Ivan Maidanski [Thu, 15 Sep 2016 15:40:21 +0000 (18:40 +0300)]
Fix calloc_explicitly_typed in case of lb*n overflow
* typd_mlc.c: Include limits.h (for SIZE_MAX).
* typd_mlc.c (GC_SIZE_MAX, GC_SQRT_SIZE_MAX): New macro (same as in
malloc.c).
* typd_mlc.c (GC_calloc_explicitly_typed): Return NULL if lb * n
overflows (same algorithm as in calloc defined in malloc.c); eliminate
lb *= n code duplication.
Ivan Maidanski [Thu, 15 Sep 2016 06:47:23 +0000 (09:47 +0300)]
Fix double multiplication of lb by n in calloc_explicitly_typed
* typd_mlc.c (GC_calloc_explicitly_typed): Do not multiply lb by n
twice (when passed to GC_malloc in case of
GC_general_register_disappearing_link fails).
Ivan Maidanski [Wed, 14 Sep 2016 22:03:15 +0000 (01:03 +0300)]
Eliminate 'constructor with 1 argument is not explicit' cppcheck warning
Note that -D CPPCHECK should be passed to cppcheck to activate
this workaround.
* include/gc_allocator.h (GC_ATTR_EXPLICIT): New macro (defined to
"explicit" keyword if at least C++11 or CPPCHECK, otherwise to empty).
* include/gc_allocator.h (gc_allocator::gc_allocator,
gc_allocator_ignore_off_page::gc_allocator_ignore_off_page,
traceable_allocator::traceable_allocator): Use GC_ATTR_EXPLICIT.
* tests/test_cpp.cc (A::A, B::B, C::C, D::D): Likewise.
* tests/test_cpp.cc (GC_ATTR_EXPLICIT): Define macro (as empty) unless
already defined in included gc_allocator.h.
Ivan Maidanski [Wed, 14 Sep 2016 06:21:02 +0000 (09:21 +0300)]
Workaround 'suspicious pointer subtraction' cppcheck warning in gc_cpp
* include/gc_cpp.h (gc_cleanup::gc_cleanup): Add extra parentheses
around (char*)this to outline that "this" pointer subtraction is
intentional (i.e., "this->" was not intended).
Ivan Maidanski [Mon, 12 Sep 2016 18:55:25 +0000 (21:55 +0300)]
Workaround 'va_list used before va_start' cppcheck error in cord
Note that -D CPPCHECK should be passed to cppcheck to activate
this workaround.
* cord/cordprnt.c (CORD_vsprintf) [CPPCHECK]: Force to use va_copy
and va_end.
* cord/cordprnt.c (CORD_vsprintf): Set res to -1 if invalid format
specifier (instead of immediate return -1); call va_end at a single
place.
Ivan Maidanski [Mon, 29 Aug 2016 12:58:00 +0000 (15:58 +0300)]
Eliminate 'checking if unsigned variable is <0' cppcheck style warning
* cord/cordbscs.c (CORD_from_fn, CORD_substr): Do not expect size_t
value to be negative (replace <=0 comparison with ==0 one).
* cord/cordxtra.c (CORD_cmp): Likewise.
* cord/cordbscs.c (CORD_substr): Remove obsolete comment (about SunOS 4
which had signed size_t type).
Ivan Maidanski [Thu, 18 Aug 2016 22:43:25 +0000 (01:43 +0300)]
Eliminate 'cast to void* from int' compiler warnings (Darwin/x64)
* darwin_stop_world.c (GC_stack_range_for, GC_suspend_thread_list,
GC_stop_world, GC_thread_resume, GC_start_world): Cast thread variable
to pointer via word type.
* darwin_stop_world.c (GC_stop_world): Cast stop_info.mach_thread to
pointer via word type.
Ivan Maidanski [Thu, 11 Aug 2016 19:55:21 +0000 (22:55 +0300)]
Improve Travis CI configuration (use both compilers, enable more features)
* .travis.yml (compiler): Add (use both gcc and clang)
* .travis.yml (install): Use autogen.sh instead of autoreconf and
automake; add options to configure (to cover more code during testing);
do not invoke make.
* .travis.yml (script): Pass -j to make.
Ivan Maidanski [Thu, 11 Aug 2016 08:56:44 +0000 (11:56 +0300)]
Workaround missing getcontext() in Docker osrf/ubuntu_32bit
* mach_dep.c [NO_GETCONTEXT] (GC_with_callee_saves_pushed): Call WARN
instead of ABORT if getcontext() failed; do not set context variable
if getcontext() failed; fallback to other register retrieval methods
(__builtin_unwind_init or setjmp) if context variable is NULL.
