gdbserver: start turning the target ops vector into a class
This is the beginning of a series of patches where the goal is to turn
the target ops vector into a class and all the target op function
pointers into methods of this class.
Currently, the target ops is a struct of function pointers. At the
end of the series, it becomes a class with methods, and the existing
low target definitions become subclasses. That is, we end up with the
following class hierarchy:
process_stratum_target either defines the default behavior for the
target ops or leaves them as pure virtual for the subclasses to
override.
The transformation is done by first introducing a helper class, called
'process_target', that is initially empty. An instance of this class
is added to the end of the current target ops vector. This new field
is called 'pt'. We will gradually carry target ops to the new class,
one by one, whereas the invocation of the target op will be converted
to a method call on 'pt'.
For instance, target op 'attach' is currently invoked as
(*the_target->attach) (args)
After moving 'attach' as a method to 'process_target', it will be
invoked as
the_target->pt->attach (args)
In this process, the concrete target vector definitions
(e.g. linux-low, win32-low, nto-low, etc.) are turned into derived
classes of 'process_target', so that they can either inherit the
default behavior of the target ops or can override the method.
We prefer to make this transition gradually rather than in a single
giant patch, to yield bite-size patches. The goal is that after each
patch gdbserver will still be buildable and testable.
The general rule of thumb when converting a target op to a method is
this:
(1) If the function call is protected with a NULL-check with an
obvious default behavior, simply implement that default behavior in
the base class (e.g.: supports_non_stop).
(2) If there is no NULL-check guard, the method becomes pure
virtual, and the derived targets are required to implement the method
(e.g.: attach).
(3) If there is a NULL-check but no apparent default behavior, or if
the NULL-check is utilized to populate a feature support packet,
introduce a 'supports_XYZ' method (e.g.: pid_to_exec_file).
The overall strategy is to preserve the existing behavior as much as
possible.
When we're done moving all the target ops into 'process_target', the
target op vector will contain nothing but the field 'pt'. At that
point, the auxiliary class 'process_target' will simply meld into
'process_stratum_target' and the method calls of the form
'the_target->pt->xyz' will be turned into 'the_target->xyz'.
The "linux-low" target has been built and reg-tested on X86_64 Linux
(Ubuntu). The "win32-low" target has been built (but not tested) via
cross-compilation to a x86_64-w64-mingw32 target. The "lynx-low" and
"nto-low" targets were neither built nor tested.
gdbserver/ChangeLog:
2020-02-20 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* target.h (class process_target): New class definition.
(struct process_stratum_target) <pt>: New field with type
'process_target*'.
* linux-low.h (class linux_process_target): Define as a derived
class of 'process_target'.
* linux-low.cc (linux_target_ops): Add a linux_process_target*
as the 'pt' field.
* lynx-low.h (class lynx_process_target): Define as a derived
class of 'process_target'.
* lynx-low.cc (lynx_target_ops): Add a lynx_process_target*
as the 'pt' field.
* nto-low.h (class nto_process_target): Define as a derived
class of 'process_target'.
* nto-low.cc (nto_target_ops): Add an nto_process_target*
as the 'pt' field.
* win32-low.h (class win32_process_target): Define as a derived
class of 'process_target'.
* win32-low.cc (win32_target_ops): Add a win32_process_target*
as the 'pt' field.
Nick Clifton [Thu, 20 Feb 2020 13:08:29 +0000 (13:08 +0000)]
Stop the BFD library from automatically converting OS and PROC specific symbol section indicies to SHN_ABS, and provide a hook for backends to decide how such indicies should be processed.
* elf-bfd.h (struct elf_backend_data): Add symbol_section_index
callback.
* elfxx-target.h (elf_backend_symbol_section_index): Provide
default definition.
(elfNN_bed): Initialise the symbol_section_index field.
* elf.c (swap_out_syms): Call symbol_section_index, if defined, on
OS and PROC specific section indicies. Warn if converting other
reserved incidies to SHN_ABS.
Andrew Burgess [Fri, 17 Jan 2020 00:10:22 +0000 (00:10 +0000)]
gdb: Allow more control over where to find python libraries
The motivation behind this commit is to make it easier to bundle the
Python *.py library files with GDB when statically linking GDB against
libpython. The Python files will be manually added into the GDB
installation tree, and GDB should be able to find them at run-time.
The installation tree will look like this:
The benefit here is that the entire installation tree can be bundled
into a single archive and copied to another machine with a different
version of Python installed, and GDB will still work, including its
Python support.
In use the new configure options would be used something like this,
first build and install a static Python library:
mkdir python
cd python
# Clone or download Python into a src/ directory.
mkdir build
export PYTHON_INSTALL_PATH=$PWD/install
cd build
../src/configure --disable-shared --prefix=$PYTHON_INSTALL_PATH
make
make install
Now build and install GDB:
mkdir binutils-gdb
cd binutils-gdb
# Clone or download GDB into a src/ directory.
mkdir build
export GDB_INSTALL_DIR=$PWD/install
cd build
../src/configure \
--prefix=$GDB_INSTALL_DIR \
--with-python=$PYTHON_INSTALL_PATH/bin/python3 \
--with-python-libdir=$GDB_INSTALL_DIR/lib
make all-gdb
make install-gdb
Finally, copy the Python libraries into the GDB install:
After this the Python src, build, and install directories are no
longer needed and can be deleted.
If the new --with-python-libdir option is not used then the existing
behaviour is left unchanged, GDB will look for the Python libraries in
the lib/ directory within the python path. The concatenation of the
python prefix and the string 'lib/' is now done at configure time,
rather than at run time in GDB as it was previous, however, this was
never something that the user had dynamic control over, so there's no
loss of functionality.
Jim Wilson [Wed, 19 Feb 2020 22:51:07 +0000 (14:51 -0800)]
RISC-V: Convert the ADD/ADDI to the compressed MV/LI if RS1 is zero.
2020-02-19 Nelson Chu <nelson.chu@sifive.com>
gas/
* testsuite/gas/riscv/c-add-addi.d: New testcase.
* testsuite/gas/riscv/c-add-addi.s: Likewise.
opcodes/
* riscv-opc.c (riscv_opcodes): Convert add/addi to the compressed
c.mv/c.li if rs1 is zero.
