuite/gas/aarch64/pr22529.d: New.
* gas/testsuite/gas/aarch64/pr22529.l: New.

include/

PR gas/22529
* opcode/aarch64.h (aarch64_opnd_qualifier): Add AARCH64_OPND_QLF_V_4B.

opcodes/

PR gas/22529
* aarch64-opc.c (aarch64_opnd_qualifiers): Add 4b variant.

6 years agoStop readelf from complaining about relocation sections with an sh_info field of...
Nick Clifton [Tue, 19 Dec 2017 09:48:40 +0000 (09:48 +0000)]
Stop readelf from complaining about relocation sections with an sh_info field of 0 in dynamic executables.

PR 22587
* readelf.c (process_section_headers): Do not complain about an
sh_info field of 0 in relocation sections of ET_EXEC or ET_DYN
type executables.

6 years agoAutomatic date update in version.in
GDB Administrator [Tue, 19 Dec 2017 00:00:18 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoPR22626, invalid dynindx used for dynamic relocs against section syms
Alan Modra [Mon, 18 Dec 2017 11:57:08 +0000 (22:27 +1030)]
PR22626, invalid dynindx used for dynamic relocs against section syms

_bfd_elf_link_renumber_dynsyms is called twice by the linker.  The
first call in bfd_elf_size_dynamic_sections is just to answer the
question as to whether there are there any dynamic symbols.  The
second call in bfd_elf_size_dynsym_hash_dynstr sets the st_shndx value
that dynamic symbols will have.  strip_excluded_output_sections is
called between these two calls.  So sections seen on the first
_bfd_elf_link_renumber_dynsyms pass might differ from those seen on
the second pass.  Unfortunately, that can result in a stripped
section's dynamic symbol being assigned a dynindx on the first pass
but not corrected to the final value (of zero, ie. not dynamic) on the
second pass.  PowerPC, x86, mips, and most other targets that emit
dynamic section symbols, just test that section symbol dynindx is
non-zero before using a given section symbol in dynamic relocations.

This patch prevents _bfd_elf_link_renumber_dynsyms from setting any
section symbol dynindx on the first pass.

PR 22626
* elflink.c (_bfd_elf_link_renumber_dynsyms): Don't set section
dynindx when section_sym_count is NULL.
(bfd_elf_size_dynamic_sections): Pass NULL section_sym_count to
preliminary _bfd_elf_link_renumber_dynsyms call.

6 years agoTest binutils_assemble return value
Alan Modra [Fri, 15 Dec 2017 06:38:56 +0000 (17:08 +1030)]
Test binutils_assemble return value

This is a followup to "binutils nm testsuite tidy".  Since the perror
in binutils_assemble has been removed, we need to take more care in
failure paths.

The patch also fixed a number of .exp files that have multiple tests,
where an assembly failure returns from the .exp file.  In most cases
it is nicer to attempt all tests.

* testsuite/binutils-all/ar.exp (unique_symbol): Don't run AR
if assembly fails.
* testsuite/binutils-all/arc/objdump.exp (check_assembly): If
objfile is empty, fail test.
* testsuite/binutils-all/arm/objdump.exp: Don't return if assembly
fails for a test, continue on to other tests.
* testsuite/binutils-all/bfin/objdump.exp: Likewise.
* testsuite/binutils-all/hppa/objdump.exp: Likewise.
* testsuite/binutils-all/m68k/objdump.exp: Likewise.
* testsuite/binutils-all/vax/objdump.exp: Likewise.
* testsuite/binutils-all/size.exp: Likewise.
* testsuite/binutils-all/nm.exp: Likewise.  Move PR12753 test.
* testsuite/binutils-all/objcopy.exp: Don't perror on assembly fail.
* testsuite/binutils-all/objdump.exp: Report assembly fails.

6 years agoDecode expression lists used with DW_AT_byte_size and other DWARF attributes.
Nick Clifton [Mon, 18 Dec 2017 17:57:18 +0000 (17:57 +0000)]
Decode expression lists used with DW_AT_byte_size and other DWARF attributes.

PR 22532
* dwarf.c (read_and_display_attr_value): Add attributes that might
use the DW_FORM_exprloc form, and if so, display the decoded
location expression list.

6 years agoResolve PR 22493 - the encoding to be used when pushing the stack pointer onto the...
Nick Clifton [Mon, 18 Dec 2017 09:37:57 +0000 (09:37 +0000)]
Resolve PR 22493 - the encoding to be used when pushing the stack pointer onto the stack.

