binutils-gdb.git
5 years ago[AArch64][gas] Add -mcpu support for Arm Ares
Kyrylo Tkachov [Tue, 8 Jan 2019 15:18:32 +0000 (15:18 +0000)]
[AArch64][gas] Add -mcpu support for Arm Ares

This adds support for the Arm Ares CPU for AArch64.
It implements the Armv8.2-A architecture with the optional features
of statistical profiling, dot product and FP16 on by default.

Note: Ares is a codename to enable early adopters and in time
we will add the final product name once it's announced.

    * config/tc-aarch64.c (aarch64_cpus): Add ares.
    * doc/c-aarch64.texi (-mcpu): Document ares value.

5 years agoAdd testcase for PR/24065.
Tamar Christina [Tue, 8 Jan 2019 14:52:53 +0000 (14:52 +0000)]
Add testcase for PR/24065.

This adds a new testcase for objcopy -O ihex to test
that it correctly works.

binutils/ChangeLog:

PR 24065
* testsuite/binutils-all/copy-6.d: New test.
* testsuite/binutils-all/objcopy.exp: Use it.

5 years agogdb: Move declarations from symtab.h to source.h
Andrew Burgess [Mon, 7 Jan 2019 09:16:16 +0000 (09:16 +0000)]
gdb: Move declarations from symtab.h to source.h

Declarations for functions in source.c are split between source.h and
symtab.h.  This commit moves the small number that are in symtab.h
into source.h.  There's just one file that needs to add an include of
source.h in order to build.

I've moved the function header comments from source.c to source.h
inline with the recommended GDB style.

gdb/ChangeLog:

* source.c (select_source_symtab): Move header comment to
declaration in source.h.
(forget_cached_source_info_for_objfile): Likewise.
(forget_cached_source_info): Likewise.
(identify_source_line): Likewise.
* source.h (identify_source_line): Move declaration from symtab.h
and add comment from source.c
(print_source_lines): Likewise.
(forget_cached_source_info_for_objfile): Likewise.
(forget_cached_source_info): Likewise.
(select_source_symtab): Likewise.
(enum print_source_lines_flag): Move definition from symtab.h.
* symtab.h (identify_source_line): Move declaration to source.h.
(print_source_lines): Likewise.
(forget_cached_source_info_for_objfile): Likewise.
(forget_cached_source_info): Likewise.
(select_source_symtab): Likewise.
(enum print_source_lines_flag): Move definition to source.h.
* tui/tui-hooks.c: Add 'source.h' include.

5 years agogdb: Handle requests to print source lines backward
Andrew Burgess [Sun, 6 Jan 2019 23:53:22 +0000 (23:53 +0000)]
gdb: Handle requests to print source lines backward

...by which I mean from high line number to low, not, actually
backward character by character!

Commit:

  commit 62f29fda90cf1d5a1899f57ef78452471c707fd6
  Date:   Tue Oct 9 22:21:05 2018 -0600

      Highlight source code using GNU Source Highlight

introduced a regression in the test gdb.linespec/explicit.exp, in
which a request is made to GDB to print a reverse sequence of lines,
from +10 to -10 from the current line number.  The expected behaviour
is that GDB prints nothing.  The above commit changed this so that GDB
now prints:

  Line number 32 out of range; /path/to/gdb/testsuite/gdb.linespec/explicit.c has 71 lines.

which is a little confusing.

This commit fixes the regression, and restores the behaviour that GDB
prints nothing.

While I was passing I noticed a call to `back` on a std::string that I
was concerned could be empty if the request for source lines returns
an empty string.  I don't know if it would be possible for a request
for lines to return an empty string, I guess it should be impossible,
in which case, maybe this should be an assertion, but adding a `empty`
check, seems like an easy and cheap safety net.

gdb/ChangeLog:

* source.c (print_source_lines_base): Handle requests to print
reverse line number sequences, and guard against empty lines
string.

5 years agogdb: Fix skip of `\r` before `\n` in source output
Andrew Burgess [Sun, 6 Jan 2019 23:31:45 +0000 (23:31 +0000)]
gdb: Fix skip of `\r` before `\n` in source output

In this commit:

  commit 62f29fda90cf1d5a1899f57ef78452471c707fd6
  Date:   Tue Oct 9 22:21:05 2018 -0600

      Highlight source code using GNU Source Highlight

A bug was introduced such that when displaying source code from a file
with lines `\r\n` GDB would print `^M` at the end of each line.

This caused a regression on the test gdb.fortran/nested-funcs.exp,
which happens to have `\r\n` line endings.

gdb/ChangeLog:

* source.c (print_source_lines_base): Fix skip of '\r' if next
character is '\n'.

5 years agoPR24065, 32-bit objcopy fails with 64-bit address ... out of range
Alan Modra [Tue, 8 Jan 2019 11:51:57 +0000 (22:21 +1030)]
PR24065, 32-bit objcopy fails with 64-bit address ... out of range

PR 23699
PR 24065
* ihex.c (ihex_write_object_contents): Properly check 32-bit
address range.

5 years agorun_dump_test source in build directory
Alan Modra [Tue, 8 Jan 2019 06:17:52 +0000 (16:47 +1030)]
run_dump_test source in build directory

Some existing tests build .s and .d files for run_dump_test, using an
absolute #source: line in the .d file.  This patch changes that scheme
a little to instead use "#source: ./..." in .d files rather than
"#source: $objdir/...", which is more useful in cases where the .d
file is not generated.

This allows RX gas test files to be built in the build directory,
rather than in a source directory (which might be read-only).

binutils/
* testsuite/lib/binutils-common.exp (run_dump_test): Don't prepend
$srcdir/$subdir to source file name if it starts with "./".
gas/
* testsuite/gas/rx/rx.exp: Create generated test source in
current directory.
* testsuite/gas/rx/Xtod.d, * testsuite/gas/rx/abs.d,
* testsuite/gas/rx/adc.d, * testsuite/gas/rx/add.d,
* testsuite/gas/rx/and.d, * testsuite/gas/rx/bclr.d,
* testsuite/gas/rx/bcnd.d, * testsuite/gas/rx/bfmov.d,
* testsuite/gas/rx/bmcnd.d, * testsuite/gas/rx/bnot.d,
* testsuite/gas/rx/bra.d, * testsuite/gas/rx/brk.d,
* testsuite/gas/rx/bset.d, * testsuite/gas/rx/bsr.d,
* testsuite/gas/rx/btst.d, * testsuite/gas/rx/clrpsw.d,
* testsuite/gas/rx/cmp.d, * testsuite/gas/rx/dabs.d,
* testsuite/gas/rx/dadd.d, * testsuite/gas/rx/dbt.d,
* testsuite/gas/rx/dcmp.d, * testsuite/gas/rx/ddiv.d,
* testsuite/gas/rx/div.d, * testsuite/gas/rx/divu.d,
* testsuite/gas/rx/dmov.d, * testsuite/gas/rx/dmul.d,
* testsuite/gas/rx/dneg.d, * testsuite/gas/rx/dpopm.d,
* testsuite/gas/rx/dpushm.d, * testsuite/gas/rx/dround.d,
* testsuite/gas/rx/dsqrt.d, * testsuite/gas/rx/dsub.d,
* testsuite/gas/rx/dtoX.d, * testsuite/gas/rx/emaca.d,
* testsuite/gas/rx/emsba.d, * testsuite/gas/rx/emul.d,
* testsuite/gas/rx/emula.d, * testsuite/gas/rx/emulu.d,
* testsuite/gas/rx/fadd.d, * testsuite/gas/rx/fcmp.d,
* testsuite/gas/rx/fdiv.d, * testsuite/gas/rx/fmul.d,
* testsuite/gas/rx/fsqrt.d, * testsuite/gas/rx/fsub.d,
* testsuite/gas/rx/ftoi.d, * testsuite/gas/rx/ftou.d,
* testsuite/gas/rx/gprel.d, * testsuite/gas/rx/int.d,
* testsuite/gas/rx/itof.d, * testsuite/gas/rx/jmp.d,
* testsuite/gas/rx/jsr.d, * testsuite/gas/rx/machi.d,
* testsuite/gas/rx/maclh.d, * testsuite/gas/rx/maclo.d,
* testsuite/gas/rx/max.d, * testsuite/gas/rx/min.d,
* testsuite/gas/rx/mov.d, * testsuite/gas/rx/movco.d,
* testsuite/gas/rx/movli.d, * testsuite/gas/rx/movu.d,
* testsuite/gas/rx/msbhi.d, * testsuite/gas/rx/msblh.d,
* testsuite/gas/rx/msblo.d, * testsuite/gas/rx/mul.d,
* testsuite/gas/rx/mulhi.d, * testsuite/gas/rx/mullh.d,
* testsuite/gas/rx/mullo.d, * testsuite/gas/rx/mvfacgu.d,
* testsuite/gas/rx/mvfachi.d, * testsuite/gas/rx/mvfaclo.d,
* testsuite/gas/rx/mvfacmi.d, * testsuite/gas/rx/mvfc.d,
* testsuite/gas/rx/mvfcp.d, * testsuite/gas/rx/mvfdc.d,
* testsuite/gas/rx/mvfdr.d, * testsuite/gas/rx/mvtacgu.d,
* testsuite/gas/rx/mvtachi.d, * testsuite/gas/rx/mvtaclo.d,
* testsuite/gas/rx/mvtc.d, * testsuite/gas/rx/mvtcp.d,
* testsuite/gas/rx/mvtdc.d, * testsuite/gas/rx/neg.d,
* testsuite/gas/rx/nop.d, * testsuite/gas/rx/not.d,
* testsuite/gas/rx/opecp.d, * testsuite/gas/rx/or.d,
* testsuite/gas/rx/pop.d, * testsuite/gas/rx/popc.d,
* testsuite/gas/rx/popm.d, * testsuite/gas/rx/push.d,
* testsuite/gas/rx/pushc.d, * testsuite/gas/rx/pushm.d,
* testsuite/gas/rx/r-bcc.d, * testsuite/gas/rx/r-bra.d,
* testsuite/gas/rx/racl.d, * testsuite/gas/rx/racw.d,
* testsuite/gas/rx/rdacl.d, * testsuite/gas/rx/rdacw.d,
* testsuite/gas/rx/revl.d, * testsuite/gas/rx/revw.d,
* testsuite/gas/rx/rmpa.d, * testsuite/gas/rx/rolc.d,
* testsuite/gas/rx/rorc.d, * testsuite/gas/rx/rotl.d,
* testsuite/gas/rx/rotr.d, * testsuite/gas/rx/round.d,
* testsuite/gas/rx/rstr.d, * testsuite/gas/rx/rte.d,
* testsuite/gas/rx/rtfi.d, * testsuite/gas/rx/rts.d,
* testsuite/gas/rx/rtsd.d, * testsuite/gas/rx/sat.d,
* testsuite/gas/rx/satr.d, * testsuite/gas/rx/save.d,
* testsuite/gas/rx/sbb.d, * testsuite/gas/rx/sccnd.d,
* testsuite/gas/rx/scmpu.d, * testsuite/gas/rx/setpsw.d,
* testsuite/gas/rx/shar.d, * testsuite/gas/rx/shll.d,
* testsuite/gas/rx/shlr.d, * testsuite/gas/rx/smovb.d,
* testsuite/gas/rx/smovf.d, * testsuite/gas/rx/smovu.d,
* testsuite/gas/rx/sstr.d, * testsuite/gas/rx/stnz.d,
* testsuite/gas/rx/stz.d, * testsuite/gas/rx/sub.d,
* testsuite/gas/rx/suntil.d, * testsuite/gas/rx/swhile.d,
* testsuite/gas/rx/tst.d, * testsuite/gas/rx/utof.d,
* testsuite/gas/rx/wait.d, * testsuite/gas/rx/xchg.d,
* testsuite/gas/rx/xor.d: Add #source line.
ld/
* testsuite/ld-elf/sec64k.exp: Use . rather than $objdir in
generated source file names.
* testsuite/ld-m68k/m68k-got.exp: Likewise.

5 years agoAutomatic date update in version.in
GDB Administrator [Tue, 8 Jan 2019 00:00:27 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoRevert "Sync libiberty sources with master version in gcc repository. Updated stabs...
Simon Marchi [Mon, 7 Jan 2019 23:05:36 +0000 (18:05 -0500)]
Revert "Sync libiberty sources with master version in gcc repository.  Updated stabs demangling and cxxfilt tests to match."

