binutils-gdb.git
6 years ago(SPARC/LEON) fix incorrect array return value printed by "finish"
Joel Brobecker [Fri, 4 May 2018 18:33:19 +0000 (13:33 -0500)]
(SPARC/LEON) fix incorrect array return value printed by "finish"

Consider the code in the gdb.ada/array_return.exp testcase, which
defines a function returning an array of 2 integers:

   type Data_Small is array (1 .. 2) of Integer;
   function Create_Small return Data_Small;

When doing a "finish" from inside function Create_Small, we expect
GDB to tell us that the return value was "(1, 1)". However, it currently
prints the wrong value:

    (gdb) finish
    Run till exit from #0  pck.create_small () at /[...]/pck.adb:5
    p () at /[...]/p.adb:10
    10         Large := Create_Large;
    Value returned is $1 = (0, 0)

This is a regression which I traced back to the following commit...

    | commit 1933fd8ee01ad2e74a9c6341bc40f54962a8f889
    | Date:   Fri May 19 03:06:19 2017 -0700
    | Subject: gdb: fix TYPE_CODE_ARRAY handling in sparc targets

... which, despite what the subject says, is not really about
TYPE_CODE_ARRAY handling, which is a bit of an implementation detail,
but about the GNU vectors extension.

The author of the patch equated TYPE_CODE_ARRAY with vectors, which
is not correct. Vectors are TYPE_CODE_ARRAY types with the TYPE_VECTOR
flag set. So at the very minimum, the patch should have been checking
for both TYPE_CODE_ARRAY and TYPE_VECTOR.

But, that's not the only thing that did not seem right to me. When
looking at the ABI, and at the summary of the implementation in GCC
of the calling conventions for that architecture:

                                size      argument     return value

      small integer              <4       int. reg.      int. reg.
      word                        4       int. reg.      int. reg.
      double word                 8       int. reg.      int. reg.

      _Complex small integer     <8       int. reg.      int. reg.
      _Complex word               8       int. reg.      int. reg.
      _Complex double word       16        memory        int. reg.

      vector integer            <=8       int. reg.       FP reg.
      vector integer             >8        memory         memory

      float                       4       int. reg.       FP reg.
      double                      8       int. reg.       FP reg.
      long double                16        memory         memory

      _Complex float              8        memory         FP reg.
      _Complex double            16        memory         FP reg.
      _Complex long double       32        memory         FP reg.

      vector float              any        memory         memory

      aggregate                 any        memory         memory

The nice thing about the patch above is that it nicely factorized
the code that determines how arguments are passed/returns. The bad
news is that the implementation, particularly for the handling of
arrays and vectors, doesn't seem to match the summary above. Hence,
the regression we observed.

So what I did was review and re-implement some of the predicate functions
according to the summary above. Because dejagnu crashes all our Solaris
machines real bad, I can't run the dejagnu testsuite there. So what I did
was test the patch with AdaCore's testsuite against leon3-elf, no
regression. I verified that this fixes the regression above while
at the same time still passing gdb.base/gnu_vector.exp (I transposed
that testcase to our testsuite), which is the testcase that was cited
in the commit above as seeing some FAIL->PASS improvements.

This patch also removes one assertion...

      gdb_assert (sparc_integral_or_pointer_p (type)
                  || (TYPE_CODE (type) == TYPE_CODE_ARRAY && len <= 8));

... because that assertion is really the "negative" of the other conditions
written in the same "if, else if, else [assert]" block in this function.
To me, this assertion forces us to maintain two versions of the same code,
and is an unnecessary burden. In particular, the above is not the
correct condition, and the ABI summary table above shows that we need
a more complex condition to describe the situations where we expect
arguments to be passed by register.

gdb/ChangeLog:

        * sparc-tdep.c (sparc_structure_return_p): Re-implement to
        match the ABI as summarized in GCC's gcc/config/sparc/sparc.c.
        (sparc_arg_by_memory_p): Renamed from sparc_arg_on_registers_p.
        Re-implement to match the ABI as summarized in GCC's
        gcc/config/sparc/sparc.c.  All callers updated.
        (sparc32_store_arguments): Remove assertion.

6 years agoMinor cleanups in printcmd.c
Tom Tromey [Fri, 27 Apr 2018 22:04:13 +0000 (16:04 -0600)]
Minor cleanups in printcmd.c

This changes decode_format to use skip_spaces, and changes printcmd.c
not to include tui.h, which apparently is not needed.

2018-05-04  Tom Tromey  <tom@tromey.com>

* printcmd.c: Don't include tui.h.
(decode_format): Use skip_spaces.

6 years agoUse previous count when 'x' command is repeated
Tom Tromey [Fri, 27 Apr 2018 21:52:44 +0000 (15:52 -0600)]
Use previous count when 'x' command is repeated

About the 'x' command, the manual says:

    If you use <RET> to repeat the 'x' command, the repeat count N is
    used again; the other arguments default as for successive uses of
    'x'.

However, PR gdb/22619 points out that this does not work.

This patch fixes the problem.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

PR gdb/22619:
* printcmd.c (last_count): New global.
(x_command): Use saved count when repeating.

testsuite/ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

PR gdb/22619:
* gdb.base/long_long.exp (gdb_test_long_long): Add test for repeat
behavior.

6 years agoRemove do_closedir_cleanup
Tom Tromey [Thu, 3 May 2018 06:18:20 +0000 (00:18 -0600)]
Remove do_closedir_cleanup

This removes both copies of do_closedir_cleanup in favor of a new
unique_ptr specialization.

Tested by the buildbot, though I'm not sure that these code paths are
exercised there.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* nto-procfs.c (do_closedir_cleanup): Remove.
(procfs_pidlist): Use gdb_dir_up.
* procfs.c (do_closedir_cleanup): Remove.
(proc_update_threads): Use gdb_dir_up.
* common/filestuff.h (struct gdb_dir_deleter): New.
(gdb_dir_up): New typedef.

6 years agoRemove cleanup from print_mention_exception
Tom Tromey [Fri, 10 Nov 2017 21:42:35 +0000 (14:42 -0700)]
Remove cleanup from print_mention_exception

This removes a cleanup from print_mention_exception by using
string_printf.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* ada-lang.c (print_mention_exception): Use std::string.

6 years agoReturn std::string from ada_exception_catchpoint_cond_string
Tom Tromey [Thu, 3 May 2018 16:23:55 +0000 (10:23 -0600)]
Return std::string from ada_exception_catchpoint_cond_string

This changes ada_exception_catchpoint_cond_string to return a
std::string, allowing for the removal of a cleanup in
create_excep_cond_exprs.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* ada-lang.c (create_excep_cond_exprs): Update.
(ada_exception_catchpoint_cond_string): Use std::string.

6 years agoRemove cleanup from old_renaming_is_invisible
Tom Tromey [Fri, 10 Nov 2017 21:39:31 +0000 (14:39 -0700)]
Remove cleanup from old_renaming_is_invisible

This removes a cleanup from ada-lang.c by changing xget_renaming_scope
to return a std::string.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* ada-lang.c (xget_renaming_scope): Return std::string.
(old_renaming_is_invisible): Update.

6 years agoUse gdb_bfd_ref_ptr in target_bfd
Tom Tromey [Fri, 4 May 2018 16:17:52 +0000 (10:17 -0600)]
Use gdb_bfd_ref_ptr in target_bfd

I noticed that target_bfd was using manual reference counting for the
BFD it held.  This patch changes it to use gdb_bfd_ref_ptr instead.

Tested by the buildbot.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* bfd-target.c (target_bfd::m_bfd): Now a gdb_bfd_ref_ptr.
(target_bfd::target_bfd, target_bfd::~target_bfd): Update.

6 years ago[spu] Fix build break
Ulrich Weigand [Fri, 4 May 2018 17:20:18 +0000 (19:20 +0200)]
[spu] Fix build break

gdb/ChangeLog:
2018-05-04  Ulrich Weigand  <uweigand@de.ibm.com>

* spu-linux-nat.c (spu_linux_nat_target::wait): Fix syntax error.

6 years agoRemove a cleanup from remote.c
Tom Tromey [Thu, 3 May 2018 06:04:11 +0000 (00:04 -0600)]
Remove a cleanup from remote.c

This removes a cleanup from remote.c by using std::string to construct
the qSupported packet.

Tested by the buildbot.

ChangeLog
2018-05-04  Tom Tromey  <tom@tromey.com>

* remote.c (remote_query_supported_append): Change type.
(remote_check_symbols): Update.

6 years agogdb/testsuite: Handle targets with lots of registers
Andrew Burgess [Sun, 8 Apr 2018 23:18:34 +0000 (00:18 +0100)]
gdb/testsuite: Handle targets with lots of registers

In gdb.base/maint.exp a test calls 'maint print registers'.  If the
target has lots of registers this may overflow expect's buffers,
causing the test to fail.

After this commit we process the output line at a time until we get back
to the GDB prompt, this should prevent buffer overrun while still
testing that the command works as required.

gdb/testsuite/ChangeLog:

* gdb.base/maint.exp: Process output from 'maint print registers'
line at a time.

6 years agoconfigure uses incorrect link order when testing libpython
Paul Pluzhnikov [Fri, 4 May 2018 14:07:58 +0000 (10:07 -0400)]
configure uses incorrect link order when testing libpython

References:

https://stackoverflow.com/a/49868387
https://sourceware.org/bugzilla/show_bug.cgi?id=11420

Configure uses "gcc -o conftest -g ... conftest.c -ldl -lncurses -lm -ldl
... -lpthread ... -lpython2.7" when deciding whether give libpython is
usable.

That of course is the wrong link order, and only works for shared libraries
(mostly by accident), and only on some systems.

gdb/ChangeLog:

PR gdb/11420
* configure.ac: Prepend libpython.
* python/python-config.py: Likewise.
* configure: Regenerate.

6 years agogdb/testsuite: Fix broken regexp in gdbstub case
Andrew Burgess [Thu, 15 Mar 2018 19:41:50 +0000 (19:41 +0000)]
gdb/testsuite: Fix broken regexp in gdbstub case

When $use_gdb_stub is true then, when we start an MI target there's a
regexp to match GDB's startup pattern.  Unfortunately the pattern is
broken, and we're also missing a timeout case in the match list (which
would have helped point out that the regexp was broken).

The changes to the regexp are:

  1. Remove '${run_match}' prefix, the issued command doesn't include
  '${run_prefix}' so expecting '${run_match}' is wrong.

  2. Replaced '\\n' with '\\\\n' in order to match literal '\n' in
  GDBs output (that is, match a backslash followed by 'n', not a
  newline character).

  3. Replaced a '.' (matching any character) with '\.' to match a '.'
  and moved the '\.' into the correct place in the regexp.

  4. Replaced '\r\n' with '[\r\n]+' to match the end of a line.  This
  change isn't esential, but matches the other end of line patterns
  within this regexp.

Here's an example of the output that the regexp should match taken
from a testfile log, the first line is the command sent to GDB, and
the remaining lines are the response from GDB:

  jump *_start
  &"jump *_start\n"
  ~"Continuing at 0x10074.\n"
  ^running
  *running,thread-id="all"
  (gdb)

gdb/testsuite/ChangeLog:

* lib/mi-support.exp (mi_run_cmd_full): Fix regexp and add a
timeout.

6 years ago-Wstringop-truncation warnings
Alan Modra [Fri, 27 Apr 2018 00:12:11 +0000 (09:42 +0930)]
-Wstringop-truncation warnings

This patch is aimed at silencing gcc8 -Wstringop-truncation warnings.

Unfortunately adding  __attribute__ ((__nonstring)) doesn't work in a
number of the places patched here, (see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85643) so if you have
recent glibc headers installed you'll need to configure binutils with
--disable-werror to compile using gcc8 or gcc9.

include/
* ansidecl.h: Import from gcc.
* coff/internal.h (struct internal_scnhdr): Add ATTRIBUTE_NONSTRING
to s_name.
(struct internal_syment): Add ATTRIBUTE_NONSTRING to _n_name.
bfd/
* elf-linux-core.h (struct elf_external_linux_prpsinfo32_ugid32),
(struct elf_external_linux_prpsinfo32_ugid16),
(struct elf_external_linux_prpsinfo64_ugid32),
(struct elf_external_linux_prpsinfo64_ugid16): Add ATTRIBUTE_NONSTRING
to pr_fname and pr_psargs fields.  Remove GCC diagnostic pragmas.
Move comment to..
* elf.c (elfcore_write_prpsinfo): ..here.  Indent nested preprocessor
directives.
* elf32-arm.c (elf32_arm_nabi_write_core_note): Add ATTRIBUTE_NONSTRING
to data.
* elf32-ppc.c (ppc_elf_write_core_note): Likewise.
* elf32-s390.c (elf_s390_write_core_note): Likewise.
* elf64-s390.c (elf_s390_write_core_note): Likewise.
* elfxx-aarch64.c (_bfd_aarch64_elf_write_core_note): Likewise.
* elf64-x86-64.c (elf_x86_64_write_core_note): Add GCC diagnostic
pragmas.
* peXXigen.c (_bfd_XXi_swap_scnhdr_out): Use strnlen to avoid
false positive gcc-8 warning.
gas/
* config/obj-evax.c (shorten_identifier): Use memcpy in place
of strncpy.
* config/obj-macho.c (obj_mach_o_make_or_get_sect): Ensure
segname and sectname fields are NUL terminated.

6 years agoppc: Fix warning messages when IBM and IEEE long double are mixed
Tulio Magno Quites Machado Filho [Wed, 2 May 2018 22:18:44 +0000 (19:18 -0300)]
ppc: Fix warning messages when IBM and IEEE long double are mixed

When IBM long double is used, the .gnu_attribute 4 is set to 1 | (1 *
4).  IEEE long double sets the same .gnu_attribute to 1 | (3 * 4).

* elf32-ppc.c (_bfd_elf_ppc_merge_fp_attributes): Fix the order
of arguments when warning about different long double types.

6 years agoFix typo in Makefile.am to make it agree with Makefile.in.
Jim Wilson [Fri, 4 May 2018 00:25:31 +0000 (17:25 -0700)]
Fix typo in Makefile.am to make it agree with Makefile.in.

ld/
* Makefile.am (earmelfb_linux_fdpiceabi.c): Fix typo in dependencies.

6 years agoAutomatic date update in version.in
GDB Administrator [Fri, 4 May 2018 00:00:30 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoUse flex's -t option instead of --stdout
Simon Marchi [Thu, 3 May 2018 21:29:14 +0000 (17:29 -0400)]
Use flex's -t option instead of --stdout

As reported in

  https://sourceware.org/ml/gdb-patches/2018-05/msg00042.html

some old versions of flex (2.5.4) don't support the --stdout switch.
Use -t, which is an alias.

gdb/ChangeLog:

* Makefile.in (%.c: %.l): Use -t instead of --stdout.

6 years agogdb/testsuite: Filter out some registers for riscv
Andrew Burgess [Sun, 1 Apr 2018 21:02:33 +0000 (22:02 +0100)]
gdb/testsuite: Filter out some registers for riscv

On riscv the cycle counter, and instructions retired counter CSRs are
read only, this causes problems in the gdb.base/callfuncs.exp test, as
the values in these CSRs change after an inferior call, the check that
no target registers have been modified then fails.

Luckily the test already has a mechanism in place for filtering out
registers that are modified (and can't be restored) by an inferior call,
so this commit adds the problem registers into this list for riscv.

In the future we may end up needing to filter out more CSRs, but right
now, for the targets I have access too, these are the only ones causing
problems.

gdb/testsuite/ChangeLog:

* gdb.base/callfuncs.exp (fetch_all_registers): Add riscv register
filter pattern.

6 years agoBFD: Prevent writing the MIPS _gp_disp symbol into symbol tables
Simon Atanasyan [Thu, 3 May 2018 16:17:46 +0000 (17:17 +0100)]
BFD: Prevent writing the MIPS _gp_disp symbol into symbol tables

The _gp_disp is a magic symbol, always implicitly defined by the linker.
It does not make a sense to write it into symbol tables for output files.
Moreover, now if the linker gets a version script, the _gp_disp symbol
gets zero version definition index.  The zero index means[1]:

"The symbol is local, not available outside the object."

But the _gp_disp symbol has GLOBAL binding.  That confuses some tools
like for example the LLD linker when they get such files as inputs.

This patch fixes the problem - it prevents writing the _gp_disp symbol
in regular and dynamic symbol tables.

This was tested by running LD test suite on a mipsel-linux board.

References:

[1] "Linux Standard Base Specification", Section "10.7.2 Symbol
    Version Table", p. 32

2018-05-03  Simon Atanasyan  <simon@atanasyan.com>

bfd/

* elf32-mips.c: (elf32_mips_fixup_symbol): New function.
(elf_backend_fixup_symbol): New macro.
* elfxx-mips.c: (mips_elf_output_extsym): Discard _gp_disp
handling.
(_bfd_mips_elf_finish_dynamic_symbol): Likewise.

ld/

* testsuite/ld-mips-elf/gp-disp-sym.d: New test.
* testsuite/ld-mips-elf/gp-disp-sym.s: New test source.
* testsuite/ld-mips-elf/mips-elf.exp: Run the new test.
* testsuite/ld-mips-elf/mips16-pic-2.ad: Update for _gp_disp
symbol removal.
* testsuite/ld-mips-elf/mips16-pic-2.nd: Likewise.
* testsuite/ld-mips-elf/pic-and-nonpic-3a.dd: Likewise.
* testsuite/ld-mips-elf/tlslib-o32-hidden.got: Likewise.
* testsuite/ld-mips-elf/tlslib-o32-ver.got: Likewise.
* testsuite/ld-mips-elf/tlslib-o32.got: Likewise.

6 years agoFix s390 GNU/Linux build
Pedro Alves [Thu, 3 May 2018 10:31:38 +0000 (11:31 +0100)]
Fix s390 GNU/Linux build

- Fixes this compile error:

  ../../src/gdb/s390-linux-nat.c:125:8: error: ‘virtual bool s390_linux_nat_target::have_continuable_watchpoint()’ can be marked override [-Werror=suggest-override]
     bool have_continuable_watchpoint () { return 1; }
  ^~~~~~~~~~~~~~~~~~~~~~~~~~~

  s390 never implemented that hook.  The declaration of the method got
  there simply via copy/paste from some other target.

  Return 'true' instead of '1' while at it.

- Fixes this link error:

  s390-linux-nat.o:(.rodata._ZTV21s390_linux_nat_target[_ZTV21s390_linux_nat_target]+0x120): undefined reference to `s390_linux_nat_target::watchpoint_addr_within_range(unsigned long, unsigned long, int)'

gdb/ChangeLog:
2018-05-03  Pedro Alves  <palves@redhat.com>

* s390-linux-nat.c
(s390_linux_nat_target::have_continuable_watchpoint): Mark with
override.  Write 'true' instead of '1'.
(s390_linux_nat_target::watchpoint_addr_within_range): Remove
declaration.

6 years agoAutomatic date update in version.in
GDB Administrator [Thu, 3 May 2018 00:00:13 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agotarget factories, target open and multiple instances of targets
Pedro Alves [Wed, 2 May 2018 23:37:32 +0000 (00:37 +0100)]
target factories, target open and multiple instances of targets

Currently, to open a target, with "target TARGET_NAME", GDB finds the
target_ops instance with "TARGET_NAME" as short name, and then calls
its target_ops::open virtual method.  In reality, there's no actual
target/name lookup, a pointer to the target_ops object was associated
with the "target TARGET_NAME" command at add_target time (when GDB is
initialized), as the command's context.

This creates a chicken and egg situation.  Consider the case of
wanting to open multiple remote connections.  We want to be able to
have one remote target_ops instance per connection, but, if we're not
connected yet, so we don't yet have an instance to call target->open()
on...

This patch fixes this by separating out common info about a target_ops
to a separate structure (shortname, longname, doc), and changing the
add_target routine to take a reference to such an object instead of a
pointer to a target_ops, and a pointer to a factory function that is
responsible to open an instance of the corresponding target when the
user types "target TARGET_NAME".

 -extern void add_target (struct target_ops *);
 +extern void add_target (const target_info &info, target_open_ftype *func);

I.e. this factory function replaces the target_ops::open virtual
method.

For static/singleton targets, nothing changes, the target_open_ftype
function pushes the global target_ops instance on the target stack.
At target_close time, the connection is tor down, but the global
target_ops object remains live.

However, targets that support being open multiple times will make
their target_open_ftype routine allocate a new target_ops instance on
the heap [e.g., new remote_target()], and push that on the stack.  At
target_close time, the new object is destroyed (by the
target_ops::close virtual method).

Both the core target and the remote targets will support being open
multiple times (others could/should too, but those were my stopping
point), but not in this patch yet.  We need to get rid of more globals
first before that'd be useful.

