Tom Tromey [Wed, 27 Sep 2017 16:53:02 +0000 (10:53 -0600)]
Change exceptions.h functions to use gdb::function_view
This changes some functions in exceptions.h to use gdb::function_view,
then fixes up the fallout. This lead to some surprising places, like
a function pointer in target_so_ops.
While writing this I found that catch_exception_ftype was unused, so I
removed this.
Note that I did not compile the windows-nat.c change, so I don't know
if it works.
Regression tested by the buildbot.
gdb/ChangeLog
2017-09-27 Tom Tromey <tom@tromey.com>
Many tdep files need to perform conversions between two floating-point
types, usually when accessing FP registers. Most targets now use the
convert_typed_floating helper routine to do so. However, a small number
still use the old method of converting via a DOUBLEST. Since we want
to get rid of DOUBLEST, these targets need to be moved to the new
method as well.
The main obstacle is that for convert_typed_floating we need an actual
*type*, not just a floatformat.
In arm-tdep.c, this is very straightforward, since there is already a
type using the ARM extended floatformat.
For sh-tdep.c and sh64-tdep.c, no such type already exists, so I've
added one to the gdbarch_tdep struct as done on other targets.
For historical reasons, the TYPE_FLOATFORMAT element is still set to hold
an array of two floatformat structs, one for big-endian and the other for
little-endian. When accessing the element via floatformat_from_type,
the code would check the type's byte order and return the appropriate
floatformat.
However, these days this is quite unnecessary, since the type's byte order
is already known at the time the type is allocated and the floatformat is
installed into TYPE_FLOATFORMAT. Therefore, we can just install the correct
version here.
Also, moves the (now trivially simple) floatformat_from_type accessor to
gdbtypes.{c,h}, since it doesn't really need to be in doublest.c now.
* doublest.h (floatformat_from_type): Move to gdbtypes.h.
* doublest.c (floatformat_from_type): Move to gdbtypes.c.
* gdbtypes.h (union type_specific): Make field floatformat hold
just a single struct floatformat, not an array.
(floatformat_from_type): Move here.
* gdbtypes.c (floatformat_from_type): Move here. Update to
changed TYPE_FLOATFORMAT definition.
(verify_floatformat): Update to changed TYPE_FLOATFORMAT.
(recursive_dump_type): Likewise.
(init_float_type): Install correct floatformat for byte order.
(arch_float_type): Likewise.
This changes the interfaces to init_type and arch_type to take the
type length in bits as input (instead of as bytes). The routines
assert that the length is a multiple of TARGET_CHAR_BIT.
For consistency, arch_flags_type is changed likewise, so that now
all type creation interfaces always use length in bits.
All callers are updated in the straightforward manner.
The assert actually found a bug in read_range_type, where the
init_integer_type routine was called with a wrong argument (probably
a bug introduced with the conversion to use init_integer_type).
Nick Clifton [Wed, 27 Sep 2017 15:21:36 +0000 (16:21 +0100)]
Add support for the new names of the RISC-V fmv.x.s and fmv.s.x instructions, vis: fmv.x.w and fmv.w.x.
PR 22179
opcodes * riscv-opc.c (riscv_opcodes): Add fmv.x.w and fmv.w.x as the new
names for the fmv.x.s and fmv.s.x instructions respectively.
gas * testsuite/gas/riscv/fmv.x.s: New file: Tests the support for the
renamed fmv.x.s and fmv.s.x instructions.
* testsuite/gas/riscv/fmv.x.d: New file: Test driver.
Tom Tromey [Sun, 10 Sep 2017 02:51:33 +0000 (20:51 -0600)]
Add add_cmd function overloads
This adds two add_cmd overloads: one whose callback takes a const char *,
and one that doesn't accept a function at all. The no-function overload
was introduced to avoid ambiguity when NULL was passed as the function.
Long term the goal is for all commands to take const arguments, and
for the non-const variants to be removed entirely.
gdb/ChangeLog
2017-09-27 Tom Tromey <tom@tromey.com>
* cli/cli-decode.c (add_cmd, set_cmd_cfunc): New function
overloads.