Tom de Vries [Wed, 19 Feb 2020 21:57:19 +0000 (22:57 +0100)]
[gdb/testsuite] Fix xpass in gdb.python/lib-types.exp
When running gdb.python/lib-types.exp, we have an xpass:
...
(gdb) python print (str (typedef_const_typedef_class1_ref_obj.type))^M
typedef_const_typedef_class1_ref^M
(gdb) XPASS: gdb.python/lib-types.exp: \
python print (str (typedef_const_typedef_class1_ref_obj.type)) \
(PRMS gcc/55641)
...
When running the same with gcc 4.8, we have an xfail instead:
...
(gdb) python print (str (typedef_const_typedef_class1_ref_obj.type))^M
const typedef_const_typedef_class1_ref^M
(gdb) XFAIL: gdb.python/lib-types.exp: \
python print (str (typedef_const_typedef_class1_ref_obj.type)) \
(PRMS gcc/55641)
...
Fix the xpass by xfailing only for the gcc 4.8 pattern.
Tom de Vries [Wed, 19 Feb 2020 21:33:42 +0000 (22:33 +0100)]
[gdb/testsuite] Fix funcall_ref.exp xpass
When running gdb.ada/funcall_ref.exp I run into two XPASSes:
...
(gdb) p get ("Hello world!")^M
$1 = (n => 12, s => "Hello world!")^M
(gdb) XPASS: gdb.ada/funcall_ref.exp: p get ("Hello world!")
ptype get ("Hello world!")^M
type = <ref> record^M
n: natural;^M
s: access array (1 .. n) of character;^M
end record^M
(gdb) XPASS: gdb.ada/funcall_ref.exp: ptype get ("Hello world!")
...
The xfails are documented in funcall_ref.exp:
...
# Currently, GCC describes such functions as returning pointers (instead of
# references).
setup_xfail *-*-*
...
Using gnatmake 4.8, we can reproduce the XFAILs:
...
(gdb) p get ("Hello world!")^M
$1 = (access foo.bar) 0x6147b0 <system.secondary_stack.chunk+48>^M
(gdb) XFAIL: gdb.ada/funcall_ref.exp: p get ("Hello world!")
ptype get ("Hello world!")^M
type = access record^M
n: natural;^M
s: access array (1 .. n) of character;^M
end record^M
(gdb) XFAIL: gdb.ada/funcall_ref.exp: ptype get ("Hello world!")
...
Fix the XPASSes by:
- removing the xfail setup
- switching the order of the two tests
- detecting the "access record" type and declaring the first test unsupported,
and skipping the second test
Tested on x86_64-linux, both with gnatmake 4.8.5 and gnatmake 7.5.0.
gdb/testsuite/ChangeLog:
2020-02-19 Tom de Vries <tdevries@suse.de>
* gdb.ada/funcall_ref.exp: Replace xfail setup by unsupported check.
This is a simple cleanup. These functions used to use the objfile's
obstack for allocation in the hash tables, but they don't anymore.
Remove the unnecessary objfile parameters, which in turn allows removing
some local variables.
Tom de Vries [Wed, 19 Feb 2020 20:33:39 +0000 (21:33 +0100)]
[gdb/testsuite] Fix corefile-buildid.exp with check-read1
When running gdb.base/corefile-buildid.exp using check-read1, I run into:
...
FAIL: gdb.base/corefile-buildid.exp: shared: info files (timeout)
FAIL: gdb.base/corefile-buildid.exp: symlink shared: info files (timeout)
FAIL: gdb.base/corefile-buildid.exp: shared sepdebug: info files (timeout)
FAIL: gdb.base/corefile-buildid.exp: symlink shared sepdebug: info files \
(timeout)
...
This is caused by attempting to match the output of an "info files" command
using a single gdb_test in check_exec_file.
Fix this by doing line-by-line matching in check_exec_file.
Tested on x86_64-linux, using make targets check and check-read1.
gdb/testsuite/ChangeLog:
2020-02-19 Tom de Vries <tdevries@suse.de>
* gdb.base/corefile-buildid.exp (check_exec_file): Match info files
output line-by-line.
Tom Tromey [Wed, 19 Feb 2020 19:18:09 +0000 (12:18 -0700)]
Fix declaration of mips_pc_is_mips
A build where CORE_ADDR is not the same as bfd_vma pointed out that
mips_pc_is_mips is declared using bfd_vma as the parameter type, but
defined using CORE_ADDR. This patch fixes the declaration.
gdb/ChangeLog
2020-02-19 Tom Tromey <tromey@adacore.com>
* mips-tdep.h (mips_pc_is_mips): Parameter type is CORE_ADDR.
Andrew Burgess [Wed, 5 Feb 2020 11:50:07 +0000 (11:50 +0000)]
Merge changes from GCC for the config/ directory
GCC's config/ChangeLog since the last time this merge was done
(in the binutils-gdb commit 0b4d000cc4e8e77c823) is included at the
end of this commit message.
It is worth noting that the binutils-gdb commit 301a9420d947da1458
added the file config/debuginfod.m4 which is not present in GCC's
config/ directory. This file is preserved, unmodified, after this
commit.
In order to regenerate all of the configure files, I configured with
--enable-maintainer-mode, and built the 'all' target. I then did the
same thing on a source tree without this patch, and only committed
those files that changed when this patch was added.
2020-01-27 Andrew Burgess <andrew.burgess@embecosm.com>
* lib-link.m4 (AC_LIB_LINKFLAGS_BODY): Add new
--with-libXXX-type=... option. Use this to guide the selection of
either a shared library or a static library.
PR bootstrap/85574
* bootstrap-lto.mk (extra-compare): Set to gcc/lto1$(exeext).
2019-04-16 Martin Liska <mliska@suse.cz>
* bootstrap-lto-lean.mk: Filter out -flto in STAGEtrain_CFLAGS.
2019-04-09 Martin Liska <mliska@suse.cz>
* bootstrap-lto-lean.mk: New file.
2019-03-02 Johannes Pfau <johannespfau@gmail.com>
* mh-mingw: Also set __USE_MINGW_ACCESS flag for C++ code.
2018-10-31 Joseph Myers <joseph@codesourcery.com>
PR bootstrap/82856
* math.m4, tls.m4: Use AC_LANG_SOURCE.