The previous commit breaks the GDB build, which is still using functions
cplus_demangle_opname & co.  Since removing these usages is not an
obvious fix, let's revert this patch until we get rid of them.

5 years agoSync libiberty sources with master version in gcc repository. Updated stabs demangli...
Nick Clifton [Mon, 7 Jan 2019 13:18:53 +0000 (13:18 +0000)]
Sync libiberty sources with master version in gcc repository.  Updated stabs demangling and cxxfilt tests to match.

PR 24044
* stabs.c (parse_stab_argtypes): Remove call to
cplus_mangle_opcode.
* testsuite/binutils-all/cxxfilt.exp: Replace tests of v2 encoding
with v3 encoding.  Add escape for known failures.

5 years ago[arm][gas] Add -mcpu support for Arm Ares
Kyrylo Tkachov [Mon, 7 Jan 2019 09:22:55 +0000 (09:22 +0000)]
[arm][gas] Add -mcpu support for Arm Ares

This adds support for the Arm Ares CPU in the arm port.
It implements the Armv8.2-A architecture with the relevant optional
features
of dot product and FP16 on by default.

Note: Ares is a codename to enable early adopters and in time
we will add the final product name once it's announced.

* config/tc-arm.c (arm_cpus): Add ares.
* doc/c-arm.texi (-mcpu): Document ares value.

5 years agos12z regen
Alan Modra [Mon, 7 Jan 2019 04:33:43 +0000 (15:03 +1030)]
s12z regen

* configure: Regenerate.
* po/POTFILES.in: Regenerate.

5 years agoAutomatic date update in version.in
GDB Administrator [Mon, 7 Jan 2019 00:01:08 +0000 (00:01 +0000)]
Automatic date update in version.in

5 years agoRemove remaining cleanups from c-exp.y
Tom Tromey [Thu, 3 Jan 2019 02:22:44 +0000 (19:22 -0700)]
Remove remaining cleanups from c-exp.y

This removes the remaining cleanups from c-exp.y by moving some
globals into c_parse_state, and changing expansion_obstack to be an
auto_obstack.

gdb/ChangeLog
2019-01-06  Tom Tromey  <tom@tromey.com>

* c-exp.y (struct c_parse_state) <macro_original_text,
expansion_obstack>: New member.
(macro_original_text, expansion_obstack): Remove globals.
(scan_macro_expansion, scanning_macro_expansion)
(finished_macro_expansion): Update.
(scan_macro_cleanup): Remove.
(yylex, c_parse): Update.

5 years agoRemove string-related cleanup from c-exp.y
Tom Tromey [Thu, 3 Jan 2019 02:13:15 +0000 (19:13 -0700)]
Remove string-related cleanup from c-exp.y

This removes a string-related cleanup from c-exp.y, by adding a new
member to c_parse_state to store the strings.

gdb/ChangeLog
2019-01-06  Tom Tromey  <tom@tromey.com>

* c-exp.y (struct c_parse_state) <strings>: New member.
(operator_stoken): Update.

5 years agoUse std::vector in type stacks
Tom Tromey [Thu, 3 Jan 2019 02:12:32 +0000 (19:12 -0700)]
Use std::vector in type stacks

This removes the use of VEC from parse.c and, at the same time,
removes some related cleanups from c-exp.y.

gdb/ChangeLog
2019-01-06  Tom Tromey  <tom@tromey.com>

* parser-defs.h (type_ptr): Remove typedef.  Don't declare VEC.
(union type_stack_elt) <typelist_val>: Now a pointer to
std::vector.
(type_stack_cleanup): Don't declare.
(push_typelist): Update.
* parse.c (pop_typelist): Return a std::vector.
(push_typelist): Take a std::vector.
(follow_types): Update.  Do not free args.
(type_stack_cleanup): Remove.
* c-exp.y (struct c_parse_state): New.
(cpstate): New global.
(type_aggregate_p, exp, ptr_operator, parameter_typelist)
(nonempty_typelist): Update.
(func_mod): Create a new vector.
(c_parse): Create a c_parse_state.
(check_parameter_typelist): Do not delete params.
(function_method): Update.  Do not delete type_list.

5 years agoFix crash in "finish"
Tom Tromey [Sat, 29 Dec 2018 17:46:07 +0000 (10:46 -0700)]
Fix crash in "finish"

PR gdb/28155 notes a crash in "finish" that occurs with a particular
source file compiled by clang.

The bug is the typical gdb problem of a missing call to check_typedef.
clang emits a function whose return type is a typedef to void.
get_return_value asserts that the return type is not void, but the
callers were not using check_typedef first.

gdb/ChangeLog
2019-01-06  Tom Tromey  <tom@tromey.com>

PR gdb/28155:
* python/py-finishbreakpoint.c (bpfinishpy_init): Use
check_typedef.
* infcmd.c (finish_command_fsm_should_stop): Use check_typedef.
(print_return_value): Likewise.

gdb/testsuite/ChangeLog
2019-01-06  Tom Tromey  <tom@tromey.com>

PR gdb/28155:
* gdb.dwarf2/typedef-void-finish.exp: New file.

5 years agoAutomatic date update in version.in
GDB Administrator [Sun, 6 Jan 2019 00:00:54 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoRemove the exception and cleanup checkers
Tom Tromey [Mon, 24 Dec 2018 21:56:03 +0000 (14:56 -0700)]
Remove the exception and cleanup checkers

Now that gdb is transitioning away from cleanups, there is no reason
to keep the cleanup and exception checker scripts around.  This patch
removes them.

gdb/ChangeLog
2019-01-05  Tom Tromey  <tom@tromey.com>

* contrib/cleanup_check.py: Remove.
* contrib/gcc-with-excheck: Remove.
* contrib/exsummary.py: Remove.
* contrib/excheck.py: Remove.

5 years agoRX: binutils - Add RXv3 support.
Yoshinori Sato [Tue, 25 Dec 2018 11:04:47 +0000 (20:04 +0900)]
RX: binutils - Add RXv3 support.

* readelf.c (get_machine_flags): Add RXv3 output.

5 years agoRX: bfd - Add RXv3 support.
Yoshinori Sato [Tue, 25 Dec 2018 10:58:54 +0000 (19:58 +0900)]
RX: bfd - Add RXv3 support.

        * bfd/archures.c: Add bfd_mach_rx_v2 and bfd_mach_rx_v3.
        * bfd/bfd-in2.h: Regenerate.
        * bfd/cpu-rx.c (arch_info_struct): Add RXv2 and RXv3 entry.
        * bfd/elf32-rx.c (elf32_rx_machine): Add RXv2 and RXv3 support.

5 years agoRX: gas - Add RXv3 instruction support.
Yoshinori Sato [Tue, 25 Dec 2018 11:52:53 +0000 (20:52 +0900)]
RX: gas - Add RXv3 instruction support.

Instruction manual.
https://www.renesas.com/us/en/doc/products/mpumcu/doc/rx_family/r01us0316ej0100-rxv3sm.pdf

* config/rx-defs.h (rx_cpu_types): Add type RXV3 and RXV3FPU.
(rx_bfield): Add prototype.
(rx_post): Likewise.
* config/rx-parse.y: Add v3 instructions and Double FPU registers.
(DSIZE): Define.
(POST): Define.
(rx_check_v3): New. check v3 type.
(rx_check_dfpu): New. check have double support.
(double_condition_table): New. dcmp<cond> contiditon.
(check_condition): Multiple condition support.
(rx_lex): RXv3 instructions support.
Add parse dcmp<cond> instruction and Double FPU registers.
(immediate): Disable optimize in dmov #imm case.
(displacement): Add double displacement in dmov instraction.
* config/tc-rx.c (rx_use_conventional_section_names):
Invert default value in rx-*-linux target.
(cpu_type): Add additional ELF flags.
(cpu_type_list): Add RXv3.
(md_parse_option): Refer elf_flags from cpu_type_list.
(md_show_usage): Add rxv3 and rxv3-dfpu.
(rx_bytesT): Add post byte.
(rx_bfield): New. generate bfmov / bfmovz "imm" field.
(rx_post): New. Set instruction post byte.
(md_assemble): Add post byte.
doc/c-rx.texi: Add cpu types.
* testsuite/gas/rx/Xtod.d: New.
* testsuite/gas/rx/Xtod.sm: New.
* testsuite/gas/rx/bfmov.d: New.
* testsuite/gas/rx/bfmov.sm: New.
* testsuite/gas/rx/dabs.d: New.
* testsuite/gas/rx/dabs.sm: New.
* testsuite/gas/rx/dadd.d: New.
* testsuite/gas/rx/dadd.sm: New.
* testsuite/gas/rx/dcmp.d: New.
* testsuite/gas/rx/dcmp.sm: New.
* testsuite/gas/rx/ddiv.d: New.
* testsuite/gas/rx/ddiv.sm: New.
* testsuite/gas/rx/dmov.d: New.
* testsuite/gas/rx/dmov.sm: New.
* testsuite/gas/rx/dmul.d: New.
* testsuite/gas/rx/dmul.sm: New.
* testsuite/gas/rx/dneg.d: New.
* testsuite/gas/rx/dneg.sm: New.
* testsuite/gas/rx/dpopm.d: New.
* testsuite/gas/rx/dpopm.sm: New.
* testsuite/gas/rx/dpushm.d: New.
* testsuite/gas/rx/dpushm.sm: New.
* testsuite/gas/rx/dround.d: New.
* testsuite/gas/rx/dround.sm: New.
* testsuite/gas/rx/dsqrt.d: New.
* testsuite/gas/rx/dsqrt.sm: New.
* testsuite/gas/rx/dsub.d: New.
* testsuite/gas/rx/dsub.sm: New.
* testsuite/gas/rx/dtoX.d: New.
* testsuite/gas/rx/dtoX.sm: New.
* testsuite/gas/rx/macros.inc: Add double FPU registers.
* testsuite/gas/rx/mvfdc.d: New.
* testsuite/gas/rx/mvfdc.sm: New.
* testsuite/gas/rx/mvfdr.d: New.
* testsuite/gas/rx/mvfdr.sm: New.
* testsuite/gas/rx/mvtdc.d: New.
* testsuite/gas/rx/mvtdc.sm: New.
* testsuite/gas/rx/rstr.d: New.
* testsuite/gas/rx/rstr.sm: New.
* testsuite/gas/rx/rx.exp: Use rxv3-dfpu option.
* testsuite/gas/rx/save.d: New.
* testsuite/gas/rx/save.sm: New.
* testsuite/gas/rx/xor.d: New.
* testsuite/gas/rx/xor.sm: Add pattern.

5 years agoRX: include - Add RXv3 support.
Yoshinori Sato [Tue, 25 Dec 2018 11:44:15 +0000 (20:44 +0900)]
RX: include - Add RXv3 support.

* elf/rx.h (EF_RX_CPU_MASK): Update new bits.
(E_FLAG_RX_V3): New RXv3 type.
* opcode/rx.h (RX_Size): Add double size.
(RX_Operand_Type): Add double FPU registers.
(RX_Opcode_ID): Add new instuctions.

5 years agoGDB crash re-running program on Windows (native)
Joel Brobecker [Sat, 5 Jan 2019 07:55:08 +0000 (11:55 +0400)]
GDB crash re-running program on Windows (native)

Running any program twice on Windows current results in GDB crashing:

    $ gdb -q any_program
    (gdb) run
    $ gdb dummy -batch -ex run -ex run
    [New Thread 684960.0xe5878]
    [New Thread 684960.0xd75ac]
    [New Thread 684960.0xddac8]
    [New Thread 684960.0xc1f50]
    [Thread 684960.0xd75ac exited with code 0]
    [Thread 684960.0xddac8 exited with code 0]
    [Thread 684960.0xc1f50 exited with code 0]
    [Inferior 1 (process 684960) exited normally]
    (gdb) run
    Segmentation fault

The crash happens while processing the CREATE_PROCESS_DEBUG_EVENT
for  the second run; in particular, we have in get_windows_debug_event:

    | case CREATE_PROCESS_DEBUG_EVENT:
    |   [...]
    |   if (main_thread_id)
    |     windows_delete_thread (ptid_t (current_event.dwProcessId, 0,
    |                                    main_thread_id),
    |                            0);

The problem is that main_thread_id is the TID of the main thread from
the *previous* inferior, and this code is trying to delete that
thread. The problem is that it is constructing a PTID by pairing
the TID of the previous inferior with the PID of the new inferior.
As a result, when we dig inside windows_delete_thread to see
how it would handle that, we see...

    | delete_thread (find_thread_ptid (ptid));

Since the PTID is bogus, we end up calling delete_thread with
a NULL thread_info. It used to be harmless, turning the delete_thread
into a nop, but the following change...

    | commit 080363310650c93ad8e93018bcb6760ba5d32d1c
    | Date:   Thu Nov 22 16:09:14 2018 +0000
    | Subject: Per-inferior thread list, thread ranges/iterators, down with ALL_THREADS, etc.

