external/binutils.git
8 years agoImplement Intel OSPKE instructions
H.J. Lu [Wed, 9 Dec 2015 16:01:57 +0000 (08:01 -0800)]
Implement Intel OSPKE instructions

This patch implements Intel OSPKE instructions documented in Intel64
and IA-32 Architectures Software Developer’s Manual Volume 2, September
2015.

gas/testsuite/

* gas/i386/i386.exp: Run ospke and x86-64-ospke.
* gas/i386/ospke.d: New file.
* gas/i386/ospke.s: Likewise.
* gas/i386/x86-64-ospke.d: Likewise.

opcodes/

* i386-dis.c (MOD_0F01_REG_5): New.
(RM_0F01_REG_5): Likewise.
(reg_table): Use MOD_0F01_REG_5.
(mod_table): Add MOD_0F01_REG_5.
(rm_table): Add RM_0F01_REG_5.
* i386-gen.c (cpu_flag_init): Add CPU_OSPKE_FLAGS.
(cpu_flags): Add CpuOSPKE.
* i386-opc.h (CpuOSPKE): New.
(i386_cpu_flags): Add cpuospke.
* i386-opc.tbl: Add rdpkru and wrpkru instructions.
* i386-init.h: Regenerated.
* i386-tbl.h: Likewise.

8 years agogas/ELF: slightly relax elf/file*.d expectations
Jan Beulich [Wed, 9 Dec 2015 13:35:07 +0000 (14:35 +0100)]
gas/ELF: slightly relax elf/file*.d expectations

Despite the re-ordering done for the file symbols, some targets manage
to put section symbols ahead of it.

8 years agovarobj zero-padded hexadecimal format
Luis Machado [Wed, 9 Dec 2015 12:56:27 +0000 (10:56 -0200)]
varobj zero-padded hexadecimal format

This set of patches add support for the zero-padded hexadecimal format for
varobj's, defined as "zero-hexadecimal".  We currently only support regular
non-zero-padded hexadecimal.

Talking with IDE developers, they would like to have this option that is
already available to GDB's print/x commands, in the CLI, as 'z'.

gdb/ChangeLog:

2015-12-09  Luis Machado  <lgustavo@codesourcery.com>

* gdb/mi/mi-cmd-var.c (mi_parse_format): Handle new "zero-hexadecimal"
format.
* gdb/varobj.c (varobj_format_string): Add "zero-hexadecimal" entry.
(format_code): Add 'z' entry.
(varobj_set_display_format): Handle FORMAT_ZHEXADECIMAL.
* gdb/varobj.h (varobj_display_formats) <FORMAT_ZHEXADECIMAL>: New enum
field.
* NEWS: Add new note to MI changes citing the new zero-hexadecimal
format for -var-set-format.

gdb/doc/ChangeLog:

2015-12-09  Luis Machado  <lgustavo@codesourcery.com>

* gdb.texinfo (GDB/MI Variable Objects): Update text to mention
-var-set-format's new zero-hexadecimal format.

gdb/testsuite/ChangeLog:

2015-12-09  Luis Machado  <lgustavo@codesourcery.com>

* gdb.mi/mi-var-display.exp: Add new checks for the zero-hexadecimal
  format and change test names to make them unique.

8 years agosparc: support %dN and %qN syntax for FP registers.
Jose E. Marchesi [Wed, 9 Dec 2015 12:32:52 +0000 (07:32 -0500)]
sparc: support %dN and %qN syntax for FP registers.

The SPARC Refence Manual documents the %dN and %qN syntax to
refer to double and quad-precision floating-point registers,
respectively.  See OSA2015 Appendix C, Assembly Language Syntax,
C1.1 Register Names.

This patch adds support for these names to GAS.  This eases the
porting of software from Solaris to GNU/Linux, as these register
names have been supported by the Solaris linker for a long time
and many assembler require that support.

gas/ChangeLog:

2015-12-09  Jose E. Marchesi  <jose.marchesi@oracle.com>

* config/tc-sparc.c (sparc_ip): Support %dN and %qN notation for
double and quad-precision floating-point registers.

8 years agoFix wrong output of x87 registers due to truncation to double on amd64
Ruslan Kabatsayev [Wed, 9 Dec 2015 12:17:40 +0000 (12:17 +0000)]
Fix wrong output of x87 registers due to truncation to double on amd64

When `info float` is used on an AMD64 system, GDB prints
floating-point values of x87 registers with raw contents like
0x361a867a8e0527397ce0 or 0xc4f988454a1ddd3cfdab wrongly.

This happens due to truncation to double, after which the former
becomes 0.0, and the latter becomes negative infinity.  This is caused
by failed detection of x86-64 host, which results in setting
gdb_host_{float,double,long_double}_format to zeros.

This commit fixes this misdetection, and adds a test to make sure
future commits don't introduce a regression here.

gdb/ChangeLog:
2015-12-09  Ruslan Kabatsayev  <b7.10110111@gmail.com>

PR gdb/18702
* configure.host: Fix detection of x86_64 host when setting
floatformats.

gdb/testsuite/ChangeLog:
2015-12-09  Ruslan Kabatsayev  <b7.10110111@gmail.com>
    Pedro Alves  <pedro@redhat.com>

PR gdb/18702
Add checking of floatformats setup on x86_64 hosts.
* gdb.arch/i386-float.S (main): Load bigval and smallval.
(smallval, bigval): New labels/constants.
* gdb.arch/i386-float.exp: Use with_test_prefix and test "info
float" after loading bigval and smallval.

8 years agoFix compile time warning building RX target.
Nick Clifton [Wed, 9 Dec 2015 12:01:19 +0000 (12:01 +0000)]
Fix compile time warning building RX target.

8 years ago[GOLD] PowerPC style fix
Alan Modra [Wed, 9 Dec 2015 00:00:18 +0000 (10:30 +1030)]
[GOLD] PowerPC style fix

* powerpc.cc (Target_powerpc::Relocate::relocate): New constant
d_offset.  Use throughout.
(Target_powerpc::relocate_relocs): Likewise.

8 years ago[GOLD] Edit PowerPC64 ELFv2 function entry code
Alan Modra [Tue, 8 Dec 2015 23:48:44 +0000 (10:18 +1030)]
[GOLD] Edit PowerPC64 ELFv2 function entry code

In an fixed position executable, the entry code does not need to be
PIC and can thus lose a dependency on r12.

* powerpc.cc (Target_powerpc::Relocate::relocate): Edit ELFv2
entry code.
(Target_powerpc::relocate_relocs): Edit relocs to suit.

8 years ago[GOLD] Relocate::relocate() params
Alan Modra [Tue, 8 Dec 2015 23:48:30 +0000 (10:18 +1030)]
[GOLD] Relocate::relocate() params

Some linker code editing needs to change multiple insns.  In some
cases multiple relocations are involved and it is not sufficient to
make the changes independently as relocations are processed, because
doing so might lead to a partial edit.  So in order to safely edit we
need all the relocations available in relocate().  Also, to emit
edited relocs corresponding to the edited code sequence we need some
way to pass information from relocate() to relocate_relocs(),
particularly if the edit depends on insns.  We can't modify input
relocs in relocate() as they are mmapped PROT_READ, nor it is
particularly clean to write relocs to the output at that stage.  So
add a Relocatable_relocs* field to relinfo to mark edited relocs.

Given that relocate is passed the raw reloc pointer, it makes sense to
remove the rel/rela parameter and r_type too.  However, that means the
mips relocate() needs to know whether SHT_REL or SHT_RELA relocs are
being processed.  So add a rel_type for mips, which also has the
benefit of removing relocate() overloading there.

This patch adds the infrastructure without making use of it.

Note that relinfo->rr will be NULL if not outputting relocations.

* object.h (struct Relocate_info): Add "rr".
* reloc.h (Relocatable_relocs::set_strategy): New accessor.
* reloc.cc (Sized_relobj_file::do_relocate_sections): Init
relinfo.rr for relocate_section and relocate_relocs.
* powerpc.cc (relocate): Add rel_type and preloc parameters.
Delete rela and r_type params, instead recalculate these from
preloc.
(relocate_relocs): Delete Relocatable_relocs* param, instead
use relinfo->rr.
* aarch64.cc: Likewise.
* arm.cc: Likewise.
* i386.cc: Likewise.
* mips.cc: Likewise.
* s390.cc: Likewise.
* sparc.cc: Likewise.
* target.h: Likewise.
* tilegx.cc: Likewise.
* x86_64.cc: Likewise.
* testsuite/testfile.cc: Likewise.
* target-reloc.h (relocate_section): Adjust to suit.
(apply_relocation, relocate_relocs): Likewise.

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

8 years agoFix static analysis warning about undefined bheaviour.
Nick Clifton [Tue, 8 Dec 2015 09:49:49 +0000 (09:49 +0000)]
Fix static analysis warning about undefined bheaviour.

PR binutils/19310
* dwarf.c (display_debug_frames): Recode range test to avoid
undefined behaviour.

8 years agogas: consistently emit diagnostics for non-zero data emission to .bss/.struct
Jan Beulich [Tue, 8 Dec 2015 09:14:49 +0000 (10:14 +0100)]
gas: consistently emit diagnostics for non-zero data emission to .bss/.struct

8 years agogas: don't get confused by .asci{i,z} after .struct
Jan Beulich [Tue, 8 Dec 2015 09:12:54 +0000 (10:12 +0100)]
gas: don't get confused by .asci{i,z} after .struct

While not allowed, this certainly shouldn't result in confusing the
programmer (by skipping lines in unexpected ways): Without returning,
demand_empty_rest_of_line() (at the end of the function) will demand
the _next_ line to be empty, and without the conditional we would
ignore the next line.

8 years agoELF: don't re-order SHF_FILE symbols
Jan Beulich [Tue, 8 Dec 2015 09:11:58 +0000 (10:11 +0100)]
ELF: don't re-order SHF_FILE symbols

.file directives may be used to identify the scope of local symbols,
the purpose of which gets subverted when re-ordering them. Only allow
the first of them to be moved to the first position.

8 years agoDOCO: Enhance the menu to select function overloads with signatures
Pierre-Marie de Rodat [Thu, 3 Sep 2015 15:34:58 +0000 (17:34 +0200)]
DOCO: Enhance the menu to select function overloads with signatures

gdb/ChangeLog:

* NEWS: Announce this enhancement and the corresponding new
option.

gdb/doc/ChangeLog:

* gdb.texinfo (Ada Mode Into): Move overloading support
description to its own node.
(Overloading support for Ada): New node.

8 years agorl78: relaxation fixes
DJ Delorie [Tue, 8 Dec 2015 06:29:25 +0000 (01:29 -0500)]
rl78: relaxation fixes

Various fixes to linker relaxation.  In general, we need to support
relaxing every branch, even if we don't relax it in the assembler,
so we can optionally defer relaxation to the linker.

* elf32-rl78.c (rl78_offset_for_reloc): Add more relocs.
(rl78_elf_relax_section): Add bc/bz/bnc/bnz/bh/bnh.  Fix reloc
choices.

* config/rl78-parse.y: Make all branches relaxable via
rl78_linkrelax_branch().
* config/tc-rl78.c (rl78_linkrelax_branch): Mark all relaxable
branches with relocs.
(options): Add OPTION_NORELAX.
(md_longopts): Add -mnorelax.
(md_parse_option): Support OPTION_NORELAX.
(op_type_T): Add bh, sk, call, and br.
(rl78_opcode_type): Likewise.
(rl78_relax_frag): Fix not-relaxing logic.  Add sk.
(md_convert_frag): Fix relocation handling.
(tc_gen_reloc): Strip relax relocs when not linker relaxing.
(md_apply_fix): Defer overflow handling for anything that needs a
PLT, to the linker.
* config/tc-rl78.h (TC_FORCE_RELOCATION): Force all relocations to
the linker when linker relaxing.
* doc/c-rl78.texi (norelax): Add.

8 years agorx: Fix p_vaddr reconstruction logic.
DJ Delorie [Tue, 8 Dec 2015 06:15:58 +0000 (01:15 -0500)]
rx: Fix p_vaddr reconstruction logic.

* elf32-rx.c (rx_elf_object_p): Ignore empty and nobits sections.

8 years agorl78: Enable MULU for all ISAs.
DJ Delorie [Tue, 8 Dec 2015 03:33:39 +0000 (22:33 -0500)]
rl78: Enable MULU for all ISAs.

Unlike other mul/div opcodes, MULU is available on all variants
of the RL78.

* rl78-decode.opc: Enable MULU for all ISAs.
* rl78-decode.c: Regenerate.

8 years agoAutomatic date update in version.in
GDB Administrator [Tue, 8 Dec 2015 00:00:22 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agold: relax alignment requirements of compressed .debug_* section checks
Jan Beulich [Mon, 7 Dec 2015 16:52:25 +0000 (17:52 +0100)]
ld: relax alignment requirements of compressed .debug_* section checks

This fixes a failure of the gabinormal linking test on some distros
(where e.g. crt1.o has a .debug_aranges section with larger alignment).

