binutils-gdb.git
4 years agoAutomatic date update in version.in
GDB Administrator [Sun, 3 May 2020 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agogold: Compile common tests with -fcommon
H.J. Lu [Sat, 2 May 2020 13:18:16 +0000 (06:18 -0700)]
gold: Compile common tests with -fcommon

Since GCC 10 defaults to -fno-common, add -fcommon to common tests to
force common behavior.

PR gold/25904
* testsuite/Makefile.am (COMMON_TEST_C_CFLAGS): New.
(common_test_1.o): New rule.
(common_test_2.o): Likewise.
(common_test_3.o): Likewise.
(plugin_common_test_1.o): Likewise.
(plugin_common_test_2.o): Likewise.
(common_test_1_v1.o): Likewise.
(common_test_1_v2.o): Likewise.
(common_test_2_pic.o): Compile with $(COMMON_TEST_C_CFLAGS).
(common_test_3_pic.o): Likewise.
* testsuite/Makefile.in: Regenerated.

4 years agoFix typo in comment of DYN_PROP_ASSOCIATED
Hannes Domani [Fri, 1 May 2020 12:50:08 +0000 (14:50 +0200)]
Fix typo in comment of DYN_PROP_ASSOCIATED

gdb/ChangeLog:

2020-05-02  Hannes Domani  <ssbssa@yahoo.de>

* gdbtypes.h (enum dynamic_prop_node_kind): Fix typo.

4 years agogdb: Fix formatting error in ChangeLog
Andrew Burgess [Sat, 2 May 2020 09:28:56 +0000 (10:28 +0100)]
gdb: Fix formatting error in ChangeLog

4 years ago[gdb/testsuite] Fix i386-mpx.exp compilation warnings
Tom de Vries [Sat, 2 May 2020 08:56:48 +0000 (10:56 +0200)]
[gdb/testsuite] Fix i386-mpx.exp compilation warnings

When running test-case gdb.arch/i386-mpx.exp with gcc-10, we get:
...
Running src/gdb/testsuite/gdb.arch/i386-mpx.exp ...
gdb compile failed, xgcc: warning: switch '-mmpx' is no longer supported
xgcc: warning: switch '-fcheck-pointer-bounds' is no longer supported
...

The test-case uses a combination of options, -mmpx and -fcheck-pointer-bounds.

The -fcheck-pointer-bounds option requires the -mmpx option:
...
$ gcc -fcheck-pointer-bounds ~/hello.c
hello.c:1:0: warning: Pointer Checker requires MPX support on this target. \
  Use -mmpx options to enable MPX.
 #include <stdio.h>

cc1: error: ‘-fcheck-pointer-bounds’ is not supported for this target
...

Both options is no longer supported in gcc-9.

Fix the warnings by testing if the option combination is supported.

Tested on x86_64-linux, with gcc-7.5.0 and gcc-10.0.1.

gdb/testsuite/ChangeLog:

2020-05-02  Tom de Vries  <tdevries@suse.de>

* lib/gdb.exp (supports_mpx_check_pointer_bounds): New proc.
* gdb.arch/i386-mpx-call.exp: Use supports_mpx_check_pointer_bounds.
* gdb.arch/i386-mpx-map.exp: Same.
* gdb.arch/i386-mpx-sigsegv.exp: Same.
* gdb.arch/i386-mpx-simple_segv.exp: Same.
* gdb.arch/i386-mpx.exp: Same.

4 years ago[gdb/testsuite] Update psym-external-decl.exp for gcc-10/clang
Tom de Vries [Sat, 2 May 2020 07:50:50 +0000 (09:50 +0200)]
[gdb/testsuite] Update psym-external-decl.exp for gcc-10/clang

When running test-case gdb.base/psym-external-decl.exp with gcc-10, we have:
...
(gdb) print aaa^M
'aaa' has unknown type; cast it to its declared type^M
(gdb) FAIL: gdb.base/psym-external-decl.exp: print aaa
...

With an an earlier version, gcc still emits the debug info for the
declaration of aaa:
...
 <0><d2>: Abbrev Number: 1 (DW_TAG_compile_unit)
    <d8>   DW_AT_name        : psym-external-decl.c
 <1><f4>: Abbrev Number: 2 (DW_TAG_variable)
    <f5>   DW_AT_name        : aaa
    <ff>   DW_AT_external    : 1
    <ff>   DW_AT_declaration : 1
...
but with gcc-10 that's no longer the case.

Fix the test-case by adding a use of aaa in psym-external-decl.c.

That still doesn't work for clang, so skip test in that case.

Tested with x86_64-linux, with gcc 7.5.0, gcc 10.0.0 and clang 5.0.2.

Also tested by reverting corresponding fix and ensuring test-case still
fails.

gdb/testsuite/ChangeLog:

2020-05-02  Tom de Vries  <tdevries@suse.de>

* gdb.base/psym-external-decl.c (main): Add use of variable aaa.

4 years agoAutomatic date update in version.in
GDB Administrator [Sat, 2 May 2020 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agogold: x86-64: Fix TLSDESC relaxation for x32
H.J. Lu [Fri, 1 May 2020 17:11:06 +0000 (10:11 -0700)]
gold: x86-64: Fix TLSDESC relaxation for x32

X32 TLSDESC sequences can be:

40 8d 05 00 00 00 00 rex lea foo@TLSDESC(%rip), %reg
...
67 ff 10 call *foo@TLSCALL(%eax)

or the same sequence as LP64:

48 8d 05 00 00 00 00 lea foo@TLSDESC(%rip), %reg
...
ff 10 call *foo@TLSCALL(%rax)

We need to support both sequences for x32.  For both GDesc -> IE/LE
transitions,

67 ff 10 call *foo@TLSCALL(%eax)

should relaxed to

0f 1f 00 nopl (%rax)

For GDesc -> LE transition,

40 8d 05 00 00 00 00 rex lea foo@TLSDESC(%rip), %reg

should relaxed to

40 c7 c0 fc ff ff ff rex movl $foo@tpoff, %reg

For GDesc -> IE transition,

40 8d 05 00 00 00 00 rex lea foo@TLSDESC(%rip), %reg

should relaxed to

40 8b 05 00 00 00 00 rex movl foo@gottpoff(%rip), %eax

PR gold/25426
* x86_64.cc (Target_x86_64<size>::Relocate::tls_desc_gd_to_ie):
For x32, relax "rex leal foo@tlsdesc(%rip), %reg" to
"rex movl foo@gottpoff(%rip), %eax" and relax ""call *(%eax)"
to "nopl (%rax)".
(Target_x86_64<size>::Relocate::tls_desc_gd_to_le): For x32,
relax "rex leal foo@tlsdesc(%rip), %reg" to
"rex movl foo@tpoff, %eax" and relax "call *foo@tlscall(%eax)"
to "nopl (%rax)".
* testsuite/Makefile.am (tls_test_gnu2.o): Depend on
gcctestdir/as.
(tls_test_file2_gnu2.o): Likewise.
(tls_test_c_gnu2.o): Likewise.
* testsuite/Makefile.in: Regenerated.

4 years agogold: x86-64: Fix TLSDESC -> LE relaxation
H.J. Lu [Fri, 1 May 2020 17:08:31 +0000 (10:08 -0700)]
gold: x86-64: Fix TLSDESC -> LE relaxation

X86-64 TLSDESC sequences can be:

4c 8d 0d 00 00 00 00 leaq foo@TLSDESC(%rip), %r9
4c 89 c8 movq %r9, %rax
ff 10 call *foo@TLSCALL(%rax)

TLSDESC -> LE relaxation can turn them into:

49 c7 c1 fc ff ff ff  mov    $0xfffffffffffffffc,%r9
4c 89 c8              mov    %r9,%rax
66 90                 xchg   %ax,%ax

We need to check and update the REX byte in this case.

PR gold/25473
* x86_64.cc (Target_x86_64<size>::Relocate::tls_desc_gd_to_ie):
Properly check r8 - r15 in "lea foo@TLSDESC(%rip), %reg".
(Target_x86_64<size>::Relocate::tls_desc_gd_to_le): Properly
relax r8 - r15 in "lea foo@TLSDESC(%rip), %reg".
* testsuite/Makefile.am (check_SCRIPTS): Add x86_64_gd_to_le.sh.
(check_DATA): Add x86_64_gd_to_le.stdout.
(MOSTLYCLEANFILES): Add x86_64_gd_to_le.
(x86_64_gd_to_le.o): New target.
(x86_64_gd_to_le): Likewise.
(x86_64_gd_to_le.stdout): Likewise.
* testsuite/Makefile.in: Regenerated.
* testsuite/x86_64_gd_to_le.s: New file.
* testsuite/x86_64_gd_to_le.sh: Likewise.

4 years agogold: Make Ordering::operator() even more complex
H.J. Lu [Fri, 1 May 2020 17:06:15 +0000 (10:06 -0700)]
gold: Make Ordering::operator() even more complex

GCC 9 generates the same function size for Ordering::operator() in
odr_violation1.cc and odr_violation2.cc on x32:

  134: 00000000    31 FUNC    WEAK   DEFAULT   64 _ZN8OrderingclEii
   40: 00000000    31 FUNC    GLOBAL DEFAULT   10 _ZN8OrderingclEii

This patch makes Ordering::operator() even more complex

   134: 00000000    31 FUNC    WEAK   DEFAULT   64 _ZN8OrderingclEii
    42: 00000000    35 FUNC    GLOBAL DEFAULT   11 _ZN8OrderingclEii

* testsuite/odr_violation2.cc (Ordering::operator()): Make
expression even more complex.

4 years agogold: Update ver_test_pr16504.sh
H.J. Lu [Fri, 1 May 2020 16:05:01 +0000 (09:05 -0700)]
gold: Update ver_test_pr16504.sh

commit df3a023bd614133fe69afb02cd0e8f3e590a36a9
Author: Alan Modra <amodra@gmail.com>
Date:   Tue Jul 23 17:54:42 2019 +0930

    SHF_GNU_MBIND requires ELFOSABI_GNU

changed readelf to print IFUNC only for ELFOSABI_GNU.  Since

$ readelf -h ver_test_pr16504.so
...
 OS/ABI:                            UNIX - System V
...

we get

$ readelf -sW ver_test_pr16504.so
...
     3: 0000000000000378    13 <OS specific>: 10 GLOBAL DEFAULT    9 foo@@VER1
...

Update ver_test_pr16504.sh to also accept "<OS specific>: 10".

* testsuite/ver_test_pr16504.sh: Updated.

4 years agogold: Increment plt_offset after setting TLSDESC PLT entry
H.J. Lu [Fri, 1 May 2020 16:00:14 +0000 (09:00 -0700)]
gold: Increment plt_offset after setting TLSDESC PLT entry

Increment plt_offset after setting the reserved TLSDESC PLT entry.

PR gold/25872
* x86_64.cc (Output_data_plt_x86_64_bnd::do_write): Increment
plt_offset after setting the reserved TLSDESC PLT entry.
(Output_data_plt_x86_64_ibt<size>::do_write): Likewise.

4 years agogold: Handle local IFUNC symbol for APLT
H.J. Lu [Fri, 1 May 2020 15:53:47 +0000 (08:53 -0700)]
gold: Handle local IFUNC symbol for APLT

Handle local IFUNC symbol for APLT like global IFUNC symbol.

PR gold/25872
* x86_64.cc (Output_data_plt_x86_64_bnd::do_address_for_local):
Handle local IFUNC symbol.
(Output_data_plt_x86_64_ibt::do_address_for_local): Likewise.

4 years ago[gdb/testsuite] Fix gdb.ada/operator_bp.exp breakpoint location FAILs
Tom de Vries [Fri, 1 May 2020 15:57:56 +0000 (17:57 +0200)]
[gdb/testsuite] Fix gdb.ada/operator_bp.exp breakpoint location FAILs

When running test-case gdb.ada/operator_bp.exp with gcc-10, I run into:
...
FAIL: gdb.ada/operator_bp.exp: break "+"
FAIL: gdb.ada/operator_bp.exp: break "-"
FAIL: gdb.ada/operator_bp.exp: break "<"
FAIL: gdb.ada/operator_bp.exp: break "<="
FAIL: gdb.ada/operator_bp.exp: break ">"
FAIL: gdb.ada/operator_bp.exp: break ">="
FAIL: gdb.ada/operator_bp.exp: break "="
FAIL: gdb.ada/operator_bp.exp: break "and"
FAIL: gdb.ada/operator_bp.exp: break "or"
FAIL: gdb.ada/operator_bp.exp: break "xor"
FAIL: gdb.ada/operator_bp.exp: break "not"
...