PR 22493
* config/tc-arm.c (encode_ldmstm): Do not use A2 encoding of the
PUSH insn when pushing the stack pointer.

6 years agox86: fold certain AVX and AVX2 templates
Jan Beulich [Mon, 18 Dec 2017 08:37:15 +0000 (09:37 +0100)]
x86: fold certain AVX and AVX2 templates

Just like for instructions in GPRs, there's no need to have separate
templates for otherwise identical insns acting on XMM or YMM registers
(or memory of the same size).

6 years agox86: fold RegXMM/RegYMM/RegZMM into RegSIMD
Jan Beulich [Mon, 18 Dec 2017 08:36:14 +0000 (09:36 +0100)]
x86: fold RegXMM/RegYMM/RegZMM into RegSIMD

... qualified by their respective sizes, allowing to drop FirstXmm0 at
the same time.

6 years agox86: drop FloatReg and FloatAcc
Jan Beulich [Mon, 18 Dec 2017 08:35:01 +0000 (09:35 +0100)]
x86: drop FloatReg and FloatAcc

Express them as Reg|Tbyte and Acc|Tbyte respectively.

6 years agox86: replace Reg8, Reg16, Reg32, and Reg64
Jan Beulich [Mon, 18 Dec 2017 08:34:00 +0000 (09:34 +0100)]
x86: replace Reg8, Reg16, Reg32, and Reg64

Use a combination of a single new Reg bit and Byte, Word, Dword, or
Qword instead.

Besides shrinking the number of operand type bits this has the benefit
of making register handling more similar to accumulator handling (a
generic flag is being accompanied by a "size qualifier"). It requires,
however, to split a few insn templates, as it is no longer correct to
have combinations like Reg32|Reg64|Byte. This slight growth in size will
hopefully be outweighed by this change paving the road for folding a
presumably much larger number of templates later on.

6 years agoimproved error message when getting an exception printing a variable
Joel Brobecker [Mon, 18 Dec 2017 03:39:33 +0000 (22:39 -0500)]
improved error message when getting an exception printing a variable

Consider the following Ada code defining a global variable whose
type is an array of static bounds (1 .. 2), but where its elements
are a variant record whose size is not statically known:

    type Ints is array (Natural range <>) of Integer;
    type Bounded_Ints (Max_Size : Natural) is record
       Length : Natural := 0;
       Objs   : Ints (1 .. Max_Size);
    end record;

    type Ints_Doubled is array (1 .. 2) of Bounded_Ints (Idem (0));

    Global : Ints_Doubled;

When compiling this program at -O2 using a GCC-6.4-based compiler
on x86_64-linux, trying to print the value of that global variable
yields:

    (gdb) p global
    $1 =

Let's look at the debugging info, which starts with the global
variable itself...

        .uleb128 0x19   # (DIE (0x25e) DW_TAG_variable)
        .long   .LASF32 # DW_AT_name: "fd__global"
        .long   0x273   # DW_AT_type

... its type is a reference to a typedef ...

        .uleb128 0x14   # (DIE (0x273) DW_TAG_reference_type)
        .byte   0x8     # DW_AT_byte_size
        .long   0x202   # DW_AT_type
        [...]
        .uleb128 0x15   # (DIE (0x202) DW_TAG_typedef)
        .long   .LASF19 # DW_AT_name: "fd__ints_doubled"
        .long   0x20d   # DW_AT_type

... of an array (1..2) ...

        .uleb128 0x2    # (DIE (0x20d) DW_TAG_array_type)
        .long   .LASF19 # DW_AT_name: "fd__ints_doubled"
        .long   0x15b   # DW_AT_type
        .long   0x221   # DW_AT_sibling
        .uleb128 0x16   # (DIE (0x21a) DW_TAG_subrange_type)
        .long   0x40    # DW_AT_type
        .sleb128 2      # DW_AT_upper_bound
        .byte   0       # end of children of DIE 0x20d

... of a struct whose name is fd__Tints_doubledC:

        .uleb128 0x10   # (DIE (0x15b) DW_TAG_structure_type)
        .long   .LASF11 # DW_AT_name: "fd__Tints_doubledC"
        .long   0x1e4   # DW_AT_GNAT_descriptive_type
                        # DW_AT_artificial
        .long   0x1e4   # DW_AT_sibling
        .uleb128 0x7    # (DIE (0x16a) DW_TAG_member)
        .long   .LASF4  # DW_AT_name: "max_size"
        [snip]

The error occurs while Ada evaluator is trying to "fix"
the element type inside the array, so as to determine its actual
size. For that, it searches for a parallel "XVZ" variable,
which, when found, contains the object's actual size.