... changed delete_thread to get the list of threads from
the inferior, which itself is now accessed via the given
thread_info. This is the corresponding diff that shows the change:

    | -  for (tp = thread_list; tp; tpprev = tp, tp = tp->next)
    | +  for (tp = thr->inf->thread_list; tp; tpprev = tp, tp = tp->next)

As a result of this, passing a NULL thread_info is no longer
an option!

Stepping back a bit, the reason behind deleting the thread late
could be found in a patch from Dec 2003, which laconically explains:

    | commit 87a45c96062d658ca83b50aa060a648bf5f5f1ff
    | Date:   Fri Dec 26 00:39:04 2003 +0000
    |
    | * win32-nat.c (get_child_debug_event): Keep main thread id around
    | even after thread exits since Windows insists on continuing to
    | report events against it.

A look at the gdb-patches archives did not provide any additional
clues (https://www.sourceware.org/ml/gdb-patches/2003-12/msg00478.html).
It is not clear whether this is still needed or not. This patch
assumes that whatever isue there was, the versions of Windows
we currently support no longer have it.

With that in mind, this commit fixes the issue by deleting the thread
when the inferior sends the exit-process event as opposed to deleting it
later, while starting a new inferior.

This also restores the printing of the thread-exit notification for
the main thread, which was missing before. Looking at the transcript
of the example shown above, we can see 4 thread creation notifications,
and only 3 notifications for thread exits. Now creation and exit
notifications are balanced.

In the handling of EXIT_THREAD_DEBUG_EVENT, the main_thread_id
check is removed because deemed unnecessary: The main thread was
introduced by a CREATE_PROCESS_DEBUG_EVENT, and thus the kernel
is expected to report its death via EXIT_PROCESS_DEBUG_EVENT.

And finally, because the behavior of delete_thread did change
(albeit when getting a value we probably never expected to receive),
this patch also adds a gdb_assert. The purpose is to provide some
immediate information in case there are other callers that mistakenly
call delete_thread with a NULL thread info. This can be useful
information when direct debugging of GDB isn't an option.

gdb/ChangeLog:

* thread.c (delete_thread_1): Add gdb_assert that THR is not
NULL. Initialize tpprev to NULL instead of assigning it
to NULL on the next statement.
* windows-nat.c (windows_delete_thread): Remove check for
main_thread_id before printing thread exit notifications.
(get_windows_debug_event) <EXIT_THREAD_DEBUG_EVENT>:
Remove thread ID check against main_thread_id.
<CREATE_PROCESS_DEBUG_EVENT>: Remove call to
windows_delete_thread.
<EXIT_PROCESS_DEBUG_EVENT>: Add call to windows_delete_thread.

5 years agoAutomatic date update in version.in
GDB Administrator [Sat, 5 Jan 2019 00:00:37 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoUse upper case for more metasyntactic variables in gdb
Tom Tromey [Mon, 24 Dec 2018 21:41:40 +0000 (14:41 -0700)]
Use upper case for more metasyntactic variables in gdb

I noticed a few places where a Usage line in gdb did not use upper
case for metasyntactic variables.  This patch fixes all the instances
I found.

Tested on x86-64 Fedora 28.

gdb/ChangeLog
2019-01-04  Tom Tromey  <tom@tromey.com>

* compile/compile.c (_initialize_compile): Use upper case for
metasyntactic variables.
* symmisc.c (_initialize_symmisc): Use upper case for
metasyntactic variables.
* psymtab.c (_initialize_psymtab): Use upper case for
metasyntactic variables.
* demangle.c (demangle_command): Use upper case for metasyntactic
variables.
(_initialize_demangler): Likewise.
* ax-gdb.c (_initialize_ax_gdb): Use upper case for metasyntactic
variables.

5 years agoFix potential buffer overrun whilst scanning macro strings.
Wu Heng [Fri, 4 Jan 2019 16:18:59 +0000 (16:18 +0000)]
Fix potential buffer overrun whilst scanning macro strings.

PR 24010
* macro.c (get_any_string): Check for end of input whilst scanning
for separators.

5 years agoFix bug parsing angle bracketed string sequences.
Wu Heng [Fri, 4 Jan 2019 15:58:02 +0000 (15:58 +0000)]
Fix bug parsing angle bracketed string sequences.

PR 24009
* read.c (stringer): Fix handling of missing '>' character at end
of <...> sequence.

5 years agoChange the default for the CSKY target to be little endian.
Lifang Xia [Fri, 4 Jan 2019 13:56:46 +0000 (13:56 +0000)]
Change the default for the CSKY target to be little endian.

* config.bfd (csky-*-elf* | csky-*-linux*): Modify the csky
default target, little endian target is more suitable.

5 years agoFix a possible integer overflow problem when examining corrupt binaries using a 32...
Nick Clifton [Fri, 4 Jan 2019 13:44:34 +0000 (13:44 +0000)]
Fix a possible integer overflow problem when examining corrupt binaries using a 32-bit binutil.

PR 24005
* objdump.c (load_specific_debug_section): Check for integer
overflow before attempting to allocate contents.

5 years agoFix ridiculously small memory leak.
Nick Clifton [Fri, 4 Jan 2019 12:11:51 +0000 (12:11 +0000)]
Fix ridiculously small memory leak.

PR 24001
* objcopy.c (copy_object): Free dhandle after writing out the
debug information.
* objdump.c (dump_bfd): Free dhandle after printing out the debug
information.

5 years agoPR24061, powerpc-ibm-aix-ar sets bogus file permissions when extracting
Alan Modra [Fri, 4 Jan 2019 01:48:36 +0000 (12:18 +1030)]
PR24061, powerpc-ibm-aix-ar sets bogus file permissions when extracting

Mode field should be read in octal, all the rest in decimal.  Do so.

PR 24061
PR 21786
* coff-rs6000.c (GET_VALUE_IN_FIELD): Add base parameter and
adjust all callers.
(EQ_VALUE_IN_FIELD): Likewise.
* coff64-rs6000.c (GET_VALUE_IN_FIELD): Likewise.

5 years agoAutomatic date update in version.in
GDB Administrator [Fri, 4 Jan 2019 00:00:29 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoUse xstrdup in tui_set_source_content
Tom Tromey [Wed, 2 Jan 2019 18:41:44 +0000 (11:41 -0700)]
Use xstrdup in tui_set_source_content

valgrind pointed out that the string copy in tui_set_source_content
was not allocating space for the trailing \0:

==3941== Invalid write of size 1
==3941==    at 0x4C3239F: strcpy (vg_replace_strmem.c:512)
==3941==    by 0x72036B: strcpy (string_fortified.h:90)
==3941==    by 0x72036B: tui_set_source_content(symtab*, int, int) (tui-source.c:203)

Looking closer, I don't think there's a need to check the line width
here, so this patch changes it to use xstrdup.

Tested by re-running the TUI under valgrind.  There are still other
valgrind reports from TUI code, but this one is gone.

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* tui/tui-source.c (tui_set_source_content): Use xstrdup.

5 years agoAvoid questionable casts in py-symtab.c
Tom Tromey [Thu, 27 Dec 2018 19:16:06 +0000 (12:16 -0700)]
Avoid questionable casts in py-symtab.c

py-symtab.c has some questionable casts of Py_None to symtab_object*.
This patch avoids these casts by instead using downcasts at the
appropriate places.

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* python/py-symtab.c (salpy_str): Update.
(struct salpy_sal_object) <symtab>: Now a PyObject.
(salpy_dealloc): Update.
(del_objfile_sal): Use gdbpy_ref.

5 years agoImprove reference counting in py-type.c
Tom Tromey [Thu, 27 Dec 2018 19:08:09 +0000 (12:08 -0700)]
Improve reference counting in py-type.c

This improves the reference counting in py-type.c by using gdbpy_ref
and gdbpy_ref::new_reference in more places.

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* python/py-type.c (convert_field): Use new_reference.  Return
gdbpy_ref.
(make_fielditem): Return gdbpy_ref.
(typy_fields): Update.
(typy_getitem): Update.
(field_name): Return gdbpy_ref.  Use new_reference.
(typy_iterator_iternext): Update.

5 years agoSimplify gdbpy_stop_recording
Tom Tromey [Thu, 27 Dec 2018 18:57:28 +0000 (11:57 -0700)]
Simplify gdbpy_stop_recording

This simplifies gdbpy_stop_recording, by having it use Py_RETURN_NONE
rather than writing it out manually, and by usin the idiomatic
GDB_PY_HANDLE_EXCEPTION.

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* python/py-record.c (gdbpy_stop_recording): Use Py_RETURN_NONE.

5 years agoRemove more uses of explicit reference counting in Python
Tom Tromey [Thu, 27 Dec 2018 18:53:20 +0000 (11:53 -0700)]
Remove more uses of explicit reference counting in Python

This changes some more place in the Python code to use gdbpy_ref
rather than explicit reference counting.  While doing this I found a
latent bug in typy_fields_items -- it was not checking for errors in
one spot.  I also changed valpy_dealloc to use Py_XDECREF rather than
an explicit "if".

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* python/py-value.c (valpy_dealloc): Use Py_XDECREF.
* python/py-type.c (typy_fields_items): Use gdbpy_ref.
* python/py-progspace.c (pspy_set_printers): Use gdbpy_ref.
(pspy_set_frame_filters, pspy_set_frame_unwinders)
(pspy_set_type_printers): Likewise.
* python/py-function.c (fnpy_init): Use gdbpy_ref.
* python/py-cmd.c (cmdpy_init): Use gdbpy_ref.
* python/py-objfile.c (objfpy_set_printers): Use gdbpy_ref.
(objfpy_set_frame_filters, objfpy_set_frame_unwinders)
(objfpy_set_type_printers): Likewise.

5 years agoUse a wrapper for PyErr_Fetch
Tom Tromey [Thu, 27 Dec 2018 18:32:01 +0000 (11:32 -0700)]
Use a wrapper for PyErr_Fetch

This introduces a new class that wraps PyErr_Fetch and PyErr_Restore,
and then changes all the callers in gdb to use it.  This reduces the
amount of explicit reference counting that is done in the Python code.
I also found and fixed a latent bug in gdbpy_print_stack -- it was not
correctly checking some error conditions, nor clearing the exception
when needed.

gdb/ChangeLog
2019-01-03  Tom Tromey  <tom@tromey.com>

* python/python.c (gdbpy_enter, ~gdbpy_enter): Update.
(gdbpy_print_stack): Use gdbpy_err_fetch.
* python/python-internal.h (class gdbpy_err_fetch): New class.
(class gdbpy_enter) <m_error_type, m_error_value,
m_error_traceback>: Remove.
<m_error>: New member.
(gdbpy_exception_to_string): Don't declare.
* python/py-varobj.c (py_varobj_iter_next): Use gdbpy_err_fetch.
* python/py-value.c (convert_value_from_python): Use
gdbpy_err_fetch.
* python/py-utils.c (gdbpy_err_fetch::to_string): Rename from
gdbpy_exception_to_string.
(gdbpy_handle_exception): Use gdbpy_err_fetch.
* python/py-prettyprint.c (print_stack_unless_memory_error): Use
gdbpy_err_fetch.

5 years agogdb: Remove cleanup from linux_nat_target::follow_fork
Andrew Burgess [Mon, 31 Dec 2018 22:20:56 +0000 (22:20 +0000)]
gdb: Remove cleanup from linux_nat_target::follow_fork

Remove cleanup from linux_nat_target::follow_fork, instead add a new
unique_ptr specialisation for holding lwp_info pointers and use this
to ensure the pointer is cleaned up when needed.

gdb/ChangeLog:

* linux-nat.c (delete_lwp_cleanup): Delete.
(struct lwp_deleter): New struct.
(lwp_info_up): New typedef.
(linux_nat_target::follow_fork): Delete cleanup, and make use of
lwp_info_up.