The first FAIL is because two breakpoint locations are expected, but there are
more than 2:
...
(gdb) break "+"
Breakpoint 2 at 0x402c3c: "+". (6 locations)
(gdb) info break
Num     Type           Disp Enb Address            What
2       breakpoint     keep y   <MULTIPLE>
2.1                         y   0x0000000000402c3c in ops."+"
                                                   at operator_bp/ops.adb:25
2.2                         y   0x0000000000402e5b in ops."+"
                                                   at operator_bp/ops.adb:119
2.3                         y   0x0000000000414207 in
  system.storage_elements."+" at s-stoele.adb:82
2.4                         y   0x0000000000414404 in
  system.storage_elements."+" at s-stoele.adb:87
2.5                         y   0x0000000000414413 in
  system.storage_elements."+" at s-stoele.adb:87
2.6                         y   0x0000000000414430 in
  system.storage_elements."+" at s-stoele.adb:81
...

This can be traced back to a extra debug info in the executable:
...
$ readelf -w ops_test | grep system__storage_elements__Oadd
    <28104>   DW_AT_name        : system__storage_elements__Oadd__2
    <2812e>   DW_AT_name        : system__storage_elements__Oadd
...

Fix the FAILs by allowing more than the required amount of breakpoint
locations.

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2020-05-01  Tom de Vries  <tdevries@suse.de>

* gdb.ada/operator_bp.exp: Allow more than required amount of
breakpoint.

4 years agoRemove gdb-gdb.gdb breakpoint on disappeared function info_command.
Philippe Waroquiers [Fri, 1 May 2020 14:47:05 +0000 (16:47 +0200)]
Remove gdb-gdb.gdb breakpoint on disappeared function info_command.

The function info_command has disappeared, so this breakpoint does not
work anymore.
"info_command" was a function for the prefix command "info",
giving the list of "info" subcommands.
It is not very clear what the removed breakpoint and its associated
command list was supposed to do.

Removed and pushed as obvious, after discussion with Tom.

4 years agoFix size recalculation of fortran arrays
Hannes Domani [Fri, 1 May 2020 12:01:02 +0000 (14:01 +0200)]
Fix size recalculation of fortran arrays

My recent change regarding size calculation of arrays of stubbed types
didn't take array strides and associated/allocated type properties into
account, which basically broke fortran arrays.

Fixed by refactoring the array size calculation of
create_array_type_with_stride into a new function, and also use it for
the stubbed array size recalculation.

gdb/ChangeLog:

2020-05-01  Hannes Domani  <ssbssa@yahoo.de>

* gdbtypes.c (update_static_array_size): New function.
(create_array_type_with_stride): Use update_static_array_size.
(check_typedef): Likewise.

4 years ago[gdb/testsuite] Fix Wunused-result warning in until-reverse.c
Tom de Vries [Fri, 1 May 2020 09:04:22 +0000 (11:04 +0200)]
[gdb/testsuite] Fix Wunused-result warning in until-reverse.c

When running test-case gdb.reverse/until-reverse.exp or
gdb.reverse/until-precsave.exp with gcc-10, we run into a Wunused-result
warning:
...
gdb compile failed, gdb.reverse/until-reverse.c: In function 'main':
gdb.reverse/until-reverse.c:40:14: warning: ignoring return value of \
  'malloc' declared with attribute 'warn_unused_result' [-Wunused-result]
   40 |       (void) malloc (1);
      |              ^~~~~~~~~~
...

Fix this by using the result of malloc as argument to a free call.

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2020-05-01  Tom de Vries  <tdevries@suse.de>

* gdb.reverse/until-reverse.c (main): Fix Wunused-result warning.

4 years agoPR25900, RISC-V: null pointer dereference
Alan Modra [Fri, 1 May 2020 06:02:00 +0000 (15:32 +0930)]
PR25900, RISC-V: null pointer dereference

PR 25900
* elfnn-riscv.c (_bfd_riscv_relax_section): Check root.type before
accessing root.u.def of symbols.  Also check root.u.def.section
is non-NULL.  Reverse tests so as to make the logic positive.

4 years agoobjdump: long run time when using -D -z
Alan Modra [Fri, 1 May 2020 06:00:12 +0000 (15:30 +0930)]
objdump: long run time when using -D -z

The zero scan is silly when the result isn't used.

* objdump.c (disassemble_bytes): Don't scan for zeros when
disassembling zeros.  Translate "resuming at file offset" message.
Formatting.  Replace some signed variables with unsigned.

4 years agoPR25882, .gnu.attributes are not checked for shared libraries
Alan Modra [Fri, 1 May 2020 05:53:17 +0000 (15:23 +0930)]
PR25882, .gnu.attributes are not checked for shared libraries

This allows backend merge_private_bfd_data to examine shared library
e_flags and/or .gnu.attributes.  ARM and PowerPC have done so when
using ld.gold for a long time.

(The tic6x change below is dead code due to the earlier FIXME,
but this is probably one of the changes needed there.)

PR 25882
bfd/
* elf32-tic6x.c (elf32_tic6x_merge_attributes): Don't transfer
Tag_ABI_PIC or Tag_ABI_PID from dynamic objects to the output.
ld/
* ldlang.c (lang_check): Call bfd_merge_private_bfd_data for
shared libraries.

4 years agoFIXME for merging of e_flags and .gnu.attributes
Alan Modra [Fri, 1 May 2020 05:50:14 +0000 (15:20 +0930)]
FIXME for merging of e_flags and .gnu.attributes

Code in the linker, present before the addition of .gnu.attributes
support, results in shared libraries not being considered by BFD when
merging e_flags and .gnu.attributes from input files to the output.
That doesn't seem correct to me, but I don't know enough about all the
various ABIs to change the behaviour with any confidence.  So this
patch merely punts on dynamic objects in merge_private_bfd_data target
functions, with a FIXME for maintainer attention.

I haven't excluded shared libraries from being considered where the
target merge_private_bfd_data (a) already has code dealing with shared
libraries, or (b) where that function just sets the output to the most
constraining arch/mach combination and other fairly trivial merges, or
(c) when the target has no shared library linker support.

In (a) are: arc, arm, aarch64, riscv, sparc.
In (b) are: bpf, cris, csky, m32r, m68k, mn10300, nios2, tilegx,
    tilepro, vax, visium, xtensa.
In (c) are: bpf, cr16, h8300, iq2000, m32c, m68hc11, m68hc12, mcore,
    mep, msp430, mt, rl78, rx, v850.

PR 25882
* elf32-bfin.c (elf32_bfin_merge_private_bfd_data): Add FIXME.
* elf32-frv.c (frv_elf_merge_private_bfd_data): Likewise.
* elfxx-mips.c (_bfd_mips_elf_merge_private_bfd_data): Likewise.
* elf32-nds32.c (nds32_elf_merge_private_bfd_data): Likewise.
* elf32-score.c (s3_elf32_score_merge_private_bfd_data): Likewise.
* elf32-score7.c (s7_elf32_score_merge_private_bfd_data): Likewise.
* elf32-sh.c (sh_elf_merge_private_data): Likewise.
* elf32-tic6x.c (elf32_tic6x_merge_attributes): Likewise.
* elf64-ia64-vms.c (elf64_ia64_merge_private_bfd_data): Likewise.
* elfnn-ia64.c (elfNN_ia64_merge_private_bfd_data): Likewise.

4 years agoppc32 merging of e_flags from dynamic objects
Alan Modra [Fri, 1 May 2020 05:47:42 +0000 (15:17 +0930)]
ppc32 merging of e_flags from dynamic objects

EF_PPC_RELOCATABLE and similar flags, if present in an input shared
library, don't have any relevance as far as the output file is
concerned.

Currently, dynamic objects aren't seen in merge_private_bfd_data.
This patch is in preparation for a change to that.

PR 25882
* elf32-ppc.c (ppc_elf_merge_private_bfd_data): Ignore e_flags
from shared libraries.

4 years agoRegen ld BLD-POTFILES.in
Alan Modra [Fri, 1 May 2020 02:19:25 +0000 (11:49 +0930)]
Regen ld BLD-POTFILES.in

* po/BLD-POTFILES.in: Regenerate.

4 years agoAutomatic date update in version.in
GDB Administrator [Fri, 1 May 2020 00:00:16 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoAdd support for NetBSD thread events (create, exit)
Kamil Rytarowski [Thu, 30 Apr 2020 14:13:33 +0000 (16:13 +0200)]
Add support for NetBSD thread events (create, exit)

Report LWP CREATE and LWP EXIT events and setup this on post_attach()
and post_startup_inferior().

Stop reinitializing the list of recognized threads in update_thread_list().

Handle LWP CREATE and EXIT events in nbsd_nat_target::wait().

gdb/ChangeLog:

        * nbsd-nat.c (nbsd_enable_proc_events)
        (nbsd_nat_target::post_startup_inferior): Add.
        (nbsd_nat_target::post_attach): Call `nbsd_enable_proc_events'.
        (nbsd_nat_target::update_thread_list): Rewrite.
        (nbsd_nat_target::wait): Handle "PTRACE_LWP_EXIT" and
        "PTRACE_LWP_CREATE".
        * nbsd-nat.h (nbsd_nat_target::post_startup_inferior): Add.

4 years agoRevert "2020-04-29 Sterling Augustine <saugustine@google.com>"
Tom de Vries [Thu, 30 Apr 2020 16:51:49 +0000 (18:51 +0200)]
Revert "2020-04-29  Sterling Augustine <saugustine@google.com>"

This reverts commit 84ed7a472551bce1ac58e0eced619433fabc956c.

The problem that the commit attempts to address has already been fixed in
commit 770479f223e "gdb: Fix toplevel types with -fdebug-types-section".

The commit itself is superfluous because it sets list_in_scope at a point that
it's already set (by start_symtab).

4 years agoRemove duplicated creation of "frame" command and "f" alias.
Philippe Waroquiers [Thu, 30 Apr 2020 16:35:33 +0000 (18:35 +0200)]
Remove duplicated creation of "frame" command and "f" alias.

"frame" and "f" are created twice by stack.c _initialize_stack.
Remove the second creation.
Regression tested on amd64/Debian.

2020-04-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>
* stack.c (_initialize_stack): Remove duplicated creation
of "frame" command and "f" alias.

4 years agoImplement debugging of WOW64 processes in gdbserver
Hannes Domani [Fri, 24 Apr 2020 15:23:59 +0000 (17:23 +0200)]
Implement debugging of WOW64 processes in gdbserver

gdbserver/ChangeLog:

2020-04-30  Hannes Domani  <ssbssa@yahoo.de>

* configure.srv <x86_64-*-mingw*, x86_64-*-cygwin*> (srv_tgtobj):
Add arch/i386.o.
* win32-arm-low.cc (arm_num_regs): New function.
(struct win32_target_ops): Use arm_num_regs.
* win32-i386-low.cc (win32_get_current_dr): Adapt for WOW64
processes.
(i386_get_thread_context): Likewise.
(i386_prepare_to_resume): Likewise.
(i386_thread_added): Likewise.
(i386_single_step): Likewise.
(i386_fetch_inferior_register): Likewise.
(i386_store_inferior_register): Likewise.
(i386_arch_setup): Likewise.
(i386_win32_num_regs): New function.
(struct win32_target_ops): Use i386_win32_num_regs.
* win32-low.cc (win32_get_thread_context): Adapt for WOW64
processes.
(win32_require_context): Likewise.
(child_add_thread): Likewise.
(do_initial_child_stuff): Likewise.
(continue_one_thread): Likewise.
(win32_process_target::resume): Likewise.
(load_psapi): Likewise.
(win32_add_all_dlls): Likewise.
(maybe_adjust_pc): Likewise.
(win32_process_target::qxfer_siginfo): Likewise.
(initialize_low): Likewise.
* win32-low.h (struct win32_target_ops): Change num_regs to
callback function.

4 years agoCalculate size of array of stubbed type
Hannes Domani [Mon, 27 Apr 2020 11:14:24 +0000 (13:14 +0200)]
Calculate size of array of stubbed type

Sizes of stubbed types are calculated on demand in check_typedef, so the
same must also be done for arrays of stubbed types.

