platform/upstream/binutils.git
9 years agoPR binutils/18257: Properly decode x86/Intel mask instructions.
Alexander Fomin [Fri, 21 Aug 2015 11:06:41 +0000 (14:06 +0300)]
PR binutils/18257: Properly decode x86/Intel mask instructions.

opcodes/

PR binutils/18257
* i386-dis.c: Use MOD_TABLE for most of mask instructions.
(MOD enum): Add MOD_VEX_W_0_0F41_P_0_LEN_1, MOD_VEX_W_1_0F41_P_0_LEN_1,
MOD_VEX_W_0_0F41_P_2_LEN_1, MOD_VEX_W_1_0F41_P_2_LEN_1,
MOD_VEX_W_0_0F42_P_0_LEN_1, MOD_VEX_W_1_0F42_P_0_LEN_1,
MOD_VEX_W_0_0F42_P_2_LEN_1, MOD_VEX_W_1_0F42_P_2_LEN_1,
MOD_VEX_W_0_0F44_P_0_LEN_1, MOD_VEX_W_1_0F44_P_0_LEN_1,
MOD_VEX_W_0_0F44_P_2_LEN_1, MOD_VEX_W_1_0F44_P_2_LEN_1,
MOD_VEX_W_0_0F45_P_0_LEN_1, MOD_VEX_W_1_0F45_P_0_LEN_1,
MOD_VEX_W_0_0F45_P_2_LEN_1, MOD_VEX_W_1_0F45_P_2_LEN_1,
MOD_VEX_W_0_0F46_P_0_LEN_1, MOD_VEX_W_1_0F46_P_0_LEN_1,
MOD_VEX_W_0_0F46_P_2_LEN_1, MOD_VEX_W_1_0F46_P_2_LEN_1,
MOD_VEX_W_0_0F47_P_0_LEN_1, MOD_VEX_W_1_0F47_P_0_LEN_1,
MOD_VEX_W_0_0F47_P_2_LEN_1, MOD_VEX_W_1_0F47_P_2_LEN_1,
MOD_VEX_W_0_0F4A_P_0_LEN_1, MOD_VEX_W_1_0F4A_P_0_LEN_1,
MOD_VEX_W_0_0F4A_P_2_LEN_1, MOD_VEX_W_1_0F4A_P_2_LEN_1,
MOD_VEX_W_0_0F4B_P_0_LEN_1, MOD_VEX_W_1_0F4B_P_0_LEN_1,
MOD_VEX_W_0_0F4B_P_2_LEN_1, MOD_VEX_W_0_0F91_P_0_LEN_0,
MOD_VEX_W_1_0F91_P_0_LEN_0, MOD_VEX_W_0_0F91_P_2_LEN_0,
MOD_VEX_W_1_0F91_P_2_LEN_0, MOD_VEX_W_0_0F92_P_0_LEN_0,
MOD_VEX_W_0_0F92_P_2_LEN_0, MOD_VEX_W_0_0F92_P_3_LEN_0,
MOD_VEX_W_1_0F92_P_3_LEN_0, MOD_VEX_W_0_0F93_P_0_LEN_0,
MOD_VEX_W_0_0F93_P_2_LEN_0, MOD_VEX_W_0_0F93_P_3_LEN_0,
MOD_VEX_W_1_0F93_P_3_LEN_0, MOD_VEX_W_0_0F98_P_0_LEN_0,
MOD_VEX_W_1_0F98_P_0_LEN_0, MOD_VEX_W_0_0F98_P_2_LEN_0,
MOD_VEX_W_1_0F98_P_2_LEN_0, MOD_VEX_W_0_0F99_P_0_LEN_0,
MOD_VEX_W_1_0F99_P_0_LEN_0, MOD_VEX_W_0_0F99_P_2_LEN_0,
MOD_VEX_W_1_0F99_P_2_LEN_0, MOD_VEX_W_0_0F3A30_P_2_LEN_0,
MOD_VEX_W_1_0F3A30_P_2_LEN_0, MOD_VEX_W_0_0F3A31_P_2_LEN_0,
MOD_VEX_W_1_0F3A31_P_2_LEN_0, MOD_VEX_W_0_0F3A32_P_2_LEN_0,
MOD_VEX_W_1_0F3A32_P_2_LEN_0, MOD_VEX_W_0_0F3A33_P_2_LEN_0,
MOD_VEX_W_1_0F3A33_P_2_LEN_0.
(vex_w_table): Replace terminals with MOD_TABLE entries for
most of mask instructions.

gas/testsuite

PR binutils/18257
* gas/i386/disassem.s: Add mask instructions with invalid ModR/M byte.
* gas/i386/x86-64-disassem.s: Likewise.
* gas/i386/disassem.d: Updated.
* gas/i386/x86-64-disassem.d: Likewise.

9 years agoAdd readahead cache to gdb's vFile:pread
Pedro Alves [Fri, 21 Aug 2015 09:13:27 +0000 (10:13 +0100)]
Add readahead cache to gdb's vFile:pread

This patch almost halves the time it takes to "target remote + run to
main" on a higher-latency connection.

E.g., I've got a ping time of ~85ms to an x86-64 machine on the gcc
compile farm (almost 2000km away from me), and I'm behind a ~16Mbit
ADSL.  When I connect to a gdbserver debugging itself on that machine
and run to main, it takes almost 55 seconds:

 [palves@gcc76] $ ./gdbserver :9999 ./gdbserver
 [palves@home] $ ssh -L 9999:localhost:9999 gcc76.fsffrance.org
 [palves@home] $ time ./gdb -data-directory=data-directory -ex "tar rem :9999" -ex "b main" -ex "c" -ex "set confirm off" -ex "quit"

 Pristine gdb 7.10.50.20150820-cvs gets us:
 ...
 Remote debugging using :9999
 Reading symbols from target:/home/palves/gdb/build/gdb/gdbserver/gdbserver...done.
 Reading symbols from target:/lib64/ld-linux-x86-64.so.2...(no debugging symbols found)...done.
 0x00007ffff7ddd190 in ?? () from target:/lib64/ld-linux-x86-64.so.2
 Breakpoint 1 at 0x41200c: file ../../../src/gdb/gdbserver/server.c, line 3635.
 Continuing.

 Breakpoint 1, main (argc=1, argv=0x7fffffffe3d8) at ../../../src/gdb/gdbserver/server.c:3635
 3635    ../../../src/gdb/gdbserver/server.c: No such file or directory.
 /home/palves/gdb/build/gdb/gdbserver/gdbserver: No such file or directory.

 real    0m54.803s
 user    0m0.329s
 sys     0m0.064s

While with the readahead cache added by this patch, it drops to:

 real    0m29.462s
 user    0m0.454s
 sys     0m0.054s

I added a few counters to show cache hit/miss, and got:

 readahead cache miss 142
 readahead cache hit 310

Tested on x86_64 Fedora 20.

gdb/ChangeLog:
2015-08-21  Pedro Alves  <palves@redhat.com>

* remote.c (struct readahead_cache): New.
(struct remote_state) <readahead_cache>: New field.
(remote_open_1): Invalidate the cache.
(readahead_cache_invalidate, readahead_cache_invalidate_fd): New
functions.
(remote_hostio_pwrite): Invalidate the readahead cache.
(remote_hostio_pread): Rename to ...
(remote_hostio_pread_vFile): ... this.
(remote_hostio_pread_from_cache): New function.
(remote_hostio_pread): Reimplement.
(remote_hostio_close): Invalidate the readahead cache.

9 years agoprocfs.c: Include "filestuff.h"
Marcin Cieslak [Wed, 19 Aug 2015 16:51:31 +0000 (16:51 +0000)]
procfs.c: Include "filestuff.h"

Fixes implicit function declaration
error in gdb/procfs.c:4927 about undeclared
make_cleanup_close().

gdb/ChangeLog:

PR build/18843
* procfs.c: Include "filestuff.h".

9 years agoRemove fields curch and cur_line from TUI_CMD_WIN
Patrick Palka [Wed, 19 Aug 2015 01:08:57 +0000 (21:08 -0400)]
Remove fields curch and cur_line from TUI_CMD_WIN

These fields are currently used to track the location of the cursor
inside the command window.  But their usefulness is questionable because
ncurses already internally keeps track of the location of the cursor,
whose coordinates we can query using the functions getyx(), getcurx() or
getcury().  It is an unnecessary pain to keep these fields in sync with
ncurses, and their meaning is not well-defined anyway.  For instance, it
is not clear whether the coordinates held in these fields are
authoritative, or whether the coordinates reported by ncurses are.

So to keep things simple, this patch removes these fields and replaces
existing reads of these fields with calls to the appropriate ncurses
querying functions, and replaces writes to these fields with calls to
wmove() (when necessary and applicable).

In the function tui_cont_sig(), I removed the call to wmove() entirely
because moving to (start_line, curch) makes no sense.  The move should
have been to (cur_line, curch) -- which would now be a no-op.

Tested on x86_64 Fedora 22, no obvious regressions.

gdb/ChangeLog:

* tui/tui-data.h (tui_command_info): Remove fields cur_line and
curch.
* tui/tui-data.c (tui_clear_win_detail) [CMD_WIN]: Don't set
cur_line or curch, instead call wmove().
(init_win_info) [CMD_WIN]: Likewise.
* tui/tui-io.c (tui_puts): Likewise. Don't read cur_line,
instead call getcury().
(tui_redisplay_readline): Don't set cur_line or curch.
(tui_mld_erase_entire_line): Don't read cur_line, instead call
getcury().
(tui_cont_sig): Remove call to wmove.
(tui_getc): Don't read cur_line or curch, instead call getcury()
or getyx().  Don't set curch.
* tui/tui-win.c (make_visible_with_new_height) [CMD_WIN]: Don't
set cur_line or curch.  Always move cursor to (0,0).

9 years agoAutomatic date update in version.in
GDB Administrator [Fri, 21 Aug 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoFix gdb.server/solib-list.exp native-extended-gdbserver regression
Pedro Alves [Thu, 20 Aug 2015 19:08:48 +0000 (20:08 +0100)]
Fix gdb.server/solib-list.exp native-extended-gdbserver regression

Commit 221e1a37 (remote non-stop: Process initially stopped threads
before other commands) caused a test regression when testing with the
native-extended-gdbserver board:

  FAIL: gdb.server/solib-list.exp: non-stop 1: non-stop interior stop (timeout)

This "interior stop" now happens before "target remote" prints the
prompt, so we should no longer explicitly expect it.

gdb/testsuite/ChangeLog:
2015-08-20  Pedro Alves  <palves@redhat.com>

* gdb.server/solib-list.exp: No longer expect an interior stop in
non-stop mode.

9 years agoremote non-stop: Process initially stopped threads before other commands
Pedro Alves [Thu, 20 Aug 2015 17:27:55 +0000 (18:27 +0100)]
remote non-stop: Process initially stopped threads before other commands

The main motivation for this is making non-stop / all-stop behave
similarly on initial connection, in order to move in the direction of
reimplementing all-stop mode with the remote target always running in
non-stop mode.

When we connect to a remote target in non-stop mode, we may find
threads either running or already stopped.  The act of connecting
itself does not force threads to stop.  To handle that, the remote
non-stop connection is currently roughly like this:

 #1 - Fetch list of remote threads (qXfer:threads:read, qfThreadInfo,
    etc).  All threads are assumed to be running until the target
    reports an asynchronous stop reply for them.

 #2 - Fetch the initial set of threads that were already stopped, with
    the '?'  packet.  (In non-stop, this is coupled with the vStopped
    mechanism to be able to retrieve the status of more than one
    thread.)

The stop replies fetched in #2 are placed in the pending stop reply
queue, and left for the regular event loop to process.  That is,
"target remote" finishes and returns _before_ those stops are
processed.

That means that it's possible to have GDB process further commands
before the initial set of stopped threads is reported to the user.

E.g., before the patch, note how the prompt is printed before the
frame:

 Remote debugging using :9999
 (gdb)
 [Thread 15296] #1 stopped.
 0x0000003615a011f0 in ?? ()

Even though thread #1 was not running, for a moment, the user can see
it as such:

 $ gdb a.out -ex "set non-stop 1" -ex "tar rem :9999"  -ex "info threads" -ex "info registers"
 Remote debugging using :9999
   Id   Target Id         Frame
 * 1    Thread 4772       (running)
 Target is executing.                 <<<<<<< info registers
 (gdb)
 [Thread 4772] #1 stopped.
 0x0000003615a011f0 in ?? ()

To fix that, this commit makes gdb process all threads found already
stopped at connection time, before giving the prompt to the user.

The fix takes a cue from fork-child.c:startup_inferior [1], and
processes the events locally in remote.c, avoiding the whole
wait_for_inferior/handle_inferior_event path.  I decided to try this
approach after noticing that:

 - several cases in handle_inferior_event miss checking stop_soon.
 - we don't want to fetch the thread list in normal_stop.

and trying to fix them was resulting in sprinkling stop_soon checks in
many places, and uglifying normal_stop even more.

While with this patch, I'm avoiding changing GDB's output other than
when the prompt is printed, I think this approach is more flexible if
we do want to change it.  And also, it's likely easier to get rid of
the MI *running event that is still sent for threads that are
initially found stopped, if we want to.

This happens to fix the testsuite too.  All non-stop tests are racy
against "target remote" / gdbserver testing currently.  That is,
sometimes the tests run, but other times they're just skipped without
any indication of PASS/FAIL.  When that happens, the logs show:

 target remote localhost:2346
 Remote debugging using localhost:2346
 (gdb)
 [Thread 25418] #1 stopped.
 0x0000003615a011f0 in ?? ()
 ^CQuit
 (gdb) Remote debugging from host 127.0.0.1
 Killing process(es): 25418
 monitor exit
 (gdb) Remote connection closed
 (gdb) testcase /home/pedro/gdb/mygit/build/../src/gdb/testsuite/gdb.threads/multi-create-ns-info-thr.exp completed in 61 seconds

The trouble here is that there's output after the prompt, and the
regex in question doesn't expect that:

   -re "Remote debugging using .*$serialport_re.*$gdb_prompt $" {
verbose "Set target to $targetname"
return 0
    }

[1] - before startup_inferior was added, we'd go through
wait_for_inferior/handle_inferior_event while going through the shell,
and that turned out problematic.