Merge from binutils-gdb:
2018-06-19 Simon Marchi <simon.marchi@ericsson.com>
* override.m4 (_GCC_AUTOCONF_VERSION): Bump from 2.64 to 2.69.
config/ChangeLog:
* ax_count_cpus.m4: New file, backported from GCC.
* bootstrap-Og.mk: New file, backported from GCC.
* bootstrap-lto-lean.mk: New file, backported from GCC.
* bootstrap-lto.mk: Changes backported from GCC.
* futex.m4: Changes backported from GCC.
* gthr.m4: Changes backported from GCC.
* lib-link.m4: Changes backported from GCC.
* mh-mingw: Changes backported from GCC.
* no-executables.m4: Changes backported from GCC.
* tls.m4: Changes backported from GCC.
* toolexeclibdir.m4: New file, backported from GCC.
bfd
* cpu-z80.c: Add machine type compatibility checking.
gas
* config/tc-z80.c (md_parse_option): Do not use an underscore
prefix for local labels in SDCC compatability mode.
(z80_start_line_hook): Remove SDCC dollar label support.
* testsuite/gas/z80/sdcc.d: Update expected disassembly.
* testsuite/gas/z80/sdcc.s: Likewise.
Tom Tromey [Wed, 19 Feb 2020 14:50:21 +0000 (07:50 -0700)]
Remove useless NULL check in python.c
I noticed that do_start_initialization, in python.c, checks the result
of xmalloc. However, xmalloc cannot fail, so this check is useless.
This patch also changes the code to use XNEWVEC.
gdb/ChangeLog
2020-02-19 Tom Tromey <tromey@adacore.com>
* python/python.c (do_start_initialization): Use XNEWVEC. Remove
NULL check.
H.J. Lu [Wed, 19 Feb 2020 12:54:45 +0000 (04:54 -0800)]
x86: Mark cvtpi2ps and cvtpi2pd as MMX
* config/tc-i386.c (output_insn): Mark cvtpi2ps and cvtpi2pd
with GNU_PROPERTY_X86_FEATURE_2_MMX.
* testsuite/gas/i386/i386.exp: Run property-3 and
x86-64-property-3.
* testsuite/gas/i386/property-3.d: New file.
* testsuite/gas/i386/property-3.s: Likewise.
* testsuite/gas/i386/x86-64-property-3.d: Likewise.
Tom de Vries [Wed, 19 Feb 2020 07:24:44 +0000 (08:24 +0100)]
[gdb/testsuite] Be quiet about missing prelink in solib-overlap.exp
When running gdb.base/solib-overlap.exp, I get:
...
Running src/gdb/testsuite/gdb.base/solib-overlap.exp ...
sh: prelink: command not found
=== gdb Summary ===
nr of untested testcases 1
...
The verbose output on stdout/stderr is due to using system to execute
prelink, which also means that the output is not captured in gdb.log and
gdb.sum.
Fix this by using exec instead of system.
Tested on x86_64-linux, with:
- no prelink installed, and
- a fake prelink installed, using "cp /usr/bin/echo ~/bin/prelink".
gdb/testsuite/ChangeLog:
2020-02-19 Tom de Vries <tdevries@suse.de>
* gdb.base/solib-overlap.exp: Use exec instead of system to execute
prelink.
Tom de Vries [Wed, 19 Feb 2020 06:05:13 +0000 (07:05 +0100)]
[gdb/testsuite] Ignore pass in gdb_caching_proc
Before commit d4295de4f3 "[gdb/testsuite] Handle missing gnatmake in
gnat_runtime_has_debug_info", calling the gdb_caching_proc
gnat_runtime_has_debug_info could generate a pass because of using
gdb_compile_ada.
This has been fixed in that commit by using a factored out variant
gdb_compile_ada_1, which does not call pass.
Additionally, fix cases like this in more generic way: by ignoring pass calls
during execution of a gdb_caching_proc.
Build and reg-tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-02-19 Tom de Vries <tdevries@suse.de>
* lib/cache.exp (ignore_pass, gdb_do_cache_wrap): New proc.
(gdb_do_cache): Use gdb_do_cache_wrap.
* gdb.base/gdb-caching-proc.exp (test_proc): Use gdb_do_cache_wrap.
Alan Modra [Wed, 19 Feb 2020 02:45:20 +0000 (13:15 +1030)]
Miscellaneous memory alloc related fixes
Some minor tidies. Allocating memory for internal relocs and symbols
after reading external relocs is slightly better with fuzzed files.
You can at least do something about silly sizes that way.
* aoutx.h (slurp_reloc_table): Allocate reloc_cache after
reading external relocs.
* ecoff.c (ecoff_slurp_reloc_table): Likewise.
* archive.c (_bfd_write_archive_contents): Don't twiddle bfd_error
after bfd_bread.
* archive64.c (_bfd_archive_64_bit_slurp_armap): Remove unnecessary
bfd_release.
* elf32-m32c.c (m32c_offset_for_reloc): Make shndx_buf a bfd_byte*.
(m32c_elf_relax_section): Likewise.
* elf32-rl78.c (rl78_offset_for_reloc): Likewise.
(rl78_elf_relax_section): Likewise.
* elf32-rx.c (rx_offset_for_reloc): Likewise.
(elf32_rx_relax_section): Likewise.
* mach-o.c (bfd_mach_o_alloc_and_read): Move earlier with better
parameter types and use..
(bfd_mach_o_read_dylinker, bfd_mach_o_read_dylib),
(bfd_mach_o_read_fvmlib, bfd_mach_o_read_str): ..in these functions.
* peicode.h (pe_bfd_object_p): Don't zero the part of opthdr
being read from file, just the extra.
* som.c (som_slurp_symbol_table): Allocate internal symbol buffer
after reading external syms. Free on failure.
Alan Modra [Wed, 19 Feb 2020 02:45:06 +0000 (13:15 +1030)]
_bfd_mul_overflow
This patch removes the bfd_alloc2 series of memory allocation functions,
replacing them with __builtin_mul_overflow followed by bfd_alloc. Why
do that? Well, a followup patch will implement _bfd_alloc_and_read
and I don't want to implement alloc2 variants as well.