8 years agoSupport Z0 packet in AArch64 multi-arch debugging
Yao Qi [Mon, 7 Dec 2015 15:56:31 +0000 (15:56 +0000)]
Support Z0 packet in AArch64 multi-arch debugging

In commit 6085d6f6, Z0 packet is disabled in aarch64 GDBserver if
the inferior is 32-bit or there may be multiple inferiors, because
Z0 packet isn't supported for arm then.  Recently, Z0 packet
is supported in arm target, so we don't have such limitation in
aarch64 GDBserver, that is to say, aarch64 GDBserver can use Z0
packet in multi-arch/multi-inferior debugging when the inferior's
arch is arm.

Part of this patch is to revert 6085d6f6, and the rest of the patch
is to move some breakpoint related arm_* functions into
linux-aarch32-low.c in order to share them between arm and aarch64.

This patch is regression tested on aarch64-linux for debugging both
aarch64 programs and arm programs respectively.

gdb/gdbserver:

2015-12-07  Yao Qi  <yao.qi@linaro.org>

* configure.srv: Append arm.o to srv_tgtobj for
aarch64*-*-linux* target.
* linux-aarch32-low.c (arm_abi_breakpoint): New macro.  Moved
from linux-arm-low.c.
(arm_eabi_breakpoint, arm_breakpoint): Likewise.
(arm_breakpoint_len, thumb_breakpoint): Likewise.
(thumb_breakpoint_len, thumb2_breakpoint): Likewise.
(thumb2_breakpoint_len): Likewise.
(arm_is_thumb_mode, arm_breakpoint_at): Likewise.
(arm_breakpoint_kinds): Likewise.
(arm_breakpoint_kind_from_pc): Likewise.
(arm_sw_breakpoint_from_kind): Likewise.
(arm_breakpoint_kind_from_current_state): Likewise.
* linux-aarch32-low.h (arm_breakpoint_kind_from_pc): Declare.
(arm_sw_breakpoint_from_kind): Declare.
(arm_breakpoint_kind_from_current_state): Declare.
(arm_breakpoint_at): Declare.
* linux-aarch64-low.c (aarch64_sw_breakpoint_from_kind): Call
arm_sw_breakpoint_from_kind if process is 32-bit.
(aarch64_breakpoint_kind_from_pc): New function.
(aarch64_breakpoint_kind_from_current_state): New function.
(the_low_target): Initialize fields breakpoint_kind_from_pc
and breakpoint_kind_from_current_state.
* linux-arm-low.c (arm_breakpoint_kinds): Move to
linux-aarch32-low.c.
(arm_abi_breakpoint, arm_eabi_breakpoint): Likewise.
(arm_breakpoint, arm_breakpoint_len): Likewise.
(thumb_breakpoint, thumb_breakpoint_len): Likewise.
(thumb2_breakpoint, thumb2_breakpoint_len): Likewise.
(arm_is_thumb_mode): Likewise.
(arm_breakpoint_at): Likewise.
(arm_breakpoint_kind_from_pc): Likewise.
(arm_sw_breakpoint_from_kind): Likewise.
(arm_breakpoint_kind_from_current_state): Likewise.

Revert:
2015-08-04  Yao Qi  <yao.qi@linaro.org>

* linux-aarch64-low.c (aarch64_supports_z_point_type): Return
0 for Z_PACKET_SW_BP if it may be used in multi-arch debugging.
* server.c (extended_protocol): Remove "static".
* server.h (extended_protocol): Declare it.

8 years agooops - accidentally omittde from previous delta.
Nick Clifton [Mon, 7 Dec 2015 14:44:46 +0000 (14:44 +0000)]
oops - accidentally omittde from previous delta.

8 years agoFix relaxation in RX linker when --no-keep-memory is specified.
Nick Clifton [Mon, 7 Dec 2015 14:43:47 +0000 (14:43 +0000)]
Fix relaxation in RX linker when --no-keep-memory is specified.

* elf32-rx.c (elf32_rx_relax_delete_bytes): Add extra parameter -
the start of the relocs for the section.  Delete code to load in
the relocs.
(elf32_rx_relax_section): Do not free the loaded relocs.

8 years agoEnhance the menu to select function overloads with signatures
Pierre-Marie de Rodat [Thu, 3 Sep 2015 15:34:58 +0000 (17:34 +0200)]
Enhance the menu to select function overloads with signatures

So far, trying to evaluate an expression involving a function call for
which GDB could find multiple function candidates outputs a menu so that
the user can select the one to run.  For instance, with the two
following functions:

    type New_Integer is new Integer;

    function F (I : Integer) return Boolean;
    function F (I : New_Integer) return Boolean;

Then we get the following GDB session:

    (gdb) print f(1)
    Multiple matches for f
    [0] cancel
    [1] foo.f at foo.adb:23
    [2] foo.f at foo.adb.28
    >

While the source location information is sufficient in order to
determine which one to select, one has to look for them in source files,
which is not convenient.

This commit tunes this menu in order to also include the list of formal
and return types (if any) in each entry.  The above then becomes:

    (gdb) print f(1)
    Multiple matches for f
    [0] cancel
    [1] foo.f (integer) return boolean at foo.adb:23
    [2] foo.f (foo.new_integer) return boolean at foo.adb.28
    >

Since this output is more verbose than previously, this change also
introduces an option (set/show ada print-signatures) to get the original
output.

gdb/ChangeLog:

* ada-lang.c (print_signatures): New.
(ada_print_symbol_signature): New.
(user_select_syms): Add signatures to the output of candidate
symbols using ada_print_symbol_signature.
(_initialize_ada_language): Add a "set/show ada
print-signatures" boolean option.

gdb/testsuite/ChangeLog:

* gdb.ada/fun_overload_menu.exp: New testcase.
* gdb.ada/fun_overload_menu/foo.adb: New testcase.

Tested on x86_64-linux, no regression.

8 years agoAdd myself as a write-after-approval GDB maintainer
Andreas Arnez [Mon, 7 Dec 2015 09:43:39 +0000 (10:43 +0100)]
Add myself as a write-after-approval GDB maintainer

gdb/ChangeLog:

* MAINTAINERS (Write After Approval): Add Andreas Arnez.

8 years agoAdd support for MSP430 F5 hardware multiply.
Nick Clifton [Mon, 7 Dec 2015 10:19:19 +0000 (10:19 +0000)]
Add support for MSP430 F5 hardware multiply.

* msp430-sim.c (sim_open): Check for needed memory at address
0x500 not 0x200.
(get_op): Add support for F5 hardware multiply addresses.
(put_op): Likewise.

8 years agoPowerPC ifunc with local symbols
Alan Modra [Mon, 7 Dec 2015 03:22:01 +0000 (13:52 +1030)]
PowerPC ifunc with local symbols

This fixes some cases where the linker would incorrectly error on plt
relocs to local ifunc symbols.  I've also tidied plt and ifunc
handling for ppc64, where check_relocs was allowing for the
possibility of plt calls via addr14/addr24 relocs but relocate_section
was not.

* elf32-ppc.c (ppc_elf_check_relocs): Don't error on local ifunc
plt call.  Wrap long lines.
(ppc_elf_relocate_section): Wrap long lines.
* elf64-ppc.c (ppc64_elf_check_relocs): Don't error on local ifunc
plt calls.  Move __tls_get_addr checks later.  Don't create plt
for addr14/addr24 relocs.
(ppc64_elf_gc_sweep_hook): Adjust to suit check_relocs changes.
(ppc64_elf_relocate_section): Correct local ifunc handling for
PLT64, PLT32 and PLT16 relocs.

8 years agoPR19323 memory allocation greater than 4G
Alan Modra [Mon, 7 Dec 2015 03:11:36 +0000 (13:41 +1030)]
PR19323 memory allocation greater than 4G

On 32-bit targets, memory requested for program/section headers on a
fuzzed binary can wrap to 0.  A bfd_alloc of zero bytes actually
returns a one byte allocation rather than a NULL pointer.  This then
leads to buffer overflows.

Making this check unconditional triggers an extremely annoying gcc-5
warning.

PR19323
* elfcode.h (elf_object_p): Check for ridiculous e_shnum and
e_phnum values.

8 years ago[GOLD] R_PPC64_ENTRY support
Alan Modra [Mon, 7 Dec 2015 02:45:24 +0000 (13:15 +1030)]
[GOLD] R_PPC64_ENTRY support

elfcpp/
* powerpc.h (R_PPC64_ENTRY): Define.
gold/
* powerpc.cc (add_2_2_12, ld_2_12, lis_2): Define.
(Target_powerpc::Scan::local, global): Handle R_PPC64_ENTRY.
(Target_powerpc::Relocate::relocate): Edit code at R_PPC64_ENTRY.

8 years agoR_PPC64_ENTRY
Alan Modra [Mon, 7 Dec 2015 02:44:53 +0000 (13:14 +1030)]
R_PPC64_ENTRY

Add a new relocation that marks large-model entry code, for edit back
to medium-model.

include/elf/
* ppc64.h (R_PPC64_ENTRY): Define.
bfd/
* reloc.c (BFD_RELOC_PPC64_ENTRY): New.
* elf64-ppc.c (reloc_howto_type ppc64_elf_howto_raw): Add
entry for R_PPC64_ENTRY.
(LD_R2_0R12, ADD_R2_R2_R12, LIS_R2, ADDIS_R2_R12): Define.
(ppc64_elf_reloc_type_lookup): Handle R_PPC64_ENTRY.
(ppc64_elf_relocate_section): Edit code at R_PPC64_ENTTY.  Use
new insn defines.
* libbfd.h: Regenerate.
* bfd-in2.h: Regenerate.

8 years agotc-ppc.c md_apply_fix tidy
Alan Modra [Mon, 7 Dec 2015 02:44:35 +0000 (13:14 +1030)]
tc-ppc.c md_apply_fix tidy

* config/tc-ppc.c (md_apply_fix): Localize variables.  Reduce casts.

8 years agoReorder some power9 insns
Alan Modra [Mon, 7 Dec 2015 02:44:05 +0000 (13:14 +1030)]
Reorder some power9 insns

The idea being to put instructions that have the same encoding adjacent
to each other.

* opcodes/ppc-opc.c (powerpc_opcodes): Sort power9 insns by
major opcode/xop.

8 years agobfd: Mark sh5*-*-* and sh64*-*-* targets as obsolete.
Kaz Kojima [Mon, 7 Dec 2015 00:58:37 +0000 (09:58 +0900)]
bfd: Mark sh5*-*-* and sh64*-*-* targets as obsolete.

8 years agoAutomatic date update in version.in
GDB Administrator [Mon, 7 Dec 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoReplace remaining references to i386-nat with x86-nat instead.
Joel Brobecker [Sun, 6 Dec 2015 17:38:24 +0000 (18:38 +0100)]
Replace remaining references to i386-nat with x86-nat instead.

i386-nat.[hc] got renamed to x86-nat.[hc] a while back, but somehow
3 references to the old file name remained past the renaming. This
fixes all of them.

gdb/ChangeLog (with Mike Stump <mikestump@comcast.net>):

        * Makefile.in (TAGS): Replace i386-nat.h by x86-nat.h.
        * x86-nat.c: Replace remaining references to i386-nat
        by reference to x86-nat instead.

8 years agoAutomatic date update in version.in
GDB Administrator [Sun, 6 Dec 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoDocument the GDB 7.10.1 release in gdb/ChangeLog
Joel Brobecker [Sat, 5 Dec 2015 15:29:09 +0000 (16:29 +0100)]
Document the GDB 7.10.1 release in gdb/ChangeLog

gdb/ChangeLog:

GDB 7.10.1 released.

8 years agogdbserver: set ptrace flags after creating inferiors
Josh Stone [Fri, 4 Dec 2015 21:28:07 +0000 (13:28 -0800)]
gdbserver: set ptrace flags after creating inferiors

Rename target_ops.arch_setup to .post_create_inferior.  In the Linux
hook, continue calling the low arch setup, then also set ptrace flags.
This corrects the possibility of running without flags, demonstrated by
a new test that would fail to catch a fork before.

gdb/gdbserver/ChangeLog:

2015-12-04  Josh Stone  <jistone@redhat.com>

* target.h (struct target_ops) <arch_setup>: Rename to ...
(struct target_ops) <post_create_inferior>: ... this.
(target_arch_setup): Rename to ...
(target_post_create_inferior): ... this, calling post_create_inferior.
* server.c (start_inferior): Update target_arch_setup calls to
target_post_create_inferior.
* linux-low.c (linux_low_ptrace_options): Forward declare.
(linux_arch_setup): Update its comment for general use.
(linux_post_create_inferior): New, run arch_setup and setup ptrace.
(struct linux_target_ops): Use linux_post_create_inferior.
* lynx-low.c (struct lynx_target_ops): Update arch_setup stub comment
to post_create_inferior.
* nto-low.c (struct nto_target_ops): Likewise.
* spu-low.c (struct spu_target_ops): Likewise.
* win32-low.c (struct win32_target_ops): Likewise.

gdb/testsuite/ChangeLog:

2015-12-04  Josh Stone  <jistone@redhat.com>

* gdb.base/catch-fork-static.exp: New.