Tested on x86_64 Fedora 20, gdbserver.

gdb/ChangeLog:
2015-08-20  Pedro Alves  <palves@redhat.com>

* infrun.c (print_target_wait_results): Make extern.
* infrun.h (print_target_wait_results): Declare.
* remote.c (set_stop_requested_callback): Delete.
(process_initial_stop_replies): New function.
(remote_start_remote): Use it.
(stop_reply_queue_length): New function.

gdb/testsuite/ChangeLog:
2015-08-20  Pedro Alves  <palves@redhat.com>

* gdb.server/connect-stopped-target.c: New file.
* gdb.server/connect-stopped-target.exp: New file.

9 years agoFix language of compilation unit with unknown file extension
Pedro Alves [Thu, 20 Aug 2015 11:30:08 +0000 (12:30 +0100)]
Fix language of compilation unit with unknown file extension

Here, in dwarfread.c:process_full_comp_unit:

      /* Set symtab language to language from DW_AT_language.  If the
 compilation is from a C file generated by language preprocessors, do
 not set the language if it was already deduced by start_subfile.  */
      if (!(cu->language == language_c
    && COMPUNIT_FILETABS (cust)->language != language_c))
COMPUNIT_FILETABS (cust)->language = cu->language;

in case start_subfile doesn't manage to deduce a language
COMPUNIT_FILETABS(cust)->language ends up as language_unknown, not
language_c.  So the condition above evals false and we never set the
language from the cu's language.

gdb/ChangeLog:
2015-08-20  Pedro Alves  <palves@redhat.com>

* dwarf2read.c (process_full_comp_unit): To tell whether
start_subfile managed to deduce a language, test for
language_unknown instead of language_c.

gdb/testsuite/ChangeLog:
2015-08-20  Pedro Alves  <palves@redhat.com>

* gdb.dwarf2/comp-unit-lang.exp: New file.
* gdb.dwarf2/comp-unit-lang.c: New file.

9 years ago[Ada] Fix parsing for expressions with attributes and characters
Pierre-Marie de Rodat [Wed, 19 Aug 2015 15:12:48 +0000 (17:12 +0200)]
[Ada] Fix parsing for expressions with attributes and characters

Before this change, trying to evaluate the following Ada expression
yielded a syntax error, even though it's completely legal:

    (gdb) p s'first = 'a'
    Error in expression, near `'.

The problem lies in the lexer (gdb/ada-lex.l): at the point we reach "'a'",
we're still in the BEFORE_QUAL_QUOTE start condition (the mechanism to
distinguish character literals from other "tick" usages: qualified
expressions and attributes), so we consider that this quote is actually a
separate "tick".

This changes resets the start condition to INITIAL in the
{TICK}[a-zA-Z][a-zA-Z]+ rule (for attributes): attributes activate this
BEFORE_QUAL_QUOTE condition and in this case the above rule is always
executed rather than the <BEFORE_QUAL_QUOTE>"'" one (in flex, it's
always the longest match that is chosen). We now have instead:

    (gdb) p s'first = 'a'
    $1 = true

gdb/ChangeLog:

* ada-lex.l: Reset the start condition to INITIAL in the rule
that matches attributes.

gdb/testsuite/ChangeLog:

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

Tested on x86_64-linux, no regression.

9 years agogold --emit-relocs
Alan Modra [Thu, 20 Aug 2015 02:32:45 +0000 (12:02 +0930)]
gold --emit-relocs

A symbol value in an ELF final linked binary is absolute, in contrast
to a relocatable object file where the value is section relative.  For
--emit-relocs it is therefore incorrect to use the value of a section
symbol as the addend when adjusting relocs against input section
symbols to output section symbols.

PR gold/18846
* target-reloc.h (relocate_relocs <RELOC_ADJUST_FOR_SECTION_RELA>):
Subtract os->address() from addend.
* powerpc.cc (relocate_relocs): Likewise.

9 years agoAutomatic date update in version.in
GDB Administrator [Thu, 20 Aug 2015 00:00:10 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agodwarf2read.c: Check type of string valued attributes prior to decoding.
Kevin Buettner [Mon, 3 Aug 2015 21:17:17 +0000 (14:17 -0700)]
dwarf2read.c: Check type of string valued attributes prior to decoding.

This change introduces a new function, dwarf2_string_attr(), which is
a wrapper for dwarf2_attr().  dwarf2read.c has been updated to
call dwarf2_string_attr in most instances where a string-valued
attribute is decoded to produce a string value.  In most cases, it
simplifies the code; in some instances, the complexity of the code
remains unchanged.

I performed this change by looking for instances where the
result of DW_STRING was used in an assignment.  Many of these
had a pattern which (roughly) looks something like this:

  struct attribute *attr = NULL;

  attr = dwarf2_attr (die, name, cu);
  if (attr != NULL && DW_STRING (attr))
    {
      const char *str;
      ...
      str = DW_STRING (attr);
      ... /* Use str in some fashion.  */
    }

Code of this form is transformed to look like this instead:

  const char *str;

  str = dwarf2_string_attr (die, name, cu)
  if (str != NULL)
    {
       ...
       /* Use str in some fashion.  */
       ...
    }

In addition to invoking dwarf2_attr() and DW_STRING(),
dwarf2_string_attr() checks to make sure that the attribute's
`form' field matches one of DW_FORM_strp, DW_FORM_string, or
DW_FORM_GNU_strp_alt.  If it does not match one of these forms,
it will return a NULL value in addition to calling complaint().

An earlier version of this patch did this type checking for one
particular instance where a string attribute was being decoded.
The situation that I was attempting to handle in that earlier patch is
this:

The Texas Instruments compiler uses the encoding for
DW_AT_MIPS_linkage_name for other purposes.  TI uses the encoding,
0x2007, for TI_AT_TI_end_line which, unlike DW_AT_MIPS_linkage_name,
does not have a string-typed value.  In this instance, GDB was attempting
to use an integer value as a string pointer, with predictable results.
(GDB would die with a segmentation fault.)

I've added a test which reproduces the problem that I was orignally
wanting to fix.  It uses DW_AT_MIPS_linkage name with an associate
value which is a string, and again, where the value is a small
integer.

My test case causes GDB to segfault in an unpatched GDB.  There
will be two PASSes in a patched GDB.

Unpatched GDB:

(gdb) ptype f
ERROR: Process no longer exists
UNRESOLVED: gdb.dwarf2/dw2-bad-mips-linkage-name.exp: ptype f
ERROR: Couldn't send ptype g to GDB.
UNRESOLVED: gdb.dwarf2/dw2-bad-mips-linkage-name.exp: ptype g

Patched GDB:

(gdb) ptype f
type = bool ()
(gdb) PASS: gdb.dwarf2/dw2-bad-mips-linkage-name.exp: ptype f
ptype g
type = bool ()
(gdb) PASS: gdb.dwarf2/dw2-bad-mips-linkage-name.exp: ptype g

I see no regressions on an x86_64 native target.

gdb/ChangeLog:

* dwarf2read.c (dwarf2_string_attr): New function.
(lookup_dwo_unit, process_psymtab_comp_unit_reader)
(dwarf2_compute_name, dwarf2_physname, find_file_and_directory)
(read_call_site_scope, namespace_name, guess_full_die_structure_name)
(anonymous_struct_prefix, prepare_one_comp_unit): Use
dwarf2_string_attr in place of dwarf2_attr and DW_STRING.

gdb/testsuite/ChangeLog:

* gdb.dwarf2/dw2-bad-mips-linkage-name.c: New file.
* gdb.dwarf2/dw2-bad-mips-linkage-name.exp: New file.

9 years ago[AArch64][6/6] LD support TLSLD load/store relocation types
Jiong Wang [Wed, 19 Aug 2015 10:26:56 +0000 (11:26 +0100)]
[AArch64][6/6] LD support TLSLD load/store relocation types

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * elfnn-aarch64.c (IS_AARCH64_TLS_RELOC): Recognize new relocation
  types, including BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC.
  (elfNN_aarch64_final_link_relocate): Likewise.
  * elfxx-aarch64.c (_bfd_aarch64_elf_put_addend): Likewise.
  (_bfd_aarch64_elf_resolve_relocation): Likewise.

ld/testsuite/
  * ld-aarch64/emit-relocs-531.s: New testcase.
  * ld-aarch64/emit-relocs-531-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-532.s: Likewise.
  * ld-aarch64/emit-relocs-533.s: Likewise.
  * ld-aarch64/emit-relocs-533-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-534.s: Likewise.
  * ld-aarch64/emit-relocs-535.s: Likewise.
  * ld-aarch64/emit-relocs-535-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-536.s: Likewise.
  * ld-aarch64/emit-relocs-537.s: Likewise.
  * ld-aarch64/emit-relocs-537-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-538.s: Likewise.
  * ld-aarch64/emit-relocs-531.d: New expectation file.
  * ld-aarch64/emit-relocs-531-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-532.d: Likewise.
  * ld-aarch64/emit-relocs-533.d: Likewise.
  * ld-aarch64/emit-relocs-533-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-534.d: Likewise.
  * ld-aarch64/emit-relocs-535.d: Likewise.
  * ld-aarch64/emit-relocs-535-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-536.d: Likewise.
  * ld-aarch64/emit-relocs-537.d: Likewise.
  * ld-aarch64/emit-relocs-537-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-538.d: Likewise.
  * ld-aarch64/aarch64-elf.exp: Run new testcases.

9 years ago[AArch64][5/6] GAS support TLSLD load/store relocation types
Jiong Wang [Wed, 19 Aug 2015 10:22:22 +0000 (11:22 +0100)]
[AArch64][5/6] GAS support TLSLD load/store relocation types

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * reloc.c: New entries, including
  BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC.
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC.
  * elfnn-aarch64.c (elfNN_aarch64_howto_table): Likewise.
  * bfd-in2.h: Regenerate.
  * libbfd.h: Regenerate.

gas/
  * config/tc-aarch64.c (reloc_table): New relocation types support for
  dtprel_lo12.
  (ldst_lo12_determine_real_reloc_type): Support
  BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC,
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12,
  BFD_RELOC_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC.
  (parse_operands): Likewise.
  (md_apply_fix): Likewise
  (aarch64_force_relocation): Likewise.
  (process_movw_reloc_info): Likewise.

gas/testsuite/
  * gas/aarch64/reloc-dtprel_lo12-ldst8.s: New testcase.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldstc.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst16.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst16.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst32.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst32.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst64.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst64.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst8.d: New expectation file.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst8.d: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst16.d: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst16.d: Likewise.
  * gas/aarch64/reloc-dtprel_lo12-ldst32.d: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst32.d: Likewise.
  * gas/aarch64/reloc-dtprel-lo12-ldst64.d: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc-ldst64.d: Likewise.

9 years ago[AArch64][4/6] LD support TLSLD move/add relocation types
Jiong Wang [Wed, 19 Aug 2015 10:18:25 +0000 (11:18 +0100)]
[AArch64][4/6] LD support TLSLD move/add relocation types

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  PR ld/18276
  * elfnn-aarch64.c (IS_AARCH64_TLS_RELOC): Recognize new relocation
  types, including BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_HI12,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G0,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G0_NC,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G1,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G1_NC,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G2.
  (elfNN_aarch64_final_link_relocate): Likewise.
  * elfxx-aarch64.c (_bfd_aarch64_elf_put_addend): Likewise.
  (_bfd_aarch64_elf_resolve_relocation): Likewise.

ld/testsuite/
  * ld-aarch64/emit-relocs-87.s: New testcase.
  * ld-aarch64/emit-relocs-88.s: Likewise.
  * ld-aarch64/emit-relocs-88-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-89.s: Likewise.
  * ld-aarch64/emit-relocs-90.s: Likewise.
  * ld-aarch64/emit-relocs-90-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-523.s: Likewise.
  * ld-aarch64/emit-relocs-524.s: Likewise.
  * ld-aarch64/emit-relocs-525.s: Likewise.
  * ld-aarch64/emit-relocs-527.s: Likewise.
  * ld-aarch64/emit-relocs-526.s: Likewise.
  * ld-aarch64/emit-relocs-528.s: Likewise.
  * ld-aarch64/emit-relocs-528-overflow.s: Likewise.
  * ld-aarch64/emit-relocs-87.d: New expectation file.
  * ld-aarch64/emit-relocs-88.d: Likewise.
  * ld-aarch64/emit-relocs-88-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-89.d: Likewise.
  * ld-aarch64/emit-relocs-90.d: Likewise.
  * ld-aarch64/emit-relocs-90-overflow.d: Likewise.
  * ld-aarch64/emit-relocs-91.d: Likewise.
  * ld-aarch64/emit-relocs-523.d: Likewise.
  * ld-aarch64/emit-relocs-524.d: Likewise.
  * ld-aarch64/emit-relocs-525.d: Likewise.
  * ld-aarch64/emit-relocs-526.d: Likewise.
  * ld-aarch64/emit-relocs-527.d: Likewise.
  * ld-aarch64/emit-relocs-528.d: Likewise.
  * ld-aarch64/emit-relocs-528-overflow.d: Likewise.
  * ld-aarch64/aarch64-elf.exp: Run new testcases.