A stubbed type is usually a structure that has only been forward declared,
but can also happen if the structure has a virtual function that's not
inline in the class definition.

For these stubbed types, the size must be recalculated once the full
definition is available.

gdb/ChangeLog:

2020-04-30  Hannes Domani  <ssbssa@yahoo.de>

PR gdb/18706
* gdbtypes.c (check_typedef): Calculate size of array of
stubbed type.

gdb/testsuite/ChangeLog:

2020-04-30  Hannes Domani  <ssbssa@yahoo.de>

PR gdb/18706
* gdb.cp/stub-array-size.cc: New test.
* gdb.cp/stub-array-size.exp: New file.
* gdb.cp/stub-array-size.h: New test.
* gdb.cp/stub-array-size2.cc: New test.

4 years agoAdjust array pretty printer tests to the new format
Hannes Domani [Thu, 30 Apr 2020 11:17:30 +0000 (13:17 +0200)]
Adjust array pretty printer tests to the new format

gdb/testsuite/ChangeLog:

2020-04-30  Hannes Domani  <ssbssa@yahoo.de>

* gdb.python/py-format-string.exp: Adjust pretty_arrays expected
output to the new format.

4 years agoAArch64: add GAS support for UDF instruction
Alex Coplan [Thu, 30 Apr 2020 14:47:30 +0000 (15:47 +0100)]
AArch64: add GAS support for UDF instruction

binutils * testsuite/binutils-all/aarch64/in-order-all.d: Update to use new
          disassembly.
        * testsuite/binutils-all/aarch64/out-of-order-all.d: Likewise.

ld/     * testsuite/ld-aarch64/erratum843419_tls_ie.d: Use udf in disassembly.
        * testsuite/ld-aarch64/farcall-b-section.d: Likewise.
        * testsuite/ld-aarch64/farcall-back.d: Likewise.
        * testsuite/ld-aarch64/farcall-bl-section.d: Likewise.

gas/   * config/tc-aarch64.c (fix_insn): Implement for AARCH64_OPND_UNDEFINED.
          (parse_operands): Implement for AARCH64_OPND_UNDEFINED.
        * testsuite/gas/aarch64/udf.s: New.
        * testsuite/gas/aarch64/udf.d: New.
        * testsuite/gas/aarch64/udf-invalid.s: New.
        * testsuite/gas/aarch64/udf-invalid.l: New.
        * testsuite/gas/aarch64/udf-invalid.d: New.

include * opcode/aarch64.h (enum aarch64_opnd): Add AARCH64_OPND_UNDEFINED.

opcodes * aarch64-opc.h (enum aarch64_field_kind): Add FLD_imm16_2.
        * aarch64-opc.c (fields): Add entry for FLD_imm16_2.
          (operand_general_constraint_met_p): validate AARCH64_OPND_UNDEFINED.
        * aarch64-tbl.h (aarch64_opcode_table): Add udf instruction, entry for
          FLD_imm16_2.
        * aarch64-asm-2.c: Regenerated.
        * aarch64-dis-2.c: Regenerated.
        * aarch64-opc-2.c: Regenerated.

4 years agoAdd generic tests for linker support of the -shared or -pie options to tests which...
Nick Clifton [Thu, 30 Apr 2020 12:49:30 +0000 (13:49 +0100)]
Add generic tests for linker support of the -shared or -pie options to tests which use them.

binutils* testsuite/lib/binutils-common.exp (check_pie_support): New
proc.

ld * testsuite/ld-elf/compress1c.d: XFAIL if thet target linker does
not support the -shared command line option.
* /ld-elf/compressed1c.d: Likewise.
* /ld-elf/compressed1e.d: Likewise.
* /ld-elf/dynamic1.d: Likewise.
* /ld-elf/dynsym1.d: Likewise.
* /ld-elf/ehdr_start-shared.d: Likewise.
* /ld-elf/exclude3b.d: Likewise.
* /ld-elf/global1.d: Likewise.
* /ld-elf/hash.d: Likewise.
* /ld-elf/local1.d: Likewise.
* /ld-elf/mbind1b.d: Likewise.
* /ld-elf/now-1.d: Likewise.
* /ld-elf/now-2.d: Likewise.
* /ld-elf/now-3.d: Likewise.
* /ld-elf/now-4.d: Likewise.
* /ld-elf/pr12975.d: Likewise.
* /ld-elf/pr13177.d: Likewise.
* /ld-elf/pr13195.d: Likewise.
* /ld-elf/pr16322.d: Likewise.
* /ld-elf/pr16498a.d: Likewise.
* /ld-elf/pr16498b.d: Likewise.
* /ld-elf/pr17615.d: Likewise.
* /ld-elf/pr19162.d: Likewise.
* /ld-elf/pr19698.d: Likewise.
* /ld-elf/pr20513c.d: Likewise.
* /ld-elf/pr20513d.d: Likewise.
* /ld-elf/pr21389a.d: Likewise.
* /ld-elf/pr21389b.d: Likewise.
* /ld-elf/pr21389c.d: Likewise.
* /ld-elf/pr21562a.d: Likewise.
* /ld-elf/pr21562b.d: Likewise.
* /ld-elf/pr21562c.d: Likewise.
* /ld-elf/pr21562d.d: Likewise.
* /ld-elf/pr21562e.d: Likewise.
* /ld-elf/pr21562f.d: Likewise.
* /ld-elf/pr21562g.d: Likewise.
* /ld-elf/pr21562h.d: Likewise.
* /ld-elf/pr21562i.d: Likewise.
* /ld-elf/pr21562j.d: Likewise.
* /ld-elf/pr21562k.d: Likewise.
* /ld-elf/pr21562l.d: Likewise.
* /ld-elf/pr21562m.d: Likewise.
* /ld-elf/pr21562n.d: Likewise.
* /ld-elf/pr21903a.d: Likewise.
* /ld-elf/pr21903b.d: Likewise.
* /ld-elf/pr22269b.d: Likewise.
* /ld-elf/pr22393-1a.d: Likewise.
* /ld-elf/pr22393-1b.d: Likewise.
* /ld-elf/pr23658-1c.d: Likewise.
* /ld-elf/pr25708.d: Likewise.
* /ld-elf/rpath-1.d: Likewise.
* /ld-elf/rpath-2.d: Likewise.
* /ld-elf/runpath-1.d: Likewise.
* /ld-elf/runpath-2.d: Likewise.
* /ld-elf/sizeofb.d: Likewise.
* /ld-elf/startofb.d: Likewise.
* /ld-elf/strtab.d: Likewise.
* /ld-elf/textaddr2.d: Likewise.
* /ld-elf/textaddr5.d: Likewise.
* /ld-elf/textaddr6.d: Likewise.
* /ld-elf/unknown2.d: Likewise.
* /ld-undefined/entry-3.d: Likewise.
* /ld-undefined/entry-4.d: Likewise.
* /ld-elf/mbind1c.d: XFAIL if the target linker does not support
the -pie option.
* /ld-elf/pie.d: Likewise.
* /ld-elf/pr19539.d: Likewise.
* /ld-elf/pr21903d.d: Likewise.
* /ld-elf/pr22269a.d: Likewise.
* /ld-elf/pr22393-1c.d: Likewise.
* /ld-elf/pr22393-1d.d: Likewise.
* /ld-elf/pr22423.d: Likewise.
* /ld-elf/loadaddr1.d: Expect to fail on the rx-linux target.
* /ld-elf/loadaddr2.d: Likewise.
* /ld-elf/pr22393-1e.d: Likewise.
* /ld-elf/pr22393-1f.d: Likewise.
* /ld-elf/textaddr1.d: Likewise.
* /ld-elf/textaddr4.d: Likewise.

4 years agoUse thiscall calling convention for class members
Hannes Domani [Mon, 27 Apr 2020 13:58:09 +0000 (15:58 +0200)]
Use thiscall calling convention for class members

Non-static member functions for Windows 32bit programs need the thiscall
calling convention, so the 'this' pointer needs to be passed in ECX.

gdb/ChangeLog:

2020-04-30  Hannes Domani  <ssbssa@yahoo.de>

PR gdb/15559
* i386-tdep.c (i386_push_dummy_call): Call
i386_thiscall_push_dummy_call.
(i386_thiscall_push_dummy_call): New function.
* i386-tdep.h (i386_thiscall_push_dummy_call): Declare.
* i386-windows-tdep.c (i386_windows_push_dummy_call): New function.
(i386_windows_init_abi): Call set_gdbarch_push_dummy_call.

4 years agold: Add rx-linux emulation. gas: Change ELF flags initial value in rx-linux
Yoshinori Sato [Thu, 30 Apr 2020 12:35:37 +0000 (13:35 +0100)]
ld: Add rx-linux emulation.  gas: Change ELF flags initial value in rx-linux

ld * emulparams/elf32rx_linux.sh: New rx-linux emulation.
* emultempl/rxlinux.em: New.
* configure.tgt: Add rx-linux.
* Makefile.am: Add eelf32rx_linux.c
* Makefile.in: Regenerate.

gas * config/tc-rx.c (elf_flags): Reset default value.
(md_parse_option): For rx-elf Initialize elf_flags with RX_ABI.

4 years agoxtensa: fix XTENSA_NDIFF handling for PR ld/25861
Max Filippov [Sat, 25 Apr 2020 07:40:25 +0000 (00:40 -0700)]
xtensa: fix XTENSA_NDIFF handling for PR ld/25861

Fields marked with XTENSA_NDIFF relocations are not negated, they only
have sign bits removed. Don't negate their values when relaxation is
performed. Don't add sign bits when the value is zero. Report overflow
when the result has negative sign but all significant bits are zero.

2020-04-29  Max Filippov  <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (relax_section): Don't negate diff_value for
XTENSA_NDIFF relocations. Don't add sign bits whe diff_value
equals 0. Report overflow when the result has negative sign but
all significant bits are zero.

ld/
* testsuite/ld-xtensa/relax-diff1.d: New test definition.
* testsuite/ld-xtensa/relax-diff1.s: New test source.
* testsuite/ld-xtensa/relax-ndiff.d: New test definition.
* testsuite/ld-xtensa/relax-ndiff.s: New test source.
* testsuite/ld-xtensa/xtensa.exp: (relax-diff1)
(relax-ndiff): New tests.

4 years agoxtensa: gas: support optional immediate simcall parameter
Max Filippov [Tue, 28 Apr 2020 11:15:05 +0000 (04:15 -0700)]
xtensa: gas: support optional immediate simcall parameter

Starting with RH.0 release Xtensa ISA adds immediate parameter to
simcall opcode. For assembly source compatibility treat "simcall"
instruction without parameter as "simcall 0" when parameter is required.

2020-04-29  Max Filippov  <jcmvbkbc@gmail.com>
gas/
* config/tc-xtensa.c (XTENSA_MARCH_EARLIEST): Define macro as 0
if it's not defined.
(microarch_earliest): New static variable.
(xg_translate_idioms): Translate "simcall" to "simcall 0" when
simcall opcode has mandatory parameter.
(xg_init_global_config): Initialize microarch_earliest.

4 years agogdb: silence shellcheck warning SC2162 (use read -r) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:35 +0000 (20:35 -0400)]
gdb: silence shellcheck warning SC2162 (use read -r) in gdbarch.sh

shellcheck reports:

    In gdbarch.sh line 53:
        while IFS='' read line
                     ^--^ SC2162: read without -r will mangle backslashes.

See the rationale at [1].  In our case, we actually want the backslashes
to be interpreted and removed.  Silence the warning using a directive.

[1] https://github.com/koalaman/shellcheck/wiki/SC2162

gdb/ChangeLog:

* gdbarch.sh (do_read): Add shellcheck disable directive for
warning SC2162.

4 years agogdb: fix shellcheck warnings SC2154 (referenced but not assigned) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:35 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2154 (referenced but not assigned) in gdbarch.sh

Fix all instances of this kind of warning:

    In gdbarch.sh line 96:
                    m ) staticdefault="${predefault}" ;;
                                       ^-----------^ SC2154: predefault is referenced but not assigned.

These warnings appear because we are doing something a bit funky when reading
the gdbarch fields.  These variables are not assigned explicitly, but
using some `eval` commands.

I don't think there is so much we can fix about those warnings.  To
silence them, I've changed `${foo}` to `${foo:-}`.  This tells the shell
to substitute with an empty string if `foo` is not defined.  This
retains the current behavior, but the warnings go away.

gdb/ChangeLog:

* gdbarch.sh: Use ${foo:-} where shellcheck would report a
"referenced but not assigned" warning.

