Ivan Maidanski [Sun, 23 Dec 2012 12:13:45 +0000 (16:13 +0400)]
Fix dwSize argument of VirtualFree call in detect_GetWriteWatch (Win32)
* os_dep.c (detect_GetWriteWatch): Pass 0 (instead of GC_page_size)
to Win32 VirtualFree if GWW_VDB (dwSize must be 0 if MEM_RELEASE
according to MSDN).
Ivan Maidanski [Sun, 25 Nov 2012 18:35:06 +0000 (22:35 +0400)]
Fix GC_CreateThread 'dwStackSize' argument type for Win64
(Apply commit b9260aa from 'master' branch.)
* include/gc.h (GC_uintptr_t): Define (for Win32/CE) even if
GC_NO_THREAD_DECLS; move definition upper than GC_CreateThread.
* include/gc.h (GC_WIN32_SIZE_T): New macro (defined as GC_uintptr_t,
Win32 threads only).
* include/gc.h (GC_CreateThread): Use GC_WIN32_SIZE_T instead of DWORD
type for "dwStackSize" argument (Win32 threads only).
* win32_threads.c (GC_CreateThread): Likewise.
* tests/test.c (fork_a_thread): Verify types of GC_CreateThread
arguments by explicit casts (only if GC_WIN32_THREADS).
Ivan Maidanski [Sun, 18 Nov 2012 19:20:23 +0000 (23:20 +0400)]
Fix GC_marker_Id elements initialization (WinCE)
(Apply commit 33bd64c from 'master' branch.)
* win32_threads.c (GC_marker_cv, GC_marker_Id): Move variable
definition up to precede GC_mark_thread (only if PARALLEL_MARK and
DONT_USE_SIGNALANDWAIT).
* win32_threads.c (GC_mark_thread): Set corresponding element of
GC_marker_Id[] to GetCurrentThreadId() value (only if PARALLEL_MARK
and DONT_USE_SIGNALANDWAIT).
* win32_threads.c (start_mark_threads): Do not set GC_marker_Id[]
elements here because GetCurrentThreadId should be invoked from the
corresponding thread; update comment (only if PARALLEL_MARK and
DONT_USE_SIGNALANDWAIT).
Ivan Maidanski [Sun, 11 Nov 2012 14:57:29 +0000 (18:57 +0400)]
Fix Win32 GC_write preventing potential infinite recursion at abort
(Apply commit 608fe46 from 'master' branch.)
* misc.c (GC_write): Define "inside_write" static variable (only
for Win32 and only if THREADS and GC_ASSERTIONS or
GC_PRINT_VERBOSE_STATS); if this variable is set then just return,
set this variable if GC_write_disabled or GC_log is
INVALID_HANDLE_VALUE just before ABORT call (to prevent infinite
recursion).
* misc.c (GC_abort): Do not call WRITE if GC_write_disabled
(only for Win32 and only if THREADS and GC_ASSERTIONS) to prevent
abort in GC_write.
Ivan Maidanski [Tue, 6 Nov 2012 03:13:29 +0000 (07:13 +0400)]
Fix unit in comment for some counters in gc_priv.h
* include/private/gc_priv.h (_GC_arrays): Fix comment for
_bytes_allocd_before_gc, _bytes_allocd, _composite_in_use and
_atomic_in_use fields (replace "words" unit with "bytes").
* include/private/gc_priv.h (GC_objfreelist): Fix comment (replace
"words" unit with "bytes").
Ivan Maidanski [Wed, 24 Oct 2012 04:29:42 +0000 (08:29 +0400)]
Fix potential double fclose in test_extras (cordtest)
* cord/tests/cordtest.c (test_extras): Explicitly clear cord handles
obtained from CORD_from_file(), invoke GC_gcollect and
GC_invoke_finalizers to force f1a, f1b, f2 handles to be closed before
removing the files; remove the second attempts to remove the files on
rename failure (print a warning instead) as it might cause double
fclose (or double free); add comments.
Ivan Maidanski [Tue, 23 Oct 2012 03:45:51 +0000 (07:45 +0400)]
Fix GC_clear_stack by declaring 'dummy' local array as volatile
* misc.c (GC_clear_stack): Declare "dummy" local array as volatile to
prevent optimizing it out by compiler; add cast of "dummy" to void*
(to un-volatile the pointer) in BZERO call (only if THREADS).
Ivan Maidanski [Fri, 19 Oct 2012 05:10:07 +0000 (09:10 +0400)]
Fix GC_unix_mmap_get_mem for open of /dev/zero failure
(Apply commit 2c62153 from 'master' branch.)