Alan Modra [Wed, 19 Feb 2020 02:44:45 +0000 (13:14 +1030)]
alloc2 used unnecessarily
The bfd_alloc2 series of functions were invented to handle cases where
nmemb * size can overflow. This patch changes some places where the
calculation can't overflow.
* elf.c (bfd_section_from_shdr): Use bfd_zalloc rather than
bfd_zalloc2.
(assign_section_numbers): Likewise.
(elf_map_symbols): Likewise, and bfd_alloc rather than bfd_alloc2.
(_bfd_elf_map_sections_to_segments): Use bfd_malloc rather than
bfd_malloc2, size_t amt, and unsigned tls_count.
(rewrite_elf_program_header): Use bfd_malloc and size_t amt.
* elflink.c (elf_create_symbuf): Use bfd_malloc.
(elf_output_implib): Use bfd_alloc.
Alan Modra [Wed, 19 Feb 2020 02:44:28 +0000 (13:14 +1030)]
bfd_get_size cache
We have calls to bfd_get_size when swapping in ELF section headers.
Since object files can have a large number of sections, it's worth
caching the file size rather than making lots of stat system calls.
* bfd.c (struct bfd): Move format and direction to other
bitfields. Add "size".
* bfdio.c (bfd_get_size): Cache size when not writing file.
* opncls.c (bfd_get_debug_link_info_1): Allow for bfd_get_size
returning zero, ie. unknown.
(bfd_get_alt_debug_link_info): Likewise.
* bfd-in2.h: Regenerate.
Alan Modra [Wed, 19 Feb 2020 02:42:00 +0000 (13:12 +1030)]
bfd_size_type to size_t
bfd_size_type was invented a long time ago in the K&R days. Many
places in binutils ought to be using size_t instead (and there are
lots of places that use long or unsigned long that really ought to use
size_t too). Note that you can't change everything over to size_t: A
32-bit host needs a larger type than size_t to support reading and
processing of 64-bit ELF object files. This patch just tidies some
of the more obvious uses of bfd_size_type that could be size_t. There
no doubt are more lurking in the source. Incidentally, practically
all functions used for output of object files can use size_t and don't
need to worry about overflow of size expressions. If you have
something like
symcount * sizeof (void *)
when symcount is counting symbols already in memory then you know that
this expression can't overflow since the size of a symbol in memory is
larger by far than that of a pointer.
Alan Modra [Wed, 19 Feb 2020 02:41:17 +0000 (13:11 +1030)]
c99 elfxx-riscv.c fix
We can't use c99 without enabling c99 support for older compilers
that don't enable c99 by default. So if you want to use c99 contructs
in binutils you'll need to first arrange for -std=c99 to be passed to
older compilers.
* elfxx-riscv.c (riscv_multi_letter_ext_valid_p): Don't use C99.
Implement RISC-V/Linux support for both RV64 and RV32 systems, including
XML target description handling based on features determined, GPR and
FPR regset support including dynamic sizing of the latter, and software
breakpoint handling. Define two NT_FPREGSET regsets of a different size
matching the FPR sizes supported for generic `gdbserver' code to pick
from according to what the OS supplies.
Also handle a glibc bug where ELF_NFPREG is defined in terms of NFPREG,
however NFPREG is nowhere defined.
2020-02-19 Maciej W. Rozycki <macro@wdc.com>
Andrew Burgess <andrew.burgess@embecosm.com>
Andrew Burgess [Wed, 19 Feb 2020 01:24:37 +0000 (01:24 +0000)]
gdb/riscv: Update API for looking up target descriptions
In preparation for adding the RISC-V gdbserver, this commit
restructures the API for looking up target descriptions.
The current API is riscv_create_target_description, which creates a
target description from a riscv_gdbarch_features, but also caches the
created target descriptions so that for a given features object we
always get back the same target description object. This is important
for GDB due to the way gdbarch objects are reused.
As the same target description is always returned to GDB, and can be
returned multiple times, it is returned as a const, however, the
current cache actually stores a non-const target description. This is
improved in this patch so that the cache holds a const target
description.
For gdbsever, this caching of the target descriptions is not needed,
the gdbserver looks up one target description to describe the target
it is actually running on and that is it. Further the gdbserver
actually needs to modify the target description that is looked up, so
for the gdbsever, returning a const target description is not
acceptable.
This commit aims to address this by creating two parallel target
description APIs, on is the old riscv_create_target_description,
however, this no longer performs any caching, and just creates a new
target description, and returns it as non-const.
The second API is riscv_lookup_target_description, this one performs
the caching, and calls riscv_create_target_description to create a
target description when needed.
In order to make sure the correct API is used in the correct place I
have guarded the code using the GDBSERVER define. For GDB the
riscv_create_target_description is static, and not generally usable
throughout GDB, only the lookup API is global. In gdbserver, the
lookup functions, and the cache are not defined or created at all,
only the riscv_create_target_description API is available.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* arch/riscv.c (struct riscv_gdbarch_features_hasher): Only define
if GDBSERVER is not defined.
(riscv_tdesc_cache): Likewise, also store const target_desc.
(STATIC_IN_GDB): Define.
(riscv_create_target_description): Update declaration with
STATIC_IN_GDB.
(riscv_lookup_target_description): New function, only define if
GDBSERVER is not defined.
* arch/riscv.h (riscv_create_target_description): Declare only
when GDBSERVER is defined.
(riscv_lookup_target_description): New declaration when GDBSERVER
is not defined.
* nat/riscv-linux-tdesc.c (riscv_linux_read_description): Rename to...
(riscv_linux_read_features): ...this, and return
riscv_gdbarch_features instead of target_desc.
* nat/riscv-linux-tdesc.h: Include 'arch/riscv.h'.
(riscv_linux_read_description): Rename to...
(riscv_linux_read_features): ...this.
* riscv-linux-nat.c (riscv_linux_nat_target::read_description):
Update to use riscv_gdbarch_features and
riscv_lookup_target_description.
* riscv-tdep.c (riscv_find_default_target_description): Use
riscv_lookup_target_description instead of
riscv_create_target_description.
Tom de Vries [Tue, 18 Feb 2020 23:05:40 +0000 (00:05 +0100)]
[gdb/testsuite] Be quiet about untested dtrace-prob.exp
When running gdb.base/dtrace-probe.exp, I get this on stdout/stderr:
...