(do_add_cmd): Rename from add_cmd. Don't call set_cmd_cfunc.
(do_const_cfunc): New function.
(cmd_cfunc_eq): New overload.
(cli_user_command_p): Check do_const_cfunc.
* cli/cli-decode.h (struct cmd_list_element) <function>: New field
const_cfunc.
* command.h (add_cmd): Add const overload and no-function
overload.
(set_cmd_cfunc): Add const overload.
(cmd_const_cfunc_ftype): Declare.
(cmd_cfunc_eq): Add const overload.
* breakpoint.c, cli-cmds.c, cli-dump.c, guile/scm-cmd.c,
python/py-cmd.c, target.c, tracepoint.c: Use no-function add_cmd
overload.
Tom Tromey [Mon, 18 Sep 2017 02:36:41 +0000 (20:36 -0600)]
Add support for __VA_OPT__
C++2a adds a "__VA_OPT__" feature that can be used to control the
pesky "," emission when the final (variable) argument of a variadic
macro is empty. This patch implements this feature for gdb. (A patch
to implement it for gcc is pending.)
gdb/ChangeLog
2017-09-27 Tom Tromey <tom@tromey.com>
* macroexp.c (get_next_token_for_substitution): New function.
(substitute_args): Call it. Check for __VA_OPT__.
gdb/testsuite/ChangeLog
2017-09-27 Tom Tromey <tom@tromey.com>
Alan Modra [Wed, 27 Sep 2017 05:44:00 +0000 (15:14 +0930)]
PR22216, infinite loop in readelf process_symbol_table
This should make readelf bombproof given a fuzzed DT_HASH. Also
removes a bogus check that would have resulted in wrong histograms.
PR 22216
* readelf.c (process_symbol_table): Check that DT_HASH symbol
chains are only visited once, and report an error if not. Display
invalid symbol index if chain is out of range. Use the same logic
when calculating histograms rather than the PR 17531 fix. Delete
bogus check that chained index is less than number of buckets.
GDB has a workaround for DWARF output by ICC, related to missing
DW_AT_declaration on incomplete types. The bug was fixed in ICC 14,
so this commit adjusts GDB accordingly.
For the version check, this adds a new parser function for the ICC
producer string. While at it, it also adds unit tests for the
producer parsing covering the new function and preexisting parsers.
gdb/ChangeLog:
2017-09-26 Walfred Tedeschi <walfred.tedeschi@intel.com>
Pedro Alves <palves@redhat.com>
* dwarf2read.c (dwarf2_cu): Remove field producer_is_icc and add
producer_is_icc_lt_14.
(producer_is_icc_lt_14): New function.
(check_producer): Add code for checking version of ICC.
(producer_is_icc): Move to producer.c.
(read_structure_type): Restrict ICC workaround to ICC<14.
* producer.c: Include selftest.h.
(producer_is_icc, producer_parsing_tests, _initialize_producer):
New functions.
* producer.h (producer_is_icc): New declaration.
* gdbthread.h: Include "common-gdbthread.h".
* inferiors.c (switch_to_thread): Use "gdb_assert" instead of
"if" when validating the ptid.
* remote-utils.c: Include "gdbthread.h".
(prepare_resume_reply): Use "switch_to_thread".
* target.c (done_accessing_memory): Likewise.
Nick Clifton [Tue, 26 Sep 2017 15:03:10 +0000 (16:03 +0100)]
Fix an assertion failure when parsing a fuzzed x86_64 ELF binary.
PR 22172
* elf64-x86-64.c (elf_x86_64_info_to_howto): Do not trigger an
assertion failure if elf_x86_64_rtype_to_howto has already issued
an error message.
and the last gdb version that can be compiled with gcc 4.9 is 7.12.1 only when
configured with --disable-build-with-cxx, it's time to obsolete/remove support
for Solaris < 10.
This patch does this, simplifying configure.nat along the way (only a single
sol2 configuration with variants for i386 and sparc).
Some configure checks for older Solaris versions can go, too, and the check
for libthread_db.so.1 removed:
* Since Solaris 10, dlopen has moved to libc and libdl.so is just a
filter on ld.so.1, so no need to check.