Unfortunately in our case, the variable exists but has been
optimized out, as seen by the presence of a variable DIE in
the debugging info, but with no address attribute:

        .uleb128 0x18   # (DIE (0x24e) DW_TAG_variable)
        .long   .LASF31 # DW_AT_name: "fd__Tints_doubledC___XVZ"
        .long   0x257   # DW_AT_type
                        # DW_AT_artificial

Discussing this with some members of AdaCore's compiler team,
it is expected that the optimizer can get rid of this variable,
and we don't want to pessimize the code just to improve debuggability,
since -O2 is about performance. So, the idea of this patch is
not to make it work, but provide a bit more information to help
users understand what kind of error is preventing GDB from being
able to print the variable's value.

The first hurdle we had to clear was the fact that ada_val_print
traps all exceptions (including QUIT ones!), and does so completly
silently. So, the fix was to add a trace of the exception being
generated. While doing so, we fix an old XXX/FIXME by only catching
errors, letting QUIT exceptions go through.

Once this is done, we now get an error message, which gives a first
clue as to what was happening:

    (gdb) p fd.global
    $1 = <error reading variable: value has been optimized out>

However, it would be more useful to know which value it was
that was optimized out. For that purpose, we enhanced
ada-lang.c::ada_to_fixed_type_1 so as to re-throw the error
with a message which indicates which variable we failed to read.

With those changes, the new output is now:

    (gdb) p fd.global
    $1 = <error reading variable: unable to read value of fd__Tints_doubledC___XVZ (value has been optimized out)>

gdb/ChangeLog:

        * ada-lang.c (ada_to_fixed_type_1): Rethrow errors with
        a more detailed exception message when getting an exception
        while trying to read the value of an XVZ variable.
        * ada-valprint.c (ada_val_print): Only catch RETURN_MASK_ERROR
        exceptions.  Print an error message when an exception is caught.

gdb/testsuite/ChangeLog:

        * gdb.dwarf2/ada-valprint-error.c: New file.
        * gdb.dwarf2/ada-valprint-error.exp: New file.

Tested on x86_64-linux

6 years ago(Ada) crash assigning to record component which is an array
Joel Brobecker [Mon, 18 Dec 2017 03:09:27 +0000 (22:09 -0500)]
(Ada) crash assigning to record component which is an array

Consider the following code, which declares a variabled called "input"
of type "parameter", which is a record with one component called "u2",
where the type of that component is a simple 3-element array of
floating point values:

   type Float_Array_3 is array (1 .. 3) of Float;
   type parameters is record
      u2 : Float_Array_3;
   end record;
   input : parameters;

Trying to assign a value to input.u2 causes GDB to crash:

    (gdb) p input.u2 := (0.25,0.5,0.75)
    [1]    20228 segmentation fault (core dumped) [...]/gdb

The crash occurs because input.u2 is described in the debugging
info as a typedef of an array. Indeed, input's type is:

 <1><ae9>: Abbrev Number: 7 (DW_TAG_structure_type)
    <aea>   DW_AT_name        : (indirect string, offset: 0x1045): target_wrapper__parameters
    [...]
 <2><af5>: Abbrev Number: 8 (DW_TAG_member)
    <af6>   DW_AT_name        : u2
    [...]
    <afb>   DW_AT_type        : <0xaca>

and, looking at DIE 0xaca to get input.u2's type, we see:

 <1><aca>: Abbrev Number: 4 (DW_TAG_typedef)
    <acb>   DW_AT_name        : (indirect string, offset: 0x1060): target_wrapper__float_array_3
    [...]
    <ad1>   DW_AT_type        : <0xad5>

We can also confirm, following the DW_AT_type attribute (0xad5), that
it's a typedef of our array:

 <1><ad5>: Abbrev Number: 5 (DW_TAG_array_type)
    <ad6>   DW_AT_name        : (indirect string, offset: 0x1060): target_wrapper__float_array_3
    [...]

In fact, this scenario uncovered 2 areas where typedef handling
is missing, thus causing a crash. The first happens inside
assign_aggregate:

   if (ada_is_direct_array_type (lhs_type))
     {
       lhs = ada_coerce_to_simple_array (lhs);
       lhs_type = value_type (lhs);
       low_index = TYPE_ARRAY_LOWER_BOUND_VALUE (lhs_type);
       high_index = TYPE_ARRAY_UPPER_BOUND_VALUE (lhs_type);
     }

Here, lhs_type is a TYPE_CODE_TYPEDEF. ada_is_direct_array_type
knows how to handle it, but TYPE_ARRAY_LOWER_BOUND_VALUE assumes
that the given type is a TYPE_CODE_ARRAY. As such, it ends up
accessing some fields in lhs_type which it shouldn't, and kaboom.