4 years agogdb: fix shellcheck warnings SC2034 (unused variable) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:35 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2034 (unused variable) in gdbarch.sh

shellcheck reports:

    In gdbarch.sh line 139:
                    fallbackdefault="0"
                    ^-------------^ SC2034: fallbackdefault appears unused. Verify use (or export if used externally).

Indeed, the `fallbackdefault` variable appears to be unused, remove the
code that sets it.

gdb/ChangeLog:

* gdbarch.sh: Remove code that sets fallbackdefault.

4 years agogdb: fix shellcheck warnings SC2166 (&& and !! instead of -a and -o) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:34 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2166 (&& and !! instead of -a and -o) in gdbarch.sh

Fix all warnings of this type:

    In gdbarch.sh line 1238:
        if [ "x${invalid_p}" = "x0" -a -n "${postdefault}" ]
                                    ^-- SC2166: Prefer [ p ] && [ q ] as [ p -a q ] is not well defined.

See the rationale here [1].

[1] https://github.com/koalaman/shellcheck/wiki/SC2166

gdb/ChangeLog:

* gdbarch.sh: Use shell operators && and || instead of
-a and -o.

4 years agogdb: fix shellcheck warnings SC2006 (use $() instead of ``) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:34 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2006 (use $() instead of ``) in gdbarch.sh

Fix all instances of:

    In gdbarch.sh line 2195:
            printf "            `echo "$function" | sed -e 's/./ /g'`  %s %s)\n" "$returntype" "$function"
                                ^-- SC2006: Use $(...) notation instead of legacy backticked `...`.

    Did you mean:
            printf "            $(echo "$function" | sed -e 's/./ /g')  %s %s)\n" "$returntype" "$function"

See here [1] for the rationale.

[1] https://github.com/koalaman/shellcheck/wiki/SC2006

gdb/ChangeLog:

* gdbarch.sh: Use $(...) instead of `...`.

4 years agogdb: fix shellcheck warnings SC2086 (missing double quotes) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:34 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2086 (missing double quotes) in gdbarch.sh

Fix all instances of:

    In gdbarch.sh line 31:
        if test ! -r ${file}
                     ^-----^ SC2086: Double quote to prevent globbing and word splitting.

    Did you mean:
        if test ! -r "${file}"

Note that some instances of these are in text that is eval'ed.  I'm
pretty sure that things could go wrong during the eval too, but that's
not something shellcheck can check.

gdb/ChangeLog:

* gdbarch.sh: Use double quotes around variables.

4 years agogdb: fix shellcheck warnings SC2059 (variables in printf format string) in gdbarch.sh
Simon Marchi [Thu, 30 Apr 2020 00:35:33 +0000 (20:35 -0400)]
gdb: fix shellcheck warnings SC2059 (variables in printf format string) in gdbarch.sh

Fix all instances of this:

    In gdbarch.sh line 2182:
                printf "  gdb_assert (!(${invalid_p}));\n"
                       ^-- SC2059: Don't use variables in the printf format string. Use printf "..%s.." "$foo".

... by doing exactly as the message suggests.

The rationale explained here [1] makes sense, if there happens to be a
format specifier in text substituted for the variable, the printf won't
do what we expect.

[1] https://github.com/koalaman/shellcheck/wiki/SC2059

gdb/ChangeLog:

* gdbarch.sh: Use %s with printf, instead of variables in the
format string.

4 years ago2020-04-29 Sterling Augustine <saugustine@google.com>
Sterling Augustine [Tue, 28 Apr 2020 18:17:50 +0000 (11:17 -0700)]
2020-04-29  Sterling Augustine <saugustine@google.com>

       * dwarf2/read.c (setup_type_unit_groups): Set list_in_scope.

4 years agoAutomatic date update in version.in
GDB Administrator [Thu, 30 Apr 2020 00:00:11 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoSync config and libiberty with GCC
H.J. Lu [Wed, 29 Apr 2020 21:25:36 +0000 (14:25 -0700)]
Sync config and libiberty with GCC

config/

2020-04-29  H.J. Lu  <hongjiu.lu@intel.com>

Sync with GCC
2020-04-28  H.J. Lu  <hongjiu.lu@intel.com>

PR bootstrap/94739
* cet.m4 (GCC_CET_HOST_FLAGS): Add -fcf-protection=none to
-Wl,-z,ibt,-z,shstk.  Check whether -fcf-protection=none
-Wl,-z,ibt,-z,shstk works first.

2020-04-25  H.J. Lu  <hongjiu.lu@intel.com>

PR bootstrap/94739
* cet.m4 (GCC_CET_HOST_FLAGS): New.

2020-04-22  Jakub Jelinek  <jakub@redhat.com>

PR libfortran/94694
PR libfortran/94586
* math.m4 (GCC_CHECK_MATH_INLINE_BUILTIN_FALLBACK1,
GCC_CHECK_MATH_INLINE_BUILTIN_FALLBACK2): New.

libiberty/

2020-04-28  H.J. Lu  <hongjiu.lu@intel.com>

PR bootstrap/94739
* configure: Regenerated.

4 years agogdb: fix duplicate test names in gdb.base/break.exp
Simon Marchi [Wed, 29 Apr 2020 20:34:54 +0000 (16:34 -0400)]
gdb: fix duplicate test names in gdb.base/break.exp

These test names are duplicated:

      2 PASS: gdb.base/break.exp: set breakpoint via non-integer convenience variable disallowed
      2 PASS: gdb.base/break.exp: set convenience variable $foo to 81.5

Wrap them with `with_test_prefix`. I've actually wrapped a bit more
tests that are related, I think it helps to give the test names a bit
more context.  The modified test names are:

    -PASS: gdb.base/break.exp: set convenience variable $foo to bp_location11
    -PASS: gdb.base/break.exp: set breakpoint via convenience variable
    -PASS: gdb.base/break.exp: set convenience variable $foo to 81.5
    -PASS: gdb.base/break.exp: set breakpoint via non-integer convenience variable disallowed
    +PASS: gdb.base/break.exp: set line breakpoint via convenience variable: set convenience variable $foo to bp_location11
    +PASS: gdb.base/break.exp: set line breakpoint via convenience variable: break $foo
    +PASS: gdb.base/break.exp: set line breakpoint via convenience variable: set convenience variable $foo to 81.5
    +PASS: gdb.base/break.exp: set line breakpoint via convenience variable: non-integer convenience variable disallowed

    -PASS: gdb.base/break.exp: set $l = 47
    -PASS: gdb.base/break.exp: break break.c:$l
    -PASS: gdb.base/break.exp: set convenience variable $foo to 81.5
    -PASS: gdb.base/break.exp: set breakpoint via non-integer convenience variable disallowed
    +PASS: gdb.base/break.exp: set line:file breakpoint via convenience variable: set $l = 47
    +PASS: gdb.base/break.exp: set line:file breakpoint via convenience variable: break break.c:$l
    +PASS: gdb.base/break.exp: set line:file breakpoint via convenience variable: set convenience variable $foo to 81.5
    +PASS: gdb.base/break.exp: set line:file breakpoint via convenience variable: non-integer convenience variable disallowed

gdb/testsuite/ChangeLog:

* gdb.base/break.exp: Use with_test_prefix.

4 years agoFix Ada crash with .debug_types
Tom Tromey [Wed, 29 Apr 2020 17:16:57 +0000 (11:16 -0600)]
Fix Ada crash with .debug_types

PR ada/25875 concerns a gdb crash when gdb.ada/arr_enum_idx_w_gap.exp
is run using the .debug_types board.

The problem turns out to be caused by weird compiler output.  In this
test, the compiler emits a top-level type that refers to an
enumeration type which is nested in a function.  However, this
function is just a declaration.

This results in gdb calling read_enumeration_type for the enum type,
but process_enumeration_scope is never called, yielding an enum with
no fields.  This causes the crash.

This patch fixes the problem by arranging to create the enum fields in
read_enumeration_type.

Tested on x86-64 Fedora 30.

gdb/ChangeLog
2020-04-29  Tom Tromey  <tromey@adacore.com>

PR ada/25875:
* dwarf2/read.c (update_enumeration_type_from_children): Compute
type fields here.
(read_enumeration_type): Call
update_enumeration_type_from_children later.  Update comments.
(process_enumeration_scope): Don't create type fields.

4 years agoSet NetBSD xml syscall file name to syscalls/netbsd.xml
Kamil Rytarowski [Wed, 29 Apr 2020 13:33:33 +0000 (15:33 +0200)]
Set NetBSD xml syscall file name to syscalls/netbsd.xml

The syscall literal names are not stable on NetBSD and can change
once a syscall is versioned. Thus these names are internal to the
system and in GDB mostly descriptive, not intended to be a stable
interface with fixed names across GDB and NetBSD versions to track
certain syscalls.

gdb/ChangeLog:

* nbsd-tdep.c: Include "xml-syscall.h".
(nbsd_init_abi): Call `set_xml_syscall_file_name'.

4 years agoAdd basic event handling in the NetBSD target
Kamil Rytarowski [Tue, 28 Apr 2020 23:57:38 +0000 (01:57 +0200)]
Add basic event handling in the NetBSD target

Implement the following events:
 - single step (TRAP_TRACE)
 - software breakpoint (TRAP_DBREG)
 - exec() (TRAP_EXEC)
 - syscall entry/exit (TRAP_SCE / TRAP_SCX)

Add support for NetBSD specific ::wait () and ::resume ().

Instruct the generic code that exec and syscall events are supported.

Define an empty nbsd_get_syscall_number as it is prerequisite for
catching syscall entry and exit events, even if it is unused.
This function is used to detect whether the gdbarch supports the
'catch syscall' feature.

gdb/ChangeLog:

       * nbsd-nat.c: Include "sys/wait.h".
       (nbsd_resume, nbsd_nat_target::resume, nbsd_wait)
       (nbsd_nat_target::wait, nbsd_nat_target::insert_exec_catchpoint)
       (nbsd_nat_target::remove_exec_catchpoint)
       (nbsd_nat_target::set_syscall_catchpoint): Add.
       * nbsd-nat.h (nbsd_nat_target::resume, nbsd_nat_target::wait)
       (nbsd_nat_target::insert_exec_catchpoint)
       (nbsd_nat_target::remove_exec_catchpoint)
       (nbsd_nat_target::set_syscall_catchpoint): Add.
       * nbsd-tdep.c (nbsd_get_syscall_number): Add.
       (nbsd_init_abi): Call `set_gdbarch_get_syscall_number' and pass
       `nbsd_get_syscall_number'.

4 years agoWhen displaying ranges, ignore duplicate offsets.
Andrew Burgess [Wed, 29 Apr 2020 16:21:10 +0000 (17:21 +0100)]
When displaying ranges, ignore duplicate offsets.

* dwarf.c (display_debug_ranges): Ignore duplicate entries in
range_entries for the same offset.

4 years agoUpdate expected disassembly after recent update.
Nick Clifton [Wed, 29 Apr 2020 16:18:56 +0000 (17:18 +0100)]
Update expected disassembly after recent update.

PR 22699
* testsuite/gas/sh/sh4al-dsp.d: Update expected disassembly.

4 years agoAlso use unsigned 8-bit immediate values for the LDRC and SETRC insns.
Nick Clifton [Wed, 29 Apr 2020 15:09:38 +0000 (16:09 +0100)]
Also use unsigned 8-bit immediate values for the LDRC and SETRC insns.

PR 22699
* sh-opc.h: Also use unsigned 8-bit immediate values for the LDRC
and SETRC insns.

4 years agoAdd a warning if an emtpty SHT_REL, SHT_RELA or SHT_PROGBITS section is detected...
Nick Clifton [Wed, 29 Apr 2020 15:01:40 +0000 (16:01 +0100)]
Add a warning if an emtpty SHT_REL, SHT_RELA or SHT_PROGBITS section is detected.  Disable all warnings unless the (new) lint mode is enabled.