* $RDYNAMIC is already handled above (and is a no-op with Solaris ld
anyway).
Both proc-service.c and sol-thread.c lose support for (Solaris-only)
PROC_SERVICE_IS_OLD.
The attached revised patch has been tested on sparcv9-sun-solaris2.10,
sparcv9-sun-solaris2.11.4, amd64-pc-solaris2.10, amd64-pc-solaris2.11.4,
and x86_64-pc-linux-gnu.
I've also started an i386-pc-solaris2.9 build to check that it really
stops as expected.
PR gdb/22185
* configure.host <*-*-solaris2.[01], *-*-solaris2.[2-9]*>: Mark as
obsolete.
Use gdb_host sol2 for i[34567]86-*-solaris2*, x86_64-*-solaris2*.
Remove i386sol2 support.
* configure.nat <i386sol2>: Remove.
<sol2-64>: Fold into ...
<sol2>: ... this.
Move common settings to default section.
Add sol-thread.o.
* configure.tgt <i[34567]86-*-solaris2.1[0-9]*,
x86_64-*-solaris2.1[0-9]*>: Rename to ...
<i[34567]86-*-solaris2*, x86_64-*-solaris2*>: ... this.
<i[34567]86-*-solaris*>: Remove.
<sparc-*-solaris2.[0-6], sparc-*-solaris2.[0-6].*>: Remove.
* configure.ac: Remove wctype in libw check.
(_MSE_INT_H): Don't define on Solaris 7-9.
<solaris*>: Remove libthread_db.so.1 check.
* configure: Regenerate.
* config.in: Regenerate.
* proc-service.c: Remove PROC_SERVICE_IS_OLD handling.
(gdb_ps_prochandle_t, gdb_ps_read_buf_t, gdb_ps_write_buf_t)
(gdb_ps_size_t): Remove.
Use base types in users.
* sol-thread.c: Likewise, also for gdb_ps_addr_t.
* NEWS (Changes since GDB 8.0): Document Solaris 2.0-9 removal.
When testing my Solaris < 10 removal patch on Solaris/SPARC, I found
that gdb mainline is currently broken there due to the recent SPARC M7
ADI patches:
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:1876:0: error: "PSR_ICC" redefined [-Werror]
#define PSR_ICC 0x00f00000
^
In file included from /usr/include/v7/sys/privregs.h:24:0,
from /usr/include/sys/regset.h:420,
from /usr/include/sys/ucontext.h:21,
from /usr/include/sys/signal.h:231,
from /usr/include/sys/procset.h:23,
from /usr/include/sys/wait.h:25,
from /usr/include/stdlib.h:21,
from build-gnulib/import/stdlib.h:36,
from /vol/src/gnu/gdb/gdb/local/gdb/common/common-defs.h:53,
from /vol/src/gnu/gdb/gdb/local/gdb/defs.h:28,
from /vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:20:
/usr/include/v7/sys/psr.h:35:0: note: this is the location of the previous definition
#define PSR_ICC 0x00F00000 /* integer condition codes */
^
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:1878:0: error: "PSR_IMPL" redefined [-Werror]
#define PSR_IMPL 0xf0000000
^
In file included from /usr/include/v7/sys/privregs.h:24:0,
from /usr/include/sys/regset.h:420,
from /usr/include/sys/ucontext.h:21,
from /usr/include/sys/signal.h:231,
from /usr/include/sys/procset.h:23,
from /usr/include/sys/wait.h:25,
from /usr/include/stdlib.h:21,
from build-gnulib/import/stdlib.h:36,
from /vol/src/gnu/gdb/gdb/local/gdb/common/common-defs.h:53,
from /vol/src/gnu/gdb/gdb/local/gdb/defs.h:28,
from /vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:20:
/usr/include/v7/sys/psr.h:41:0: note: this is the location of the previous definition
#define PSR_IMPL 0xF0000000 /* implementation */
^
Comparing Solaris 11.4 <v7/sys/psr.h> and sparc64-tdep.c, there are more
inconsistencies:
Apart from the capitalization differences that trip g++, the names
differ (PSR_VER vs. PSR_VERS), PSR_XCC is included in Solaris' PSR_RSV,
and there's no PSR_V8PLUS on Solaris either.