5 years agogdb: Remove cleanup from linux-fork.c:inferior_call_waitpid
Andrew Burgess [Mon, 31 Dec 2018 21:23:14 +0000 (21:23 +0000)]
gdb: Remove cleanup from linux-fork.c:inferior_call_waitpid

Replace cleanup in linux-fork.c:inferior_call_waitpid with a RAII
object.

gdb/ChangeLog:

* linux-fork.c (class scoped_switch_fork_info): New class.
(inferior_call_waitpid): Update to use scoped_switch_fork_info.

5 years agogdb: Remove final cleanup from find_overload_match
Andrew Burgess [Mon, 31 Dec 2018 17:44:17 +0000 (17:44 +0000)]
gdb: Remove final cleanup from find_overload_match

This patch removes the setup of a null_cleanup in
valops.c:find_overload_match, and all the calls to do_cleanups.

gdb/ChangeLog:

* valops.c (find_overload_match): Remove use of null_cleanup, and
calls to do_cleanups.

5 years agogdb: Remove a cleanup from find_overload_match
Andrew Burgess [Mon, 31 Dec 2018 17:41:38 +0000 (17:41 +0000)]
gdb: Remove a cleanup from find_overload_match

This patch changes cp-support.c:cp_func_name to return a
'gdb::unique_xmalloc_ptr<char>' instead of a 'char *'.  This allows a
cleanup to be removed from valops.c:find_overload_match.

gdb/ChangeLog:

* compile/compile-cplus-types.c
(compile_cplus_instance::decl_name): Handle changes to
cp_func_name.
* cp-support.c (cp_func_name): Update header comment, update
return type.
* cp-support.h (cp_func_name): Update return type in declaration.
* valops.c (find_overload_match): Move temp_func local to top
level of function and change its type.  Use temp_func to hold and
delete temporary string obtained from cp_func_name.

5 years agogdb/remote: Remove a cleanup in remote_check_symbols
Andrew Burgess [Mon, 31 Dec 2018 14:05:09 +0000 (14:05 +0000)]
gdb/remote: Remove a cleanup in remote_check_symbols

Convert one of the variables that requires a cleanup from a 'char *'
to a 'gdb::char_vector' in remote_target::remote_check_symbols.

Tested on x86-64/Linux with target_board native-gdbserver and
native-extended-gdbserver.

gdb/ChangeLog:

* remote.c (remote_target::remote_check_symbols): Convert `msg` to
gdb::char_vector, remove cleanup, and update uses of `msg`.

5 years agoRISC-V: Fix wrong use of s0 register name.
Jim Wilson [Thu, 3 Jan 2019 19:12:17 +0000 (11:12 -0800)]
RISC-V: Fix wrong use of s0 register name.

s0 is listed as both an int register name and an FP register name.  The FP reg
name is wrong.  This looks like a simple editting error, and has an easy fix.
Tested with riscv64-linux build and check, with no regressions.

gdb/
* riscv-tdep.c (riscv_freg_feature): Drop s0 name from f8.

5 years agoS12Z: opcodes: Separate the decoding of operations from their display.
John Darrington [Thu, 3 Jan 2019 17:30:40 +0000 (18:30 +0100)]
S12Z: opcodes: Separate the decoding of operations from their display.

This change adds an abstraction layer between the decoding of machine
operations and their disassembled textual representation.  This allows
the decoding routines to be re-used for other purposes (at the expense)
of slightly slower running time.

ChangeLog: opcodes/

   * s12z-opc.c: New file.
   * s12z-opc.h: New file.
   * s12z-dis.c: Removed all code not directly related to display
     of instructions.  Used the interface provided by the new files
     instead.
   * Makefile.am (TARGET_LIBOPCODES_CFILES) Add s12z-opc.c.
   * Makefile.in: regenerate.
   * configure.ac (bfd_s12z_arch): Correct the dependencies.
   * configure: regenerate.

5 years agoAdd 'extern C' if simulator is written in C++
Павел Крюков [Mon, 31 Dec 2018 11:53:54 +0000 (14:53 +0300)]
Add 'extern C' if simulator is written in C++

sim/common/Changelog:
2018-12-31  Pavel I. Kryukov  <kryukov@frtk.ru>

        * sim-base.h: Add 'extern C' if header is compiled with C++

5 years agoDon't build readline/libreadline.a, when --with-system-readline is supplied
Дилян Палаузов [Wed, 12 Dec 2018 07:46:03 +0000 (07:46 +0000)]
Don't build readline/libreadline.a, when --with-system-readline is supplied

https://sourceware.org/bugzilla/show_bug.cgi?id=18632

The bundled libreadline is always built, even if the system is
./configure'd --with-system-readline and the build libreadline.a is not
used.

Proposed patch:

Fix ./configure.ac not to proceed readline/, when --with-system-
readline is provided

5 years agoAutomatic date update in version.in
GDB Administrator [Thu, 3 Jan 2019 00:00:34 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoRemove a cleanup from target-descriptions.c
Tom Tromey [Wed, 2 Jan 2019 15:03:13 +0000 (08:03 -0700)]
Remove a cleanup from target-descriptions.c

This removes a cleanup from target-descriptions.c, by changing it to
use a unique_ptr instead.  Note that a deletion adapter is used, even
though target_desc is allocated with new, to avoid moving target_desc
to target-descriptions.h.

gdb/ChangeLog
2019-01-02  Tom Tromey  <tom@tromey.com>

* xml-tdesc.c (xml_cache): Hold a target_desc_up.
(tdesc_parse_xml): Remove cleanups.
* target-descriptions.h (make_cleanup_free_target_description):
Don't declare.
(target_desc_deleter): New struct.
(target_desc_up): New typedef.
* target-descriptions.c (target_desc_deleter::operator()): Rename
from free_target_description.
(make_cleanup_free_target_description): Remove.

5 years agoRemove cleanups from linespec.c
Tom Tromey [Wed, 2 Jan 2019 16:17:43 +0000 (09:17 -0700)]
Remove cleanups from linespec.c

This removes the remaining cleanups from linespec.c.  This adds a
constructor and destructor to linespec_parser, but in a minimal way --
the parser could still benefit from a bit more C++-ification.

gdb/ChangeLog
2019-01-02  Tom Tromey  <tom@tromey.com>

* linespec.c (struct linespec_parser): Rename from ls_parser.  Add
constructor, destructor.
(linespec_parser): Remove typedef.
(~linespec_parser): Rename from linespec_parser_delete.
(linespec_lex_to_end, linespec_complete_label)
(linespec_complete): Update.
(decode_line_full): Remove cleanups.
(decode_line_1): Update.

5 years agoChange inferior_to_inferior_object to return a gdbpy_ref
Tom Tromey [Wed, 2 Jan 2019 16:58:14 +0000 (09:58 -0700)]
Change inferior_to_inferior_object to return a gdbpy_ref

Most callers of inferior_to_inferior_object already use a gdbpy_ref,
so this changes inferior_to_inferior_object to return one.  Doing this
revealed that create_thread_object was not correctly handling the case
where inferior_to_inferior_object failed, so this patch fixes this as
well.

gdb/ChangeLog
2019-01-02  Tom Tromey  <tom@tromey.com>

* python/python-internal.h (inferior_to_inferior_object): Change
return type.
* python/py-exitedevent.c (create_exited_event_object): Update.
* python/py-inferior.c (inferior_to_inferior_object): Return
gdbpy_ref.
(python_new_inferior, python_inferior_deleted)
(thread_to_thread_object, delete_thread_object)
(build_inferior_list, gdbpy_selected_inferior): Update.
* python/py-infthread.c (create_thread_object): Update.  Also fail
if inferior_to_inferior_object fails.

5 years agoPlace displaced step data directly in inferior structure
Simon Marchi [Wed, 2 Jan 2019 22:31:08 +0000 (17:31 -0500)]
Place displaced step data directly in inferior structure

This patch moves the per-inferior data related to displaced stepping to
be directly in the inferior structure, rather than in a container on the
side.

On notable difference is that previously, we deleted the state on
inferior exit, which guaranteed a clean state if re-using the inferior
for a new run or attach.  We now need to reset the state manually.

At the same time, I changed step_saved_copy to be a gdb::byte_vector, so
it is automatically freed on destruction (which should plug the leak
reported here [1]).

[1] https://sourceware.org/ml/gdb-patches/2018-11/msg00202.html

gdb/ChangeLog:

* inferior.h (class inferior) <displaced_step_state>: New field.
* infrun.h (struct displaced_step_state): Move here from
infrun.c.  Initialize fields, add constructor.
<inf>: Remove field.
<reset>: New method.
* infrun.c (struct displaced_step_inferior_state): Move to
infrun.h.
(displaced_step_inferior_states): Remove.
(get_displaced_stepping_state): Adust.
(displaced_step_in_progress_any_inferior): Adjust.
(displaced_step_in_progress_thread): Adjust.
(displaced_step_in_progress): Adjust.
(add_displaced_stepping_state): Remove.
(get_displaced_step_closure_by_addr): Adjust.
(remove_displaced_stepping_state): Remove.
(infrun_inferior_exit): Call displaced_step_state.reset.
(use_displaced_stepping): Don't check for NULL.
(displaced_step_prepare_throw): Call
get_displaced_stepping_state.
(displaced_step_fixup): Don't check for NULL.
(prepare_for_detach): Don't check for NULL.

5 years agoFix leak of struct call_thread_fsm in call_function_by_hand_dummy.
Philippe Waroquiers [Tue, 1 Jan 2019 19:54:52 +0000 (20:54 +0100)]
Fix leak of struct call_thread_fsm in call_function_by_hand_dummy.

When the call does not complete, the call_thread_fsm allocated
by new_call_thread_fsm is not cleaned up and deleted, which causes
the following leak e.g. in gdb.base/callfuncs.exp:

==29263== 560 bytes in 7 blocks are definitely lost in loss record 2,833 of 3,341
==29263==    at 0x4C2E0BC: calloc (vg_replace_malloc.c:762)
==29263==    by 0x405110: xcalloc (common-utils.c:84)
==29263==    by 0x4E67EB: xcnew<call_thread_fsm> (poison.h:122)
==29263==    by 0x4E67EB: new_call_thread_fsm (infcall.c:516)
==29263==    by 0x4E67EB: call_function_by_hand_dummy(value*, type*, gdb::array_view<value*>, void (*)(void*, int), void*) (infcall.c:1154)
==29263==    by 0x4E784E: call_function_by_hand(value*, type*, gdb::array_view<value*>) (infcall.c:693)
==29263==    by 0x496111: eval_call(expression*, noside, int, value**, char const*, type*) [clone .isra.5] (eval.c:835)

Fix the leak by similarly doing cleanup/destroy when restoring
previous state machine.

Tested on debian/amd64, natively and under valgrind.

2019-01-02  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* infcall.c (call_function_by_hand_dummy): cleanup/destroy sm
 in case of call that did not complete.

5 years agoFix search of debug files for remote debuggee
Andrey Utkin [Fri, 9 Nov 2018 19:09:40 +0000 (19:09 +0000)]
Fix search of debug files for remote debuggee

When using remote debugging server, and when debuggee filename is
inferred via qXfer:exec-file:read request, or sysroot starts with
"target:", this "target:" prefix of filepaths is not treated correctly
during debug file search - it appears in the middle of the looked up
paths.

In the following example, unpatched GDB can't find separate debug files for
neither the executable, nor standard libraries:

     $ gdb -ex 'set debug separate-debug-file 1' -ex 'set sysroot target:/' -ex 'set debug-file-directory /usr/lib/debug:/home/j/hide' -ex 'target remote :3333' -ex 'break main' -ex 'continue' -ex 'bt' -ex 'info sharedlibrary' -ex 'set confirm off' -ex 'quit'
    GNU gdb (Gentoo 9999 vanilla) 8.2.50.20181109-git
    Copyright (C) 2018 Free Software Foundation, Inc.
    License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
    This is free software: you are free to change and redistribute it.
    There is NO WARRANTY, to the extent permitted by law.
    Type "show copying" and "show warranty" for details.
    This GDB was configured as "x86_64-pc-linux-gnu".
    Type "show configuration" for configuration details.
    For bug reporting instructions, please see:
    <https://bugs.gentoo.org/>.
    Find the GDB manual and other documentation resources online at:
        <http://www.gnu.org/software/gdb/documentation/>.

    For help, type "help".
    Type "apropos word" to search for commands related to "word".
    Remote debugging using :3333
    Reading /home/j/test from remote target...
    warning: File transfers from remote targets can be slow. Use "set sysroot" to access files locally instead.
    Reading /home/j/test from remote target...
    Reading symbols from target:/home/j/test...