9 years ago[AArch64][3/6] GAS support TLSLD move/add relocation types
Jiong Wang [Wed, 19 Aug 2015 10:02:34 +0000 (11:02 +0100)]
[AArch64][3/6] GAS support TLSLD move/add relocation types

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * reloc.c (BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_HI12,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G0,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G0_NC,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G1,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G1_NC,
  BFD_RELOC_AARCH64_TLSLD_MOVW_DTPREL_G2): New entries.
  * elfnn-aarch64.c (elfNN_aarch64_howto_table): Likewise.
  * bfd-in2.h: Regenerate.
  * libbfd.h: Regenerate.

gas/
  * config/tc-aarch64.c (reloc_table): New relocation modifiers,
  "dtprel_hi12", "dtprel_g0", "dtprel_g0_nc", "dtprel_g1",
  "dtprel_g1_nc", "dtprel_g2".
  (md_apply_fix): Support new relocation types.
  (aarch64_force_relocation): Likewise.
  (process_movw_reloc_info): Likewise.

gas/testsuite/
  * gas/aarch64/reloc-dtprel_g0.s: New testcase.
  * gas/aarch64/reloc-dtprel_g0-ilp32.s: Likewise.
  * gas/aarch64/reloc-dtprel_g0_nc.s: Likewise.
  * gas/aarch64/reloc-dtprel_g0_nc-ilp32.s: Likewise.
  * gas/aarch64/reloc-dtprel_g1.s: Likewise.
  * gas/aarch64/reloc-dtprel_g1-ilp32.s: Likewise.
  * gas/aarch64/reloc-dtprel_g1_nc.s: Likewise.
  * gas/aarch64/reloc-dtprel_g2.s: Likewise.
  * gas/aarch64/reloc-dtprel_hi12.s: Likewise.
  * gas/aarch64/reloc-dtprel_hi12-ilp32.s: Likewise.
  * gas/aarch64/reloc-dtprel_g0.d: New expectation file.
  * gas/aarch64/reloc-dtprel_g0-ilp32.d: Likewise.
  * gas/aarch64/reloc-dtprel_g0_nc.d: Likewise.
  * gas/aarch64/reloc-dtprel_g0_nc-ilp32.d: Likewise.
  * gas/aarch64/reloc-dtprel_g1.d: Likewise.
  * gas/aarch64/reloc-dtprel_g1-ilp32.d: Likewise.
  * gas/aarch64/reloc-dtprel_g1_nc.d: Likewise.
  * gas/aarch64/reloc-dtprel_g2.d: Likewise.
  * gas/aarch64/reloc-dtprel_hi12.d: Likewise.
  * gas/aarch64/reloc-dtprel_hi12-ilp32.d: Likewise.

9 years ago[AArch64][2/6] LD support BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC
Jiong Wang [Wed, 19 Aug 2015 09:58:13 +0000 (10:58 +0100)]
[AArch64][2/6] LD support BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * elfnn-aarch64.c (IS_AARCH64_TLS_RELOC): Recognize
  BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC.
  (aarch64_reloc_got_type): Likewise.
  (elfNN_aarch64_final_link_relocate): Likewise.
  (elfNN_aarch64_relocate_section): Likewise.
  * elfxx-aarch64.c (_bfd_aarch64_elf_put_addend): Likewise.
  (_bfd_aarch64_elf_resolve_relocation): Likewise.

ld/testsuite/
  * ld-aarch64/emit-relocs-530.s: New testcase.
  * ld-aarch64/emit-relocs-92.s: Likewise.
  * ld-aarch64/emit-relocs-530.d: New expectation file.
  * ld-aarch64/emit-relocs-92.d: Likewise.
  * ld-aarch64/aarch64-elf.exp: Run new testcases.

9 years ago[AArch64][1/6] GAS support BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC
Jiong Wang [Wed, 19 Aug 2015 09:57:34 +0000 (10:57 +0100)]
[AArch64][1/6] GAS support BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC

2015-08-19  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * reloc.c (BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC): New entry.
  * bfd-in2.h: Regenerate.
  * libbfd.h: Regenerate.
  * elfnn-aarch64.c (elfNN_aarch64_howto_table): New entry for
  BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC.

gas/
  * config/tc-aarch64.c (reloc_table): New relocation modifiers.
  (md_apply_fix): Support BFD_RELOC_AARCH64_TLSLD_ADD_DTPREL_LO12_NC.
  (aarch64_force_relocation): Likewise.

gas/testsuite/
  * gas/aarch64/reloc-dtprel_lo12_nc.s: New testcase.
  * gas/aarch64/reloc-dtprel_lo12_nc-ilp32.s: Likewise.
  * gas/aarch64/reloc-dtprel_lo12_nc.d: New expectation file.
  * gas/aarch64/reloc-dtprel_lo12_nc-ilp32.d: Likewise.

9 years agoPrelimit number of bytes to read in "vFile:pread:"
Gary Benson [Wed, 19 Aug 2015 12:53:24 +0000 (13:53 +0100)]
Prelimit number of bytes to read in "vFile:pread:"

While handling "vFile:pread:" packets, gdbserver would read the
number of bytes requested regardless of whether this would fit
into the reply packet.  gdbserver would then return a packet's
worth of data and discard the remainder.  When accessing large
binaries GDB (via BFD) routinely makes large "vFile:pread:"
requests, resulting in gdbserver allocating large unnecessary
buffers and reading some portions of the file many times over.

This commit causes gdbserver to limit the number of bytes to be
read to a sensible maximum prior to allocating buffers and reading
data.

gdb/gdbserver/ChangeLog:

* hostio.c (handle_pread): Do not attempt to read more data
than hostio_reply_with_data can fit in a packet.

9 years agoRemove link_info.pic
Alan Modra [Wed, 19 Aug 2015 03:02:54 +0000 (12:32 +0930)]
Remove link_info.pic

Adding type_pie to output_type allows us to remove link_info.pic and
with some reordering of the enum, produces better code.

include/
* bfdlink.h (enum output_type): Delete type_executable, add type_pde
and type_pie.  Reorder.
(struct bfd_link_info): Delete pic field.
(bfd_link_executable, bfd_link_pde, bfd_link_pie, bfd_link_pic): Adjust.
ld/
* emultempl/aix.em: Don't set link_info.pic.
* emultempl/pe.em: Likewise.
* emultempl/pep.em: Likewise.
* emultempl/sunos.em: Likewise.
* lexsup.c (parse_args): Likewise.  Set type_pie for -pie.
* plugin.c (set_tv_header <LDPT_LINKER_OUTPUT>): Simplify.

9 years agoAdd bfd_link_pde, and simplify some tests of link_info.type
Alan Modra [Wed, 19 Aug 2015 02:27:40 +0000 (11:57 +0930)]
Add bfd_link_pde, and simplify some tests of link_info.type

include/
* bfdlink.h (bfd_link_pde): Define.
bfd/
* elf-s390-common.c: Simplify expressions using
bfd_linke_executable, bfd_link_pie and bfd_link_pic.
* elf32-arm.c: Likewise.
* elf32-bfin.c: Likewise.
* elf32-frv.c: Likewise.
* elf32-m68k.c: Likewise.
* elf32-nios2.c: Likewise.
* elf32-ppc.c: Likewise.
* elf32-s390.c: Likewise.
* elf32-sh.c: Likewise.
* elf64-alpha.c: Likewise.
* elf64-ppc.c: Likewise.
* elf64-s390.c: Likewise.
* elflink.c: Likewise.

9 years agoFix symbol versioning problems in PR 18703.
Cary Coutant [Mon, 27 Jul 2015 22:09:08 +0000 (15:09 -0700)]
Fix symbol versioning problems in PR 18703.

If a symbol is defined with ".symver foo,foo@VER", the assembler
creates two symbols in the object: one unversioned, and one with
the (non-default) version "VER". If foo is listed in a version
script, gold would then make the first of those symbols the
default version, and would ignore the second symbol as a
duplicate, without making it a non-default version. While this is
arguably reasonable behavior, it doesn't match Gnu ld behavior,
so this patch fixes that by allowing the second definition to
override the first by resetting the "default version" indication.

Several test cases from the Gnu ld testsuite also exposed another
related problem, where a symbol defined with ".symver foo,foo@",
placed into a shared library, is not handled properly by gold.
This patch also fixes that case, binding the symbol to the base
version.

gold/
PR gold/18703
* dynobj.cc (Versions::record_version): Handle symbol defined with
base version.
(Versions::symbol_section_contents): Likewise.
* symtab.h (Symbol::set_is_not_default): New class method.
(Symbol_table::resolve): Add is_default_version parameter.
(Symbol_table::should_override): Likewise.
* resolve.cc (Symbol_table::resolve): Add is_default_version parameter,
and pass to should_override. Adjust all callers and explicit
instantiations.
(Symbol_table::should_override): Add is_default_value parameter;
allow default version in a dynamic object to override existing
definition from same object.
* symtab.cc (Symbol_table::add_from_object): Handle case where same
symbol is defined as unversioned and non-default version in the same
object.
* testsuite/Makefile.am (ver_test_13): New test case.
* testsuite/Makefile.in: Regenerate.
* testsuite/ver_test_4.cc: Add test for symbol with base version.
* testsuite/ver_test_4.sh: Likewise.
* testsuite/ver_test_13.c: New source file.
* testsuite/ver_test_13.script: New version script.
* testsuite/ver_test_13.sh: New test case.

9 years agoAutomatic date update in version.in
GDB Administrator [Wed, 19 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agogdbserver/linux-aarch32-low: build failure when NT_ARM_VFP not defined
Joel Brobecker [Tue, 11 Aug 2015 19:28:19 +0000 (15:28 -0400)]
gdbserver/linux-aarch32-low: build failure when NT_ARM_VFP not defined

On some older versions of GNU/Linux, gdbserver now fails to build
due to an undefined reference to NT_ARM_VFP. Same issue on Android,
where this macros is undefined until Android API level 21 (Android
5.0 "Lollipop").

This patch modifies linux-aarch32-low.c to define that macros when
not already defined.

gdb/gdbserver/ChangeLog:

        * linux-aarch32-low.c (NT_ARM_VFP): Define if not already defined.

9 years agoPR mi/18833 gdb.execute ("set param value", to_string=True) will crash gdb if using MI
Doug Evans [Tue, 18 Aug 2015 21:02:03 +0000 (14:02 -0700)]
PR mi/18833 gdb.execute ("set param value", to_string=True) will crash gdb if using MI

gdb/ChangeLog:

* cli/cli-logging.c (pop_output_files): Don't restore redirection
if MI-like.
* mi/mi-out.c: #include "vec.h".
(ui_filep): New type.
(DEV_VEC_P (ui_filep)): New type.
(struct ui_out_data) <buffer, original_buffer>: Delete.
(struct ui_out_data) <streams>: New member.
(mi_ui_out_impl): Add data_destroy field.
(mi_field_string, mi_field_fmt): Update.
(mi_flush, mi_redirect, field_separator): Update.
(mi_open, mi_close): Update.
(mi_out_buffered, mi_out_rewind, mi_out_put): Update.
(mi_out_data_ctor, mi_out_data_dtor): New functions.
(mi_out_new): Call mi_out_data_ctor.

testsuite/gdb/ChangeLog:

* lib/gdb.exp (skip_python_tests_prompt): Renamed from
skip_python_tests.  New arg prompt_regexp.
(skip_python_tests): New function.
* lib/mi-support.exp (mi_skip_python_tests): New function.
* gdb.python/py-mi-objfile-gdb.py: New file.
* gdb.python/py-mi-objfile.c: New file.
* gdb.python/py-mi-objfile.exp: New file.

9 years agoAdd another test for PR ld/18841
H.J. Lu [Tue, 18 Aug 2015 17:43:19 +0000 (10:43 -0700)]
Add another test for PR ld/18841

PR ld/18841
* ld-ifunc/ifunc.exp: Add another test for PR ld/18841.
* ld-ifunc/pr18841c.c: New file.

9 years agoFix mis-parsing of hex register numbers in 'T' stop replies.
Sandra Loosemore [Tue, 18 Aug 2015 17:29:54 +0000 (10:29 -0700)]
Fix mis-parsing of hex register numbers in 'T' stop replies.

2015-08-18  Sandra Loosemore  <sandra@codesourcery.com>

gdb/
* remote.c (strprefix): New.
(remote_parse_stop_reply): Use strprefix instead of strncmp
to ensure exact match of keyword.

9 years agoReturn reloc_class_ifunc for reloc against IFUNC
H.J. Lu [Tue, 18 Aug 2015 16:47:59 +0000 (09:47 -0700)]
Return reloc_class_ifunc for reloc against IFUNC

elf_XXX_reloc_type_class should return reloc_class_ifunc for relocation
against STT_GNU_IFUNC symbol.

bfd/

PR ld/18841
* elf-bfd.h (elf_link_hash_table): Add dynsym.
* elf32-i386.c (elf_i386_reloc_type_class): Return
reloc_class_ifunc for relocation against STT_GNU_IFUNC symbol.
* elf64-x86-64.c (elf_x86_64_reloc_type_class): Likewise.
* elflink.c (_bfd_elf_link_create_dynamic_sections): Set dynsym.
(bfd_elf_size_dynsym_hash_dynstr): Use dynsym.
(elf_final_link_info): Remove dynsym_sec.
(elf_link_output_extsym): Replace dynsym_sec with dynsym.
(bfd_elf_final_link): Remove reference to dynsym_sec.  Replace
dynsym_sec with dynsym.

ld/testsuite/

PR ld/18841
* ld-ifunc/ifunc.exp: Add a test for PR ld/18841.
* ld-ifunc/pr18841.out: New file.
* ld-ifunc/pr18841a.c: Likewise.
* ld-ifunc/pr18841b.c: Likewise.

9 years agogdb/doc: Fix build of 'info' manual.
Andrew Burgess [Tue, 18 Aug 2015 16:09:41 +0000 (17:09 +0100)]
gdb/doc: Fix build of 'info' manual.

In commit 18989b3c5642e2ba533e02737797dcf06134229b I broke the creation
of gdb's info manual; I added a new section without adding a suitable
menu entry.

This commit adds the missing menu entry and fixes the build of gdb's
info manual.

gdb/doc/ChangeLog:

* gdb.texinfo (GDB Files): Add 'File Caching' menu entry.

9 years ago[AArch64] Cleanup TLS relocation types which don't go through GOT table
Jiong Wang [Tue, 18 Aug 2015 15:50:56 +0000 (16:50 +0100)]
[AArch64] Cleanup TLS relocation types which don't go through GOT table

This patch done two types of cleanup:

  * in aarch64_reloc_got_type and elfNN_aarch64_relocate_section

    We don't need those redundant "case" check, as they can be merged
    with the "default" which just "break".

  * in elfNN_aarch64_gc_sweep_hook and elfNN_aarch64_check_relocs

    All TLS local executable relocations and some local dynamic
    relocations (those calculate module offset) actually don't need GOT
    entry, so remove them from GOT entry counting.

2015-08-18  Jiong Wang  <jiong.wang@arm.com>

bfd/
  * elfnn-aarch64.c (aarch64_reloc_got_type): Delete useless check.
  (elfNN_aarch64_relocate_section): Likewise.
  (elfNN_aarch64_gc_sweep_hook): Likewise.
  (elfNN_aarch64_check_relocs): Likewise.