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c: In function `int adi_tag_fd()':
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:296:63: error: format `%d' expects argument of type `int', but argument 4 has type `pid_t {aka long int}' [-Werror=format=]
snprintf (cl_name, sizeof(cl_name), "/proc/%d/adi/tags", pid);
^
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c: In function `bool adi_is_addr_mapped(CORE_ADDR, std::size_t)':
/vol/src/gnu/gdb/gdb/local/gdb/sparc64-tdep.c:314:64: error: format `%d' expects argument of type `int', but argument 4 has type `pid_t {aka long int}' [-Werror=format=]
snprintf (filename, sizeof filename, "/proc/%d/adi/maps", pid);
^
You cannot always print a pid_t, which can be either int or long on
Solaris, as an int.
Obviously, the ADI patch which modifies code shared between all SPARC
targets, hasn't been tested on anything but Linux/SPARC.
The patch below includes the minimal fixes necessary to unbreak the
Solaris/SPARC build.
However, as detailed in the PR, there's more breakage here: apart from
not bothering to implement ADI support on Solaris, the code contains
several more changes to shared/common SPARC code that are simply wrong
on anything but Linux/SPARC.
The patch was tested on sparcv9-sun-solaris2.10 and
sparcv9-sun-solaris2.11.4 (build and gdb/gdb gdb/gdb smoke test only).
Alan Modra [Tue, 26 Sep 2017 12:17:24 +0000 (21:47 +0930)]
PR22209, invalid memory read in find_abstract_instance_name
This patch adds bounds checking for DW_FORM_ref_addr die refs, and
calculates them relative to the first .debug_info section. See the
big comment for why calculating relative to the current .debug_info
section was wrong for relocatable object files.
PR 22209
* dwarf2.c (struct comp_unit): Delete sec_info_ptr field.
(find_abstract_instance_name): Calculate DW_FORM_ref_addr relative
to stash->info_ptr_memory, and check die_ref is within that memory.
Set info_ptr_end correctly when another CU is refd. Check die_ref
for DW_FORM_ref4 etc. is within CU.
LD/testsuite: Fix undefined symbol tests for MIPS and S+core targets
Fix commit 647d4de92e06 ("Test undefined symbols in shared libraries")
and remove "undefined symbols in shared lib (dyn reloc)" test failures
for MIPS and S+core targets.
This test assumes 3 dynamic relocation entries, however the respective
psABIs define an extra single dummy *_NONE relocation at the beginning
of the dynamic relocation table, which results in test output like the
following for MIPS:
compare-sections.exp has two cases that are not handled appropriately:
1) value read with msb set
2) error while patching that section
This patch adapts the "get value of read-only section" test to print
the value as an unsigned integer to fix 1) and test for the error
message to not set the written variable if read-only section cannot
be written to so as to solve 2).
2017-09-26 Thomas Preud'homme <thomas.preudhomme@arm.com>
Pedro Alves <palves@redhat.com>
gdb/testsuite/
* gdb.base/compare-sections.exp (get value of read-only section): Read
as unsigned value.
(corrupt read-only section): Likewise and don't set written if patching
failed.
Alan Modra [Tue, 26 Sep 2017 01:58:04 +0000 (11:28 +0930)]
PR21732, String-concatenated message can not be translated
Preprocessor tokens like DWARF_VMA_FMT and BFD_VMA_FMT are not
understood by gettext, and cause the translation string to be
truncated. Even if it were not truncated, we wouldn't want a value
that changes depending on host or configure options to be put in a
translation string.
PR 21732
* dwarf.c (READ_ULEB): Don't use DWARF_VMA_FMT in translated string.
(READ_SLEB): Likewise.
* readelf.c (bfd_vmatoa): New function.
(get_data, get_dynamic_data): Use it for printing error messages.
projects / thirdparty / binutils-gdb.git / log