Running src/gdb/testsuite/gdb.base/dtrace-probe.exp ...
gdb compile failed, ld: error in \
build/gdb/testsuite/outputs/gdb.base/dtrace-probe/dtrace-probe.o\
(.eh_frame); no .eh_frame_hdr table will be created
ld: crt1.o: in function `_start':
start.S:110: undefined reference to `main'
ld: build/gdb/testsuite/outputs/gdb.base/dtrace-probe/dtrace-probe-p.o:\
(.SUNW_dof+0x88): undefined reference to `main'
ld: build/gdb/testsuite/outputs/gdb.base/dtrace-probe/dtrace-probe-p.o:\
(.SUNW_dof+0xb8): undefined reference to `main'
collect2: error: ld returned 1 exit status
=== gdb Summary ===
nr of untested testcases 1
...
There is no reason to be this verbose about the failure to compile.
Fix this by using quiet as additional option to gdb_compile in
dtrace_build_usdt_test_program. Note that the error message still occurs in
gdb.log.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-02-19 Tom de Vries <tdevries@suse.de>
* lib/dtrace.exp (dtrace_build_usdt_test_program): Use quiet as
gdb_compile option.
Simon Marchi [Tue, 18 Feb 2020 22:30:51 +0000 (17:30 -0500)]
gdb: change print format of flag enums with value 0
If a flag enum has value 0 and the enumeration type does not have an
enumerator with value 0, we currently print:
$1 = (unknown: 0x0)
I don't like the display of "unknown" here, since for flags, 0 is a
an expected value. It just means that no flags are set. This patch
makes it so that we print it as a simple 0 in this situation:
$1 = 0
If there is an enumerator with value 0, it is still printed using that
enumerator, for example (from the test):
$1 = FE_NONE
gdb/ChangeLog:
* valprint.c (generic_val_print_enum_1): When printing a flag
enum with value 0 and there is no enumerator with value 0, print
just "0" instead of "(unknown: 0x0)".
Simon Marchi [Tue, 18 Feb 2020 22:30:21 +0000 (17:30 -0500)]
gdb: print unknown part of flag enum in hex
When we print the "unknown" part of a flag enum, it is printed in
decimal. I think it would be more useful if it was printed in hex, as
it helps to determine which bits are set more than a decimal value.
gdb/ChangeLog:
* valprint.c (generic_val_print_enum_1): Print unknown part of
flag enum in hex.
gdb/testsuite/ChangeLog:
* gdb.base/printcmds.exp (test_print_enums): Expect hex values
for "unknown".
Simon Marchi [Tue, 18 Feb 2020 22:29:23 +0000 (17:29 -0500)]
gdb: allow duplicate enumerators in flag enums
I have come across some uses cases where it would be desirable to treat
an enum that has duplicate values as a "flag enum". For example, this
one here [1]:
/* Alias for header backward compatibility. */
MEMBARRIER_CMD_SHARED = MEMBARRIER_CMD_GLOBAL,
};
The last enumerator is kept for backwards compatibility. Without this
patch, this enumeration wouldn't be considered a flag enum, because two
enumerators collide. With this patch, it would be considered a flag
enum, and the value 3 would be printed as:
Although if people prefer, we could display both MEMBARRIER_CMD_GLOBAL
and MEMBARRIER_CMD_SHARED in the result. It wouldn't be wrong, and
could perhaps be useful in case a bit may have multiple meanings
(depending on some other bit value).
Simon Marchi [Tue, 18 Feb 2020 22:28:23 +0000 (17:28 -0500)]
gdb: fix printing of flag enums with multi-bit enumerators
GDB has this feature where if an enum looks like it is meant to
represent binary flags, it will present the values of that type as a
bitwise OR of the flags that are set in the value.
The original motivation for this patch is to fix this behavior:
enum hello { AAA = 0x1, BBB = 0xf0 };
(gdb) p (enum hello) 0x11
$1 = (AAA | BBB)
This is wrong because the bits set in BBB (0xf0) are not all set in the
value 0x11, but GDB presents it as if they all were.
I think that enumerations with enumerators that have more than one bit
set should simply not qualify as "flag enum", as far as this
heuristic is concerned. I'm not sure what it means to have flags of
more than one bit. So this is what this patch implements.
I have added an assert in generic_val_print_enum_1 to make sure the flag
enum types respect that, in case they are used by other debug info
readers, in the future.
I've enhanced the gdb.base/printcmds.exp test to cover this case. I've
also added tests for printing flag enums with value 0, both when the
enumeration has and doesn't have an enumerator for value 0.
gdb/ChangeLog:
* dwarf2/read.c: Include "count-one-bits.h".
(update_enumeration_type_from_children): If an enumerator has
multiple bits set, don't treat the enumeration as a "flag enum".
* valprint.c (generic_val_print_enum_1): Assert that enumerators
of flag enums have 0 or 1 bit set.
gdb/testsuite/ChangeLog:
* gdb.base/printcmds.c (enum flag_enum): Prefix enumerators with
FE_, add FE_NONE.
(three): Update.
(enum flag_enum_without_zero): New enum.
(flag_enum_without_zero): New variable.
(enum not_flag_enum): New enum.
(three_not_flag): New variable.
* gdb.base/printcmds.exp (test_artificial_arrays): Update.
(test_print_enums): Add more tests for printing flag enums.
Tom de Vries [Tue, 18 Feb 2020 09:18:36 +0000 (10:18 +0100)]
[gdb/testsuite] Handle missing gnatmake in gnat_runtime_has_debug_info
After de-installing gnatmake, I get on stdout/stderr:
...
Running src/gdb/testsuite/gdb.base/gdb-caching-proc.exp ...
FAIL: gdb-caching-proc.exp: failed to compile gnat-debug-info test binary
...
FAIL: gdb-caching-proc.exp: failed to compile gnat-debug-info test binary
...
In gdb.sum, we see these FAILs (each paired with an UNSUPPORTED as well)
followed by:
...
PASS: gdb-caching-proc.exp: gnat_runtime_has_debug_info consistency
...
Likewise, after re-installing gnatmake, I get a PASS for each of the
UNSUPPORTEDs, and the FAILs disappear.