8 years agoAutomatic date update in version.in
GDB Administrator [Sat, 5 Dec 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoOptimize R_386_GOT32/R_386_GOT32X only if addend is 0
H.J. Lu [Fri, 4 Dec 2015 16:43:45 +0000 (08:43 -0800)]
Optimize R_386_GOT32/R_386_GOT32X only if addend is 0

Linker can't optimize R_386_GOT32 and R_386_GOT32X relocations if addend
isn't 0.  It isn't valid to convert

movl foo@GOT+1(%ecx), %eax

to

leal foo@GOTOFF+1(%ecx), %eax

nor to convert

movq foo@GOTPCREL+1(%rip), %rax

to

leaq foo(%rip), %rax

for x86-64.  We should check if addend is 0 before optimizing R_386_GOT32
and R_386_GOT32X relocations.  Testcases are added for i386 and x86-64.

bfd/

* elf32-i386.c (elf_i386_convert_load): Skip if addend isn't 0.
(elf_i386_relocate_section): Skip R_386_GOT32X optimization if
addend isn't 0.

ld/testsuite/

* ld-i386/i386.exp: Run mov2a, mov2b and mov3.
* ld-i386/mov2.s: New file.
* ld-i386/mov2a.d: Likewise.
* ld-i386/mov2b.d: Likewise.
* ld-i386/mov3.d: Likewise.
* ld-i386/mov3.s: Likewise.
* ld-x86-64/mov2.s: Likewise.
* ld-x86-64/mov2a.d: Likewise.
* ld-x86-64/mov2b.d: Likewise.
* ld-x86-64/mov2c.d: Likewise.
* ld-x86-64/mov2d.d: Likewise.
* ld-x86-64/x86-64.exp: Run mov2a, mov2b, mov2c and mov2d.

8 years agoFix GAS testsuite failures for COFF/PE based ARM targets.
Nick Clifton [Fri, 4 Dec 2015 15:07:10 +0000 (15:07 +0000)]
Fix GAS testsuite failures for COFF/PE based ARM targets.

PR gas/19276
gas * config/tc-arm.h (SUB_SEGMENT_ALIGN): Do not define for COFF/PE
targets.

testsuite * gas/arm/align64.d: Skip for COFF/PE targets.
* gas/arm/bundle-lock.d: Adjust for COFF/PE targets.

8 years agoRemove useless loop in elf.c
Tristan Gingold [Thu, 3 Dec 2015 10:57:29 +0000 (11:57 +0100)]
Remove useless loop in elf.c

8 years agoFix failures in the GAS testsuite for the ARC architecture.
Claudiu Zissulescu [Fri, 4 Dec 2015 10:49:57 +0000 (10:49 +0000)]
Fix failures in the GAS testsuite for the ARC architecture.

gas * config/tc-arc.c (arc_option): Sets all internal gas options when
parsing .cpu directive.
(declare_register_set): Declare all 64 registers.
(md_section_align): Refactor.
(md_pcrel_from_section): Remove assert.
(pseudo_operand_match): Fix pseudo operand match.
(find_reloc): Use flags filed, extend matching.
* config/tc-arc.h (TC_VALIDATE_FIX): Don't fixup any PLT
relocation.

testsuite * gas/arc/bic.d: Update test.
* gas/arc/add_s-err.s: New file.
* gas/arc/cpu-warn1.s: Likewise.
* gas/arc/pcl-relocs.d: Likewise.
* gas/arc/pcl-relocs.s: Likewise.
* gas/arc/pcrel-relocs.d: Likewise.
* gas/arc/pcrel-relocs.s: Likewise.
* gas/arc/pic-relocs.d: Likewise.
* gas/arc/pic-relocs.s: Likewise.
* gas/arc/plt-relocs.d: Likewise.
* gas/arc/plt-relocs.s: Likewise.
* gas/arc/pseudos.d: Likewise.
* gas/arc/pseudos.s: Likewise.
* gas/arc/sda-relocs.d: Likewise.
* gas/arc/sda-relocs.s: Likewise.
* gas/arc/sda-relocs2.d: Likewise.
* gas/arc/sda-relocs2.s: Likewise.
* gas/arc/tls-relocs.d: Likewise.
* gas/arc/tls-relocs.s: Likewise.

opcode * arc.h (arc_reloc_equiv_tab): Replace flagcode with flags[32].

opcodes * arc-dis.c (special_flag_p): Match full mnemonic.
* arc-opc.c (print_insn_arc): Check section size to read
appropriate number of bytes. Fix printing.
* arc-tbl.h: Fix instruction table. Allow clri/seti instruction without
arguments.

8 years agoAutomatic date update in version.in
GDB Administrator [Fri, 4 Dec 2015 00:00:12 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoTake addend into account when making GOT entries for local symbols.
Vladimir Radosavljevic [Thu, 3 Dec 2015 23:29:17 +0000 (15:29 -0800)]
Take addend into account when making GOT entries for local symbols.

gold/
* object.cc (Sized_relobj::do_for_all_local_got_entries): Use
Local_got_entry_key for searching in local_got_offsets_.
* object.h (class Local_got_entry_key): New class.
(Relobj::local_has_got_offset): New overloaded method.
(Relobj::local_got_offset): Likewise.
(Relobj::set_local_got_offset): Likewise.
(Relobj::do_local_has_got_offset): Add addend argument.
(Relobj::do_local_got_offset): Likewise.
(Relobj::do_set_local_got_offset): Likewise.
(Sized_relobj::do_local_has_got_offset): Add addend argument, and use
Local_got_entry_key for searching through local_got_offsets_.
(Sized_relobj::do_local_got_offset): Likewise.
(Sized_relobj::do_set_local_got_offset): Likewise.
(Sized_relobj::Local_got_offsets): Change type of the key from
unsigned int to Local_got_entry_key, and add hash and equal_to.
* output.cc (Got_entry::write): Take addend into account for
calculating value of the local symbol for GOT.
(Output_data_got::add_local): New definition of overloaded method.
(Output_data_got::add_local_with_rel): Likewise.
(Output_data_got::add_local_pair_with_rel): Likewise.
* output.h (Output_data_got::add_local): New declaration of overloaded
method.

8 years agoRemove duplicate arch/arm.h include in linux-arm-low.c.
Antoine Tremblay [Thu, 3 Dec 2015 18:56:37 +0000 (13:56 -0500)]
Remove duplicate arch/arm.h include in linux-arm-low.c.

A duplicate include arm/arm.h was introduced, remove it.
Pushed as obvious.

gdb/gdbserver/ChangeLog:

* linux-arm-low.c: Remove duplicate arch/arm.h include.

8 years agoRun gdb.base/sizeof.exp with board having gdb,noinferiorio
Yao Qi [Thu, 3 Dec 2015 17:12:41 +0000 (17:12 +0000)]
Run gdb.base/sizeof.exp with board having gdb,noinferiorio

In my remote cross testing (x86_64 host and aarch64 target), the test
gdb.base/sizeof.exp is skipped because gdb,noinferiorio is defined in
my gdbserver board file.  Tests are skipped because the test checks
the expected value from the program's output, but I don't see why must
do it this way.  With my patch applied, we can save the result in variable
in the program, and check the variable then.  Then, the test doesn't rely
on inferiorio.

gdb/testsuite:

2015-12-03  Yao Qi  <yao.qi@linaro.org>

* gdb.base/sizeof.c: Don't include stdio.h and
../lib/unbuffer_output.c.
(main): New variable 'size' and 'value'.  Remove printf and
gdb_unbuffer_output.  Assign return value to size and value.
* gdb.base/sizeof.exp: Remove the checking to gdb,noinferiorio
at the beginning.
(check_sizeof): Check the result by printing variable 'size'.
(check_valueof): Check the result by printing variable 'value'.

8 years agoDarwin: add new mach-o header flags.
Tristan Gingold [Thu, 3 Dec 2015 10:56:09 +0000 (11:56 +0100)]
Darwin: add new mach-o header flags.

binutils/
* od-macho.c (bfd_mach_o_header_flags_name): Add name
for flags until BFD_MACH_O_MH_APP_EXTENSION_SAFE.

include/mach-o/
* loader.h (bfd_mach_o_header_flags): Add
BFD_MACH_O_MH_APP_EXTENSION_SAFE.

8 years agoAutomatic date update in version.in
GDB Administrator [Thu, 3 Dec 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoaddr2line vs. inlined C functions called from C++
Alan Modra [Wed, 2 Dec 2015 05:16:48 +0000 (15:46 +1030)]
addr2line vs. inlined C functions called from C++

In this case the inlined function doesn't have DW_AT_linkage_name in
.debug_info, but the language is C++ so find_nearest_line goes looking
in the symbol table.  Since the function is inlined the enclosing
non-inline function symbol is returned from _bfd_elf_find_function,
which is wrong.  This patch only uses a symbol if its address matches.

PR binutils/19315
* dwarf2.c (_bfd_elf_find_function): Return symbol matched.
(_bfd_dwarf2_find_nearest_line): Check symbol returned above
against dwarf range.
* elf-bfd.h (_bfd_elf_find_function): Update prototype.

8 years agoMake --enable-initfini-array the default
Alan Modra [Wed, 2 Dec 2015 08:53:41 +0000 (19:23 +1030)]
Make --enable-initfini-array the default

* configure.ac (--enable-initfini-array): Remove run test.  Default
to "yes".  Change help string to --disable-initfini-array.
* configure: Regenerate.

8 years agoFix powerpc64 segfault caused by zero r_symndx relocs.
Alan Modra [Wed, 2 Dec 2015 08:19:53 +0000 (18:49 +1030)]
Fix powerpc64 segfault caused by zero r_symndx relocs.

Fixes a segfault in ppc64_elf_tls_optimize found when testing
R_PPC64_ENTRY, and potential for trouble in other places found by
code inspection.

* elf64-ppc.c (ppc64_elf_tls_optimize): Don't segfault on NULL
symbol section or output section.
(ppc64_elf_edit_toc): Similarly for ld -R objects.
(ppc64_elf_size_stubs): Likewise.

8 years agoFix ldah being disassembled as ldaexh
Andre Vieira [Wed, 25 Nov 2015 13:56:55 +0000 (13:56 +0000)]
Fix ldah being disassembled as ldaexh

2015-12-02  Andre Vieira  <andre.simoesdiasvieira@arm.com>

opcodes/
    * arm-dis.c (arm_opcodes): <ldaexh>: Fix typo...
    <ldah>: ... to this.

gas/testsuite/
    * gas/arm/armv8-a.d: <ldaexh>: Rename mismatched mnemonics ...
    <ldah>: ... to this.

8 years agoAutomatic date update in version.in
GDB Administrator [Wed, 2 Dec 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoProperly check symbol defined by assignment in linker script
H.J. Lu [Tue, 1 Dec 2015 22:45:51 +0000 (14:45 -0800)]
Properly check symbol defined by assignment in linker script

Symbol defined by a linker assignment may have type bfd_link_hash_new
or bfd_link_hash_undefined.  And h->def_regular is always set.
elf_i386_convert_load and elf_x86_64_convert_load should check
h->def_regular as well as bfd_link_hash_undefined and bfd_link_hash_new
to see if a symbol is defined by a linker script.

bfd/

PR ld/19319
* elf32-i386.c (elf_i386_convert_load): Check h->def_regular
instead of bfd_link_hash_new.
* elf64-x86-64.c (elf_x86_64_convert_load): Likewise.  Skip
relocation overflow for bfd_link_hash_undefined and
bfd_link_hash_new if h->def_regular is set.

ld/testsuite/

PR ld/19319
* ld-i386/i386.exp: Run pr19319 test.
* ld-x86-64/x86-64.exp: Likewise.
* ld-i386/pr19319.dd: New file.
* ld-i386/pr19319a.S: Likewise.
* ld-i386/pr19319b.S: Likewise.
* ld-x86-64/pr19319.dd: Likewise.
* ld-x86-64/pr19319a.S: Likewise.
* ld-x86-64/pr19319b.S: Likewise.

8 years agoAvoid "operation may be undefined" warning in remote.c
Ulrich Weigand [Tue, 1 Dec 2015 17:04:39 +0000 (18:04 +0100)]
Avoid "operation may be undefined" warning in remote.c

GCC 4.1 gives the following warning:
gdb/remote.c: In function 'remote_parse_stop_reply':
gdb/remote.c:6549: warning: operation on 'p' may be undefined
on this line of code:

event->ptid = read_ptid (++p, &p);

Since p actually isn't used afterwards anyway, simply use NULL.

gdb/
* remote.c (remote_parse_stop_reply): Avoid GCC 4.1 "operation
may be undefined" warning.

8 years agoFix uninitialized variable warnings in remote.c
Ulrich Weigand [Tue, 1 Dec 2015 16:49:27 +0000 (17:49 +0100)]
Fix uninitialized variable warnings in remote.c

Fix a couple of places where a struct thread_item was added to a
vector while the item.name field was uninitialized.

gdb/
* remote.c (remote_newthread_step): Initialize item.name.
(remote_get_threads_with_qthreadinfo): Likewise.