Native targets are somewhat special, given find_default_run_target &
friends.  Those routines also expect to return a target_ops pointer,
even before we've open the target.  However, we'll never need more
than one instance of the native target, so we can assume/require that
native targets are global/simpletons, and have the backends register a
pointer to the native target_ops.  Since all native targets inherit
inf_child_target, we can centralize that registration.  See
add_inf_child_target, get_native_target/set_native_target and
find_default_run_target.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* aarch64-fbsd-nat.c (_initialize_aarch64_fbsd_nat): Use
add_inf_child_target.
* aarch64-linux-nat.c (_initialize_aarch64_linux_nat): Use
add_inf_child_target.
* aix-thread.c (aix_thread_target_info): New.
(aix_thread_target) <shortname, longname, doc>: Delete.
<info>: New.
* alpha-bsd-nat.c (_initialize_alphabsd_nat): Use
add_inf_child_target.
* alpha-linux-nat.c (_initialize_alpha_linux_nat): Use
add_inf_child_target.
* amd64-fbsd-nat.c (_initialize_amd64fbsd_nat): Use
add_inf_child_target.
* amd64-linux-nat.c (_initialize_amd64_linux_nat): Use
add_inf_child_target.
* amd64-nbsd-nat.c (_initialize_amd64nbsd_nat): Use
add_inf_child_target.
* amd64-obsd-nat.c (_initialize_amd64obsd_nat): Use
add_inf_child_target.
* arm-fbsd-nat.c (_initialize_arm_fbsd_nat): Use
add_inf_child_target.
* arm-linux-nat.c (_initialize_arm_linux_nat): Use
add_inf_child_target.
* arm-nbsd-nat.c (_initialize_arm_netbsd_nat): Use
add_inf_child_target.
* bfd-target.c (target_bfd_target_info): New.
(target_bfd) <shortname, longname, doc>: Delete.
<info>: New.
* bsd-kvm.c (bsd_kvm_target_info): New.
(bsd_kvm_target) <shortname, longname, doc>: Delete.
<info>: New.
(bsd_kvm_target::open): Rename to ...
(bsd_kvm_target_open): ... this.  Adjust.
* bsd-uthread.c (bsd_uthread_target_info): New.
(bsd_uthread_target) <shortname, longname, doc>: Delete.
<info>: New.
* corefile.c (core_file_command): Adjust.
* corelow.c (core_target_info): New.
(core_target) <shortname, longname, doc>: Delete.
<info>: New.
(core_target::open): Rename to ...
(core_target_open): ... this.  Adjust.
* ctf.c (ctf_target_info): New.
(ctf_target) <shortname, longname, doc>: Delete.
<info>: New.
(ctf_target::open): Rename to ...
(ctf_target_open): ... this.
(_initialize_ctf): Adjust.
* exec.c (exec_target_info): New.
(exec_target) <shortname, longname, doc>: Delete.
<info>: New.
(exec_target::open): Rename to ...
(exec_target_open): ... this.
* gdbcore.h (core_target_open): Declare.
* go32-nat.c (_initialize_go32_nat): Use add_inf_child_target.
* hppa-linux-nat.c (_initialize_hppa_linux_nat): Use
add_inf_child_target.
* hppa-nbsd-nat.c (_initialize_hppanbsd_nat): Use
add_inf_child_target.
* hppa-obsd-nat.c (_initialize_hppaobsd_nat): Use
add_inf_child_target.
* i386-darwin-nat.c (_initialize_i386_darwin_nat): Use
add_inf_child_target.
* i386-fbsd-nat.c (_initialize_i386fbsd_nat): Use
add_inf_child_target.
* i386-gnu-nat.c (_initialize_i386gnu_nat): Use
add_inf_child_target.
* i386-linux-nat.c (_initialize_i386_linux_nat): Use
add_inf_child_target.
* i386-nbsd-nat.c (_initialize_i386nbsd_nat): Use
add_inf_child_target.
* i386-obsd-nat.c (_initialize_i386obsd_nat): Use
add_inf_child_target.
* ia64-linux-nat.c (_initialize_ia64_linux_nat): Use
add_inf_child_target.
* inf-child.c (inf_child_target_info): New.
(inf_child_target::info): New.
(inf_child_open_target): Remove 'target' parameter.  Use
get_native_target instead.
(inf_child_target::open): Delete.
(add_inf_child_target): New.
* inf-child.h (inf_child_target) <shortname, longname, doc, open>:
Delete.
<info>: New.
(add_inf_child_target): Declare.
(inf_child_open_target): Declare.
* linux-thread-db.c (thread_db_target_info): New.
(thread_db_target) <shortname, longname, doc>: Delete.
<info>: New.
* m32r-linux-nat.c (_initialize_m32r_linux_nat): Use
add_inf_child_target.
* m68k-bsd-nat.c (_initialize_m68kbsd_nat): Use
add_inf_child_target.
* m68k-linux-nat.c (_initialize_m68k_linux_nat): Use
add_inf_child_target.
* m88k-bsd-nat.c (_initialize_m88kbsd_nat): Use
add_inf_child_target.
* make-target-delegates (print_class): Adjust.
* mips-fbsd-nat.c (_initialize_mips_fbsd_nat): Use
add_inf_child_target.
* mips-linux-nat.c (_initialize_mips_linux_nat): Use
add_inf_child_target.
* mips-nbsd-nat.c (_initialize_mipsnbsd_nat): Use
add_inf_child_target.
* mips64-obsd-nat.c (_initialize_mips64obsd_nat): Use
add_inf_child_target.
* nto-procfs.c (nto_native_target_info): New.
(nto_procfs_target_native) <shortname, longname, doc>:
Delete.
<info>: New.
(nto_procfs_target_info): New.
(nto_procfs_target_procfs) <shortname, longname, doc>:
Delete.
<info>: New.
(init_procfs_targets): Adjust.
* ppc-fbsd-nat.c (_initialize_ppcfbsd_nat): Use
add_inf_child_target.
* ppc-linux-nat.c (_initialize_ppc_linux_nat): Use
add_inf_child_target.
* ppc-nbsd-nat.c (_initialize_ppcnbsd_nat): Use
add_inf_child_target.
* ppc-obsd-nat.c (_initialize_ppcobsd_nat): Use
add_inf_child_target.
* ravenscar-thread.c (ravenscar_target_info): New.
(ravenscar_thread_target) <shortname, longname, doc>:
Delete.
<info>: New.
* record-btrace.c (record_btrace_target_info):
(record_btrace_target) <shortname, longname, doc>: Delete.
<info>: New.
(record_btrace_target::open): Rename to ...
(record_btrace_target_open): ... this.  Adjust.
* record-full.c (record_longname, record_doc): New.
(record_full_base_target) <shortname, longname, doc>: Delete.
<info>: New.
(record_full_target_info): New.
(record_full_target): <shortname>: Delete.
<info>: New.
(record_full_core_open_1, record_full_open_1): Update comments.
(record_full_base_target::open): Rename to ...
(record_full_open): ... this.
(cmd_record_full_restore): Update.
(_initialize_record_full): Update.
* remote-sim.c (remote_sim_target_info): New.
(gdbsim_target) <shortname, longname, doc>: Delete.
<info>: New.
(gdbsim_target::open): Rename to ...
(gdbsim_target_open): ... this.
(_initialize_remote_sim): Adjust.
* remote.c (remote_doc): New.
(remote_target_info): New.
(remote_target) <shortname, longname, doc>: Delete.
<info>: New.
(extended_remote_target_info): New.
(extended_remote_target) <shortname, longname, doc>: Delete.
<info>: New.
(remote_target::open_1): Make static.  Adjust.
* rs6000-nat.c (_initialize_rs6000_nat): Use add_inf_child_target.
* s390-linux-nat.c (_initialize_s390_nat): Use
add_inf_child_target.
* sh-nbsd-nat.c (_initialize_shnbsd_nat): Use
add_inf_child_target.
* sol-thread.c (thread_db_target_info): New.
(sol_thread_target) <shortname, longname, doc>: Delete.
<info>: New.
* sparc-linux-nat.c (_initialize_sparc_linux_nat): Use
add_inf_child_target.
* sparc-nbsd-nat.c (_initialize_sparcnbsd_nat): Use
add_inf_child_target.
* sparc64-fbsd-nat.c (_initialize_sparc64fbsd_nat): Use
add_inf_child_target.
* sparc64-linux-nat.c (_initialize_sparc64_linux_nat): Use
add_inf_child_target.
* sparc64-nbsd-nat.c (_initialize_sparc64nbsd_nat): Use
add_inf_child_target.
* sparc64-obsd-nat.c (_initialize_sparc64obsd_nat): Use
add_inf_child_target.
* spu-linux-nat.c (_initialize_spu_nat): Use
add_inf_child_target.
* spu-multiarch.c (spu_multiarch_target_info): New.
(spu_multiarch_target) <shortname, longname, doc>: Delete.
<info>: New.
* target-delegates.c: Regenerate.
* target.c: Include <unordered_map>.
(target_ops_p): Delete.
(DEF_VEC_P(target_ops_p)): Delete.
(target_factories): New.
(test_target_info): New.
(test_target_ops::info): New.
(open_target): Adjust to use target_factories.
(add_target_with_completer): Rename to ...
(add_target): ... this.  Change prototype.  Register target_info
and open callback in target_factories.  Register target_info in
command context instead of target_ops.
(add_target): Delete old implementation.
(add_deprecated_target_alias): Change prototype.  Adjust.
(the_native_target): New.
(set_native_target, get_native_target): New.
(find_default_run_target): Use the_native_target.
(find_attach_target, find_run_target): Simplify.
(target_ops::open): Delete.
(dummy_target_info): New.
(dummy_target::shortname, dummy_target::longname)
(dummy_target::doc): Delete.
(dummy_target::info): New.
(debug_target::shortname, debug_target::longname)
(debug_target::doc): Delete.
(debug_target::info): New.
* target.h (struct target_info): New.
(target_ops::~target_ops): Add comment.
(target_ops::info): New.
(target_ops::shortname, target_ops::longname, target_ops::doc): No
longer virtual.  Implement in terms of target_info.
(set_native_target, get_native_target): Declare.
(target_open_ftype): New.
(add_target, add_target_with_completer)
(add_deprecated_target_alias): Change prototype.
(test_target) <shortname, longname, doc>: Delete.
<info>: New.
* tilegx-linux-nat.c (_initialize_tile_linux_nat): Use
add_inf_child_target.
* tracefile-tfile.c (tfile_target_info): New.
(tfile_target) <shortname, longname, doc>: Delete.
<info>: New.
(tfile_target::open): Rename to ...
(tfile_target_open): ... this.
(_initialize_tracefile_tfile): Adjust.
* vax-bsd-nat.c (_initialize_vaxbsd_nat): Use
add_inf_child_target.
* windows-nat.c (_initialize_windows_nat): Use
add_inf_child_target.
* xtensa-linux-nat.c (_initialize_xtensa_linux_nat): Use
add_inf_child_target.

6 years agolinux_nat_target: More low methods
Pedro Alves [Wed, 2 May 2018 23:37:27 +0000 (00:37 +0100)]
linux_nat_target: More low methods

This converts the remaining linux-nat.c hooks low_ methods like had
been started in a previous patch.  The linux_nat_set_foo routines are
all gone with this.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* linux-nat.h (linux_nat_target) <low_new_thread,
low_delete_thread, low_new_fork, low_forget_process,
low_prepare_to_resume, low_siginfo_fixup, low_status_is_event>:
New virtual methods.
(linux_nat_set_new_thread, linux_nat_set_delete_thread)
(linux_nat_new_fork_ftype, linux_nat_set_new_fork)
(linux_nat_forget_process_ftype, linux_nat_set_forget_process)
(linux_nat_forget_process, linux_nat_set_siginfo_fixup)
(linux_nat_set_prepare_to_resume, linux_nat_set_status_is_event):
Delete.
* linux-fork.c (delete_fork): Adjust to call low method.
* linux-nat.c (linux_nat_new_thread, linux_nat_delete_thread)
(linux_nat_new_fork, linux_nat_forget_process_hook)
(linux_nat_prepare_to_resume, linux_nat_siginfo_fixup)
(linux_nat_status_is_event):
(linux_nat_target::follow_fork, lwp_free, add_lwp, detach_one_lwp)
(linux_resume_one_lwp_throw, linux_handle_extended_wait): Adjust
to call low method.
(sigtrap_is_event): Rename to ...
(linux_nat_target::low_status_is_event): ... this.
(linux_nat_set_status_is_event): Delete.
(save_stop_reason, linux_nat_wait_1)
(linux_nat_target::mourn_inferior, siginfo_fixup): Adjust to call
low methods.
(linux_nat_set_new_thread, linux_nat_set_delete_thread)
(linux_nat_set_new_fork, linux_nat_set_forget_process)
(linux_nat_forget_process, linux_nat_set_siginfo_fixup)
(linux_nat_set_prepare_to_resume): Delete.
* aarch64-linux-nat.c: All linux_nat_set_* callbacks converted to
low virtual methods.
* amd64-linux-nat.c: Likewise.
* arm-linux-nat.c: Likewise.
* i386-linux-nat.c: Likewise.
* ia64-linux-nat.c: Likewise.
* mips-linux-nat.c: Likewise.
* ppc-linux-nat.c: Likewise.
* s390-linux-nat.c: Likewise.
* sparc64-linux-nat.c: Likewise.
* x86-linux-nat.c: Likewise.
* x86-linux-nat.h: Include "nat/x86-linux.h".
(x86_linux_nat_target) <low_new_fork, low_forget_process,
low_prepare_to_resume, low_new_thread, low_delete_thread>:
Override methods.

6 years agotarget_ops: Use bool throughout
Pedro Alves [Wed, 2 May 2018 23:37:26 +0000 (00:37 +0100)]
target_ops: Use bool throughout

After the previous target_ops/C++ patches are all squashed and merged,
this one can go in separately.

This patch adjusts all the target methods to return bool instead of int
when they're returning a boolean.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* target.h (target_ops)
<stopped_by_sw_breakpoint, supports_stopped_by_sw_breakpoint,
stopped_by_hw_breakpoint, supports_stopped_by_hw_breakpoint,
stopped_by_watchpoint, have_continuable_watchpoint,
stopped_data_address, watchpoint_addr_within_range,
can_accel_watchpoint_condition, can_run, thread_alive,
has_all_memory, has_memory, has_stack, has_registers,
has_execution, can_async_p, is_async_p, supports_non_stop,
always_non_stop_p, can_execute_reverse, supports_multi_process,
supports_enable_disable_tracepoint,
supports_disable_randomization, supports_string_tracing,
supports_evaluation_of_breakpoint_conditions,
can_run_breakpoint_commands, filesystem_is_local,
can_download_tracepoint, get_trace_state_variable_value,
set_trace_notes, get_tib_address, use_agent, can_use_agent,
record_is_replaying, record_will_replay,
augmented_libraries_svr4_read>: Adjust to return bool.
* aarch64-linux-nat.c: All implementations adjusted.
* aix-thread.c: All implementations adjusted.
* arm-linux-nat.c: All implementations adjusted.
* breakpoint.c: All implementations adjusted.
* bsd-kvm.c: All implementations adjusted.
* bsd-uthread.c: All implementations adjusted.
* corelow.c: All implementations adjusted.
* ctf.c: All implementations adjusted.
* darwin-nat.c: All implementations adjusted.
* darwin-nat.h: All implementations adjusted.
* exec.c: All implementations adjusted.
* fbsd-nat.c: All implementations adjusted.
* fbsd-nat.h: All implementations adjusted.
* gnu-nat.c: All implementations adjusted.
* gnu-nat.h: All implementations adjusted.
* go32-nat.c: All implementations adjusted.
* ia64-linux-nat.c: All implementations adjusted.
* inf-child.c: All implementations adjusted.
* inf-child.h: All implementations adjusted.
* inf-ptrace.c: All implementations adjusted.
* inf-ptrace.h: All implementations adjusted.
* linux-nat.c: All implementations adjusted.
* linux-nat.h: All implementations adjusted.
* mips-linux-nat.c: All implementations adjusted.
* nto-procfs.c: All implementations adjusted.
* ppc-linux-nat.c: All implementations adjusted.
* procfs.c: All implementations adjusted.
* ravenscar-thread.c: All implementations adjusted.
* record-btrace.c: All implementations adjusted.
* record-full.c: All implementations adjusted.
* remote-sim.c: All implementations adjusted.
* remote.c: All implementations adjusted.
* s390-linux-nat.c: All implementations adjusted.
* sol-thread.c: All implementations adjusted.
* spu-multiarch.c: All implementations adjusted.
* target-delegates.c: All implementations adjusted.
* target.c: All implementations adjusted.
* target.h: All implementations adjusted.
* tracefile-tfile.c: All implementations adjusted.
* tracefile.c: All implementations adjusted.
* tracefile.h: All implementations adjusted.
* windows-nat.c: All implementations adjusted.
* x86-linux-nat.h: All implementations adjusted.
* x86-nat.h: All implementations adjusted.

6 years agomake-target-delegates: line break between return type and function name
Pedro Alves [Wed, 2 May 2018 23:37:23 +0000 (00:37 +0100)]
make-target-delegates: line break between return type and function name

Before the target_ops C++ification, this wasn't necessary simply
because the methods were wrapped in ()'s, like
  '(*to_my_long_method_name) (target_ops *)',
so
  std::vector<long_type_name>(*to_my_long_method_name) ()TARGET_DEFAULT_IGNORE ()

still parsed correctly.  With the (*) gone, we need this.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* make-target-delegates (scan_target_h): Don't trim lines here.
Replace sequences of tabs and/or whitespace with a single
whitespace.
(top level, parsing methods): Trim each line before processing it
here.

6 years agoConvert struct target_ops to C++
Pedro Alves [Wed, 2 May 2018 23:37:22 +0000 (00:37 +0100)]
Convert struct target_ops to C++

I.e., use C++ virtual methods and inheritance instead of tables of
function pointers.

Unfortunately, there's no way to do a smooth transition.  ALL native
targets in the tree must be converted at the same time.  I've tested
all I could with cross compilers and with help from GCC compile farm,
but naturally I haven't been able to test many of the ports.  Still, I
made a best effort to port everything over, and while I expect some
build problems due to typos and such, which should be trivial to fix,
I don't expect any design problems.

* Implementation notes:

- The flattened current_target is gone.  References to current_target
  or current_target.beneath are replaced with references to
  target_stack (the top of the stack) directly.

- To keep "set debug target" working, this adds a new debug_stratum
  layer that sits on top of the stack, prints the debug, and delegates
  to the target beneath.

  In addition, this makes the shortname and longname properties of
  target_ops be virtual methods instead of data fields, and makes the
  debug target defer those to the target beneath.  This is so that
  debug code sprinkled around that does "if (debugtarget) ..."  can
  transparently print the name of the target beneath.

  A patch later in the series actually splits out the
  shortname/longname methods to a separate structure, but I preferred
  to keep that chance separate as it is associated with changing a bit
  the design of how targets are registered and open.

- Since you can't check whether a C++ virtual method is overridden,
  the old method of checking whether a target_ops implements a method
  by comparing the function pointer must be replaced with something
  else.

  Some cases are fixed by adding a parallel "can_do_foo" target_ops
  methods.  E.g.,:

    +  for (t = target_stack; t != NULL; t = t->beneath)
 {
    -      if (t->to_create_inferior != NULL)
    +      if (t->can_create_inferior ())
    break;
 }

  Others are fixed by changing void return type to bool or int return
  type, and have the default implementation return false or -1, to
  indicate lack of support.

- make-target-delegates was adjusted to generate C++ classes and
  methods.

  It needed tweaks to grok "virtual" in front of the target method
  name, and for the fact that methods are no longer function pointers.
  (In particular, the current code parsing the return type was simple
  because it could simply parse up until the '(' in '(*to_foo)'.

  It now generates a couple C++ classes that inherit target_ops:
  dummy_target and debug_target.

  Since we need to generate the class declarations as well, i.e., we
  need to emit methods twice, we now generate the code in two passes.

- The core_target global is renamed to avoid conflict with the
  "core_target" class.

- ctf/tfile targets

  init_tracefile_ops is replaced by a base class that is inherited by
  both ctf and tfile.

- bsd-uthread

  The bsd_uthread_ops_hack hack is gone.  It's not needed because
  nothing was extending a target created by bsd_uthread_target.

- remote/extended-remote targets

  This is a first pass, just enough to C++ify target_ops.

  A later pass will convert more free functions to methods, and make
  remote_state be truly per remote instance, allowing multiple
  simultaneous instances of remote targets.

- inf-child/"native" is converted to an actual base class
  (inf_child_target), that is inherited by all native targets.

- GNU/Linux

  The old weird double-target linux_ops mechanism in linux-nat.c, is
  gone, replaced by adding a few virtual methods to linux-nat.h's
  target_ops, called low_XXX, that the concrete linux-nat
  implementations override.  Sort of like gdbserver's
  linux_target_ops, but simpler, for requiring only one
  target_ops-like hierarchy, which spares implementing the same method
  twice when we need to forward the method to a low implementation.
  The low target simply reimplements the target_ops method directly in
  that case.

  There are a few remaining linux-nat.c hooks that would be better
  converted to low_ methods like above too.  E.g.:

   linux_nat_set_new_thread (t, x86_linux_new_thread);
   linux_nat_set_new_fork (t, x86_linux_new_fork);
   linux_nat_set_forget_process

  That'll be done in a follow up patch.

- We can no longer use functions like x86_use_watchpoints to install
  custom methods on an arbitrary base target.

  The patch replaces instances of such a pattern with template mixins.
  For example memory_breakpoint_target defined in target.h, or
  x86_nat_target in x86-nat.h.

- linux_trad_target, MIPS and Alpha GNU/Linux

  The code in the new linux-nat-trad.h/c files which was split off of
  inf-ptrace.h/c recently, is converted to a C++ base class, and used
  by the MIPS and Alpha GNU/Linux ports.

- BSD targets

  The

    $architecture x NetBSD/OpenBSD/FreeBSD

  support matrix complicates things a bit.  There's common BSD target
  code, and there's common architecture-specific code shared between
  the different BSDs.  Currently, all that is stiched together to form
  a final target, via the i386bsd_target, x86bsd_target,
  fbsd_nat_add_target functions etc.

  This introduces new fbsd_nat_target, obsd_nat_target and
  nbsd_nat_target classes that serve as base/prototype target for the
  corresponding BSD variant.

  And introduces generic i386/AMD64 BSD targets, to be used as
  template mixin to build a final target.  Similarly, a generic SPARC
  target is added, used by both BSD and Linux ports.