9 years agogdb: Add debug tracing for bfd cache activity.
Andrew Burgess [Mon, 13 Apr 2015 15:56:23 +0000 (16:56 +0100)]
gdb: Add debug tracing for bfd cache activity.

This patch adds a new debug flag bfd-cache, which when set to non-zero
produces debugging log messages relating to gdb's bfd cache.

gdb/ChangeLog:

* gdb_bfd.c (debug_bfd_cache): New variable.
(show_bfd_cache_debug): New function.
(gdb_bfd_open): Add debug logging.
(gdb_bfd_ref): Likewise.
(gdb_bfd_unref): Likewise.
(_initialize_gdb_bfd): Add new set/show command.
* NEWS: Mention new command.

gdb/doc/ChangeLog:

* gdb.texinfo (File Caching): Document "set/show debug bfd-cache".

9 years agogdb: New maintenance command to disable bfd sharing.
Andrew Burgess [Mon, 13 Apr 2015 15:31:21 +0000 (16:31 +0100)]
gdb: New maintenance command to disable bfd sharing.

In some rare maintainer cases it is desirable to be able to disable bfd
sharing.  This patch adds new commands maintenance set/show commands for
bfd-sharing, allowing gdb's bfd cache to be turned off.

gdb/ChangeLog:

* gdb_bfd.c (bfd_sharing): New variable.
(show_bfd_sharing): New function.
(gdb_bfd_open): Check bfd_sharing variable.
(_initialize_gdb_bfd): Add new set/show command.
* NEWS: Mention new command.

gdb/doc/ChangeLog:

* gdb.texinfo (Maintenance Commands): Move documentation of "main
info bfds" to...
(File Caching): A New section.  Outline bfd caching, and add new
description for "main set/show bfd-sharing".

9 years agogdb: Improve cache matching criteria for the bfd cache.
Andrew Burgess [Mon, 13 Apr 2015 13:53:48 +0000 (14:53 +0100)]
gdb: Improve cache matching criteria for the bfd cache.

Within gdb open bfd objects are reused where possible if an attempt is
made to reopen a file that is already being debugged.  To spot if the on
disc file has changed gdb currently examines the mtime of the file and
compares it to the mtime of the open bfd in the cache.

A problem exists when the on disc file is being rapidly regenerated, as
happens, for example, with automated testing.  In some cases the file is
generated so quickly that the mtime appears not to change, while the on
disc file has changed.

This patch extends the bfd cache to also hold the file size of the file,
the inode of the file, and the device id of the file; gdb can then
compare filename, file size, mtime, inode, and device id to determine if
an existing bfd object can be reused.

gdb/ChangeLog:

* gdb_bfd.c (struct gdb_bfd_data): Add size, inode, and device id
field.
(struct gdb_bfd_cache_search): Likewise.
(eq_bfd): Compare the size, inode, and device id fields.
(gdb_bfd_open): Initialise the size, inode, and device id fields.
(gdb_bfd_ref): Likewise.
(gdb_bfd_unref): Likewise.

9 years agoAdd output_type to bfd_link_info
H.J. Lu [Tue, 18 Aug 2015 12:51:03 +0000 (05:51 -0700)]
Add output_type to bfd_link_info

The "shared" field in bfd_link_info is set for both DSO and and PIE.
There are separate fields for executable and relocatable outputs.  This
patch adds an "output_type" field:

enum output_type
{
  type_unknown = 0,
  type_executable,
  type_dll,
  type_relocatable
};

and a "pic" field to bfd_link_info to replace shared, executable and
relocatable fields so that we can use the "output_type" field to check
for output type and the "pic" field check if output is PIC.  Macros,
bfd_link_executable, bfd_link_dll, bfd_link_relocatable, bfd_link_pic
and bfd_link_pie, are provided to check for output features.

bfd/

* bfd/aoutx.h: Replace shared, executable, relocatable and pie
fields with bfd_link_executable, bfd_link_dll,
bfd_link_relocatable, bfd_link_pic and bfd_link_pie.
* bfd/bout.c: Likewise.
* bfd/coff-alpha.c: Likewise.
* bfd/coff-arm.c: Likewise.
* bfd/coff-i386.c: Likewise.
* bfd/coff-i960.c: Likewise.
* bfd/coff-m68k.c: Likewise.
* bfd/coff-mcore.c: Likewise.
* bfd/coff-mips.c: Likewise.
* bfd/coff-ppc.c: Likewise.
* bfd/coff-rs6000.c: Likewise.
* bfd/coff-sh.c: Likewise.
* bfd/coff-tic80.c: Likewise.
* bfd/coff-x86_64.c: Likewise.
* bfd/coff64-rs6000.c: Likewise.
* bfd/coffgen.c: Likewise.
* bfd/cofflink.c: Likewise.
* bfd/ecoff.c: Likewise.
* bfd/ecofflink.c: Likewise.
* bfd/elf-bfd.h: Likewise.
* bfd/elf-eh-frame.c: Likewise.
* bfd/elf-ifunc.c: Likewise.
* bfd/elf-m10200.c: Likewise.
* bfd/elf-m10300.c: Likewise.
* bfd/elf-s390-common.c: Likewise.
* bfd/elf-vxworks.c: Likewise.
* bfd/elf.c: Likewise.
* bfd/elf32-arm.c: Likewise.
* bfd/elf32-avr.c: Likewise.
* bfd/elf32-bfin.c: Likewise.
* bfd/elf32-cr16.c: Likewise.
* bfd/elf32-cr16c.c: Likewise.
* bfd/elf32-cris.c: Likewise.
* bfd/elf32-crx.c: Likewise.
* bfd/elf32-d10v.c: Likewise.
* bfd/elf32-dlx.c: Likewise.
* bfd/elf32-epiphany.c: Likewise.
* bfd/elf32-fr30.c: Likewise.
* bfd/elf32-frv.c: Likewise.
* bfd/elf32-ft32.c: Likewise.
* bfd/elf32-h8300.c: Likewise.
* bfd/elf32-hppa.c: Likewise.
* bfd/elf32-i370.c: Likewise.
* bfd/elf32-i386.c: Likewise.
* bfd/elf32-i860.c: Likewise.
* bfd/elf32-ip2k.c: Likewise.
* bfd/elf32-iq2000.c: Likewise.
* bfd/elf32-lm32.c: Likewise.
* bfd/elf32-m32c.c: Likewise.
* bfd/elf32-m32r.c: Likewise.
* bfd/elf32-m68hc11.c: Likewise.
* bfd/elf32-m68hc1x.c: Likewise.
* bfd/elf32-m68k.c: Likewise.
* bfd/elf32-mcore.c: Likewise.
* bfd/elf32-mep.c: Likewise.
* bfd/elf32-metag.c: Likewise.
* bfd/elf32-microblaze.c: Likewise.
* bfd/elf32-moxie.c: Likewise.
* bfd/elf32-msp430.c: Likewise.
* bfd/elf32-mt.c: Likewise.
* bfd/elf32-nds32.c: Likewise.
* bfd/elf32-nios2.c: Likewise.
* bfd/elf32-or1k.c: Likewise.
* bfd/elf32-ppc.c: Likewise.
* bfd/elf32-rl78.c: Likewise.
* bfd/elf32-rx.c: Likewise.
* bfd/elf32-s390.c: Likewise.
* bfd/elf32-score.c: Likewise.
* bfd/elf32-score7.c: Likewise.
* bfd/elf32-sh-symbian.c: Likewise.
* bfd/elf32-sh.c: Likewise.
* bfd/elf32-sh64.c: Likewise.
* bfd/elf32-spu.c: Likewise.
* bfd/elf32-tic6x.c: Likewise.
* bfd/elf32-tilepro.c: Likewise.
* bfd/elf32-v850.c: Likewise.
* bfd/elf32-vax.c: Likewise.
* bfd/elf32-visium.c: Likewise.
* bfd/elf32-xc16x.c: Likewise.
* bfd/elf32-xstormy16.c: Likewise.
* bfd/elf32-xtensa.c: Likewise.
* bfd/elf64-alpha.c: Likewise.
* bfd/elf64-hppa.c: Likewise.
* bfd/elf64-ia64-vms.c: Likewise.
* bfd/elf64-mmix.c: Likewise.
* bfd/elf64-ppc.c: Likewise.
* bfd/elf64-s390.c: Likewise.
* bfd/elf64-sh64.c: Likewise.
* bfd/elf64-x86-64.c: Likewise.
* bfd/elflink.c: Likewise.
* bfd/elfnn-aarch64.c: Likewise.
* bfd/elfnn-ia64.c: Likewise.
* bfd/elfxx-mips.c: Likewise.
* bfd/elfxx-sparc.c: Likewise.
* bfd/elfxx-tilegx.c: Likewise.
* bfd/i386linux.c: Likewise.
* bfd/linker.c: Likewise.
* bfd/m68klinux.c: Likewise.
* bfd/pdp11.c: Likewise.
* bfd/pe-mips.c: Likewise.
* bfd/peXXigen.c: Likewise.
* bfd/reloc.c: Likewise.
* bfd/reloc16.c: Likewise.
* bfd/sparclinux.c: Likewise.
* bfd/sunos.c: Likewise.
* bfd/vms-alpha.c: Likewise.
* bfd/xcofflink.c: Likewise.

include/

* include/bfdlink.h (output_type): New enum.
(bfd_link_executable): New macro.
(bfd_link_dll): Likewise.
(bfd_link_relocatable): Likewise.
(bfd_link_pic): Likewise.
(bfd_link_pie): Likewise.
(bfd_link_info): Remove shared, executable, pie and relocatable.
Add output_type and pic.

ld/

* ld/ldctor.c: Replace shared, executable, relocatable and pie
fields with bfd_link_executable, bfd_link_dll,
bfd_link_relocatable, bfd_link_pic and bfd_link_pie.
* ld/ldemul.c: Likewise.
* ld/ldfile.c: Likewise.
* ld/ldlang.c: Likewise.
* ld/ldmain.c: Likewise.
* ld/ldwrite.c: Likewise.
* ld/lexsup.c: Likewise.
* ld/pe-dll.c: Likewise.
* ld/plugin.c: Likewise.
* ld/emultempl/aarch64elf.em: Likewise.
* ld/emultempl/aix.em: Likewise.
* ld/emultempl/alphaelf.em: Likewise.
* ld/emultempl/armcoff.em: Likewise.
* ld/emultempl/armelf.em: Likewise.
* ld/emultempl/avrelf.em: Likewise.
* ld/emultempl/beos.em: Likewise.
* ld/emultempl/cr16elf.em: Likewise.
* ld/emultempl/elf-generic.em: Likewise.
* ld/emultempl/elf32.em: Likewise.
* ld/emultempl/genelf.em: Likewise.
* ld/emultempl/generic.em: Likewise.
* ld/emultempl/gld960.em: Likewise.
* ld/emultempl/gld960c.em: Likewise.
* ld/emultempl/hppaelf.em: Likewise.
* ld/emultempl/irix.em: Likewise.
* ld/emultempl/linux.em: Likewise.
* ld/emultempl/lnk960.em: Likewise.
* ld/emultempl/m68hc1xelf.em: Likewise.
* ld/emultempl/m68kcoff.em: Likewise.
* ld/emultempl/m68kelf.em: Likewise.
* ld/emultempl/metagelf.em: Likewise.
* ld/emultempl/mipself.em: Likewise.
* ld/emultempl/mmo.em: Likewise.
* ld/emultempl/msp430.em: Likewise.
* ld/emultempl/nds32elf.em: Likewise.
* ld/emultempl/needrelax.em: Likewise.
* ld/emultempl/nios2elf.em: Likewise.
* ld/emultempl/pe.em: Likewise.
* ld/emultempl/pep.em: Likewise.
* ld/emultempl/ppc32elf.em: Likewise.
* ld/emultempl/ppc64elf.em: Likewise.
* ld/emultempl/sh64elf.em: Likewise.
* ld/emultempl/solaris2.em: Likewise.
* ld/emultempl/spuelf.em: Likewise.
* ld/emultempl/sunos.em: Likewise.
* ld/emultempl/tic6xdsbt.em: Likewise.
* ld/emultempl/ticoff.em: Likewise.
* ld/emultempl/v850elf.em: Likewise.
* ld/emultempl/vms.em: Likewise.
* ld/emultempl/vxworks.em: Likewise.

9 years agox86/Linux: disable all-stop on top of non-stop
Pedro Alves [Tue, 18 Aug 2015 10:04:30 +0000 (11:04 +0100)]
x86/Linux: disable all-stop on top of non-stop

Markus reported that ASNS breaks target record-btrace.  In particular,
the gdb.btrace/multi-thread-step.exp test fails (both with BTS and PT
tracing) with a crash in py-inferior.c:

 Program received signal SIGSEGV, Segmentation fault.

 0x00000000006aa40d in add_thread_object (tp=0x27d32d0)

     at /users/mmetzger/team/gdb/git/gdb/python/py-inferior.c:337

 337       entry->next = inf_obj->threads;

My machine doesn't support BTS nor PT, so I missed this...

Disabling ASNS temporarily on x86 until this is addressed.

Tested on x86_64 Fedora 20.

gdb/ChangeLog:
2015-08-18  Pedro Alves  <palves@redhat.com>

* linux-nat.c (linux_nat_always_non_stop_p): If the linux_ops
target implements to_always_non_stop_p, call it.
* x86-linux-nat.c (x86_linux_always_non_stop_p): New function.
(x86_linux_create_target): Install it as to_always_non_stop_p
method.

9 years agoUse $SED in bfd Makefile
Alan Modra [Tue, 18 Aug 2015 07:14:48 +0000 (16:44 +0930)]
Use $SED in bfd Makefile

PR 18667
* Makefile.am: Use $(SED) in place of sed throughout.
* Makefile.in: Regenerate.

9 years agoPPC64: Allow .TOC. in linker script to override backend calculated value
Alan Modra [Tue, 18 Aug 2015 07:13:18 +0000 (16:43 +0930)]
PPC64: Allow .TOC. in linker script to override backend calculated value

bfd/
* elf64-ppc.c (ppc64_elf_func_desc_adjust): Don't redefine .TOC.
if already defined, and set linker_def.
(ppc64_elf_set_toc): Use .TOC. value if defined other than by
the backend.
ld/
* ldexp.c (exp_fold_tree_1): Clear linker_def on symbol assignment.