The FAIL comes from gnat_runtime_has_debug_info, the PASS/UNSUPPORTED comes
from gdb_compile_ada.
Fix this by removing the corresponding fail call in
gnat_runtime_has_debug_info, as well as using a new variant gdb_compile_ada_1
that doesn't call pass/unsupported.
Tested on x86_64-linux, with gnatmake installed and de-installed.
gdb/testsuite/ChangeLog:
2020-02-18 Tom de Vries <tdevries@suse.de>
* lib/ada.exp (gdb_compile_ada_1): Factor out of ...
(gdb_compile_ada): ... here.
(gnat_runtime_has_debug_info): Remove fail call for gdb_compile_ada
failure. Use gdb_compile_ada_1 instead of gdb_compile_ada.
Tom Tromey [Mon, 17 Feb 2020 17:00:26 +0000 (10:00 -0700)]
Fix gdbserver-without-gdb build
An earlier patch changed gdbserver to use the already-built top-level
gnulib and gdbsupport. However, if one did a build that did not
include gdb, then gdbserver would fail to build.
The problem is that configure.ac only adds gnulib and gdbsupport to
the build when gdb is being built. This patch fixes the problem by
arranging for this to happen when gdbserver is built.
ChangeLog
2020-02-17 Tom Tromey <tom@tromey.com>
* configure: Rebuild.
* configure.ac (configdirs): Add gnulib and gdbsupport when building
gdbserver.
H.J. Lu [Mon, 17 Feb 2020 15:12:10 +0000 (07:12 -0800)]
x86: Remove CpuABM and add CpuPOPCNT
AMD ABM has 2 instructions: popcnt and lzcnt. ABM CPUID feature bit has
been reused for lzcnt and a POPCNT CPUID feature bit is added for popcnt
which used to be the part of SSE4.2. This patch removes CpuABM and adds
CpuPOPCNT. It changes ABM to enable both lzcnt and popcnt, changes SSE4.2
to also enable popcnt.
gas/
* config/tc-i386.c (cpu_arch): Add .popcnt.
* doc/c-i386.texi: Remove abm and .abm. Add popcnt and .popcnt.
Add a tab before @samp{.sse4a}.
opcodes/
* i386-gen.c (cpu_flag_init): Replace CpuABM with
CpuLZCNT|CpuPOPCNT. Add CpuPOPCNT to CPU_SSE4_2_FLAGS. Add
CPU_POPCNT_FLAGS.
(cpu_flags): Remove CpuABM. Add CpuPOPCNT.
* i386-opc.h (CpuABM): Removed.
(CpuPOPCNT): New.
(i386_cpu_flags): Remove cpuabm. Add cpupopcnt.
* i386-opc.tbl: Replace CpuABM|CpuSSE4_2 with CpuPOPCNT on
popcnt. Remove CpuABM from lzcnt.
* i386-init.h: Regenerated.
* i386-tbl.h: Likewise.
now will match in behavior with the AVX512 counterparts in that not
only the 2nd vcvtsi2sd won't assemble, but also the first. The last
two, otoh, will continue to assemble fine (due to the lack of any
memory operand size specifier). As a result, another way to make
things behave more consistently would be to avoid the folding and
add IgnoreSize to the CpuNo64 AVX512 variants. A 3rd way to do so
would be to add Cpu386 to any such insn template.
While doing this also make the usual cosmetic adjustments for the
insns touched anyway. Additionally drop the redundant Cpu64 from
the SAE forms of VCVT{,U}SI2SD - they won't assemble outside of
64-bit mode due to there not being anything to match the Reg64
operand.
Jan Beulich [Mon, 17 Feb 2020 07:59:07 +0000 (08:59 +0100)]
x86: fold AddrPrefixOpReg templates
There's no need to have separate Cpu64 and CpuNo64 templates: There
already is special logic handling the attribute, and all that's needed
is rejecting 16-bit address registers in 64-bit mode. Suppress suffix
guessing and group all involved logic together, outside of suffix
processing (arguably it doesn't even belong in process_suffix()).
Also, since no AddrPrefixOpReg template permits any suffixes, move the
No_*Suf specifiers for them to a central place. Along with this drop
the no longer relevant NoRex64 from there.
Jan Beulich [Mon, 17 Feb 2020 07:57:54 +0000 (08:57 +0100)]
x86/Intel: don't swap operands of MONITOR{,X} and MWAIT{,X}
Generally, the documentation doesn't allow for any explicit operands
to be specified with MONITOR/MWAIT. To permit the more legible
overriding of the address size via specifying operands, the option is
being retained even in Intel mode, but operand swapping is being
suppressed by this patch. This is both because it makes no sense here
(all of the operands are inputs) and because, as a result, old gcc
(prior to 4.8) actually expects it this way with -mintel-syntax (and
hence gets fixed by this change rather than, as claimed by a reply in
the bug report, broken).
Jan Beulich [Mon, 17 Feb 2020 07:56:18 +0000 (08:56 +0100)]
x86/Intel: improve diagnostics for ambiguous VCVT* operands
Conversions which shrink element size and which have a memory source
can't be disambiguated between their 128- and 256-bit variants by
looking at the register operand. "operand size mismatch", however, is a
pretty misleading diagnostic. Generalize the logic introduced for
VFPCLASSP{S,D} such that, with suitable similar adjustments to the
respective templates, it'll cover these cases too.
For VCVTNEPS2BF16 also fold the two previously separate AVX512VL
templates to achieve the intended effect. This is then also accompanied
by a respective addition to the inval-avx512f testcase.
H.J. Lu [Sun, 16 Feb 2020 16:36:51 +0000 (08:36 -0800)]
x86: Don't disable SSE4a when disabling SSE4
commit 7deea9aad8 changed nosse4 to include CpuSSE4a. But AMD SSE4a is
a superset of SSE3 and Intel SSE4 is a superset of SSSE3. Disable Intel
SSE4 shouldn't disable AMD SSE4a. This patch restores nosse4. It also
adds .sse4a and nosse4a.
gas/
* config/tc-i386.c (cpu_arch): Add .sse4a and nosse4a. Restore
nosse4.
* doc/c-i386.texi: Document sse4a and nosse4a.