- bsd_kvm_add_target, BSD libkvm target

  I considered making bsd_kvm_supply_pcb a virtual method, and then
  have each port inherit bsd_kvm_target and override that method, but
  that was resulting in lots of unjustified churn, so I left the
  function pointer mechanism alone.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>
    John Baldwin  <jhb@freebsd.org>

* target.h (enum strata) <debug_stratum>: New.
(struct target_ops) <all delegation methods>: Replace by C++
virtual methods, and drop "to_" prefix.  All references updated
throughout.
<to_shortname, to_longname, to_doc, to_data,
to_have_steppable_watchpoint, to_have_continuable_watchpoint,
to_has_thread_control, to_attach_no_wait>: Delete, replaced by
virtual methods.  All references updated throughout.
<can_attach, supports_terminal_ours, can_create_inferior,
get_thread_control_capabilities, attach_no_wait>: New
virtual methods.
<insert_breakpoint, remove_breakpoint>: Now
TARGET_DEFAULT_NORETURN methods.
<info_proc>: Now returns bool.
<to_magic>: Delete.
(OPS_MAGIC): Delete.
(current_target): Delete.  All references replaced by references
to ...
(target_stack): ... this.  New.
(target_shortname, target_longname): Adjust.
(target_can_run): Now a function declaration.
(default_child_has_all_memory, default_child_has_memory)
(default_child_has_stack, default_child_has_registers)
(default_child_has_execution): Remove target_ops parameter.
(complete_target_initialization): Delete.
(memory_breakpoint_target): New template class.
(test_target_ops): Refactor as a C++ class with virtual methods.
* make-target-delegates (NAME_PART): Tighten.
(POINTER_PART, CP_SYMBOL): New.
(SIMPLE_RETURN_PART): Reimplement.
(VEC_RETURN_PART): Expect less.
(RETURN_PART, VIRTUAL_PART): New.
(METHOD): Adjust to C++ virtual methods.
(scan_target_h): Remove reference to C99.
(dname): Output "target_ops::" prefix.
(write_function_header): Adjust to output a C++ class method.
(write_declaration): New.
(write_delegator): Adjust to output a C++ class method.
(tdname): Output "dummy_target::" prefix.
(write_tdefault, write_debugmethod): Adjust to output a C++ class
method.
(tdefault_names, debug_names): Delete.
(return_types, tdefaults, styles, argtypes_array): New.
(top level): All methods are delegators.
(print_class): New.
(top level): Print dummy_target and debug_target classes.
* target-delegates.c: Regenerate.
* target-debug.h (target_debug_print_enum_info_proc_what)
(target_debug_print_thread_control_capabilities)
(target_debug_print_thread_info_p): New.
* target.c (dummy_target): Delete.
(the_dummy_target, the_debug_target): New.
(target_stack): Now extern.
(set_targetdebug): Push/unpush debug target.
(default_child_has_all_memory, default_child_has_memory)
(default_child_has_stack, default_child_has_registers)
(default_child_has_execution): Remove target_ops parameter.
(complete_target_initialization): Delete.
(add_target_with_completer): No longer call
complete_target_initialization.
(target_supports_terminal_ours): Use regular delegation.
(update_current_target): Delete.
(push_target): No longer check magic number.  Don't call
update_current_target.
(unpush_target): Don't call update_current_target.
(target_is_pushed): No longer check magic number.
(target_require_runnable): Skip for all stratums over
process_stratum.
(target_ops::info_proc): New.
(target_info_proc): Use find_target_at and
find_default_run_target.
(target_supports_disable_randomization): Use regular delegation.
(target_get_osdata): Use find_target_at.
(target_ops::open, target_ops::close, target_ops::can_attach)
(target_ops::attach, target_ops::can_create_inferior)
(target_ops::create_inferior, target_ops::can_run)
(target_can_run): New.
(default_fileio_target): Use regular delegation.
(target_ops::fileio_open, target_ops::fileio_pwrite)
(target_ops::fileio_pread, target_ops::fileio_fstat)
(target_ops::fileio_close, target_ops::fileio_unlink)
(target_ops::fileio_readlink): New.
(target_fileio_open_1, target_fileio_unlink)
(target_fileio_readlink): Always call the target method.  Handle
FILEIO_ENOSYS.
(return_zero, return_zero_has_execution): Delete.
(init_dummy_target): Delete.
(dummy_target::dummy_target, dummy_target::shortname)
(dummy_target::longname, dummy_target::doc)
(debug_target::debug_target, debug_target::shortname)
(debug_target::longname, debug_target::doc): New.
(target_supports_delete_record): Use regular delegation.
(setup_target_debug): Delete.
(maintenance_print_target_stack): Skip debug_stratum.
(initialize_targets): Instantiate the_dummy_target and
the_debug_target.
* auxv.c (target_auxv_parse): Remove 'ops' parameter.  Adjust to
use target_stack.
(target_auxv_search, fprint_target_auxv): Adjust.
(info_auxv_command): Adjust to use target_stack.
* auxv.h (target_auxv_parse): Remove 'ops' parameter.
* exceptions.c (print_flush): Handle a NULL target_stack.
* regcache.c (target_ops_no_register): Refactor as class with
virtual methods.

* exec.c (exec_target): New class.
(exec_ops): Now an exec_target.
(exec_open, exec_close_1, exec_get_section_table)
(exec_xfer_partial, exec_files_info, exec_has_memory)
(exec_make_note_section): Refactor as exec_target methods.
(exec_file_clear, ignore, exec_remove_breakpoint, init_exec_ops):
Delete.
(exec_target::find_memory_regions): New.
(_initialize_exec): Don't call init_exec_ops.
* gdbcore.h (exec_file_clear): Delete.

* corefile.c (core_target): Delete.
(core_file_command): Adjust.
* corelow.c (core_target): New class.
(the_core_target): New.
(core_close): Remove target_ops parameter.
(core_close_cleanup): Adjust.
(core_target::close): New.
(core_open, core_detach, get_core_registers, core_files_info)
(core_xfer_partial, core_thread_alive, core_read_description)
(core_pid_to_str, core_thread_name, core_has_memory)
(core_has_stack, core_has_registers, core_info_proc): Rework as
core_target methods.
(ignore, core_remove_breakpoint, init_core_ops): Delete.
(_initialize_corelow): Initialize the_core_target.
* gdbcore.h (core_target): Delete.
(the_core_target): New.

* ctf.c: (ctf_target): New class.
(ctf_ops): Now a ctf_target.
(ctf_open, ctf_close, ctf_files_info, ctf_fetch_registers)
(ctf_xfer_partial, ctf_get_trace_state_variable_value)
(ctf_trace_find, ctf_traceframe_info): Refactor as ctf_target
methods.
(init_ctf_ops): Delete.
(_initialize_ctf): Don't call it.
* tracefile-tfile.c (tfile_target): New class.
(tfile_ops): Now a tfile_target.
(tfile_open, tfile_close, tfile_files_info)
(tfile_get_tracepoint_status, tfile_trace_find)
(tfile_fetch_registers, tfile_xfer_partial)
(tfile_get_trace_state_variable_value, tfile_traceframe_info):
Refactor as tfile_target methods.
(tfile_xfer_partial_features): Remove target_ops parameter.
(init_tfile_ops): Delete.
(_initialize_tracefile_tfile): Don't call it.
* tracefile.c (tracefile_has_all_memory, tracefile_has_memory)
(tracefile_has_stack, tracefile_has_registers)
(tracefile_thread_alive, tracefile_get_trace_status): Refactor as
tracefile_target methods.
(init_tracefile_ops): Delete.
(tracefile_target::tracefile_target): New.
* tracefile.h: Include "target.h".
(tracefile_target): New class.
(init_tracefile_ops): Delete.

* spu-multiarch.c (spu_multiarch_target): New class.
(spu_ops): Now a spu_multiarch_target.
(spu_thread_architecture, spu_region_ok_for_hw_watchpoint)
(spu_fetch_registers, spu_store_registers, spu_xfer_partial)
(spu_search_memory, spu_mourn_inferior): Refactor as
spu_multiarch_target methods.
(init_spu_ops): Delete.
(_initialize_spu_multiarch): Remove references to init_spu_ops,
complete_target_initialization.

* ravenscar-thread.c (ravenscar_thread_target): New class.
(ravenscar_ops): Now a ravenscar_thread_target.
(ravenscar_resume, ravenscar_wait, ravenscar_update_thread_list)
(ravenscar_thread_alive, ravenscar_pid_to_str)
(ravenscar_fetch_registers, ravenscar_store_registers)
(ravenscar_prepare_to_store, ravenscar_stopped_by_sw_breakpoint)
(ravenscar_stopped_by_hw_breakpoint)
(ravenscar_stopped_by_watchpoint, ravenscar_stopped_data_address)
(ravenscar_mourn_inferior, ravenscar_core_of_thread)
(ravenscar_get_ada_task_ptid): Refactor as ravenscar_thread_target
methods.
(init_ravenscar_thread_ops): Delete.
(_initialize_ravenscar): Remove references to
init_ravenscar_thread_ops and complete_target_initialization.

* bsd-uthread.c (bsd_uthread_ops_hack): Delete.
(bsd_uthread_target): New class.
(bsd_uthread_ops): Now a bsd_uthread_target.
(bsd_uthread_activate): Adjust to refer to bsd_uthread_ops.
(bsd_uthread_close, bsd_uthread_mourn_inferior)
(bsd_uthread_fetch_registers, bsd_uthread_store_registers)
(bsd_uthread_wait, bsd_uthread_resume, bsd_uthread_thread_alive)
(bsd_uthread_update_thread_list, bsd_uthread_extra_thread_info)
(bsd_uthread_pid_to_str): Refactor as bsd_uthread_target methods.
(bsd_uthread_target): Delete function.
(_initialize_bsd_uthread): Remove reference to
complete_target_initialization.

* bfd-target.c (target_bfd_data): Delete.  Fields folded into ...
(target_bfd): ... this new class.
(target_bfd_xfer_partial, target_bfd_get_section_table)
(target_bfd_close): Refactor as target_bfd methods.
(target_bfd::~target_bfd): New.
(target_bfd_reopen): Adjust.
(target_bfd::close): New.

* record-btrace.c (record_btrace_target): New class.
(record_btrace_ops): Now a record_btrace_target.
(record_btrace_open, record_btrace_stop_recording)
(record_btrace_disconnect, record_btrace_close)
(record_btrace_async, record_btrace_info)
(record_btrace_insn_history, record_btrace_insn_history_range)
(record_btrace_insn_history_from, record_btrace_call_history)
(record_btrace_call_history_range)
(record_btrace_call_history_from, record_btrace_record_method)
(record_btrace_is_replaying, record_btrace_will_replay)
(record_btrace_xfer_partial, record_btrace_insert_breakpoint)
(record_btrace_remove_breakpoint, record_btrace_fetch_registers)
(record_btrace_store_registers, record_btrace_prepare_to_store)
(record_btrace_to_get_unwinder)
(record_btrace_to_get_tailcall_unwinder, record_btrace_resume)
(record_btrace_commit_resume, record_btrace_wait)
(record_btrace_stop, record_btrace_can_execute_reverse)
(record_btrace_stopped_by_sw_breakpoint)
(record_btrace_supports_stopped_by_sw_breakpoint)
(record_btrace_stopped_by_hw_breakpoint)
(record_btrace_supports_stopped_by_hw_breakpoint)
(record_btrace_update_thread_list, record_btrace_thread_alive)
(record_btrace_goto_begin, record_btrace_goto_end)
(record_btrace_goto, record_btrace_stop_replaying_all)
(record_btrace_execution_direction)
(record_btrace_prepare_to_generate_core)
(record_btrace_done_generating_core): Refactor as
record_btrace_target methods.
(init_record_btrace_ops): Delete.
(_initialize_record_btrace): Remove reference to
init_record_btrace_ops.
* record-full.c (RECORD_FULL_IS_REPLAY): Adjust to always refer to
the execution_direction global.
(record_full_base_target, record_full_target)
(record_full_core_target): New classes.
(record_full_ops): Now a record_full_target.
(record_full_core_ops): Now a record_full_core_target.
(record_full_target::detach, record_full_target::disconnect)
(record_full_core_target::disconnect)
(record_full_target::mourn_inferior, record_full_target::kill):
New.
(record_full_open, record_full_close, record_full_async): Refactor
as methods of the record_full_base_target class.
(record_full_resume, record_full_commit_resume): Refactor
as methods of the record_full_target class.
(record_full_wait, record_full_stopped_by_watchpoint)
(record_full_stopped_data_address)
(record_full_stopped_by_sw_breakpoint)
(record_full_supports_stopped_by_sw_breakpoint)
(record_full_stopped_by_hw_breakpoint)
(record_full_supports_stopped_by_hw_breakpoint): Refactor as
methods of the record_full_base_target class.
(record_full_store_registers, record_full_xfer_partial)
(record_full_insert_breakpoint, record_full_remove_breakpoint):
Refactor as methods of the record_full_target class.
(record_full_can_execute_reverse, record_full_get_bookmark)
(record_full_goto_bookmark, record_full_execution_direction)
(record_full_record_method, record_full_info, record_full_delete)
(record_full_is_replaying, record_full_will_replay)
(record_full_goto_begin, record_full_goto_end, record_full_goto)
(record_full_stop_replaying): Refactor as methods of the
record_full_base_target class.
(record_full_core_resume, record_full_core_kill)
(record_full_core_fetch_registers)
(record_full_core_prepare_to_store)
(record_full_core_store_registers, record_full_core_xfer_partial)
(record_full_core_insert_breakpoint)
(record_full_core_remove_breakpoint)
(record_full_core_has_execution): Refactor
as methods of the record_full_core_target class.
(record_full_base_target::supports_delete_record): New.
(init_record_full_ops): Delete.
(init_record_full_core_ops): Delete.
(record_full_save): Refactor as method of the
record_full_base_target class.
(_initialize_record_full): Remove references to
init_record_full_ops and init_record_full_core_ops.

* remote.c (remote_target, extended_remote_target): New classes.
(remote_ops): Now a remote_target.
(extended_remote_ops): Now an extended_remote_target.
(remote_insert_fork_catchpoint, remote_remove_fork_catchpoint)
(remote_insert_vfork_catchpoint, remote_remove_vfork_catchpoint)
(remote_insert_exec_catchpoint, remote_remove_exec_catchpoint)
(remote_pass_signals, remote_set_syscall_catchpoint)
(remote_program_signals, )
(remote_thread_always_alive): Remove target_ops parameter.
(remote_thread_alive, remote_thread_name)
(remote_update_thread_list, remote_threads_extra_info)
(remote_static_tracepoint_marker_at)
(remote_static_tracepoint_markers_by_strid)
(remote_get_ada_task_ptid, remote_close, remote_start_remote)
(remote_open): Refactor as methods of remote_target.
(extended_remote_open, extended_remote_detach)
(extended_remote_attach, extended_remote_post_attach):
(extended_remote_supports_disable_randomization)
(extended_remote_create_inferior): : Refactor as method of
extended_remote_target.
(remote_set_permissions, remote_open_1, remote_detach)
(remote_follow_fork, remote_follow_exec, remote_disconnect)
(remote_resume, remote_commit_resume, remote_stop)
(remote_interrupt, remote_pass_ctrlc, remote_terminal_inferior)
(remote_terminal_ours, remote_wait, remote_fetch_registers)
(remote_prepare_to_store, remote_store_registers)
(remote_flash_erase, remote_flash_done, remote_files_info)
(remote_kill, remote_mourn, remote_insert_breakpoint)
(remote_remove_breakpoint, remote_insert_watchpoint)
(remote_watchpoint_addr_within_range)
(remote_remove_watchpoint, remote_region_ok_for_hw_watchpoint)
(remote_check_watch_resources, remote_stopped_by_sw_breakpoint)
(remote_supports_stopped_by_sw_breakpoint)
(remote_stopped_by_hw_breakpoint)
(remote_supports_stopped_by_hw_breakpoint)
(remote_stopped_by_watchpoint, remote_stopped_data_address)
(remote_insert_hw_breakpoint, remote_remove_hw_breakpoint)
(remote_verify_memory): Refactor as methods of remote_target.
(remote_write_qxfer, remote_read_qxfer): Remove target_ops
parameter.
(remote_xfer_partial, remote_get_memory_xfer_limit)
(remote_search_memory, remote_rcmd, remote_memory_map)
(remote_pid_to_str, remote_get_thread_local_address)
(remote_get_tib_address, remote_read_description): Refactor as
methods of remote_target.
(remote_target::fileio_open, remote_target::fileio_pwrite)
(remote_target::fileio_pread, remote_target::fileio_close): New.
(remote_hostio_readlink, remote_hostio_fstat)
(remote_filesystem_is_local, remote_can_execute_reverse)
(remote_supports_non_stop, remote_supports_disable_randomization)
(remote_supports_multi_process, remote_supports_cond_breakpoints)
(remote_supports_enable_disable_tracepoint)
(remote_supports_string_tracing)
(remote_can_run_breakpoint_commands, remote_trace_init)
(remote_download_tracepoint, remote_can_download_tracepoint)
(remote_download_trace_state_variable, remote_enable_tracepoint)
(remote_disable_tracepoint, remote_trace_set_readonly_regions)
(remote_trace_start, remote_get_trace_status)
(remote_get_tracepoint_status, remote_trace_stop)
(remote_trace_find, remote_get_trace_state_variable_value)
(remote_save_trace_data, remote_get_raw_trace_data)
(remote_set_disconnected_tracing, remote_core_of_thread)
(remote_set_circular_trace_buffer, remote_traceframe_info)
(remote_get_min_fast_tracepoint_insn_len)
(remote_set_trace_buffer_size, remote_set_trace_notes)
(remote_use_agent, remote_can_use_agent, remote_enable_btrace)
(remote_disable_btrace, remote_teardown_btrace)
(remote_read_btrace, remote_btrace_conf)
(remote_augmented_libraries_svr4_read, remote_load)
(remote_pid_to_exec_file, remote_can_do_single_step)
(remote_execution_direction, remote_thread_handle_to_thread_info):
Refactor as methods of remote_target.
(init_remote_ops, init_extended_remote_ops): Delete.
(remote_can_async_p, remote_is_async_p, remote_async)
(remote_thread_events, remote_upload_tracepoints)
(remote_upload_trace_state_variables): Refactor as methods of
remote_target.
(_initialize_remote): Remove references to init_remote_ops and
init_extended_remote_ops.

* remote-sim.c (gdbsim_target): New class.
(gdbsim_fetch_register, gdbsim_store_register, gdbsim_kill)
(gdbsim_load, gdbsim_create_inferior, gdbsim_open, gdbsim_close)
(gdbsim_detach, gdbsim_resume, gdbsim_interrupt)
(gdbsim_wait, gdbsim_prepare_to_store, gdbsim_xfer_partial)
(gdbsim_files_info, gdbsim_mourn_inferior, gdbsim_thread_alive)
(gdbsim_pid_to_str, gdbsim_has_all_memory, gdbsim_has_memory):
Refactor as methods of gdbsim_target.
(gdbsim_ops): Now a gdbsim_target.
(init_gdbsim_ops): Delete.
(gdbsim_cntrl_c): Adjust.
(_initialize_remote_sim): Remove reference to init_gdbsim_ops.