9 years agoAutomatic date update in version.in
GDB Administrator [Tue, 18 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoui-out.c (default_ui_out_impl): Add comment.
Doug Evans [Mon, 17 Aug 2015 20:07:11 +0000 (13:07 -0700)]
ui-out.c (default_ui_out_impl): Add comment.

gdb/ChangeLog:

* ui-out.c (default_ui_out_impl): Add comment.

9 years ago[D] Implement looking up members of D enums.
Iain Buclaw [Mon, 17 Aug 2015 19:45:42 +0000 (21:45 +0200)]
[D] Implement looking up members of D enums.

In D, all named enums are explicitly scoped (the C++ equivalent of enum class)
so they should be handled as such in the language-specific symbol lookup
routines.  However so as to support D compilers that don't emit enums as
DW_AT_enum_class, need to make sure that appropriate checks for
TYPE_DECLARED_CLASS are done.

gdb/ChangeLog

* d-exp.y (type_aggregate_p): New function.
(PrimaryExpression : TypeExp '.' IdentifierExp): Use it.
(classify_inner_name): Likewise.
* d-namespace.c (d_lookup_nested_symbol): Handle TYPE_CODE_ENUM.

9 years agoMove strace -m/explicit location test to strace.exp
Keith Seitz [Thu, 13 Aug 2015 18:56:50 +0000 (11:56 -0700)]
Move strace -m/explicit location test to strace.exp

One of the build slaves shows this error running explicit.exp:

(gdb) strace -m gdbfoobarbaz
Remote failure reply: E.In-process agent library not loaded in process.
Fast and static tracepoints unavailable.
(gdb) FAIL: gdb.linespec/explicit.exp: strace -m gdbfoobarbaz

There are two big problems with this test:
1) The expected output is actually not what the test is meant to test for.
2) This test should really only run where it is supported.

This is most easily fixed by moving the test to gdb.trace/strace.exp.

gdb/testsuite/ChangeLog

* gdb.linespec/explicit.exp: Move strace test from here ...
* gdb.trace/strace.exp: ... to here.

9 years agoRegen binutils/configure
Alan Modra [Mon, 17 Aug 2015 02:31:58 +0000 (12:01 +0930)]
Regen binutils/configure

Looks like previous regen wasn't using current sources.

* configure: Regenerate.

9 years agoTrailing space in opcodes/ generated files
Alan Modra [Mon, 17 Aug 2015 02:21:44 +0000 (11:51 +0930)]
Trailing space in opcodes/ generated files

HJ recently removed trailing space in binutils files, but unfortunately
they return when regenerating files in opcodes.  This fixes the regen
process.

* cgen.sh: Trim trailing space from cgen output.
* ia64-gen.c (print_dependency_table): Don't generate trailing space.
(print_dis_table): Likewise.
* opc2c.c (dump_lines): Likewise.
(orig_filename): Warning fix.
* ia64-asmtab.c: Regenerate.

9 years agoAutomatic date update in version.in
GDB Administrator [Mon, 17 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoRationalize ARM .align
Alan Modra [Sun, 16 Aug 2015 23:35:42 +0000 (09:05 +0930)]
Rationalize ARM .align

* gas/config/tc-arm.c (s_align): Delete.
(md_pseudo_table): Use s_align_ptwo for "align".
* gas/config/tc-arm.h (TC_ALIGN_ZERO_IS_DEFAULT): Define.
* read.c (s_align): Modify for TC_ALIGN_ZERO_IS_DEFAULT.

9 years agopsymtab.c (add_psymbol_to_bcache): Remove "val" arg.
Doug Evans [Sun, 16 Aug 2015 05:08:47 +0000 (22:08 -0700)]
psymtab.c (add_psymbol_to_bcache): Remove "val" arg.

gdb/ChangeLog:

* psymtab.c (add_psymbol_to_bcache): Remove "val" arg.  All callers
updated.
(add_psymbol_to_list): Ditto.

9 years agoAutomatic date update in version.in
GDB Administrator [Sun, 16 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoAdd end_psymtab_common, have all debug info readers call it.
Doug Evans [Sat, 15 Aug 2015 23:46:20 +0000 (16:46 -0700)]
Add end_psymtab_common, have all debug info readers call it.

gdb/ChangeLog:

* dbxread.c (dbx_end_psymtab): Renamed from end_psymtab.  All callers
updated.  Call end_psymtab_common.
* dwarf2read.c (process_psymtab_comp_unit_reader): Call
end_psymtab_common.
(build_type_psymtabs_reader): Ditto.
* psympriv.h (sort_pst_symbols): Delete.
(end_psymtab_common): Declare.
* psymtab.c (sort_pst_symbols): Make static.
(end_psymtab_common): New function.
* xcoffread.c (xcoff_end_psymtab): Call end_psymtab_common.

9 years agoUse macros for some enum bit field sizes.
Doug Evans [Sat, 15 Aug 2015 23:25:53 +0000 (16:25 -0700)]
Use macros for some enum bit field sizes.

gdb/ChangeLog:

* defs.h (LANGUAGE_BITS): Define.
* psympriv.h (partial_symbol) <domain>: Use SYMBOL_DOMAIN_BITS.
(partial_symbol) <aclass>: Use SYMBOL_ACLASS_BITS.
* symtab.h (general_symbol_info> <language>: Usage LANGUAGE_BITS.
(minimal_symbol_type): Add nr_minsym_types.
(MINSYM_TYPE_BITS): Define.
(minimal_symbol) <type>: Use MINSYM_TYPE_BITS.
(domain_enum_tag): Add NR_DOMAINS.
(SYMBOL_DOMAIN_BITS): Change from 4 to 3.
(SYMBOL_ACLASS_BITS): Define from 6 to 5.

9 years agoobjfiles.h,psympriv.h,psymtab.c: Whitespace.
Doug Evans [Sat, 15 Aug 2015 22:51:00 +0000 (15:51 -0700)]
objfiles.h,psympriv.h,psymtab.c: Whitespace.

gdb/ChangeLog:

* objfiles.h: Whitespace cleanup.
* psympriv.h: Whitespace cleanup.
* psymtab.c: Whitespace/coding convention cleanup.

9 years agoFix invoking "[kill|detach] inferiors" on inferiors that are not running
Patrick Palka [Tue, 28 Jul 2015 19:00:58 +0000 (15:00 -0400)]
Fix invoking "[kill|detach] inferiors" on inferiors that are not running

Invoking either of the above commands on an inferior that's not running
triggers the following assert failure:

  .../binutils-gdb/gdb/thread.c:514: internal-error: any_thread_of_process: Assertion `pid != 0' failed.

The fix is straightforward.  This patch also adds a test to check the
basic functionality of these commands, along with testing this fix in
particular.  Tested on x86_64 Linux.

gdb/ChangeLog:

* inferior.c (detach_inferior_command): Don't call
any_thread_of_process when pid is 0.
(kill_inferior_command): Likewise.

gdb/testsuite/ChangeLog:

* gdb.base/kill-detach-inferiors-cmd.exp: New test file.
* gdb.base/kill-detach-inferiors-cmd.c: New test file.

9 years agoperftest/utils.py (select_file): Kill any existing inferior before selecting a new...
Doug Evans [Sat, 15 Aug 2015 06:29:04 +0000 (23:29 -0700)]
perftest/utils.py (select_file): Kill any existing inferior before selecting a new file.

gdb/testsuite/ChangeLog:

* gdb.perf/lib/perftest/utils.py (select_file): Kill any existing
inferior before selecting a new file.

9 years agoNew /s modifier for the disassemble command.
Doug Evans [Sat, 15 Aug 2015 04:45:54 +0000 (21:45 -0700)]
New /s modifier for the disassemble command.

The "source centric" /m option to the disassemble command is often
unhelpful, e.g., in the presence of optimized code.
This patch adds a /s modifier that is better.
For one, /m only prints instructions from the originating source file,
leaving out instructions from e.g., inlined functions from other files.

gdb/ChangeLog:

PR gdb/11833
* NEWS: Document new /s modifier for the disassemble command.
* cli/cli-cmds.c (disassemble_command): Add support for /s.
(_initialize_cli_cmds): Update online docs of disassemble command.
* disasm.c: #include "source.h".
(struct deprecated_dis_line_entry): Renamed from dis_line_entry.
All uses updated.
(dis_line_entry): New struct.
(hash_dis_line_entry, eq_dis_line_entry): New functions.
(allocate_dis_line_table): New functions.
(maybe_add_dis_line_entry, line_has_code_p): New functions.
(dump_insns): New arg end_pc.  All callers updated.
(do_mixed_source_and_assembly_deprecated): Renamed from
do_mixed_source_and_assembly.  All callers updated.
(do_mixed_source_and_assembly): New function.
(gdb_disassembly): Handle /s (DISASSEMBLY_SOURCE).
* disasm.h (DISASSEMBLY_SOURCE_DEPRECATED): Renamed from
DISASSEMBLY_SOURCE.  All uses updated.
(DISASSEMBLY_SOURCE): New macro.
* mi/mi-cmd-disas.c (mi_cmd_disassemble): New modes 4,5.

gdb/doc/ChangeLog:

* gdb.texinfo (Machine Code): Update docs for mixed source/assembly
disassembly.
(GDB/MI Data Manipulation): Update docs for new disassembly modes.

gdb/testsuite/ChangeLog:

* gdb.mi/mi-disassemble.exp: Update.
* gdb.base/disasm-optim.S: New file.
* gdb.base/disasm-optim.c: New file.
* gdb.base/disasm-optim.h: New file.
* gdb.base/disasm-optim.exp: New file.

9 years agoRename `typename' in d-exp.y to avoid C++ reserved word
Keith Seitz [Sat, 15 Aug 2015 00:28:11 +0000 (17:28 -0700)]
Rename `typename' in d-exp.y to avoid C++ reserved word

A recent patch introduced a variable named `typename' into d-exp.y,
and one of the --enable-with-cxx build slaves consequently failed to compile
this.  This patch simply adds an underscore into the name to avoid the
reserved word.

gdb/ChangeLog

* d-exp.y (PrimaryExpression : TypeExp '.' IdentifierExp): Rename
`typename' to `type_name' to avoid C++ reserved word.

9 years agoAutomatic date update in version.in
GDB Administrator [Sat, 15 Aug 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoRename location accessor macro parameters to silence ARI
Keith Seitz [Fri, 14 Aug 2015 22:00:04 +0000 (15:00 -0700)]
Rename location accessor macro parameters to silence ARI

The locations patch I recently committed contains macro definitions
such as:

This causes an ARI error to be emitted by the server ("Do not use PTR, ISO C
90 implies `void *'").  While this ARI error is bogus in this context,
it is just easiest to squash the error completely by renaming the macro
parameters.

gdb/ChangeLog

* location.c (EL_TYPE, EL_LINESPEC, EL_PROBE, EL_ADDRESS)
(EL_EXPLICIT, EL_STRING): Change macro parameter to "P" to
silence ARI errors.

9 years agoAdd missing ChangeLog entry for previous commit.
Keith Seitz [Fri, 14 Aug 2015 21:54:37 +0000 (14:54 -0700)]
Add missing ChangeLog entry for previous commit.

9 years agomi_make_breakpoint: add "evaluated-by" option
Keith Seitz [Fri, 14 Aug 2015 20:45:06 +0000 (13:45 -0700)]
mi_make_breakpoint: add "evaluated-by" option

For some time now, GDB has permitted target-side evaluation of
breakpoint conditions.  On targets that support this feature, GDB
may output an "evaluated-by" field into the breakpoint reply.

This patch adds handling for this option, and outputs a default
pattern to optionally recognize (and ignore) this pattern in the
reply.

gdb/testsuite/ChangeLog

* lib/mi-support.exp (mi_make_breakpoint): Add option/handling for
"evaluated-by".

9 years agoFix ARI warnings in d-exp.y
Iain Buclaw [Fri, 14 Aug 2015 19:25:17 +0000 (21:25 +0200)]
Fix ARI warnings in d-exp.y

This fixes four ARI warnings found in d-exp.y.

This is comprised of three uses of the && or || at the end of a line, and one
use of sprintf.

gdb/ChangeLog

* d-exp.y (PrimaryExpression : TypeExp '.' IdentifierExp): Use
xstrprintf instead of malloc and sprintf.
(PrimaryExpression : IdentifierExp): Avoid operator at end of line.
(lex_one_token): Likewise.

9 years agoAdd support for DT_MIPS_RLD_MAP_REL.
Matthew Fortune [Thu, 11 Jun 2015 09:37:31 +0000 (10:37 +0100)]
Add support for DT_MIPS_RLD_MAP_REL.

This tag allows debugging of MIPS position independent executables
and provides access to shared library information.

gdb/gdbserver/

* linux-low.c (get_r_debug): Handle DT_MIPS_RLD_MAP_REL.

gdb/

* solib-svr4.c (read_program_header): Add base_addr argument to
report the runtime address of the segment.
(find_program_interpreter): Update read_program_header call to pass
a NULL pointer for the new argument.
(scan_dyntag): Add ptr_addr argument to report the runtime address
of the tag payload.
(scan_dyntag_auxv): Likewise and use thew new base_addr argument of
read_program_header to get the base address of the dynamic segment.
(elf_locate_base): Update uses of scan_dyntag, scan_dyntag_auxv and
read_program_header.
(elf_locate_base): Scan for and handle DT_MIPS_RLD_MAP_REL.

9 years agoAdd myself to gdb MAINTAINERS
Matthew Fortune [Fri, 14 Aug 2015 12:07:23 +0000 (13:07 +0100)]
Add myself to gdb MAINTAINERS

gdb/
* MAINTAINERS (Write After Approval): Add Matthew Fortune.

9 years agoRevert "Fix encoding or OpenRisk1000 PC relative relocations."
Alan Modra [Wed, 12 Aug 2015 09:37:26 +0000 (19:07 +0930)]
Revert "Fix encoding or OpenRisk1000 PC relative relocations."