Simon Marchi [Fri, 14 Feb 2020 21:45:40 +0000 (16:45 -0500)]
gdb: introduce displaced_step_closure_up type alias
To help with readability, add the type displaced_step_closure_up, an
alias for std::unique_ptr<displaced_step_closure>, and use it throughout
the code base.
gdb/ChangeLog:
* aarch64-tdep.c (aarch64_displaced_step_copy_insn): Use
displaced_step_closure_up.
* aarch64-tdep.h (aarch64_displaced_step_copy_insn): Likewise.
(struct displaced_step_closure_up):
* amd64-tdep.c (amd64_displaced_step_copy_insn): Likewise.
* amd64-tdep.h (amd64_displaced_step_copy_insn): Likewise.
* arm-linux-tdep.c (arm_linux_displaced_step_copy_insn):
Likewise.
* gdbarch.sh (displaced_step_copy_insn): Likewise.
* gdbarch.c, gdbarch.h: Re-generate.
* i386-linux-tdep.c (i386_linux_displaced_step_copy_insn): Use
displaced_step_closure_up.
* i386-tdep.c (i386_displaced_step_copy_insn): Likewise.
* i386-tdep.h (i386_displaced_step_copy_insn): Likewise.
* infrun.h (displaced_step_closure_up): New type alias.
(struct displaced_step_inferior_state) <step_closure>: Change
type to displaced_step_closure_up.
* rs6000-tdep.c (ppc_displaced_step_copy_insn): Use
displaced_step_closure_up.
* s390-tdep.c (s390_displaced_step_copy_insn): Likewise.
Tom Tromey [Fri, 14 Feb 2020 21:34:20 +0000 (14:34 -0700)]
Change gdbserver to use existing gnulib and libiberty
This changes gdbserver so that it no longer builds its own gnulib and
libiberty. Instead, it now relies on the ones that were already built
at the top level.
gdbsupport is still built specially for gdbserver. This is more
complicated and will be tackled in a subsequent patch.
ChangeLog
2020-02-14 Tom Tromey <tom@tromey.com>
* Makefile.in: Rebuild.
* Makefile.def: Make gdbserver require gnulib and libiberty.
gdbserver/ChangeLog
2020-02-14 Tom Tromey <tom@tromey.com>
Tom Tromey [Fri, 14 Feb 2020 21:14:38 +0000 (14:14 -0700)]
Have testsuite find gdbserver in new location
This updates the gdb testsuite to look for gdbserver in its new
location. The old location is also checked for, on the theory that
perhaps someone sets GDB to a full path for install testing.
gdb/testsuite/ChangeLog
2020-02-14 Tom Tromey <tom@tromey.com>
* lib/gdbserver-support.exp (find_gdbserver): Find gdbserver in
build directory.
* boards/gdbserver-base.exp: Update path to gdbserver.
Simon Marchi [Fri, 14 Feb 2020 20:29:08 +0000 (15:29 -0500)]
gdb: make gdbarch_displaced_step_copy_insn return an std::unique_ptr
This callback dynamically allocates a specialized displaced_step_closure, and
gives the ownership of the object to its caller. So I think it would make
sense for the callback to return an std::unique_ptr, this is what this patch
implements.
Simon Marchi [Fri, 14 Feb 2020 20:11:58 +0000 (15:11 -0500)]
gdb: cleanup of displaced_step_inferior_state::reset/displaced_step_clear
displaced_step_inferior_state::reset and displaced_step_clear appear to
have the same goal, but they don't do the same thing.
displaced_step_inferior_state::reset clears more things than
displaced_step_clear, but it misses free'ing the closure, which
displaced_step_clear does.
This patch replaces displaced_step_clear's implementation with just a call to
displaced_step_inferior_state::reset. It then changes
displaced_step_inferior_state::step_closure to be a unique_ptr, to indicate the
fact that displaced_step_inferior_state owns the closure (and so that it is
automatically freed when the field is reset).
The test gdb.base/step-over-syscall.exp caught a problem when doing this, which
I consider to be a latent bug which my cleanup exposes. In
handle_inferior_event, in the TARGET_WAITKIND_FORKED case, if we displaced-step
over a fork syscall, we make sure to restore the memory that we used as a
displaced-stepping buffer in the child. We do so using the
displaced_step_inferior_state of the parent. However, we do it after calling
displaced_step_fixup for the parent, which clears the information in the
parent's displaced_step_inferior_state. It worked fine before, because
displaced_step_clear didn't completely clear the displaced_step_inferior_state
structure, so the required information (in this case the gdbarch) was
still available after clearing.
I fixed it by making GDB restore the child's memory before calling the
displaced_step_fixup on the parent. This way, the data in the
displaced_step_inferior_state structure is still valid when we use it for the
child. This is the error you would get in
gdb.base/step-over-syscall.exp without this fix:
Simon Marchi [Fri, 14 Feb 2020 19:41:07 +0000 (14:41 -0500)]
gnulib: import count-one-bits module and use it
For a fix I intend to submit, I would need a function that counts the
number of set bits in a word. There is __builtin_popcount that is
supported by gcc and clang, but there is also a gnulib module that wraps
that and provides a fallback for other compilers, so I think it would be
good to use it.
I also noticed that there is a bitcount function in arch/arm.c, so I
thought that as a first step I would replace that one with the gnulib
count-one-bits module. This is what this patch does.
The gnulib module provides multiple functions, with various parameter
length (unsigned int, unsigned long int, unsigned long long int), I
chose the one that made sense for each call site based on the argument
type.
* arm-tdep.c: Include count-one-bits.h.
(cleanup_block_store_pc): Use count_one_bits.
(cleanup_block_load_pc): Use count_one_bits.
(arm_copy_block_xfer): Use count_one_bits.
(thumb2_copy_block_xfer): Use count_one_bits.
(thumb_copy_pop_pc_16bit): Use count_one_bits.
* arch/arm-get-next-pcs.c: Include count-one-bits.h.
(thumb_get_next_pcs_raw): Use count_one_bits.
(arm_get_next_pcs_raw): Use count_one_bits_l.
* arch/arm.c (bitcount): Remove.
* arch/arm.h (bitcount): Remove.
Jan Beulich [Fri, 14 Feb 2020 13:27:28 +0000 (14:27 +0100)]
x86: replace adhoc (partly wrong) ambiguous operand checking for MOVSX/MOVZX
For these to get treatment consistent with other operand size checking
the special logic shouldn't live in md_assemble(), but process_suffix().
