Ivan Maidanski [Thu, 4 Apr 2019 21:34:04 +0000 (00:34 +0300)]
Update documentation about the incremental collector support
(a cherry-pick of commit 5d4fd4de from 'master')
* doc/README.darwin (Important Usage Notes): Remove outdated note about
flaky incremental mode.
* doc/README.darwin (Implementation Information): Remove information
about old incremental collector implementation.
* doc/README.win32 (Watcom compiler): Refine information about the
incremental mode support.
* doc/scale.md (Options for enhanced scalability): Refine when the
dedicated marker threads are created.
Ivan Maidanski [Tue, 26 Feb 2019 22:32:53 +0000 (01:32 +0300)]
Avoid potential race in hb_sz access between realloc and reclaim_block
Issue #240 (bdwgc).
GC_realloc might be changing the block size while GC_reclaim_block
is examining it. The change to the size field is benign, i.e.
GC_reclaim would work correctly with either value, since we are not
changing the number of objects in the block. But seeing a half-updated
value (though unlikely to occur in practice) could be probably bad.
Using unordered atomic fetch of hb_sz field should solve the issue.
* reclaim.c (GC_block_nearly_full, GC_reclaim_small_nonempty_block):
Add sz argument; use sz instead of hhdr->hb_sz.
* reclaim.c (GC_reclaim_clear, GC_reclaim_uninit, GC_reclaim_check):
Skip the assertion about hhdr->hb_sz if THREADS.
* reclaim.c [ENABLE_DISCLAIM] (GC_disclaim_and_reclaim): Likewise.
* reclaim.c [AO_HAVE_load] (GC_reclaim_block): Use AO_load to access
hhdr->hb_sz; add comments.
* reclaim.c (GC_reclaim_block): Pass sz to
GC_reclaim_small_nonempty_block() and GC_block_nearly_full().
* reclaim.c (GC_continue_reclaim, GC_reclaim_all): Pass hhdr->hb_sz to
GC_reclaim_small_nonempty_block().
* reclaim.c [!EAGER_SWEEP && ENABLE_DISCLAIM]
(GC_reclaim_unconditionally_marked): Likewise.
Ivan Maidanski [Tue, 26 Feb 2019 21:48:12 +0000 (00:48 +0300)]
Fix 'duplicate symbol' error for tests using multiple static libs (OS X)
Issue #263 (bdwgc).
If configured with --enable-static, libtool passes libgc.a to gcc
twice (with a relative path and with an absolute one) when linking
tests that depend on libcord.a, libgccpp.a or libstaticrootslib_test.a
which, in turn, depends on libgc.a. Double specification of libgc.a
seems to confuse ld tool of MacOS 10.14.
The workaround is to omit libgc.la in *_LDADD specification of the
mentioned tests when the shared libraries are not requested, at least.
* configure.ac (ENABLE_SHARED): New AM_CONDITIONAL.
* cord/cord.am (cordtest_LDADD): Do not add libgc.la unless
ENABLE_SHARED; add comment.
* tests/tests.am (staticrootstest_LDADD): Likewise.
* tests/tests.am [CPLUSPLUS && !AVOID_CPP_LIB] (test_cpp_LDADD):
Likewise.
Jonas Echterhoff [Thu, 21 Feb 2019 10:49:16 +0000 (11:49 +0100)]
Avoid a full GC when growing finalizer tables if in incremental mode
(a cherry-pick of commits 6e0590ef1, c23bad10a from 'master')
Issue Unity-Technologies#37 (bdwgc).
* finalize.c (GC_grow_table): Do not call GC_try_to_collect_inner
if GC_incremental; add comment.
* tests/test_cpp.cc (C::Test): Call GC_gcollect() if incremental mode
and nFreed is too low.
Ivan Maidanski [Fri, 15 Feb 2019 20:40:49 +0000 (23:40 +0300)]
Avoid unexpected heap growth in gctest caused by GC_disable
(fix of commit 8fc1ebf6d)
Issue #264 (bdwgc).
This commit eliminates the scenario when one thread disables garbage
collections while other threads intensively allocate memory objects.
* tests/test.c (run_one_test): Do not call GC_disable and GC_enable.
* tests/test.c (run_single_threaded_test): New function (which tests
GC_disable and GC_enable).
* tests/test.c (main): Call run_single_threaded_test() (before
check_heap_stats).
Ivan Maidanski [Thu, 14 Feb 2019 19:55:20 +0000 (22:55 +0300)]
Fix thread_info() count argument value (OS X)
* darwin_stop_world.c [DEBUG_THREADS || GC_ASSERTIONS]
(GC_thread_resume): Set outCount to THREAD_BASIC_INFO_COUNT (instead
of THREAD_INFO_MAX) to match the size of info passed to thread_info().
Hamayama [Thu, 7 Feb 2019 21:51:25 +0000 (00:51 +0300)]
Fix GetThreadContext stale register values use if WoW64 (Win32)
Issue #262 (bdwgc).
* misc.c [MSWIN32 && !MSWINRT_FLAVOR && !MSWIN_XBOX1]
(GC_win32_MessageBoxA): Do not define unless SMALL_CONFIG.
* misc.c [MSWIN32 && !_WIN64 && GC_WIN32_THREADS && CHECK_NOT_WOW64]
(GC_init): Do not call IsWow64Process() and GC_win32_MessageBoxA().
* win32_threads.c [I386] (isWow64): New static variable.
* win32_threads.c [I386] (GC_push_stack_for): If isWow64 then set also
CONTEXT_EXCEPTION_REQUEST and CONTEXT_SEGMENTS bits in ContextFlags;
if isWow64, and CONTEXT_EXCEPTION_REPORTING and
CONTEXT_EXCEPTION_ACTIVE are set on return from GetThreadContext then
call GetThreadSelectorEntry and use StackLimit of FS selector to set
sp local variable (instead of context.Esp); add comment.