* amd64-linux-nat.c (amd64_linux_nat_target): New class.
(the_amd64_linux_nat_target): New.
(amd64_linux_fetch_inferior_registers)
(amd64_linux_store_inferior_registers): Refactor as methods of
amd64_linux_nat_target.
(_initialize_amd64_linux_nat): Adjust.  Set linux_target.
* i386-linux-nat.c: Don't include "linux-nat.h".
(i386_linux_nat_target): New class.
(the_i386_linux_nat_target): New.
(i386_linux_fetch_inferior_registers)
(i386_linux_store_inferior_registers, i386_linux_resume): Refactor
as methods of i386_linux_nat_target.
(_initialize_i386_linux_nat): Adjust.  Set linux_target.
* inf-child.c (inf_child_ops): Delete.
(inf_child_fetch_inferior_registers)
(inf_child_store_inferior_registers): Delete.
(inf_child_post_attach, inf_child_prepare_to_store): Refactor as
methods of inf_child_target.
(inf_child_target::supports_terminal_ours)
(inf_child_target::terminal_init)
(inf_child_target::terminal_inferior)
(inf_child_target::terminal_ours_for_output)
(inf_child_target::terminal_ours, inf_child_target::interrupt)
(inf_child_target::pass_ctrlc, inf_child_target::terminal_info):
New.
(inf_child_open, inf_child_disconnect, inf_child_close)
(inf_child_mourn_inferior, inf_child_maybe_unpush_target)
(inf_child_post_startup_inferior, inf_child_can_run)
(inf_child_pid_to_exec_file): Refactor as methods of
inf_child_target.
(inf_child_follow_fork): Delete.
(inf_child_target::can_create_inferior)
(inf_child_target::can_attach): New.
(inf_child_target::has_all_memory, inf_child_target::has_memory)
(inf_child_target::has_stack, inf_child_target::has_registers)
(inf_child_target::has_execution): New.
(inf_child_fileio_open, inf_child_fileio_pwrite)
(inf_child_fileio_pread, inf_child_fileio_fstat)
(inf_child_fileio_close, inf_child_fileio_unlink)
(inf_child_fileio_readlink, inf_child_use_agent)
(inf_child_can_use_agent): Refactor as methods of
inf_child_target.
(return_zero, inf_child_target): Delete.
(inf_child_target::inf_child_target): New.
* inf-child.h: Include "target.h".
(inf_child_target): Delete function prototype.
(inf_child_target): New class.
(inf_child_open_target, inf_child_mourn_inferior)
(inf_child_maybe_unpush_target): Delete.
* inf-ptrace.c (inf_ptrace_target::~inf_ptrace_target): New.
(inf_ptrace_follow_fork, inf_ptrace_insert_fork_catchpoint)
(inf_ptrace_remove_fork_catchpoint, inf_ptrace_create_inferior)
(inf_ptrace_post_startup_inferior, inf_ptrace_mourn_inferior)
(inf_ptrace_attach, inf_ptrace_post_attach, inf_ptrace_detach)
(inf_ptrace_detach_success, inf_ptrace_kill, inf_ptrace_resume)
(inf_ptrace_wait, inf_ptrace_xfer_partial)
(inf_ptrace_thread_alive, inf_ptrace_files_info)
(inf_ptrace_pid_to_str, inf_ptrace_auxv_parse): Refactor as
methods of inf_ptrace_target.
(inf_ptrace_target): Delete function.
* inf-ptrace.h: Include "inf-child.h".
(inf_ptrace_target): Delete function declaration.
(inf_ptrace_target): New class.
(inf_ptrace_trad_target, inf_ptrace_detach_success): Delete.
* linux-nat.c (linux_target): New.
(linux_ops, linux_ops_saved, super_xfer_partial): Delete.
(linux_nat_target::~linux_nat_target): New.
(linux_child_post_attach, linux_child_post_startup_inferior)
(linux_child_follow_fork, linux_child_insert_fork_catchpoint)
(linux_child_remove_fork_catchpoint)
(linux_child_insert_vfork_catchpoint)
(linux_child_remove_vfork_catchpoint)
(linux_child_insert_exec_catchpoint)
(linux_child_remove_exec_catchpoint)
(linux_child_set_syscall_catchpoint, linux_nat_pass_signals)
(linux_nat_create_inferior, linux_nat_attach, linux_nat_detach)
(linux_nat_resume, linux_nat_stopped_by_watchpoint)
(linux_nat_stopped_data_address)
(linux_nat_stopped_by_sw_breakpoint)
(linux_nat_supports_stopped_by_sw_breakpoint)
(linux_nat_stopped_by_hw_breakpoint)
(linux_nat_supports_stopped_by_hw_breakpoint, linux_nat_wait)
(linux_nat_kill, linux_nat_mourn_inferior)
(linux_nat_xfer_partial, linux_nat_thread_alive)
(linux_nat_update_thread_list, linux_nat_pid_to_str)
(linux_nat_thread_name, linux_child_pid_to_exec_file)
(linux_child_static_tracepoint_markers_by_strid)
(linux_nat_is_async_p, linux_nat_can_async_p)
(linux_nat_supports_non_stop, linux_nat_always_non_stop_p)
(linux_nat_supports_multi_process)
(linux_nat_supports_disable_randomization, linux_nat_async)
(linux_nat_stop, linux_nat_close, linux_nat_thread_address_space)
(linux_nat_core_of_thread, linux_nat_filesystem_is_local)
(linux_nat_fileio_open, linux_nat_fileio_readlink)
(linux_nat_fileio_unlink, linux_nat_thread_events): Refactor as
methods of linux_nat_target.
(linux_nat_wait_1, linux_xfer_siginfo, linux_proc_xfer_partial)
(linux_proc_xfer_spu, linux_nat_xfer_osdata): Remove target_ops
parameter.
(check_stopped_by_watchpoint): Adjust.
(linux_xfer_partial): Delete.
(linux_target_install_ops, linux_target, linux_nat_add_target):
Delete.
(linux_nat_target::linux_nat_target): New.
* linux-nat.h: Include "inf-ptrace.h".
(linux_nat_target): New.
(linux_target, linux_target_install_ops, linux_nat_add_target):
Delete function declarations.
(linux_target): Declare global.
* linux-thread-db.c (thread_db_target): New.
(thread_db_target::thread_db_target): New.
(thread_db_ops): Delete.
(the_thread_db_target): New.
(thread_db_detach, thread_db_wait, thread_db_mourn_inferior)
(thread_db_update_thread_list, thread_db_pid_to_str)
(thread_db_extra_thread_info)
(thread_db_thread_handle_to_thread_info)
(thread_db_get_thread_local_address, thread_db_get_ada_task_ptid)
(thread_db_resume): Refactor as methods of thread_db_target.
(init_thread_db_ops): Delete.
(_initialize_thread_db): Remove reference to init_thread_db_ops.
* x86-linux-nat.c: Don't include "linux-nat.h".
(super_post_startup_inferior): Delete.
(x86_linux_nat_target::~x86_linux_nat_target): New.
(x86_linux_child_post_startup_inferior)
(x86_linux_read_description, x86_linux_enable_btrace)
(x86_linux_disable_btrace, x86_linux_teardown_btrace)
(x86_linux_read_btrace, x86_linux_btrace_conf): Refactor as
methods of x86_linux_nat_target.
(x86_linux_create_target): Delete.  Bits folded ...
(x86_linux_add_target): ... here.  Now takes a linux_nat_target
pointer.
* x86-linux-nat.h: Include "linux-nat.h" and "x86-nat.h".
(x86_linux_nat_target): New class.
(x86_linux_create_target): Delete.
(x86_linux_add_target): Now takes a linux_nat_target pointer.
* x86-nat.c (x86_insert_watchpoint, x86_remove_watchpoint)
(x86_region_ok_for_watchpoint, x86_stopped_data_address)
(x86_stopped_by_watchpoint, x86_insert_hw_breakpoint)
(x86_remove_hw_breakpoint, x86_can_use_hw_breakpoint)
(x86_stopped_by_hw_breakpoint): Remove target_ops parameter and
make extern.
(x86_use_watchpoints): Delete.
* x86-nat.h: Include "breakpoint.h" and "target.h".
(x86_use_watchpoints): Delete.
(x86_can_use_hw_breakpoint, x86_region_ok_for_hw_watchpoint)
(x86_stopped_by_watchpoint, x86_stopped_data_address)
(x86_insert_watchpoint, x86_remove_watchpoint)
(x86_insert_hw_breakpoint, x86_remove_hw_breakpoint)
(x86_stopped_by_hw_breakpoint): New declarations.
(x86_nat_target): New template class.

* ppc-linux-nat.c (ppc_linux_nat_target): New class.
(the_ppc_linux_nat_target): New.
(ppc_linux_fetch_inferior_registers)
(ppc_linux_can_use_hw_breakpoint)
(ppc_linux_region_ok_for_hw_watchpoint)
(ppc_linux_ranged_break_num_registers)
(ppc_linux_insert_hw_breakpoint, ppc_linux_remove_hw_breakpoint)
(ppc_linux_insert_mask_watchpoint)
(ppc_linux_remove_mask_watchpoint)
(ppc_linux_can_accel_watchpoint_condition)
(ppc_linux_insert_watchpoint, ppc_linux_remove_watchpoint)
(ppc_linux_stopped_data_address, ppc_linux_stopped_by_watchpoint)
(ppc_linux_watchpoint_addr_within_range)
(ppc_linux_masked_watch_num_registers)
(ppc_linux_store_inferior_registers, ppc_linux_auxv_parse)
(ppc_linux_read_description): Refactor as methods of
ppc_linux_nat_target.
(_initialize_ppc_linux_nat): Adjust.  Set linux_target.

* procfs.c (procfs_xfer_partial): Delete forward declaration.
(procfs_target): New class.
(the_procfs_target): New.
(procfs_target): Delete function.
(procfs_auxv_parse, procfs_attach, procfs_detach)
(procfs_fetch_registers, procfs_store_registers, procfs_wait)
(procfs_xfer_partial, procfs_resume, procfs_pass_signals)
(procfs_files_info, procfs_kill_inferior, procfs_mourn_inferior)
(procfs_create_inferior, procfs_update_thread_list)
(procfs_thread_alive, procfs_pid_to_str)
(procfs_can_use_hw_breakpoint, procfs_stopped_by_watchpoint)
(procfs_stopped_data_address, procfs_insert_watchpoint)
(procfs_remove_watchpoint, procfs_region_ok_for_hw_watchpoint)
(proc_find_memory_regions, procfs_info_proc)
(procfs_make_note_section): Refactor as methods of procfs_target.
(_initialize_procfs): Adjust.
* sol-thread.c (sol_thread_target): New class.
(sol_thread_ops): Now a sol_thread_target.
(sol_thread_detach, sol_thread_resume, sol_thread_wait)
(sol_thread_fetch_registers, sol_thread_store_registers)
(sol_thread_xfer_partial, sol_thread_mourn_inferior)
(sol_thread_alive, solaris_pid_to_str, sol_update_thread_list)
(sol_get_ada_task_ptid): Refactor as methods of sol_thread_target.
(init_sol_thread_ops): Delete.
(_initialize_sol_thread): Adjust.  Remove references to
init_sol_thread_ops and complete_target_initialization.

* windows-nat.c (windows_nat_target): New class.
(windows_fetch_inferior_registers)
(windows_store_inferior_registers, windows_resume, windows_wait)
(windows_attach, windows_detach, windows_pid_to_exec_file)
(windows_files_info, windows_create_inferior)
(windows_mourn_inferior, windows_interrupt, windows_kill_inferior)
(windows_close, windows_pid_to_str, windows_xfer_partial)
(windows_get_tib_address, windows_get_ada_task_ptid)
(windows_thread_name, windows_thread_alive): Refactor as
windows_nat_target methods.
(do_initial_windows_stuff): Adjust.
(windows_target): Delete function.
(_initialize_windows_nat): Adjust.

* darwin-nat.c (darwin_resume, darwin_wait_to, darwin_interrupt)
(darwin_mourn_inferior, darwin_kill_inferior)
(darwin_create_inferior, darwin_attach, darwin_detach)
(darwin_pid_to_str, darwin_thread_alive, darwin_xfer_partial)
(darwin_pid_to_exec_file, darwin_get_ada_task_ptid)
(darwin_supports_multi_process): Refactor as darwin_nat_target
methods.
(darwin_resume_to, darwin_files_info): Delete.
(_initialize_darwin_inferior): Rename to ...
(_initialize_darwin_nat): ... this.  Adjust to C++ification.
* darwin-nat.h: Include "inf-child.h".
(darwin_nat_target): New class.
(darwin_complete_target): Delete.
* i386-darwin-nat.c (i386_darwin_nat_target): New class.
(darwin_target): New.
(i386_darwin_fetch_inferior_registers)
(i386_darwin_store_inferior_registers): Refactor as methods of
darwin_nat_target.
(darwin_complete_target): Delete, with ...
(_initialize_i386_darwin_nat): ... bits factored out here.

* alpha-linux-nat.c (alpha_linux_nat_target): New class.
(the_alpha_linux_nat_target): New.
(alpha_linux_register_u_offset): Refactor as
alpha_linux_nat_target method.
(_initialize_alpha_linux_nat): Adjust.
* linux-nat-trad.c (inf_ptrace_register_u_offset): Delete.
(inf_ptrace_fetch_register, inf_ptrace_fetch_registers)
(inf_ptrace_store_register, inf_ptrace_store_registers): Refact as
methods of linux_nat_trad_target.
(linux_trad_target): Delete.
* linux-nat-trad.h (linux_trad_target): Delete function.
(linux_nat_trad_target): New class.
* mips-linux-nat.c (mips_linux_nat_target): New class.
(super_fetch_registers, super_store_registers, super_close):
Delete.
(the_mips_linux_nat_target): New.
(mips64_linux_regsets_fetch_registers)
(mips64_linux_regsets_store_registers)
(mips64_linux_fetch_registers, mips64_linux_store_registers)
(mips_linux_register_u_offset, mips_linux_read_description)
(mips_linux_can_use_hw_breakpoint)
(mips_linux_stopped_by_watchpoint)
(mips_linux_stopped_data_address)
(mips_linux_region_ok_for_hw_watchpoint)
(mips_linux_insert_watchpoint, mips_linux_remove_watchpoint)
(mips_linux_close): Refactor as methods of mips_linux_nat.
(_initialize_mips_linux_nat): Adjust to C++ification.

* aix-thread.c (aix_thread_target): New class.
(aix_thread_ops): Now an aix_thread_target.
(aix_thread_detach, aix_thread_resume, aix_thread_wait)
(aix_thread_fetch_registers, aix_thread_store_registers)
(aix_thread_xfer_partial, aix_thread_mourn_inferior)
(aix_thread_thread_alive, aix_thread_pid_to_str)
(aix_thread_extra_thread_info, aix_thread_get_ada_task_ptid):
Refactor as methods of aix_thread_target.
(init_aix_thread_ops): Delete.
(_initialize_aix_thread): Remove references to init_aix_thread_ops
and complete_target_initialization.
* rs6000-nat.c (rs6000_xfer_shared_libraries): Delete.
(rs6000_nat_target): New class.
(the_rs6000_nat_target): New.
(rs6000_fetch_inferior_registers, rs6000_store_inferior_registers)
(rs6000_xfer_partial, rs6000_wait, rs6000_create_inferior)
(rs6000_xfer_shared_libraries): Refactor as rs6000_nat_target methods.
(super_create_inferior): Delete.
(_initialize_rs6000_nat): Adjust to C++ification.

* arm-linux-nat.c (arm_linux_nat_target): New class.
(the_arm_linux_nat_target): New.
(arm_linux_fetch_inferior_registers)
(arm_linux_store_inferior_registers, arm_linux_read_description)
(arm_linux_can_use_hw_breakpoint, arm_linux_insert_hw_breakpoint)
(arm_linux_remove_hw_breakpoint)
(arm_linux_region_ok_for_hw_watchpoint)
(arm_linux_insert_watchpoint, arm_linux_remove_watchpoint)
(arm_linux_stopped_data_address, arm_linux_stopped_by_watchpoint)
(arm_linux_watchpoint_addr_within_range): Refactor as methods of
arm_linux_nat_target.
(_initialize_arm_linux_nat): Adjust to C++ification.

* aarch64-linux-nat.c (aarch64_linux_nat_target): New class.
(the_aarch64_linux_nat_target): New.
(aarch64_linux_fetch_inferior_registers)
(aarch64_linux_store_inferior_registers)
(aarch64_linux_child_post_startup_inferior)
(aarch64_linux_read_description)
(aarch64_linux_can_use_hw_breakpoint)
(aarch64_linux_insert_hw_breakpoint)
(aarch64_linux_remove_hw_breakpoint)
(aarch64_linux_insert_watchpoint, aarch64_linux_remove_watchpoint)
(aarch64_linux_region_ok_for_hw_watchpoint)
(aarch64_linux_stopped_data_address)
(aarch64_linux_stopped_by_watchpoint)
(aarch64_linux_watchpoint_addr_within_range)
(aarch64_linux_can_do_single_step): Refactor as methods of
aarch64_linux_nat_target.
(super_post_startup_inferior): Delete.
(_initialize_aarch64_linux_nat): Adjust to C++ification.

* hppa-linux-nat.c (hppa_linux_nat_target): New class.
(the_hppa_linux_nat_target): New.
(hppa_linux_fetch_inferior_registers)
(hppa_linux_store_inferior_registers): Refactor as methods of
hppa_linux_nat_target.
(_initialize_hppa_linux_nat): Adjust to C++ification.

* ia64-linux-nat.c (ia64_linux_nat_target): New class.
(the_ia64_linux_nat_target): New.
(ia64_linux_insert_watchpoint, ia64_linux_remove_watchpoint)
(ia64_linux_stopped_data_address)
(ia64_linux_stopped_by_watchpoint, ia64_linux_fetch_registers)
(ia64_linux_store_registers, ia64_linux_xfer_partial): Refactor as
ia64_linux_nat_target methods.
(super_xfer_partial): Delete.
(_initialize_ia64_linux_nat): Adjust to C++ification.

* m32r-linux-nat.c (m32r_linux_nat_target): New class.
(the_m32r_linux_nat_target): New.
(m32r_linux_fetch_inferior_registers)
(m32r_linux_store_inferior_registers): Refactor as
m32r_linux_nat_target methods.
(_initialize_m32r_linux_nat): Adjust to C++ification.

* m68k-linux-nat.c (m68k_linux_nat_target): New class.
(the_m68k_linux_nat_target): New.
(m68k_linux_fetch_inferior_registers)
(m68k_linux_store_inferior_registers): Refactor as
m68k_linux_nat_target methods.
(_initialize_m68k_linux_nat): Adjust to C++ification.

* s390-linux-nat.c (s390_linux_nat_target): New class.
(the_s390_linux_nat_target): New.
(s390_linux_fetch_inferior_registers)
(s390_linux_store_inferior_registers, s390_stopped_by_watchpoint)
(s390_insert_watchpoint, s390_remove_watchpoint)
(s390_can_use_hw_breakpoint, s390_insert_hw_breakpoint)
(s390_remove_hw_breakpoint, s390_region_ok_for_hw_watchpoint)
(s390_auxv_parse, s390_read_description): Refactor as methods of
s390_linux_nat_target.
(_initialize_s390_nat): Adjust to C++ification.

* sparc-linux-nat.c (sparc_linux_nat_target): New class.
(the_sparc_linux_nat_target): New.
(_initialize_sparc_linux_nat): Adjust to C++ification.
* sparc-nat.c (sparc_fetch_inferior_registers)
(sparc_store_inferior_registers): Remove target_ops parameter.
* sparc-nat.h (sparc_fetch_inferior_registers)
(sparc_store_inferior_registers): Remove target_ops parameter.
* sparc64-linux-nat.c (sparc64_linux_nat_target): New class.
(the_sparc64_linux_nat_target): New.
(_initialize_sparc64_linux_nat): Adjust to C++ification.

* spu-linux-nat.c (spu_linux_nat_target): New class.
(the_spu_linux_nat_target): New.
(spu_child_post_startup_inferior, spu_child_post_attach)
(spu_child_wait, spu_fetch_inferior_registers)
(spu_store_inferior_registers, spu_xfer_partial)
(spu_can_use_hw_breakpoint): Refactor as spu_linux_nat_target
methods.
(_initialize_spu_nat): Adjust to C++ification.

* tilegx-linux-nat.c (tilegx_linux_nat_target): New class.
(the_tilegx_linux_nat_target): New.
(fetch_inferior_registers, store_inferior_registers):
Refactor as methods.
(_initialize_tile_linux_nat): Adjust to C++ification.

* xtensa-linux-nat.c (xtensa_linux_nat_target): New class.
(the_xtensa_linux_nat_target): New.
(xtensa_linux_fetch_inferior_registers)
(xtensa_linux_store_inferior_registers): Refactor as
xtensa_linux_nat_target methods.
(_initialize_xtensa_linux_nat): Adjust to C++ification.

* fbsd-nat.c (USE_SIGTRAP_SIGINFO): Delete.
(fbsd_pid_to_exec_file, fbsd_find_memory_regions)
(fbsd_find_memory_regions, fbsd_info_proc, fbsd_xfer_partial)
(fbsd_thread_alive, fbsd_pid_to_str, fbsd_thread_name)
(fbsd_update_thread_list, fbsd_resume, fbsd_wait)
(fbsd_stopped_by_sw_breakpoint)
(fbsd_supports_stopped_by_sw_breakpoint, fbsd_follow_fork)
(fbsd_insert_fork_catchpoint, fbsd_remove_fork_catchpoint)
(fbsd_insert_vfork_catchpoint, fbsd_remove_vfork_catchpoint)
(fbsd_post_startup_inferior, fbsd_post_attach)
(fbsd_insert_exec_catchpoint, fbsd_remove_exec_catchpoint)
(fbsd_set_syscall_catchpoint)
(super_xfer_partial, super_resume, super_wait)
(fbsd_supports_stopped_by_hw_breakpoint): Delete.
(fbsd_handle_debug_trap): Remove target_ops parameter.
(fbsd_nat_add_target): Delete.
* fbsd-nat.h: Include "inf-ptrace.h".
(fbsd_nat_add_target): Delete.
(USE_SIGTRAP_SIGINFO): Define.
(fbsd_nat_target): New class.

* amd64-bsd-nat.c (amd64bsd_fetch_inferior_registers)
(amd64bsd_store_inferior_registers): Remove target_ops parameter.
(amd64bsd_target): Delete.
* amd64-bsd-nat.h: New file.
* amd64-fbsd-nat.c: Include "amd64-bsd-nat.h" instead of
"x86-bsd-nat.h".
(amd64_fbsd_nat_target): New class.
(the_amd64_fbsd_nat_target): New.
(amd64fbsd_read_description): Refactor as method of
amd64_fbsd_nat_target.
(amd64_fbsd_nat_target::supports_stopped_by_hw_breakpoint): New.
(_initialize_amd64fbsd_nat): Adjust to C++ification.
* amd64-nat.h (amd64bsd_target): Delete function declaration.
* i386-bsd-nat.c (i386bsd_fetch_inferior_registers)
(i386bsd_store_inferior_registers): Remove target_ops parameter.
(i386bsd_target): Delete.
* i386-bsd-nat.h (i386bsd_target): Delete function declaration.
(i386bsd_fetch_inferior_registers)
(i386bsd_store_inferior_registers): Declare.
(i386_bsd_nat_target): New class.
* i386-fbsd-nat.c (i386_fbsd_nat_target): New class.
(the_i386_fbsd_nat_target): New.
(i386fbsd_resume, i386fbsd_read_description): Refactor as
i386_fbsd_nat_target methods.
(i386_fbsd_nat_target::supports_stopped_by_hw_breakpoint): New.
(_initialize_i386fbsd_nat): Adjust to C++ification.
* x86-bsd-nat.c (super_mourn_inferior): Delete.
(x86bsd_mourn_inferior, x86bsd_target): Delete.
(_initialize_x86_bsd_nat): Adjust to C++ification.
* x86-bsd-nat.h: Include "x86-nat.h".
(x86bsd_target): Delete declaration.
(x86bsd_nat_target): New class.