    Looking for separate debug info (debug link) for target:/home/j/test
      Trying target:/home/j/test.debug
    Reading /home/j/test.debug from remote target...
      Trying target:/home/j/.debug/test.debug
    Reading /home/j/.debug/test.debug from remote target...
      Trying /usr/lib/debug/target:/home/j/test.debug
      Trying /home/j/hide/target:/home/j/test.debug
    (No debugging symbols found in target:/home/j/test)
    Reading /lib64/ld-linux-x86-64.so.2 from remote target...
    Reading /lib64/ld-linux-x86-64.so.2 from remote target...
    Reading symbols from target:/lib64/ld-linux-x86-64.so.2...

    Looking for separate debug info (debug link) for target:/lib64/ld-linux-x86-64.so.2
      Trying target:/lib64/ld-2.27.so.debug
    Reading /lib64/ld-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/ld-2.27.so.debug
    Reading /lib64/.debug/ld-2.27.so.debug from remote target...
      Trying /usr/lib/debug/target:/lib64/ld-2.27.so.debug
      Trying /home/j/hide/target:/lib64/ld-2.27.so.debug
    (No debugging symbols found in target:/lib64/ld-linux-x86-64.so.2)

    Looking for separate debug info (build-id) for system-supplied DSO at 0x7ffff7ffa000
      Trying /usr/lib/debug/.build-id/fd/03d584bc1a90ba28be457635a02662c9f9c1f2.debug
      Trying /home/j/hide/.build-id/fd/03d584bc1a90ba28be457635a02662c9f9c1f2.debug
    0x00007ffff7dd7000 in ?? () from target:/lib64/ld-linux-x86-64.so.2
    Breakpoint 1 at 0x4005eb
    Continuing.
    Reading /lib64/libpthread.so.0 from remote target...
    Reading /lib64/libc.so.6 from remote target...

    Looking for separate debug info (debug link) for target:/lib64/libpthread.so.0
      Trying target:/lib64/libpthread-2.27.so.debug
    Reading /lib64/libpthread-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/libpthread-2.27.so.debug
    Reading /lib64/.debug/libpthread-2.27.so.debug from remote target...
      Trying /usr/lib/debug/target:/lib64/libpthread-2.27.so.debug
      Trying /home/j/hide/target:/lib64/libpthread-2.27.so.debug

    Looking for separate debug info (debug link) for target:/lib64/libc.so.6
      Trying target:/lib64/libc-2.27.so.debug
    Reading /lib64/libc-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/libc-2.27.so.debug
    Reading /lib64/.debug/libc-2.27.so.debug from remote target...
      Trying /usr/lib/debug/target:/lib64/libc-2.27.so.debug
      Trying /home/j/hide/target:/lib64/libc-2.27.so.debug

    Breakpoint 1, 0x00000000004005eb in main ()
    #0  0x00000000004005eb in main ()
    From                To                  Syms Read   Shared Object Library
    0x00007ffff7dd6e80  0x00007ffff7df4650  Yes (*)     target:/lib64/ld-linux-x86-64.so.2
    0x00007ffff7bbbb70  0x00007ffff7bcbfee  Yes (*)     target:/lib64/libpthread.so.0
    0x00007ffff780f200  0x00007ffff7962d7c  Yes (*)     target:/lib64/libc.so.6
    (*): Shared library is missing debugging information.

With current fix, the paths used always have target: in the beginning
and this helps to find all debug files:

     $ gdb -ex 'set debug separate-debug-file 1' -ex 'set sysroot target:/' -ex 'set debug-file-directory /usr/lib/debug:/home/j/hide' -ex 'target remote :3333' -ex 'break main' -ex 'continue' -ex 'bt' -ex 'info sharedlibrary' -ex 'set confirm off' -ex 'quit'
    GNU gdb (Gentoo 9999 vanilla) 8.2.50.20181109-git
    Copyright (C) 2018 Free Software Foundation, Inc.
    License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
    This is free software: you are free to change and redistribute it.
    There is NO WARRANTY, to the extent permitted by law.
    Type "show copying" and "show warranty" for details.
    This GDB was configured as "x86_64-pc-linux-gnu".
    Type "show configuration" for configuration details.
    For bug reporting instructions, please see:
    <https://bugs.gentoo.org/>.
    Find the GDB manual and other documentation resources online at:
        <http://www.gnu.org/software/gdb/documentation/>.

    For help, type "help".
    Type "apropos word" to search for commands related to "word".
    Remote debugging using :3333
    Reading /home/j/test from remote target...
    warning: File transfers from remote targets can be slow. Use "set sysroot" to access files locally instead.
    Reading /home/j/test from remote target...
    Reading symbols from target:/home/j/test...

    Looking for separate debug info (debug link) for target:/home/j/test
      Trying target:/home/j/test.debug
    Reading /home/j/test.debug from remote target...
      Trying target:/home/j/.debug/test.debug
    Reading /home/j/.debug/test.debug from remote target...
      Trying target:/usr/lib/debug//home/j/test.debug
    Reading /usr/lib/debug//home/j/test.debug from remote target...
      Trying target:/home/j/hide//home/j/test.debug
    Reading /home/j/hide//home/j/test.debug from remote target...
    Reading /home/j/hide//home/j/test.debug from remote target...
    Reading symbols from target:/home/j/hide//home/j/test.debug...
    Reading /lib64/ld-linux-x86-64.so.2 from remote target...
    Reading /lib64/ld-linux-x86-64.so.2 from remote target...
    Reading symbols from target:/lib64/ld-linux-x86-64.so.2...

    Looking for separate debug info (debug link) for target:/lib64/ld-linux-x86-64.so.2
      Trying target:/lib64/ld-2.27.so.debug
    Reading /lib64/ld-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/ld-2.27.so.debug
    Reading /lib64/.debug/ld-2.27.so.debug from remote target...
      Trying target:/usr/lib/debug//lib64/ld-2.27.so.debug
    Reading /usr/lib/debug//lib64/ld-2.27.so.debug from remote target...
    Reading /usr/lib/debug//lib64/ld-2.27.so.debug from remote target...
    Reading symbols from target:/usr/lib/debug//lib64/ld-2.27.so.debug...

    Looking for separate debug info (build-id) for system-supplied DSO at 0x7ffff7ffa000
      Trying /usr/lib/debug/.build-id/fd/03d584bc1a90ba28be457635a02662c9f9c1f2.debug
      Trying /home/j/hide/.build-id/fd/03d584bc1a90ba28be457635a02662c9f9c1f2.debug
    0x00007ffff7dd7000 in _start () from target:/lib64/ld-linux-x86-64.so.2
    Breakpoint 1 at 0x4005ef: file test/test16.c, line 13.
    Continuing.
    Reading /lib64/libpthread.so.0 from remote target...
    Reading /lib64/libc.so.6 from remote target...

    Looking for separate debug info (debug link) for target:/lib64/libpthread.so.0
      Trying target:/lib64/libpthread-2.27.so.debug
    Reading /lib64/libpthread-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/libpthread-2.27.so.debug
    Reading /lib64/.debug/libpthread-2.27.so.debug from remote target...
      Trying target:/usr/lib/debug//lib64/libpthread-2.27.so.debug
    Reading /usr/lib/debug//lib64/libpthread-2.27.so.debug from remote target...
    Reading /usr/lib/debug//lib64/libpthread-2.27.so.debug from remote target...

    Looking for separate debug info (debug link) for target:/lib64/libc.so.6
      Trying target:/lib64/libc-2.27.so.debug
    Reading /lib64/libc-2.27.so.debug from remote target...
      Trying target:/lib64/.debug/libc-2.27.so.debug
    Reading /lib64/.debug/libc-2.27.so.debug from remote target...
      Trying target:/usr/lib/debug//lib64/libc-2.27.so.debug
    Reading /usr/lib/debug//lib64/libc-2.27.so.debug from remote target...
    Reading /usr/lib/debug//lib64/libc-2.27.so.debug from remote target...

    Breakpoint 1, main () at test/test16.c:13
    13          for ( i=0; i<10; ++i)
    #0  main () at test/test16.c:13
    From                To                  Syms Read   Shared Object Library
    0x00007ffff7dd6e80  0x00007ffff7df4650  Yes         target:/lib64/ld-linux-x86-64.so.2
    0x00007ffff7bbbb70  0x00007ffff7bcbfee  Yes         target:/lib64/libpthread.so.0
    0x00007ffff780f200  0x00007ffff7962d7c  Yes         target:/lib64/libc.so.6

gdb/ChangeLog:

2018-11-09  Andrey Utkin  <autkin@undo.io>

        * symfile.c (find_separate_debug_file): Fix search of debug files for
        remote debuggee.

Signed-off-by: Andrey Utkin <autkin@undo.io>
5 years agoStyle improvements in gdb/python
Tom Tromey [Wed, 2 Jan 2019 16:30:49 +0000 (09:30 -0700)]
Style improvements in gdb/python

This fixes a few minor style issues I found in gdb/python: some
unnecessary casts, the removal of an unnecessary local variable, and
one instance of incorrect formatting.

Tested by rebuilding and re-running gdb.python.

gdb/ChangeLog
2019-01-02  Tom Tromey  <tom@tromey.com>

* python/py-inferior.c (gdbpy_initialize_inferior): Fix
indentation.
* python/py-frame.c (frapy_older): Remove cast.
(frapy_newer): Likewise.
* python/py-breakpoint.c (local_setattro): Remove cast.
* python/py-arch.c (archpy_name): Remove local variable.
* python/py-type.c (gdbpy_lookup_type): Remove cast.

5 years agoFix year range in unittests/.../char/empty.cc copyright header
Joel Brobecker [Wed, 2 Jan 2019 06:11:41 +0000 (10:11 +0400)]
Fix year range in unittests/.../char/empty.cc copyright header

The year range in the copyright header of that file was incorrect:

    // Copyright (C) 3 Free Software Foundation, Inc.

Tracing the origin of this file down, I found that it was copied
from GCC's libstdc++-v3 testsuite:

    Adapt and integrate string_view tests
    https://www.sourceware.org/ml/gdb-patches/2018-04/msg00113.html

Looking at the version in GCC, I found the same issue, so sent
a fix there:

    https://gcc.gnu.org/ml/gcc-patches/2019-01/msg00000.html

Now that the fix is in GCC, this commit applies the same fix
to our copy.

gdb/ChangeLog:

* unittests/basic_string_view/element_access/char/empty.cc:
Fix year range in copyright header.

5 years agoAutomatic date update in version.in
GDB Administrator [Wed, 2 Jan 2019 00:00:16 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agogdb/riscv: Split ISA and ABI features
Andrew Burgess [Thu, 13 Dec 2018 17:59:12 +0000 (17:59 +0000)]
gdb/riscv: Split ISA and ABI features

The goal of this commit is to allow RV64 binaries compiled for the 'F'
extension to run on a target that supports both the 'F' and 'D'
extensions.

The 'D' extension depends on the 'F' extension and chapter 9 of the
RISC-V ISA manual implies that running a program compiled for 'F' on
a 'D' target should be fine.

To support this the gdbarch now holds two feature sets, one represents
the features that are present on the target, and one represents the
features requested in the ELF flags.

The existing error checks are relaxed slightly to allow binaries
compiled for 32-bit 'F' extension to run on targets with the 64-bit
'D' extension.

A new set of functions called riscv_abi_{xlen,flen} are added to
compliment the existing riscv_isa_{xlen,flen}, and some callers to the
isa functions now call the abi functions when that is appropriate.

In riscv_call_arg_struct two asserts are removed, these asserts no
longer make sense.  The asserts were both like this:

    gdb_assert (TYPE_LENGTH (ainfo->type)
                <= (cinfo->flen + cinfo->xlen));

And were made in two cases, when passing structures like these:

   struct {
     integer field1;
     float   field2;
   };

or,

   struct {
     float   field1;
     integer field2;
   };

When running on an RV64 target which only has 32-bit float then the
integer field could be 64-bits, while if the float field is 32-bits
the overall size of the structure can be 128-bits (with 32-bits of
padding).  In this case the assertion would fail, however, the code
isn't incorrect, so its safe to just remove the assertion.

This was tested by running on an RV64IMFDC target using a compiler
configured for RV64IMFC, and comparing the results with those obtained
when using a compiler configured for RV64IMFDC.  The only regressions
I see (now) are in gdb.base/store.exp and are related too different
code generation choices GCC makes between the two targets.