* readelf.c (warn): New function - like elfcomm.c version but only
produces output if warnings are enabled.
(struct options): Add --lint and --enable-checks.
(usage): Add entry for --lint.
(parse_args): Handle -L.  If checks are enabled but no dumps have
been selected then enable all dumps.
(process_section_headers): Replace long if-then-else sequence with
a switch.  Add warning messages for empty SHT_REL, SHT_RELA and
SHT_PROGBITS sections.
(process_file): Do not complain if the file is an archive and lint
mode has been enabled.
* elfcomm.c (error): Make the function weak.
(warn): Likewise.
* NEWS: Mention the new feature.
* doc/binutils.texi: Document the new feature.
* dwarf.h (report_leb_status): Add file name and line number
parameters.  Include them in the diagnostic output.
(READ_ULEB): Pass file and line number to report_leb_status.
(READ_SLEB): Likewise.
* dwarf.c (read_and_print_leb128): Pass file and line number to
report_leb_status.
* testsuite/binutils-all/readelf.exp: Add test of new feature.
* testsuite/binutils-all/zero-sec.s: New test source file.
* testsuite/binutils-all/zero-sec.r: Expected output from new
test.

4 years agoRemove some dead code
Tom Tromey [Wed, 29 Apr 2020 14:10:28 +0000 (08:10 -0600)]
Remove some dead code

print_block_frame_labels has been commented out since 2010.
I don't think we need it; this patch removes it.

2020-04-29  Tom Tromey  <tom@tromey.com>

* stack.c (print_block_frame_labels): Remove.

4 years agobfd: Fix 64-bit relocation handling for a.out
Gunther Nikl [Wed, 29 Apr 2020 13:42:41 +0000 (14:42 +0100)]
bfd: Fix 64-bit relocation handling for a.out

* aoutx.h (swap_std_reloc_out): Special case 64 bit relocations.
(aout_link_reloc_link_order): Likewise. Make r_length an unsigned.

4 years agoUpdated Serbian translation for the binutils sub-directory, and Swedish translation...
Nick Clifton [Wed, 29 Apr 2020 12:23:32 +0000 (13:23 +0100)]
Updated Serbian translation for the binutils sub-directory, and Swedish translation for the opcodes sub-directory.

4 years agoFix the disassmbly of SH instructions which have an unsigned 8-bit immediate operand.
Nick Clifton [Wed, 29 Apr 2020 12:13:55 +0000 (13:13 +0100)]
Fix the disassmbly of SH instructions which have an unsigned 8-bit immediate operand.

PR 22699
opcodes * sh-opc.h (IMM0_8): Replace with IMM0_8S and IMM0_8U.  Use
IMM0_8S for arithmetic insns and IMM0_8U for logical insns.
* sh-dis.c (print_insn_sh): Change IMM0_8 case to IMM0_8S and add
IMM0_8U case.

gas * config/tc-sh.c (build_Mytes): Change operand type IMM0_8 to
IMM0_8S and add support for IMM0_8U.
* testsuite/gas/sh/sh4a.s: Add test of a logical insn using an
unsigned 8-bit immediate.
* testsuite/gas/sh/sh4a.d: Extended expected disassembly.

4 years agoRemove restriction on skipping some linker tests because their default image base...
Stephen Casner [Wed, 29 Apr 2020 11:24:37 +0000 (12:24 +0100)]
Remove restriction on skipping some linker tests because their default image base is too high.

PR 25829
* testsuite/ld-scripts/default-script.exp: Add --image-base=0 to
LDFLAGS for targets *-*-mingw64 x86_64-*-cygwin.
* testsuite/ld-scripts/default-script1.d: No longer have to skip
test for those targets.
* testsuite/ld-scripts/default-script2.d: Likewise.
* testsuite/ld-scripts/default-script3.d: Likewise.
* testsuite/ld-scripts/default-script4.d: Likewise.

4 years ago[gdb/testsuite] Add xfails for PR gcc/90232
Tom de Vries [Wed, 29 Apr 2020 11:22:21 +0000 (13:22 +0200)]
[gdb/testsuite] Add xfails for PR gcc/90232

With target board debug-types, we have these FAILs:
...
FAIL: gdb.guile/scm-symtab.exp: test simple_struct in static symbols
FAIL: gdb.python/py-symtab.exp: test simple_struct in static symbols
...
due to PR gcc/90232, as explained in commit 15cd93d05e8 "[gdb/symtab] Handle
struct decl with DW_AT_signature".

Marks these as XFAILs.

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2020-04-29  Tom de Vries  <tdevries@suse.de>

* lib/gdb.exp (debug_types): New proc.
* gdb.guile/scm-symtab.exp: Add xfail for PR gcc/90232.
* gdb.python/py-symtab.exp: Same.

4 years agoFix array pretty formatter
Hannes Domani [Sun, 26 Apr 2020 13:28:46 +0000 (15:28 +0200)]
Fix array pretty formatter

Currently, printing with array pretty formatting makes the output actually
less readable than without:

(gdb) p -array on -- {{1,2,3},{4,5,6}}
$1 =   {    {1,
    2,
    3},
      {4,
    5,
    6}}
(gdb) p -array on -array-indexes on -- {{1,2,3},{4,5,6}}
$2 =   {[0] =     {[0] = 1,
    [1] = 2,
    [2] = 3},
  [1] =     {[0] = 4,
    [1] = 5,
    [2] = 6}}

These changes now also put the first element and the array end bracket on a new
line, similar to the structure pretty formatter:

(gdb) p -array on -- {{1,2,3},{4,5,6}}
$1 = {
  {
    1,
    2,
    3
  },
  {
    4,
    5,
    6
  }
}
(gdb) p -array on -array-indexes on -- {{1,2,3},{4,5,6}}
$2 = {
  [0] = {
    [0] = 1,
    [1] = 2,
    [2] = 3
  },
  [1] = {
    [0] = 4,
    [1] = 5,
    [2] = 6
  }
}

gdb/ChangeLog:

2020-04-29  Hannes Domani  <ssbssa@yahoo.de>

PR gdb/17320
* ada-valprint.c (val_print_packed_array_elements): Move array
end bracket to new line.
(ada_val_print_string): Remove extra spaces before first array
element.
* c-valprint.c (c_value_print_array): Likewise.
* m2-valprint.c (m2_print_array_contents): Likewise.
(m2_value_print_inner): Likewise.
* p-valprint.c (pascal_value_print_inner): Likewise.
* valprint.c (generic_val_print_array): Likewise.
(value_print_array_elements): Move first array element and array
end bracket to new line.

gdb/testsuite/ChangeLog:

2020-04-29  Hannes Domani  <ssbssa@yahoo.de>

PR gdb/17320
* gdb.base/pretty-array.c: New test.
* gdb.base/pretty-array.exp: New file.

4 years ago[gdb] Fix range loop index in find_method
Tom de Vries [Wed, 29 Apr 2020 09:39:36 +0000 (11:39 +0200)]
[gdb] Fix range loop index in find_method

With target board debug-types, we have:
...
FAIL: gdb.cp/cpexprs.exp: list policy1::function
...

This is a regression triggered by commit 770479f223e "gdb: Fix toplevel types
with -fdebug-types-section".

However, the FAIL is caused by commit 4dedf84da98 "Change
decode_compound_collector to use std::vector" which changes a VEC_iterate loop
into a range loop:
...
-  for (ix = 0; VEC_iterate (symbolp, sym_classes, ix, sym); ++ix)
+  unsigned int ix = 0;
+  for (const auto &sym : *sym_classes)
...
but fails to ensure that the increment of ix happens every iteration.

Fix this by calculating the index variable at the start of the loop body:
...
  for (const auto &elt : *sym_classes)
    {
      unsigned int ix = &elt - &*sym_classes->begin ();
...

Tested on x86_64-linux, with native and target board debug-types.

gdb/ChangeLog:

2020-04-29  Tom de Vries  <tdevries@suse.de>

PR symtab/25889
* linespec.c (find_method): Fix ix calculation.

gdb/testsuite/ChangeLog:

2020-04-29  Tom de Vries  <tdevries@suse.de>

PR symtab/25889
* gdb.cp/cpexprs.exp: Adapt for inclusion.
* gdb.cp/cpexprs-debug-types.exp: New file.  Set -fdebug-types-section
and include cpexprs.exp.

4 years agoAutomatic date update in version.in
GDB Administrator [Wed, 29 Apr 2020 00:00:12 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoAdd definitions of system calls to catch in native NetBSD targets
Kamil Rytarowski [Tue, 28 Apr 2020 23:46:41 +0000 (01:46 +0200)]
Add definitions of system calls to catch in native NetBSD targets

All platforms on NetBSD use a shared system call table, so use a
single XML file to describe the system calls available on each NetBSD
platform.

gdb/ChangeLog:

       * syscalls/update-netbsd.sh: New file.
       * syscalls/netbsd.xml: Regenerate.
       * data-directory/Makefile.in: Register `netbsd.xml' in
       `SYSCALLS_FILES'

4 years agogdb: fix shellcheck warning in update-freebsd.sh
Simon Marchi [Tue, 28 Apr 2020 18:29:39 +0000 (14:29 -0400)]
gdb: fix shellcheck warning in update-freebsd.sh

shellcheck reports:

    In update-freebsd.sh line 72:
    }' $1 >> freebsd.xml.tmp
       ^-- SC2086: Double quote to prevent globbing and word splitting.

    Did you mean:
    }' "$1" >> freebsd.xml.tmp

    For more information:
      https://www.shellcheck.net/wiki/SC2086 -- Double quote to prevent globbing ...

Add double quotes to fix it.

gdb/ChangeLog:

* syscalls/update-freebsd.sh: Add double quotes.

4 years agoAllow Python commands to be in class_tui
Tom Tromey [Tue, 28 Apr 2020 14:54:17 +0000 (08:54 -0600)]
Allow Python commands to be in class_tui

Now that Python code can create TUI windows, it seemed appropriate to
allow Python commands to appear in the "TUI" help class.  This patch
adds this capability.

gdb/ChangeLog
2020-04-28  Tom Tromey  <tom@tromey.com>

* NEWS: Update.
* python/py-cmd.c (gdbpy_initialize_commands): Add COMMAND_TUI.
(cmdpy_init): Allow class_tui.

gdb/doc/ChangeLog
2020-04-28  Tom Tromey  <tom@tromey.com>

* python.texi (Commands In Python): Document gdb.COMMAND_TUI.

4 years agoAdd missing ChangeLog entries
Tom de Vries [Tue, 28 Apr 2020 14:34:38 +0000 (16:34 +0200)]
Add missing ChangeLog entries

4 years agogdb: Fix toplevel types with -fdebug-types-section
Mark Williams [Tue, 28 Apr 2020 14:12:45 +0000 (16:12 +0200)]
gdb: Fix toplevel types with -fdebug-types-section

When debugging a program compiled with -fdebug-types-section,
only the first top-level type in each file is visible to gdb.

The problem was caused by moving the assignment to list_in_scope
from process_full_comp_unit and process_full_type_unit to
start_symtab.  This was fine for process_full_comp_unit, because
symtabs and comp units are one-to-one.  But there can be many type
units per symtab (one for each type), and we only call start_symtab
for the first one.  This adds the necessary assignments on the paths
where start_symtab is not called.

gdb/Changelog:

2020-04-28 Mark Williams <mark@myosotissp.com>

PR gdb/24480
* dwarf2read.c: Add missing assingments to list_in_scope when
start_symtab was already called.

gdb/testsuite/Changelog:

2020-04-28 Mark Williams <mark@myosotissp.com>

PR gdb/24480
* dw4-toplevel-types.exp: Test for top level types.
* dw4-toplevel-types.cc: Test for top level types.

4 years agogdb: use gdb:hash_enum as hash function in offset_map_type
Simon Marchi [Tue, 28 Apr 2020 13:49:58 +0000 (09:49 -0400)]
gdb: use gdb:hash_enum as hash function in offset_map_type

When building with g++ 4.8, we get this error (just an excerpt, because
g++ outputs a very long error message):

      CXX    dwarf2/read.o
    ...
    /home/smarchi/src/binutils-gdb/gdb/dwarf2/read.c:14616:31:   required from here
    /usr/include/c++/4.8/bits/hashtable_policy.h:1070:12: error: invalid use of incomplete type ‘struct std::hash<sect_offset>’
         struct _Hash_code_base<_Key, _Value, _ExtractKey, _H1, _H2,

This is the same problem and fix as in commit f23f598e28ad ("[gdb] Fix
build breaker with gcc 4.8").  Pass an explicit hash function rather
than relying on the default std::hash<sect_offset>.

gdb/ChangeLog:

PR gdb/25881
* dwarf2/read.c (offset_map_type): Use
gdb:hash_enum<sect_offset> as hash function.