8 years agoRun gdb.base/disp-step-syscall.exp for aarch64-linux
Yao Qi [Tue, 1 Dec 2015 12:37:04 +0000 (12:37 +0000)]
Run gdb.base/disp-step-syscall.exp for aarch64-linux

This patch handles target aarch64*-*-linux* for syscall instruction.

gdb/testsuite:

2015-12-01  Yao Qi  <yao.qi@linaro.org>

* gdb.base/disp-step-syscall.exp: Define syscall instruction
for aarch64*-*-linux* target.

8 years agoTrim unused params from aout adjust_sizes_and_vmas
Alan Modra [Tue, 1 Dec 2015 03:26:09 +0000 (13:56 +1030)]
Trim unused params from aout adjust_sizes_and_vmas

* aoutx.h (adjust_sizes_and_vmas): Remove unused text_size and
text_end parameters.  Update callers.
* aout-tic30.c: Update adjust_sizes_and_vmas callers.
* hp300hpux.c: Likewise.
* i386lynx.c: Likewise.
* libaout.h: Likewise.
* netbsd.h: Likewise.
* pdp11.c: Likewise.
* riscix.c: Likewise.

8 years agoInvoke aout N_* macros with pointer to struct internal_exec
Alan Modra [Tue, 1 Dec 2015 01:10:52 +0000 (11:40 +1030)]
Invoke aout N_* macros with pointer to struct internal_exec

No functional changes here.

BTW, some of these headers don't seem to be used anywhere:
include/aout/dynix3.h, include/aout/encap.h, include/aout/hp.h,
gas/config/aout_gnu.h

bfd/
* aout-adobe.c: Invoke aout N_* macros with pointer to
struct internal_exec.
* aout-arm.c: Likewise.
* aout-cris.c: Likewise.
* aout-target.h: Likewise.
* aout-tic30.c: Likewise.
* aoutf1.h: Likewise.
* aoutx.h: Likewise.
* bout.c: Likewise.
* freebsd.h: Likewise.
* gen-aout.c: Likewise.
* hp300hpux.c: Likewise.
* i386aout.c: Likewise.
* i386linux.c: Likewise.
* i386lynx.c: Likewise.
* i386mach3.c: Likewise.
* i386os9k.c: Likewise.
* libaout.h: Likewise.
* m68klinux.c: Likewise.
* m88kmach3.c: Likewise.
* mipsbsd.c: Likewise.
* netbsd.h: Likewise.
* pc532-mach.c: Likewise.
* pdp11.c: Likewise.
* riscix.c: Likewise.
* sparclinux.c: Likewise.
* sparclynx.c: Likewise.
gas/
* config/aout_gnu.h: Invoke aout N_* macros with pointer to
struct internal_exec.
include/
* bout.h: Invoke aout N_* macros with pointer to
struct internal_exec.
* os9k.h: Likewise.
include/aout/
* adobe.h: Invoke aout N_* macros with pointer to
struct internal_exec.
* aout64.h: Likewise.
* dynix3.h: Likewise.
* encap.h: Likewise.
* hp.h: Likewise.
* hp300hpux.h: Likewise.
* sun4.h: Likewise.

8 years agoDon't use BFD_TRADITIONAL_FORMAT flag in COFF support
Alan Modra [Tue, 1 Dec 2015 01:10:25 +0000 (11:40 +1030)]
Don't use BFD_TRADITIONAL_FORMAT flag in COFF support

info->traditional_format is available, or can be easily made
available.  This relegates BFD_TRADITIONAL_FORMAT to AOUT use only.

* coff-rs6000.c (_bfd_xcoff_put_symbol_name): Replace abfd param
with info param.  Test info->traditional_format rather than
BFD_TRADITIONAL_FORMAT flag.
* coff64-rs6000.c (_bfd_xcoff64_put_symbol_name): Likewise.
* libxcoff.h (struct xcoff_backend_data_rec): Update
_xcoff_put_symbol_name prototype.
(bfd_xcoff_put_symbol_name): Add info param.
* xcofflink.c (xcoff_find_tc0): Update bfd_xcoff_put_symbol_name call.
(xcoff_write_global_symbol): Likewise.
(xcoff_link_input_bfd): Test info->traditional_format rather than
BFD_TRADITIONAL_FORMAT flag.
* cofflink.c (_bfd_coff_final_link): Likewise.
(_bfd_coff_link_input_bfd, _bfd_coff_write_global_sym): Likewise.

8 years agobinutils/configure update
Alan Modra [Tue, 1 Dec 2015 01:10:02 +0000 (11:40 +1030)]
binutils/configure update

Missed from f8c2a965.

* configure: Regenerate.

8 years agoSRC-POTFILES.in update
Alan Modra [Tue, 1 Dec 2015 01:09:37 +0000 (11:39 +1030)]
SRC-POTFILES.in update

* po/SRC-POTFILES.in: Regenerate.

8 years agoRe: ARC port broken reloc processing
Alan Modra [Tue, 1 Dec 2015 03:06:30 +0000 (13:36 +1030)]
Re: ARC port broken reloc processing

* elf32-arc.c (ARC_ELF_HOWTO): Delete.
(arc_elf_howto): New function.
(bfd_elf32_bfd_reloc_type_lookup): Use it in place of existing
init code.
(bfd_elf32_bfd_reloc_name_lookup): Use arc_elf_howto.
(arc_info_to_howto_rel, elf_arc_relocate_section): Likwise.
(elf_arc_check_relocs): Likewise.

8 years agoARC port broken reloc processing
Alan Modra [Tue, 1 Dec 2015 00:40:23 +0000 (11:10 +1030)]
ARC port broken reloc processing

This initialises howto.dst_mask so that relocations in debug sections
are applied by the generic reloc processing used by objdump to display
debug sections.

* elf32-arc.c (arc_elf_howto_init): Init dst_mask.

8 years agoAutomatic date update in version.in
GDB Administrator [Tue, 1 Dec 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoRemove too simple breakpoint_reinsert_addr implementations.
Antoine Tremblay [Thu, 19 Nov 2015 16:29:10 +0000 (11:29 -0500)]
Remove too simple breakpoint_reinsert_addr implementations.

This patch removes too simple implementations of the breakpoint_reinsert_addr
operation.

The only reason to keep them around was to support thread events when
PTRACE_EVENT_CLONE was not present but this support has been removed in a
previous patch.

No regressions, tested on ubuntu 14.04 ARMv7 and x86.
With gdbserver-{native,extended} / { -marm -mthumb }

Also compilation was tested on aarch64, bfin, cris, crisv32,
m32r, mips, nios2, ppc, s390, sparc, tic6x, tile,  xtensa.

gdb/gdbserver/ChangeLog:

* linux-arm-low.c (arm_reinsert_addr): Remove function.
(struct linux_target_ops <breakpoint_reinsert_addr>: Set to NULL.
* linux-cris-low.c (cris_reinsert_addr> Remove function.
(struct linux_target_ops) <breakpoint_reinsert_addr>: Set to NULL.
* linux-crisv32-low.c (cris_reinsert_addr): Remove function.
(struct linux_target_ops) <breakpoint_reinsert_addr>: Set to NULL.
* linux-mips-low.c (mips_reinsert_addr): Remove function.
(struct linux_target_ops) <breakpoint_reinsert_addr>: Set to NULL.
* linux-nios2-low.c (nios2_reinsert_addr): Remove function.
(struct linux_target_ops) <breakpoint_reinsert_addr>: Set to NULL.
* linux-sparc-low.c (sparc_reinsert_addr): Remove function.
(struct linux_target_ops) <breakpoint_reinsert_addr>: Set to NULL.

8 years agoRemove support for thread events without PTRACE_EVENT_CLONE in GDBServer.
Antoine Tremblay [Thu, 19 Nov 2015 16:29:10 +0000 (11:29 -0500)]
Remove support for thread events without PTRACE_EVENT_CLONE in GDBServer.

This patch removes support for thread events if PTRACE_EVENT_CLONE is not
supported in GDBServer.

Before, on systems that did not support PTRACE_EVENT_CLONE, both GDB and
GDBServer coordinated with libthread_db.so to insert breakpoints at magic
locations in libpthread.so, in order to break at thread creation and thread
death.

Simple software single stepping support was implemented to step over these
breakpoints in case there was no hardware single stepping support. However,
these simple software single stepping implementations were not fit for any other
use as discussed in :
https://sourceware.org/ml/gdb-patches/2015-04/msg01110.html

These too simple implementations conflict with ongoing work to make proper
implementations of software single stepping in GDBServer.

The problem is that if some implementations are correct and others are not and
only there for the thread magic breakpoint, we can't enable features based
solely software single step support since some would be broken.

To keep the incorrect implementations and allow the new proper ones at the same
time we would need to implement fallback code and it quickly becomes ugly and
confusing with multiple checks for legacy software single step or proper
software single step.

However, PTRACE_EVENT_CLONE was first introduced in Linux 2.5.46,
released in November 2002.

So I think it's reasonable to just remove support for kernels that don't support
PTRACE_EVENT_CLONE, and sidestep the libthread_db breakpoints issues entirely.

This thread on the mailling list discusses the issue :
https://sourceware.org/ml/gdb/2015-10/msg00078.html

No regressions, tested on ubuntu 14.04 ARMv7 and x86.
With gdbserver-{native,extended} / { -marm -mthumb }

gdb/gdbserver/ChangeLog:

* linux-low.c (linux_look_up_symbols): Don't call
linux_supports_traceclone.
* linux-low.h (thread_db_init): Remove use_events argument.
* thread-db.c (thread_db_use_event): Remove global variable.
(struct thread_db) <td_thr_event_enable_p>: Remove field.
(struct thread_db) <td_create_bp>: Remove field.
(thread_db_create_event): Remove function.
(thread_db_enable_reporting): Likewise.
(find_one_thread): Don't check for thread_db_use_events.
(attach_thread): Likewise.
(thread_db_load_search): Remove td_thr_event_enable_p initialization.
(try_thread_db_load_1): Don't check for thread_db_use_events.
(thread_db_init): Remove use_events argument and thread events
handling.
(remove_thread_event_breakpoints): Remove function.
(thread_db_detach): Remove call to remove_thred_event_breakpoints.

8 years agoRefactor queries for hardware and software single stepping support in GDBServer.
Antoine Tremblay [Thu, 19 Nov 2015 16:29:10 +0000 (11:29 -0500)]
Refactor queries for hardware and software single stepping support in GDBServer.

Before this patch there was only one call: can_hardware_single_step. Its
implementation was a check on breakpoint_reinsert_addr if NULL it assumed
that the target could hardware single step.

This patch prepares for the case where this is not true anymore.

In order to improve software single stepping in GDBServer the
breakpoint_reinsert_addr operation of targets that had a very simple
software implementation used only for stepping over thread creation events
will be removed.

This will create a case where a target does not support hardware single
step and has the operation breakpoint_reinsert_addr set to NULL, thus
can_hardware_single_step needs to be implemented another way.

A new target operation supports_hardware_single_step is introduced and is
to return true if the target does support such a feature, support for the
feature is manually hardcoded.

Note that the hardware single step support was enabled as per the current
behavior, I did not check if tile for example really has ptrace singlestep
support but since the current implementation assumed it had, I kept it
that way.

No regressions on Ubuntu 14.04 on ARMv7 and x86.
With gdbserver-{native,extended} / { -marm -mthumb }

Compilation tested on: aarch64,arm,bfind,crisv32,m32r,ppc,s390,tic6x,tile,
xtensa.
Not tested : sh.

gdb/gdbserver/ChangeLog:

* linux-aarch64-low.c (aarch64_supports_hardware_single_step):
New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-arm-low.c (arm_supports_hardware_single_step): New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-bfin-low.c (bfin_supports_hardware_single_step): New function.
(struct linux_target_ops) <bfin_supports_hardware_single_step>:
Initialize.
* linux-crisv32-low.c (cris_supports_hardware_single_step):
New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-low.c (can_hardware_single_step): Use
supports_hardware_single_step.
(can_software_single_step): New function.
(start_step_over): Call can_software_single_step.
(linux_supports_hardware_single_step): New function.
(struct target_ops) <supports_software_single_step>: Initialize.
* linux-low.h (struct linux_target_ops)
<supports_hardware_single_step>: Initialize.
* linux-m32r-low.c (m32r_supports_hardware_single_step): New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-ppc-low.c (ppc_supports_hardware_single_step): New function.
(struct linux_target_ops) <supports_hardware_single_step> Initialize.
* linux-s390-low.c (s390_supports_hardware_single_step): New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-sh-low.c (sh_supports_hardware_single_step): New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-tic6x-low.c (tic6x_supports_hardware_single_step): New function.
(struct linux_target_ops) <tic6x_supports_hardware_single_step>:
Initialize.
* linux-tile-low.c (tile_supports_hardware_single_step): New function.
(struct linux_target_ops) <tile_supports_hardware_single_step>:
Initialize.
* linux-x86-low.c (x86_supports_hardware_single_step) New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* linux-xtensa-low.c (xtensa_supports_hardware_single_step):
New function.
(struct linux_target_ops) <supports_hardware_single_step>: Initialize.
* target.h (struct target_ops): <supports_software_single_step>:
New field.
(target_supports_software_single_step): New macro.

