binutils-gdb/gdbsupport/common-defs.h
Tom de Vries 9a56fae565 [gdb/build] Use -fno-hoist-adjacent-loads for gcc <= 13
When building gdb with gcc 12 and -fsanitize=threads while renabling
background dwarf reading by setting dwarf_synchronous to false, I run into:
...
(gdb) file amd64-watchpoint-downgrade
Reading symbols from amd64-watchpoint-downgrade...
(gdb) watch global_var
==================
WARNING: ThreadSanitizer: data race (pid=20124)
  Read of size 8 at 0x7b80000500d8 by main thread:
    #0 cooked_index_entry::full_name(obstack*, bool) const cooked-index.c:220
    #1 cooked_index::get_main_name(obstack*, language*) const cooked-index.c:735
    #2 cooked_index_worker::wait(cooked_state, bool) cooked-index.c:559
    #3 cooked_index::wait(cooked_state, bool) cooked-index.c:631
    #4 cooked_index_functions::wait(objfile*, bool) cooked-index.h:729
    #5 cooked_index_functions::compute_main_name(objfile*) cooked-index.h:806
    #6 objfile::compute_main_name() symfile-debug.c:461
    #7 find_main_name symtab.c:6503
    #8 main_language() symtab.c:6608
    #9 set_initial_language_callback symfile.c:1634
    #10 get_current_language() language.c:96
    ...

  Previous write of size 8 at 0x7b80000500d8 by thread T1:
    #0 cooked_index_shard::finalize(parent_map_map const*) \
         dwarf2/cooked-index.c:409
    #1 operator() cooked-index.c:663
    ...

  ...

SUMMARY: ThreadSanitizer: data race cooked-index.c:220 in \
  cooked_index_entry::full_name(obstack*, bool) const
==================
Hardware watchpoint 1: global_var
(gdb) PASS: gdb.arch/amd64-watchpoint-downgrade.exp: watch global_var
...

This was also reported in PR31715.

This is due do gcc PR110799 [1], generating wrong code with
-fhoist-adjacent-loads, and causing a false positive for
-fsanitize=threads.

Work around the gcc PR by forcing -fno-hoist-adjacent-loads for gcc <= 13
and -fsanitize=threads.

Tested in that same configuration on x86_64-linux.  Remaining ThreadSanitizer
problems are the ones reported in PR31626 (gdb.rust/dwindex.exp) and
PR32247 (gdb.trace/basic-libipa.exp).

PR gdb/31715
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31715

Tested-By: Bernd Edlinger <bernd.edlinger@hotmail.de>
Approved-By: Tom Tromey <tom@tromey.com>

[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110799
2024-10-18 00:18:24 +02:00