Finally, this commit does not make any attempt to support running
binaries compiled for RV32 on an RV64 target, though nothing in here
should prevent that being supported in the future.

gdb/ChangeLog:

* arch/riscv.h (struct riscv_gdbarch_features) <hw_float_abi>:
Delete.
<operator==>: Update with for removed field.
<hash>: Likewise.
* riscv-tdep.h (struct gdbarch_tdep) <features>: Renamed to...
<isa_features>: ...this.
<abi_features>: New field.
(riscv_isa_flen): Update comment.
(riscv_abi_xlen): New declaration.
(riscv_abi_flen): New declaration.
* riscv-tdep.c (riscv_isa_xlen): Update to get answer from
isa_features.
(riscv_abi_xlen): New function.
(riscv_isa_flen): Update to get answer from isa_features.
(riscv_abi_flen): New function.
(riscv_has_fp_abi): Update to get answer from abi_features.
(riscv_call_info::riscv_call_info): Use abi xlen and flen, not isa
xlen and flen.
(riscv_call_info) <xlen, flen>: Update comment.
(riscv_call_arg_struct): Remove invalid assertions
(riscv_features_from_gdbarch_info): Update now hw_float_abi field
is removed.
(riscv_gdbarch_init): Gather isa features and abi features
separately, ensure both match on the gdbarch when reusing an old
gdbarch.  Relax an error check to allow 32-bit abi float to run on
a target with 64-bit float hardware.

5 years agoFix 'Invalid read of size 4' in search_command_helper
Philippe Waroquiers [Tue, 1 Jan 2019 14:45:09 +0000 (15:45 +0100)]
Fix 'Invalid read of size 4' in search_command_helper

Valgrind detects the below error in gdb.base/list.exp.
==14763== Invalid read of size 4
==14763==    at 0x60B584: search_command_helper(char const*, int, bool) [clone .constprop.91] (source.c:1601)
==14763==    by 0x408888: cmd_func(cmd_list_element*, char const*, int) (cli-decode.c:1892)
==14763==    by 0x668550: execute_command(char const*, int) (top.c:630)
==14763==    by 0x4B2F7B: command_handler(char const*) (event-top.c:583)
==14763==    by 0x4B326C: command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) (event-top.c:772)
...
==14763==  Address 0x6d9f09c is 4 bytes before a block of size 156 alloc'd
==14763==    at 0x4C2E2B3: realloc (vg_replace_malloc.c:836)
==14763==    by 0x41904C: xrealloc (common-utils.c:62)
==14763==    by 0x60A300: find_source_lines(symtab*, int) (source.c:1203)
==14763==    by 0x608219: source_cache::get_plain_source_lines(symtab*, int, int, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*) (source-cache.c:51)
==14763==    by 0x60A46B: print_source_lines_base(symtab*, int, int, enum_flags<print_source_lines_flag>) (source.c:1350)
==14763==    by 0x404E2D: list_command(char const*, int) (cli-cmds.c:1080)
....

Add the missing condition to end the loop once line 1 has been
reversed-searched.

gdb/ChangeLog
2019-01-01  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* source.c (search_command_helper): Stop reverse search
when line 1 has been searched.

5 years agoFix leak in record-full.c
Philippe Waroquiers [Tue, 1 Jan 2019 13:12:30 +0000 (14:12 +0100)]
Fix leak in record-full.c

valgrind detects leaks in several gdb.reverse tests,
such as the below in gdb.reverse/watch-precsave.exp.

Fix the leak by rewriting the loop that frees
record_full_core_buf_list.

gdb/ChangeLog
2019-01-01  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* record-full.c (record_full_base_target::close): Rewrite
record_full_core_buf_list free logic.

==18847== VALGRIND_GDB_ERROR_BEGIN
==18847== 4,120 (24 direct, 4,096 indirect) bytes in 1 blocks are definitely lost in loss record 3,094 of 3,199
==18847==    at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==18847==    by 0x405097: xmalloc (common-utils.c:44)
==18847==    by 0x5AF8EA: xnew<record_full_core_buf_entry> (poison.h:110)
==18847==    by 0x5AF8EA: record_full_core_target::xfer_partial(target_object, char const*, unsigned char*, unsigned char const*, unsigned long, unsigned long, unsigned long*) (record-full.c:2182)
==18847==    by 0x64677D: raw_memory_xfer_partial(target_ops*, unsigned char*, unsigned char const*, unsigned long, long, unsigned long*) (target.c:956)
==18847==    by 0x64691E: memory_xfer_partial_1(target_ops*, target_object, unsigned char*, unsigned char const*, unsigned long, unsigned long, unsigned long*) (target.c:1086)

5 years agoFix leak in print_one_catch_syscall.
Philippe Waroquiers [Mon, 31 Dec 2018 17:07:26 +0000 (18:07 +0100)]
Fix leak in print_one_catch_syscall.

The last text produced was not freed, causing the below leak
(e.g. in gdb.base/catch-syscall.exp):

==24970== 56 bytes in 12 blocks are definitely lost in loss record 626 of 3,289
==24970==    at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==24970==    by 0x66B9C3F: __vasprintf_chk (vasprintf_chk.c:80)
==24970==    by 0x405181: vasprintf (stdio2.h:210)
==24970==    by 0x405181: xstrvprintf(char const*, __va_list_tag*) (common-utils.c:122)
==24970==    by 0x40524B: xstrprintf(char const*, ...) (common-utils.c:113)
==24970==    by 0x3B49DB: print_one_catch_syscall(breakpoint*, bp_location**) (break-catch-syscall.c:275)
==24970==    by 0x3C698F: print_one_breakpoint_location(breakpoint*, bp_location*, int, bp_location**, int) (breakpoint.c:6076)
==24970==    by 0x3C75B1: print_one_breakpoint(breakpoint*, bp_location**, int) (breakpoint.c:6373)
==24970==    by 0x3C7D0E: breakpoint_1(char const*, int, int (*)(breakpoint const*)) (breakpoint.c:6571)
==24970==    by 0x3C822C: info_breakpoints_command(char const*, int) (breakpoint.c:6625)

2019-01-01  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* break-catch-syscall.c (print_one_catch_syscall): xfree
the last text.

5 years agoUpdate year range in copyright notice of binutils files
Alan Modra [Tue, 1 Jan 2019 10:31:27 +0000 (21:01 +1030)]
Update year range in copyright notice of binutils files

5 years agoChangeLog rotation
Alan Modra [Tue, 1 Jan 2019 10:53:15 +0000 (21:23 +1030)]
ChangeLog rotation

5 years agoupdate copyright year printed by gdb, gdbserver and gdbreplay
Joel Brobecker [Tue, 1 Jan 2019 06:09:59 +0000 (10:09 +0400)]
update copyright year printed by gdb, gdbserver and gdbreplay

gdb/ChangeLog:

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

gdb/gdbserver/ChangeLog:

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

5 years agoUpdate copyright year range in all GDB files.
Joel Brobecker [Tue, 1 Jan 2019 06:01:51 +0000 (10:01 +0400)]
Update copyright year range in all GDB files.

This commit applies all changes made after running the gdb/copyright.py
script.

Note that one file was flagged by the script, due to an invalid
copyright header
(gdb/unittests/basic_string_view/element_access/char/empty.cc).
As the file was copied from GCC's libstdc++-v3 testsuite, this commit
leaves this file untouched for the time being; a patch to fix the header
was sent to gcc-patches first.

gdb/ChangeLog:

Update copyright year range in all GDB files.

5 years agorotate gdb/ChangeLog
Joel Brobecker [Tue, 1 Jan 2019 05:59:27 +0000 (09:59 +0400)]
rotate gdb/ChangeLog

5 years agoAutomatic date update in version.in
GDB Administrator [Tue, 1 Jan 2019 00:00:27 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoUse gdb::unique_xmalloc_ptr<char> in command_line_input to fix a leak
Philippe Waroquiers [Sun, 30 Dec 2018 19:41:49 +0000 (20:41 +0100)]
Use gdb::unique_xmalloc_ptr<char> in command_line_input to fix a leak

Following the change of logic where the input_handler gets a
gdb::unique_xmalloc_ptr<char>, a call to readline directly
followed by a call to handle_line_of_input is missing a free,
and causes the below leak.

Use gdb::unique_xmalloc_ptr<char> to solve the leak.

==16291== VALGRIND_GDB_ERROR_BEGIN
==16291== 64 bytes in 1 blocks are definitely lost in loss record 1,815 of 4,111
==16291==    at 0x4C2E2B3: realloc (vg_replace_malloc.c:836)
==16291==    by 0x41EB1C: xrealloc (common-utils.c:62)
==16291==    by 0x41DBD3: buffer_grow(buffer*, char const*, unsigned long) [clone .part.1] (buffer.c:40)
==16291==    by 0x66E8FF: buffer_grow_char (buffer.h:40)
==16291==    by 0x66E8FF: gdb_readline_no_editing (top.c:798)
==16291==    by 0x66E8FF: command_line_input(char const*, char const*) (top.c:1249)
==16291==    by 0x66EBD8: read_command_file(_IO_FILE*) (top.c:421)
==16291==    by 0x412C0C: script_from_file(_IO_FILE*, char const*) (cli-script.c:1547)
==16291==    by 0x40BE90: source_script_from_stream (cli-cmds.c:569)
==16291==    by 0x40BE90: source_script_with_search(char const*, int, int) (cli-cmds.c:606)
==16291==    by 0x54D567: catch_command_errors(void (*)(char const*, int), char const*, int) (main.c:379)
==16291==    by 0x54EA84: captured_main_1 (main.c:994)
==16291==    by 0x54EA84: captured_main (main.c:1167)
==16291==    by 0x54EA84: gdb_main(captured_main_args*) (main.c:1193)
==16291==    by 0x29DA27: main (gdb.c:32)
==16291==
==16291== VALGRIND_GDB_ERROR_END

gdb/ChangeLog
2018-12-31  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* top.c (command_line_input): Use unique_xmalloc_ptr to
manage memory allocated by readline.

5 years agoPR24042, Global-buffer-overflow problem in output_rel_find
Alan Modra [Mon, 31 Dec 2018 06:36:25 +0000 (17:06 +1030)]
PR24042, Global-buffer-overflow problem in output_rel_find

place_orphan handled ELF SHT_REL/SHT_RELA specially, output_rel_find
didn't.  This mismatch was a bug and also meant it was possible to
craft an object where ld accessed section->name out of bounds.

PR 24042
* emultempl/elf32.em (output_rel_find): Drop "sec" param.  Add
"rela".
(gld${EMULATION_NAME}_place_orphan): Use sh_type to calculate
"rela" param of output_rel_find when ELF.  Tidy uses of elfinput.

5 years agoPR24041, Invalid Memory Address Dereference in elf_link_add_object_symbols
Alan Modra [Mon, 31 Dec 2018 05:10:08 +0000 (15:40 +1030)]
PR24041, Invalid Memory Address Dereference in elf_link_add_object_symbols

PR 24041
* elflink.c (elf_link_add_object_symbols): Don't segfault on
crafted ET_DYN with no program headers.

5 years ago[PowerPC64] Nop out ld 2,24(1) after old-style __tls_get_addr
Alan Modra [Mon, 31 Dec 2018 01:41:42 +0000 (12:11 +1030)]
[PowerPC64] Nop out ld 2,24(1) after old-style __tls_get_addr

When optimising inline plt calls to __tls_get_addr without tls marker
relocs, ld should zap any toc restore insn after the bctrl, to stop a
load-hit-store stall.

* elf64-ppc.c (ppc64_elf_relocate_section <tls_ldgd_opt>): When
editing an old-style __tls_get_addr call, replace a toc restore
insn with a nop.

5 years agoAutomatic date update in version.in
GDB Administrator [Mon, 31 Dec 2018 00:00:18 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoChange input_handler to take a unique_xmalloc_ptr
Tom Tromey [Sat, 29 Dec 2018 19:42:18 +0000 (12:42 -0700)]
Change input_handler to take a unique_xmalloc_ptr

This changes ui::input_handler to take a unique_xmalloc_ptr.  This
clarifies the ownership transfer of input_handler's argument.

gdb/ChangeLog
2018-12-30  Tom Tromey  <tom@tromey.com>

* event-top.h (command_line_handler): Update.
* top.c (class gdb_readline_wrapper_cleanup) <m_handler_orig>:
Update.
(gdb_readline_wrapper_line): Update.
* top.h (struct ui) <input_handler>: Take a unique_xmalloc_ptr.
(handle_line_of_input): Update.
* event-top.c: Update.
(gdb_readline_no_editing_callback): Update.
(command_line_handler): Take a unique_xmalloc_ptr.
(handle_line_of_input): Take a const char *.
(command_line_append_input_line): Take a const char *.