And there's more logic involved than simply zapping the suffix.
Note however that MOVS[BW]* and MOVZ[BW]* still won't be fully
consistent, due to the objection to fold MOVS* templates just like was
done for MOVZ* in c07315e0c6 ("x86: allow suffix-less movzw and 64-bit
movzb").
Note further that it is against my own intentions to have MOVSX/MOVZX
silently default to a byte source in AT&T mode. This should happen only
when the destination register is a 16-bit one. In all other cases there
is an ambiguity, and the user should be warned. But it was explicitly
requested for this to be done in a way inconsistent with everything
else.
Note finally that the assembler change points out (and this patch fixes)
a wrong Intel syntax test introduced by bc31405ebb2c ("x86-64: Properly
encode and decode movsxd"): When source code specifies a 16-bit
destination register, disassembly expectations shouldn't have been to
find a 32-bit one.
Jan Beulich [Fri, 14 Feb 2020 13:04:23 +0000 (14:04 +0100)]
x86: adjust segment override prefix emission
Since we already suppress the prefix altogether when it's the default
one for the chosen addressing mode, let's do so also when instruction
prefix and override specified with the memory operand match. (Note that
insn prefix specified segment overrides never get discarded.)
Jan Beulich [Fri, 14 Feb 2020 13:02:05 +0000 (14:02 +0100)]
x86: extend LEA's segment override warning
For one both possible forms should be warned about. And then, to guard
against future surprises, qualify the original opcode check by excluding
VEX/EVEX-like templates.
Eli Zaretskii [Fri, 14 Feb 2020 09:53:55 +0000 (11:53 +0200)]
Fix argv[] in programs invoked by gdbserver on MS-Windows
gdbserver/ChangeLog
2020-02-14 Eli Zaretskii <eliz@gnu.org>
* win32-low.c (create_process): Prepend PROGRAM to ARGS when
preparing the command line for CreateProcess.
(win32_create_inferior): Reflect the program name in debugging
output that shows the process and its command line.
I see for some program at gdb startup:
...
Samples: 102K of event 'cycles:pu', Event count (approx.): 91710925103
Overhead Command Shared Object Symbol
15.21% gdb gdb [.]
lnp_state_machine::handle_special
...
where the divisions are the places we stall. The following
micro-optimizes things but it smells like m_line_header->line_range
is constant, likewise probably m_line_header->maximum_ops_per_instruction
so eventually the divisions could be avoided completely with some
lookup table.
Well. Micro-optimizing with this patch improves things
(don't expect [load] CSE over the gdbarch_adjust_dwarf2_line call).
Build and reg-tested on x86_64-linux.
gdb/ChangeLog:
2020-02-14 Richard Biener <rguenther@suse.de>
* dwarf2/read.c (lnp_state_machine::handle_special_opcode): Apply CSE
on expression with division operators.
H.J. Lu [Thu, 13 Feb 2020 21:44:17 +0000 (13:44 -0800)]
x86: Resolve PLT32 reloc aganst local symbol to section
Since PLT entry isn't needed for branch to local symbol, we can resolve
R_386_PLT32/R_X86_64_PLT32 relocation aganst local symbol to section,
similar to R_386_PC32/R_X86_64_PC32.
2020-02-13 Fangrui Song <maskray@google.com>
H.J. Lu <hongjiu.lu@intel.com>
PR gas/25551
* config/tc-i386.c (tc_i386_fix_adjustable): Don't check
BFD_RELOC_386_PLT32 nor BFD_RELOC_X86_64_PLT32.
* testsuite/gas/i386/i386.exp: Run relax-5 and x86-64-relax-4.
* testsuite/gas/i386/relax-5.d: New file.
* testsuite/gas/i386/relax-5.s: Likewise.
* testsuite/gas/i386/x86-64-relax-4.d: Likewise.
* testsuite/gas/i386/x86-64-relax-4.s: Likewise.
Simon Marchi [Thu, 13 Feb 2020 21:27:02 +0000 (16:27 -0500)]
gdbsupport: rename source files to .cc
This patch renames the .c source files in gdbsupport to .cc.
In the gdb directory, there is an argument against renaming the source
files, which is that it makes using some git commands more difficult to
do archeology. Some commands have some kind of "follow" option that
makes git try to follow renames, but it doesn't work in all situations.
Given that we have just moved the gdbsupport directory, that argument
doesn't hold for source files in that directory. I therefore suggest
renaming them to .cc, so that they are automatically recognized as C++
by various tools and editors.
The original motivation behind this is that when building gdbsupport
with clang, I get:
CC agent.o
clang: error: treating 'c' input as 'c++' when in C++ mode, this behavior is deprecated [-Werror,-Wdeprecated]
In the gdb/ directory, we make clang happy by passing "-x c++". We
could do this in gdbsupport too, but I think that renaming the files is
a better long-term solution.
gdbserver still does its own build of gdbsupport, so a few changes in
its Makefile are necessary.
gdbsupport/ChangeLog:
* Makefile.am: Rename source files from .c to .cc.
(CC, CFLAGS): Don't override.
(AM_CFLAGS): Rename to ...
(AM_CXXFLAGS): ... this.
* Makefile.in: Re-generate.
* %.c: Rename to %.cc.
gdbserver/ChangeLog:
* Makefile.in: Rename gdbsupport source files from .c to .cc.
Tom de Vries [Thu, 13 Feb 2020 14:42:07 +0000 (15:42 +0100)]
[gdb/testsuite] Remove stale exec in gdb_compile_ada
When running test-case gdb.ada/ptype_tagged_param.exp, I get:
...
PASS: gdb.ada/ptype_tagged_param.exp: compilation foo.adb
...
However, when I then de-install gnatmake and run again, I get the same result.
This is due to a stale exec. After removing the stale exec, I get:
...
UNSUPPORTED: gdb.ada/ptype_tagged_param.exp: compilation foo.adb
...
Fix this removing the stale exec in gdb_compile_ada before compilation.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-02-13 Tom de Vries <tdevries@suse.de>
* lib/ada.exp (gdb_compile_ada): Delete stale exec before compilation.
projects / thirdparty / binutils-gdb.git / log