4 years agoRebase libiberty source with latest changes from gcc.
Nick Clifton [Tue, 28 Apr 2020 10:56:06 +0000 (11:56 +0100)]
Rebase libiberty source with latest changes from gcc.

PR 25876
PR demangler/94797
* cp-demangle.c (cplus_demangle_operators): Add ss <=> operator.
* testsuite/demangle-expected: Add operator<=> test.

4 years agoFix typo (thead -> thread)
Tankut Baris Aktemur [Tue, 28 Apr 2020 09:30:52 +0000 (11:30 +0200)]
Fix typo (thead -> thread)

gdb/stubs/ChangeLog:
2020-04-28  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* ia64vms-stub.c: Fix typo in comment (thead -> thread).

gdb/testsuite/ChangeLog:
2020-04-28  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdb.threads/stop-with-handle.exp: Fix typo in comment
(theads -> threads).

gdbsupport/ChangeLog:
2020-04-28  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* gdb-sigmask.h: Fix typo (pthead_sigmask -> pthread_sigmask).

4 years ago[gdb/testsuite] Add PR number to KFAIL in gdb.opt/inline-cmds.exp
Tom de Vries [Tue, 28 Apr 2020 06:33:40 +0000 (08:33 +0200)]
[gdb/testsuite] Add PR number to KFAIL in gdb.opt/inline-cmds.exp

With test-case gdb.opt/inline-cmds.exp, we have:
...
KFAIL: gdb.opt/inline-cmds.exp: next to second func1 (PRMS: gdb/NNNN)
...

I've filed PR25884 for this failure.

Set the KFAIL PR accordingly.

gdb/testsuite/ChangeLog:

2020-04-28  Tom de Vries  <tdevries@suse.de>

* gdb.opt/inline-cmds.exp: Set KFAIL PR.

4 years ago[gdb/testsuite] Remove KFAIL from gdb.base/info-macros.exp
Tom de Vries [Tue, 28 Apr 2020 04:54:55 +0000 (06:54 +0200)]
[gdb/testsuite] Remove KFAIL from gdb.base/info-macros.exp

When running test-case gdb.base/info-macros.exp, we have:
...
(gdb) KFAIL: gdb.base/info-macros.exp: info macros info-macros.c:42 \
  (PRMS: gdb/NNNN)
...

The described failure mode however:
...
set test "info macros info-macros.c:42"

set r1 ".*define DEF_MACROS"
set r2 ".*define ONE"
setup_kfail "gdb/NNNN" *-*-*
gdb_test "$test" "$r1$r2"
...
does not match the actual output, given that both defines are in fact
printed.

The pattern fails to match because it's missing a trailing ".*".

Fix this by removing the KFAIL and adding the missing trailing ".*".

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2020-04-28  Tom de Vries  <tdevries@suse.de>

* gdb.base/info-macros.exp: Remove KFAIL.  Add missing trailing ".*".

4 years ago[gdb/testsuite] Add PR number in KFAIL in gdb.ada/array_ptr_renaming.exp
Tom de Vries [Tue, 28 Apr 2020 04:22:36 +0000 (06:22 +0200)]
[gdb/testsuite] Add PR number in KFAIL in gdb.ada/array_ptr_renaming.exp

When running test-case gdb.ada/array_ptr_renaming.exp we run into:
...
(gdb) print ntp^M
$3 = (3 => 30, 40)^M
(gdb) KFAIL: gdb.ada/array_ptr_renaming.exp: print ntp (PRMS: gdb/NNNN)
...

I've opened PR25883 for this failure.  Reference the PR from the KFAIL, such
that we have:
...
KFAIL: gdb.ada/array_ptr_renaming.exp: print ntp (PRMS: gdb/25883)
...

gdb/testsuite/ChangeLog:

2020-04-28  Tom de Vries  <tdevries@suse.de>

* gdb.ada/array_ptr_renaming.exp: Add PR number in KFAIL.

4 years ago[gdb/symtab] Handle struct decl with DW_AT_signature
Tom de Vries [Tue, 28 Apr 2020 04:12:35 +0000 (06:12 +0200)]
[gdb/symtab] Handle struct decl with DW_AT_signature

Consider a test-case with sources 36.c:
...
struct s { int i; };
extern void f (void);
int main (void) {
  struct s a;
  f ();
  return 0;
}
...
and 36b.c:
...
struct s { int j; };
void f (void) {
  struct s b;
}
...
compiled like this:
...
$ gcc 36.c 36b.c -g
...

It contains DWARF like this:
...
 <0><d2>: Abbrev Number: 1 (DW_TAG_compile_unit)
    <d8>   DW_AT_name        : 36.c
 <1><f4>: Abbrev Number: 2 (DW_TAG_structure_type)
    <f5>   DW_AT_name        : s
 <2><fe>: Abbrev Number: 3 (DW_TAG_member)
    <ff>   DW_AT_name        : i
 <1><110>: Abbrev Number: 5 (DW_TAG_subprogram)
    <111>   DW_AT_name        : main
 <2><12d>: Abbrev Number: 6 (DW_TAG_variable)
    <12e>   DW_AT_name        : a
    <132>   DW_AT_type        : <0xf4>
 <0><146>: Abbrev Number: 1 (DW_TAG_compile_unit)
    <14c>   DW_AT_name        : 36b.c
 <1><168>: Abbrev Number: 2 (DW_TAG_structure_type)
    <169>   DW_AT_name        : s
 <2><172>: Abbrev Number: 3 (DW_TAG_member)
    <173>   DW_AT_name        : j
 <1><184>: Abbrev Number: 5 (DW_TAG_subprogram)
    <185>   DW_AT_name        : f
 <2><19b>: Abbrev Number: 6 (DW_TAG_variable)
    <19c>   DW_AT_name        : b
    <1a0>   DW_AT_type        : <0x168>
...

And when printing "struct s", we get first a random one (with int j), and then
context-specific ones (with int i in main, and int j in f):
...
$ gdb -batch a.out \
  -ex "ptype struct s" \
  -ex start \
  -ex "ptype struct s" \
  -ex "break f" -ex continue \
  -ex "ptype struct s" \
  | grep "int [ij];"
    int j;
    int i;
    int j;
...
Same for -readnow.

However, if we use -fdebug-types-section:
...
$ gcc 36.c 36b.c -g -fdebug-types-section
...
we get:
...
$ gdb ... | grep "int [ij];"
    int j;
    int i;
    int i;
$ gdb -readnow ... | grep "int [ij];"
    int j;
    int j;
    int j;
...

This is due to the fact that both "struct s" DIEs have been moved to the
.debug_types section:
...
  Compilation Unit @ offset 0x0:
   Signature:     0xfd1462823bb6f7b7
 <0><17>: Abbrev Number: 1 (DW_TAG_type_unit)
 <1><1d>: Abbrev Number: 2 (DW_TAG_structure_type)
    <1e>   DW_AT_name        : s
 <2><27>: Abbrev Number: 3 (DW_TAG_member)
    <28>   DW_AT_name        : i
  Compilation Unit @ offset 0x3a:
   Signature:     0x534310fbefba324d
 <0><51>: Abbrev Number: 1 (DW_TAG_type_unit)
 <1><57>: Abbrev Number: 2 (DW_TAG_structure_type)
    <58>   DW_AT_name        : s
 <2><61>: Abbrev Number: 3 (DW_TAG_member)
    <62>   DW_AT_name        : j
...
and there's no longer a "struct s" DIE in the 36.c and
and 36b.c CUs to specify which "struct s" belongs in the CU.  This is gcc
PR90232.

However, using a tentative patch for gcc that adds these DIEs (according to
DWARF standard: If the complete declaration of a type has been placed in a
separate type unit, an incomplete declaration of that type in the compilation
unit may provide the unique 64-bit signature of the type using a
DW_AT_signature attribute):
...
  <0><d2>: Abbrev Number: 5 (DW_TAG_compile_unit)
     <d8>   DW_AT_name        : 36.c
+ <1><f4>: Abbrev Number: 6 (DW_TAG_structure_type)
+    <f5>   DW_AT_name        : s
+    <f7>   DW_AT_signature   : signature: 0xfd1462823bb6f7b7
+    <ff>   DW_AT_declaration : 1
  <0><13c>: Abbrev Number: 5 (DW_TAG_compile_unit)
     <142>   DW_AT_name        : 36b.c
+ <1><15e>: Abbrev Number: 6 (DW_TAG_structure_type)
+    <15f>   DW_AT_name        : s
+    <161>   DW_AT_signature   : signature: 0x534310fbefba324d
+    <169>   DW_AT_declaration : 1
...
still does not help, because they're declarations, so new_symbol is not called
for them in process_structure_scope.

Fix this by calling new_symbol for these decls.

Build and tested on x86_64-linux.

Also tested with target board enabling by default -fdebug-types-section
-gdwarf-4, and with gcc with aforementioned tentative patch.  In this
configuration, the patch reduces number of FAILs from 2888 to 238.

gdb/ChangeLog:

2020-04-28  Tom de Vries  <tdevries@suse.de>

* dwarf2/read.c (process_structure_scope): Add symbol for struct decl
with DW_AT_signature.

gdb/testsuite/ChangeLog:

2020-04-28  Tom de Vries  <tdevries@suse.de>

* gdb.dwarf2/main-foo.c: New test.
* gdb.dwarf2/struct-with-sig.exp: New file.

4 years agoAutomatic date update in version.in
GDB Administrator [Tue, 28 Apr 2020 00:00:10 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoalpha-vms: divide by zero
Alan Modra [Mon, 27 Apr 2020 23:02:13 +0000 (08:32 +0930)]
alpha-vms: divide by zero

The zero check was on the wrong operand.  And, yes, the second operand
popped is supposed to be divided by the first operand popped.

* vms-alpha.c (_bfd_vms_slurp_etir): Correct divide by zero check.
Emit warning message.

4 years agox86: Add i386 PE big-object support
Tamar Christina [Mon, 27 Apr 2020 16:39:31 +0000 (17:39 +0100)]
x86: Add i386 PE big-object support

The 64-bit version of binutils got support for the PE COFF BIG OBJ format a
couple of years ago.   The BIG OBJ format is a slightly different COFF format
which extends the size of the number of section field in the header from a
uint16_t to a uint32_t and so greatly increases the number of sections allowed.

However the 32-bit version of bfd never got support for this.  The GHC Haskell
compiler generates a great deal of symbols due to it's use of
-ffunction-sections and -fdata-sections.

This meant that we could not build the 32-bit version of the GHC Compiler for
many releases now as binutils didn't have this support.

This patch adds the support to the 32-bit port of binutils as well and also does
come cleanup in the code.

bfd/ChangeLog:

* coff-i386.c (COFF_WITH_PE_BIGOBJ): New.
* coff-x86_64.c (COFF_WITH_PE_BIGOBJ): New.
* config.bfd (targ_selvecs): Rename x86_64_pe_be_vec
to x86_64_pe_big_vec as it not a big-endian format.
(vec i386_pe_big_vec): New.
* configure.ac: Likewise.
* targets.c: Likewise.
* configure: Regenerate.
* pe-i386.c (TARGET_SYM_BIG, TARGET_NAME_BIG,
COFF_WITH_PE_BIGOBJ): New.
* pe-x86_64.c (TARGET_SYM_BIG, TARGET_NAME_BIG):
New.
(x86_64_pe_be_vec): Moved.

gas/ChangeLog:

* NEWS: Add news entry for big-obj.
* config/tc-i386.c (i386_target_format): Support new format.
* doc/c-i386.texi: Add i386 support.
* testsuite/gas/pe/big-obj.d: Rename test to not be x64 specific.
* testsuite/gas/pe/pe.exp (big-obj): Make test run on i386 as well.

ld/ChangeLog:

* pe-dll.c (pe_detail_list):  Add pe-bigobj-i386.

4 years agogdb, gdbserver: remove configure check for fs_base/gs_base in user_regs_struct
Simon Marchi [Mon, 27 Apr 2020 14:46:51 +0000 (10:46 -0400)]
gdb, gdbserver: remove configure check for fs_base/gs_base in user_regs_struct

I recently stumbled on this code mentioning Linux kernel 2.6.25, and
thought it could be time for some spring cleaning (newer GDBs probably
don't need to supports 12-year old kernels).  I then found that the
"legacy" case is probably broken anyway, which gives an even better
motivation for its removal.