We fixed this issue by making sure that the TYPE_CODE_TYPEDEF
layer gets stripped.

Once this is done, we hit a different kind of error, also leading to
a SEGV, this time in assign_component. The code looks like this:

  if (TYPE_CODE (value_type (lhs)) == TYPE_CODE_ARRAY)
    [...]
  else
    [...]

Because once again lhs is a TYPE_CODE_TYPEDEF, the check fail,
and we end up assuming that lhs is a struct, executing the "else"
block, which is:

  else
    {
      elt = ada_index_struct_field (index, lhs, 0, value_type (lhs));
      elt = ada_to_fixed_value (elt);
    }

Since lhs is not a struct, ada_index_struct_field returns NULL,
which ada_to_fixed_value does not handle well, hence another crash.

This patch fixes this other issue the same way, by stripping
TYPE_CODE_TYPEDEF layers.

gdb/ChangeLog:

        * ada-lang.c (assign_component): Strip any TYPE_CODE_TYPEDEF
        layer from lhs' type.
        (assign_aggregate): Likewise.

gdb/testsuite:

        * gdb.ada/assign_arr: New testcase.

Tested on x86_64-linux.

6 years agoAda: fix bad handling in ada_convert_actual
Xavier Roirand [Mon, 18 Dec 2017 02:59:07 +0000 (21:59 -0500)]
Ada: fix bad handling in ada_convert_actual

Using this small example:

procedure Foo is

   type Integer_Access is access all Integer;

   procedure P (A : Integer_Access) is
   begin
      null;
   end P;

begin
   P (null);
end Foo;

and doing this debug session:

(gdb) b p
Breakpoint 1 at 0x402d67: file foo.adb, line 7.
(gdb) print p(null)

Breakpoint 1, foo.p (a=0x641010) at foo.adb:10
...                  ^^^^^^^^^^

shows that something goes wrong between the initial null value and the
received parameter value in the 'f' function.
The value for the parameter 'a' we get is the address of the value we
would expect instead of the value itself. This can be checked by doing:

(gdb) p *a
$1 = 0

Before this fix, in ada_convert_value, this function was looking to the
actual value (the null value here) to determine if the formal (parameter
'a' in the procedure 'P' in this exemple) requires a pointer or not which
is a wrong assumption and leads to push the address of the value to the
inferior instead of the value itself.

This is fixed by this patch.

gdb/ChangeLog:

        * ada-lang.c (ada_convert_actual): Change the way actual value
        are passed to the inferior when the inferior expects a pointer type.

gdb/testsuite/ChangeLog:

        * gdb.ada/funcall_ptr: New testcase.

Tested on x86_64-linux.

6 years agoAutomatic date update in version.in
GDB Administrator [Mon, 18 Dec 2017 00:00:20 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agox86: Check pseudo prefix without instruction
H.J. Lu [Sun, 17 Dec 2017 17:40:54 +0000 (09:40 -0800)]
x86: Check pseudo prefix without instruction

Pseudo prefixes must be used on an instruction.  Issue an error when
pseudo prefix is used without instruction.

PR gas/22623
* gas/config/tc-i386.c (output_insn): Check pseudo prefix
without instruction.
* testsuite/gas/i386/i386.exp: Run inval-pseudo.
* testsuite/gas/i386/inval-pseudo.l: New file.
* testsuite/gas/i386/inval-pseudo.s: Likewise.

6 years agoAutomatic date update in version.in
GDB Administrator [Sun, 17 Dec 2017 00:00:31 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agogdb: Fix function parameter alignments in or1k-tdep.c.
Stafford Horne [Sat, 16 Dec 2017 22:20:51 +0000 (07:20 +0900)]
gdb: Fix function parameter alignments in or1k-tdep.c.

As suggested by Joel Brobecker <brobecker@adacore.com> and as per fsf
coding standards.  Also fix a few more issues with directly printing
pointers.

gdb/ChangeLog:

* gdb/or1k-tdep.c (show_or1k_debug): Fix function parameter alignment.
(or1k_analyse_inst): Likewise.
(or1k_single_step_through_delay): Likewise.
(or1k_frame_cache): Fix parameter alignment and use paddress()
instead of %x.