* os_dep.c (GC_unix_mmap_get_mem): Check open("/dev/zero") (or open
of similar file on Symbian) result and abort (with the appropriate
message) on failure (only if USE_MMAP_ANON is undefined).
Ivan Maidanski [Wed, 17 Oct 2012 18:07:54 +0000 (22:07 +0400)]
Fix vsprintf_args cleanup in CORD_vsprintf
* cord/cordprnt.c (CORD_vsprintf): Invoke va_end (before return) for
vsprintf_args initialized by [__]va_copy (only if __va_copy defined
or GCC but not DJGPP).
Ivan Maidanski [Tue, 11 Sep 2012 04:30:25 +0000 (08:30 +0400)]
Fix min_bytes_allocd preventing potential infinite loop in GC_allocobj
* alloc.c (min_bytes_allocd): Do not return zero in case of big
GC_free_space_divisor value (return 1 instead to prevent infinite loop
in GC_allocobj if GC_adj_bytes_allocd returns zero); update comment.
Ivan Maidanski [Thu, 30 Aug 2012 04:39:25 +0000 (08:39 +0400)]
.gitignore: remove path prefix for config.h and stamp-h1
* .gitignore: Add include/config.h.in (to ignore); remove path for
config.h and stamp-h1 (to ignore files both located in /include/private
(old behavior) and in /include folders).
Ivan Maidanski [Wed, 1 Aug 2012 22:02:08 +0000 (02:02 +0400)]
Fix GC_clear_stack_inner by using GC_approx_sp
* misc.c (GC_clear_stack_inner): Use GC_approx_sp() instead of
"dummy[0]" set to "&dummy" value (that worked as expected only for
STACK_GROWS_DOWN case).
Ivan Maidanski [Thu, 2 Aug 2012 03:42:43 +0000 (07:42 +0400)]
Fix all address-of-dummy operations by using GC_approx_sp() instead
(a joint commit of commits 'd6acbd' and '57b94a3' in master branch)
* alloc.c (min_bytes_allocd, GC_stopped_mark): Use GC_approx_sp()
instead of "&dummy"; remove 'dummy' local variable.
* dyn_load.c (GC_cond_add_roots): Likewise.
* misc.c (GC_init): Likewise.
* os_dep.c (GC_get_stack_base, GC_get_main_stack_base): Likewise.
* pthread_stop_world.c (GC_suspend_handler_inner,
nacl_pre_syscall_hook, __nacl_suspend_thread_if_needed): Likewise.
* pthread_support.c (GC_thr_init): Likewise.
* ptr_chck.c (GC_on_stack): Likewise.
* win32_threads.c (GC_push_stack_for): Likewise.
* extra/setjmp_t.c (main): Define volatile 'sp' local variable, store
its address to it and use it instead of "&dummy"; remove 'dummy' local
variable.
* mach_dep.c (GC_with_callee_saves_pushed): Use volatile for 'dummy'
local variable.
* misc.c (GC_clear_stack_inner): Store address of volatile 'dummy'
local array (i.e. 'sp' value) to its first element (and use it in the
comparison of addresses) to prevent any harmful optimizations as C
officially disallows comparisons of pointers to different objects
(e.g., some Mac OS X clang releases might turn a conditional
expression that uses 'dummy' address into a constant); update comment.
* misc.c (GC_call_with_stack_base): Use "&base" instead of "&dummy"
(it is safe to use address of base here); remove dummy variable.
Ivan Maidanski [Fri, 20 Jul 2012 20:36:34 +0000 (00:36 +0400)]
Fix stop_info.stack_ptr assignment in GC_suspend_all for OpenBSD
(commit '7ff92dc' from master branch)
* os_dep.c (GC_get_stack_base): Abort if pthread_stackseg_np fails
(if GC_OPENBSD_THREADS).
* pthread_stop_world.c (GC_suspend_all): Get correct stack_ptr by
calling pthread_stackseg_np (subtracting ss_size from ss_sp) instead
of retrieving it from OpenBSD pthread implementation-dependent context
(if GC_OPENBSD_THREADS); remove comment.
Ivan Maidanski [Sun, 17 Jun 2012 16:25:09 +0000 (20:25 +0400)]
Fix thread-related tests for pthreads-w32
* tests/initsecondarythread.c: Include "private/config.h" if
HAVE_CONFIG_H (mostly to have GC_WIN32_PTHREADS defined for
pthreads-w32 target).
* tests/thread_leak_test.c: Likewise.
* tests/threadkey_test.c: Likewise.