This reverts commit dbac553d28887561e3f154654ec8e70195d89943.

PR ld/18759
* elf32-or1k.c: Revert 2015-08-11 change.

9 years agoAutomatic date update in version.in
GDB Administrator [Fri, 14 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years ago[D] Move classification of symbols from the grammar to the lexer.
Iain Buclaw [Thu, 13 Aug 2015 19:35:09 +0000 (21:35 +0200)]
[D] Move classification of symbols from the grammar to the lexer.

This makes it so that alternating '.' and identifier tokens are resolved to
symbols as early as possible, which should all the addition of D properties -
such as EXP.sizeof and EXP.typeof - without the shift/reduce conflicts that
would occur in the current parsing strategy.

gdb/ChangeLog

* d-exp.y (%union): Add voidval.
(%token): Add UNKNOWN_NAME as a token to represent an unclassified
name in the lexing stage.
(PostfixExpression): Move symbol completion handling in grammar here
from PrimaryExpression.
(PrimaryExpression): Move routines to handle resolving identifier
tokens in the grammar here from push_expression_name.
(IdentifierExp): Remove the handling of alternating '.' and identifier
tokens.
(TypeExp): Allow TypeExp to be wrapped in parenthesis in the grammar.
(BasicType): Remove C-style typename rules.
(d_type_from_name, d_module_from_name, push_variable)
(push_fieldnames, push_type_name, push_module_name)
(push_expression_name): Remove.
(lex_one_token): Rename from yylex.  Replace pstate with par_state.
(token_and_value): New type.
(token_fifo, popping, name_obstack): New globals.
(classify_name): New function.
(classify_inner_name): Likewise.
(yylex): Likewise.
(d_parse): Initialize token_fifo, popping and name_obstack.

9 years ago[D] Support looking up symbols in the current and imported modules.
Iain Buclaw [Thu, 13 Aug 2015 18:34:15 +0000 (20:34 +0200)]
[D] Support looking up symbols in the current and imported modules.

In D, there is the notion of modules, and importing from one to the other,
whether it is a basic, selective or renamed import declaration.

module A;
import X;
void foo() {
  import Y : bar;
}

If the compiler emits DW_TAG_imported_declaration at the appropriate locations,
then we can make use of what gdb stores in using_direct when performing
nonlocal symbol lookups.

gdb/ChangeLog

* Makefile.in (SFILES): Add d-namespace.c.
(COMMON_OBS): Add d-namespace.o.
* d-lang.c (d_language_defn): Use d_lookup_symbol_nonlocal as the
la_lookup_symbol_nonlocal callback function pointer.
* d-lang.h (d_lookup_symbol_nonlocal): New declaration.
(d_lookup_nested_symbol): New declaration.
* d-namespace.c: New file.

9 years agogdb.base/dso2dso.exp: Improve testcase documentation.
Joel Brobecker [Wed, 12 Aug 2015 20:40:54 +0000 (13:40 -0700)]
gdb.base/dso2dso.exp: Improve testcase documentation.

gdb/testsuite/ChangeLog:

        * gdb.base/dso2dso.exp: Improve the testcase's documentation.

9 years agoFix Python frame unwinder issue caught by Valgrind
Pedro Alves [Thu, 13 Aug 2015 17:56:42 +0000 (18:56 +0100)]
Fix Python frame unwinder issue caught by Valgrind

Valgrind shows:

 ==17026== Invalid write of size 8
 ==17026==    at 0x54AA80: pending_frame_invalidate (py-unwind.c:477)
 ==17026==    by 0x5AB934: do_my_cleanups (cleanups.c:155)
 ==17026==    by 0x5AB9AF: do_cleanups (cleanups.c:177)
 ==17026==    by 0x54B009: pyuw_sniffer (py-unwind.c:606)
 ==17026==    by 0x755DAC: frame_unwind_try_unwinder (frame-unwind.c:105)
 ==17026==    by 0x755EEE: frame_unwind_find_by_frame (frame-unwind.c:160)
 ==17026==    by 0x750FFA: compute_frame_id (frame.c:454)
 ==17026==    by 0x753BD6: get_prev_frame_if_no_cycle (frame.c:1781)
 ==17026==    by 0x754292: get_prev_frame_always_1 (frame.c:1955)
 ==17026==    by 0x7542DA: get_prev_frame_always (frame.c:1971)
 ==17026==    by 0x7547BE: get_prev_frame (frame.c:2213)
 ==17026==    by 0x7532BD: unwind_to_current_frame (frame.c:1450)
 ==17026==  Address 0xd27b570 is 16 bytes inside a block of size 32 free'd
 ==17026==    at 0x4A07577: free (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
 ==17026==    by 0x54B276: gdb_Py_DECREF (python-internal.h:185)
 ==17026==    by 0x54B298: py_decref (py-utils.c:34)
 ==17026==    by 0x5AB934: do_my_cleanups (cleanups.c:155)
 ==17026==    by 0x5AB9AF: do_cleanups (cleanups.c:177)
 ==17026==    by 0x54B009: pyuw_sniffer (py-unwind.c:606)
 ==17026==    by 0x755DAC: frame_unwind_try_unwinder (frame-unwind.c:105)
 ==17026==    by 0x755EEE: frame_unwind_find_by_frame (frame-unwind.c:160)
 ==17026==    by 0x750FFA: compute_frame_id (frame.c:454)
 ==17026==    by 0x753BD6: get_prev_frame_if_no_cycle (frame.c:1781)
 ==17026==    by 0x754292: get_prev_frame_always_1 (frame.c:1955)
 ==17026==    by 0x7542DA: get_prev_frame_always (frame.c:1971)
 ==17026==

Simply invalidate the object before releasing it.

gdb/ChangeLog:
2015-08-13  Pedro Alves  <palves@redhat.com>

* python/py-unwind.c (pyuw_sniffer): Install the invalidate
cleanup after the decref cleanup, not before.

9 years agoMass rename `explicit' -> `explicit_loc'.
Keith Seitz [Thu, 13 Aug 2015 16:48:12 +0000 (09:48 -0700)]
Mass rename `explicit' -> `explicit_loc'.

BuildBot reminded me that "explicit" is a reserved keyword in C++.
This patch simply renames all the (illegal) uses of "explicit". This should
fix the build errors with --enable-build-with-cxx bots.

gdb/ChangeLog

* break-catch-throw.c (re_set_exception_catchpoint) Rename
reserved C++ keyword "explicit" to "explicit_loc".
* breakpoint.c (create_overlay_event_breakpoint)
(create_longjmp_master_breakpoint)
(create_std_terminate_master_breakpoint)
(create_exception_master_breakpoint, update_static_tracepoint):
Rename reserved C++ keyword "explicit" to "explicit_loc".
* completer.c (collect_explicit_location_matches)
(explicit_location_completer): Rename reserved C++ keyword
"explicit" to "explicit_loc".
* linespec.c (struct linespec) <explicit>: Rename to "explicit_loc".
(canonicalize_linespec, create_sals_line_offset)
(convert_linespec_to_sals, convert_explicit_location_to_sals)
(event_location_to_sals, decode_objc): Rename reserved C++ keyword
"explicit" to "explicit_loc".
* location.c (struct event_location) <explicit>: Rename to
"explicit_loc".
(initialize_explicit_location, new_explicit_location)
(explicit_location_to_string_internal, explicit_location_to_linespec):
Rename reserved C++ keyword "explicit" to "explicit_loc".
* location.h (explicit_location_to_string)
(explicit_location_to_linespec, initialize_explicit_location)
(new_explicit_location): Rename reserved C++ keyword "explicit"
to "explicit_loc".
* mi/mi-cmd-break.c (mi_cmd_break_insert_1): Rename reserved C++
keyword "explicit" to "explicit_loc".

9 years agoIssue an error for read-only segment with dynamic IFUNC relocations
H.J. Lu [Thu, 13 Aug 2015 11:31:38 +0000 (04:31 -0700)]
Issue an error for read-only segment with dynamic IFUNC relocations

To load an ELF binary with DT_TEXTREL tag, the dynamic linker calls
__mprotect on the read-only segment with PROT_READ|PROT_WRITE before
applying dynamic relocation.  It leads to segfault when performing
IFUNC relocations since the read-only segment has no execute permission.
This patch changes x86 linker to issue an error for read-only segment
with dynamic IFUNC relocations.  Other backends with IFUNC support
may need a similar change.

bfd/

PR ld/18801
* elf32-i386.c (elf_i386_size_dynamic_sections): Issue an error
for read-only segment with dynamic IFUNC relocations.
* elf64-x86-64.c (elf_x86_64_size_dynamic_sections): Likewise.

ld/testsuite/

PR ld/18801
* ld-i386/i386.exp: Run pr18801.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr18801.d: New file.
* ld-i386/pr18801.s: Likewise.
* ld-x86-64/pr18801.d: Likewise.
* ld-x86-64/pr18801.s: Likewise.

9 years agoFixes for unpredictable nops and 26-bit versions of teq,tst,cmn,cmp.
Andre Vieira [Thu, 13 Aug 2015 10:39:08 +0000 (11:39 +0100)]
Fixes for unpredictable nops and 26-bit versions of teq,tst,cmn,cmp.

opcodes * arm-dis.c (print_insn_arm): Disassembling for all targets V6
and higher with ARM instruction set will now mark the 26-bit
versions of teq,tst,cmn and cmp as UNPREDICTABLE.
(arm_opcodes): Fix for unpredictable nop being recognized as a teq.

test    * gas/arm/nops.d: New.
* gas/arm/nops.s: New.
* gas/arm/inst.d: Changed expectation file for 26-bit  teq,
tst, cmn and cmp.

9 years agoAdd ChangeLog for check_ifunc_attribute_available
H.J. Lu [Thu, 13 Aug 2015 09:48:55 +0000 (02:48 -0700)]
Add ChangeLog for check_ifunc_attribute_available

9 years agoAdd ChangeLog entries for the previous commit
Pierre-Marie de Rodat [Thu, 13 Aug 2015 07:54:09 +0000 (09:54 +0200)]
Add ChangeLog entries for the previous commit

9 years ago[Ada] Add support for subprogram renamings
Pierre-Marie de Rodat [Wed, 22 Jul 2015 13:30:57 +0000 (15:30 +0200)]
[Ada] Add support for subprogram renamings

Consider the following declaration:

    function Foo (I : Integer) return Integer renames Pack.Bar;

As Foo is not materialized as a routine whose name is derived from Foo,
GDB currently cannot use it:

    (gdb) print foo(0)
    No definition of "foo" in current context.

However, compilers can emit DW_TAG_imported_declaration in order to
materialize the fact that Foo is actually another name for Pack.Bar.
This commit enhances the DWARF reader to record global renamings (it
used to put global ones in a static block) and enhances the Ada engine
to leverage this information during symbol lookup.

gdb/ChangeLog:

* ada-lang.c: Include namespace.h
(aux_add_nonlocal_symbols): Fix a function name in comment.
(ada_add_block_renamings): New.
(add_nonlocal_symbols): Add global renamings handling.
(ada_lookup_symbol_list_worker): Move the symbol lookup part
to...
(ada_add_all_symbols): ... this new function.
(ada_add_block_symbols): Try to match the input name against the
"using directives list", perform a recursive symbol lookup on
the matched declarations.
* block.h (struct block): Move the_namespace to top-level as
namespace_info. Remove the language_specific field.
(BLOCK_NAMESPACE): Update access to the namespace_info field.
* buildsym.h (using_directives): Rename into...
(local_using_directives): ... this.
(global_using_directives): New.
(struct context_stack): Rename the using_directives field into
local_using_directives.
* buildsym.c (finish_block_internal): Deal with the proper
using directives repository (local or global).
(prepare_for_building): Reset local_using_directives. Assert
that there is no pending global using directive.
(reset_symtab_globals): Reset global_using_directives and
local_using_directives.
(end_symtab_get_static_block): Don't ignore symtabs that have
only using directives.
(push_context): Update references to local_using_directives.
(buildsym_init): Do not reset using_directives.
* cp-support.c: Include namespace.h.
* cp-support.h (struct using_direct): Move to namespace.h.
(cp_add_using_directives): Move to namespace.h.
* cp-namespace.c: Include namespace.h
(cp_add_using_directive): Move to namespace.c, rename it to
add_using_directive, add a "using_directives" argument and use
it as the pending using directives repository.  All callers
updated.
* dwarf2read.c (using_directives): New.
(read_import_statement): Call using_directives.
(read_func_scope): Update references to local_using_directives.
(read_lexical_block_scope): Likewise.
(read_namespace): Update the heading comment, call
using_directives.
* namespace.h: New file.
* namespace.c: New file.
* Makefile.in (SFILES): Add namespace.c.
(COMMON_OBS): Add namespace.o

gdb/testsuite/ChangeLog:

* gdb.ada/fun_renaming.exp: New testcase.
* gdb.ada/fun_renaming/fun_renaming.adb: New file.
* gdb.ada/fun_renaming/pack.adb: New file.
* gdb.ada/fun_renaming/pack.ads: New file.

Tested on x86_64-linux.  Support for this in GCC is in the pipeline: see
<https://gcc.gnu.org/ml/gcc-patches/2015-07/msg02166.html>.

9 years agogas 0f handling
Alan Modra [Thu, 13 Aug 2015 06:27:15 +0000 (15:57 +0930)]
gas 0f handling

_start:
 .byte 0f-_start
0:

Fixes
..:2: Error: floating point number invalid
..:2: Error: junk at end of line, first unrecognized character is `_'

* expr.c (operand): Rewrite handling of operands starting with "0f".
If atof_generic only parses "-" or "+", treat as expression.

9 years agogas 0b vs 0b0 vs 00b
Alan Modra [Thu, 13 Aug 2015 06:25:31 +0000 (15:55 +0930)]
gas 0b vs 0b0 vs 00b

* expr.c (integer_constant): Return O_absent expression if eol.
(operand): For targets with both LOCAL_LABELS_FB and
NUMBERS_WITH_SUFFIX set, treat "0b" not followed by binary
digits as a local label reference.  Correct handling of 0b prefix.
If a suffix is not allowed, error on 0B.