* aarch64-fbsd-nat.c (aarch64_fbsd_nat_target): New class.
(the_aarch64_fbsd_nat_target): New.
(aarch64_fbsd_fetch_inferior_registers)
(aarch64_fbsd_store_inferior_registers): Refactor as methods of
aarch64_fbsd_nat_target.
(_initialize_aarch64_fbsd_nat): Adjust to C++ification.
* alpha-bsd-nat.c (alpha_bsd_nat_target): New class.
(the_alpha_bsd_nat_target): New.
(alphabsd_fetch_inferior_registers)
(alphabsd_store_inferior_registers): Refactor as
alpha_bsd_nat_target methods.
(_initialize_alphabsd_nat): Refactor as methods of
alpha_bsd_nat_target.
* amd64-nbsd-nat.c: Include "amd64-bsd-nat.h".
(the_amd64_nbsd_nat_target): New.
(_initialize_amd64nbsd_nat): Adjust to C++ification.
* amd64-obsd-nat.c: Include "amd64-bsd-nat.h".
(the_amd64_obsd_nat_target): New.
(_initialize_amd64obsd_nat): Adjust to C++ification.
* arm-fbsd-nat.c (arm_fbsd_nat_target): New.
(the_arm_fbsd_nat_target): New.
(arm_fbsd_fetch_inferior_registers)
(arm_fbsd_store_inferior_registers, arm_fbsd_read_description):
(_initialize_arm_fbsd_nat): Refactor as methods of
arm_fbsd_nat_target.
(_initialize_arm_fbsd_nat): Adjust to C++ification.
* arm-nbsd-nat.c (arm_netbsd_nat_target): New class.
(the_arm_netbsd_nat_target): New.
(armnbsd_fetch_registers, armnbsd_store_registers): Refactor as
arm_netbsd_nat_target.
(_initialize_arm_netbsd_nat): Adjust to C++ification.
* hppa-nbsd-nat.c (hppa_nbsd_nat_target): New class.
(the_hppa_nbsd_nat_target): New.
(hppanbsd_fetch_registers, hppanbsd_store_registers): Refactor as
hppa_nbsd_nat_target methods.
(_initialize_hppanbsd_nat): Adjust to C++ification.
* hppa-obsd-nat.c (hppa_obsd_nat_target): New class.
(the_hppa_obsd_nat_target): New.
(hppaobsd_fetch_registers, hppaobsd_store_registers): Refactor as
methods of hppa_obsd_nat_target.
(_initialize_hppaobsd_nat): Adjust to C++ification.  Use
add_target.
* i386-nbsd-nat.c (the_i386_nbsd_nat_target): New.
(_initialize_i386nbsd_nat): Adjust to C++ification.  Use
add_target.
* i386-obsd-nat.c (the_i386_obsd_nat_target): New.
(_initialize_i386obsd_nat): Use add_target.
* m68k-bsd-nat.c (m68k_bsd_nat_target): New class.
(the_m68k_bsd_nat_target): New.
(m68kbsd_fetch_inferior_registers)
(m68kbsd_store_inferior_registers): Refactor as methods of
m68k_bsd_nat_target.
(_initialize_m68kbsd_nat): Adjust to C++ification.
* mips-fbsd-nat.c (mips_fbsd_nat_target): New class.
(the_mips_fbsd_nat_target): New.
(mips_fbsd_fetch_inferior_registers)
(mips_fbsd_store_inferior_registers): Refactor as methods of
mips_fbsd_nat_target.
(_initialize_mips_fbsd_nat): Adjust to C++ification.  Use
add_target.
* mips-nbsd-nat.c (mips_nbsd_nat_target): New class.
(the_mips_nbsd_nat_target): New.
(mipsnbsd_fetch_inferior_registers)
(mipsnbsd_store_inferior_registers): Refactor as methods of
mips_nbsd_nat_target.
(_initialize_mipsnbsd_nat): Adjust to C++ification.
* mips64-obsd-nat.c (mips64_obsd_nat_target): New class.
(the_mips64_obsd_nat_target): New.
(mips64obsd_fetch_inferior_registers)
(mips64obsd_store_inferior_registers): Refactor as methods of
mips64_obsd_nat_target.
(_initialize_mips64obsd_nat): Adjust to C++ification.  Use
add_target.
* nbsd-nat.c (nbsd_pid_to_exec_file): Refactor as method of
nbsd_nat_target.
* nbsd-nat.h: Include "inf-ptrace.h".
(nbsd_nat_target): New class.
* obsd-nat.c (obsd_pid_to_str, obsd_update_thread_list)
(obsd_wait): Refactor as methods of obsd_nat_target.
(obsd_add_target): Delete.
* obsd-nat.h: Include "inf-ptrace.h".
(obsd_nat_target): New class.
* ppc-fbsd-nat.c (ppc_fbsd_nat_target): New class.
(the_ppc_fbsd_nat_target): New.
(ppcfbsd_fetch_inferior_registers)
(ppcfbsd_store_inferior_registers): Refactor as methods of
ppc_fbsd_nat_target.
(_initialize_ppcfbsd_nat): Adjust to C++ification.  Use
add_target.
* ppc-nbsd-nat.c (ppc_nbsd_nat_target): New class.
(the_ppc_nbsd_nat_target): New.
(ppcnbsd_fetch_inferior_registers)
(ppcnbsd_store_inferior_registers): Refactor as methods of
ppc_nbsd_nat_target.
(_initialize_ppcnbsd_nat): Adjust to C++ification.
* ppc-obsd-nat.c (ppc_obsd_nat_target): New class.
(the_ppc_obsd_nat_target): New.
(ppcobsd_fetch_registers, ppcobsd_store_registers): Refactor as
methods of ppc_obsd_nat_target.
(_initialize_ppcobsd_nat): Adjust to C++ification.  Use
add_target.
* sh-nbsd-nat.c (sh_nbsd_nat_target): New class.
(the_sh_nbsd_nat_target): New.
(shnbsd_fetch_inferior_registers)
(shnbsd_store_inferior_registers): Refactor as methods of
sh_nbsd_nat_target.
(_initialize_shnbsd_nat): Adjust to C++ification.
* sparc-nat.c (sparc_xfer_wcookie): Make extern.
(inf_ptrace_xfer_partial): Delete.
(sparc_xfer_partial, sparc_target): Delete.
* sparc-nat.h (sparc_fetch_inferior_registers)
(sparc_store_inferior_registers, sparc_xfer_wcookie): Declare.
(sparc_target): Delete function declaration.
(sparc_target): New template class.
* sparc-nbsd-nat.c (the_sparc_nbsd_nat_target): New.
(_initialize_sparcnbsd_nat): Adjust to C++ification.
* sparc64-fbsd-nat.c (the_sparc64_fbsd_nat_target): New.
(_initialize_sparc64fbsd_nat): Adjust to C++ification.  Use
add_target.
* sparc64-nbsd-nat.c (the_sparc64_nbsd_nat_target): New.
(_initialize_sparc64nbsd_nat): Adjust to C++ification.
* sparc64-obsd-nat.c (the_sparc64_obsd_nat_target): New.
(_initialize_sparc64obsd_nat): Adjust to C++ification.  Use
add_target.
* vax-bsd-nat.c (vax_bsd_nat_target): New class.
(the_vax_bsd_nat_target): New.
(vaxbsd_fetch_inferior_registers)
(vaxbsd_store_inferior_registers): Refactor as vax_bsd_nat_target
methods.
(_initialize_vaxbsd_nat): Adjust to C++ification.

* bsd-kvm.c (bsd_kvm_target): New class.
(bsd_kvm_ops): Now a bsd_kvm_target.
(bsd_kvm_open, bsd_kvm_close, bsd_kvm_xfer_partial)
(bsd_kvm_files_info, bsd_kvm_fetch_registers)
(bsd_kvm_thread_alive, bsd_kvm_pid_to_str): Refactor as methods of
bsd_kvm_target.
(bsd_kvm_return_one): Delete.
(bsd_kvm_add_target): Adjust to C++ification.

* nto-procfs.c (nto_procfs_target, nto_procfs_target_native)
(nto_procfs_target_procfs): New classes.
(procfs_open_1, procfs_thread_alive, procfs_update_thread_list)
(procfs_files_info, procfs_pid_to_exec_file, procfs_attach)
(procfs_post_attach, procfs_wait, procfs_fetch_registers)
(procfs_xfer_partial, procfs_detach, procfs_insert_breakpoint)
(procfs_remove_breakpoint, procfs_insert_hw_breakpoint)
(procfs_remove_hw_breakpoint, procfs_resume)
(procfs_mourn_inferior, procfs_create_inferior, procfs_interrupt)
(procfs_kill_inferior, procfs_store_registers)
(procfs_pass_signals, procfs_pid_to_str, procfs_can_run): Refactor
as methods of nto_procfs_target.
(nto_procfs_ops): Now an nto_procfs_target_procfs.
(nto_native_ops): Delete.
(procfs_open, procfs_native_open): Delete.
(nto_native_ops): Now an nto_procfs_target_native.
(init_procfs_targets): Adjust to C++ification.
(procfs_can_use_hw_breakpoint, procfs_remove_hw_watchpoint)
(procfs_insert_hw_watchpoint, procfs_stopped_by_watchpoint):
Refactor as methods of nto_procfs_target.

* go32-nat.c (go32_nat_target): New class.
(the_go32_nat_target): New.
(go32_attach, go32_resume, go32_wait, go32_fetch_registers)
(go32_store_registers, go32_xfer_partial, go32_files_info)
(go32_kill_inferior, go32_create_inferior, go32_mourn_inferior)
(go32_terminal_init, go32_terminal_info, go32_terminal_inferior)
(go32_terminal_ours, go32_pass_ctrlc, go32_thread_alive)
(go32_pid_to_str): Refactor as methods of go32_nat_target.
(go32_target): Delete.
(_initialize_go32_nat): Adjust to C++ification.

* gnu-nat.c (gnu_wait, gnu_resume, gnu_kill_inferior)
(gnu_mourn_inferior, gnu_create_inferior, gnu_attach, gnu_detach)
(gnu_stop, gnu_thread_alive, gnu_xfer_partial)
(gnu_find_memory_regions, gnu_pid_to_str): Refactor as methods of
gnu_nat_target.
(gnu_target): Delete.
* gnu-nat.h (gnu_target): Delete.
(gnu_nat_target): New class.
* i386-gnu-nat.c (gnu_base_target): New.
(i386_gnu_nat_target): New class.
(the_i386_gnu_nat_target): New.
(_initialize_i386gnu_nat): Adjust to C++ification.

gdb/testsuite/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* gdb.base/breakpoint-in-ro-region.exp: Adjust to to_resume and
to_log_command renames.
* gdb.base/sss-bp-on-user-bp-2.exp: Likewise.

6 years agoEliminate target_ops::to_xclose
Pedro Alves [Wed, 2 May 2018 23:37:09 +0000 (00:37 +0100)]
Eliminate target_ops::to_xclose

In the multi-target branch, I found no need for the target_close vs
target_xclose distinction.  Heap-allocated targets simply delete
themselves in their target_close implementation, while
singleton/static targets don't.

The target_ops C++ification patches will add more commentary around
target_ops's destructor, but there's no destructor yet...

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* bfd-target.c (target_bfd_xclose): Rename to ...
(target_bfd_close): ... this.
(target_bfd_reopen): Adjust.
* target.c (target_close): Remove references to to_xclose.
* target.h (target_ops::to_xclose): Delete.
(target_ops::to_close): Update comments.

6 years agoMake inf_ptrace_trad Linux-only, move to separate file
Pedro Alves [Wed, 2 May 2018 23:37:08 +0000 (00:37 +0100)]
Make inf_ptrace_trad Linux-only, move to separate file

There are only two inf_ptrace_trad_target users, MIPS GNU/Linux and
Alpha GNU/Linux.  They both call it via linux_trad_target.

Move this code out of inf-ptrace.c to a GNU/Linux-specific new file.

Making this code be GNU/Linux-specific simplifies C++ification of
target_ops, because we can make the trad target inherit linux_nat
instead of inheriting inf_ptrace.  That'll be visible in a later patch.

Note this makes linux_target_install_ops an extern function, but that
is temporary -- the function will disappear once target_ops is made a
C++ class with virtual methods, later in the series.  Also, I did not
rename the functions in the new file for a similar reason.  They'll be
renamed again anyway in a couple of patches.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* alpha-linux-nat.c: Include "linux-nat-trad.h" instead of
"linux-nat.h".
* configure.nat (alpha-linux, linux-mips): Add linux-nat-trad.o.
* inf-ptrace.c (inf_ptrace_register_u_offset)
(inf_ptrace_fetch_register, inf_ptrace_fetch_registers)
(inf_ptrace_store_register, inf_ptrace_store_registers)
(inf_ptrace_trad_target): Move to ...
* linux-nat-trad.c: ... this new file.
* linux-nat-trad.h: New file.
* linux-nat.c (linux_target_install_ops): Make extern.
(linux_trad_target): Delete.
* linux-nat.h (linux_trad_target): Delete declaration.
(linux_target_install_ops): Declare.
* mips-linux-nat.c: Include "linux-nat-trad.h" instead of
"linux-nat.h".

6 years agoMore procfs.c simplification
Pedro Alves [Wed, 2 May 2018 23:37:07 +0000 (00:37 +0100)]
More procfs.c simplification

There are only two architectures using procfs.c (i386/SPARC), and none
of their corresponding nat files overrides any target method.  Move
the add_target calls to procfs.c directly.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* i386-sol2-nat.c (_initialize_amd64_sol2_nat): Don't call
procfs_target/add_target here.
* procfs.c (procfs_target): Make static.
(_initialize_procfs): Call add_target here.
* procfs.h (struct target_ops): Remove forward declaration.
(procfs_target): Remove declaration.
* sparc-sol2-nat.c (_initialize_sparc_sol2_nat): Delete.

6 years agoEliminate procfs.c:procfs_use_watchpoints
Pedro Alves [Wed, 2 May 2018 23:37:07 +0000 (00:37 +0100)]
Eliminate procfs.c:procfs_use_watchpoints

Now that procfs.c is only ever used by Solaris, and, both x86 and
SPARC Solaris support watchpoints (*), we don't need the separate
procfs_use_watchpoints function.  Getting rid of it simplifies
C++ification of target_ops.

(*) and I assume that any other Solaris port would use the same kernel
debug API interfaces for watchpoints.  Otherwise, we can worry about
it if it ever happens.

gdb/ChangeLog:
2018-05-02  Pedro Alves  <palves@redhat.com>

* procfs.c (procfs_stopped_by_watchpoint)
(procfs_insert_watchpoint, procfs_remove_watchpoint)
(procfs_region_ok_for_hw_watchpoint, procfs_stopped_data_address):
Forward declare.
(procfs_use_watchpoints): Delete, move contents...
(procfs_target): ... here.
* procfs.h (procfs_use_watchpoints): Delete declaration.
* i386-sol2-nat.c (_initialize_amd64_sol2_nat): Don't call
procfs_use_watchpoints.
* sparc-sol2-nat.c (_initialize_sparc_sol2_nat): Don't call
procfs_use_watchpoints.

6 years agoSet test message in py-parameter.exp
Tom Tromey [Wed, 2 May 2018 22:37:57 +0000 (16:37 -0600)]
Set test message in py-parameter.exp

Pedro pointed out that a test in py-parameter.exp had an empty
message.  This fixes it.

testsuite/ChangeLog
2018-05-02  Tom Tromey  <tom@tromey.com>

* gdb.python/py-parameter.exp: Set test message.

6 years agoHandle var_zuinteger and var_zuinteger_unlimited from Python
Tom Tromey [Thu, 26 Apr 2018 22:18:07 +0000 (16:18 -0600)]
Handle var_zuinteger and var_zuinteger_unlimited from Python

PR python/20084 points out that the Python API doesn't handle the
var_zuinteger and var_zuinteger_unlimited parameter types.

This patch adds support for these types.

Regression tested on x86-64 Fedora 26.

ChangeLog
2018-05-02  Tom Tromey  <tom@tromey.com>

PR python/20084:
* python/python.c (gdbpy_parameter_value): Handle var_zuinteger
and var_zuinteger_unlimited.
* python/py-param.c (struct parm_constant): Add PARAM_ZUINTEGER
and PARAM_ZUINTEGER_UNLIMITED.
(set_parameter_value): Handle var_zuinteger and
var_zuinteger_unlimited.
(add_setshow_generic): Likewise.
(parmpy_init): Likewise.

doc/ChangeLog
2018-05-02  Tom Tromey  <tom@tromey.com>

PR python/20084:
* python.texi (Parameters In Python): Document PARAM_ZUINTEGER and
PARAM_ZUINTEGER_UNLIMITED.

testsuite/ChangeLog
2018-05-02  Tom Tromey  <tom@tromey.com>

PR python/20084:
* gdb.python/py-parameter.exp: Add PARAM_ZUINTEGER and
PARAM_ZUINTEGER_UNLIMITED tests.

6 years agoAutomatic date update in version.in
GDB Administrator [Wed, 2 May 2018 00:00:37 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoFix unintialized memory in aarch64 opcodes.
Tamar Christina [Tue, 1 May 2018 16:11:11 +0000 (17:11 +0100)]
Fix unintialized memory in aarch64 opcodes.

This patch fixes an issue where the memory for the opcode structure is not zero'd before
the first exit branch. So there is one failure mode for which uninitialized memory
is returned.

This causes weird failures when the return code is not checked before inst is used.

opcodes/

* aarch64-dis.c (aarch64_opcode_decode): Moved memory clear code.

6 years agoUpdated Spanish translation for the gas sub-directory.
Nick Clifton [Tue, 1 May 2018 15:42:41 +0000 (16:42 +0100)]
Updated Spanish translation for the gas sub-directory.

6 years agoBring in support for the NFP target in the config.sub file.
Nick Clifton [Tue, 1 May 2018 15:35:04 +0000 (16:35 +0100)]
Bring in support for the NFP target in the config.sub file.

* config.guess: Synchronize with config project master sources.
* config.sub: Likewise.

6 years agoAdd the Netronome Flow Processor as a build target to the top-level configure.ac...
Francois H. Theron [Tue, 1 May 2018 15:23:21 +0000 (16:23 +0100)]
Add the Netronome Flow Processor as a build target to the top-level configure.ac file.

* configure.ac: Added "nfp" target.
* configure: Regenerate.

6 years agorust: Fix null deref when casting (PR 23124)
Dan Robertson [Sat, 28 Apr 2018 03:18:00 +0000 (03:18 +0000)]
rust: Fix null deref when casting (PR 23124)

Fix a null dereference when casting a value to a unit type.

ChangeLog
2018-04-28  Dan Robertson  <danlrobertson89@gmail.com>

PR rust/23124
* gdb/rust-exp.y (convert_params_to_types): Ensure that the params
pointer is not null before dereferencing it.

testsuite/ChangeLog
2018-04-28  Dan Robertson  <danlrobertson89@gmail.com>

PR rust/23124
* gdb.rust/expr.exp: Test that the unit type is correctly parsed
when casting.

6 years agoAutomatic date update in version.in
GDB Administrator [Tue, 1 May 2018 00:00:37 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years ago[Ada/ravenscar] error during "continue" after task/thread switch
Joel Brobecker [Mon, 30 Apr 2018 22:05:42 +0000 (17:05 -0500)]
[Ada/ravenscar] error during "continue" after task/thread switch

When debugging a program using the Ada ravenscar profile, resuming
a program's execution after having switched to a different task
sometimes yields the following error:

     (gdb) cont
     Continuing.
     Cannot execute this command while the target is running.
     Use the "interrupt" command to stop the target
     and then try again.

In short, the Ravenscar profile is a standardized subset of Ada which
allows tasking (often mapped to threads). We often use it on baremetal
targets where there is no OS support. Thread support is implemented
as a thread target_ops layer. It sits on top of the "remote" layer,
so we can do thread debugging against baremetal targets to which GDB
is connected via "target remote".

What happens, when the user request the program to resume execution,
is the following:

  - the ravenscar-thread target_ops layer gets the order to resume
    the program's execution. The current thread is not the active
    thread in the inferior, and the "remote" layer doesn't know
    about that thread anyway. So what we do is (see ravenscar_resume):

       + switch inferior_ptid to the ptid of the actually active thread;
       + ask the layer beneath us to actually do the resume.

  - Once that's done, the resuming itself is done. But execute_command
    (in top.c) actually does a bit more. More precisely, it unconditionally
    checks to see if the language may no longer be matching the current
    frame:

        check_frame_language_change ();

The problem, here, is that we haven't received the "stop" event
from the inferior, yet. This part will be handled by the event loop,
which is done later. So, checking for the language-change here
doesn't make sense, since we don't really have a frame. In our
case, the error comes from the fact that we end up trying to read
the registers, which causes the error while the remote protocol
is waiting for the event showing the inferior stopped.

This apparently used to work, but it is believed that this was only
accidental. In other words, we had enough information already cached
within GDB that we were able to perform the entire call to
check_frame_language_change without actually querying the target.
On PowerPC targets, this started to fail as a side-effect of a minor
change in the way we get to the regcache during the handling of
software-single-step (which seems fine).

This patch fixes the issue by only calling check_frame_language_change
in cases the inferior isn't running. Otherwise, it skips it, knowing
that the event loop should eventually get to it.

gdb/ChangeLog:

        * top.c (execute_command): Do not call check_frame_language_change
        if the inferior is running.

Tested on x86_64-linux, no regression. Also tested on aarch64-elf,
arm-elf, leon3-elf, and ppc-elf, but using AdaCore's testsuite.

6 years agoRemove a use of is_mi_like_p from darwin-nat-info.c
Tom Tromey [Mon, 30 Apr 2018 02:53:38 +0000 (20:53 -0600)]
Remove a use of is_mi_like_p from darwin-nat-info.c

This removes a use of is_mi_like_p from darwin-nat-info.c.
This is not needed because MI already ignores ui_out::text.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* darwin-nat-info.c (darwin_debug_regions_recurse): Remove use of
is_mi_like_p.

6 years agoRemove some is_mi_like_p from breakpoint code
Tom Tromey [Mon, 30 Apr 2018 02:52:26 +0000 (20:52 -0600)]
Remove some is_mi_like_p from breakpoint code

This removes some uses of is_mi_like_p from the breakpoint code.  The
break-catch-throw.c change brings it into line with what other
breakpoint classes do.  The other changes simply replace printf calls
with ui_out::text or ui_out::message calls.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* breakpoint.c (mention): Remove use of is_mi_like_p.
(print_mention_ranged_breakpoint): Likewise.
* break-catch-throw.c (print_it_exception_catchpoint): Remove use
of is_mi_like_p.

6 years agoRemove a use of is_mi_like_p from tracepoint.c
Tom Tromey [Mon, 30 Apr 2018 02:50:43 +0000 (20:50 -0600)]
Remove a use of is_mi_like_p from tracepoint.c

This removes a use of is_mi_like_p and changes a printf_filtered into
a call to ui_out::text.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* tracepoint.c (tvariables_info_1): Remove use of is_mi_like_p.

6 years agoRemove some uses of is_mi_like_p from spu-tdep.c
Tom Tromey [Sun, 29 Apr 2018 04:39:08 +0000 (22:39 -0600)]
Remove some uses of is_mi_like_p from spu-tdep.c

There were a few spots in spu-tdep.c where a use of is_mi_like_p was
not needed.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* spu-tdep.c (info_spu_mailbox_list, info_spu_dma_cmdlist)
(info_spu_event_command): Remove some uses of is_mi_like_p.