Ivan Maidanski [Sat, 16 Jun 2012 08:34:06 +0000 (12:34 +0400)]
Fix test_cpp (ensure the collector recognizes pointers to interiors)
* tests/test_cpp.cc (main): Call GC_set_all_interior_pointers(1)
before GC_INIT to ensure that the collector considers pointers to
object interiors as valid ones (such a pointer could emerge as
a result of a type cast to subclass in case of multiple inheritance);
add comment.
Tsugutomo Enami [Thu, 31 May 2012 23:12:24 +0000 (03:12 +0400)]
Fix GC_FirstDLOpenedLinkMap() for NetBSD 6 release
* dyn_load.c: Include sys/param.h and dlfcn.h on NetBSD.
* dyn_load.c (GC_FirstDLOpenedLinkMap): Obtain link map using dlinfo()
on NetBSD if RTLD_DI_LINKMAP feature present (defined).
Ivan Maidanski [Sat, 2 Jun 2012 08:33:03 +0000 (12:33 +0400)]
Fix visibility of GC_push_all/conditional, GC_push_other_roots symbols
* include/private/gc_priv.h (GC_push_all, GC_push_other_roots):
Declare as GC_API_PRIV (instead of GC_INNER) to make the symbol
externally visible to some well-known 3rd-party software (e.g., ECL).
* include/private/gc_priv.h (GC_push_conditional): Declare as
GC_API_PRIV (only if GC_DISABLE_INCREMENTAL is undefined).
* mark.c (GC_push_all, GC_push_conditional): Remove GC_INNER (to match
the declaration).
* os_dep.c (GC_push_other_roots): Likewise.
Ivan Maidanski [Sat, 2 Jun 2012 07:13:19 +0000 (11:13 +0400)]
Fix visibility of GC_clear/set_mark_bit (unhide symbols)
* include/private/gc_priv.h (GC_clear_mark_bit, GC_set_mark_bit):
Declare as GC_API_PRIV (instead of GC_INNER) to make the symbol
externally visible to 3rd-party software (e.g., ECL).
* mark.c (GC_set_mark_bit, GC_clear_mark_bit): Remove GC_INNER (to
match the declaration).
Ivan Maidanski [Wed, 16 May 2012 04:34:10 +0000 (08:34 +0400)]
Fix assertion in GC_malloc_[atomic_]uncollectable (THREADS case only)
* malloc.c (GC_malloc_uncollectable): Do not check assertion on
hb_n_marks to be zero if multi-threaded; add comment.
* mallocx.c (GC_malloc_atomic_uncollectable): Likewise.
* alloc.c (GC_check_fl_marks): Re-read each pointer atomically before
following the pointed-to link and bail out if the result is different
(this can happen if the thread has popped the object off the
free-list); the function is a no-op if AO_load is unavailable.
Ivan Maidanski [Fri, 20 Apr 2012 05:12:47 +0000 (09:12 +0400)]
Code refactoring of GC_check_tls_for/GC_check_fl_marks
(Apply commit e67ab08 from 'master' branch)
* alloc.c (GC_check_fl_marks): Change prototype (pass pointer to
"freelist" element instead of value); do not define the function if
THREAD_LOCAL_ALLOC undefined.
* include/private/gc_priv.h (GC_check_fl_marks): Likewise.
* alloc.c (GC_check_fl_marks): Skip check if the argument points to
a special (non-pointer) value; update comment; rename "q" local
variable to "list".
* thread_local_alloc.c (GC_check_tls_for): Update code according to
GC_check_fl_marks functionality change (remove checks for special
value).
Fix GC_with_callee_saves_pushed for clang (disable __builtin_unwind_init)
* include/private/gcconfig.h (HAVE_BUILTIN_UNWIND_INIT): Do not define
for clang (as __builtin_unwind_init seems to be a no-op in the latest
clang release).
Ivan Maidanski [Mon, 2 Apr 2012 18:31:22 +0000 (22:31 +0400)]
FIX: Disable incremental mode on Darwin if fork handling requested
* doc/README.macros (NO_HANDLE_FORK): Replace back with HANDLE_FORK
and update.
* include/private/gcconfig.h (CAN_HANDLE_FORK): Define also for Darwin
in case of MPROTECT_VDB.
* misc.c (GC_set_handle_fork): Handle Darwin properly; add comment.
* os_dep.c (GC_dirty_init): Do not turn on GC_dirty_maintained for
Darwin if GC_handle_fork is on; add comment; remove FIXME.
* pthread_support.c (GC_fork_prepare_proc): Remove WARN for Darwin
(since GC_dirty_maintained is off in that case).
* pthread_support.c (GC_fork_child_proc): Remove comment (for Darwin).