5 years agoFix 'help set/show style' strange layouts/results.
Philippe Waroquiers [Sat, 29 Dec 2018 13:21:31 +0000 (14:21 +0100)]
Fix 'help set/show style' strange layouts/results.

The layout for 'help set address|variable' is strange, e.g.:
  (gdb) help set style address
  style address

  List of show Address display styling
  Configure address colors and display intensity subcommands:

  show Address display styling
  Configure address colors and display intensity background -- Set the background color for this property
  show Address display styling
  Configure address colors and display intensity foreground -- Set the foreground color for this property
  show Address display styling
  Configure address colors and display intensity intensity -- Set the display intensity color for this property

  Type "help show Address display styling
  Configure address colors and display intensity" followed by show Address display styling
  Configure address colors and display intensity subcommand name for full documentation.
  Type "apropos word" to search for commands related to "word".
  Command name abbreviations are allowed if unambiguous.
  (gdb)

The help for 'set style function|filename' gives help for 'Show':
  (gdb) help set style filename
  Filename display styling
  Configure filename colors and display intensity.

  List of show style filename subcommands:

  show style filename background -- Set the background color for this property
  show style filename foreground -- Set the foreground color for this property
  show style filename intensity -- Set the display intensity color for this property

The help for 'show style function|filename' is equally strange, as it speaks
about commands, instead of sub commands:
  (gdb) help show style filename
  Filename display styling
  Configure filename colors and display intensity.

  List of commands:

  background -- Show the background color for this property
  foreground -- Show the foreground color for this property
  intensity -- Show the display intensity color for this property

  Type "help" followed by command name for full documentation.
  Type "apropos word" to search for commands related to "word".
  Command name abbreviations are allowed if unambiguous.
  (gdb)

This patch fixes all this.

Note that the 'set style' and 'show style' have the same prefix_doc:
  (gdb) help show style
  Style-specific settings
  Configure various style-related variables, such as colors
  ...
  (gdb) help set style
  Style-specific settings
  Configure various style-related variables, such as colors
  ...

Other similar commands (such as set|show history) have typically
a more specific prefix:
  (gdb) help show history
  Generic command for showing command history parameters.
  ...
  (gdb) help set history
  Generic command for setting command history parameters.
  ...

This could be fixed by having set_prefix_doc and show_prefix_doc instead of
the single prefix_doc argument to cli_style_option::add_setshow_commands.
That could be improved if deemed better.

2018-12-29  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* cli/cli-style.c (cli_style_option::add_setshow_commands):
Initialize m_set_prefix with "set", instead of re-assigning
m_show_prefix.  Use m_set_prefix for set_list and m_show_prefix
for show_list.
(_initialize_cli_style): Correct the order of arguments in
variable_name_style.add_setshow_commands and
address_style.add_setshow_commands calls.

5 years agoAutomatic date update in version.in
GDB Administrator [Sun, 30 Dec 2018 00:00:54 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoFix the build when GNU Source Highlight is not available
Tom Tromey [Sat, 29 Dec 2018 04:07:24 +0000 (21:07 -0700)]
Fix the build when GNU Source Highlight is not available

The builder pointed out that, when GNU Source Highlight is not
available, get_language_name is not used.  This patch makes it
conditional, fixing the build problem.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* source-cache.c (get_language_name): Conditionally compile.

5 years agoAutomatic date update in version.in
GDB Administrator [Sat, 29 Dec 2018 00:00:35 +0000 (00:00 +0000)]
Automatic date update in version.in

5 years agoFix a crash in jit.c
Tom Tromey [Wed, 19 Dec 2018 18:44:28 +0000 (11:44 -0700)]
Fix a crash in jit.c

A user at Mozilla pointed out a crash in jit.c.  In his situation, an
inferior using the JIT API exec'd an executable that did not use it.
This caused an assertion failure when jit.c:free_objfile_data called
delete_breakpoint with NULL.

This patch fixes the problem in the obvious way.  New test case
included.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* jit.c (free_objfile_data): Only delete breakpoint if non-null.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>
    Simon Marchi <simark@simark.ca>

* gdb.base/jit-exec.exp: New file.
* gdb.base/jit-exec.c: New file.
* gdb.base/jit-execd.c: New file.

5 years agoDocument the "set style" commands
Tom Tromey [Mon, 26 Nov 2018 23:19:17 +0000 (16:19 -0700)]
Document the "set style" commands

This documents the new "set style" commands.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* NEWS: Mention terminal styling.

gdb/doc/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.texinfo (Output Styling): New node.

5 years agoHighlight source code using GNU Source Highlight
Tom Tromey [Wed, 10 Oct 2018 04:21:05 +0000 (22:21 -0600)]
Highlight source code using GNU Source Highlight

This changes gdb to highlight source using GNU Source Highlight, if it
is available.

This affects the output of the "list" command and also the TUI source
window.

No new test because I didn't see a way to make it work when Source
Highlight is not found.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* utils.h (can_emit_style_escape): Declare.
* utils.c (can_emit_style_escape): No longer static.
* cli/cli-style.c (set_style_enabled): New function.
(_initialize_cli_style): Use it.
* tui/tui-winsource.c (tui_show_source_line): Use tui_puts.
(tui_alloc_source_buffer): Change how source lines are allocated.
* tui/tui-source.c (copy_source_line): New function.
(tui_set_source_content): Use source cache.
* tui/tui-io.h (tui_puts): Update.
* tui/tui-io.c (tui_puts_internal): Add window parameter.
(tui_puts): Likewise.
(tui_redisplay_readline): Update.
* tui/tui-data.c (free_content_elements): Change how source window
contents are freed.
* source.c (forget_cached_source_info): Clear the source cache.
(print_source_lines_base): Use the source cache.
* source-cache.h: New file.
* source-cache.c: New file.
* configure.ac: Check for GNU Source Highlight library.
* configure: Update.
* config.in: Update.
* Makefile.in (SRCHIGH_LIBS, SRCHIGH_CFLAGS): New variables.
(INTERNAL_CFLAGS_BASE): Add SRCHIGH_CFLAGS.
(CLIBS): Add SRCHIGH_LIBS.
(COMMON_SFILES): Add source-cache.c.
(HFILES_NO_SRCDIR): Add source-cache.h.

5 years agoUse wclrtoeol in tui_show_source_line
Tom Tromey [Tue, 9 Oct 2018 20:16:07 +0000 (14:16 -0600)]
Use wclrtoeol in tui_show_source_line

This changes tui_show_source_line to use wclrtoeol rather than
manually emitting a sequence of spaces.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* tui/tui-winsource.c (tui_show_source_line): Use wclrtoeol.

5 years agoMake ANSI terminal escape sequences work in TUI
Tom Tromey [Mon, 3 Sep 2018 18:42:59 +0000 (12:42 -0600)]
Make ANSI terminal escape sequences work in TUI

PR tui/14126 notes that ANSI terminal escape sequences don't affect
the colors shown in the TUI.  A simple way to see this is to try the
extended-prompt example from the gdb manual.

Curses does not pass escape sequences through to the terminal.
Instead, it replaces non-printable characters with a visible
representation, for example "^[" for the ESC character.

This patch fixes the problem by adding a simple ANSI terminal sequence
parser to gdb.  These sequences are decoded and those that are
recognized are turned into the appropriate curses calls.

The curses approach to color handling is unusual and so there are some
oddities in the implementation.

Standard curses has no notion of the default colors of the terminal.
So, if you set the foreground color, it is not possible to reset it --
you have to pick some other color.  ncurses provides an extension to
handle this, so this patch updates configure and uses it when
available.

Second, in curses, colors always come in pairs: you cannot set just
the foreground.  This patch handles this by tracking actually-used
pairs of colors and keeping a table of these for reuse.

Third, there are a limited number of such pairs available.  In this
patch, if you try to use too many color combinations, gdb will just
ignore some color changes.

Finally, in addition to limiting the number of color pairs, curses
also limits the number of colors.  This means that, when using
extended 8- or 24-bit color sequences, it may be possible to exhaust
the curses color table.

I am very sour on the curses design now.

I do not know how to write a test for this, so I did not.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

PR tui/14126:
* tui/tui.c (tui_enable): Call start_color and
use_default_colors.
* tui/tui-io.c (struct color_pair): New.
(color_pair_map, last_color_pair, last_style): New globals.
(tui_setup_io): Clean up color map when shutting down.
(curses_colors): New constant.
(get_color_pair, apply_ansi_escape): New functions.
(tui_write): Rewrite.
(tui_puts_internal): New function, from tui_puts.  Add "height"
parameter.
(tui_puts): Use tui_puts_internal.
(tui_redisplay_readline): Use tui_puts_internal.
(_initialize_tui_io): New function.
(color_map): New globals.
(get_color): New function.
* configure.ac: Check for use_default_colors.
* config.in, configure: Rebuild.

5 years agoStyle addresses
Tom Tromey [Fri, 19 Oct 2018 22:22:35 +0000 (16:22 -0600)]
Style addresses

This changes gdb to style addresses.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* ui-out.h (enum class ui_out_style_kind) <ADDRESS>: New
constant.
* ui-out.c (ui_out::field_core_addr): Add styling.
* stack.c (print_frame): Add styling.
* printcmd.c (print_address): Add styling.
(print_address_demangle, info_address_command): Likewise.
* cli/cli-style.h (address_style): Declare.
* cli/cli-style.c (address_style): New global.
(_initialize_cli_style): Register new commands.
* cli-out.c (cli_ui_out::do_field_string): Update.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Update test to check for address styling.

5 years agoStyle the "Reading symbols" message
Tom Tromey [Thu, 6 Sep 2018 20:49:39 +0000 (14:49 -0600)]
Style the "Reading symbols" message

The "Reading symbols" message does not use ui-out (perhaps it
should?), so this styles it using the low-level API.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* symfile.c (symbol_file_add_with_addrs): Style file name.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Add test for styling of "Reading symbols"
message.

5 years agoStyle the gdb welcome message
Tom Tromey [Thu, 6 Sep 2018 20:44:17 +0000 (14:44 -0600)]
Style the gdb welcome message

This changes gdb to style the welcome message that is shown by
default.  The styling is only done interactively.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* top.c (print_gdb_version): Style gdb version number.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Add test for version number styling.

5 years agoStyle print_address_symbolic
Tom Tromey [Thu, 6 Sep 2018 05:39:34 +0000 (23:39 -0600)]
Style print_address_symbolic

print_address_symbolic does not use ui-out, so it did not style
function names.  This patch changes it to use the low-level style code
directly.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* printcmd.c (print_address_symbolic): Style function name.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Add test for print_address_symbolic.

5 years agoStyle locations when setting a breakpoint
Tom Tromey [Thu, 6 Sep 2018 05:32:16 +0000 (23:32 -0600)]
Style locations when setting a breakpoint

say_where does not use ui-out, so function and file names printed by
it were not styled.  This patch changes say_where to use the low-level
style code directly.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* breakpoint.c (say_where): Style file name.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Add test for breakpoint setting.

5 years agoStyle variable names
Tom Tromey [Wed, 5 Sep 2018 18:12:19 +0000 (12:12 -0600)]
Style variable names

This adds style support for variable names.  For the time being, this
is only done in backtraces, not in ptype or print; those places do not
use ui-out and so would need ad hoc changes.

This also adds styling to the names printed for local variables in
"backtrace full".  This code does not use ui-out, so the styling is
done using the low-level API.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* ui-out.h (enum class ui_out_style_kind) <VARIABLE>: New global.
* stack.c (print_frame_arg): Style name.
* printcmd.c (print_variable_and_value): Style variable name.
* cli/cli-style.h (variable_name_style): Declare.
* cli/cli-style.c (variable_name_style): New global.
(_initialize_cli_style): Update.
* cli-out.c (cli_ui_out::do_field_string): Update.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* gdb.base/style.exp: Add test for variable names.

5 years agoReset terminal styles
Tom Tromey [Sat, 17 Nov 2018 18:49:25 +0000 (11:49 -0700)]
Reset terminal styles

This adds a function that can be used to reset terminal styles,
regardless of what style the low-level output routines currently think
is applied.