6 years agoRemove some uses of is_mi_like_p from py-framefilter.c
Tom Tromey [Sun, 29 Apr 2018 04:30:08 +0000 (22:30 -0600)]
Remove some uses of is_mi_like_p from py-framefilter.c

Some uses of is_mi_like_p in py-framefilter.c were not needed.  In
general a call to ui_out::text, ui_out::message, or ui_out::spaces
does not need to be guarded -- these are already ignored by MI.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* python/py-framefilter.c (py_print_single_arg)
(enumerate_locals, py_print_args, py_print_frame): Remove some
uses of is_mi_like_p.

6 years agoMake do_is_mi_like_p const.
Tom Tromey [Tue, 24 Apr 2018 02:52:36 +0000 (20:52 -0600)]
Make do_is_mi_like_p const.

This changes ui_out to make is_mi_like_p and do_is_mi_like_p "const".

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* ui-out.c: Update.
* cli-out.h (cli_ui_out::do_is_mi_like_p): Update.
* ui-out.h (ui_out::is_mi_like_p): Now const.
(ui_out::do_is_mi_like_p): Now const.
* mi/mi-out.h (mi_ui_out::do_is_mi_like_p): Update.

6 years agoChange Python code to use new_reference
Tom Tromey [Mon, 30 Apr 2018 03:09:02 +0000 (21:09 -0600)]
Change Python code to use new_reference

This changes a few spots in the Python code to use new_reference
rather than the manual incref+constructor that was previously being
done.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* varobj.c (varobj_set_visualizer): Use new_reference.
* python/python.c (gdbpy_decode_line): Use new_reference.
* python/py-cmd.c (cmdpy_function, cmdpy_completer_helper): Use
new_reference.

6 years agoUse new_reference for struct value
Tom Tromey [Mon, 30 Apr 2018 03:07:03 +0000 (21:07 -0600)]
Use new_reference for struct value

value_incref returned its argument just as a convenience, which in the
end turned out to only be used in precisely the cases where
new_reference helps.  So, this patch changes value_incref to return
void and changes some value-using code to use new_reference.

I also noticed that the comments for value_incref and value_decref
were swapped, so this patch fixes those.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* varobj.c (install_new_value): Use new_reference.
* value.h (value_incref): Return void.  Swap intro comment with
value_decref.
* value.c (set_value_parent): Use new_reference.
(value_incref): Return void.  Update intro comment.
(release_value): Use new_reference.
* dwarf2loc.c (dwarf2_evaluate_loc_desc_full): Use new_reference.

6 years agoRemove new_bfd_ref
Tom Tromey [Mon, 30 Apr 2018 03:02:19 +0000 (21:02 -0600)]
Remove new_bfd_ref

For gdb_bfd_ref_ptr, gdb already had a convenience function like the
new gdb_ref_ptr::new_reference -- called new_bfd_ref.  This patch
removes it in favor of the new common function.

While doing this I also noticed that the comment for gdb_bfd_open was
incorrect (in a way related to reference counting), so this patch
updates the comment as well.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* symfile-mem.c (symbol_file_add_from_memory): Use new_reference.
* gdb_bfd.h (new_bfd_ref): Remove.
(gdb_bfd_open): Update comment.
* gdb_bfd.c (gdb_bfd_open, gdb_bfd_fopen, gdb_bfd_openr)
(gdb_bfd_openw, gdb_bfd_openr_iovec, gdb_bfd_record_inclusion)
(gdb_bfd_fdopenr): Use new_reference.
* exec.c (exec_file_attach): Use new_reference.

6 years agoIntroduce ref_ptr::new_reference
Tom Tromey [Mon, 30 Apr 2018 02:59:21 +0000 (20:59 -0600)]
Introduce ref_ptr::new_reference

I noticed a common pattern with gdb::ref_ptr, where callers would
"incref" and then create a new wrapper object, like:

    Py_INCREF (obj);
    gdbpy_ref<> ref (obj);

The ref_ptr constructor intentionally does not acquire a new
reference, but it seemed to me that it would be reasonable to add a
static member function that does so.

In this patch I chose to call the function "new_reference".  I
considered "acquire_reference" as well, but "new" seemed less
ambiguous than "acquire" to me.

ChangeLog
2018-04-30  Tom Tromey  <tom@tromey.com>

* common/gdb_ref_ptr.h (ref_ptr::new_reference): New static
method.

6 years agoRemove long_long_align_bit gdbarch attribute
Tom Tromey [Mon, 23 Apr 2018 19:41:27 +0000 (13:41 -0600)]
Remove long_long_align_bit gdbarch attribute

This removes the long_long_align_bit gdbarch attribute in favor of
type_align.  This uncovered two possible issues.

First, arc-tdep.c claimed that long long alignment was 32 bits, but as
discussed on the list, ARC has a maximum alignment of 32 bits, so I've
added an arc_type_align function to account for this.

Second, jit.c, the sole user of long_long_align_bit, was confusing
"long long" with uint64_t.  The relevant structure is defined in the
JIT API part of the manual as:

     struct jit_code_entry
     {
       struct jit_code_entry *next_entry;
       struct jit_code_entry *prev_entry;
       const char *symfile_addr;
       uint64_t symfile_size;
     };

I've changed this code to use uint64_t.

2018-04-30  Tom Tromey  <tom@tromey.com>

* jit.c (jit_read_code_entry): Use type_align.
* i386-tdep.c (i386_gdbarch_init): Don't call
set_gdbarch_long_long_align_bit.
* gdbarch.sh: Remove long_long_align_bit.
* gdbarch.c, gdbarch.h: Rebuild.
* arc-tdep.c (arc_type_align): New function.
(arc_gdbarch_init): Use arc_type_align.  Don't call
set_gdbarch_long_long_align_bit.

6 years agoRemove rust_type_alignment
Tom Tromey [Mon, 23 Apr 2018 19:23:02 +0000 (13:23 -0600)]
Remove rust_type_alignment

rust_type_alignment is not needed now that gdb has type alignment
code.  So, this removes it.

2018-04-30  Tom Tromey  <tom@tromey.com>

* rust-lang.c (rust_type_alignment): Remove.
(rust_composite_type): Use type_align.

6 years agoExpose type alignment on gdb.Type
Tom Tromey [Fri, 20 Apr 2018 21:43:56 +0000 (15:43 -0600)]
Expose type alignment on gdb.Type

This adds an "alignof" attribute to gdb.Type in the Python API.

2018-04-30  Tom Tromey  <tom@tromey.com>

* NEWS: Mention Type.align.
* python/py-type.c (typy_get_alignof): New function.
(type_object_getset): Add "alignof".

2018-04-30  Tom Tromey  <tom@tromey.com>

* python.texi (Types In Python): Document Type.align.

2018-04-30  Tom Tromey  <tom@tromey.com>

* gdb.python/py-type.exp: Check align attribute.
* gdb.python/py-type.c: New "aligncheck" global.

6 years agoHandle alignof and _Alignof
Tom Tromey [Fri, 20 Apr 2018 19:40:29 +0000 (13:40 -0600)]
Handle alignof and _Alignof

This adds alignof and _Alignof to the C/C++ expression parser, and
adds new tests to test the features.  The tests are written to try to
ensure that gdb's knowledge of alignment rules stays in sync with the
compiler's.

2018-04-30  Tom Tromey  <tom@tromey.com>

PR exp/17095:
* NEWS: Update.
* std-operator.def (UNOP_ALIGNOF): New operator.
* expprint.c (dump_subexp_body_standard) <case UNOP_ALIGNOF>:
New.
* eval.c (evaluate_subexp_standard) <case UNOP_ALIGNOF>: New.
* c-lang.c (c_op_print_tab): Add alignof.
* c-exp.y (ALIGNOF): New token.
(exp): Add "ALIGNOF" production.
(ident_tokens): Add _Alignof and alignof.

2018-04-30  Tom Tromey  <tom@tromey.com>

PR exp/17095:
* gdb.dwarf2/dw2-align.exp: New file.
* gdb.cp/align.exp: New file.
* gdb.base/align.exp: New file.
* lib/gdb.exp (gdb_int128_helper): New proc.
(has_int128_c, has_int128_cxx): New caching procs.

6 years agoAdd initial type alignment support
Tom Tromey [Fri, 20 Apr 2018 17:50:09 +0000 (11:50 -0600)]
Add initial type alignment support

This adds some basic type alignment support to gdb.  It changes struct
type to store the alignment, and updates dwarf2read.c to handle
DW_AT_alignment.  It also adds a new gdbarch method and updates
i386-tdep.c.

None of this new functionality is used anywhere yet, so tests will
wait until the next patch.

2018-04-30  Tom Tromey  <tom@tromey.com>

* i386-tdep.c (i386_type_align): New function.
(i386_gdbarch_init): Update.
* gdbarch.sh (type_align): New method.
* gdbarch.c, gdbarch.h: Rebuild.
* arch-utils.h (default_type_align): Declare.
* arch-utils.c (default_type_align): New function.
* gdbtypes.h (TYPE_ALIGN_BITS): New define.
(struct type) <align_log2>: New field.
<instance_flags>: Now a bitfield.
(TYPE_RAW_ALIGN): New macro.
(type_align, type_raw_align, set_type_align): Declare.
* gdbtypes.c (type_align, type_raw_align, set_type_align): New
functions.
* dwarf2read.c (quirk_rust_enum): Set type alignment.
(get_alignment, maybe_set_alignment): New functions.
(read_structure_type, read_enumeration_type, read_array_type)
(read_set_type, read_tag_pointer_type, read_tag_reference_type)
(read_subrange_type, read_base_type): Set type alignment.

6 years agoThis patch adds support to objdump for disassembly of NFP (Netronome Flow Processor...
Francois H. Theron [Mon, 30 Apr 2018 16:02:59 +0000 (17:02 +0100)]
This patch adds support to objdump for disassembly of NFP (Netronome Flow Processor) ELF files (.nffw) as well as some basic readelf support.

bfd * Makefile.am: Added NFP files to build.
* archures.c: Added bfd_arch_nfp
* config.bfd: Added NFP support.
* configure.ac: Added NFP support.
* cpu-nfp.c: New, for NFP support.
* elf-bfd.h: Added elf_section_info()
* elf64-nfp.c: New, for NFP support.
* po/SRC-POTFILES.in: Added NFP source files.
* targets.c: Added nfp_elf64_vec
* bfd-in2.h: Regenerate.
* Makefile.in: Regenerate.
* configure: Regenerate.

binutils* readelf.c: Very basic support for EM_NFP and its section types.
* testsuite/binutils-all/nfp: New directory.
* testsuite/binutils-all/nfp/objdump.exp: New file.  Run new
tests.
* testsuite/binutils-all/nfp/test2_ctx8.d: New file.
* testsuite/binutils-all/nfp/test2_no-pc_ctx4.d: New file.
* testsuite/binutils-all/nfp/test1.d: New file.
* testsuite/binutils-all/nfp/nfp6000.nffw: New file.
* testsuite/binutils-all/nfp/test2_nfp6000.nffw: New file.
* NEWS: Mention the new support.

include * dis-asm.h: Added print_nfp_disassembler_options prototype.
* elf/common.h: Added EM_NFP, officially assigned. See Google Group
Generic System V Application Binary Interface.
* elf/nfp.h: New, for NFP support.
* opcode/nfp.h: New, for NFP support.

opcodes Makefile.am: Added nfp-dis.c.
configure.ac: Added bfd_nfp_arch.
disassemble.h: Added print_insn_nfp prototype.
disassemble.c: Added ARCH_nfp and call to print_insn_nfp
nfp-dis.c: New, for NFP support.
po/POTFILES.in: Added nfp-dis.c to the list.
Makefile.in: Regenerate.
configure: Regenerate.

6 years agoUse bool in read_index_from_section
Simon Marchi [Mon, 30 Apr 2018 15:06:57 +0000 (11:06 -0400)]
Use bool in read_index_from_section

gdb/ChangeLog:

* dwarf2read.c (read_index_from_section): Use bool.

6 years agoAutomatic date update in version.in
GDB Administrator [Mon, 30 Apr 2018 00:00:32 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoproc-events.c: fix compilation on Solaris
Fabian Groffen [Sun, 29 Apr 2018 16:04:54 +0000 (12:04 -0400)]
proc-events.c: fix compilation on Solaris

This patch adds a guard around the usage of SYS_uuidsys, which is
not available on (at least) Solaris 10 and OpenIndiana.

gdb/ChangeLog:

PR gdb/22950
* proc-events.c (init_syscall_table): Guard usage os SYS_uuidsys
with #ifdef.

6 years agoFix race when building ada-lex.c
John Reiser [Sun, 29 Apr 2018 15:57:38 +0000 (11:57 -0400)]
Fix race when building ada-lex.c

Prevent a race when building ada-lex.c, and any target of rules .c:.l or
.c:.y.  The target should be written only at the last step, else SIGINT
(^C) can leave an inconsistent state.  Being .PRECIOUS makes it even
worse.

gdb/ChangeLog:

PR build/22873
* gdb/Makefile.in: (.c:.l, .c:.y): Write the target only in the
last step, and do it atomically.

6 years agoAutomatic date update in version.in
GDB Administrator [Sun, 29 Apr 2018 00:01:17 +0000 (00:01 +0000)]
Automatic date update in version.in

6 years agoAutomatic date update in version.in
GDB Administrator [Sat, 28 Apr 2018 00:00:28 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoAdd libcc1 v1 compatibility to C compile feature
Alexandre Oliva [Fri, 27 Apr 2018 18:17:02 +0000 (11:17 -0700)]
Add libcc1 v1 compatibility to C compile feature

This patch adds v1 compatibiltiy to the C compile feature.  The only change
in v1 concerns the handling of integer types, which permits GDB to specify
the built-in name for the type.

As far as I know, the C frontend is still on v0, so this patch is purely
precautionary. [By default C++ compile uses the equivalent of the C
frontend's int_type and float_type (aka the "v1" versions).]

gdb/ChangeLog:

* compile/compile-c-types.c (convert_int, convert_float):
Update for C FE v1.

6 years agoAdd inclusive range support for Rust
Tom Tromey [Thu, 29 Mar 2018 20:14:07 +0000 (14:14 -0600)]
Add inclusive range support for Rust

This is version 2 of the patch to add inclusive range support for
Rust.  I believe it addresses all review comments.

Rust recently stabilized the inclusive range feature:

    https://github.com/rust-lang/rust/issues/28237

An inclusive range is an expression like "..= EXPR" or "EXPR ..=
EXPR".  It is like an ordinary range, except the upper bound is
inclusive, not exclusive.

This patch adds support for this feature to gdb.

Regression tested on x86-64 Fedora 27.

2018-04-27  Tom Tromey  <tom@tromey.com>

PR rust/22545:
* rust-lang.c (rust_inclusive_range_type_p): New function.
(rust_range): Handle inclusive ranges.
(rust_compute_range): Likewise.
* rust-exp.y (struct rust_op) <inclusive>: New field.
(DOTDOTEQ): New constant.
(range_expr): Add "..=" productions.
(operator_tokens): Add "..=" token.
(ast_range): Add "inclusive" parameter.
(convert_ast_to_expression) <case OP_RANGE>: Handle inclusive
ranges.
* parse.c (operator_length_standard) <case OP_RANGE>: Handle new
bounds values.
* expression.h (enum range_type) <NONE_BOUND_DEFAULT_EXCLUSIVE,
LOW_BOUND_DEFAULT_EXCLUSIVE>: New constants.
Update comments.
* expprint.c (print_subexp_standard): Handle new bounds values.
(dump_subexp_body_standard): Likewise.

2018-04-27  Tom Tromey  <tom@tromey.com>

PR rust/22545:
* gdb.rust/simple.exp: Add inclusive range tests.

6 years agoEnable -Wsuggest-override
Tom Tromey [Fri, 27 Apr 2018 04:46:54 +0000 (22:46 -0600)]
Enable -Wsuggest-override

I noticed the existence of -Wsuggest-override and so this patch
enables it for gdb.  It found a few spots that could use "override".
Also I went ahead and removed all uses of the "OVERRIDE" macro.

Using override is beneficial because it makes it harder to change a
base class and then forget to change a derived class.

Tested by the buildbot.

ChangeLog
2018-04-27  Tom Tromey  <tom@tromey.com>

* configure: Rebuild.
* warning.m4 (AM_GDB_WARNINGS): Add -Wsuggest-override.
* dwarf2loc.c (class dwarf_evaluate_loc_desc): Use "override", not
"OVERRIDE".
(class symbol_needs_eval_context): Likewise.
* dwarf2read.c (mock_mapped_index::symbol_name_count)
(mock_mapped_index::symbol_name_at): Use "override".  Remove
"virtual".
* dwarf2-frame.c (dwarf_expr_executor::get_addr_index): Use
"override".
(class dwarf_expr_executor): Use "override", not "OVERRIDE".
* aarch64-tdep.c (instruction_reader::read): Use "override".
(instruction_reader_test::read): Likewise.
* arm-tdep.c (instruction_reader::read): Use "override".
(instruction_reader_thumb::read): Likewise.

gdbserver/ChangeLog
2018-04-27  Tom Tromey  <tom@tromey.com>

* configure: Rebuild.

6 years agoMIPS/LD/testsuite: Update `run_dump_test' cases for non-DSO targets
Maciej W. Rozycki [Fri, 27 Apr 2018 14:25:20 +0000 (15:25 +0100)]
MIPS/LD/testsuite: Update `run_dump_test' cases for non-DSO targets

Mark these `run_dump_test' cases across `ld-mips-elf/mips-elf.exp' that
are run unconditionally and require shared library support for exclusion
for targets that do not have such support, removing these failures:

FAIL: MIPS BAL/JALX in PIC mode
FAIL: microMIPS BAL/JALX in PIC mode
FAIL: MIPS BAL/JALX in PIC mode (ignore branch ISA)
FAIL: microMIPS BAL/JALX in PIC mode (ignore branch ISA)
FAIL: ld-mips-elf/hash1a
FAIL: ld-mips-elf/hash1b
FAIL: ld-mips-elf/hash1c

with `mipsel-ps2-elf' and `mips64el-ps2-elf' targets.  Tests that are
guarded with `linux_gnu' will have to be reviewed separately.

ld/
* testsuite/ld-mips-elf/bal-jalx-pic.d: Only run for
`check_shared_lib_support' targets.
* testsuite/ld-mips-elf/bal-jalx-pic-n32.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-n64.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-micromips.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-micromips-n32.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-micromips-n64.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore-n32.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore-n64.d: Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore-micromips.d:
Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore-micromips-n32.d:
Likewise.
* testsuite/ld-mips-elf/bal-jalx-pic-ignore-micromips-n64.d:
Likewise.
* testsuite/ld-mips-elf/hash1a.d: Likewise.
* testsuite/ld-mips-elf/hash1b.d: Likewise.
* testsuite/ld-mips-elf/hash1c.d: Likewise.
* testsuite/ld-mips-elf/relax-jalr-n32-shared.d: Likewise.
* testsuite/ld-mips-elf/relax-jalr-n64-shared.d: Likewise.

6 years agotestsuite: Support filtering targets by TCL procedure in `run_dump_test'
Maciej W. Rozycki [Fri, 27 Apr 2018 14:25:20 +0000 (15:25 +0100)]
testsuite: Support filtering targets by TCL procedure in `run_dump_test'

Implement a more complex way of selecting targets to include or exclude
with `run_dump_test' cases, by extending the syntax for the `target',
`not-target', `skip' and `not-skip' options (with the binutils and GAS
test suites) and the `target', `alltargets' and `notarget' options (with
the LD test suite) to also accept a name of a TCL procedure instead of a
target triplet glob matching expression.  The result, 1 or 0, of the
procedure determines whether the test is to be run or not.  This mimics
and expands `dg-require-effective-target' from the GCC test suite.

Names of TCL procedures are supplied in square brackets `[]' as with TCL
procedure calls, observing that target triplet glob matching expressions
do not normally start and end with matching square brackets both at a
time.  Arguments for procedures are allowed if required.

Having a way to specify a complex condition for a `run_dump_test' case
to run has the advantage of keeping it local within the test case itself
where tool options related to the check might be also present, removing
the need to wrap `run_dump_test' calls into an `if' block whose only
reason is to do a feature check, and ultimately lets one have the test
reported as UNSUPPORTED automagically if required (not currently
supported by the `run_dump_test' options used for LD).

binutils/
* testsuite/lib/binutils-common.exp (match_target): New procedure.
* testsuite/lib/utils-lib.exp (run_dump_test): Use it in place
of `istarget' for matching with `target', `not-target', `skip'
and `not-skip' options.

gas/
* testsuite/lib/gas-defs.exp (run_dump_test): Use `match_target'
in place of `istarget' for matching with `target', `not-target',
`skip' and `not-skip' options.

ld/
* testsuite/lib/ld-lib.exp (run_dump_test): Use `match_target'
in place of `istarget' for matching with `target', `alltargets'
and `notarget' options.

6 years agoRevert "Enable Intel MOVDIRI, MOVDIR64B instructions."
Igor Tsimbalist [Fri, 27 Apr 2018 12:34:13 +0000 (14:34 +0200)]
Revert "Enable Intel MOVDIRI, MOVDIR64B instructions."

This reverts commit a914a7c95895161c99533d5919b8504b37ea54a0.

6 years agoRegenerate some files for recent ARM patches
Alan Modra [Fri, 27 Apr 2018 09:54:13 +0000 (19:24 +0930)]
Regenerate some files for recent ARM patches

bfd/
* bfd-in2.h: Regenerate.
* libbfd.h: Regenerate.
ld/
* po/BLD-POTFILES.in: Regenerate.

6 years agoPR23123, PowerPC32 ifunc regression
Alan Modra [Fri, 27 Apr 2018 06:16:40 +0000 (15:46 +0930)]
PR23123, PowerPC32 ifunc regression

Two of the gcc ifunc tests fail for ppc32, due to my pr22374 fix being
a little too enthusiastic in trimming PLT entries.  ppc64 doesn't have
the same failures because ppc64_elf_check_relocs happens to set
needs_plt for any ifunc reloc.