8 years agoFix instruction skipping when using software single step in GDBServer
Antoine Tremblay [Mon, 30 Nov 2015 20:16:22 +0000 (15:16 -0500)]
Fix instruction skipping when using software single step in GDBServer

Without this patch, when doing a software single step, with for example
a conditional breakpoint, gdbserver would wrongly avance the pc of
breakpoint_len and skips an instruction.

This is due to gdbserver assuming that it's hardware single stepping.
When it resumes from the breakpoint address it expects the trap to be
caused by ptrace and if it's rather caused by a software breakpoint
it assumes this is a permanent breakpoint and that it needs to skip
over it.

However when software single stepping, this breakpoint is legitimate as
it's the reinsert breakpoint gdbserver has put in place to break at
the next instruction. Thus gdbserver wrongly advances the pc and skips
an instruction.

This patch fixes this behavior so that gdbserver checks if it is a
reinsert breakpoint from software single stepping. If it is it won't
advance the pc. And if there's no reinsert breakpoint there we assume
then that it's a permanent breakpoint and advance the pc.

Here's a commented log of what would happen before and after the fix on
gdbserver :

/* Here there is a conditional breakpoint at 0x10428 that needs to be
stepped over. */

Need step over [LWP 11204]? yes, found breakpoint at 0x10428
...
/* e7f001f0 is a breakpoint instruction on arm
   Here gdbserver writes the software breakpoint we would like to hit
*/
Writing e7f001f0 to 0x0001042c in process 11204
...
Resuming lwp 11220 (continue, signal 0, stop not expected)
  pending reinsert at 0x10428
stop pc is 00010428
  continue from pc 0x10428
...

/* Here gdbserver hit the software breakpoint that was in place
   for the step over */

stop pc is 0001042c
pc is 0x1042c
step-over for LWP 11220.11220 executed software breakpoint
Finished step over.
Could not find fast tracepoint jump at 0x10428 in list (reinserting).

/* Here gdbserver writes back the original instruction */
Writing e50b3008 to 0x0001042c in process 11220
Step-over finished.
Need step over [LWP 11220]? No

/* Here because gdbserver assumes this is a permenant breakpoint it advances
the pc of breakpoint_len, in this case 4 bytes, so we have just skipped
the instruction that was written back here :
Writing e50b3008 to 0x0001042c in process 11220
*/

stop pc is 00010430
pc is 0x10430
Need step over [LWP 11220]? No, no breakpoint found at 0x10430
Proceeding, no step-over needed
proceed_one_lwp: lwp 11220
stop pc is 00010430

This patch fixes this situation and we get the right behavior :

Writing e50b3008 to 0x0001042c in process 11245
Hit a gdbserver breakpoint.
Hit a gdbserver breakpoint.
Step-over finished.
proceeding all threads.
Need step over [LWP 11245]? No
stop pc is 0001042c
pc is 0x1042c
Need step over [LWP 11245]? No, no breakpoint found at 0x1042c
Proceeding, no step-over needed
proceed_one_lwp: lwp 11245
stop pc is 0001042c
pc is 0x1042c
Resuming lwp 11245 (continue, signal 0, stop not expected)
stop pc is 0001042c
  continue from pc 0x1042c

It also works if the value at 0x0001042c is a permanent breakpoint.
If so gdbserver will finish the step over, remove the reinserted breakpoint,
resume at that location and on the next SIGTRAP gdbserver will trigger
the advance PC condition as reinsert_breakpoint_inserted_here will be false.

I also tested this against bp-permanent.exp on arm (with a work in progress
software single step patchset) without any regressions.

It's also tested against x86 bp-permanent.exp without any regression.

So both software and hardware single step are tested.

No regressions on Ubuntu 14.04 on ARMv7 and x86.
With gdbserver-{native,extended} / { -marm -mthumb }

gdb/gdbserver/ChangeLog:

* linux-low.c (linux_wait_1): Fix pc advance condition.
* mem-break.c (reinsert_breakpoint_inserted_here): New function.
* mem-break.h (reinsert_breakpoint_inserted_here): New declaration.

8 years agoFix breakpoint size when stepping over a permanent breakpoint in GDBServer.
Antoine Tremblay [Mon, 30 Nov 2015 20:08:04 +0000 (15:08 -0500)]
Fix breakpoint size when stepping over a permanent breakpoint in GDBServer.

When manually stepping over a permanent breakpoint on ARM we need to fetch the
right breakpoint size based on the current instruction set used.

Since this is not encoded in the stop_pc, the instruction mode needs to be
fetched from the CPSR register.

This is done by introducing a new target operation called :
breakpoint_kind_from_current_state.

For other targets that do not need this, breakpoint_kind_from_pc is used.

No regressions, tested on ubuntu 14.04 ARMv7 and x86.
With gdbserver-{native,extended} / { -marm -mthumb }

gdb/gdbserver/ChangeLog:

* linux-arm-low.c (arm_is_thumb_mode): New function.
(arm_breakpoint_at): Use arm_is_thumb_mode.
(arm_breakpoint_kind_from_current_state): New function.
(struct linux_target_ops) <breakpoint_kind_from_current_state>:
Initialize.
* linux-low.c (linux_wait_1): Call breakpoint_kind_from_current_state.
(linux_breakpoint_kind_from_current_state): New function.
(struct target_ops <breakpoint_kind_from_current_state>: Initialize.
* linux-low.h (struct linux_target_ops)
<breakpoint_kind_from_current_state>: New field.
* target.h (struct target_ops): Likewise.
(target_breakpoint_kind_from_current_state): New macro.

8 years agogdbserver: don't exit until GDB disconnects
Pedro Alves [Mon, 30 Nov 2015 16:05:27 +0000 (16:05 +0000)]
gdbserver: don't exit until GDB disconnects

When testing with "target remote" with "maint set target-non-stop on",
we regressions like this:

  Running /home/pedro/gdb/mygit/build/../src/gdb/testsuite/gdb.threads/continue-pending-after-query.exp ...
  FAIL: gdb.threads/continue-pending-after-query.exp: iter 4: continue until exit
  FAIL: gdb.threads/continue-pending-after-query.exp: iter 6: continue until exit
  FAIL: gdb.threads/continue-pending-after-query.exp: iter 10: continue until exit

  === gdb Summary ===

  # of expected passes            28
  # of unexpected failures        3

where gdb.log shows:

  continue
  Continuing.
  Remote communication error.  Target disconnected.: Connection reset by peer.
  (gdb) FAIL: gdb.threads/continue-pending-after-query.exp: iter 4: continue until exit

Enabling gdb + gdbserver debug logs we see:

  gdbserver:  <<<< exiting linux_wait_1
  gdbserver: handling possible serial event
  gdbserver: Writing resume reply for LWP 11089.11089:0
  gdbserver: handling possible serial event
  gdbserver: GDBserver exiting

GDB: Packet received: OK
GDB: infrun: prepare_to_wait
GDB: Sending packet: $vStopped#55...Packet received: W0;process:2b51
GDB: Sending packet: $vStopped#55...Packet received: OK
GDB: infrun: target_wait (-1.0.0, status) =
GDB: infrun:   -1.0.0 [Thread 0],
GDB: infrun:   status->kind = no-resumed
GDB: Sending packet: $Hgp2b51.2b51#41...Remote connection closed
    (gdb) FAIL: gdb.threads/continue-pending-after-query.exp: iter 1: continue until exit

Notice the "Packet received: W0;process:2b51" followed by
vStopped->OK.

That means the process exit notification was successfully sent to GDB
and GDB fetched it.  That makes gdbserver exit, in
server.c:process_serial_event:

  if (!extended_protocol && have_ran && !target_running ())
    {
      /* In non-stop, defer exiting until GDB had a chance to query
 the whole vStopped list (until it gets an OK).  */
      if (QUEUE_is_empty (notif_event_p, notif_stop.queue))
{
  /* Be transparent when GDB is connected through stdio -- no
     need to spam GDB's console.  */
  if (!remote_connection_is_stdio ())
    fprintf (stderr, "GDBserver exiting\n");
  remote_close ();
  exit (0);
}
    }

However, GDB is still busy processing an earlier "no-resumed" event,
and sends a "Hg" packet, which errors out with "Remote connection
closed".  IOW, it's not enough to wait for GDB to query the whole
vStopped list, gdbserver needs to wait until the exit event is really
processed.

The fix is to make gdbserver not disconnect until gdb does.

Tested on x86_64 Fedora, native gdbserver, remote + extended-remote +
with and without "maint set target-non-stop on".

gdb/gdbserver/ChangeLog:
2015-10-14  Pedro Alves  <palves@redhat.com>

* remote-utils.c (readchar): Don't print "Got EOF" unless
debugging gdbserver.
* server.c (captured_main): Exit gdbserver if gdb disconnects when
in "target remote" mode and there are no processes left to debug.
(process_serial_event): Remove 'have_ran' static local and remove
logic that exits gdbserver in "target remote" mode.

8 years agogdbserver/linux: Always wake up event loop after resume
Pedro Alves [Mon, 30 Nov 2015 16:05:26 +0000 (16:05 +0000)]
gdbserver/linux: Always wake up event loop after resume

Running killed-outside.exp in with "maint set target-non-stop on"
hangs currently.  This test has the inferior process die with a
SIGKILL while stopped.  gdbserver gets a SIGCHLD and reacts by
retrieveing the SIGKILL events out of waitpid.  But because the
process is not resumed from GDB's perspective, the event is left
pending.  When GDB resumes the process afterwards, the process is not
really resumed because it already has the event pending.  But nothing
wakes up the event loop to consume the event.

Handle this in the same way nat/linux-nat.c:linux_nat_resume handles
this.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* linux-low.c (linux_resume): Wake up the event loop before
returning.

8 years agogdbserver:prepare_access_memory: pick another thread
Pedro Alves [Mon, 30 Nov 2015 16:05:26 +0000 (16:05 +0000)]
gdbserver:prepare_access_memory: pick another thread

Say GDB wants to access the inferior process's memory.  The current
remote general thread is 3, but GDB's switched to thread 2.  Because
both threads are of the same process, GDB skips making the remote
thread be thread 2 as well (sending an Hg packet) before accessing
memory (remote.c:set_general_process).  However, if thread 3 has
exited meanwhile, thread 3 no longer exists on the server and
gdbserver points current_thread to NULL.  The result is the memory
access fails, even through the process still exists.

Fix this by making prepare_to_access memory select the thread to
access memory through.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* mem-break.c (check_gdb_bp_preconditions): Remove current_thread
check.
(set_gdb_breakpoint): If prepare_to_access_memory fails, set *ERR
to -1.
* target.c (struct thread_search): New structure.
(thread_search_callback): New function.
(prev_general_thread): New global.
(prepare_to_access_memory, done_accessing_memory): New functions.
* target.h (prepare_to_access_memory, done_accessing_memory):
Replace macros with function declarations.

8 years agoImplement TARGET_WAITKIND_NO_RESUMED in the remote protocol
Pedro Alves [Mon, 30 Nov 2015 16:05:25 +0000 (16:05 +0000)]
Implement TARGET_WAITKIND_NO_RESUMED in the remote protocol

Testing with "maint set target-non-stop on" causes regressions in
tests that rely on TARGET_WAITKIND_NO_RESUMED, which isn't modelled on
the RSP.  In real all-stop, gdbserver detects the situation and
reporst error to GDB, and so the tests (e.g.,
gdb.threads/no-unwaited-for-left.exp) at fail quickly.  But with
"maint set target-non-stop on", GDB instead hangs forever waiting for
a stop reply that never comes, and so the tests take longer to time
out.

This adds a new "N" stop reply packet that maps 1-1 to
TARGET_WAITKIND_NO_RESUMED.

gdb/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

PR 14618
* NEWS (New remote packets): Mention the N stop reply.
* remote.c (remote_protocol_features): Add "no-resumed" entry.
(remote_query_supported): Report no-resumed+ support.
(remote_parse_stop_reply): Handle 'N'.
(process_stop_reply): Handle TARGET_WAITKIND_NO_RESUMED.
(remote_wait_as): Handle 'N' / TARGET_WAITKIND_NO_RESUMED.
(_initialize_remote): Register "set/show remote
no-resumed-stop-reply" commands.

gdb/doc/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

PR 14618
* gdb.texinfo (Stop Reply Packets): Document the N stop reply.
(Remote Configuration): Add the "set/show remote
no-resumed-stop-reply" to the available settings table.
(General Query Packets): Document the "no-resumed" qSupported
feature.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

PR 14618
* linux-low.c (linux_wait_1): If the last resumed thread is gone,
report TARGET_WAITKIND_NO_RESUMED.
* remote-utils.c (prepare_resume_reply): Handle
TARGET_WAITKIND_NO_RESUMED.
* server.c (report_no_resumed): New global.
(handle_query) <qSupported>: Handle "no-resumed+".  Report
"no-resumed+" support.
(resume): When the target reports TARGET_WAITKIND_NO_RESUMED, only
return error if the client doesn't support no-resumed events.
(push_stop_notification): New function.
(handle_target_event): Use it.  Report TARGET_WAITKIND_NO_RESUMED
events if the client supports them.

gdb/testsuite/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdb.threads/no-unwaited-for-left.exp: Remove setup_kfail calls.