Ivan Maidanski [Mon, 2 Apr 2012 16:18:12 +0000 (20:18 +0400)]
Add public GC_set_handle_fork to control forked child handling support
(Apply commit 1e882b9 from 'master' branch)
* include/gc.h (GC_set_handle_fork): New API function.
* misc.c (GC_set_handle_fork): Likewise.
* include/private/gc_priv.h (GC_handle_fork): New internal variable
declaration (only if CAN_HANDLE_FORK).
* misc.c (GC_handle_fork): New internal variable (defined only if
CAN_HANDLE_FORK); initialize to TRUE if HANDLE_FORK.
* include/private/gcconfig.h (HANDLE_FORK): Replace with
CAN_HANDLE_FORK.
* pthread_support.c (HANDLE_FORK): Likewise.
* win32_threads.c (HANDLE_FORK): Likewise.
* include/private/gcconfig.h (CAN_HANDLE_FORK): Always define macro if
HANDLE_FORK.
* pthread_support.c (GC_thr_init): Replace HANDLE_FORK with
CAN_HANDLE_FORK; call pthread_atfork only if GC_handle_fork; update
the comment.
* win32_threads.c (GC_thr_init): Likewise.
* tests/test.c (NO_TEST_HANDLE_FORK): Define new macro if fork
handling is not supported (or is a no-op) on the target.
* tests/test.c (INIT_FORK_SUPPORT): New macro (invoke
GC_set_handle_fork unless NO_TEST_HANDLE_FORK).
* tests/test.c (GC_OPT_INIT): New macro (defined to GC_INIT or empty).
* tests/test.c (GC_COND_INIT): Use INIT_FORK_SUPPORT and GC_OPT_INIT.
* tests/test.c (run_one_test): Test NO_TEST_HANDLE_FORK (instead of
target-specific macros).
* win32_threads.c (GC_remove_all_threads_but_me, GC_fork_prepare_proc,
GC_fork_parent_proc, GC_fork_child_proc): Do not test GC_PTHREADS.
* configure.ac (HANDLE_FORK, NO_HANDLE_FORK): Update message.
Ivan Maidanski [Fri, 30 Mar 2012 03:53:06 +0000 (07:53 +0400)]
Remove redundant public GC_HAVE_GET_HEAP_USAGE_SAFE macro
* include/gc.h (GC_HAVE_GET_HEAP_USAGE_SAFE): Remove public HAVE_
macro (since function availability could be determined by autotools);
update the comment.
Andy Wingo [Tue, 27 Mar 2012 11:20:37 +0000 (13:20 +0200)]
configure: add --enable-handle-fork option
* configure.ac: Add --enable-handle-fork option; if explicitly passed,
enables or disables the pthread_atfork support; if not passed, the
default, as determined by gcconfig.h, is to enable HANDLE_FORK if it
is supported on the target platform.
Ivan Maidanski [Sun, 25 Mar 2012 06:32:05 +0000 (10:32 +0400)]
Fix visibility of some GC internal symbols used by GCJ
(Includes reverting part of commits 68b9f27, 91d5df5, b2345fb)
* allchblk.c (GC_hblkfreelist, GC_free_bytes): Do not define array as
STATIC again but only if GC_GCJ_SUPPORT (for GNU GCJ client).
* blacklst.c (GC_is_black_listed): Do not define as GC_INNER again
(for GNU GCJ client).
* include/private/gc_priv.h (GC_is_black_listed): Likewise.
* include/private/gc_priv.h (GC_hblkfreelist): Restore removed
declaration (but only if GC_GCJ_SUPPORT).
* reclaim.c (GC_n_set_marks): Do not define as STATIC again (for GCJ);
update the comment.
Ivan Maidanski [Fri, 16 Mar 2012 15:11:08 +0000 (19:11 +0400)]
Fix GC_ALPHA_VERSION comparison in gctest and tools files
(Apply commit 3b6bc1e from 'master' branch)
* tests/test.c (GC_ALPHA_VERSION): Define to GC_TMP_ALPHA_VERSION if
not defined by gc_version.h.
* extra/add_gc_prefix.c (GC_ALPHA_VERSION): Likewise.
* extra/gcname.c (GC_ALPHA_VERSION): Likewise.
Hans Boehm [Thu, 15 Mar 2012 17:09:05 +0000 (21:09 +0400)]
Speedup calloc size overflow check by preventing division if small values
* malloc.c (GC_SQRT_SIZE_MAX): New macro.
* malloc.c (calloc): Add fast initial size overflow check to avoid
integer division for reasonably small values passed.