deliverable/binutils-gdb.git
4 years agoFix the thread-pool.c compilation
Tom Tromey [Thu, 12 Dec 2019 13:09:40 +0000 (06:09 -0700)] 
Fix the thread-pool.c compilation

A recent commit removed DIAGNOSTIC_IGNORE_UNUSED_FUNCTION, which was
used in thread-pool.c.  This patch changes this code to use
ATTRIBUTE_UNUSED instead.

Tested by rebuilding.

gdb/ChangeLog
2019-12-12  Tom Tromey  <tromey@adacore.com>

* gdbsupport/thread-pool.c (set_thread_name): Use
ATTRIBUTE_UNUSED.

Change-Id: I56d46eaac73690565d0e52db1791411567a918dd

4 years agoFix unused function error
Luis Machado [Wed, 11 Dec 2019 14:55:49 +0000 (11:55 -0300)] 
Fix unused function error

Attempting to build GDB in Ubuntu 16.04.6 LTS on x86_64, I ran into warnings
that caused the build to fail:

binutils-gdb/gdb/gdbsupport/safe-strerror.c:44:1: error: ‘char* select_strerror_r(char*, char*)’ defined but not used [-Werror=unused-function]  select_strerror_r (char *res, char *)

The diagnostic macro DIAGNOSTIC_IGNORE_UNUSED_FUNCTION seems to expand
correctly to its respective pragma, but this doesn't seem to have an effect on
the warning. I tried to use the pragma explicitly and got the same result.

ATTRIBUTE_UNUSED works fine in this case if you put it in both functions,
which should fix warnings for both gdb and gdbserver builds.

The compiler version is gcc (Ubuntu 5.4.0-6ubuntu1~16.04.11) 5.4.0 20160609.

This is likely the result of PR64079 in GCC, which was fixed by commit
9e96f1e1b9731c4e1ef4fbbbf0997319973f0537.

To prevent other developers from attempting to use this macro, only to get
confused by it not working as expected, it seems better to not define this
particular macro.

gdb/ChangeLog:

2019-12-12  Luis Machado  <luis.machado@linaro.org>

* gdbsupport/safe-strerror.c: Don't include diagnostics.h.
(select_strerror_r): Use ATTRIBUTE_UNUSED instead of the diagnostics
macros.

include/ChangeLog:

2019-12-12  Luis Machado  <luis.machado@linaro.org>

* diagnostics.h (DIAGNOSTIC_IGNORE_UNUSED_FUNCTION). Remove
definitions.

Change-Id: Iad6123d61d76d111e3ef8d24aa8c60112304c749

4 years agogas signed overflow fixes
Alan Modra [Wed, 11 Dec 2019 22:59:45 +0000 (09:29 +1030)] 
gas signed overflow fixes

* config/tc-aarch64.c (get_aarch64_insn): Avoid signed overflow.
* config/tc-metag.c (parse_dalu): Likewise.
* config/tc-tic4x.c (md_pcrel_from): Likewise.
* config/tc-tic6x.c (tic6x_output_unwinding): Likewise.
* config/tc-csky.c (parse_fexp): Use an unsigned char temp buffer.
Don't use register keyword.  Avoid signed overflow and remove now
unneccesary char masks.  Formatting.
* config/tc-ia64.c (operand_match): Don't use shifts to sign extend.
* config/tc-mep.c (mep_apply_fix): Likewise.
* config/tc-pru.c (md_apply_fix): Likewise.
* config/tc-riscv.c (load_const): Likewise.
* config/tc-nios2.c (md_apply_fix): Likewise.  Don't potentially
truncate fixup before right shift.  Tidy BFD_RELOC_NIOS2_HIADJ16
calculation.

4 years agoobj-evax.c tidy
Alan Modra [Thu, 12 Dec 2019 06:17:21 +0000 (16:47 +1030)] 
obj-evax.c tidy

This started out as fixing decode_16, which used a char to index a 256
byte decodings array.  When char is signed that could result in an out
of bounds array access.  The rest of the patch is for consistency and
just general cleanup.

* config/obj-evax.c (crc32, encode_32, encode_16, decode_16):
Remove unnecessary prototypes.
(number_of_codings): Delete, use ARRAY_SIZE instead throughout.
(codings, decodings): Make arrays of unsigned char.
(crc32): Use unsigned variables.  Delete unnecessary mask.
(encode_32, encode_16): Return unsigned char*, and make static
buffer an unsigned char array.
(decode_16): Make arg an unsigned char*.  Remove useless casts.
(shorten_identifier): Use unsigned char crc_chars.
(is_truncated_identifier): Make ptr an unsigned char*.

4 years agocsky: tidy csky_chars_to_number
Alan Modra [Thu, 12 Dec 2019 06:14:24 +0000 (16:44 +1030)] 
csky: tidy csky_chars_to_number

* csky-dis.c (csky_chars_to_number): Remove abort and unnecessary
mask.

4 years agoRemove duplicate cast
Alan Modra [Thu, 12 Dec 2019 03:50:07 +0000 (14:20 +1030)] 
Remove duplicate cast

* libbfd.c (bfd_get): Don't cast result of bfd_get_8.
* bfd-in2.h: Regenerate.

4 years agoFix TUI test suite regexps
Tom Tromey [Thu, 12 Dec 2019 02:17:47 +0000 (19:17 -0700)] 
Fix TUI test suite regexps

Testing on another TUI series showed that some of the regexps in the
TUI test suite have been incorrect for a while.  In particular, "|"
was meant literally in these tests, but was interpreted as pattern
alternation due to lack of quoting.

This patch fixes the bad tests.  I am checking this in.

gdb/testsuite/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* gdb.tui/resize.exp: Fix regexp.
* gdb.tui/regs.exp: Fix regexps.
* gdb.tui/main.exp: Fix regexp.

Change-Id: Ib6661361171ac120bb92f4a8aec7efa4bcaa36b9

4 years agoAutomatic date update in version.in
GDB Administrator [Thu, 12 Dec 2019 00:00:15 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years agoChange "winheight" resizing to use new layout code
Tom Tromey [Sat, 26 Oct 2019 22:37:32 +0000 (16:37 -0600)] 
Change "winheight" resizing to use new layout code

The "winheight" command resizes a specified window, resizing the other
windows in the layout to adapt.  In the current code, this is done by
examining each possible layout separately.  The new layout code has a
more general approach to handling this, and this patch simply removes
the old code in favor of a call into the new layout engine.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-win.c (tui_set_win_height_command): Call
tui_adjust_window_height.
(tui_adjust_win_heights, new_height_ok): Remove.
* tui/tui-layout.h (tui_adjust_window_height): Declare.
* tui/tui-layout.c (tui_adjust_window_height): New function.

Change-Id: I6bb681375a46adc8d29fd06f581deed4e078e78a

4 years agoRe-apply the current layout when resizing
Tom Tromey [Thu, 10 Oct 2019 03:18:31 +0000 (21:18 -0600)] 
Re-apply the current layout when resizing

The TUI has separate code for each possible layout to handle the case
where the terminal window is resized.  With the new layout code, this
can all be replaced with a call to tui_apply_current_layout, which
simply re-applies the current layout.

This results in some small differences in behavior when resizing, so
some tests are updated.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-win.c (tui_resize_all): Remove code, call
tui_apply_current_layout.

gdb/testsuite/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* gdb.tui/resize.exp: Update.
* gdb.tui/empty.exp (layouts): Update.

Change-Id: I3dc6c02a753d495d9ab5e8213d550a147198ce6f

4 years agoFirst use of tui_layout
Tom Tromey [Wed, 9 Oct 2019 22:35:41 +0000 (16:35 -0600)] 
First use of tui_layout

This patch introduces the first use of tui_layout, by changing
show_layout to clone and use the appropriate tui_layout.

This resulted in one minor layout change, and also in the unintended
-- but good -- side effect that the title of each boxed window is now
visible.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-layout.h (tui_apply_current_layout): Declare.
* tui/tui-layout.c (standard_layouts, applied_layout): New
globals.
(tui_apply_current_layout): New function.
(show_layout): Set applied_layout.  Call
tui_apply_current_layout.
(show_source_command, show_disasm_command)
(show_source_disasm_command, show_data)
(show_source_or_disasm_and_command): Remove.
(initialize_layouts): New function.
(_initialize_tui_layout): Call initialize_layouts.

gdb/testsuite/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* gdb.tui/regs.exp: Update.
* gdb.tui/empty.exp (layouts): Update.
* gdb.tui/basic.exp: Update.
* lib/tuiterm.exp (_check_box): Don't check bottom border.

Change-Id: If1ee06ee58f4803e8c213f4ab0f5bb59f4650ec2

4 years agoIntroduce new layout code
Tom Tromey [Tue, 8 Oct 2019 00:03:02 +0000 (18:03 -0600)] 
Introduce new layout code

This introduces a new approach to window layout for the TUI.  The idea
behind this code is that a layout should be specified in a declarative
way, and then be applied by generic code that does not need to know
the specifics of every possible layout.

This patch itself does not change any behavior, because the new layout
engine isn't yet connected to anything.  That is, this merely
introduces the implementation.

This generic approach makes the code more maintainable.  It also
enables some future changes:

* New window types are simpler to add;
* User-specified layouts are possible; and
* Horizontal layouts are more attainable

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-layout.h (class tui_layout_base)
(class tui_layout_window, class tui_layout_split): New.
* tui/tui-layout.c (tui_get_window_by_name)
(tui_layout_window::clone, tui_layout_window::apply)
(tui_layout_window::get_sizes, tui_layout_window::add_split)
(tui_layout_split::add_window, tui_layout_split::clone)
(tui_layout_split::get_sizes)
(tui_layout_split::set_weights_from_heights)
(tui_layout_split::adjust_size, tui_layout_split::apply): New
functions.
(tui_layout_split::add_split, tui_layout_split::add_split)
(tui_layout_split::set_weights_from_heights)
(tui_layout_split::set_weights_from_heights): New functions.

Change-Id: I3a4cae666327b617d862aaa356f8179f945c6a4e

4 years agoRemove struct tui_point
Tom Tromey [Sat, 26 Oct 2019 22:00:01 +0000 (16:00 -0600)] 
Remove struct tui_point

struct tui_point does not help very much.  It is only used for
storage, and never passed between functions.  I think it makes the
code more verbose without any corresponding benefit, so this patch
removes it.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-wingeneral.c (tui_gen_win_info::make_window): Update.
* tui/tui-win.c (tui_adjust_win_heights, tui_resize_all): Update.
* tui/tui-layout.c (tui_gen_win_info::resize): Update.
* tui/tui-data.h (struct tui_point): Remove.
(struct tui_gen_win_info) <origin>: Remove.
<x, y>: New fields.
* tui/tui-command.c (tui_cmd_window::resize): Update.