8 years agoinfrun: Fix TARGET_WAITKIND_NO_RESUMED handling in non-stop mode
Pedro Alves [Mon, 30 Nov 2015 16:05:24 +0000 (16:05 +0000)]
infrun: Fix TARGET_WAITKIND_NO_RESUMED handling in non-stop mode

Running the testsuite against gdbserver with "maint set target-non-stop on"
stumbled on a set of problems.  See code comments for details.

This handles my concerns expressed in PR14618.

gdb/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

PR 14618
* infrun.c (handle_no_resumed): New function.
(handle_inferior_event_1) <TARGET_WAITKIND_NO_RESUMED>: Defer to
handle_no_resumed.

8 years agotestsuite: Range stepping and non-stop mode
Pedro Alves [Mon, 30 Nov 2015 16:05:23 +0000 (16:05 +0000)]
testsuite: Range stepping and non-stop mode

The range-stepping tests fail with "maint set target-non-stop on" mode
because exec_cmd_expect_vCont_count doesn't know that in non-stop
mode, vCont's reply is simply "OK".

gdb/testsuite/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* lib/range-stepping-support.exp (exec_cmd_expect_vCont_count):
Handle non-stop mode vCont replies.

8 years agogdbserver: fix killed-outside.exp
Pedro Alves [Mon, 30 Nov 2015 16:05:23 +0000 (16:05 +0000)]
gdbserver: fix killed-outside.exp

killed-outside.exp regresses with "maint set target-non-stop on".  The
logs show:

 (gdb) continue
 Continuing.
 infrun: clear_proceed_status_thread (Thread 9028.9028)
 infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT)
 infrun: proceed: resuming Thread 9028.9028
 Sending packet: $Z0,3615a03966,1#4b...  Notification received: Stop:X9;process:2344
 Packet received: E01
 Sending packet: $Z0,3615a13970,1#47...Packet received: E01
 Sending packet: $Z0,3615a14891,1#4a...Packet received: E01
 infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [Thread 9028.9028] at 0x4005e4
 Sending packet: $vCont;c:p2344.2344#1a...Packet received: E.target not running.
 Sending packet: $qXfer:threads:read::0,fff#03...Packet received: l<threads>\n</threads>\n
 Sending packet: $vStopped#55...Packet received: OK
 Unexpected vCont reply in non-stop mode: E.target not running.
 (gdb) remote_async_inferior_event_handler
 infrun: target_wait (-1.0.0, status) =
 infrun:   9028.0.0 [process 9028],
 infrun:   status->kind = signalled, signal = GDB_SIGNAL_KILL
 infrun: TARGET_WAITKIND_SIGNALLED

 Program terminated with signal SIGKILL, Killed.
 The program no longer exists.
 infrun: stop_waiting
 infrun: clear_step_over_info
 infrun: stop_all_threads
 remote_thread_exit_events(1)

Note the "Unexpected vCont reply" error.

I traced it to a problem in status_pending_p_callback.  It resumes an
LWP when it shouldn't.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* linux-low.c (thread_still_has_status_pending_p): Don't check
vCont;t here.
(lwp_resumed): New function.
(status_pending_p_callback): Return early if the LWP is not
supposed to be resumed.

8 years agoRemote thread create/exit events
Pedro Alves [Mon, 30 Nov 2015 16:05:21 +0000 (16:05 +0000)]
Remote thread create/exit events

When testing with "maint set target-non-stop on", a few
threading-related tests expose an issue that requires new RSP packets.

Say there are 3 threads running, 1-3.  If GDB tries to stop thread 1,
2 and 3, and then waits for their stops, but meanwhile say, thread 2
exits, GDB hangs forever waiting for a stop for thread 2 that won't
ever happen.

This patch fixes the issue by adding support for thread exit events to
the protocol.  However, we don't want these always enabled, as they're
useless most of the time, and would slow down remote debugging.  So I
made it so that GDB can enable/disable them, and then made gdb do that
around the cases that need it, which currently is only
infrun.c:stop_all_threads.

In turn, if we have thread exit events, then the extra "thread x
exited" traffic slows down attach-many-short-lived-threads.exp enough
that gdb has trouble keeping up with new threads that are spawned
while gdb tries to stop existing ones.  To fix that I added support
for the counterpart thread created events too.  Enabling those when we
try to stop threads ensures that new threads never get a chance to
themselves start new threads, killing the race.

gdb/doc/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdb.texinfo (Remote Configuration): List "set/show remote
thread-events" command in configuration table.
(Stop Reply Packets): Document "T05 create" stop
reason and 'w' stop reply.
(General Query Packets): Document QThreadEvents packet.  Document
QThreadEvents qSupported feature.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* linux-low.c (handle_extended_wait): Assert that the LWP's
waitstatus is TARGET_WAITKIND_IGNORE.  If GDB wants to hear about
thread create events, leave the new child's status pending.
(linux_low_filter_event): If GDB wants to hear about thread exit
events, leave the LWP marked dead and don't delete it.
(linux_wait_for_event_filtered): Don't check for thread exit.
(filter_exit_event): New function.
(linux_wait_1): Use it, when returning an exit event.
(linux_resume_one_lwp_throw): Assert that the LWP's
waitstatus is TARGET_WAITKIND_IGNORE.
* remote-utils.c (prepare_resume_reply): Handle
TARGET_WAITKIND_THREAD_CREATED and TARGET_WAITKIND_THREAD_EXITED.
* server.c (report_thread_events): New global.
(handle_general_set): Handle QThreadEvents.
(handle_query) <qSupported>: Handle and report QThreadEvents+;
(handle_target_event): Handle TARGET_WAITKIND_THREAD_CREATED and
TARGET_WAITKIND_THREAD_EXITED.
* server.h (report_thread_events): Declare.

gdb/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* NEWS (New commands): Mention "set/show remote thread-events"
commands.
(New remote packets): Mention thread created/exited stop reasons
and QThreadEvents packet.
* infrun.c (disable_thread_events): New function.
(stop_all_threads): Disable/enable thread create/exit events.
Handle TARGET_WAITKIND_THREAD_EXITED.
(handle_inferior_event_1): Handle TARGET_WAITKIND_THREAD_CREATED
and TARGET_WAITKIND_THREAD_EXITED.
* remote.c (remove_child_of_pending_fork): Also remove threads of
threads that have TARGET_WAITKIND_THREAD_EXITED events.
(remote_parse_stop_reply): Handle "create" magic register.  Handle
'w' stop reply.
(initialize_remote): Install remote_thread_events as
to_thread_events target hook.
(remote_thread_events): New function.
* target-delegates.c: Regenerate.
* target.c (target_thread_events): New function.
* target.h (struct target_ops) <to_thread_events>: New field.
(target_thread_events): Declare.
* target/waitstatus.c (target_waitstatus_to_string): Handle
TARGET_WAITKIND_THREAD_CREATED and TARGET_WAITKIND_THREAD_EXITED.
* target/waitstatus.h (enum target_waitkind)
<TARGET_WAITKIND_THREAD_CREATED, TARGET_WAITKIND_THREAD_EXITED):
New values.

8 years agoMake dprintf-non-stop.exp cope with remote testing
Pedro Alves [Mon, 30 Nov 2015 16:05:19 +0000 (16:05 +0000)]
Make dprintf-non-stop.exp cope with remote testing

Testing with the extended-remote board with "maint set target-non-stop
on" shows a dprintf-non-stop.exp regression.  The issue is simply that
the test is expecting output that is only valid for the native target:

 native:

  [process 8676] #1 stopped.

 remote:

  [Thread 8900.8900] #1 stopped.

In order to expose this without "maint set target-non-stop on", this
restarts gdb with non-stop mode already enabled.

gdb/testsuite/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdb.base/dprintf-non-stop.exp: Use build_executable instead of
prepare_for_testing.  Start gdb with "set non-stop on" appended to
GDBFLAGS.  Lax expected stop output.

8 years agogdbserver resume_stop handling bug
Pedro Alves [Mon, 30 Nov 2015 16:05:19 +0000 (16:05 +0000)]
gdbserver resume_stop handling bug

Running attach-many-short-lived-threads.exp with the extended-remote
board with "maint set target-non-stop on" times out -- the attach
never completes.  Enabling infrun debug logs, we see that GDB is stuck
stopping all threads:

 infrun: target_wait (-1.0.0, status) =
 infrun:   1639.22213.0 [Thread 1639.22213],
 infrun:   status->kind = stopped, signal = GDB_SIGNAL_0
 infrun:   Thread 1639.22260 not executing
 infrun:   Thread 1639.22256 not executing
 infrun:   Thread 1639.22258 not executing
 infrun:   Thread 1639.22257 not executing
 infrun:   Thread 1639.22259 not executing
 infrun:   Thread 1639.22255 not executing
 infrun:   Thread 1639.22253 executing, already stopping
 infrun:   Thread 1639.22251 executing, already stopping
 infrun:   Thread 1639.22252 executing, already stopping
 infrun:   Thread 1639.22250 executing, already stopping
 infrun:   Thread 1639.22254 executing, already stopping
 infrun:   Thread 1639.22247 executing, already stopping
 infrun:   Thread 1639.22213 not executing
 infrun:   Thread 1639.22207 not executing
 infrun:   Thread 1639.22201 not executing
 infrun:   Thread 1639.22219 not executing
 infrun:   Thread 1639.1639 not executing
 ** HANG HERE **

GDB is waiting for the stop replies of any of those "already stopping"
threads.  Take 22253 for example.  On the gdbserver logs we see:

 ...
 resume_stop request for LWP 22253
 stopping LWP 22253
 Sending sigstop to lwp 22253
 linux_resume done
 ...

and:

 my_waitpid (-1, 0x40000001)
 my_waitpid (-1, 0x80000001): status(3057f), 22253
 LWFE: waitpid(-1, ...) returned 22253, ERRNO-OK
 LLW: waitpid 22253 received Trace/breakpoint trap (stopped)
 pc is 0x3615ef4ce1
 HEW: Got clone event from LWP 22253, new child is LWP 22259

but from here on, we never see any other event for LWP 22253.  In
particular, we never see the expected SIGSTOP (from "Sending sigstop"
above).  The issue is that linux_resume_stopped_resumed_lwps never
re-resumes the 22253 after the clone event.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* linux-low.c (resume_stopped_resumed_lwps): Don't check whether
the thread's last_resume_kind was resume_stop.

8 years agogdbserver crash if gdb attaches too fast
Pedro Alves [Mon, 30 Nov 2015 16:05:18 +0000 (16:05 +0000)]
gdbserver crash if gdb attaches too fast

With "maint set target-non-stop on", the attach tests occasionally
crash gdbserver.

Basically, gdb attaches with vAttach;PID, and then shortly after reads
the xml target description for that process, to figure out the
process' architecture.  On the gdbserver side, the target description
is only filled in when the first process/thread in the thread group
reports its initial PTRACE_ATTACH SIGSTOP.  So if GDB is fast enough,
it can read the target description _before_ that initial stop, and
then gdbserver dies dereferencing a NULL tdesc pointer.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* linux-low.c (linux_attach): In non-stop mode, wait for one stop
before returning.

8 years agoNew vCtrlC packet, non-stop mode equivalent of \003
Pedro Alves [Mon, 30 Nov 2015 16:05:17 +0000 (16:05 +0000)]
New vCtrlC packet, non-stop mode equivalent of \003

There's currently no non-stop equivalent of the all-stop ^C (\003)
"packet" that GDB sends when a ctrl-c is pressed while a foreground
command is active.  There's vCont;t, but that's defined to cause a
"signal 0" stop.

This fixes many tests that type ^C, when testing with extended-remote
with "maint set target-non-stop on".  E.g.:

 Continuing.
 talk to me baby
 PASS: gdb.base/interrupt.exp: process is alive
 a
 a
 PASS: gdb.base/interrupt.exp: child process ate our char
 ^C
 [Thread 22730.22730] #1 stopped.
 0x0000003615ee6650 in __read_nocancel () at ../sysdeps/unix/syscall-template.S:81
 81      T_PSEUDO (SYSCALL_SYMBOL, SYSCALL_NAME, SYSCALL_NARGS)
 (gdb) FAIL: gdb.base/interrupt.exp: send_gdb control C
 p func1 ()

gdb/
2015-11-30  Pedro Alves  <palves@redhat.com>

* NEWS (New remote packets): Mention vCtrlC.
* remote.c (PACKET_vCtrlC): New enum value.
(async_remote_interrupt): Call target_interrupt instead of
target_stop.
(remote_interrupt_as): Remove 'ptid' parameter.
(remote_interrupt_ns): New function.
(remote_stop): Adjust.
(remote_interrupt): If the target is in non-stop mode, try
interrupting with vCtrlC.
(initialize_remote): Install set remote ctrl-c packet.

gdb/doc/
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdb.texinfo (Bootstrapping): Add "interrupting remote targets"
anchor.
(Packets): Document vCtrlC.

gdb/gdbserver/
2015-11-30  Pedro Alves  <palves@redhat.com>

* server.c (handle_v_requests): Handle vCtrlC.