* win32_threads.c [I386 && DEBUG_THREADS] (GC_push_stack_for): Call
GC_log_printf() to report TIB stack limit/base and the case when
CONTEXT_EXCEPTION_REQUEST is not supported.
* win32_threads.c [I386] (GC_thr_init): Set isWow64 by IsWow64Process()
if the later is available.
Ivan Maidanski [Mon, 21 Jan 2019 22:01:45 +0000 (01:01 +0300)]
Fix 'unexpected mark stack overflow' abort in push_all_stack
Issue #260 (bdwgc).
* mark.c [!NEED_FIXUP_POINTER] (GC_push_all_stack): Call
GC_push_all_eager() instead of GC_push_all() if GC_mark_stack_top is
rather close to GC_mark_stack_limit.
Ivan Maidanski [Fri, 4 Jan 2019 16:37:50 +0000 (19:37 +0300)]
Fix BSD_TIME variant of MS_TIME_DIFF for the case of a.tv_usec < b.tv_usec
(fix of commit 6b2796b69)
* include/private/gc_priv.h [!NO_CLOCK && BSD_TIME] (MS_TIME_DIFF):
Cast the result of a.tv_usec-b.tv_usec to long (so that the result of
the division should be negative if a.tv_usec is less than b.tv_usec).
Ivan Maidanski [Sat, 5 Jan 2019 08:52:53 +0000 (11:52 +0300)]
Fix invalid initializer of CLOCK_TYPE variables if BSD_TIME
(fix of commits 7e59301c1, 06ecaefca)
* alloc.c [!NO_CLOCK] (GC_start_time): Initialize to
CLOCK_TYPE_INITIALIZER (instead of 0).
* alloc.c [!NO_CLOCK] (GC_try_to_collect_inner, GC_stopped_mark,
GC_finish_collection): Initialize local variable of CLOCK_TYPE type to
CLOCK_TYPE_INITIALIZER (instead of 0); remove comment (duplicating that
in gc_priv.h).
* reclaim.c [!NO_CLOCK] (GC_reclaim_all): Likewise.
* include/private/gc_priv.h [!NO_CLOCK] (CLOCK_TYPE_INITIALIZER):
Define macro.
Ivan Maidanski [Thu, 20 Dec 2018 08:34:12 +0000 (11:34 +0300)]
Allow to force executable pages allocation in gctest
Pass -D TEST_PAGES_EXECUTABLE to CFLAGS to enable the executable pages
allocation in gctest even if the collector is compiled with
NO_EXECUTE_PERMISSION.
Ivan Maidanski [Thu, 6 Dec 2018 08:23:52 +0000 (11:23 +0300)]
Eliminate 'assigned value never used' compiler warning in test_cpp WinMain
* tests/test_cpp.cc [(MSWIN32 && !__MINGW32__ || MSWINCE)
&& !NO_WINMAIN_ENTRY] (WinMain): Remove first argv[argc]=cmd (when cmd
points to a space or tab character).
Ivan Maidanski [Wed, 28 Nov 2018 09:09:59 +0000 (12:09 +0300)]
Travis CI: Pass extra/gc.c to cppcheck --enable=unusedFunction
(a cherry-pick of commit 53aa42b3 from 'master')
This fixes a dozen of 'the function is never used' cppcheck style
warnings (for GC_core_gcj_malloc, GC_durango_get_mem,
GC_FreeBSDGetDataStart, GC_haiku_get_mem, GC_mprotect_resume,
GC_mprotect_stop, GC_remove_roots_subregion, GC_win32_get_mem,
GC_wince_get_mem).
Ivan Maidanski [Thu, 29 Nov 2018 08:46:48 +0000 (11:46 +0300)]
Fix GC_register_disclaim_proc for leak-finding mode
Issue #252 (bdwgc).
This makes the behavior of GC_register_disclaim_proc() and
GC_finalized_malloc() somewhat consistent with
GC_register_disappearing_link() and GC_register_finalizer() when
find-leak is on. The documentation is updated accordingly.
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_finalized_disclaim): Add assertion
that GC_find_leak is off.
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_register_disclaim_proc): Do not
assign ok_disclaim_proc, ok_mark_unconditionally fields if GC_find_leak.
* include/gc_disclaim.h (GC_register_disclaim_proc,
GC_finalized_malloc): Refine comment about leak-find mode and GC_free
invocation.
Ivan Maidanski [Thu, 22 Nov 2018 11:34:03 +0000 (14:34 +0300)]
Workaround 'nonlocal var will use ptr to local var' cppcheck false positive
* extra/msvc_dbg.c (sym_namebuf_u): New union type.
* extra/msvc_dbg.c (GetSymbolNameFromAddress): Use sym_namebuf_u
instead of anonymous union.
* os_dep.c [MPROTECT_VDB && DARWIN] (mp_reply_s, mp_msg_s): Move struct
type definition out of GC_mprotect_thread (adding mp_ prefix).
Ivan Maidanski [Fri, 9 Nov 2018 17:08:20 +0000 (20:08 +0300)]
Eliminate 'uninitialized var' cppcheck false positive in mach_dep, os_dep
* mach_dep.c (GC_with_callee_saves_pushed): Change type of "context"
local variable from void* volatile to volatile ptr_t.
* os_dep.c [(SVR4 || AIX || DGUX || (LINUX && SPARC)) && !PCR]
(GC_SysVGetDataStart): Change type of "result" local variable from
char* volatile to volatile ptr_t.