Change-Id: I3f77920585b9ea9e2b4b189f3f3ae32d4da0c252

4 years agoIntroduce the tui_gen_win_info::min_height method
Tom Tromey [Mon, 7 Oct 2019 23:36:49 +0000 (17:36 -0600)] 
Introduce the tui_gen_win_info::min_height method

This introduces a new method, tui_gen_win_info::min_height, to fetch
the minimum height of a window.  This is used in the subsequent
unified layout patch.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-stack.h (struct tui_locator_window) <min_height>:
Implement.
* tui/tui-regs.h (struct tui_data_item_window) <min_height>:
Implement.
* tui/tui-data.h (struct tui_gen_win_info) <min_height>: New
method.
(struct tui_win_info) <min_height>: Implement.

Change-Id: Id33baffdf041fde072e15c1ff89b75f8b8118adb

4 years agoMove can_box to tui_gen_win_info
Tom Tromey [Wed, 6 Nov 2019 00:12:04 +0000 (17:12 -0700)] 
Move can_box to tui_gen_win_info

This moves the can_box method to tui_gen_win_info, so that it will be
available on the tui_locator_window class.  This will be used in a
subsequent patch.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-data.h (struct tui_gen_win_info) <can_box>: New method.
(struct tui_win_info) <can_box>: Update.

Change-Id: Idfa58af41341607932d3c39415f6a35ee9b5d3dc

4 years agoMove max_height method to tui_gen_win_info
Tom Tromey [Mon, 7 Oct 2019 22:59:02 +0000 (16:59 -0600)] 
Move max_height method to tui_gen_win_info

This moves the max_height method to tui_gen_win_info and implements it
in the subclasses.  This is used by a subsequent patch, which will
normalize window layout across all window types.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tom@tromey.com>

* tui/tui-stack.h (struct tui_locator_window) <max_height>: New
method.
* tui/tui-regs.h (struct tui_data_item_window) <max_height>: New
method.
* tui/tui-data.h (struct tui_gen_win_info) <max_height>: New
method.
(struct tui_win_info) <max_height>: Now override.

Change-Id: I4ba3e8899bc4668328d3d78e3c1674c61882450d

4 years agoUpdate NEWS post GDB 9 branch creation.
Joel Brobecker [Wed, 11 Dec 2019 20:44:47 +0000 (21:44 +0100)] 
Update NEWS post GDB 9 branch creation.

gdb/ChangeLog:

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

4 years agoBump version to 10.0.50.DATE-git.
Joel Brobecker [Wed, 11 Dec 2019 20:37:51 +0000 (21:37 +0100)] 
Bump version to 10.0.50.DATE-git.

Now that the GDB 9 branch has been created, we can
bump the version number.

gdb/ChangeLog:

GDB 9 branch created (27f7b2f64062ac9e52afc60509263c2702a9ebd0):
* version.in: Bump version to 10.0.50.DATE-git.

4 years agoFix typo, get_Frame_id -> get_frame_id
Simon Marchi [Wed, 11 Dec 2019 16:30:15 +0000 (11:30 -0500)] 
Fix typo, get_Frame_id -> get_frame_id

gdb/doc/ChangeLog:

* gdb.texinfo (Writing JIT Debug Info Readers): Fix typo.

4 years ago[gas][arm] Add -mwarn-restrict-it
Andre Vieira [Wed, 11 Dec 2019 15:53:26 +0000 (15:53 +0000)] 
[gas][arm] Add -mwarn-restrict-it

Add a -m{no-}warn-restrict-it option to control IT related warnings in
ARMv8-A and ARMv8-R.  This is disabled by default.

Committed on behalf of Wilco Dijkstra.

gas/ChangeLog:
2019-12-11  Wilco Dijkstra  <wdijkstr@arm.com>

* config/tc-arm.c (warn_on_restrict_it): Add new variable.
(it_fsm_post_encode): Check warn_on_restrict_it.
(arm_option_table): Add -mwarn-restrict-it/-mno-warn-restrict-it.
* testsuite/gas/arm/armv8-2-fp16-scalar-bad.d: Add -mwarn-restrict-it.
* testsuite/gas/arm/armv8-2-fp16-scalar-bad-ext.d: Likewise.
* testsuite/gas/arm/armv8-a-bad.d: Likewise.
* testsuite/gas/arm/armv8-a-it-bad.d: Likewise.
* testsuite/gas/arm/armv8-r-bad.d: Likewise.
* testsuite/gas/arm/armv8-r-it-bad.d: Likewise.
* testsuite/gas/arm/sp-pc-validations-bad-t-v8a.d: Likewise.
* testsuite/gas/arm/udf.d: Likewise.

4 years agoFix build on macOS
Tom Tromey [Tue, 10 Dec 2019 21:16:19 +0000 (14:16 -0700)] 
Fix build on macOS

PR build/25268 points out that the build fails on macOS, because on
macOS the "pthread_setname_np" function takes a single argument.

This patch fixes the problem, by introducing a new adapter function
that handles both styles of pthread_setname_np.

This change also meant moving the pthread_setname_np call to the
thread function, because macOS only permits setting the name of the
current thread.  This means that there can be a brief window when gdb
will see the wrong name; but I think this is a minor concern.

Tested by rebuilding on x86-64 Fedora 30, and on macOS High Sierra.
On Linux I also debugged gdb to ensure that the thread names are still
set correctly.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tromey@adacore.com>

PR build/25268:
* gdbsupport/thread-pool.c (set_thread_name): New function.
(thread_pool::set_thread_count): Don't call pthread_setname_np.
(thread_pool::thread_function): Call set_thread_name.

Change-Id: Id7bf28d99ca27a893a9fc87ebb90b15a9c2a9cb4

4 years agoFix the build after bfd_get_signed_8 change
Tom Tromey [Wed, 11 Dec 2019 14:58:46 +0000 (07:58 -0700)] 
Fix the build after bfd_get_signed_8 change

A recent commit changed bfd_get_signed_8 to extend the result to a
bfd_signed_vma.  This caused a compiler error in one spot in my
--enable-targets=all gdb build, where the result of bfd_get_signed_8
was passed to printf.

This patch fixes the build.  Tested by rebuilding.

gdb/ChangeLog
2019-12-11  Tom Tromey  <tromey@adacore.com>

* fbsd-tdep.c (fbsd_core_info_proc_status): Cast result of
bfd_get_signed_8.

Change-Id: Ic015f5fd3d88da6b5da8f7b4e1d11d5c981333db

4 years agoRemove more shifts for sign/zero extension
Alan Modra [Wed, 11 Dec 2019 06:15:14 +0000 (16:45 +1030)] 
Remove more shifts for sign/zero extension

cpu/
* epiphany.cpu (f-sdisp11): Don't sign extend with shifts.
* lm32.cpu (f-branch, f-vall): Likewise.
* m32.cpu (f-lab-8-16): Likewise.
opcodes/
* arc-dis.c (BITS): Don't truncate high bits with shifts.
* nios2-dis.c (nios2_print_insn_arg): Don't sign extend with shifts.
* tic54x-dis.c (print_instruction): Likewise.
* tilegx-opc.c (parse_insn_tilegx): Likewise.
* tilepro-opc.c (parse_insn_tilepro): Likewise.
* visium-dis.c (disassem_class0): Likewise.
* pdp11-dis.c (sign_extend): Likewise.
(SIGN_BITS): Delete.
* epiphany-ibld.c: Regenerate.
* lm32-ibld.c: Regenerate.
* m32c-ibld.c: Regenerate.

4 years agobfd signed overflow fixes
Alan Modra [Wed, 11 Dec 2019 03:02:25 +0000 (13:32 +1030)] 
bfd signed overflow fixes

Aimed at quietening ubsan.

include/
* opcode/mmix.h (PUSHGO_INSN_BYTE): Make unsigned.
(GO_INSN_BYTE, SETL_INSN_BYTE, INCML_INSN_BYTE, INCMH_INSN_BYTE),
(INCH_INSN_BYTE, SWYM_INSN_BYTE, JMP_INSN_BYTE): Likewise.
bfd/
* elf32-rx.c (elf32_rx_relax_section): Avoid signed overflow.
* libaout.h (N_SET_INFO, N_SET_FLAGS): Likewise.
* netbsd.h (write_object_contents): Likewise.
* elf32-arm.c (bfd_elf32_arm_vfp11_erratum_scan): Likewise.
* libhppa.h (HPPA_R_CONSTANT): Don't signed extend with shifts.
(stm32l4xx_create_replacing_stub_vldm): Don't truncate high bits
with shifts.
* elf32-nds32.h (R_NDS32_RELAX_ENTRY_DISABLE_RELAX_FLAG): Define
using 1u shifted left.  Ditto for other macros.
* mmo.c (LOP): Make unsigned.

4 years agoMake bfd_get_8 return a bfd_vma like other bfd_get_* functions
Alan Modra [Wed, 11 Dec 2019 10:36:17 +0000 (21:06 +1030)] 
Make bfd_get_8 return a bfd_vma like other bfd_get_* functions

And similarly for bfd_get_signed_8.

bfd/
* libbfd.c (bfd_get_8): Return a bfd_vma.
(bfd_get_signed_8): Return a bfd_signed_vma.
* bfd-in2.h: Regenerate.
binutils/
* od-xcoff.c (dump_dumpx_core): Adjust for bfd_h_get_8 change.

4 years agox86: further refine SSE check (SSE4a, SHA, GFNI)
Jan Beulich [Wed, 11 Dec 2019 08:42:29 +0000 (09:42 +0100)] 
x86: further refine SSE check (SSE4a, SHA, GFNI)

In  ("x86: extend SSE check to PCLMULQDQ, AES, and GFNI insns") I went
both a little too far and not quite far enough:
- GFNI insns also have AVX512 variants, which also shouldn't get
  diagnosed,
- SSE4a insns should get diagnosed just like SSE4.x ones,
- SHA insns should get diagnosed just like PCLMULQDQ or AES ones.