8 years agoremote: stop reason and watchpoint data address per thread
Pedro Alves [Mon, 30 Nov 2015 16:05:16 +0000 (16:05 +0000)]
remote: stop reason and watchpoint data address per thread

Running local-watch-wrong-thread.exp with "maint set target-non-stop
on" exposes that gdb/remote.c only records whether the target stopped
for a breakpoint/watchpoint plus the watchpoint data address *for the
last reported remote event*.  But in non-stop mode, we need to keep
that info per-thread, as each thread can end up with its own
last-status pending.

gdb/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* remote.c (struct remote_state) <remote_watch_data_address,
stop_reason>: Delete fields.
(struct private_thread_info) <stop_reason, watch_data_address>:
New fields.
(resume_clear_thread_private_info): New function.
(append_pending_thread_resumptions): Call it.
(remote_resume): Clear all threads' private info.
(process_stop_reply): Adjust.
(remote_wait_as): Don't reference remote_state's stop_reason
field.
(remote_stopped_by_sw_breakpoint)
(remote_stopped_by_hw_breakpoint, remote_stopped_by_watchpoint)
(remote_stopped_data_address): Adjust to refer get data from the
current thread.

8 years agogdbserver crash running gdb.threads/non-ldr-exc-1.exp
Pedro Alves [Mon, 30 Nov 2015 16:05:15 +0000 (16:05 +0000)]
gdbserver crash running gdb.threads/non-ldr-exc-1.exp

This fixes a gdbserver crash when running
gdb.threads/non-ldr-exc-1.exp with "maint set target-non-stop on".
The problem is that qSymbol is called when gdbserver has
current_thread == NULL.

gdb/gdbserver/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdbthread.h (find_any_thread_of_pid): Declare.
* inferiors.c (thread_of_pid, find_any_thread_of_pid): New
functions.
* server.c (handle_query): If current_thread is NULL, look for
another thread of the selected process.

8 years agoattach + target always in non-stop mode: stop all threads
Pedro Alves [Mon, 30 Nov 2015 16:05:14 +0000 (16:05 +0000)]
attach + target always in non-stop mode: stop all threads

When running with "maint set target-non-stop on", and in all-stop
mode, nothing is stopping all threads after attaching.  vAttach in
non-stop can leave all threads running and GDB has to explicitly pause
them.

This is not visible with the native target, as in that case, attach
always stops all threads (the core re-resumes them in case of
"attach&").

In addition, it's not defined which thread manages to report the
initial attach stop, so always pick the lowest one (otherwise
multi-attach.exp regresses).

gdb/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* infcmd.c (attach_post_wait): If the target is always in non-stop
mode, and the UI is in all-stop mode, stop all threads and pick
the one with lowest number as current.

8 years agoRemote all-stop-on-top-of-non-stop
Pedro Alves [Mon, 30 Nov 2015 16:05:13 +0000 (16:05 +0000)]
Remote all-stop-on-top-of-non-stop

This is the first pass at implementing support for all-stop mode
running against the remote target using the non-stop variant of the
protocol.

The trickiest part here is the initial connection setup/synching.  We
need to fetch all inferiors' target descriptions etc. before stopping
threads, because stop_all_threads needs to read the threads' registers
(to record each thread's stop_pc).  But OTOH, the initial inferior
setup (target_post_attach, post_create_inferior, etc.), only works
correctly if the inferior is stopped...  So I've split that initial
setup part from attach_command_post_wait to a separate function, and
added a "still needs setup" flag to the inferior structure.  This is
similar to gdbserver/linux-low.c's handling of discovering the
process's target description).  Then if on connection all threads of
the remote inferior are running, when we go about stopping them, as
soon as they stop we call setup_inferior, from within
stop_all_threads.

Also, in all-stop, we need to process all the initial stop replies to
learn about all the pending signal the threads may already be stopped
for, and pick the one to report as current.  This is exposed by
gdb.threads/reconnect-signal.exp.

gdb/
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdbthread.h (switch_to_thread_no_regs): Declare.
* infcmd.c (setup_inferior): New function, factored out from ...
(attach_command_post_wait): ... this.  Rename to ...
(attach_post_wait): ... this.  Replace parameter async_exec with
attach_post_wait_mode parameter.  Adjust.
(enum attach_post_wait_mode): New enum.
(struct attach_command_continuation_args): Replace 'async_exec'
field with 'mode' field.
(attach_command_continuation): Adjust.
(attach_command): Add comment.  Mark the inferior as needing
setup.  Adjust to use enum attach_post_wait_mode.
(notice_new_inferior): Use switch_to_thread_no_regs.  Adjust to
use enum attach_post_wait_mode.
* inferior.h (setup_inferior): Declare.
(struct inferior) <needs_setup>: New field.
* infrun.c (set_last_target_status): Make extern.
(stop_all_threads): Make extern.  Setup inferior, if necessary.
* infrun.h (set_last_target_status, stop_all_threads): Declare.
* remote-notif.c (remote_async_get_pending_events_handler)
(handle_notification): Replace non_stop checks with
target_is_non_stop_p() checks.
* remote.c (remote_notice_new_inferior): Remove non_stop check.
(remote_update_thread_list): Replace non_stop check with
target_is_non_stop_p() check.
(print_one_stopped_thread): New function.
(process_initial_stop_replies): New 'from_tty' parameter.
"Notice" all new live inferiors after storing initial stops as
pending status in each corresponding thread.  If all-stop, stop
all threads, try picking a signalled thread as current, and print
the status of that one thread.  Record the last target status.
(remote_start_remote): Replace non_stop checks with
target_is_non_stop_p() checks.  Don't query for the remote current
thread of use qOffsets here.  Pass from_tty to
process_initial_stop_replies.
(extended_remote_attach): Replace non_stop checks with
target_is_non_stop_p() checks.
(extended_remote_post_attach): Send qOffsets here.
(remote_vcont_resume, remote_resume, remote_stop)
(remote_interrupt, remote_parse_stop_reply, remote_wait): Replace
non_stop checks with target_is_non_stop_p() checks.
(remote_async): If target is non-stop, mark/clear the pending
events token.
* thread.c (switch_to_thread_no_regs): New function.

8 years agoFix mi-nonstop.exp with extended-remote
Pedro Alves [Mon, 30 Nov 2015 16:05:12 +0000 (16:05 +0000)]
Fix mi-nonstop.exp with extended-remote

Testing with "maint set target-non-stop on" makes mi-nonstop.exp run
with the extended-remote board.  That reveals that mi-nonstop.exp is
using the wrong predicate to check for "using remote protocol".

This is not visible today because non-stop tests all fail to run with
extended-remote board, because they spawn gdb and then do "set
non-stop on".  However, with that board, gdb connects to the gdbserver
from within mi_gdb_start, and changing non-stop when already connected
doesn't work.  Fix that by instead enabling non-stop mode on gdb's
command line.

gdb/testsuite/ChangeLog:
2015-11-30  Pedro Alves  <palves@redhat.com>

* gdb.mi/mi-nonstop.exp: Append "set non-stop on" to GDBFLAGS
instead of issuing "-gdb-set non-stop 1" after starting gdb.
Use mi_is_target_remote instead of checking "is_remote target".
* lib/gdb.exp (gdb_is_target_remote): Rename to ...
(gdb_is_target_remote_prompt): ... this, and add 'prompt_regexp'
parameter.
(gdb_is_target_remote): Reimplement.
* lib/mi-support.exp (mi_is_target_remote): New procedure.

8 years agoFix ChangeLog entry
Pedro Alves [Mon, 30 Nov 2015 18:32:24 +0000 (18:32 +0000)]
Fix ChangeLog entry

There should be only one date in multi-author entries.

8 years agoobjdump: Handle 32-bit base address in debug_ranges / debug_loc.
Andrew Burgess [Fri, 16 Oct 2015 09:58:02 +0000 (11:58 +0200)]
objdump: Handle 32-bit base address in debug_ranges / debug_loc.

When the DWARF address size is 32-bit, but the host machine is 64-bit,
objdump fails to spot base addresses specified in the .debug_ranges and
.debug_loc lists.

As an example, here is the output when dumping an example .debug_ranges
section with the pre-patched objdump:

    Contents of the .debug_ranges section:

        Offset   Begin    End
        00000000 ffffffff 00000004 (start > end)
        00000000 00000000 00000004
        00000000 ffffffff 00000008 (start > end)
        00000000 00000000 00000004
        00000000 <End of list>

And this is what the same section looks like when dumped with the
patched version of objdump:

    Contents of the .debug_ranges section:

        Offset   Begin    End
        00000000 ffffffff 00000004 (base address)
        00000000 00000004 00000008
        00000000 ffffffff 00000008 (base address)
        00000000 00000008 0000000c
        00000000 <End of list>

binutils/ChangeLog:

* dwarf.c (is_max_address): New function.
(display_loc_list): Remove out of date comment, use
is_max_address.
(display_debug_ranges): Likewise.

binutils/testsuite/ChangeLog:

* binutils-all/objdump.exp: Add test for .debug_ranges decode.
* binutils-all/dw2-ranges.S: New file.
* binutils-all/dw2-ranges.W: New file.

8 years agoAutomatic date update in version.in
GDB Administrator [Mon, 30 Nov 2015 00:00:08 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoUpdate my email address in MAINTAINERS.
Cary Coutant [Sun, 29 Nov 2015 01:19:52 +0000 (17:19 -0800)]
Update my email address in MAINTAINERS.

binutils/
* MAINTAINERS: Update my email address.

8 years agoAutomatic date update in version.in
GDB Administrator [Sun, 29 Nov 2015 00:00:07 +0000 (00:00 +0000)]
Automatic date update in version.in

8 years agoAdjust GDB to demangler API change
Pedro Alves [Sat, 28 Nov 2015 16:39:32 +0000 (16:39 +0000)]
Adjust GDB to demangler API change

Before commit 3a8724032abf, DEMANGLE_COMPONENT_CAST was used for both
casts and conversion operators.  We now have
DEMANGLE_COMPONENT_CONVERSION for the latter.

gdb/ChangeLog:
2014-11-28  Pedro Alves  <palves@redhat.com>

* cp-name-parser.y (conversion_op): Use
DEMANGLE_COMPONENT_CONVERSION instead of DEMANGLE_COMPONENT_CAST.

8 years agolibsanitizer merge from upstream r250806, compiler part.
Maxim Ostapenko [Sat, 28 Nov 2015 16:39:32 +0000 (16:39 +0000)]
libsanitizer merge from upstream r250806, compiler part.

gcc/

* asan.c (asan_emit_stack_protection): Don't pass local stack to
asan_stack_malloc_[n] anymore. Check if asan_stack_malloc_[n] returned
NULL and use local stack than.
(asan_finish_file): Insert __asan_version_mismatch_check_v[n] call
in addition to __asan_init.
* sanitizer.def (BUILT_IN_ASAN_INIT): Rename to __asan_init.
(BUILT_IN_ASAN_VERSION_MISMATCH_CHECK): Add new builtin call.
* asan.h (asan_intercepted_p): Handle new string builtins.
* ubsan.c (ubsan_use_new_style_p): New function.
(ubsan_instrument_float_cast): If location is unknown, assign
input_location to loc. Propagate loc to ubsan_create_data if
ubsan_use_new_style_p returned true.

config/

* bootstrap-asan.mk: Replace ASAN_OPTIONS=detect_leaks with
LSAN_OPTIONS=detect_leaks.

gcc/testsuite/

* c-c++-common/ubsan/float-cast-overflow-10.c: Adjust test.
* c-c++-common/ubsan/float-cast-overflow-8.c: Likewise.
* c-c++-common/ubsan/float-cast-overflow-9.c: Likewise.
* g++.dg/asan/default-options-1.C: Likewise.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@229112 138bc75d-0d04-0410-961f-82ee72b054a4

8 years agoPort libvtv to Solaris
Rainer Orth [Sat, 28 Nov 2015 16:39:32 +0000 (16:39 +0000)]
Port libvtv to Solaris

libstdc++-v3:
* acinclude.m4 (GLIBCXX_ENABLE_VTABLE_VERIFY) <solaris2*>: Use
-Wl,-R in VTV_CXXLINKFLAGS.
* configure: Regenerate.

* testsuite/18_support/bad_exception/23591_thread-1.c: Use
-fvtable-verify=none on Solaris 12+.

libgcc:
* Makefile.in (VTV_CFLAGS): New variable.
(vtv_start$(objext), vtv_end$(objext), vtv_end$(objext))
(vtv_start_preinit$(objext), vtv_end_preinit$(objext)): Use it.
* config.host (*-*-solaris2*): Add t-crtstuff-pic to tmake_file.
Add vtv_start.o, vtv_end.o, vtv_start_preinit.o, vtv_end_preinit.o
to extra_parts if $enable_vtable_verify = yes.

libvtv:
* configure.tgt (*-*-solaris2.[1-9]*): Declare supported.
* configure.ac: Call AC_USE_SYSTEM_EXTENSIONS.
<*-*-solaris2*>: Check for init priority support.
Check for getexecname, __fortify_fail, _obstack_begin.
(VTV_NO_OBSTACK): New conditional.
* configure: Regenerate.
* Makefile.am [VTV_NO_OBSTACK] (obstack.c): Use new condition.
Create empty config.h
* Makefile.in: Regenerate.