In short, this patch removes the configure checks that check if
user_regs_struct contains the fs_base/gs_base fields and adjusts all
uses of the HAVE_STRUCT_USER_REGS_STRUCT_{FS,GS}_BASE macros.  The
longer explanation/rationale follows.

Apparently, Linux kernels since 2.6.25 (that's from 2008) have been
reliably providing fs_base and gs_base as part of user_regs_struct.
Commit df5d438e33d7 in the Linux kernel [1] seems related.  This means
that we can get these values by reading registers with PTRACE_GETREGS.
Previously, these values were obtained using a separate
PTRACE_ARCH_PRCTL ptrace call.

First, I'm not even sure the configure check was really right in the
first place.

The user_regs_struct used by GDB comes from
/usr/include/x86_64-linux-gnu/sys/user.h (or equivalent on other
distros) and is provided by glibc.  glibc has had the fs_base/gs_base
fields in there for a very long time, at least since this commit from
2001 [2].  The Linux kernel also has its version of user_regs_struct,
which I think was exported to user-space at some point.  It included the
fs_base/gs_base fields since at least this 2002 commit [3].  In any
case, my conclusion is that the fields were there long before the
aforementioned Linux kernel commit.  The kernel commit didn't add these
fields, it only made sure that they have reliable values when obtained
with PTRACE_GETREGS.

So, checking for the presence of the fs_base/gs_base fields in struct
user_regs_struct doesn't sound like a good way of knowing if we can
reliably get the fs_base/gs_base values from PTRACE_GETREGS.  My guess
is that if we were using that strategy on a < 2.6.25 kernel, things
would not work correctly:

- configure would find that the user_regs_struct has the fs_base/gs_base
  fields (which are probided by glibc anyway)
- we would be reading the fs_base/gs_base values using PTRACE_GETREGS,
  for which the kernel would provide unreliable values

Second, I have tried to see how things worked by forcing GDB to not use
fs_base/gs_base from PTRACE_GETREGS (forcing it to use the "legacy"
code, by configuring with

  ac_cv_member_struct_user_regs_struct_gs_base=no ac_cv_member_struct_user_regs_struct_fs_base=no

Doing so breaks writing registers back to the inferior.  For example,
calling an inferior functions gives an internal error:

    (gdb) p malloc(10)
    /home/smarchi/src/binutils-gdb/gdb/i387-tdep.c:1408: internal-error: invalid i387 regnum 152

The relevant last frames where this error happens are:

    #8  0x0000563123d262fc in internal_error (file=0x563123e93fd8 "/home/smarchi/src/binutils-gdb/gdb/i387-tdep.c", line=1408, fmt=0x563123e94482 "invalid i387 regnum %d") at /home/smarchi/src/binutils-gdb/gdbsupport/errors.cc:55
    #9  0x0000563123047d0d in i387_collect_xsave (regcache=0x5631269453f0, regnum=152, xsave=0x7ffd38402a20, gcore=0) at /home/smarchi/src/binutils-gdb/gdb/i387-tdep.c:1408
    #10 0x0000563122c69e8a in amd64_collect_xsave (regcache=0x5631269453f0, regnum=152, xsave=0x7ffd38402a20, gcore=0) at /home/smarchi/src/binutils-gdb/gdb/amd64-tdep.c:3448
    #11 0x0000563122c5e94c in amd64_linux_nat_target::store_registers (this=0x56312515fd10 <the_amd64_linux_nat_target>, regcache=0x5631269453f0, regnum=152) at /home/smarchi/src/binutils-gdb/gdb/amd64-linux-nat.c:335
    #12 0x00005631234c8c80 in target_store_registers (regcache=0x5631269453f0, regno=152) at /home/smarchi/src/binutils-gdb/gdb/target.c:3485
    #13 0x00005631232e8df7 in regcache::raw_write (this=0x5631269453f0, regnum=152, buf=0x56312759e468 "@\225\372\367\377\177") at /home/smarchi/src/binutils-gdb/gdb/regcache.c:765
    #14 0x00005631232e8f0c in regcache::cooked_write (this=0x5631269453f0, regnum=152, buf=0x56312759e468 "@\225\372\367\377\177") at /home/smarchi/src/binutils-gdb/gdb/regcache.c:778
    #15 0x00005631232e75ec in regcache::restore (this=0x5631269453f0, src=0x5631275eb130) at /home/smarchi/src/binutils-gdb/gdb/regcache.c:283
    #16 0x0000563123083fc4 in infcall_suspend_state::restore (this=0x5631273ed930, gdbarch=0x56312718cf20, tp=0x5631270bca90, regcache=0x5631269453f0) at /home/smarchi/src/binutils-gdb/gdb/infrun.c:9103
    #17 0x0000563123081eed in restore_infcall_suspend_state (inf_state=0x5631273ed930) at /home/smarchi/src/binutils-gdb/gdb/infrun.c:9151

The problem seems to be that amd64_linux_nat_target::store_registers
calls amd64_native_gregset_supplies_p to know whether gregset provides
fs_base.  When !HAVE_STRUCT_USER_REGS_STRUCT_FS_BASE,
amd64_native_gregset_supplies_p returns false.  store_registers
therefore assumes that it must be an "xstate" register.  This is of
course wrong, and that leads to the failed assertion when
i387_collect_xsave doesn't recognize the register.

amd64_linux_nat_target::store_registers could probably be fixed to
handle this case, but I don't think it's worth it, given that it would
only be to support very old kernels.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=df5d438e33d7fc914ba9b6e0d6b019a8966c5fcc
[2] https://sourceware.org/git/?p=glibc.git;a=commit;h=c9cf6ddeebb7bb
[3] https://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git/commit/?id=88e4bc32686ebd0b1111a94f93eba2d334241f68

gdb/ChangeLog:

* configure.ac: Remove check for fs_base/gs_base in
user_regs_struct.
* configure: Re-generate.
* config.in: Re-generate.
* amd64-nat.c (amd64_native_gregset_reg_offset): Adjust.
* amd64-linux-nat.c (amd64_linux_nat_target::fetch_registers,
amd64_linux_nat_target::store_registers, ps_get_thread_area, ): Adjust.

gdbserver/ChangeLog:

* configure.ac: Remove check for fs_base/gs_base in
user_regs_struct.
* configure: Re-generate.
* config.in: Re-generate.
* linux-x86-low.cc (x86_64_regmap, x86_fill_gregset,
x86_store_gregset): Adjust.

4 years agoExpand dynamic type documentation
Tom Tromey [Mon, 27 Apr 2020 14:23:53 +0000 (08:23 -0600)]
Expand dynamic type documentation

This expands the Python dynamic type documentation, as suggested by
Christian.

gdb/doc/ChangeLog
2020-04-27  Tom Tromey  <tromey@adacore.com>

* python.texi (Types In Python): Mention missing fields.  Add
dynamic type example.

4 years agogdbsupport: include cstdlib in common-defs.h
Simon Marchi [Mon, 27 Apr 2020 13:19:48 +0000 (09:19 -0400)]
gdbsupport: include cstdlib in common-defs.h

In PR 25731 [1], the following build failure was reported:

    ../../binutils-gdb/gdb/gdbtypes.c:1254:10: error: no member named 'abs' in namespace 'std'; did you mean simply 'abs'?
                = ((std::abs (stride) * element_count) + 7) / 8;
                    ^~~~~~~~
                    abs
    /usr/include/stdlib.h:129:6: note: 'abs' declared here
    int      abs(int) __pure2;
             ^
The original report was using:

    $ gcc -v
    Apple LLVM version 8.0.0 (clang-800.0.42.1)
    Target: x86_64-apple-darwin15.6.0

Note that I was _not_ able to reproduce using:

    $ g++ --version
    Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include/c++/4.2.1
    Apple clang version 11.0.0 (clang-1100.0.33.17)
    Target: x86_64-apple-darwin19.3.0

The proposed fix is to include <cstdlib> in addition to <stdlib.h>.

Here's an excerpt of [2] relevant to this problem:

    These headers [speaking of the .h form] are allowed to also declare
    the same names in the std namespace, and the corresponding cxxx
    headers are allowed to also declare the same names in the global
    namespace: including <cstdlib> definitely provides std::malloc and
    may also provide ::malloc.  Including <stdlib.h> definitely provides
    ::malloc and may also provide std::malloc

Since we use std::abs, we should not assume that our include of stdlib.h
declares an `abs` function in the `std` namespace.

If we replace the include of stdlib.h with cstdlib, then we fall in the
opposite situation.  A standard C++ library may decide to only put the
declarations in the std namespace, requiring us to prefix all standard
functions with `std::`.  I'm not against that, but for the moment I think the
safest way forward is to just include both.

Note that I don't know what effect this patch can have on any stdlib.h fix
provided by gnulib.

[1] https://sourceware.org/bugzilla/show_bug.cgi?id=25731
[2] https://en.cppreference.com/w/cpp/header#C_compatibility_headers

gdbsupport/ChangeLog:

* common-defs.h: Include cstdlib.h.

4 years agoFix remaining inline/tailcall unwinding breakage for x86_64
Luis Machado [Sat, 25 Apr 2020 03:32:44 +0000 (00:32 -0300)]
Fix remaining inline/tailcall unwinding breakage for x86_64

Commit 5939967b355ba6a940887d19847b7893a4506067 fixed inline
frame unwinding breakage for some targets (aarch64, riscv, s390...)
but regressed a few amd64 testcases related to tailcalls.

Given the following example situation...

Frame #-1 - sentinel frame
Frame # 0 - inline frame
Frame # 1 - normal frame

... suppose we're at level #1 and call into dwarf2_tailcall_sniffer_first.

We'll attempt to fetch PC, which used to be done via the gdbarch_unwind_pc call
(before 5939967b355ba6a940887d19847b7893a4506067), but now it is being handled
by the get_frame_register function.

gdbarch_unwind_pc will attempt to use frame #1's cache to retrieve information
about the PC. Here's where different architectures behave differently.

x86_64 will find a dwarf rule to retrieve PC from memory, at a CFA + offset
location. So the PC value is readily available and there is no need to
create a lazy value.

For aarch64 (and others), GCC doesn't emit an explicit location for PC, so we
eventually will find that PC is DWARF2_FRAME_REG_UNSPECIFIED. This is known
and is handled by GDB by assuming GCC really meant DWARF2_FRAME_REG_SAME_VALUE.

This means we'll attempt to fetch the register value from frame #0, via a call
to frame_unwind_got_register, which will trigger the creation of a lazy value
that requires a valid frame id for frame #0.

We don't have a valid id for frame #0 yet, so we assert.

Given the above, the following patch attempts to handle the situation without
being too hacky. We verify if the next frame is an inline frame and if its
frame id has been computed already. If it hasn't been computed yet, then we
use the safer get_frame_register function, otherwise we use the regular
gdbarch_unwind_pc hook.

gdb/ChangeLog:

2020-04-27  Luis Machado  <luis.machado@linaro.org>

* dwarf2/frame-tailcall.c (dwarf2_tailcall_sniffer_first): Handle
problematic inline frame unwinding situation.
* frame.c (frame_id_computed_p): New function.
* frame.h (frame_id_computed_p): New prototype.

4 years agoGAS: Allow automatically assigned entries in the file table to be reassigned if the...
Nick Clifton [Mon, 27 Apr 2020 10:35:25 +0000 (11:35 +0100)]
GAS: Allow automatically assigned entries in the file table to be reassigned if the source file specifically requests to use the assigned slot.

PR 25878
* dwarf2dbg.c (struct file_entry): Add auto_assigned field.
(assign_file_to_slot): New function.  Fills in an entry in the
files table.
(allocate_filenum): Use new function.
(allocate_filename_to_slot): Use new function.  If the specified
slot entry is already in use, but was chosen automatically then
reassign the automatic entry.

4 years agoAutomatic date update in version.in
GDB Administrator [Mon, 27 Apr 2020 00:00:15 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoRemove class_pseudo
Tom Tromey [Sun, 26 Apr 2020 19:36:04 +0000 (13:36 -0600)]
Remove class_pseudo

The class_pseudo constant is unused, so this removes it.
Tested by rebuilding.

gdb/ChangeLog
2020-04-26  Tom Tromey  <tom@tromey.com>

* command.h (enum command_class) <class_pseudo>: Remove.