4 years agoImplement 'print -raw-values' and 'set print raw-values on|off'
Philippe Waroquiers [Wed, 7 Aug 2019 18:50:54 +0000 (20:50 +0200)] 
Implement 'print -raw-values' and 'set print raw-values on|off'

The option framework documentation was speaking about a 'print -raw'
option, but this option does not exist.

This patch implements -raw-values option that tells to ignore the
active pretty printers when printing a value.
As we already have -raw-frame-arguments, I thought -raw-values
was more clear, in particular to differentiate
   set print raw-values and set print raw-frame-arguments.

gdb/doc/ChangeLog
2019-12-11  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.texinfo (Command Options): Use -p and -pretty in the example,
as -r is ambiguous.  Update the print - TAB TAB completion result.
(Data): Document new option -raw-values.  Use -p and -pretty in the
 example, as -r is ambiguous.
(Print Settings): Document set print raw values.
(Pretty-Printer Commands): Document interaction between enabled
pretty printers and -raw-values/-raw-frame-arguments.

gdb/ChangeLog
2019-12-11  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* NEWS: Document -raw-values option and the related setting commands.
* printcmd.c (print_command_parse_format): Do not set opts->raw off,
only set it on when /r is given.
* valprint.c (value_print_option_defs): New element raw-values.
* Makefile.in: Add the new file.

gdb/testsuite/ChangeLog
2019-12-11  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.base/options.exp: Add -raw-values in the print completion list.
* gdb.python/py-prettyprint.exp: Add tests for -raw-values.

4 years agoRe: ubsan: ns32k: left shift cannot be represented in type
Alan Modra [Wed, 11 Dec 2019 02:41:07 +0000 (13:11 +1030)] 
Re: ubsan: ns32k: left shift cannot be represented in type

* ns32k-dis.c (sign_extend): Correct last patch.

4 years agoubsan: vax: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 22:27:45 +0000 (08:57 +1030)] 
ubsan: vax: left shift cannot be represented in type 'int'

* vax-dis.c (NEXTLONG): Avoid signed overflow.

4 years agoubsan: v850: left shift cannot be represented in type 'long'
Alan Modra [Tue, 10 Dec 2019 22:23:50 +0000 (08:53 +1030)] 
ubsan: v850: left shift cannot be represented in type 'long'

* v850-dis.c (get_operand_value): Use unsigned arithmetic.  Don't
sign extend using shifts.

4 years agoubsan: tic6x: shift left of int
Alan Modra [Tue, 10 Dec 2019 22:22:22 +0000 (08:52 +1030)] 
ubsan: tic6x: shift left of int

* tic6x-dis.c (tic6x_extract_32): Avoid signed overflow.

4 years agoubsan: tic4x: segv and signed shifts
Alan Modra [Tue, 10 Dec 2019 22:17:35 +0000 (08:47 +1030)] 
ubsan: tic4x: segv and signed shifts

* tic4x-dis.c (tic4x_print_register): Formatting.  Don't segfault
on NULL registertable entry.
(tic4x_hash_opcode): Use unsigned arithmetic.

4 years agoubsan: s12z: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 22:07:53 +0000 (08:37 +1030)] 
ubsan: s12z: left shift cannot be represented in type 'int'

* s12z-opc.c (z_decode_signed_value): Avoid signed overflow.

4 years agoubsan: ns32k: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 22:04:57 +0000 (08:34 +1030)] 
ubsan: ns32k: left shift cannot be represented in type 'int'

* ns32k-dis.c (bit_extract): Use unsigned arithmetic.
(bit_extract_simple, sign_extend): Likewise.

4 years agoubsan: nios2: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 22:02:18 +0000 (08:32 +1030)] 
ubsan: nios2: left shift cannot be represented in type 'int'

* nios2-dis.c (nios2_print_insn_arg): Use 1u << 31.

4 years agoubsan: moxie: left shift of negative value
Alan Modra [Tue, 10 Dec 2019 21:56:58 +0000 (08:26 +1030)] 
ubsan: moxie: left shift of negative value

* moxie-dis.c (INST2OFFSET): Don't sign extend using shifts.

4 years agoubsan: m68k: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 21:53:33 +0000 (08:23 +1030)] 
ubsan: m68k: left shift cannot be represented in type 'int'

* m68k-dis.c (COERCE32): Cast value first.
(NEXTLONG, NEXTULONG): Avoid signed overflow.

4 years agoubsan: h8300: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 13:23:57 +0000 (23:53 +1030)] 
ubsan: h8300: left shift cannot be represented in type 'int'

This is
  *cst = (data[0] << 24) + (data[1] << 16) + (data[2] << 8) + data[3];
data is unsigned char which promotes to int.

* h8300-dis.c (extract_immediate): Avoid signed overflow.
(bfd_h8_disassemble): Likewise.

4 years agoussan: d30v: index out of bounds
Alan Modra [Tue, 10 Dec 2019 13:20:02 +0000 (23:50 +1030)] 
ussan: d30v: index out of bounds

* d30v-dis.c (print_insn): Make opind unsigned.  Don't access
past end of operands array.

4 years agoubsan: csky: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 13:07:03 +0000 (23:37 +1030)] 
ubsan: csky: left shift cannot be represented in type 'int'

In the following buf is an unsigned char array, so elements are
promoted to int before arithmetic operations.

  if (dis_info.info->endian == BFD_ENDIAN_BIG)
    while (n--)
      val |= buf[n] << (n*8);
  else
    for (i = 0; i < n; i++)
      val |= buf[i] << (i*8);

* csky-dis.c (csky_chars_to_number): Rewrite.  Avoid signed
overflow when collecting bytes of a number.

4 years agoubsan: cris: signed integer overflow
Alan Modra [Tue, 10 Dec 2019 12:52:10 +0000 (23:22 +1030)] 
ubsan: cris: signed integer overflow

This was the following in print_with_operands
case 4:
  number
    = buffer[2] + buffer[3] * 256 + buffer[4] * 65536
    + buffer[5] * 0x1000000;
and buffer[5] * 0x1000000 can indeed overflow.  So to fix this we need
to use unsigned arithmetic where overflow semantics are specified.
But number is a long, and the expression is int which will be sign
extended to long.  If we make the expression unsigned it will be zero
extended.  So make number an int32_t and rearrange a little for some
of the places that need fixing.

* cris-dis.c (print_with_operands): Avoid signed integer
overflow when collecting bytes of a 32-bit integer.

4 years agoubsan: cr16: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 12:32:37 +0000 (23:02 +1030)] 
ubsan: cr16: left shift cannot be represented in type 'int'

This was:
  unsigned long mask = SBM (instruction->match_bits);
with
  #define SBM(offs)  ((((1 << (32 - offs)) -1) << (offs)))

Well, there are a couple of problems.  Firstly, the expression uses
int values (1 rather than 1u or 1ul) resulting in the ubsan error, and
secondly, a zero offs will result in a 32-bit shift which is undefined
if ints are only 32 bits.

* cr16-dis.c (EXTRACT, SBM): Rewrite.
(cr16_match_opcode): Delete duplicate bcond test.

4 years agoubsan: bfin: shift exponent is too large
Alan Modra [Tue, 10 Dec 2019 12:02:06 +0000 (22:32 +1030)] 
ubsan: bfin: shift exponent is too large

This was the following in fmtconst_val, x is unsigned int.
    x = SIGNEXTEND (x, constant_formats[cf].nbits);
Problem is, the SIGNEXTEND macro assumed its arg was a long and sign
extended by shifting left then shifting right, and didn't cast the
arg.  So don't do the silly shift thing.  It's not guaranteed to work
anyway according to the C standard.  ">>" might do a logical shift
even if its args are signed.

* bfin-dis.c (HOST_LONG_WORD_SIZE, XFIELD): Delete.
(SIGNBIT): New.
(MASKBITS, SIGNEXTEND): Rewrite.
(fmtconst): Don't use ? expression now that SIGNEXTEND uses
unsigned arithmetic, instead assign result of SIGNEXTEND back
to x.
(fmtconst_val): Use 1u in shift expression.

4 years agoubsan: arc: shift exponent 32 is too large for 32-bit type 'int'
Alan Modra [Tue, 10 Dec 2019 08:39:47 +0000 (19:09 +1030)] 
ubsan: arc: shift exponent 32 is too large for 32-bit type 'int'

When operand->bits is 32, the following results in UB.
value = (insn >> operand->shift) & ((1 << operand->bits) - 1);

* arc-dis.c (find_format_from_table): Use ull constant when
shifting by up to 32.

4 years agoaarch64 disassembler infinite loop
Alan Modra [Tue, 10 Dec 2019 08:34:37 +0000 (19:04 +1030)] 
aarch64 disassembler infinite loop

Assembling this to an object and trying to disassemble results in
objdump -d looping forever.

 .inst 0x45205120

PR 25270
* aarch64-dis.c (aarch64_decode_variant_using_iclass): Return
false when field is zero for sve_size_tsz_bhs.

4 years agoubsan: left shift of cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 08:28:38 +0000 (18:58 +1030)] 
ubsan: left shift of cannot be represented in type 'int'

* dis-asm.h (INSN_HAS_RELOC, DISASSEMBLE_DATA),
(USER_SPECIFIED_MACHINE_TYPE, WIDE_OUTPUT): Make unsigned.
* opcode/tic80.h (TIC80_OPERAND_*): Likewise.

4 years agoubsan: xtensa: left shift cannot be represented in type 'int'
Alan Modra [Tue, 10 Dec 2019 07:42:28 +0000 (18:12 +1030)] 
ubsan: xtensa: left shift cannot be represented in type 'int'

In Operand_soffsetx4_decode:
  soffsetx4_0 = 0x4 + ((((int) offset_0 << 14) >> 14) << 2);
and other places.

Don't sign extend with shifts!  This file also has many occurrences of
truncation via shifts, which aren't a problem due to using uint32, but
I dislike on principle enough to fix.

* xtensa-modules.c (Field_* functions): Don't mask using shifts.
(Operand_soffsetx4_decode, Operand_simm4_decode),
(Operand_simm8_decode, Operand_simm8x256_decode),
(Operand_simm12b_decode, Operand_label8_decode),
(Operand_label12_decode, Operand_soffset_decode),
(Operand_xt_wbr15_label_decode, Operand_xt_wbr18_label_decode): Don't
sign extend using shifts.
(Operand_immrx4_decode, Operand_uimm16x4_decode): Avoid UB in
constant.