* vtv_rts.cc [HAVE_GETEXECNAME] (program_invocation_name): New
variable.
(read_section_offset_and_length) [HAVE_GETEXECNAME]: Set it.
(dl_iterate_phdr_callback) [HAVE_GETEXECNAME]: Set it.

(__fortify_fail): Wrap in HAVE___FORTIFY_FAIL
[!HAVE___FORTIFY_FAIL]: Provide non-Cygwin implementation.

(read_section_offset_and_length): Assert sh_size >= VTV_PAGE_SIZE.
(iterate_modules): Fix typo.
Use VTV_PAGE_SIZE.
(dl_iterate_phdr_callback): Fix typo.
Use VTV_PAGE_SIZE.
(__VLTChangePermission): Fix typos.

include:
* vtv-change-permission.h (VTV_PAGE_SIZE) [__sun__ && __svr4__ &&
__sparc__]: Define.

gcc:
* config/sol2.h (SUPPORTS_INIT_PRIORITY): Move up.
(STARTFILE_VTV_SPEC, ENDFILE_VTV_SPEC): Define.
(STARTFILE_SPEC): Use %(startfile_vtv).
(ENDFILE_SPEC): Use %(endfile_vtv).
(SUBTARGET_EXTRA_SPECS): Handle STARTFILE_VTV_SPEC,
ENDFILE_VTV_SPEC.

* gcc.c (LINK_COMMAND_SPEC): Move VTABLE_VERIFICATION_SPEC after %{L*}.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@230865 138bc75d-0d04-0410-961f-82ee72b054a4

8 years agoSH FDPIC backend support
Rich Felker [Sat, 28 Nov 2015 16:39:31 +0000 (16:39 +0000)]
SH FDPIC backend support

gcc/ChangeLog
* config.gcc: Handle --enable-fdpic.
* config/sh/constraints.md (Ccl): New constraint.
* config/sh/linux.h (SUBTARGET_LINK_EMUL_SUFFIX): Handle -mfdpic.
* config/sh/sh-c.c (sh_cpu_cpp_builtins): Add __FDPIC__ and
__SH_FDPIC__.
* config/sh/sh-mem.cc (expand_block_move): Support FDPIC for calls
to library functions.
* config/sh/sh-protos.h (function_symbol_result): New struct.
(function_symbol): Return function_symbol_result.
(sh_get_fdpic_reg_initial_val, sh_load_function_descriptor): New
declarations.
* config/sh/sh.c (TARGET_ASM_INTEGER, sh_assemble_integer):
Implement target hook.
  (TARGET_CANNOT_FORCE_CONST_MEM, sh_cannot_force_const_mem_p):
Likewise.
(sh_option_override): Force -fPIC if FDPIC is in effect.
  (sh_asm_output_addr_const_extra): Add UNSPEC_GOTFUNCDESC and
UNSPEC_GOTOFFFUNCDESC cases.
(prepare_move_operands): Use FDPIC initial GOT register for
TLS-related GOT access; inhibit cross-section address offset
constants for FDPIC.
(sh_assemble_integer): New function.
(sh_cannot_copy_insn_p): Inhibit copying insns that are FDPIC
PC-relative call sites.
(expand_ashiftrt): Adapt invocation of function_symbol.
(sh_expand_prologue): Inhibit PC-relative GOT address load for
FDPIC.
(nonpic_symbol_mentioned_p): Add cases for UNSPEC_GOTFUNCDESC and
UNSPEC_GOTOFFFUNCDESC.
(legitimize_pic_address): Resolve function symbols to function
descriptors for FDPIC.  Do not use GOT-relative addressing for
local data that may be read-only on FDPIC.
(sh_emit_storesi, sh_emit_storehi): New functions.
(sh_trampoline_init): Generate FDPIC trampolines.
(sh_function_ok_for_sibcall): Add TARGET_FDPIC check.
(sh_expand_sym_label2reg): Don't assume sibcalls are local.
(sh_output_mi_thunk): Generate FDPIC call.
(function_symbol): Return function_symbol_result.  For
SFUNC_STATIC on FDPIC, generate call site labels to use
PC-relative addressing rather than GOT-relative addressing.
(sh_conditional_register_usage): Make PIC register fixed and call used
when FDPIC is in effect.
(sh_legitimate_constant_p): Impose FDPIC
constant constraints.
(sh_cannot_force_const_mem_p, sh_load_function_descriptor)
(sh_get_fdpic_reg_initial_val): New functions.
* config/sh/sh.h (SUBTARGET_ASM_SPEC, SUBTARGET_LINK_EMUL_SUFFIX):
Handle -mfdpic.
(FDPIC_SELF_SPECS, SUBTARGET_DRIVER_SELF_SPECS)
(PIC_OFFSET_TABLE_REG_CALL_CLOBBERED)
(SH_OFFSETS_MUST_BE_WITHIN_SECTIONS_P): New macros.
(DRIVER_SELF_SPECS): Add SUBTARGET_DRIVER_SELF_SPECS and
FDPIC_SELF_SPECS.
(TRAMPOLINE_SIZE): Select trampoline size for FDPIC.
(ASM_PREFERRED_EH_DATA_FORMAT): Add EH format constraints for
FDPIC.
(ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX): Handle FDPIC case.
* config/sh/sh.md (UNSPEC_GOTFUNCDESC, UNSPEC_GOTOFFFUNCDESC): New
constants.
(calli_fdpic, call_valuei_fdpic, sibcalli_fdpic)
(sibcalli_pcrel_fdpic, sibcall_pcrel_fdpic, sibcall_valuei_fdpic)
(sibcall_valuei_pcrel_fdpic, sibcall_value_pcrel_fdpic)
(sym2GOTFUNCDESC, symGOTFUNCDESC2reg, sym2GOTOFFFUNCDESC)
(symGOTOFFFUNCDESC2reg): New patterns.
(udivsi3_i1, udivsi3_i4, udivsi3_i4_single, udivsi3,
*divsi_inv_call_combine, divsi3_i4, divsi3_i4_single, divsi3)
(ashlsi3, ashlsi3_d_call, ashrsi3_n, lshrsi3, lshrsi3_d_call)
(calli, call_valuei, call, call_value, sibcalli, sibcalli_pcrel)
(sibcall_pcrel, sibcall, sibcall_valuei, sibcall_valuei_pcrel)
(sibcall_value_pcrel, sibcall_value, GOTaddr2picreg, symGOT_load)
(symGOTOFF2reg, block_move_real, block_lump_real)
(block_move_real_i4, block_lump_real_i4): Add support for FDPIC
calls.
(mulsi3, ic_invalidate_line, initialize_trampoline, call_pop,
call_value_pop): Adjust for new function_symbol signature.
* config/sh/sh.opt (-mfdpic): New option.
* doc/install.texi (Options specification): Document
--enable-fdpic.
* doc/invoke.texi (SH Options): Document -mfdpic.

include/ChangeLog:
* longlong.h (udiv_qrnnd): Add FDPIC compatible version for SH.

libitm/ChangeLog:
* config/sh/sjlj.S (_ITM_beginTransaction): Bypass PLT calling
GTM_begin_transaction for compatibility with FDPIC.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@229438 138bc75d-0d04-0410-961f-82ee72b054a4

8 years agoPR other/61321 - demangler crash on casts in template parameters
Pedro Alves [Sat, 28 Nov 2015 16:39:31 +0000 (16:39 +0000)]
PR other/61321 - demangler crash on casts in template parameters

The fix for bug 59195:

 [C++ demangler handles conversion operator incorrectly]
 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59195

unfortunately makes the demangler crash due to infinite recursion, in
case of casts in template parameters.

For example, with:

 template<int> struct A {};
 template <typename Y> void function_temp(A<sizeof ((Y)(999))>) {}
 template void function_temp<int>(A<sizeof (int)>);

The 'function_temp<int>' instantiation above mangles to:

  _Z13function_tempIiEv1AIXszcvT_Li999EEE

The demangler parses this as:

typed name
  template
    name 'function_temp'
    template argument list
      builtin type int
  function type
    builtin type void
    argument list
      template                          (*)
        name 'A'
        template argument list
          unary operator
            operator sizeof
            unary operator
              cast
                template parameter 0    (**)
              literal
                builtin type int
                name '999'

And after the fix for 59195, due to:

 static void
 d_print_cast (struct d_print_info *dpi, int options,
       const struct demangle_component *dc)
 {
 ...
   /* For a cast operator, we need the template parameters from
      the enclosing template in scope for processing the type.  */
   if (dpi->current_template != NULL)
     {
       dpt.next = dpi->templates;
       dpi->templates = &dpt;
       dpt.template_decl = dpi->current_template;
     }

when printing the template argument list of A (what should be "<sizeof
(int)>"), the template parameter 0 (that is, "T_", the '**' above) now
refers to the first parameter of the the template argument list of the
'A' template (the '*' above), exactly what we were already trying to
print.  This leads to infinite recursion, and stack exaustion.  The
template parameter 0 should actually refer to the first parameter of
the 'function_temp' template.

Where it reads "for the cast operator" in the comment in d_print_cast
(above), it's really talking about a conversion operator, like:

  struct A { template <typename U> explicit operator U(); };

We don't want to inject the template parameters from the enclosing
template in scope when processing a cast _expression_, only when
handling a conversion operator.

The problem is that DEMANGLE_COMPONENT_CAST is currently ambiguous,
and means _both_ 'conversion operator' and 'cast expression'.

Fix this by adding a new DEMANGLE_COMPONENT_CONVERSION component type,
which does what DEMANGLE_COMPONENT_CAST does today, and making
DEMANGLE_COMPONENT_CAST just simply print its component subtree.

I think we could instead reuse DEMANGLE_COMPONENT_CAST and in
d_print_comp_inner still do:

 @@ -5001,9 +5013,9 @@ d_print_comp_inner (struct d_print_info *dpi, int options,
        d_print_comp (dpi, options, dc->u.s_extended_operator.name);
        return;

     case DEMANGLE_COMPONENT_CAST:
       d_append_string (dpi, "operator ");
 -     d_print_cast (dpi, options, dc);
 +     d_print_conversion (dpi, options, dc);
       return;

leaving the unary cast case below calling d_print_cast, but seems to
me that spliting the component types makes it easier to reason about
the code.

g++'s testsuite actually generates three symbols that crash the
demangler in the same way.  I've added those as tests in the demangler
testsuite as well.

And then this fixes PR other/61233 too, which happens to be a
demangler crash originally reported to GDB, at:
https://sourceware.org/bugzilla/show_bug.cgi?id=16957

Bootstrapped and regtested on x86_64 Fedora 20.

Also ran this through GDB's testsuite.  GDB will require a small
update to use DEMANGLE_COMPONENT_CONVERSION in one place it's using
DEMANGLE_COMPONENT_CAST in its sources.

libiberty/
2015-11-27  Pedro Alves  <palves@redhat.com>

        PR other/61321
        PR other/61233
        * demangle.h (enum demangle_component_type)
        <DEMANGLE_COMPONENT_CONVERSION>: New value.
        * cp-demangle.c (d_demangle_callback, d_make_comp): Handle
        DEMANGLE_COMPONENT_CONVERSION.
        (is_ctor_dtor_or_conversion): Handle DEMANGLE_COMPONENT_CONVERSION
        instead of DEMANGLE_COMPONENT_CAST.
        (d_operator_name): Return a DEMANGLE_COMPONENT_CONVERSION
        component if handling a conversion.
        (d_count_templates_scopes, d_print_comp_inner): Handle
        DEMANGLE_COMPONENT_CONVERSION.
        (d_print_comp_inner): Handle DEMANGLE_COMPONENT_CONVERSION instead
        of DEMANGLE_COMPONENT_CAST.
        (d_print_cast): Rename as ...
        (d_print_conversion): ... this.  Adjust comments.
        (d_print_cast): Rewrite - simply print the left subcomponent.
        * cp-demint.c (cplus_demangle_fill_component): Handle
        DEMANGLE_COMPONENT_CONVERSION.

        * testsuite/demangle-expected: Add tests.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@231020 138bc75d-0d04-0410-961f-82ee72b054a4

8 years agolibiberty TAGS
Mike Stump [Sat, 28 Nov 2015 16:39:31 +0000 (16:39 +0000)]
libiberty TAGS

  * Makefile.in (etags tags TAGS): Use && instead of ;.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@230270 138bc75d-0d04-0410-961f-82ee72b054a4

8 years ago(Makefiles): PATCH to include libcpp and libiberty in GCC etags
Jason Merrill [Sat, 28 Nov 2015 16:39:30 +0000 (16:39 +0000)]
(Makefiles): PATCH to include libcpp and libiberty in GCC etags

gcc/c/
* Make-lang.in (c.tags): Also include libcpp TAGS.
gcc/cp/
* Make-lang.in (c++.tags): Also include libcpp TAGS.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@229504 138bc75d-0d04-0410-961f-82ee72b054a4