228 lines
7.2 KiB
C

/* Common definitions.
Copyright (C) 1986-2024 Free Software Foundation, Inc.
This file is part of GDB.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>. */
#ifndef COMMON_COMMON_DEFS_H
#define COMMON_COMMON_DEFS_H
#if defined (__SANITIZE_THREAD__) && defined (__GNUC__) \
&& !defined (__clang__) && __GNUC__ <= 13
/* Work around PR gcc/110799. */
#pragma GCC optimize("-fno-hoist-adjacent-loads")
#endif
#include <gdbsupport/config.h>
#undef PACKAGE_NAME
#undef PACKAGE
#undef PACKAGE_VERSION
#undef PACKAGE_STRING
#undef PACKAGE_TARNAME
#include "gnulib/config.h"
/* From:
https://www.gnu.org/software/gnulib/manual/html_node/stdint_002eh.html
"On some hosts that predate C++11, when using C++ one must define
__STDC_CONSTANT_MACROS to make visible the definitions of constant
macros such as INTMAX_C, and one must define __STDC_LIMIT_MACROS to
make visible the definitions of limit macros such as INTMAX_MAX.".
And:
https://www.gnu.org/software/gnulib/manual/html_node/inttypes_002eh.html
"On some hosts that predate C++11, when using C++ one must define
__STDC_FORMAT_MACROS to make visible the declarations of format
macros such as PRIdMAX."
Must do this before including any system header, since other system
headers may include stdint.h/inttypes.h. */
#define __STDC_CONSTANT_MACROS 1
#define __STDC_LIMIT_MACROS 1
#define __STDC_FORMAT_MACROS 1
/* Some distros enable _FORTIFY_SOURCE by default, which on occasion
has caused build failures with -Wunused-result when a patch is
developed on a distro that does not enable _FORTIFY_SOURCE. We
enable it here in order to try to catch these problems earlier;
plus this seems like a reasonable safety measure. The check for
optimization is required because _FORTIFY_SOURCE only works when
optimization is enabled. If _FORTIFY_SOURCE is already defined,
then we don't do anything. Also, on MinGW, fortify requires
linking to -lssp, and to avoid the hassle of checking for
that and linking to it statically, we just don't define
_FORTIFY_SOURCE there. */
#if (!defined _FORTIFY_SOURCE && defined __OPTIMIZE__ && __OPTIMIZE__ > 0 \
&& !defined(__MINGW32__))
#define _FORTIFY_SOURCE 2
#endif
/* We don't support Windows versions before XP, so we define
_WIN32_WINNT correspondingly to ensure the Windows API headers
expose the required symbols.
NOTE: this must be kept in sync with common.m4. */
#if defined (__MINGW32__) || defined (__CYGWIN__)
# ifdef _WIN32_WINNT
# if _WIN32_WINNT < 0x0501
# undef _WIN32_WINNT
# define _WIN32_WINNT 0x0501
# endif
# else
# define _WIN32_WINNT 0x0501
# endif
#endif /* __MINGW32__ || __CYGWIN__ */
#include <stdarg.h>
#include <stdio.h>
/* Include both cstdlib and stdlib.h to ensure we have standard functions
defined both in the std:: namespace and in the global namespace. */
#include <cstdlib>
#include <stdlib.h>
#include <stddef.h>
#include <stdint.h>
#include <string.h>
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif
#include <errno.h>
#if HAVE_ALLOCA_H
#include <alloca.h>
#endif
#include "ansidecl.h"
/* This is defined by ansidecl.h, but we prefer gnulib's version. On
MinGW, gnulib might enable __USE_MINGW_ANSI_STDIO, which may or not
require use of attribute gnu_printf instead of printf. gnulib
checks that at configure time. Since _GL_ATTRIBUTE_FORMAT_PRINTF_STANDARD
is compatible with ATTRIBUTE_PRINTF, simply use it. */
#undef ATTRIBUTE_PRINTF
#define ATTRIBUTE_PRINTF _GL_ATTRIBUTE_FORMAT_PRINTF_STANDARD
/* This is defined by ansidecl.h, but we disable the attribute.
Say a developer starts out with:
...
extern void foo (void *ptr) __attribute__((nonnull (1)));
void foo (void *ptr) {}
...
with the idea in mind to catch:
...
foo (nullptr);
...
at compile time with -Werror=nonnull, and then adds:
...
void foo (void *ptr) {
+ gdb_assert (ptr != nullptr);
}
...
to catch:
...
foo (variable_with_nullptr_value);
...
at runtime as well.
Said developer then verifies that the assert works (using -O0), and commits
the code.
Some other developer then checks out the code and accidentally writes some
variant of:
...
foo (variable_with_nullptr_value);
...
and builds with -O2, and ... the assert doesn't trigger, because it's
optimized away by gcc.
There's no suppported recipe to prevent the assertion from being optimized
away (other than: build with -O0, or remove the nonnull attribute). Note
that -fno-delete-null-pointer-checks does not help. A patch was submitted
to improve gcc documentation to point this out more clearly (
https://gcc.gnu.org/pipermail/gcc-patches/2021-July/576218.html ). The
patch also mentions a possible workaround that obfuscates the pointer
using:
...
void foo (void *ptr) {
+ asm ("" : "+r"(ptr));
gdb_assert (ptr != nullptr);
}
...
but that still requires the developer to manually add this in all cases
where that's necessary.
A warning was added to detect the situation: -Wnonnull-compare, which does
help in detecting those cases, but each new gcc release may indicate a new
batch of locations that needs fixing, which means we've added a maintenance
burden.
We could try to deal with the problem more proactively by introducing a
gdb_assert variant like:
...
void gdb_assert_non_null (void *ptr) {
asm ("" : "+r"(ptr));
gdb_assert (ptr != nullptr);
}
void foo (void *ptr) {
gdb_assert_nonnull (ptr);
}
...
and make it a coding style to use it everywhere, but again, maintenance
burden.
With all these things considered, for now we go with the solution with the
least maintenance burden: disable the attribute, such that we reliably deal
with it everywhere. */
#undef ATTRIBUTE_NONNULL
#define ATTRIBUTE_NONNULL(m)
#define ATTRIBUTE_UNUSED_RESULT __attribute__ ((__warn_unused_result__))
#define ATTRIBUTE_USED __attribute__ ((__used__))
#include "libiberty.h"
#include "pathmax.h"
#include "gdb/signals.h"
#include "gdb_locale.h"
#include "ptid.h"
#include "common-types.h"
#include "common-utils.h"
#include "gdb_assert.h"
#include "errors.h"
#include "print-utils.h"
#include "common-debug.h"
#include "cleanups.h"
#include "common-exceptions.h"
#include "gdbsupport/poison.h"
/* Pull in gdb::unique_xmalloc_ptr. */
#include "gdbsupport/gdb_unique_ptr.h"
/* sbrk on macOS is not useful for our purposes, since sbrk(0) always
returns the same value. brk/sbrk on macOS is just an emulation
that always returns a pointer to a 4MB section reserved for
that. */
#if defined (HAVE_SBRK) && !__APPLE__
#define HAVE_USEFUL_SBRK 1
#endif
#endif /* COMMON_COMMON_DEFS_H */