4 years agoubsan: epiphany: left shift of negative value
Alan Modra [Tue, 10 Dec 2019 07:36:09 +0000 (18:06 +1030)] 
ubsan: epiphany: left shift of negative value

Two places in epiphany_cgen_extract_operand, "value" is a long.
        value = ((((value) << (1))) + (pc));

cpu/
* epiphany.cpu (f-simm8, f-simm24): Use multiply rather than
shift left to avoid UB on left shift of negative values.
opcodes/
* epiphany-ibld.c: Regenerate.

4 years agoubsan: ia64: left shift of negative value
Alan Modra [Tue, 10 Dec 2019 07:27:14 +0000 (17:57 +1030)] 
ubsan: ia64: left shift of negative value

Here, since val is signed:
   *valuep = (val << scale);

* cpu-ia64-opc.c (ext_imms_scaled): Avoid undefined left shift
of negative values by using unsigned vars.

4 years agoAutomatic date update in version.in
GDB Administrator [Wed, 11 Dec 2019 00:00:25 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years agoMinor fix to gdb.prompt documentation
Tom Tromey [Wed, 20 Nov 2019 23:07:46 +0000 (16:07 -0700)] 
Minor fix to gdb.prompt documentation

I noticed that an example in the gdb.prompt documentation used the
wrong kind of quotes -- because it is code, it should use a plain
ASCII quotation mark.  I also slightly shortened the sample text here,
so it would more clearly fit on a single line.

gdb/doc/ChangeLog
2019-12-10  Tom Tromey  <tom@tromey.com>

* python.texi (gdb.prompt): Use correct quotes in example.
Shorten sample text.

Change-Id: I4153928c0d88001244ad410f3943c952a6ebfeb1

4 years agoOpenMP parallel region scope tests
Kevin Buettner [Wed, 20 Sep 2017 22:40:14 +0000 (15:40 -0700)] 
OpenMP parallel region scope tests

Add tests which check for accessibility of variables from within
various OpenMP parallel regions.

Tested on Fedora 27, 28, 29, 30, and 31.  I also tested with my OpenMP
work on Fedora 30.  The test has been annotated with setup_xfail and
setup_kfail statements so that there are no unexpected failures on any
of these platforms when using gcc.  Better still, for my own testing
anyway, is that there are also no XPASSes or KPASSes either.  So,
regardless of platform, when using gcc, and regardless of whether my
(not yet public) OpenMP work is used, seeing a FAIL indicates a real
problem.

Fedora 27 results:

        # of expected passes            85
        # of expected failures          65

(Note: I have not retested F27 since v1 of the patch; it's possible
that the numbers will be slightly different for v2.)

Fedora 28, 29, 30 results:

        # of expected passes            131
        # of expected failures          4
        # of known failures             16

Fedora 30, 31 results w/ my OpenMP work:

        # of expected passes            151

The above results all use gcc, either the system gcc or a development
gcc (when testing against my OpenMP work in GDB).  I've also tested
with clang 9.0.0 and icc 19.0.5.281 20190815 on Fedora 31.

Fedora 31, clang:

FAIL: gdb.threads/omp-par-scope.exp: single_scope: first thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: single_scope: first thread: print s3
FAIL: gdb.threads/omp-par-scope.exp: single_scope: first thread: print i1
FAIL: gdb.threads/omp-par-scope.exp: single_scope: first thread: print i3
FAIL: gdb.threads/omp-par-scope.exp: single_scope: second thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: single_scope: second thread: print s3
FAIL: gdb.threads/omp-par-scope.exp: single_scope: second thread: print i1
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i02
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i11
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i12
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i22
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: second thread: print i11
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: second thread: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: after parallel: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print num
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print l
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print k
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 2nd stop: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 2nd stop: print num
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print num
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print l
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print k
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 4th stop: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 4th stop: print num
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: outer_threads: outer stop: print file_scope_var
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: outer_threads: outer stop: print i
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: outer_threads: outer stop: print j

Fedora 31, icc:

FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i12
FAIL: gdb.threads/omp-par-scope.exp: multi_scope: first thread: print i22
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 1st thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 1st thread: print i
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 1st thread: print j
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 2nd thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 2nd thread: print i
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 2nd thread: print j
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 2nd thread: print k
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 1st call: 2nd thread: print z
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 1st thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 1st thread: print i
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 1st thread: print j
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 2nd thread: print s1
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 2nd thread: print i
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 2nd thread: print j
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 2nd thread: print k
FAIL: gdb.threads/omp-par-scope.exp: nested_func: 2nd call: 2nd thread: print z
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print l
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 1st stop: print k
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print l
FAIL: gdb.threads/omp-par-scope.exp: nested_parallel: inner_threads: 3rd stop: print k

For both clang and icc, it turns out that there are some problems with
the DWARF that these compilers generate.  Of the two, icc does at
least nest the subprogram of the outlined function representing the
parallel region within the function that it's defined, but does not
handle inner scopes if they exist.  clang places the subprogram for
the outlined function at the same level as the containing function, so
variables declared within the function aren't visible at all.

I could call setup_xfail to avoid FAILs for clang and icc also, but I don't
want to further complicate the test.

gdb/testsuite/ChangeLog:

* gdb.threads/omp-par-scope.c: New file.
* gdb/threads/omp-par-scope.exp: New file.

Change-Id: Icb9c991730d84ca7509380af817dfcc778e764ea

4 years agoAdd gdb_caching_proc support_nested_function_tests to lib/gdb.exp
Kevin Buettner [Sat, 9 Nov 2019 18:09:41 +0000 (11:09 -0700)] 
Add gdb_caching_proc support_nested_function_tests to lib/gdb.exp

This commit adds the gdb_caching_proc, support_nested_function_tests,
to lib/gdb.exp.  It tests to see whether or not the C compiler has
support for nested function calls.

gdb/testsuite/ChangeLog:

* lib/gdb.exp (support_nested_function_tests): New proc.

Change-Id: Ic2c93bc4cc200e07e104a2398f89a9c0514bdc75

4 years agoAdd gdb_compile_openmp to lib/gdb.exp
Kevin Buettner [Thu, 22 Aug 2019 02:32:12 +0000 (19:32 -0700)] 
Add gdb_compile_openmp to lib/gdb.exp

gdb/testsuite/ChangeLog:

* lib/gdb.exp (gdb_compile_openmp): New proc.
(build_executable_from_specs): Add an "openmp" option.
(gdb_compile_pthreads): Add non-executable case.

Change-Id: I94048b8b0940c707ce0529a6bcfa6e4eace49101

4 years agoSuppress the "unused function" warning for select_strerror_r
Christian Biesinger [Tue, 10 Dec 2019 19:42:40 +0000 (13:42 -0600)] 
Suppress the "unused function" warning for select_strerror_r

We only ever use one of the two overloads, so to avoid breaking -Werror
builds, supress the warning.

gdb/ChangeLog:

2019-12-10  Christian Biesinger  <cbiesinger@google.com>

* gdbsupport/safe-strerror.c: Supress the unused function warning
for select_strerror_r.

Change-Id: I344869a382bb36fe181b5b2a31838d1d20f58169

4 years agoReplace the remaining uses of strerror with safe_strerror
Christian Biesinger [Wed, 6 Nov 2019 19:57:52 +0000 (13:57 -0600)] 
Replace the remaining uses of strerror with safe_strerror

To do that, this patch makes IPA compile safe-strerror as well. Because
it doesn't use Gnulib, it calls the Glibc version of strerror_r directly.

Consequently this patch also removes the configure checks for strerror.

gdb/ChangeLog:

2019-12-10  Christian Biesinger  <cbiesinger@google.com>

* config.in: Regenerate.
* configure: Regenerate.
* gdbsupport/agent.c (gdb_connect_sync_socket): Call
safe_strerror instead of strerror.
* gdbsupport/common.m4: Don't check for strerror.
* gdbsupport/safe-strerror.c: Support both the glibc version
of strerror_r and the XSI version.

gdb/gdbserver/ChangeLog:

2019-12-10  Christian Biesinger  <cbiesinger@google.com>

* Makefile.in: Add safe-strerror.c to gdbreplay and IPA, and change
UNDO_GNULIB_CFLAGS to undo strerror_r instead of strerror.
* config.in: Regenerate.
* configure: Regenerate.
* configure.ac: Don't check for strerror.
* linux-i386-ipa.c (initialize_fast_tracepoint_trampoline_buffer):
Call safe_strerror instead of strerror.
* server.h (strerror): Remove this now-unnecessary declaration.
* tracepoint.c (init_named_socket): Call safe_strerror instead of
strerror.
(gdb_agent_helper_thread): Likewise.
* utils.c (perror_with_name): Likewise.

Change-Id: I74848f072dcde75cb55c435ef9398dc8f958cd73

4 years ago[gas][arm] Set context table for '.arch_extension'
Andre Vieira [Tue, 10 Dec 2019 16:10:17 +0000 (16:10 +0000)] 
[gas][arm] Set context table for '.arch_extension'

This patch fixes .arch_extension behaviour.

Currently, context table for '.arch_extension' is only set while
"-march" processing, but it is not set while .arch processing,
so following code is rejected

  .syntax unified
  .thumb
  .arch armv8.1-m.main
  .arch_extension mve.fp

  unless -march=armv8.1-m.main is given.

Committing on behalf of Vladimir Murzin

gas/ChangeLog:
2019-12-10  Vladimir Murzin  <vladimir.murzin@arm.com>

* config/tc-arm.c (s_arm_arch): Set selected_ctx_ext_table.
* testsuite/gas/arm/mve-arch-ext.s: New.
* testsuite/gas/arm/mve-arch-ext.d: New.

4 years agoNormalize Ada ptype to use a single "?"
Tom Tromey [Tue, 3 Dec 2019 20:31:21 +0000 (13:31 -0700)] 
Normalize Ada ptype to use a single "?"

Sometimes -- notably with unchecked unions -- the Ada "ptype" code
will print a "?" or "??" to indicate something unknown.  The choice of
what was printed was somewhat arbitrary, and in one case, Ada would
print an empty string rather than "?".

This patch normalizes the Ada code to use "?" rather than an empty
string or "??".  My reasoning here is that a single question mark is
enough to convey unknown-ness.

gdb/ChangeLog
2019-12-10  Tom Tromey  <tromey@adacore.com>

* ada-typeprint.c (print_choices): Use a single "?".
(print_variant_part): Print "?" if the discriminant name
is not known.

gdb/testsuite/ChangeLog
2019-12-10  Tom Tromey  <tromey@adacore.com>

* gdb.ada/unchecked_union.exp: New file.
* gdb.ada/unchecked_union/pck.adb: New file.
* gdb.ada/unchecked_union/pck.ads: New file.
* gdb.ada/unchecked_union/unchecked_union.adb: New file.
* gdb-utils.exp (string_to_regexp): Also quote "?".