PR 23123
PR 22374
* elf32-ppc.c (ppc_elf_adjust_dynamic_symbol): Don't drop plt
relocs for ifuncs.
* elf64-ppc.c (ppc64_elf_adjust_dynamic_symbol): Comment fixes.

6 years agoAutomatic date update in version.in
GDB Administrator [Fri, 27 Apr 2018 00:00:36 +0000 (00:00 +0000)]
Automatic date update in version.in

6 years agoFix remote 'g' command error handling (PR remote/9665)
Andrzej Kaczmarek [Thu, 26 Apr 2018 22:47:25 +0000 (23:47 +0100)]
Fix remote 'g' command error handling (PR remote/9665)

'g' command returns hex-string as response so simply checking for 'E'
to determine if it failed is not enough and can trigger spurious error
messages.  For example, invalid behaviour can be easily triggered on
Cortex-M as follows:

  (gdb) set $r0 = 0xe0
  Sending packet: $P0=e0000000#72...Packet received: OK
  Packet P (set-register) is supported
  Sending packet: $g#67...Packet received: E0000000849A0020...
  Remote failure reply: E0000000849A0020...

This patch fixes the problem by calling putpkt()/getpkt() directly and
checking result with packet_check_result().  This works fine since Enn
response has odd number of bytes while proper response has even number
of bytes.

Also, remote_send() is now not used anywhere so it can be removed.

gdb/Changelog:
2018-04-26  Andrzej Kaczmarek  <andrzej.kaczmarek@codecoup.pl>

PR remote/9665
* remote.c (send_g_packet): Use putpkt/getpkt/packet_check_result
instead of remote_send.
(remote_send): Remove.

6 years agoEnable Intel MOVDIRI, MOVDIR64B instructions.
Igor Tsimbalist [Wed, 25 Apr 2018 15:02:06 +0000 (17:02 +0200)]
Enable Intel MOVDIRI, MOVDIR64B instructions.

gas/
* config/tc-i386.c (cpu_arch): Add .movdir, .movdir64b.
(cpu_noarch): Likewise.
(process_suffix): Add check for register size.
* doc/c-i386.texi: Document movdiri, movdir64b.
* testsuite/gas/i386/i386.exp: Run MOVDIR{I,64B} tests.
* testsuite/gas/i386/movdir-intel.d: New test.
* testsuite/gas/i386/movdir.d: Likewise.
* testsuite/gas/i386/movdir.s: Likewise.
* testsuite/gas/i386/movdir64b-reg.s: Likewise.
* testsuite/gas/i386/movdir64b-reg.l: Likewise.
* testsuite/gas/i386/x86-64-movdir-intel.d: Likewise.
* testsuite/gas/i386/x86-64-movdir.d: Likewise.
* testsuite/gas/i386/x86-64-movdir.s: Likewise.
* testsuite/gas/i386/x86-64-movdir64b-reg.s: Likewise.
* testsuite/gas/i386/x86-64-movdir64b-reg.l: Likewise.

opcodes/
* i386-dis.c (enum): Add PREFIX_0F38F8, PREFIX_0F38F9.
(prefix_table): New instructions (see prefix above).
Add Gva macro and handling in OP_G.
* i386-gen.c (cpu_flag_init): Add CPU_MOVDIRI_FLAGS,
CPU_MOVDIR64B_FLAGS.
(cpu_flags): Likewise.
(opcode_modifiers): Add AddrPrefixOpReg.
(i386_opcode_modifier): Likewise.
* i386-opc.h (enum): Add CpuMOVDIRI, CpuMOVDIR64B.
(i386_cpu_flags): Likewise.
* i386-opc.tbl: Add movidir{i,64b}.
* i386-init.h: Regenerate.
* i386-tbl.h: Likewise.

6 years agoExtend the assembler so that it can automatically generate GNU Build attribute notes...
Nick Clifton [Thu, 26 Apr 2018 14:12:42 +0000 (15:12 +0100)]
Extend the assembler so that it can automatically generate GNU Build attribute notes if none are present in the input files.

gas * as.c (flag_generate_build_notes): New variable.
(show_usage): Add entry for --generate-missing-build-notes.
(parse_args): Parse --generate-missing-build-notes.
* as.h: Export flag_generate_build_notes.
* symbols.c (save_symbol_name): Ensure that the name parameter is
not NULL.
* write.c (create_obj_attrs_section): Reformat.
(create_note_reloc): New function - creates a relocation for a
field in a GNU Build attribute note.
(maybe_generate_build_notes): New function - created GNU Build
attribute notes if none are present in the output file.
(write_object_file): Call maybe_generate_build_notes.
* configure.ac (--enable-generate-build-notes): New option.
* NEWS: Announce the new feature.
* doc/as.textinfo: Document the new option.
* config.in: Regenerate.
* configure: Regenerate.

binutils* readelf.c (is_32bit_abs_reloc): Support R_PARISC_DIR32 as a
32-bit absolute reloc for the HPPA target.
* testsuite/binutils-all/note-5.d: New test.
* testsuite/binutils-all/note-5.s: Source file for new test.
* testsuite/binutils-all/objcopy.exp: Run new test.

6 years ago[ld/testsuite] Fix pr2404 output.
Christophe Lyon [Thu, 26 Apr 2018 13:44:33 +0000 (13:44 +0000)]
[ld/testsuite] Fix pr2404 output.

2018-04-26  Christophe Lyon  <christophe.lyon@linaro.org>

* testsuite/ld-elf/pr2404b.c (main): Adjust printf to account for
new variable name.
* testsuite/ld-elf/pr2404.out: Adjust expected output accordingly.

6 years agoFix resolving GNU ifunc bp locations when inferior runs resolver
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
Fix resolving GNU ifunc bp locations when inferior runs resolver

I noticed that if you set a breakpoint on an ifunc before the ifunc is
resolved, and then let the program call the ifunc, thus resolving it,
GDB end up with a location for that original breakpoint that is
pointing to the ifunc target, but it is left pointing to the first
address of the function, instead of after its prologue.  After
prologue is what you get if you create a new breakpoint at that point.

1) With no debug info for the target function:

  1.a) Set before resolving, and then program continued passed resolving:

    Num     Type           Disp Enb Address            What
    1       breakpoint     keep y   0x0000000000400753 <final>

  1.b) Breakpoint set after inferior resolved ifunc:

    Num     Type           Disp Enb Address            What
    2       breakpoint     keep y   0x0000000000400757 <final+4>

2) With debug info for the target function:

   1.a) Set before resolving, and then program continued passed resolving:

     Num     Type           Disp Enb Address            What
     1       breakpoint     keep y   0x0000000000400753 in final at gdb/testsuite/gdb.base/gnu-ifunc-final.c:20

   1.b) Breakpoint set after inferior resolved ifunc:

     Num     Type           Disp Enb Address            What
     2       breakpoint     keep y   0x000000000040075a in final at gdb/testsuite/gdb.base/gnu-ifunc-final.c:21

The problem is that elf_gnu_ifunc_resolver_return_stop (called by the
internal breakpoint that traps the resolver returning) does not agree
with linespec.c:minsym_found.  It does not skip to the function's
start line (i.e., past the prologue).  We can now use the
find_function_start_sal overload added by the previous commmit to fix
this.

New tests included, which fail before the patch, and pass afterwards.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* elfread.c (elf_gnu_ifunc_resolver_return_stop): Use
find_function_start_sal instead of find_pc_line.

gdb/testsuite/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* gdb.base/gnu-ifunc.exp (set-break): Test that GDB resolves
ifunc breakpoint locations correctly of ifunc breakpoints set
while the program resolves the ifunc.

6 years agoExtend GNU ifunc testcases
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
Extend GNU ifunc testcases

This patch extends/rewrites the gdb.base/gnu-ifunc.exp testcase to
cover the many different fixes in earlier patches.  (This was actually
what encovered most of the problems.)

The current testcase uses an ifunc symbol with the same name as the
ifunc resolver symbol and makes sure to compile the ifunc resolver
without debug info.  That does not model how ifuncs are implemented in
gcc/ifunc nowadays.  Instead, what we have is that the glibc ifunc
resolvers nowadays are written in C and end up with debug info.

Also, in some cases the ifunc target is written in assembly, but in
other cases it's written in C.  In the case of target function written
in C, if the target function has debug info, when we set a break on
the ifunc, we want to set it past the prologue of the target function.
Currently GDB gets that wrong.

To make sure we cover all the different scenarios, the testcase is
tweaked to cover all the different combinations of

 - An ifunc resolver with the same name as the user-visible symbol vs
   an ifunc resolver with a different name as the user-visible symbol.

 - ifunc resolver compiled with and without debug info.

 - ifunc target function compiled with and without debug info.

The testcase currently sets breakpoints on ifuncs, calls ifunc
functions, steps into ifunc functions, etc.  After this series, this
all works and the testcase passes cleanly.

While working on this, I noticed that "b gnu_ifunc" before and after
the inferior resolved the ifunc would end up with a breakpoint with
different locations.  That's now covered by new tests inside the new
"set-break" procedure.

It also tests other things like making sure we can't call an ifunc
without a return-type case if we don't know the type of the target.
And making sure that we pass enough arguments when we do know the
type.

gdb/testsuite/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* gdb.base/gnu-ifunc-final.c: New file.
* gdb.base/gnu-ifunc.c (final): Delete, moved to gnu-ifunc-final.c.
* gdb.base/gnu-ifunc.exp (executable): Delete.
(staticexecutable): Adjust.
(lib_opts, exec_opts): Delete.
(make_binsuffix, build, set-break): New procedures.
(misc_tests): New, with tests factored out from the top level.
(top level): Test different combinations of ifunc resolver name,
resolver with and with debug info, and ifunc target with and
without debug info.  Wrap static tests with with_target_prefix.

6 years agoPPC64: always make synthetic .text symbols for GNU ifunc symbols
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
PPC64: always make synthetic .text symbols for GNU ifunc symbols

If you create an ifunc using GCC's __attribute__ ifunc, like:

 extern int gnu_ifunc (int arg);
 static int gnu_ifunc_target (int arg) { return 0; }
 __typeof (gnu_ifunc) *gnu_ifunc_resolver (unsigned long hwcap) { return gnu_ifunc_target; }
 __typeof (gnu_ifunc) gnu_ifunc __attribute__ ((ifunc ("gnu_ifunc_resolver")));

then you end up with two (function descriptor) symbols, one for the
ifunc itself, and another for the resolver:

   (...)
   12: 0000000000020060    104 FUNC    GLOBAL DEFAULT       18 gnu_ifunc_resolver
   (...)
   16: 0000000000020060    104 GNU_IFUNC GLOBAL DEFAULT       18 gnu_ifunc
   (...)

Both ifunc and resolver symbols have the same address/value, so
ppc64_elf_get_synthetic_symtab only creates a synthetic text symbol
for one of them.  In the case above, it ends up being created for the
resolver, only:

  (gdb) maint print msymbols
  (...)
  [ 7] t 0x980 .frame_dummy section .text
  [ 8] T 0x9e4 .gnu_ifunc_resolver section .text
  [ 9] T 0xa58 __glink_PLTresolve section .text
  (...)

GDB needs to know when a program stepped into an ifunc resolver, so
that it can know whether to step past the resolver into the target
function without the user noticing.  The way GDB does it is my
checking whether the current PC points to an ifunc symbol (since
resolver and ifunc have the same address by design).

The problem is then that ppc64_elf_get_synthetic_symtab never creates
the synchetic symbol for the ifunc, so GDB stops stepping at the
resolver (in a test added by the following patch):

  (gdb) step
  gnu_ifunc_resolver (hwcap=21) at gdb/testsuite/gdb.base/gnu-ifunc-lib.c:33
  33      {
  (gdb) FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=1: final_debug=0: step

After this commit, we get:

  [ 8] i 0x9e4 .gnu_ifunc section .text
  [ 9] T 0x9e4 .gnu_ifunc_resolver section .text

And stepping an ifunc call takes to the final function:
  (gdb) step
  0x00000000100009e8 in .final ()
  (gdb) PASS: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=1: final_debug=0: step

An alternative to touching bfd I considered was for GDB to check
whether there's an ifunc data symbol / function descriptor that points
to the current PC, whenever the program stops, but discarded it
because we'd have to do a linear scan over .opd over an over to find a
matching function descriptor for the current PC.  At that point I
considered caching that info, but quickly dismissed it as then that
has no advantage (memory or performance) over just creating the
synthetic ifunc text symbol in the first place.

I ran the binutils and ld testsuites on PPC64 ELFv1 (machine gcc110 on
the GCC compile farm), and saw no regressions.  This commit is part of
a GDB patch series that includes GDB tests that fail without this fix.

bfd/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* elf64-ppc.c (ppc64_elf_get_synthetic_symtab): Don't consider
ifunc and non-ifunc symbols duplicates.

6 years agoFor PPC64/ELFv1: Introduce mst_data_gnu_ifunc
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
For PPC64/ELFv1: Introduce mst_data_gnu_ifunc

Running the new tests added later in the series on PPC64 (ELFv1)
revealed that the current ifunc support needs a bit of a design rework
to work properly on PPC64/ELFv1, as most of the new tests fail.  The
ifunc support only kind of works today if the ifunc symbol and the
resolver have the same name, as is currently tested by the
gdb.base/gnu-ifunc.exp testcase, which is unlike how ifuncs are
written nowadays.

The crux of the problem is that ifunc symbols are really function
descriptors, not text symbols:

   44: 0000000000020060    104 FUNC    GLOBAL DEFAULT       18 gnu_ifunc_resolver
   54: 0000000000020060    104 GNU_IFUNC GLOBAL DEFAULT     18 gnu_ifunc

But, currently GDB only knows about ifunc symbols that are text
symbols.  GDB's support happens to work in practice for PPC64 when the
ifunc and resolver are one and only, like in the current
gdb.base/gnu-ifunc.exp testcase:

   15: 0000000000020060    104 GNU_IFUNC GLOBAL DEFAULT       18 gnu_ifunc

because in that case, the synthetic ".gnu_ifunc" entry point text
symbol that bfd creates from the actual GNU ifunc "gnu_ifunc" function
(descriptor) symbol ends up with the the "is a gnu ifunc" flag set /
copied over:

  (gdb) maint print msymbols
  ...
  [ 8] i 0x9c4 .gnu_ifunc section .text                <<< mst_text_gnu_ifunc
  ...
  [29] D 0x20060 gnu_ifunc section .opd  crtstuff.c    <<< mst_data

But, if the resolver gets a distinct symbol/name from the ifunc
symbol, then we end up with this:

  (gdb) maint print msymbols
  [ 8] T 0x9e4 .gnu_ifunc_resolver section .text               <<< mst_text
  ...
  [29] D 0x20060 gnu_ifunc section .opd  crtstuff.c            <<< mst_data
  [30] D 0x20060 gnu_ifunc_resolver section .opd  crtstuff.c   <<< mst_data

I have a follow up bfd patch that turns that into:

   (gdb) maint print msymbols
+  [ 8] i 0x9e4 .gnu_ifunc section .text               <<< mst_text_gnu_ifunc
   [ 8] T 0x9e4 .gnu_ifunc_resolver section .text      <<< mst_text
   ...
   [29] D 0x20060 gnu_ifunc section .opd  crtstuff.c
   [30] D 0x20060 gnu_ifunc_resolver section .opd  crtstuff.c

but that won't help everything.  We still need this patch.

Specifically, when we do a symbol lookup by name, like e.g., to call a
function (see c-exp.y hunk), e.g., "p gnu_ifunc()", then we need to
know that the found "gnu_ifunc" minimal symbol is an ifunc in order to
do some special processing.  But, on PPC, that lookup by name finds
the function descriptor symbol, which presently is just a mst_data
symbol, while at present, we look for mst_text_gnu_ifunc symbols to
decide whether to do special GNU ifunc processing.  In most of those
places, we could try to resolve the function descriptor with
gdbarch_convert_from_func_ptr_addr, and then lookup the minimal symbol
at the resolved PC, see if that finds a minimal symbol of type
mst_text_gnu_ifunc.  If so, then we could assume that the original
mst_dadta / function descriptor "gnu_ifunc" symbol was an ifunc.  I
tried it, and it mostly works, even if it's not the most efficient.

However, there's one case that can't work with such a design -- it's
that of the user calling the ifunc resolver directly to debug it, like
"p gnu_ifunc_resolver(0)", expecting that to return the function
pointer of the final function (which is exercised by the new tests
added later).  In this case, with the not-fully-working solution, we'd
resolve the function descriptor, find that there's an
mst_text_gnu_ifunc symbol for the resolved address, and proceed
calling the function as if we tried to call "gnu_ifunc", the
user-visible GNU ifunc symbol, instead of the resolver.  I.e., it'd be
impossible to call the resolver directly as a normal function.

Introducing mst_data_gnu_ifunc eliminates the need for several
gdbarch_convert_from_func_ptr_addr calls, and, fixes the "call
resolver directly" use case mentioned above too.  It's the cleanest
approach I could think of.

In sum, we make GNU ifunc function descriptor symbols get a new
"mst_data_gnu_ifunc" minimal symbol type instead of the bare mst_data
type.  So when symbol lookup by name finds such a minimal symbol, we
know we found an ifunc symbol, without resolving the entry/text
symbol.  If the user calls the the resolver symbol instead, like "p
gnu_ifunc_resolver(0)", then we'll find the regular mst_data symbol
for "gnu_ifunc_resolver", and we'll call the resolver function as just
another regular function.

With this, most of the GNU ifunc tests added by a later patch pass on
PPC64 too.  The following bfd patch fixes the remaining issues.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* breakpoint.c (set_breakpoint_location_function): Handle
mst_data_gnu_ifunc.
* c-exp.y (variable production): Handle mst_data_gnu_ifunc.
* elfread.c (elf_symtab_read): Give data symbols with
BSF_GNU_INDIRECT_FUNCTION set mst_data_gnu_ifunc type.
(elf_rel_plt_read): Update comment.
* linespec.c (convert_linespec_to_sals): Handle
mst_data_gnu_ifunc.
(minsym_found): Handle mst_data_gnu_ifunc.
* minsyms.c (msymbol_is_function, minimal_symbol_reader::record)
(find_solib_trampoline_target): Handle mst_data_gnu_ifunc.
* parse.c (find_minsym_type_and_address): Handle
mst_data_gnu_ifunc.
* symmisc.c (dump_msymbols): Handle mst_data_gnu_ifunc.
* symtab.c (find_gnu_ifunc): Handle mst_data_gnu_ifunc.
* symtab.h (minimal_symbol_type) <mst_text_gnu_ifunc>: Update
comment.
<mst_data_gnu_ifunc>: New enumerator.

6 years agoFix stepping past GNU ifunc resolvers (introduce lookup_msym_prefer)
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
Fix stepping past GNU ifunc resolvers (introduce lookup_msym_prefer)

When we're stepping (with "step"), we want to skip trampoline-like
functions automatically, including GNU ifunc resolvers.  That is done
by infrun.c calling into:

  in_solib_dynsym_resolve_code
    -> svr4_in_dynsym_resolve_code
      -> in_gnu_ifunc_stub

A problem here is that if there's a regular text symbol at the same
address as the ifunc symbol, the minimal symbol lookup in
in_gnu_ifunc_stub may miss the GNU ifunc symbol:

(...)
    41: 000000000000071a    53 FUNC    GLOBAL DEFAULT   11 gnu_ifunc_resolver
(...)
    50: 000000000000071a    53 IFUNC   GLOBAL DEFAULT   11 gnu_ifunc
(...)

This causes this FAIL in the tests added later in the series:

 (gdb) PASS: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=0: final_debug=0: resolver received HWCAP
 set step-mode on
 (gdb) PASS: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=0: final_debug=0: set step-mode on
 step
 0x00007ffff7bd371a in gnu_ifunc_resolver () from build/gdb/testsuite/outputs/gdb.base/gnu-ifunc/gnu-ifunc-lib-1-0-0.so
 (gdb) FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=0: final_debug=0: step

Above, GDB simply thought that it stepped into a regular function, so
it stopped stepping, while it should have continued stepping past the
resolver.

The fix is to teach minimal symbol lookup to prefer GNU ifunc symbols
if desired.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* minsyms.c (lookup_minimal_symbol_by_pc_section_1): Rename to ...
(lookup_minimal_symbol_by_pc_section): ... this.  Replace
'want_trampoline' parameter by a lookup_msym_prefer parameter.
Handle it.
(lookup_minimal_symbol_by_pc_section): Delete old implementation.
(lookup_minimal_symbol_by_pc): Adjust.
(in_gnu_ifunc_stub): Prefer GNU ifunc symbols.
(lookup_solib_trampoline_symbol_by_pc): Adjust.
* minsyms.h (lookup_msym_prefer): New enum.
(lookup_minimal_symbol_by_pc_section): Replace 'want_trampoline'
parameter by a lookup_msym_prefer parameter.

6 years agoFor PPC64: elf_gnu_ifunc_record_cache: handle plt symbols in .text section
Pedro Alves [Thu, 26 Apr 2018 12:01:27 +0000 (13:01 +0100)]
For PPC64: elf_gnu_ifunc_record_cache: handle plt symbols in .text section

elf_gnu_ifunc_record_cache doesn't ever record anything on PPC64
(tested on gcc110 on the compile farm, CentOS 7.4, ELFv1), because
that expects to find PLT symbols in the .plt section, while there we
get:

  (gdb) info symbol 'gnu_ifunc@plt'
  gnu_ifunc@plt in section .text
                           ^^^^^

I guess that may be related to the comment in ppc-linux-tdep.c that
says "For secure PLT, stub is in .text".

In any case, this commit fixes the issue by making the function look
at the symbol name instead of at the section.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* elfread.c (elf_gnu_ifunc_record_cache): Check if the symbol name
ends in "@plt" instead of looking at the symbol's section.