This is used to make "echo" and "printf" work properly when emitting
ANSI terminal escapes -- now gdb will reset the style at the end of
the command.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* utils.h (reset_terminal_style): Declare.
* utils.c (can_emit_style_escape): New function.
(set_output_style): Use it.
(reset_terminal_style): New function.
* printcmd.c (printf_command): Call reset_terminal_style.
* cli/cli-cmds.c (echo_command): Call reset_terminal_style.

5 years agoAdd output styles to gdb
Tom Tromey [Tue, 4 Sep 2018 04:56:33 +0000 (22:56 -0600)]
Add output styles to gdb

This adds some output styling to the CLI.

A style is currently a foreground color, a background color, and an
intensity (dim or bold).  (This list could be expanded depending on
terminal capabilities.)

A style can be applied while printing.  For ui-out, this is done by
passing the style constant as an argument.  For low-level cases,
fprintf_styled and fputs_styled are provided.

Users can control the style via a number of new set/show commands.  In
the interest of not typing many nearly-identical documentation
strings, I automated this.  On the down side, this is not very
i18n-friendly.

I've chose some default colors to use.  I think it would be good to
enable this by default, so that when users start the new gdb, they
will see the new feature.

Stylizing is done if TERM is set and is not "dumb".  This could be
improved when the TUI is available by using the curses has_colors
call.  That is, the lowest layer could call this without committing to
using curses everywhere; see my other patch for TUI colorizing.

I considered adding a new "set_style" method to ui_file.  However,
because the implementation had to interact with the pager code, I
didn't take this approach.  But, one idea might be to put the isatty
check there and then have it defer to the lower layers.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* utils.h (set_output_style, fprintf_styled)
(fputs_styled): Declare.
* utils.c (applied_style, desired_style): New globals.
(emit_style_escape, set_output_style): New function.
(prompt_for_continue): Emit style escapes.
(fputs_maybe_filtered): Likewise.
(fputs_styled, fprintf_styled): New functions.
* ui-out.h (enum class ui_out_style_kind): New.
(class ui_out) <field_string, field_stream, do_field_string>: Add
style parameter.
* ui-out.c (ui_out::field_stream, ui_out::field_string): Add style
parameter.
* tui/tui-out.h (class tui_ui_out) <do_field_string>: Add style
parameter.
* tui/tui-out.c (tui_ui_out::do_field_string): Add style
parameter.
(tui_ui_out::do_field_string): Update.
* tracepoint.c (print_one_static_tracepoint_marker): Style
output.
* stack.c (print_frame_info, print_frame): Style output.
* source.c (print_source_lines_base): Style output.
* skip.c (info_skip_command): Style output.
* record-btrace.c (btrace_call_history_src_line): Style output.
(btrace_call_history): Likewise.
* python/py-framefilter.c (py_print_frame): Style output.
* mi/mi-out.h (class mi_ui_out) <do_field_string>: Add style
parameter.
* mi/mi-out.c (mi_ui_out::do_table_header)
(mi_ui_out::do_field_int): Update.
(mi_ui_out::do_field_string): Update.
* disasm.c (gdb_pretty_print_disassembler::pretty_print_insn):
Style output.
* cli/cli-style.h: New file.
* cli/cli-style.c: New file.
* cli-out.h (class cli_ui_out) <do_field_string>: Add style
parameter.
* cli-out.c (cli_ui_out::do_table_header)
(cli_ui_out::do_field_int, cli_ui_out::do_field_skip): Update.
(cli_ui_out::do_field_string): Add style parameter.  Style the
output.
* breakpoint.c (print_breakpoint_location): Style output.
(update_static_tracepoint): Likewise.
* Makefile.in (SUBDIR_CLI_SRCS): Add cli-style.c.
(HFILES_NO_SRCDIR): Add cli-style.h.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

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

5 years agoChange gdb test suite's TERM setting
Tom Tromey [Tue, 13 Nov 2018 18:59:03 +0000 (11:59 -0700)]
Change gdb test suite's TERM setting

This changes the gdb test suite to set TERM to "dumb" by default.
This setting disables terminal styling, so that the existing tests do
not need to be updated.

gdb/testsuite/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* lib/gdb.exp (gdb_init): Set the TERM environment variable to
"dumb".
* gdb.base/readline.exp (operate_and_get_next): Save and restore
the TERM environment variable.

5 years agoIntroduce ui_file_style
Tom Tromey [Sat, 10 Nov 2018 00:29:50 +0000 (17:29 -0700)]
Introduce ui_file_style

This introduces the new ui_file_style class and various helpers.  This
class represents a terminal style and provides methods for parsing and
emitting the corresponding ANSI terminal escape sequences.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* unittests/style-selftests.c: New file.
* ui-style.c: New file.
* ui-style.h: New file.
* ui-file.h: Include ui-style.h.
* Makefile.in (COMMON_SFILES): Add ui-style.c.
(HFILES_NO_SRCDIR): Add ui-style.h.
(SUBDIR_UNITTESTS_SRCS): Add style-selftests.c.

5 years agoAdd a "context" argument to add_setshow_enum_cmd
Tom Tromey [Thu, 6 Sep 2018 20:03:38 +0000 (14:03 -0600)]
Add a "context" argument to add_setshow_enum_cmd

This adds a "context" argument to add_setshow_enum_cmd.  Now
add_setshow_enum_cmd will call set_cmd_context on both of the new
commands.  This is used in a later patch.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* command.h (add_setshow_enum_cmd): Add "context" argument.
* cli/cli-decode.c (add_setshow_enum_cmd): Add "context"
argument.  Call set_cmd_context.

5 years agoChange wrap buffering to use a std::string
Tom Tromey [Wed, 5 Sep 2018 17:25:28 +0000 (11:25 -0600)]
Change wrap buffering to use a std::string

Currently wrap buffering is implemented by allocating a string that is
the same width as the window, and then writing characters into it.
However, if gdb emits terminal escapes, then these could possibly
overflow the buffer.

To prevent this, change the wrap buffer to be a std::string and update
the various uses.

This also changes utils.c to always emit characters to the wrap
buffer.  This simplifies future patches which emit terminal escape
sequences, and also makes it possible for the "echo" and "printf"
commands to be used to emit terminal escapes and have these work in
the TUI.

gdb/ChangeLog
2018-12-28  Tom Tromey  <tom@tromey.com>

* utils.c (filter_initialized): New global.
(wrap_buffer): Now a std::string.
(wrap_pointer): Remove.
(flush_wrap_buffer): New function.
(filtered_printing_initialized, set_width, wrap_here)
(fputs_maybe_filtered): Update.

5 years agoFix leak of set/show verbose doc, avoid xfree of static string
Philippe Waroquiers [Fri, 28 Dec 2018 11:19:59 +0000 (12:19 +0100)]
Fix leak of set/show verbose doc, avoid xfree of static string

In the tests
  py-pp-registration/gdb.log
  default/gdb.log
  foll-fork/gdb.log
  setshow/gdb.log
  break-interp/gdb.log
Valgrind detects a leak of the doc strings for the set and show verbose cmd.

Here is the stacktrace of the leaked set doc:
==25548== 15 bytes in 1 blocks are definitely lost in loss record 101 of 3,120
==25548==    at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==25548==    by 0x409C27: xmalloc (common-utils.c:44)
==25548==    by 0x778AF9: xstrdup (xstrdup.c:34)
==25548==    by 0x3F860F: add_setshow_cmd_full(char const*, command_class, var_types, void*, char const*, char const*, char const*, void (*)(char const*, int, cmd_list_element*), void (*)(ui_file*, int, cmd_list_element*, char const*), cmd_list_element**, cmd_list_element**, cmd_list_element**, cmd_list_element**) [clone .constprop.10] (cli-decode.c:495)
==25548==    by 0x3F8ADB: add_setshow_boolean_cmd(char const*, command_class, int*, char const*, char const*, char const*, void (*)(char const*, int, cmd_list_element*), void (*)(ui_file*, int, cmd_list_element*, char const*), cmd_list_element**, cmd_list_element**) (cli-decode.c:593)
==25548==    by 0x3F7442: _initialize_cli_cmds() (cli-cmds.c:1768)
==25548==    by 0x69EED3: initialize_all_files() (init.c:365)
==25548==    by 0x658A84: gdb_init(char*) (top.c:2163)
==25548==    by 0x5403E1: captured_main_1 (main.c:863)
==25548==    by 0x5403E1: captured_main (main.c:1167)
==25548==    by 0x5403E1: gdb_main(captured_main_args*) (main.c:1193)
==25548==    by 0x289CA7: main (gdb.c:32)

The leak is created by top.c set_verbose 'elaborate joke':
the doc string is changed according to the verbosity:
  (gdb) help set verbose
  Set verbosity.
  (gdb) set verbose on
  (gdb) help set verbose
  Set verbose printing of informational messages.
  (gdb)

set_verbose creates the leak as it replaces the string allocated in
the above stacktrace by a static (non translated) string:
...
  if (info_verbose)
    {
      c->doc = "Set verbose printing of informational messages.";
...

Also, this can possibly trigger a call to 'free' of a static string,
as c->doc_allocated is kept true, while the string is not allocated anymore.

This patch:
 * fixes the leak by freeing the previous docs if doc_allocated.
 * internationalize the messages.
 * properly sets doc_allocated to 0 once doc strings are static.

gdb/ChangeLog
2018-12-28  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* top.c (set_verbose): Free previous docs if doc_allocated.
Internationalize messages.  Set doc_allocated to 0.

5 years agoAvoid internal errors when stepping outside 'main' on MinGW
Eli Zaretskii [Fri, 28 Dec 2018 07:02:04 +0000 (09:02 +0200)]
Avoid internal errors when stepping outside 'main' on MinGW

When one steps with "next" past the 'main's 'return' statement
in MinGW programs built by mingw.org's tools, PC lands in a
function whose symbol is not in any symtab.  GDB then looks
up the nearest symbol, and should find none, because all those
with addresses below PC are not real functions.  Having
unresolved symbols, whose address is zero, in minsyms tricked
GDB into using these bogus symbols, which then caused
assertion violation and internal_error.  See the discussion at
https://sourceware.org/ml/gdb-patches/2018-12/msg00176.html
for more details.

gdb/ChangeLog
2018-12-28  Eli Zaretskii  <eliz@gnu.org>

* coffread.c (coff_symtab_read): Don't record in minsyms symbols
that are unresolved.  This avoids triggering an internal error
when stepping outside of 'main' in MinGW programs.

5 years agoPR24015, glibc-2.28 on little-endian mips32 broken
Alan Modra [Thu, 27 Dec 2018 23:11:44 +0000 (09:41 +1030)]
PR24015, glibc-2.28 on little-endian mips32 broken

Commit 2bf2bf23da exposed a bug on targets that create common sections
other than the standard ELF SHN_COMMON.  If these are output by ld -r,
then their type becomes SHT_PROGBITS unless the target handles them
specially (eg. by elf_backend_special_sections), and if they are
merged into .bss/.sbss by ld -r then that section becomes SHT_PROGBITS.

Worse, if they are output by ld -r, then their size is increased by
bfd_generic_define_common_symbol during final link, which leads to
bogus file contents being copied to output.

For mips, it seems to me that the .scommon section should not be
output for ld -r, but I haven't made that change in this patch.

PR 24015
* elf.c (bfd_elf_get_default_section_type): Make common sections
SHT_NOBITS.
* linker.c (bfd_generic_define_common_symbol): Clear
SEC_HAS_CONTENTS.

5 years agoPR23966, mingw failure due to 32-bit long
Alan Modra [Thu, 27 Dec 2018 23:04:28 +0000 (09:34 +1030)]
PR23966, mingw failure due to 32-bit long

PR 23966
* libbfd.c (SSIZE_MAX): Define.
(bfd_malloc, bfd_realloc): Don't cast size to long to check for
"negative" values, compare against SSIZE_MAX instead.

5 years agoPR24028, PPC_INT_FMT
Alan Modra [Thu, 27 Dec 2018 23:01:42 +0000 (09:31 +1030)]
PR24028, PPC_INT_FMT

PPC_INT_FMT is redundant now that bfd.h pulls in inttypes.h if
available.  Apparently MacOS Mojave defines int64_t as long long even
though long is also 64 bits, which confuses the logic selecting
PPC_INT_FMT (and BFD_PRI64 too).  Hopefully inttypes.h is available on
Mojave.

PR 24028
include/
* opcode/ppc.h (PPC_INT_FMT): Delete.
opcodes/
* ppc-dis.c (print_insn_powerpc): Replace PPC_INT_FMT uses with
PRId64/PRIx64.