Change-Id: I3403040780a155ffa2c44c8e6a04ba86bc810e29

4 years agoAutomatic date update in version.in
GDB Administrator [Tue, 10 Dec 2019 00:00:15 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years agoPR24960, Memory leak from disassembler
Alan Modra [Mon, 9 Dec 2019 10:04:49 +0000 (20:34 +1030)] 
PR24960, Memory leak from disassembler

PR 24960
include/
* dis-asm.h (disassemble_free_target): Declare.
opcodes/
* disassemble.c (disassemble_free_target): New function.
binutils/
* objdump.c (disassemble_data): Call disassemble_free_target.

4 years agoUse disassemble_info.private_data in place of insn_sets
Alan Modra [Mon, 9 Dec 2019 12:16:26 +0000 (22:46 +1030)] 
Use disassemble_info.private_data in place of insn_sets

No cgen target uses private_data.  This patch removes a
disassemble_info field that is only used by cgen, and instead uses
private_data.  It also removes a macro that is no longer used.

include/
* dis-asm.h (struct disassemble_info): Delete insn_sets.
(INIT_DISASSEMBLE_INFO_NO_ARCH): Don't define.
opcodes/
* cgen-dis.in (print_insn_@arch@): Replace insn_sets with private_data.
* disassemble.c (disassemble_init_for_target): Likewise.
* bpf-dis.c: Regenerate.
* epiphany-dis.c: Regenerate.
* fr30-dis.c: Regenerate.
* frv-dis.c: Regenerate.
* ip2k-dis.c: Regenerate.
* iq2000-dis.c: Regenerate.
* lm32-dis.c: Regenerate.
* m32c-dis.c: Regenerate.
* m32r-dis.c: Regenerate.
* mep-dis.c: Regenerate.
* mt-dis.c: Regenerate.
* or1k-dis.c: Regenerate.
* xc16x-dis.c: Regenerate.
* xstormy16-dis.c: Regenerate.

4 years agoRemove backup ppc struct dis_private.
Alan Modra [Mon, 9 Dec 2019 12:27:15 +0000 (22:57 +1030)] 
Remove backup ppc struct dis_private.

ppc-dis.c used a global struct whenever malloc failed to provide the
eight bytes of memory necessary for struct dis_private.  Which is
quite ridiculous.  If that malloc failed there is zero chance some
other malloc won't fail too.

* ppc-dis.c (private): Delete variable.
(get_powerpc_dialect): Don't segfault on NULL info->private_data.
(powerpc_init_dialect): Don't use global private.

4 years agoDelete duplicate struct objdump_disasm_info field
Alan Modra [Mon, 9 Dec 2019 10:17:04 +0000 (20:47 +1030)] 
Delete duplicate struct objdump_disasm_info field

objdump_disasm_info.sec duplicates disassemble_info.section

* objdump.c (struct objdump_disasm_info): Delete "sec".
(find_symbol_for_address): Use inf->section rather than aux->sec.
(objdump_print_addr, disassemble_bytes): Likewise.
(disassemble_section): Don't set aux->sec.

4 years agos12z-opc.c formatting fixes
Alan Modra [Sat, 7 Dec 2019 05:34:33 +0000 (16:04 +1030)] 
s12z-opc.c formatting fixes

Wrap overlong lines, whitespace fixes, and for function definitions
start a line with the name of the function.

* s12z-opc.c: Formatting.

4 years agoFix scripted probe breakpoints
George Barrett [Mon, 9 Dec 2019 21:28:39 +0000 (08:28 +1100)] 
Fix scripted probe breakpoints

The documentation for make-breakpoint from the Guile API and the `spec'
variant of the gdb.Breakpoint constructor from the Python API state that
the format acceptable for location strings is the same as that accepted
by the break command. However, using the -probe qualifier at the
beginning of the location string causes a GDB internal error as it
attempts to decode a probe location in the wrong code path. Without this
functionality, there doesn't appear to be another way to set breakpoints
on probe points from Python or Guile scripts.

This patch introduces a new helper function that returns a
breakpoint_ops instance appropriate for a parsed location and updates
the Guile and Python bindings to use said function, rather than the
current hard-coded use of bkpt_breakpoint_ops. Since this logic is
duplicated in the handling of the `break' and `trace' commands, those
are also updated to call into the new helper function.

gdb/ChangeLog:
2019-12-10  George Barrett  <bob@bob131.so>

Fix scripted probe breakpoints.
* breakpoint.c (tracepoint_probe_breakpoint_ops): Move
declaration forward.
(breakpoint_ops_for_event_location_type)
(breakpoint_ops_for_event_location): Add function definitions.
(break_command_1, trace_command): Use
breakpoint_ops_for_event_location.
* breakpoint.h (breakpoint_ops_for_event_location): Add function
declarations.
* guile/scm-breakpoint.c (gdbscm_register_breakpoint_x): Use
breakpoint_ops_for_event_location.
* python/py-breakpoint.c (bppy_init): Use
breakpoint_ops_for_event_location.

gdb/testsuite/ChangeLog:
2019-12-10  George Barrett  <bob@bob131.so>

Test scripted probe breakpoints.
* gdb.guile/scm-breakpoint.c (main): Add probe point.
* gdb.python/py-breakpoint.c (main): Likewise.
* gdb.guile/scm-breakpoint.exp (test_bkpt_probe): Add probe
specifier test.
* gdb.python/py-breakpoint.exp (test_bkpt_probe): Likewise.

4 years agogdb: rank an lvalue argument incompatible for an rvalue parameter
Tankut Baris Aktemur [Mon, 9 Dec 2019 16:07:47 +0000 (17:07 +0100)] 
gdb: rank an lvalue argument incompatible for an rvalue parameter

Passing an lvalue argument to a function that takes an rvalue parameter
is not allowed per C++ rules.  Consider this function:

    int g (int &&x) { return x; }

Calling g as in

    int i = 5;
    int j = g (i);

is illegal.  For instance, GCC 9.2.1 yields

~~~
test.cpp: In function ‘int main()’:
test.cpp:6:14: error: cannot bind rvalue reference of type ‘int&&’ to
lvalue of type ‘int’
    6 |   int j = g (i);
      |              ^
~~~

GDB currently allows this function call:

~~~
(gdb) print g(i)
$1 = 5
~~~

Fix this by ranking an lvalue argument incompatible with an rvalue
parameter.  The behavior after this patch is:

~~~
(gdb) print g(i)
Cannot resolve function g to any overloaded instance
~~~

Tested with GCC 9.2.1.

gdb/ChangeLog:
2019-12-09  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdbtypes.c (rank_one_type): Return INCOMPATIBLE_TYPE_BADNESS
when ranking an lvalue argument for an rvalue parameter.

gdb/testsuite/ChangeLog:
2019-12-09  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdb.cp/rvalue-ref-overload.cc (g): New function that takes
an rvalue parameter.
* gdb.cp/rvalue-ref-overload.exp: Test calling it with an lvalue
parameter.

Change-Id: I4a6dfc7dac63efa1e3b9f8f391e4b736fbdccdc1

4 years agogdb/fortran: Improve output pattern in gdb.mi/mi-fortran-modules.exp
Andrew Burgess [Tue, 3 Dec 2019 20:41:27 +0000 (20:41 +0000)] 
gdb/fortran: Improve output pattern in gdb.mi/mi-fortran-modules.exp

Extend the output pattern in mi-fortran-modules.exp to skip some
system modules that appear with versions of GFortran after 7.x.x.

gdb/testsuite/ChangeLog:

* gdb.mi/mi-fortran-modules.exp: Add patterns to skip system
modules.

Change-Id: I64aaa395e554a32e8267ffa096faee53c19c0b9e

4 years agogdb/testsuite: kfail some tests if using broken gcc
Andrew Burgess [Tue, 3 Dec 2019 19:15:06 +0000 (19:15 +0000)] 
gdb/testsuite: kfail some tests if using broken gcc

In some cases the Fortran stride information generated by GCC is wrong
with versions of GCC after 7.x.x.  This commit adds kfails for the
tests in question with known bad versions of gcc.

The bug has been reported to GCC here:

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92775

gdb/testsuite/ChangeLog:

* gdb.fortran/derived-type-striding.exp: KFAIL if we are using a
broken version of GCC.

Change-Id: Iaef08e5e2c87ab3d6983b88f749d40e01aea2bc6

4 years agogdb/testsuite/fortran: Fix info-modules/info-types for gfortran 8+
Andrew Burgess [Tue, 3 Dec 2019 13:18:43 +0000 (13:18 +0000)] 
gdb/testsuite/fortran: Fix info-modules/info-types for gfortran 8+

The gdb.fortran/info-modules.exp and gdb.fortran/info-types.exp tests
are failing on versions of gfortran after 7.3 due to the inclusion of
extra "system" modules and type that were not being matched by the
current test patterns.

Rather than building increasingly complex patterns that would always
be at risk of breaking with future versions of GCC I have instead
added a new library that parses the output of the following commands:

  info types
  info variables
  info functions
  info modules
  info module functions
  info module variables

into a data structure, the test can than run checks against the
contents of this data structure.

The benefit is that we can simply ignore extra results that we don't
care about.

There is a small risk that a bug in GDB might allow us to start
reporting incorrect results in such a way that the new library will
not spot the error.  However, I have tried to mitigate this risk by
adding extra procedures into the test library (see check_no_entry) and
we can add more in future if we wanted to be even more defensive.

I tested this test file with gFortran 7.3.1, 8.3.0, and 9.2.0, I now
see 100% pass in all cases.

gdb/testsuite/ChangeLog:

* gdb.fortran/info-modules.exp: Rewrite to make use of new
sym-info-cmds library.
* gdb.fortran/info-types.exp: Likewise.
* lib/sym-info-cmds.exp: New file.

Change-Id: Iff81624f51b5afb6c95393932f3d94472d7c2970

4 years agogdb/darwin-nat.c: Fix template argument for scoped_restore_tmpl
Wataru Ashihara [Sun, 8 Dec 2019 08:00:43 +0000 (17:00 +0900)] 
gdb/darwin-nat.c: Fix template argument for scoped_restore_tmpl

This should be the type of startup_with_shell, whose type was changed
from int to bool at commit 80fd28264.