6 years agoFactor out minsym_found/find_function_start_sal overload
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Factor out minsym_found/find_function_start_sal overload

I need to make the ifunc resolving code in elfread.c skip the target
function's prologue like minsym_found does.  I thought of factoring
that out to a separate function, but turns out there's already a
comment in find_function_start_sal that says that should agree with
minsym_found...

Instead of making sure the code agrees with a comment, factor out the
common code to a separate function and use it from both places.

Note that the current find_function_start_sal does a bit more than
minsym_found's equivalent (the "We always should ..." bit), though
that's probably a latent bug.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* linespec.c (minsym_found): Use find_function_start_sal CORE_ADDR
overload.
* symtab.c (find_function_start_sal(CORE_ADDR, obj_section *,bool)):
New, factored out from ...
(find_function_start_sal(symbol *, int)): ... this.  Reimplement
and use bool.
* symtab.h (find_function_start_sal(CORE_ADDR, obj_section *,bool)):
New.
(find_function_start_sal(symbol *, int)): Change boolean parameter
type to bool.

6 years agoEliminate find_pc_partial_function_gnu_ifunc
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Eliminate find_pc_partial_function_gnu_ifunc

Not used anywhere any longer.

If this is ever reinstated, note that this case:

  cache_pc_function_is_gnu_ifunc = TYPE_GNU_IFUNC (SYMBOL_TYPE (f));

was incorrect in that regular symbols never have type marked as GNU
ifunc type, only minimal symbols.  At some point I had some fix that
checking the matching minsym here.  But in the end I ended up just
eliminating need for this function, so that fix was not necessary.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* blockframe.c (cache_pc_function_is_gnu_ifunc): Delete.  Remove
all references.
(find_pc_partial_function_gnu_ifunc): Rename to ...
(find_pc_partial_function): ... this, and remove references to
'is_gnu_ifunc_p'.
(find_pc_partial_function): Delete old implementation.
* symtab.h (find_pc_partial_function_gnu_ifunc): Delete.

6 years agoBreakpoints, don't skip prologue of ifunc resolvers with debug info
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Breakpoints, don't skip prologue of ifunc resolvers with debug info

Without this patch, some of the tests added to gdb.base/gnu-ifunc.exp
by a following patch fail like so:

  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=0: set-break: before resolving: break gnu_ifunc
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=0: set-break: before resolving: info breakpoints
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=0: set-break: after resolving: break gnu_ifunc
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=0: set-break: after resolving: info breakpoints
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=1: set-break: before resolving: break gnu_ifunc
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=1: set-break: before resolving: info breakpoints
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=1: set-break: after resolving: break gnu_ifunc
  FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=1: set-break: after resolving: info breakpoints

All of them trigger iff:

 - you have debug info for the ifunc resolver.
 - the resolver and the user-visible symbol have the same name.

If you have an ifunc that has a resolver with the same name as the
user visible symbol, debug info for the resolver masks out the ifunc
minsym.  When you set a breakpoint by name on the user visible symbol,
GDB finds the DWARF symbol for the resolver, and thinking that it's a
regular function, sets a breakpoint location past its prologue.

Like so, location 1.2, before the ifunc is resolved by the inferior:

  (gdb) break gnu_ifunc
  Breakpoint 2 at 0x7ffff7bd36ea (2 locations)
  (gdb) info breakpoints
  Num     Type           Disp Enb Address            What
  1       breakpoint     keep y   <MULTIPLE>
  1.1                         y     0x00007ffff7bd36ea <gnu_ifunc>
  1.2                         y     0x00007ffff7bd36f2 in gnu_ifunc at src/gdb/testsuite/gdb.base/gnu-ifunc-lib.c:34
  (gdb)

And like so, location 2.2, if you set the breakpoint after the ifunc
is resolved by the inferior (to "final"):

  (gdb) break gnu_ifunc
  Breakpoint 5 at 0x400757 (2 locations)
  (gdb) info breakpoints
  Num     Type           Disp Enb Address            What
  2       breakpoint     keep y   <MULTIPLE>
  2.1                         y     0x000000000040075a in final at src/gdb/testsuite/gdb.base/gnu-ifunc-resd.c:21
  2.2                         y     0x00007ffff7bd36f2 in gnu_ifunc at src/gdb/testsuite/gdb.base/gnu-ifunc-lib.c:34
  (gdb)

I don't think this is right because when users set a breakpoint at an
ifunc, they don't care about debugging the resolver.  Instead what you
should is a single location for the ifunc in the first case, and a
single location of the ifunc target in the second case.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* linespec.c (struct bound_minimal_symbol_search_key): New.
(convert_linespec_to_sals): Sort minimal symbols earlier.  Don't
skip first line if we found a GNU ifunc minimal symbol by name.
(compare_msymbols): Change parameters to work with a destructured
lhs minsym.
(compare_msymbols_for_qsort, compare_msymbols_for_bsearch): New
functions.

6 years agoFix setting breakpoints on ifunc functions after they're already resolved
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Fix setting breakpoints on ifunc functions after they're already resolved

This fixes setting breakpoints on ifunc functions by name after the
ifunc has already been resolved.

In that case, if you have debug info for the ifunc resolver, without
the fix, then gdb puts a breakpoint past the prologue of the resolver,
instead of setting a breakpoint at the ifunc target:

  break gnu_ifunc
  Breakpoint 4 at 0x7ffff7bd36f2: file src/gdb/testsuite/gdb.base/gnu-ifunc-lib.c, line 34.
  (gdb) continue
  Continuing.
  [Inferior 1 (process 13300) exited normally]
  (gdb)

above we should have stopped at "final", but didn't because we never
resolved the ifunc to the final location.

If you don't have debug info for the resolver, GDB manages to resolve
the ifunc target, but, it should be setting a breakpoint after the
prologue of the final function, and instead what you get is that GDB
sets a breakpoint on the first address of the target function.  With
the gnu-ifunc.exp tests added by a later patch, we get, without the
fix:

  (gdb) break gnu_ifunc
  Breakpoint 4 at 0x400753
  (gdb) continue
  Continuing.

  Breakpoint 4, final (arg=1) at src/gdb/testsuite/gdb.base/gnu-ifunc-final.c:20
  20 {

vs, fixed:

  (gdb) break gnu_ifunc
  Breakpoint 4 at 0x40075a: file src/gdb/testsuite/gdb.base/gnu-ifunc-final.c, line 21.
  (gdb) continue
  Continuing.

  Breakpoint 4, final (arg=2) at src/gdb/testsuite/gdb.base/gnu-ifunc-final.c:21
  21   return arg + 1;
  (gdb)

Fix the problems above by moving the ifunc target resolving to
linespec.c, before we skip a function's prologue.  We need to save
something in the sal, so that set_breakpoint_location_function knows
that it needs to create a bp_gnu_ifunc_resolver bp_location.  Might as
well just save a pointer to the minsym.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* breakpoint.c (set_breakpoint_location_function): Don't resolve
ifunc targets here.  Instead, if we have an ifunc minsym, use its
address/name.
(add_location_to_breakpoint): Store the minsym and the objfile in
the breakpoint location.
* breakpoint.h (bp_location) <msymbol, objfile>: New fields.
* linespec.c (minsym_found): Resolve GNU ifunc targets here.
Record the minsym in the sal.
* symtab.h (symtab_and_line) <msymbol>: New field.

6 years agoFix elf_gnu_ifunc_resolve_by_got buglet
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Fix elf_gnu_ifunc_resolve_by_got buglet

The next patch will add a call to elf_gnu_ifunc_resolve_by_got that
trips on a latent buglet -- the function is writing to its output
parameter even if the address wasn't found, confusing the caller.  The
function's intro comment says:

  /* Try to find the target resolved function entry address of a STT_GNU_IFUNC
     function NAME.  If the address is found it is stored to *ADDR_P (if ADDR_P
     is not NULL) and the function returns 1.  It returns 0 otherwise.

So fix the function accordingly.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* elfread.c (elf_gnu_ifunc_resolve_by_got): Don't write to *ADDR_P
unless we actually resolved the ifunc.

6 years agoCalling ifunc functions when resolver has debug info, user symbol same name
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Calling ifunc functions when resolver has debug info, user symbol same name

If the GNU ifunc resolver has the same name as the user visible
symbol, and the resolver has debug info, then the DWARF info for the
resolver masks the ifunc minsym.  In that scenario, if you try calling
the ifunc from GDB, you call the resolver instead.  With the
gnu-ifunc.exp testcase added in a following patch, you'd see:

  (gdb) p gnu_ifunc (3)
  $1 = (int (*)(int)) 0x400753 <final>
  (gdb) FAIL: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=1: resolved_debug=0: p gnu_ifunc (3)
                                                       ^^^^^^^^^^^^^^^^

That is, we called the ifunc resolver manually, which returned a
pointer to the ifunc target function ("final").  The "final" symbol is
the function that GDB should have called automatically,

  ~~~~~~~~~~~~
  int
  final (int arg)
  {
    return arg + 1;
  }
  ~~~~~~~~~

which is what happens if you don't have debug info for the resolver:

  (gdb) p gnu_ifunc (3)
  $1 = 4
  (gdb) PASS: gdb.base/gnu-ifunc.exp: resolver_attr=0: resolver_debug=0: resolved_debug=1: p gnu_ifunc (3)
                                                       ^^^^^^^^^^^^^^^^

or if the resolver's symbol has a different name from the ifunc (as is
the case with modern uses of ifunc via __attribute__ ifunc, such as
glibc uses):

  (gdb) p gnu_ifunc (3)
  $1 = 4
  (gdb) PASS: gdb.base/gnu-ifunc.exp: resolver_attr=1: resolver_debug=1: resolved_debug=0: p gnu_ifunc (3)
                                      ^^^^^^^^^^^^^^^

in which case after this patch, you can still call the resolver
directly if you want:

  (gdb) p gnu_ifunc_resolver (3)
  $1 = (int (*)(int)) 0x400753 <final>

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* c-exp.y (variable production): Prefer ifunc minsyms over
regular function symbols.
* symtab.c (find_gnu_ifunc): New function.
* minsyms.h (lookup_msym_prefer): New enum.
(lookup_minimal_symbol_by_pc_section): Replace 'want_trampoline'
parameter by a lookup_msym_prefer parameter.
* symtab.h (find_gnu_ifunc): New declaration.

6 years agoCalling ifunc functions when target has no debug info but resolver has
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Calling ifunc functions when target has no debug info but resolver has

After the previous patch, on Fedora 27 (glibc 2.26), if you try
calling strlen in the inferior, you now get:

  (top-gdb) p strlen ("hello")
  '__strlen_avx2' has unknown return type; cast the call to its declared return type

This is correct, because __strlen_avx2 is written in assembly.

We can improve on this though -- if the final ifunc resolved/target
function has no debug info, but the ifunc _resolver_ does have debug
info, we can try extracting the final function's type from the type
that the resolver returns.  E.g.,:

  typedef size_t (*strlen_t) (const char*);

  size_t my_strlen (const char *) { /* some implementation */ }
  strlen_t strlen_resolver (unsigned long hwcap) { return my_strlen; }

  extern size_t strlen (const char *s);
  __typeof (strlen) strlen __attribute__ ((ifunc ("strlen_resolver")));

In the strlen example above, the resolver returns strlen_t, which is a
typedef for pointer to a function that returns size_t.  "strlen_t" is
the type of both the user-visible "strlen", and of the the target
function that implements it.

This patch teaches GDB to extract that type.

This is done for actual inferior function calls (in infcall.c), and
for ptype (in eval_call).  By the time we get to either of these
places, we've already lost the original symbol/minsym, and only have
values and types to work with.  Hence the changes to c-exp.y and
evaluate_var_msym_value, to ensure that we propagate the ifunc
minsymbol's info.

The change to make ifunc symbols have no/unknown return type exposes a
latent problem -- gdb.compile/compile-ifunc.exp calls a no-debug-info
function, but we did not warn about it.  The test is fixed by this
commit too.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* blockframe.c (find_gnu_ifunc_target_type): New function.
(find_function_type): New.
* eval.c (evaluate_var_msym_value): For GNU ifunc types, always
return a value with a memory address.
(eval_call): For calls to GNU ifunc functions, try to find the
type of the target function from the type that the resolver
returns.
* gdbtypes.c (objfile_type): Don't install a return type for ifunc
symbols.
* infcall.c (find_function_return_type): Delete.
(find_function_addr): Add 'function_type' parameter.  For calls to
GNU ifunc functions, try to find the type of the target function
from the type that the resolver returns, and return it via
FUNCTION_TYPE.
(call_function_by_hand_dummy): Adjust to use the function type
returned by find_function_addr.
(find_function_addr): Add 'function_type' parameter and move
description here.
* symtab.h (find_function_type, find_gnu_ifunc_target_type): New
declarations.

gdb/testsuite/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* gdb.compile/compile-ifunc.exp: Also expect "function has unknown
return type" warnings.

6 years agoFix calling ifunc functions when resolver has debug info and different name
Pedro Alves [Thu, 26 Apr 2018 12:01:26 +0000 (13:01 +0100)]
Fix calling ifunc functions when resolver has debug info and different name

Currently, on Fedora 27 (glibc 2.26), if you try to call strlen in the
inferior you get:

 (gdb) p strlen ("hello")
 $1 = (size_t (*)(const char *)) 0x7ffff554aac0 <__strlen_avx2>

strlen is an ifunc function, and what we see above is the result of
calling the ifunc resolver in the inferior.  That returns a pointer to
the actual target function that implements strlen on my machine.  GDB
should have turned around and called the resolver automatically
without the user noticing.

This is was caused by commit:

  commit bf223d3e808e6fec9ee165d3d48beb74837796de
  Date: Mon Aug 21 11:34:32 2017 +0100

      Handle function aliases better (PR gdb/19487, errno printing)

which added the find_function_alias_target call to c-exp.y, to try to
find an alias with debug info for a minsym.  For ifunc symbols, that
finds the ifunc's resolver if it has debug info (in the example it's
called "strlen_ifunc"), with the result that GDB calls that as a
regular function.

After this commit, we get now get:

  (top-gdb) p strlen ("hello")
  '__strlen_avx2' has unknown return type; cast the call to its declared return type

Which is correct, because __strlen_avx2 is written in assembly.
That'll be improved in a following patch, though.

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* c-exp.y (variable production): Skip finding an alias for ifunc
symbols.

6 years agoFix breakpoints in ifunc after inferior resolved it (@got.plt symbol creation)
Pedro Alves [Thu, 26 Apr 2018 12:02:26 +0000 (13:02 +0100)]
Fix breakpoints in ifunc after inferior resolved it (@got.plt symbol creation)

Setting a breakpoint on an ifunc symbol after the ifunc has already
been resolved by the inferior should result in creating a breakpoint
location at the ifunc target.  However, that's not what happens on
current Fedora:

  (gdb) n
  53        i = gnu_ifunc (1);    /* break-at-call */
  (gdb)
  54        assert (i == 2);
  (gdb) b gnu_ifunc
  Breakpoint 2 at gnu-indirect-function resolver at 0x7ffff7bd36ee
  (gdb) info breakpoints
  Num     Type                   Disp Enb Address            What
  2       STT_GNU_IFUNC resolver keep y   0x00007ffff7bd36ee <gnu_ifunc+4>

The problem is that elf_gnu_ifunc_resolve_by_got never manages to
resolve an ifunc target.  The reason is that GDB never actually
creates the internal got.plt symbols:

 (gdb) p 'gnu_ifunc@got.plt'
 No symbol "gnu_ifunc@got.plt" in current context.

and this is because GDB expects that rela.plt has relocations for
.plt, while it actually has relocations for .got.plt:

 Relocation section [10] '.rela.plt' for section [22] '.got.plt' at offset 0x570 contains 2 entries:
   Offset              Type            Value               Addend Name
   0x0000000000601018  X86_64_JUMP_SLOT 000000000000000000      +0 __assert_fail
   0x0000000000601020  X86_64_JUMP_SLOT 000000000000000000      +0 gnu_ifunc

Using an older system on the GCC compile farm (machine gcc15, an
x86-64 running Debian 6.0.8, with GNU ld 2.20.1), we see that it used
to be that we'd get a .rela.plt section for .plt:

 Relocation section [ 9] '.rela.plt' for section [11] '.plt' at offset 0x578 contains 3 entries:
   Offset              Type            Value               Addend Name
   0x0000000000600cc0  X86_64_JUMP_SLOT 000000000000000000      +0 __assert_fail
   0x0000000000600cc8  X86_64_JUMP_SLOT 000000000000000000      +0 __libc_start_main
   0x0000000000600cd0  X86_64_JUMP_SLOT 000000000000000000      +0 gnu_ifunc

Those offsets did point into .got.plt, as seen with objdump -h:

  20 .got.plt      00000030  0000000000600ca8  0000000000600ca8  00000ca8  2**3
         CONTENTS, ALLOC, LOAD, DATA

I also tested on gcc110 on the compile farm (PPC64 running CentOS
7.4.1708, with GNU ld 2.25.1), and there we see instead:

 Relocation section [ 9] '.rela.plt' for section [23] '.plt' at offset 0x5d0 contains 4 entries:
   Offset              Type            Value               Addend Name
   0x0000000010020148  PPC64_JMP_SLOT  000000000000000000      +0 __libc_start_main
   0x0000000010020160  PPC64_JMP_SLOT  000000000000000000      +0 __gmon_start__
   0x0000000010020178  PPC64_JMP_SLOT  000000000000000000      +0 __assert_fail
   0x0000000010020190  PPC64_JMP_SLOT  000000000000000000      +0 gnu_ifunc

But note that those offsets point into .plt, not .got.plt, as seen
with objdump -h:

 22 .plt          00000078  0000000010020130  0000000010020130  00010130  2**3
                  ALLOC

This commit makes us support all the different combinations above.

With that addressed, we now get:

 (gdb) p 'gnu_ifunc@got.plt'
 $1 = (<text from jump slot in .got.plt, no debug info>) 0x400753 <final>

And setting a breakpoint on the ifunc finds the ifunc target:

 (gdb) b gnu_ifunc
 Breakpoint 2 at 0x400753
 (gdb) info breakpoints
 Num     Type           Disp Enb Address            What
 2       breakpoint     keep y   0x0000000000400753 <final>

gdb/ChangeLog:
2018-04-26  Pedro Alves  <palves@redhat.com>

* elfread.c (elf_rel_plt_read): Look for relocations for .got.plt too.

6 years agox86: fold various non-memory operand AVX512VL templates
Jan Beulich [Thu, 26 Apr 2018 06:55:02 +0000 (08:55 +0200)]
x86: fold various non-memory operand AVX512VL templates

There's little point carrying up to three templates per insn flavor
when the sole difference is operand size and the dependency on AVX512VL
being enabled. Instead the need for AVX512VL can be derived from an
operand allowing for ZMMword as well as one or both or XMMword and
YMMword (irrespective of whether this is a register or memory operand).
Without further abstraction to deal with the different Disp8MemShift
values between the templates, only a limited set (mostly ones only
allowing for non-memory operands) can be folded, which is being done
here.

Also drop IgnoreSize wherever possible from anything that's being
touched anyway.

6 years agox86: also optimize zeroing-masking variants of insns
Jan Beulich [Thu, 26 Apr 2018 06:53:20 +0000 (08:53 +0200)]
x86: also optimize zeroing-masking variants of insns

When zeroing an element of a register it doesn't matter whether the zero
results from the actual operation (xor, sub, or nand) or from the
zeroing-masking taking effect due to a clear mask register bit.

6 years agox86: properly force / avoid forcing EVEX encoding
Jan Beulich [Thu, 26 Apr 2018 06:49:41 +0000 (08:49 +0200)]
x86: properly force / avoid forcing EVEX encoding

Pseudo prefixes are supposed to be a hint only - when the specific
encoding can't be used to encode an insn, silently override it. But
this overriding must only happen after the respective check, to
avoid forcing EVEX encoding because of something that isn't a valid
register name in the given context.

6 years agox86: CpuXSAVE is a prereq for various other features
Jan Beulich [Thu, 26 Apr 2018 06:48:56 +0000 (08:48 +0200)]
x86: CpuXSAVE is a prereq for various other features

All of AVX, LWP, MPX, and PKU require XSAVE, and hence it as well as
XRSTOR should be enabled when enabling these ISA extensions. Leverage
these implications to shorten some of the cpu_flag_init[] entries.

6 years agox86: drop CpuRegMMX, CpuReg[XYZ]MM, and CpuRegMask
Jan Beulich [Thu, 26 Apr 2018 06:48:01 +0000 (08:48 +0200)]
x86: drop CpuRegMMX, CpuReg[XYZ]MM, and CpuRegMask

It's not clear to me why they had been introduced - the respective
comments in opcodes/i386-gen.c are certainly wrong: ymm<N> registers
are very well supported (and necessary) with just AVX512F.

6 years agox86: don't recognize bnd<N> as registers without CpuMPX
Jan Beulich [Thu, 26 Apr 2018 06:46:39 +0000 (08:46 +0200)]
x86: don't recognize bnd<N> as registers without CpuMPX

This is just like for all other extended/optional register sets.

6 years agox86: x87-related adjustments
Jan Beulich [Thu, 26 Apr 2018 06:45:35 +0000 (08:45 +0200)]
x86: x87-related adjustments

Neither 287 wrt 8087 nor 387 wrt 287 are proper supersets - in each case
some insns get removed from the ISA (they become NOPs, but code intended
for newer co-processors should not use them).

Furthermore with .no87, ST should not be recognized as a register name.

6 years agox86: fix indentation in build_modrm_byte()
Jan Beulich [Thu, 26 Apr 2018 06:31:14 +0000 (08:31 +0200)]
x86: fix indentation in build_modrm_byte()

The VEX3SOURCES code was (originally) written with just space
indentation, which is not in line with general coding style as well as
the style later in the function.

6 years agox86: move and fold common code in build_modrm_byte()
Jan Beulich [Thu, 26 Apr 2018 06:30:45 +0000 (08:30 +0200)]
x86: move and fold common code in build_modrm_byte()

The source and reg_slot calculations in the VEX3SOURCES only depend on
the number of immediate operands.