9 years agoCorrect local label doc
Alan Modra [Thu, 13 Aug 2015 06:21:51 +0000 (15:51 +0930)]
Correct local label doc

* doc/as.texinfo (Local Labels): Allowed range of N in local
labels is non-negative integers, not positive integers.

9 years agogdb.base/dso2dso.exp sometimes broken
Keith Seitz [Thu, 13 Aug 2015 01:31:11 +0000 (18:31 -0700)]
gdb.base/dso2dso.exp sometimes broken

Keith reported that gdb.base/dso2dso.exp is broken, with the following
error:

| $ make check RUNTESTFLAGS=dso2dso.exp
| [snip]
| Running ../../../src/gdb/testsuite/gdb.base/dso2dso.exp ...
| ERROR: tcl error sourcing ../../../src/gdb/testsuite/gdb.base/dso2dso.exp.
| ERROR: couldn't open
| "../../../src/gdb/testsuite/gdb.base/../../../src/gdb/testsuite/gdb.base/dso2dso-dso1.c":
| no such file or directory
|     while executing
| "error "$message""
|     (procedure "gdb_get_line_number" line 14)
|     invoked from within
| "gdb_get_line_number "STOP HERE" $srcfile_libdso1"
|     (file "../../../src/gdb/testsuite/gdb.base/dso2dso.exp" line 60)
|     invoked from within
| "source ../../../src/gdb/testsuite/gdb.base/dso2dso.exp"
|     ("uplevel" body line 1)
|     invoked from within
| "uplevel #0 source ../../../src/gdb/testsuite/gdb.base/dso2dso.exp"
|     invoked from within
| "catch "uplevel #0 source $test_file_name""

This happens because gdb_get_line_number will prepend $srcdir/$subdir
if the given filename does not start with "/", and this happens when
GDB was configured using a relative path to the configure script.
When using an absolute path like I do, we avoid the pre-pending that
Keith is seeing.

gdb/testsuite/ChangeLog:

        Keith Seitz  <keiths@redhat.com>:
        * gdb.base/dso2dso.exp: Pass basename of source file in call
        to gdb_get_line_number.

Tested on x86_64-linux with both scenarios.

9 years agoAutomatic date update in version.in
GDB Administrator [Thu, 13 Aug 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

9 years agoSet EI_OSABI to ELFOSABI_GNU for local IFUNC symbols
H.J. Lu [Wed, 12 Aug 2015 20:31:29 +0000 (13:31 -0700)]
Set EI_OSABI to ELFOSABI_GNU for local IFUNC symbols

Since the backend elf_add_symbol_hook isn't called on local symbols,
the EI_OSABI field isn't to ELFOSABI_GNU where are local IFUNC symbols.
This patch changes the x86 backends to set has_gnu_symbols if there are
relocations against IFUNC symbols.  Other backends with IFUNC support
may need a similar change.

This patch also changes the type of has_gnu_symbols from bfd_boolean to
enum elf_gnu_symbols.

bfd/

PR ld/18815
* elf-bfd.h (elf_gnu_symbols): New enum.
(elf_obj_tdata): Use elf_gnu_symbols on has_gnu_symbols.
* elf-s390-common.c (elf_s390_add_symbol_hook): Set
has_gnu_symbols to elf_gnu_symbol_any.
* elf32-arm.c (elf32_arm_add_symbol_hook): Likewise.
* elf32-m68k.c (elf_m68k_add_symbol_hook): Likewise.
* elf32-ppc.c (ppc_elf_add_symbol_hook): Likewise.
* elf32-sparc.c (elf32_sparc_add_symbol_hook): Likewise.
* elf64-ppc.c (ppc64_elf_add_symbol_hook): Likewise.
* elf64-sparc.c (elf64_sparc_add_symbol_hook): Likewise.
* lfxx-aarch64.c (_bfd_aarch64_elf_add_symbol_hook): Likewise.
* elf32-i386.c (elf_i386_check_relocs): Update has_gnu_symbols
if there are relocations against IFUNC symbols.
(elf_i386_add_symbol_hook): Don't check STT_GNU_IFUNC here.
* elf64-x86-64. (elf_x86_64_check_relocs): Update has_gnu_symbols
if there are relocations against IFUNC symbols.
(elf_x86_64_add_symbol_hook): Don't check STT_GNU_IFUNC here.

ld/testsuite/

PR ld/18815
* ld-i386/i386.exp: Run pr18815.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr18815.d: New file.
* ld-i386/pr18815.s: Likewise.
* ld-x86-64/pr18815.d: Likewise.
* ld-x86-64/pr18815.s: Likewise.

9 years ago[amd64] Invalid return address after displaced stepping
Joel Brobecker [Wed, 12 Aug 2015 16:33:19 +0000 (09:33 -0700)]
[amd64] Invalid return address after displaced stepping

Making all-stop run on top of non-stop caused a small regression
in behavior. This was observed on x86_64-linux. The attached testcase
is in C whereas the investigation was done with an Ada program,
but it's the same scenario, and using a C testcase allows wider testing.
Basically: I am debugging a single-threaded program, and currently
stopped inside a function provided by a shared-library, at a line
calling a subprogram provided by a second shared library, and trying
to "next" over that function call.

Before we changed the default all-stop behavior, we had:

    7             Impl_Initialize;  -- Stop here and try "next" over this line
    (gdb) n
    8             return 5;  <<-- OK

But now, "next" just stops much earlier:

    (gdb) n
    0x00007ffff7bd8560 in impl.initialize@plt () from /[...]/lib/libpck.so

What happens is that next stops at a call instruction, which calls
the function's PLT, and GDB fails to notice that the inferior stepped
into a subroutine, and so decides that we're done. We can see another
symptom of the same issue by looking at the backtrace at the point
GDB stopped:

    (gdb) bt
    #0  0x00007ffff7bd8560 in impl.initialize@plt ()
       from /[...]/lib/libpck.so
    #1  0x00000000f7bd86f9 in ?? ()
    #2  0x00007fffffffdf50 in ?? ()
    #3  0x0000000000401893 in a () at /[...]/a.adb:7
    Backtrace stopped: frame did not save the PC

With a functioning GDB, the backtrace looks like the following instead:

    #0  0x00007ffff7bd8560 in impl.initialize@plt ()
       from /[...]/lib/libpck.so
    #1  0x00007ffff7bd86f9 in sub () at /[...]/pck.adb:7
    #2  0x0000000000401893 in a () at /[...]/a.adb:7

Note how, for frame #1, the address looks quite similar, except
for the high-order bits not being set:

    #1  0x00007ffff7bd86f9 in sub () at /[...]/pck.adb:7   <<<--  OK
    #1  0x00000000f7bd86f9 in ?? ()                        <<<--  WRONG
              ^^^^
              ||||
              Wrong

Investigating this further led me to displaced stepping.
As we are "next"-ing from a location where a breakpoint is inserted,
we need to step out of it, and since we're on non-stop mode, we need
to do it using displaced stepping. And looking at
amd64-tdep.c:amd64_displaced_step_fixup, I found the code that handles
the return address:

    regcache_cooked_read_unsigned (regs, AMD64_RSP_REGNUM, &rsp);
    retaddr = read_memory_unsigned_integer (rsp, retaddr_len, byte_order);
    retaddr = (retaddr - insn_offset) & 0xffffffffUL;

The mask used to compute retaddr looks wrong to me, keeping only
4 bytes instead of 8, and explains why the high order bits of
the backtrace are unset. What happens is that, after the displaced
stepping has completed, GDB restores that return address at the location
where the program expects it.  But because the top half bits of
the address have been masked out, the return address is now invalid.
The incorrect behavior of the "next" command and the backtrace at
that location are the first symptoms of that.  Another symptom is
that this actually alters the behavior of the program, where a "cont"
from there soon leads to a SEGV when the inferior tries to jump back
to that incorrect return address:

    (gdb) c
    Continuing.

    Program received signal SIGSEGV, Segmentation fault.
    0x00000000f7bd86f9 in ?? ()
    ^^^^^^^^^^^^^^^^^^

This patch fixes the issue by using a mask that seems more appropriate
for this architecture.

gdb/ChangeLog:

        * amd64-tdep.c (amd64_displaced_step_fixup): Fix the mask used to
        compute RETADDR.

gdb/testsuite/ChangeLog:

        * gdb.base/dso2dso-dso2.c, gdb.base/dso2dso-dso2.h,
        gdb.base/dso2dso-dso1.c, gdb.base/dso2dso-dso1.h, gdb.base/dso2dso.c,
        gdb.base/dso2dso.exp: New files.

Tested on x86_64-linux, no regression.

9 years agoInitialize `location' in gdbpy_decode_line
Keith Seitz [Wed, 12 Aug 2015 18:31:24 +0000 (11:31 -0700)]
Initialize `location' in gdbpy_decode_line

BuildBot flagged an uninitialized variable coming from one of the patches
in my recently committed locations/explicit patchset.

The following patch fixes this.

gdb/ChangeLog

* python/python.c (gdbpy_decode_line): Initialize `location' to NULL
and only call decode_line_1 when it is non-NULL.

diff --git a/gdb/python/python.c b/gdb/python/python.c
index c28f98b..14da62c 100644
--- a/gdb/python/python.c
+++ b/gdb/python/python.c
@@ -730,7 +730,7 @@ gdbpy_decode_line (PyObject *self, PyObject *args)
   PyObject *result = NULL;
   PyObject *return_result = NULL;
   PyObject *unparsed = NULL;
-  struct event_location *location;
+  struct event_location *location = NULL;

   if (! PyArg_ParseTuple (args, "|s", &arg))
     return NULL;
@@ -747,7 +747,7 @@ gdbpy_decode_line (PyObject *self, PyObject *args)

   TRY
     {
-      if (arg)
+      if (location != NULL)
  sals = decode_line_1 (location, 0, 0, 0);
       else
  {

9 years agoxtensa: add --auto-litpools option
Max Filippov [Wed, 29 Jul 2015 14:42:54 +0000 (17:42 +0300)]
xtensa: add --auto-litpools option

Auto-litpools is the automated version of text-section-literals: literal
pool candidate frags are planted every N frags and during relaxation
they are turned into actual literal pools where literals are moved to
become reachable for their first reference by L32R instruction.

2015-08-12  David Weatherford  <weath@cadence.com>
gas/
* config/tc-xtensa.c (struct litpool_frag, struct litpool_seg):
New structures.
(xtensa_maybe_create_literal_pool_frag): New function.
(litpool_seg_list, auto_litpools, auto_litpool_limit)
(litpool_buf, litpool_slotbuf): New static variables.
(option_auto_litpools, option_no_auto_litpools)
(option_auto_litpool_limit): New enum identifiers.
(md_longopts): Add entries for auto-litpools, no-auto-litpools
and auto-litpool-limit.
(md_parse_option): Handle option_auto_litpools,
option_no_auto_litpools and option_auto_litpool_limit.
(md_show_usage): Add help for --[no-]auto-litpools and
--auto-litpool-limit.
(xtensa_mark_literal_pool_location): Record a place for literal
pool with a call to xtensa_maybe_create_literal_pool_frag.
(get_literal_pool_location): Find highest priority literal pool
or convert candidate to literal pool when auto-litpools are used.
(xg_assemble_vliw_tokens): Create literal pool after jump
instruction.
(xtensa_check_frag_count): Create candidate literal pool every
auto_litpool_limit frags.
(xtensa_relax_frag): Add jump around literals to non-empty
literal pool.
(xtensa_move_literals): Estimate literal pool addresses and move
unreachable literals closer to their users, converting candidate
to literal pool if needed.
(xtensa_switch_to_non_abs_literal_fragment): Only emit error
about missing .literal_position in case auto-litpools are not
used.
* config/tc-xtensa.h (xtensa_relax_statesE): New relaxation
state: RELAX_LITERAL_POOL_CANDIDATE_BEGIN.
* doc/as.texinfo (Xtensa options):  Document --auto-litpools and
--no-auto-litpools options.
* doc/c-xtensa.texi (Xtensa options): Likewise.

2015-08-12  Max Filippov  <jcmvbkbc@gmail.com>
gas/testsuite/
* gas/xtensa/all.exp: Add auto-litpools to the list of xtensa
tests.
* gas/xtensa/auto-litpools.s: New file: auto-litpools test.
* gas/xtensa/auto-litpools.s: New file: auto-litpools test
result pattern.

9 years agoGuarantee save-and-restore of GDBFLAGS on gdb.base/checkpoint-ns.exp
Sergio Durigan Junior [Wed, 12 Aug 2015 16:32:16 +0000 (12:32 -0400)]
Guarantee save-and-restore of GDBFLAGS on gdb.base/checkpoint-ns.exp

Keith found out that several tests were failing when testing the
native-gdbserver board on Fedora (x86_64).  Strangely, these failures
had not been reported by our BuildBot.  Later, he found that the reason
for this was because the failures only happened when running the
testsuite without FORCE_PARALLEL (i.e., on serial mode; maybe it would
be worth having a builder testing things on serial...).  Then, he
decided to start bisecting the changes to see which one introduced the
failure (it was not trivial to know this only by looking at gdb.log).

After a lot of time, he found that Pedro's commit
e1316e60d4d1fe406efc6e7536b2bdb43733e9d2 was the culprit.  There was
nothing wrong in the code, but the new gdb.base/checkpoint-ns.exp
testcase did something that left the GDBFLAGS variable in an
inconsistent state.  This test works by modifying this variable to set
non-stop on, sourcing gdb.base/checkpoint.exp (which does the hard
work), and then restoring the old value on GDBFLAGS.  However, this was
not working because gdb.base/checkpoint.exp bails out if it is being
tested on gdbserver, and when it calls "continue" the control goes back
to the function calling the tests, and not to
gdb.base/checkpoint-ns.exp.

The fix is simple: just wrap the "source" call, and make
gdb.base/checkpoint-ns.exp aware of the "continue"/"return" calls made
by gdb.base/checkpoint.exp.

gdb/testsuite/ChangeLog:
2015-08-12  Sergio Durigan Junior  <sergiodj@redhat.com>
    Pedro Alves  <palves@redhat.com>
    Keith Seitz  <keiths@redhat.com>

* gdb.base/checkpoint-ns.exp: Use save_vars to save and restore
GDBFLAGS.