This fixes the build on macOS:

      CXX    darwin-nat.o
    In file included from ../../gdb/darwin-nat.c:22:
    In file included from ../../gdb/top.h:25:
    In file included from ../../gdb/value.h:23:
    In file included from ../../gdb/frame.h:72:
    In file included from ../../gdb/language.h:26:
    In file included from ../../gdb/symtab.h:33:
    ../../gdb/gdbsupport/gdb_optional.h:155:19: error: no matching constructor for initialization of 'scoped_restore_tmpl<int>'
        new (&m_item) T (std::forward<Args>(args)...);
                      ^  ~~~~~~~~~~~~~~~~~~~~~~~~
    ../../gdb/darwin-nat.c:1995:31: note: in instantiation of function template specialization 'gdb::optional<scoped_restore_tmpl<int> >::emplace<bool *, int>' requested here
              restore_startup_with_shell.emplace (&startup_with_shell, 0);
                                         ^
    ../../gdb/gdbsupport/scoped_restore.h:69:3: note: candidate constructor template not viable: no known conversion from 'bool *' to 'int *' for 1st argument
      scoped_restore_tmpl (T *var, T2 value)
      ^
    ../../gdb/gdbsupport/scoped_restore.h:57:3: note: candidate constructor not viable: requires single argument 'var', but 2 arguments were provided
      scoped_restore_tmpl (T *var)
      ^
    ../../gdb/gdbsupport/scoped_restore.h:76:3: note: candidate constructor not viable: requires single argument 'other', but 2 arguments were provided
      scoped_restore_tmpl (const scoped_restore_tmpl<T> &other)
      ^
    1 error generated.

gdb/ChangeLog
2019-12-08  Wataru Ashihara  <wataash@wataash.com>

* darwin-nat.c (darwin_nat_target::create_inferior): Fix
template argument for scoped_restore_tmpl from bool to int.

Change-Id: Ia0202efd34dbce69b6af5d035fa55ed89215138a

4 years agox86/Intel: fold "xmmword" with "oword"
Jan Beulich [Mon, 9 Dec 2019 12:32:11 +0000 (13:32 +0100)] 
x86/Intel: fold "xmmword" with "oword"

These are full aliases of one another, so there's no real need to use
distinct O_md* values for them.

4 years agox86/Intel: support "mmword ptr"
Jan Beulich [Mon, 9 Dec 2019 12:31:39 +0000 (13:31 +0100)] 
x86/Intel: support "mmword ptr"

This is an alias of "qword ptr", commonly used with MMX insns.

At this occasion also test (alongside the newly supported "mmword")
- "zmmword" used as expression,
- PADDB with "oword ptr" (aliasing "xmmword ptr").

4 years agox86/Intel: fix "near ptr" / "far ptr" handling
Jan Beulich [Mon, 9 Dec 2019 12:31:07 +0000 (13:31 +0100)] 
x86/Intel: fix "near ptr" / "far ptr" handling

Commit dc2be329b950 ("i386: Only check suffix in instruction mnemonic")
broke rejecting of these for floating point insns. Fix this by setting
the "byte" operand attribute, which will now (again) cause an error.

Furthermore the diagnostic for the "far ptr" case in general and for the
"near ptr" case in the non-float cases became "invalid instruction
suffix" instead of the intended "operand size mismatch". Fix this by
also setting the "tbyte" operand attribute (no insn template accepts
both byte and tbyte operands).

4 years agox86/Intel: drop pointless suffix setting for "tbyte ptr"
Jan Beulich [Mon, 9 Dec 2019 12:29:44 +0000 (13:29 +0100)] 
x86/Intel: drop pointless suffix setting for "tbyte ptr"

There are extremely few insns accepting "tbyte ptr" operand, so the
"tbyte" operand flag checking done by match_operand_size() is already
sufficient; the setting of the suffix has become meaningless anyway
with dc2be329b950 ("i386: Only check suffix in instruction mnemonic").
Fold the code with that setting the "byte" operand flag to force an
error (no insn at all accepts both "byte ptr" and tbyte ptr" operands,
except for AnySize ones where the two (conflicting) recorded types
don't matter (operand_size_match() doesn't call match_operand_size() in
this case).

4 years agox86/Intel: drop pointless suffix setting for "fword ptr"
Jan Beulich [Mon, 9 Dec 2019 12:29:14 +0000 (13:29 +0100)] 
x86/Intel: drop pointless suffix setting for "fword ptr"

No floating point insn accepts an "fword ptr" operand, so the "fword"
operand flag checking done by match_mem_size() is already sufficient;
the setting of the suffix has become meaningless anyway with
dc2be329b950 ("i386: Only check suffix in instruction mnemonic").

4 years agox86/Intel: drop pointless special casing of LxS
Jan Beulich [Mon, 9 Dec 2019 12:28:43 +0000 (13:28 +0100)] 
x86/Intel: drop pointless special casing of LxS

LDS et al don't accept "word ptr" operands anyway, as per their insn
templates. Hence there's no need to special case this here; the check
has become dysfunctional anyway by dc2be329b950 ("i386: Only check
suffix in instruction mnemonic").

4 years agoAutomatic date update in version.in
GDB Administrator [Mon, 9 Dec 2019 00:00:26 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years agoS12Z disassembler memory leak
Alan Modra [Sat, 7 Dec 2019 00:55:52 +0000 (11:25 +1030)] 
S12Z disassembler memory leak

* s12z-opc.c (exg_sex_discrim): Don't leak memory on invalid
registers.

4 years agoaarch64*-*-*ilp32 gas tests
Alan Modra [Sat, 7 Dec 2019 12:01:43 +0000 (22:31 +1030)] 
aarch64*-*-*ilp32 gas tests

The new pac_negate_ra_state test adds yet another failure on aarch64
ipl32 targets.  This patch fixes that particular problem and a few
more that are trivial to fix.

* testsuite/gas/aarch64/bfloat16.d: Match 32-bit and 64-bit output.
* testsuite/gas/aarch64/dgh.d: Likewise.
* testsuite/gas/aarch64/f32mm.d: Likewise.
* testsuite/gas/aarch64/f64mm.d: Likewise.
* testsuite/gas/aarch64/i8mm.d: Likewise.
* testsuite/gas/aarch64/pac_ab_key.d: Likewise.
* testsuite/gas/aarch64/pac_negate_ra_state.d: Likewise.
* testsuite/gas/aarch64/reloc-prel_g0.d: Likewise.
* testsuite/gas/aarch64/reloc-prel_g0_nc.d: Likewise.
* testsuite/gas/aarch64/reloc-prel_g1.d: Likewise.
* testsuite/gas/aarch64/sve-bfloat-movprfx.d: Likewise.
* testsuite/gas/aarch64/sve-movprfx-mm.d: Likewise.
* testsuite/gas/aarch64/sve2.d: Likewise.

4 years agoOptional matching for run_dump_test
Alan Modra [Sat, 7 Dec 2019 12:02:52 +0000 (22:32 +1030)] 
Optional matching for run_dump_test

This adds an alternative to using #... in dump files, useful where we
only want to allow specific extra output.  DW_CFA_nop in CIEs and FDEs
to pad out to required alignment (larger for 64-bit than 32-bit) is
an example where these optional match patterns are useful.

binutils/
* testsuite/lib/binutils-common.exp (regexp_diff): Support #?REGEXP.
ld/
* testsuite/ld-elf/eh4.d: Match optional padding DW_CFA_nop in FDEs.
* testsuite/ld-elf/eh5.d: Likewise, and extra CIEs emitted on
embedded targets.

4 years agoFix inter-CU references using intra-CU form in imported-unit
Tom de Vries [Sun, 8 Dec 2019 10:25:28 +0000 (11:25 +0100)] 
Fix inter-CU references using intra-CU form in imported-unit

When running the gdb testsuite with the cc-with-dwz board, I run into:
...
Running gdb/testsuite/gdb.dwarf2/imported-unit.exp ...
gdb compile failed, dwz: gdb.dwarf2/imported-unit/imported-unit: \
  Couldn't find DIE referenced by DW_AT_abstract_origin
cc-with-tweaks.sh: dwz did not modify gdb.dwarf2/imported-unit/imported-unit.
...

The problem is that the DW_AT_abstract_origin reference here:
...
 <0><d2>: Abbrev Number: 2 (DW_TAG_compile_unit)
 <1><e6>: Abbrev Number: 4 (DW_TAG_subprogram)
    <e7>   DW_AT_abstract_origin: <0x142>
    <eb>   DW_AT_low_pc      : 0x4004b2
    <f3>   DW_AT_high_pc     : 0x4004c8
...
referring to a DIE in another compilation unit here:
...
 <0><129>: Abbrev Number: 2 (DW_TAG_compile_unit)
  <1><142>: Abbrev Number: 4 (DW_TAG_subprogram)
    <143>   DW_AT_name        : main
    <148>   DW_AT_type        : <0x13b>
    <14c>   DW_AT_external    : 1
...
is encoded using intra-CU reference form DW_FORM_ref4 instead of intra-CU
reference DW_FORM_ref_addr:
...
   4      DW_TAG_subprogram    [has children]
    DW_AT_abstract_origin DW_FORM_ref4
    DW_AT_low_pc       DW_FORM_addr
    DW_AT_high_pc      DW_FORM_addr
    DW_AT value: 0     DW_FORM value: 0
...

Fix this in the DWARF assembler by making all inter-CU references use the '%'
label prefix.

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2019-12-08  Tom de Vries  <tdevries@suse.de>

* gdb.dwarf2/imported-unit.exp: Fix inter-CU references.

Change-Id: I690ff18c3943705ed478453531b176ff74700f3c

4 years agoAutomatic date update in version.in
GDB Administrator [Sun, 8 Dec 2019 00:01:12 +0000 (00:01 +0000)] 
Automatic date update in version.in

4 years agoCore file build-id support
Keith Seitz [Sat, 7 Dec 2019 17:35:03 +0000 (09:35 -0800)] 
Core file build-id support

This patch uses new BFD support for detecting build-ids in core
files.

After this patch, it is possible to run gdb with only the
core file, and gdb will automatically load the executable and
debug info [example from tests]:

$ gdb -nx -q
(gdb) core-file corefile-buildid.core
[New LWP 29471]
Reading symbols from gdb.base/corefile-buildid/debugdir-exec/.build-id/36/fe5722c5a7ca3ac746a84e223c6a2a69193a24...
Core was generated by `outputs/gdb.base/coref'.
Program terminated with signal SIGABRT, Aborted.
(gdb)

This work is based on functionality available in Fedora originally
written by Jan Kratochvil.