4 years agoreadelf: NULL dereference
Alan Modra [Sun, 26 Apr 2020 09:00:50 +0000 (18:30 +0930)]
readelf: NULL dereference

This fixes another missing error check.

* readelf.c (get_num_dynamic_syms): Check DT_MIPS_XHASH was
read before dereferencing, and gracefully return.  Remove
gnu_hash_error variable.  Free gnu hash arrays if number of
syms found is zero.

4 years agoFix comments and whitespace in lookup_cmd_composition
Philippe Waroquiers [Sun, 26 Apr 2020 14:01:52 +0000 (16:01 +0200)]
Fix comments and whitespace in lookup_cmd_composition

2020-04-26  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* cli/cli-decode.c (lookup_cmd_composition): Fix comments
and whitespace.

4 years agoImprove -mlfence-after-load
liuhongt [Mon, 16 Mar 2020 03:03:12 +0000 (11:03 +0800)]
Improve -mlfence-after-load

  1.Implict load for POP/POPF/POPA/XLATB, no load for Anysize insns
  2. Add -mlfence-before-ret=shl/yes, adjust operand size of
  or/not/shl according to ret's.
  3. Issue warning for REP CMPS/SCAS since they would affect control
  flow behavior.
  4. Adjust testcases and documents.

gas/Changelog:
* config/tc-i386.c (lfence_before_ret_shl): New member.
(load_insn_p): implict load for POP/POPA/POPF/XLATB, no load
for Anysize insns.
(insert_after_load): Issue warning for REP CMPS/SCAS.
(insert_before_before): Handle iret, Handle
-mlfence-before-ret=shl, Adjust operand size of or/not/shl to ret's,
(md_parse_option): Change -mlfence-before-ret=[none|not|or] to
-mlfence-before-ret=[none/not/or/shl/yes].
Enable -mlfence-before-ret=shl when
-mlfence-beofre-indirect-branch=all and no explict -mlfence-before-ret option.
(md_show_usage): Ditto.
* doc/c-i386.texi: Ditto.
* testsuite/gas/i386/i386.exp: Add new testcases.
* testsuite/gas/i386/lfence-load-b.d: New.
* testsuite/gas/i386/lfence-load-b.e: New.
* testsuite/gas/i386/lfence-load.d: Modified.
* testsuite/gas/i386/lfence-load.e: New.
* testsuite/gas/i386/lfence-load.s: Modified.
* testsuite/gas/i386/lfence-ret-a.d: Modified.
* testsuite/gas/i386/lfence-ret-b.d: Modified.
* testsuite/gas/i386/lfence-ret-c.d: New.
* testsuite/gas/i386/lfence-ret-d.d: New.
* testsuite/gas/i386/lfence-ret.s: Modified.
* testsuite/gas/i386/x86-64-lfence-load-b.d: New.
* testsuite/gas/i386/x86-64-lfence-load.d: Modified.
* testsuite/gas/i386/x86-64-lfence-load.s: Modified.
* testsuite/gas/i386/x86-64-lfence-ret-a.d: Modified.
* testsuite/gas/i386/x86-64-lfence-ret-b.d: Modified.
* testsuite/gas/i386/x86-64-lfence-ret-c.d: New.
* testsuite/gas/i386/x86-64-lfence-ret-d.d: New
* testsuite/gas/i386/x86-64-lfence-ret-e.d: New.
* testsuite/gas/i386/x86-64-lfence-ret.e: New.
* testsuite/gas/i386/x86-64-lfence-ret.s: New.

4 years agoAutomatic date update in version.in
GDB Administrator [Sun, 26 Apr 2020 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoRemove unused code block in inf_ptrace_target::wait
Kamil Rytarowski [Fri, 24 Apr 2020 16:10:07 +0000 (18:10 +0200)]
Remove unused code block in inf_ptrace_target::wait

Remove unused PT_GET_PROCESS_STATE block. It used to be used
by OpenBSD, but it is now reimplemented independently in
obsd-nat.c.

gdb/ChangeLog:

       * inf-ptrace.c (inf_ptrace_target::wait): Remove
       `PT_GET_PROCESS_STATE' block.

Change-Id: I9b872df8517b658c0dfe889fc1e4a7009bc5c076

4 years ago[gdb/testsuite] Add target board debug-types
Tom de Vries [Sat, 25 Apr 2020 15:19:26 +0000 (17:19 +0200)]
[gdb/testsuite] Add target board debug-types

This patch adds a target board debug-types that switches on
-fdebug-types-section by default.

This -fdebug-types-section option is a gcc option that enables the generation
of a .debug_types section, which is only effective for DWARF version 4.

There are two other boards that enable this: dwarf4-gdb-index and fisson, but
while those test some meaningful combination of options, this board is
intended to test only -fdebug-types-section.

Current results with gcc 7.5.0 are:
...
 === gdb Summary ===

 # of expected passes            75832
 # of unexpected failures        2841
 # of expected failures          130
 # of known failures             75
 # of unresolved testcases       22
 # of untested testcases         37
 # of unsupported tests          83
...

Related known issues:
- PR gcc/90232 - "gcc drops top-level dies with -fdebug-types-section"
- PR gdb/25875 - "segv in ada_discrete_type_low_bound"
- PR gdb/14148 - "-fdebug-types-section regresses static scope of types"

Tested on x86_64-linux.

gdb/testsuite/ChangeLog:

2020-04-25  Tom de Vries  <tdevries@suse.de>

* boards/debug-types.exp: New file.

4 years agogdb/testsuite: Remove build paths from test names
Andrew Burgess [Thu, 23 Apr 2020 09:44:30 +0000 (10:44 +0100)]
gdb/testsuite: Remove build paths from test names

Having paths in test names makes it harder to compare results from
different builds of GDB.

gdb/testsuite/ChangeLog:

* gdb.btrace/multi-inferior.exp: Avoid paths in test names.

4 years agoAutomatic date update in version.in
GDB Administrator [Sat, 25 Apr 2020 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

4 years agoRemove symbol_get_demangled_name
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Remove symbol_get_demangled_name

Now that symbol_get_demangled_name is only used by general_symbol_info
methods, and because these methods already check the symbol's language
to decide what to return, symbol_get_demangled_name is no longer
needed.  This patch removes it.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* symtab.h (symbol_get_demangled_name): Don't declare.
* symtab.c (symbol_get_demangled_name): Remove.
(general_symbol_info::natural_name)
(general_symbol_info::demangled_name): Update.

4 years agoFix Rust test cases
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Fix Rust test cases

PR rust/25025 notes that some Rust test cases fail.

Debugging gdb revealed that the Rust compiler emits different linkage
names that demangle to the same result.  Enabling complaints when
reading the test case is enough to show it:

    During symbol reading: Computed physname <generics::identity<f64>> does not match demangled <generics::identity> (from linkage <_ZN8generics8identity17h8540b320af6656d6E>) - DIE at 0x424 [in module /home/tromey/gdb/build/gdb/testsuite/outputs/gdb.rust/generics/generics]
    During symbol reading: Computed physname <generics::identity<u32>> does not match demangled <generics::identity> (from linkage <_ZN8generics8identity17hae302fad0c33bd7dE>) - DIE at 0x459 [in module /home/tromey/gdb/build/gdb/testsuite/outputs/gdb.rust/generics/generics]
    ...

This patch changes the DWARF reader to prefer the computed physname,
rather than the output of the demangler, for Rust.  This fixes the
bug.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

PR rust/25025:
* dwarf2/read.c (dwarf2_physname): Do not demangle for Rust.

4 years agoUse the linkage name if it exists
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Use the linkage name if it exists

The DWARF reader has had some odd code since the "physname" patches landed.

In particular, these patches caused PR symtab/12707; namely, they made
it so "set print demangle off" no longer works.

This patch attempts to fix the problem.  It arranges to store the
linkage name on the symbol if it exists, and it changes the DWARF
reader so that the demangled name is no longer (usually) stored in the
symbol's "linkage name" field.

c-linkage-name.exp needed a tweak, because it started working
correctly.  This conforms to what I think ought to happen, so this
seems like an improvement here.

compile-object-load.c needed a small change to use
symbol_matches_search_name rather than directly examining the linkage
name.  Looking directly at the name does the wrong thing for C++.

There is still some name-related confusion in the DWARF reader:

* "physname" often refers to the logical name and not what I would
  consider to be the "physical" name;

* dwarf2_full_name, dwarf2_name, and dwarf2_physname all exist and
  return different strings -- but this seems like at least one name
  too many.  For example, Fortran requires dwarf2_full_name, but other
  languages do not.

* To my surprise, dwarf2_physname prefers the form emitted by the
  demangler over the one that it computes.  This seems backward to me,
  given that the partial symbol reader prefers the opposite, and it
  seems to me that this choice may perform better as well.

I didn't attempt to clean up these things.  It would be good to do,
but whenever I contemplate it I get caught up in dreams of truly
rewriting the DWARF reader instead.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

PR symtab/12707:
* dwarf2/read.c (add_partial_symbol): Use the linkage name if it
exists.
(new_symbol): Likewise.
* compile/compile-object-load.c (get_out_value_type): Use
symbol_matches_search_name.

gdb/testsuite/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

PR symtab/12707:
* gdb.python/py-symbol.exp: Update expected results for
linkage_name test.
* gdb.cp/print-demangle.exp: New file.
* gdb.base/c-linkage-name.exp: Fix test.
* gdb.guile/scm-symbol.exp: Update expected results for
linkage_name test.

4 years agoDon't call compute_and_set_names for partial symbols
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Don't call compute_and_set_names for partial symbols

As mentioned in another thread, there's currently no need to call
compute_and_set_names for partial symbols.  Because the DWARF partial
symbol reader constructs demangled names, this call can only demangle
a name by mistake.

So, this patch changes the DWARF reader to simply set the linkage name
on the new symbol.  This is equivalent to what was done before.  There
should be no user-visible change from this patch, aside from gdb
speeding up a bit.

... there *should* be, but this regressed
dw2-namespaceless-anonymous.exp.  However, upon examination, I think
that test is incorrect.  It puts a mangled name into DW_AT_name, and
it puts the variable at the top level, not in a namespace.  This isn't
what C++ compilers ought to do.  So, this patch also updates the test
case.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* dwarf2/read.c (add_partial_symbol): Do not call
compute_and_set_names.

gdb/testsuite/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* gdb.dwarf2/dw2-namespaceless-anonymous.S: Remove.
* gdb.dwarf2/dw2-namespaceless-anonymous.c: New file.
* gdb.dwarf2/dw2-namespaceless-anonymous.exp: Use DWARF
assembler.

4 years agoUse the new add_psymbol_to_list overload
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Use the new add_psymbol_to_list overload

This changes the DWARF reader to use the new add_psymbol_to_list
overload.  There should be no visible changes due to this patch.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* dwarf2/read.c (add_partial_symbol): Use new add_psymbol_to_list
overload.

4 years agoIntroduce new add_psymbol_to_list overload
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Introduce new add_psymbol_to_list overload

This adds a new overload of add_psymbol_to_list.  This one takes an
already constructed psymbol and adds it to the bcache and the
appropriate list.

This seemed cleaner than continuing to add parameters to the existing
add_psymbol_to_list, and is more in line with how full symbols are
constructed.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* psymtab.c (add_psymbol_to_bcache): Simplify calling convention.
(add_psymbol_to_list): New overload.  Make old overload call new
one.
* psympriv.h (add_psymbol_to_list): New overload.

4 years agoAdd attribute::value_as_string method
Tom Tromey [Fri, 24 Apr 2020 21:35:01 +0000 (15:35 -0600)]
Add attribute::value_as_string method

The full DIE reader checks that an attribute has a "string" form in
some spots, but the partial DIE reader does not.  This patch brings
the two readers in sync for one specific case, namely when examining
the linkage name.  This avoids regressions in an existing DWARF test
case.

A full fix for this problem would be preferable.  An accessor like
DW_STRING should always check the form.  However, I haven't attempted
that in this series.

Also the fact that the partial and full readers can disagree like this
is a design flaw.

gdb/ChangeLog
2020-04-24  Tom Tromey  <tom@tromey.com>

* dwarf2/read.c (partial_die_info::read) <case
DW_AT_linkage_name>: Use value_as_string.
(dwarf2_string_attr): Use value_as_string.
* dwarf2/attribute.h (struct attribute) <value_as_string>: Declare
method.
* dwarf2/attribute.c (attribute::value_as_string): New method.