9 years ago[MIPS] Map 'move' to 'or'.
Simon Dardis [Wed, 12 Aug 2015 16:06:35 +0000 (17:06 +0100)]
[MIPS] Map 'move' to 'or'.

The MIPS assembly idiom 'move' now maps to the 'or' machine instruction. This
change affects microMIPS, MIPS32, MIPS64.

2015-08-12  Simon Dardis  <simon.dardis@imgtec.com>

opcodes/

* micromips-opc.c (micromips_opcodes): Re-order table so that move
based on 'or' is first.
* mips-opc.c (mips_builtin_opcodes): Ditto.

bfd/

* elfxx-mips.c (STUB_MOVE): Change to use 'or' only.
(mips_o32_exec_plt0_entry, mips_n32_exec_plt0_entry,
mips_n64_exec_plt0_entry, micromips_insn32_o32_exec_plt0_entry):
Update to use 'or' instead of 'addu/daddu'.
(_bfd_mips_elf_finish_dynamic_symbol): Update usage of STUB_MOVE.
(move_insns_32): Reorder table.

gas/

* config/tc-mips.c (move_register): Change to use 'or' only.
(s_cpload, s_cpsetup, s_cprestore, s_cpreturn): Update to
use or for move.

gas/testsuite/

* gas/mips/elf-rel23.d: Update test.
* gas/mips/elf-rel23.d: Ditto.
* gas/mips/elf-rel23a.d: Ditto.
* gas/mips/elf-rel23b.d: Ditto.
* gas/mips/elf_e_flags1.d: Ditto.
* gas/mips/elf_e_flags2.d: Ditto.
* gas/mips/elf_e_flags3.d: Ditto.
* gas/mips/elf_e_flags4.d: Ditto.
* gas/mips/loc-swap-dis.d: Ditto.
* gas/mips/micromips-insn32.d: Ditto.
* gas/mips/micromips-noinsn32.d: Ditto.
* gas/mips/micromips-trap.d: Ditto.
* gas/mips/micromips.d: Ditto.
* gas/mips/mips-abi32-pic.d: Ditto.
* gas/mips/mips-abi32.d: Ditto.
* gas/mips/mips-gp32-fp32-pic.d: Ditto.
* gas/mips/mips-gp32-fp32.d: Ditto.
* gas/mips/mips-gp32-fp64-pic.d: Ditto.
* gas/mips/mips-gp32-fp64.d: Ditto.
* gas/mips/mips-gp64-fp32-pic.d: Ditto.
* gas/mips/mips-gp64-fp32.d: Ditto.
* gas/mips/mips-gp64-fp64-pic.d: Ditto.
* gas/mips/mips-gp64-fp64.d: Ditto.
* gas/mips/mipsr6@loc-swap-dis.d: Ditto.
* gas/mips/tls-o32.d: Ditto.
* gas/mips/uld2-eb.d: Ditto.
* gas/mips/uld2-el.d: Ditto.
* gas/mips/ulw2-eb-ilocks.d: Ditto.
* gas/mips/ulw2-eb.d: Ditto.
* gas/mips/ulw2-el-ilocks.d: Ditto.
* gas/mips/ulw2-el.d: Ditto.
* gas/mips/move.d: New test.
* gas/mips/move.s: Ditto.
* gas/mips/micromips32-move.d: Ditto.
* gas/mips/micromips32-move.s: Ditto.
* gas/mips/mips.exp: Run the new tests.

gold/

* mips.cc (plt0_entry_o32, plt0_entry_n32, plt0_entry_n64,
lazy_stub_normal_1, lazy_stub_normal_1_n64,
lazy_stub_normal_2, lazy_stub_normal_2_n64, lazy_stub_big,
lazy_stub_big_n64, lazy_stub_micromips32_normal_1_n64,
lazy_stub_micromips32_normal_2_n64, lazy_stub_micromips32_big,
lazy_stub_micromips32_big_n64): Update to use 'or' for move instead
of 'addu/daddu'.

ld/testsuite/

* ld-mips-elf/compressed-plt-1-n32-mips16.od: Update test.
* ld-mips-elf/compressed-plt-1-n32-umips.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-mips16-got.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-mips16-only.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-mips16-word.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-mips16.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-se.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-umips-got.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-umips-word.od: Ditto.
* ld-mips-elf/compressed-plt-1-o32-umips.od: Ditto.
* ld-mips-elf/jalx-2.dd: Ditto.
* ld-mips-elf/mips16-pic-3.dd: Ditto.
* ld-mips-elf/pic-and-nonpic-3a.dd: Ditto.
* ld-mips-elf/pic-and-nonpic-3b.dd: Ditto.
* ld-mips-elf/pic-and-nonpic-5b.dd: Ditto.
* ld-mips-elf/pic-and-nonpic-6-n32.dd: Ditto.
* ld-mips-elf/pic-and-nonpic-6-o32.dd: Ditto.
* ld-mips-elf/stub-dynsym-1-10000.d: Ditto.
* ld-mips-elf/stub-dynsym-1-2fe80.d: Ditto.
* ld-mips-elf/stub-dynsym-1-7fff.d: Ditto.
* ld-mips-elf/stub-dynsym-1-8000.d: Ditto.
* ld-mips-elf/stub-dynsym-1-fff0.d: Ditto.
* ld-mips-elf/tlsbin-o32.d: Ditto.
* ld-mips-elf/tlsdyn-o32-1.d: Ditto.
* ld-mips-elf/tlsdyn-o32-2.d: Ditto.
* ld-mips-elf/tlsdyn-o32-3.d: Ditto.
* ld-mips-elf/tlsdyn-o32.d: Ditto.
* ld-mips-elf/tlslib-o32.d: Ditto.

9 years agoUse save_vars to replace existing manipulation of globals in tests
Patrick Palka [Fri, 19 Jun 2015 01:47:55 +0000 (21:47 -0400)]
Use save_vars to replace existing manipulation of globals in tests

gdb/testsuite/ChangeLog:

* gdb.base/gdbhistsize-history.exp
(test_histsize_history_setting): Use save_vars.
* gdb.base/gdbinit-history.exp (test_gdbinit_history_setting):
Use save_vars.
(test_no_truncation_of_unlimited_history_file): Use save_vars.
* gdb.base/readline.exp: Use save_vars.

9 years agoRemove ChangeLog separator
Alan Modra [Wed, 12 Aug 2015 09:36:01 +0000 (19:06 +0930)]
Remove ChangeLog separator

9 years agoIntroduce save_vars, a testsuite proc for safely manipulating globals
Patrick Palka [Fri, 19 Jun 2015 01:33:26 +0000 (21:33 -0400)]
Introduce save_vars, a testsuite proc for safely manipulating globals

gdb/testsuite/ChangeLog:

* lib/gdb.exp (save_vars): New proc.

9 years ago[AArch64] Fix test failures on elf configuration
Jiong Wang [Wed, 12 Aug 2015 12:17:50 +0000 (13:17 +0100)]
[AArch64] Fix test failures on elf configuration

This patch fixed those failures on elf configuration by:

   * Improve the ILP32 target selector "aarch64_choose_ilp32_emul",
     makes it more robust. Target triples copied from configure.tgt

   * Updated emit-relocs-86/-overflow.d to use aarch64_choose_ilp32_emul
     which is following what have done with emit-relocs-28.

   * Those instruction encoding mismatch is because those encoding
     contains pc-relative address. As for elf, we may have different
     start address. relaxed encodind check, especially for
     aarch64-farcall-b/bl-plt, as the main purpose of those check are
     ELF text/data layout, we just want to make sure veneer to plt stub
     is generated.

2015-08-12  Jiong Wang  <jiong.wang@arm.com>

ld/testsuite/
  * ld-aarch64/aarch64-elf.exp (aarch64_choose_ilp32_emul): Support all
  four triple shapes: aarch64-*-linux*, aarch64-*-elf,
  aarch64_be-*-linux*, aarch64_be-*-elf.
  * ld-aarch64/emit-relocs-86.d: Use aarch64_choose_ilp32_emul.
  * ld-aarch64/emit-relocs-86-overflow.d: Likewise.
  * ld-aarch64/ld-aarch64/farcall-b-plt.d: Relax instrucion encoding
  check when they reflect address.
  * ld-aarch64/ld-aarch64/farcall-bl-plt.d: Likewise.

9 years agoSync ansidecl.h with GCC
H.J. Lu [Wed, 12 Aug 2015 12:02:21 +0000 (05:02 -0700)]
Sync ansidecl.h with GCC

Sync with GCC
2015-08-11  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* ansidecl.h (GCC_FINAL): New macro.

9 years agoRemove trailing spaces in ld
H.J. Lu [Wed, 12 Aug 2015 11:46:43 +0000 (04:46 -0700)]
Remove trailing spaces in ld

9 years agoRemove trailing spaces in opcodes
H.J. Lu [Wed, 12 Aug 2015 11:45:07 +0000 (04:45 -0700)]
Remove trailing spaces in opcodes

9 years agoRemove trailing spaces in gprof
H.J. Lu [Wed, 12 Aug 2015 11:43:32 +0000 (04:43 -0700)]
Remove trailing spaces in gprof

9 years agoRemove trailing spaces in binutils
H.J. Lu [Wed, 12 Aug 2015 11:42:37 +0000 (04:42 -0700)]
Remove trailing spaces in binutils

9 years agoRemove trailing spaces in gas
H.J. Lu [Wed, 12 Aug 2015 11:40:42 +0000 (04:40 -0700)]
Remove trailing spaces in gas

9 years agoRemove trailing spaces in bfd
H.J. Lu [Wed, 12 Aug 2015 11:32:43 +0000 (04:32 -0700)]
Remove trailing spaces in bfd

9 years agoUpdate Swedish translation in gprof.
Nick Clifton [Wed, 12 Aug 2015 10:50:26 +0000 (11:50 +0100)]
Update Swedish translation in gprof.

* po/sv.po: Updated Swedish translation.

9 years ago[regression] Do not read from catchpoint/watchpoint locations' addresses when checkin...
Luis Machado [Wed, 12 Aug 2015 08:36:09 +0000 (05:36 -0300)]
[regression] Do not read from catchpoint/watchpoint locations' addresses when checking for a permanent breakpoint

While running bare-metal tests with GDB i noticed some failures in
gdb.base/break.exp, related to the use of the catch commands.

It turns out GDB tries to access memory address 0x0 whenever one tries
to insert a catchpoint, which should obviously not happen.

This was introduced with the changes for permanent breakpoints. In special,
bp_loc_is_permanent tries to check if there is a breakpoint inserted at
the same address as the current breakpoint's location's address. In the
case of catchpoints, this is 0x0.

(top-gdb) catch fork
Sending packet: $m0,1#fa...Packet received: E01
Catchpoint 4 (fork)

(top-gdb) catch vfork
Sending packet: $m0,1#fa...Packet received: E01
Catchpoint 5 (vfork)

It is not obvious to detect because this fails silently for Linux. For our
bare-metal testing, though, this fails with a clear error message from the
target about not being able to read such address.

The attached patch addresses this by bailing out of bp_loc_is_permanent (...)
if the location address is not meaningful. I also took the opportunity to
update the comment for breakpoint_address_is_meaningful, which mentioned
breakpoint addresses as opposed to their locations' addresses.

gdb/ChangeLog:

2015-08-11  Luis Machado  <lgustavo@codesourcery.com>

* breakpoint.c (bp_loc_is_permanent): Return 0 when breakpoint
location address is not meaningful.
(breakpoint_address_is_meaningful): Update comment.

9 years agoProperly skip IFUNC relocations in debug sections
H.J. Lu [Wed, 12 Aug 2015 02:04:38 +0000 (19:04 -0700)]
Properly skip IFUNC relocations in debug sections

Use "continue" instead of "break" to skip IFUNC relocations in debug
sections.

* elf32-i386.c (elf_i386_relocate_section): Properly skip IFUNC
relocations in debug sections.
* elf64-x86-64.c (elf_x86_64_relocate_section): Likewise.

9 years ago[AArch64] Rename relocs-257-symbolic-func ld test to relocs-1027-symbolic-func.
Thomas Preud'homme [Wed, 12 Aug 2015 01:31:04 +0000 (09:31 +0800)]
[AArch64] Rename relocs-257-symbolic-func ld test to relocs-1027-symbolic-func.

2015-08-12  Thomas Preud'homme  <thomas.preudhomme@arm.com>

ld/testsuite/
    * ld-aarch64/aarch64-elf.exp: Rename relocs-257-symbolic-func to
    relocs-1027-symbolic-func.
    * ld-aarch64/relocs-257-symbolic-func.d: Rename to ...
    * ld-aarch64/relocs-1027-symbolic-func.d: ... this.
    * ld-aarch64/relocs-257-symbolic-func.s: Rename to ...
    * ld-aarch64/relocs-1027-symbolic-func.s: ... this.

9 years agoExplicit locations: documentation updates
Keith Seitz [Wed, 12 Aug 2015 00:09:36 +0000 (17:09 -0700)]
Explicit locations: documentation updates

This patch adds documentation for explicit locations to both the
User Manual and gdb's online help system.

gdb/ChangeLog:

* NEWS: Mention explicit locations.
* breakpoint.c [LOCATION_HELP_STRING]: New macro.
[BREAK_ARGS_HELP]: Use LOCATION_HELP_STRING.
(_initialize_breakpoint): Update documentation for
"clear", "break", "trace", "strace", "ftrace", and "dprintf".

gdb/doc/ChangeLog:

* gdb.texinfo (Thread-Specific Breakpoints, Printing Source Lines):
Use "location(s)"instead of "linespec(s)".
(Specifying a Location): Rewrite.
Add subsections describing linespec, address, and explicit locations.
Add node/menu for each subsection.
(Source and Machine Code, C Preprocessor Macros)
(Create and Delete Trace points)
(Extensions for Ada Tasks): Use "location(s)" instead of "linespec(s)".
(Continuing at a Different Address): Remove "linespec" examples.
Add reference to "Specify a Location"
(The -break-insert Command): Rewrite.  Add anchor.
Add reference to appropriate manual section discussing locations.
(The -dprintf-insert Command): Refer to -break-insert for
specification of 'location'.

gdb/testsuite/ChangeLog:

* gdb.base/help.exp: Update help_breakpoint_text.