Regression tested on buildbot.

gdb/ChangeLog:
2019-12-07  Keith Seitz  <keiths@redhat.com>

* build-id.c (build_id_bfd_get): Permit bfd_core, too.
(build_id_to_debug_bfd): Make static, rewriting to use
build_id_to_bfd_suffix.
(build_id_to_bfd_suffix): Copy of build_id_to_debug_bfd,
adding `suffix' parameter. Append SUFFIX to file names
when searching for matching files.
(build_id_to_debug_bfd): Use build_id_to_bfd_suffix.
(build_id_to_exec_bfd): Likewise.
* build-id.h (build_id_to_debug_bfd): Clarify that function
searches for BFD of debug info file.
(build_id_to_exec_bfd): Declare.
* corelow.c: Include build-id.h.
(locate_exec_from_corefile_build_id): New function.
(core_target_open): If no executable BFD is found,
search for a core file BFD using build-id.

gdb/testsuite/ChangeLog:
2019-12-07  Keith Seitz  <keiths@redhat.com>

* gdb.base/corefile-buildid-shlib-shr.c: New file.
* gdb.base/corefile-buildid-shlib.c: New file.
* gdb.base/corefile-buildid.c: New file.
* gdb.base/corefile-buildid.exp: New file.

Change-Id: I15e9e8e58f10c68b5cae55e2eba58df1e8aef529

4 years agoPR25236, common sym versioning
Alan Modra [Fri, 6 Dec 2019 00:51:45 +0000 (11:21 +1030)] 
PR25236, common sym versioning

In cases where a relocatable object file has a common symbol, no other
file has a definition, and there is a matching common symbol found in
a shared library then ld will output a definition using the largest of
size and alignment for the commons.  This patch fixes a bug in ld that
ignored common symbols when assigning versions, resulting in such
symbols being given VER_NDX_LOCAL versions.

PR 25236
* elflink.c (_bfd_elf_link_assign_sym_version): Assign versions
for ELF_COMMON_DEF_P symbols.
(elf_link_output_extsym, _bfd_elf_add_default_symbol): Adjust to
suit.

4 years agoAutomatic date update in version.in
GDB Administrator [Sat, 7 Dec 2019 00:00:19 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years ago[ARM, sim] Fix build error and warnings
Luis Machado [Tue, 26 Nov 2019 15:52:56 +0000 (12:52 -0300)] 
[ARM, sim] Fix build error and warnings

Newer GCC's have switched to -fno-common by default, and this breaks the build
for the ARM sim, like this:

binutils-gdb.git~gdb-8.3-release/sim/arm/maverick.c:65: multiple definition of `DSPsc'; libsim.a(wrapper.o):binutils-gdb.git~gdb-8.3-release/sim/arm/wrapper.c:134: first defined here
binutils-gdb.git~gdb-8.3-release/sim/arm/maverick.c:64: multiple definition of `DSPacc'; libsim.a(wrapper.o):binutils-gdb.git~gdb-8.3-release/sim/arm/wrapper.c:133: first defined here
binutils-gdb.git~gdb-8.3-release/sim/arm/maverick.c:63: multiple definition of `DSPregs'; libsim.a(wrapper.o):binutils-gdb.git~gdb-8.3-release/sim/arm/wrapper.c:132: first defined here

I also noticed a few warnings due to mismatching types, as follows:

../../../../repos/binutils-gdb/sim/arm/wrapper.c: In function ‘sim_create_inferior’:
../../../../repos/binutils-gdb/sim/arm/wrapper.c:335:16: warning: assignment discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
       for (arg = argv; *arg != NULL; arg++)
                ^
../../../../repos/binutils-gdb/sim/arm/wrapper.c:342:8: warning: assignment discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
    arg = argv;
        ^
../../../../repos/binutils-gdb/sim/arm/wrapper.c:345:13: warning: assignment discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
    for (arg = argv; *arg != NULL; arg++)
             ^
The following patch fixes both of the above.

sim/arm/ChangeLog:

2019-12-06  Luis Machado  <luis.machado@linaro.org>

* armemu.c (isize): Move this declaration ...
* arminit.c (isize): ... here.
* maverick.h: New file.
* wrapper.c: Include "maverick.h".
(<struct maverick_regs>, <union maverick_acc_regs>): Remove and update
comment.
(sim_create_inferior): Cast variables to proper type.
* maverick.c: Include "maverick.h".
(<struct maverick_regs>, <union maverick_acc_regs>): Move
declarations to maverick.h and update comment.
(DSPsc, DSPacc, DSPregs): Adjust comment.

Change-Id: I21db699d3b61b2de8c44053e47be4387285af28f

4 years agoPut bcache inside "namespace gdb"
Christian Biesinger [Mon, 2 Dec 2019 21:05:54 +0000 (15:05 -0600)] 
Put bcache inside "namespace gdb"

This avoids a conflict with a system "struct bcache" on
Solaris (see e.g.
https://www.isi.edu/nsnam/archive/ns-users/webarch/2001/msg05393.html)

Note that the Solaris conflict for now only surfaces with
--enable-targets=all (which the build bot doesn't use).

gdb/ChangeLog:

2019-12-06  Christian Biesinger  <cbiesinger@google.com>

* bcache.c: Put in namespace gdb.
* bcache.h: Likewise.
* gdbtypes.c (check_types_worklist): Update.
(types_deeply_equal): Update.
* macrotab.c (struct macro_table) <bcache>: Update.
(new_macro_table): Update.
* macrotab.h (struct bcache): Put this forward declaration
inside namespace gdb.
(new_macro_table): Update.
* objfiles.h (struct objfile_per_bfd_storage) <filename_cache>:
Update.
<macro_cache>: Update.
* psymtab.h: (psymtab_storage) <psymbol_cache>: Update.

Change-Id: I843d5e91f7ccb3db6d1099a8214c15a74510256f

4 years ago[gdb/symtab] Prefer var def over decl
Tom de Vries [Fri, 6 Dec 2019 17:51:49 +0000 (18:51 +0100)] 
[gdb/symtab] Prefer var def over decl

Consider the DWARF as generated by gcc with the tentative patch to fix gcc
PR91507 - "wrong debug for completed array with previous incomplete
declaration":
...
 <1><f4>: Abbrev Number: 2 (DW_TAG_array_type)
    <f5>   DW_AT_type        : <0xff>
    <f9>   DW_AT_sibling     : <0xff>
 <2><fd>: Abbrev Number: 3 (DW_TAG_subrange_type)
 <2><fe>: Abbrev Number: 0
 <1><ff>: Abbrev Number: 4 (DW_TAG_pointer_type)
    <100>   DW_AT_byte_size   : 8
    <101>   DW_AT_type        : <0x105>
 <1><105>: Abbrev Number: 5 (DW_TAG_base_type)
    <106>   DW_AT_byte_size   : 1
    <107>   DW_AT_encoding    : 6       (signed char)
    <108>   DW_AT_name        : (indirect string, offset: 0x19f): char
 <1><10c>: Abbrev Number: 6 (DW_TAG_variable)
    <10d>   DW_AT_name        : zzz
    <111>   DW_AT_decl_file   : 1
    <112>   DW_AT_decl_line   : 1
    <113>   DW_AT_decl_column : 14
    <114>   DW_AT_type        : <0xf4>
    <118>   DW_AT_external    : 1
    <118>   DW_AT_declaration : 1
 <1><118>: Abbrev Number: 2 (DW_TAG_array_type)
    <119>   DW_AT_type        : <0xff>
    <11d>   DW_AT_sibling     : <0x128>
 <1><12f>: Abbrev Number: 8 (DW_TAG_variable)
    <130>   DW_AT_specification: <0x10c>
    <134>   DW_AT_decl_line   : 2
    <135>   DW_AT_decl_column : 7
    <136>   DW_AT_type        : <0x118>
    <13a>   DW_AT_location    : 9 byte block: 3 30 10 60 0 0 0 0 0      (DW_OP_addr: 601030)
...

The DWARF will result in two entries in the symbol table, a decl with type
char *[] and a def with type char*[2].

When trying to print the value of zzz:
...
$ gdb a.spec.out -batch -ex "p zzz"
...
the decl (rather than the def) will be found in the symbol table, which is
missing the location information, and consequently we get:
...
$1 = 0x601030 <zzz>
...

[ There is a fallback mechanism that finds the address of the variable in the
minimal symbol table, but that's not used here, because the type of the decl
does not specify a size.  We could use the symbol size here to get the size
of the type, but that's currently not done: PR exp/24989.  Still, fixing that
PR would not fix the generic case, where minimal symbol info is not
available. ]

Fix this by preferring defs over decls when searching in the symbol table.

Build and reg-tested on x86_64-linux.

gdb/ChangeLog:

2019-12-06  Tom de Vries  <tdevries@suse.de>

PR symtab/24971
* block.c (best_symbol, better_symbol): New function.
(block_lookup_symbol_primary, block_lookup_symbol): Prefer def over
decl.

gdb/testsuite/ChangeLog:

2019-12-06  Tom de Vries  <tdevries@suse.de>

* gdb.dwarf2/varval.exp: Add decl before def test.

Change-Id: Id92326cb8ef9903b121ef9e320658eb565d0f5a9

4 years ago[gas] Implement .cfi_negate_ra_state directive
Kyrylo Tkachov [Fri, 6 Dec 2019 10:46:32 +0000 (10:46 +0000)] 
[gas] Implement .cfi_negate_ra_state directive

This patch implements the .cfi_negate_ra_state to be consistent with
LLVM (https://reviews.llvm.org/D50136). The relevant DWARF code DW_CFA_AARCH64_negate_ra_state
is multiplexed on top of DW_CFA_GNU_window_save, as per
https://gcc.gnu.org/ml/gcc-patches/2017-08/msg00753.html

I believe this is the simplest patch implementing this and is needed to
allow users to build, for example, the Linux kernel with Armv8.3-A
pointer authentication support with Clang while using gas as the
assembler, which is a common usecase.

gas/
2019-12-06  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

    * dw2gencfi.c (cfi_pseudo_table): Add cfi_negate_ra_state.
    * testsuite/gas/aarch64/pac_negate_ra_state.s: New file.
    * testsuite/gas/aarch64/pac_negate_ra_state.d: Likewise.

4 years agoRe: Only give FDE encoding warnings if --eh-frame-hdr was specified.
Alan Modra [Fri, 6 Dec 2019 03:08:42 +0000 (13:38 +1030)] 
Re: Only give FDE encoding warnings if --eh-frame-hdr was specified.

For mips*-linux targets, fixes the following fails.
FAIL: MIPS eh-frame 1, n32
FAIL: MIPS eh-frame 1, n64
FAIL: MIPS eh-frame 2, n32
FAIL: MIPS eh-frame 2, n64

* testsuite/ld-mips-elf/eh-frame1-n32.d: Pass --eh-frame-hdr to ld.
* testsuite/ld-mips-elf/eh-frame1-n64.d: Likewise.
* testsuite/ld-mips-elf/eh-frame2-n32.d: Likewise.
* testsuite/ld-mips-elf/eh-frame2-n64.d: Likewise.

4 years agogdb/testsuite: do minor clean-up in gdb.cp/rvalue-ref-overload.exp
Tankut Baris Aktemur [Tue, 12 Nov 2019 15:06:29 +0000 (16:06 +0100)] 
gdb/testsuite: do minor clean-up in gdb.cp/rvalue-ref-overload.exp

Simplify the expected test outputs.  This is a minor cleanup; no
functional change is intended.

gdb/testsuite/ChangeLog:
2019-12-06  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdb.cp/rvalue-ref-overload.exp: Minor cleanup.

Change-Id: Ie760a2856cae3be0eeed5496765a5f1cd102d6b7

4 years agogdb: fix overload resolution for see-through references
Tankut Baris Aktemur [Tue, 12 Nov 2019 14:12:43 +0000 (15:12 +0100)] 
gdb: fix overload resolution for see-through references

The overload resolution mechanism assigns badness values to the
necessary conversions to be made on types to pick a champion.  A
badness value consists of a "rank" that scores the conversion and a
"subrank" to differentiate conversions of the same kind.

An auxiliary function, 'sum_ranks', is used for adding two badness
values.  In all of its uses, except two, 'sum_ranks' is used for
populating the subrank of a badness value.  The two exceptions are in
'rank_one_type':

~~~
  /* See through references, since we can almost make non-references
     references.  */

  if (TYPE_IS_REFERENCE (arg))
    return (sum_ranks (rank_one_type (parm, TYPE_TARGET_TYPE (arg), NULL),
       REFERENCE_CONVERSION_BADNESS));
  if (TYPE_IS_REFERENCE (parm))
    return (sum_ranks (rank_one_type (TYPE_TARGET_TYPE (parm), arg, NULL),
       REFERENCE_CONVERSION_BADNESS));
~~~

Here, the result of a recursive call is combined with
REFERENCE_CONVERSION_BADNESS.  This leads to the problem of
over-punishment by combining two ranks.  Consider this:

    void an_overloaded_function (const foo &);
    void an_overloaded_function (const foo &&);
    ...
    foo arg;
    an_overloaded_function(arg);

When ranking 'an_overloaded_function (const foo &)', the badness
values REFERENCE_CONVERSION_BADNESS and CV_CONVERSION_BADNESS are
combined, whereas 'rank_one_type' assigns only the
REFERENCE_CONVERSION_BADNESS value to 'an_overloaded_function (const
foo &&)' (there is a different execution flow for that).  This yields
in GDB picking the latter function as the overload champion instead of
the former.

In fact, the 'rank_one_type' function should have given
'an_overloaded_function (const foo &)' the CV_CONVERSION_BADNESS
value, with the see-through referencing increasing the subrank a
little bit.  This can be achieved by introducing a new badness value,
REFERENCE_SEE_THROUGH_BADNESS, which bumps up the subrank only, and
using it in the two "exceptional" cases of 'sum_ranks'.

gdb/ChangeLog:
2019-12-06  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdbtypes.h: Define the REFERENCE_SEE_THROUGH_BADNESS value.
* gdbtypes.c (rank_one_type): Use REFERENCE_SEE_THROUGH_BADNESS
for ranking see-through reference cases.

gdb/testsuite/ChangeLog:
2019-12-06  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdb.cp/rvalue-ref-overload.cc: Add a case that involves both
CV and reference conversion for overload resolution.
* gdb.cp/rvalue-ref-overload.exp: Test it.

Change-Id: I39ae6505ab85ad0bd21915368c82540ceeb3aae9

4 years agoFix crash when command arg is missing in faas/taas/tfaas commands.
Philippe Waroquiers [Thu, 5 Dec 2019 22:41:58 +0000 (23:41 +0100)] 
Fix crash when command arg is missing in faas/taas/tfaas commands.

GDB crashes when doing:
  (gdb) faas
  Aborted

Do the needed check to avoid crashing.

gdb/ChangeLog
2019-12-06  Philippe Waroquiers  <philippe.waroquiers@skynet.be>
* stack.c (faas_command): Check a command is provided.
* thread.c (taas_command, tfaas_command): Likewise.

gdb/testsuite/ChangeLog
2019-12-06  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.threads/pthreads.exp: Test taas and tfaas without command.
* gdb.base/frameapply.exp: Test faas without command.

4 years agoAutomatic date update in version.in
GDB Administrator [Fri, 6 Dec 2019 00:00:39 +0000 (00:00 +0000)] 
Automatic date update in version.in

4 years agoFix leaks when pruning inferiors.
Philippe Waroquiers [Sun, 1 Dec 2019 14:40:17 +0000 (15:40 +0100)] 
Fix leaks when pruning inferiors.

Valgrind detects various inferior related leaks, such as:
  ==31877== 5,530 (56 direct, 5,474 indirect) bytes in 1 blocks are definitely lost in loss record 7,131 of 7,355
  ==31877==    at 0x4C2E18C: calloc (vg_replace_malloc.c:760)
  ==31877==    by 0x23E580: xcalloc (alloc.c:100)
  ==31877==    by 0x4794A9: xcnewvec<void*> (poison.h:158)
  ==31877==    by 0x4794A9: registry_alloc_data(registry_data_registry*, registry_fields*) (registry.c:51)
  ==31877==    by 0x3A537C: inferior_alloc_data (inferior.c:43)
  ==31877==    by 0x3A537C: inferior::inferior(int) (inferior.c:92)
  ==31877==    by 0x3A5426: add_inferior_silent(int) (inferior.c:98)
  ==31877==    by 0x3A5530: add_inferior(int) (inferior.c:122)
  ...

Origin of the leaks is in prune_inferiors: prune_inferiors is first removing
the inferior to prune from the inferior list, then calls delete_inferior.
But delete_inferior will only really destroy the inferior when it finds
it into the inferior list.
As delete_inferior is removing the inferior to delete from the inferior list,
ensure prune_inferiors only calls delete_inferior, without touching the
inferior list.

gdb/ChangeLog
2019-12-05  Philippe Waroquiers  <philippe.waroquiers@skynet.be>
* inferior.c (prune_inferiors):  Only call delete_inferior.
Do not modify the inferior list.

4 years agoOnly give FDE encoding warnings if --eh-frame-hdr was specified.
Sandra Loosemore [Thu, 5 Dec 2019 21:20:12 +0000 (13:20 -0800)] 
Only give FDE encoding warnings if --eh-frame-hdr was specified.

This bug was observed on nios2-linux-gnu with some C++ programs
linked with -pie or -shared.  The nios2 ABI doesn't include appropriate
relocations in this instance and GCC is also being patched not to pass
--eh-frame-hdr to the linker in those cases.

2019-12-05  Sandra Loosemore  <sandra@codesourcery.com>

        bfd/
        * elf-eh-frame.c (_bfd_elf_discard_section_eh_frame): Make
        FDE encoding warning conditional.

4 years agoRemove gdbarch parameter of lookup_typename
Simon Marchi [Thu, 5 Dec 2019 18:44:30 +0000 (13:44 -0500)] 
Remove gdbarch parameter of lookup_typename

I noticed that the gdbarch parameter of lookup_typename was unused, so I
removed it (as well as from lookup_signed_typename and
lookup_unsigned_typename) and updated all callers.

Tested by rebuilding.

gdb/ChangeLog:

* c-exp.y: Update calls to lookup_typename,
lookup_signed_typename and lookup_unsigned_typename.
* c-lang.c (evaluate_subexp_c): Likewise.
* cp-namespace.c (cp_lookup_symbol_imports_or_template):
Likewise.
* eval.c (binop_promote): Likewise.
* gdbtypes.c (lookup_typename): Remove gdbarch parameter.
(lookup_unsigned_typename): Likewise.
(lookup_signed_typename): Likewise.
* gdbtypes.h (lookup_unsigned_typename): Likewise.
(lookup_signed_typename): Likewise.
(lookup_typename): Likewise.
* guile/scm-type.c (tyscm_lookup_typename): Update calls to
lookup_typename, lookup_signed_typename,
lookup_unsigned_typename.
* m2-exp.y: Likewise.
* printcmd.c (printf_wide_c_string): Likewise.
(ui_printf): Likewise.
* python/py-type.c (typy_lookup_typename): Likewise.
* python/py-xmethods.c (python_xmethod_worker::invoke):
Likewise.
* rust-exp.y: Likewise.

4 years agoFix a problem computing the size fields in the PE format header.
Nick Clifton [Thu, 5 Dec 2019 13:56:07 +0000 (13:56 +0000)] 
Fix a problem computing the size fields in the PE format header.

PR 25029
* peXXigen.c (_bfd_XXi_swap_aouthdr_out): Ignore empty sections
when computing the sizes stored in the headers.

4 years agoRe: PR25244, --print-memory-usage, division by zero if MEMORY length is zero
Alan Modra [Thu, 5 Dec 2019 10:59:21 +0000 (21:29 +1030)] 
Re: PR25244, --print-memory-usage, division by zero if MEMORY length is zero

Do print the linefeed when length is zero.

PR 25244
* ldlang.c (lang_print_memory_usage): Correct last patch.

4 years agoArm64: simplify Crypto arch extension handling
Jan Beulich [Thu, 5 Dec 2019 07:44:22 +0000 (08:44 +0100)] 
Arm64: simplify Crypto arch extension handling

This, at the assembler level, is just a "brace" feature covering both
AES and SHA2. Hence there's no need for it to have a separate feature
flag, freeing up a bit for future re-use. Along these lines there are
also a number of dead definitions/variables in the opcode table file.

This page took 0.053523 seconds and 4 git commands to generate.