* mach_dep.c (GC_with_callee_saves_pushed): Reformat code.
Ivan Maidanski [Mon, 8 Aug 2016 18:49:20 +0000 (21:49 +0300)]
Fix 'unknown type name GC_INNER' compilation error (FreeBSD)
(fix commit f13a9559)
See issue #132. The reason of compilation failure is GC_INNER macro
used before its definition. The solution is to move
GC_FreeBSDGetDataStart prototype from gcconfig.h to gc_priv.h.
Ivan Maidanski [Tue, 2 Aug 2016 09:01:48 +0000 (12:01 +0300)]
Do not allow SHORT_DBG_HDRS if KEEP_BACK_PTRS or MAKE_BACK_GRAPH
Otherwise GC_HAS_DEBUG_INFO (defined as (p&1)) might return true
if a non-pointer is stored at the beginning of the tested object
leading further to its corruption by GC_store_back_pointer.
See issue #125 for details.
* include/private/dbg_mlc.h [KEEP_BACK_PTRS || MAKE_BACK_GRAPH]
(GC_HAS_DEBUG_INFO): Add #error (with the appropriate message) in
case of SHORT_DBG_HDRS defined.
Ivan Maidanski [Thu, 28 Jul 2016 09:06:42 +0000 (12:06 +0300)]
Fix tag collision between ENABLE_DISCLAIM and KEEP_BACK_PTRS
* fnlz_mlc.c (FINALIZER_CLOSURE_FLAG): New macro (defined to 0x2 in
case KEEP_BACK_PTRS or MAKE_BACK_GRAPH, otherwise to 0x1).
* fnlz_mlc.c (GC_finalized_disclaim, GC_finalized_malloc): Use
FINALIZER_CLOSURE_FLAG instead of 0x1.
Process all PT_LOAD segments before PT_GNU_RELRO segments (Glibc)
This is needed in case a PT_GNU_RELRO segment is placed before its
corresponding PT_LOAD segments. (It might be that this is guaranteed
never to be the case, but at least this way we do not have to worry
about it.)
* dyn_load.c [HAVE_DL_ITERATE_PHDR] (GC_register_dynlib_callback):
Process PT_LOAD segments in a separate pass before PT_GNU_RELRO ones;
replace FIXME with TODO; reformat code.
Handle load_segs overflow in register_dynlib_callback gracefully
* dyn_load.c [HAVE_DL_ITERATE_PHDR and PT_GNU_RELRO]
(GC_register_dynlib_callback): If n_load_segs reaches MAX_LOAD_SEGS
then call WARN (with the appropriate message) and call
GC_add_roots_inner to register the segment directly instead of ABORT.
Do not warn of missing PT_GNU_RELRO segment when custom DSO filter used
It is most likely that it is not found just because the segment had
been excluded.
Alternatively, we could have registered all segments, and checked the
callback afterwards, but then we could break programs that rely on
GC_has_static_roots_func to avoid overflowing the maximum number of
roots. In addition, it would make the logic slightly more
complicated, probably without a very good reason since the chance of
this warning to show without the segment being excluded is likely to
be none.
* dyn_load.c [HAVE_DL_ITERATE_PHDR] (GC_register_dynlib_callback):
Do not call WARN() if GC_has_static_roots callback is set.
Enable thread-local storage usage for GC_malloc/calloc_explicitly_typed
Use GC_malloc_kind instead of explicit allocation algorithm
implementation.
* typd_mlc.c: Include gc_inline.h to declare GC_malloc_kind.
* typd_mlc.c (GC_arobjfreelist): Remove.
* typd_mlc.c (GC_init_explicit_typing): Do not use GC_arobjfreelist.
* typd_mlc.c (GC_malloc_explicitly_typed, GC_calloc_explicitly_typed):
Change type of "op" local variable from ptr_t to word*.
* typd_mlc.c (GC_malloc_explicitly_typed, GC_calloc_explicitly_typed):
Use GC_malloc_kind instead of explicit allocator (those implementation
was same as for GC_malloc_kind).
* typd_mlc.c (GC_calloc_explicitly_typed): Remove"register" keyword for
"descr_type" local variable; reformat code slightly; use EXPECT to
check GC_general_register_disappearing_link result for out-of-memory.
Mike Frysinger [Sat, 29 Aug 2015 23:14:24 +0000 (19:14 -0400)]
Check for execinfo.h by configure
The current header depends on glibc/uClibc version checks to determine
whether execinfo.h exists which breaks other C libs. Instead, add an
explicit configure check for it.
* configure.ac: Check execinfo.h presence, define
GC_MISSING_EXECINFO_H otherwise.
* include/gc_config_macros.h [__GLIBC__] (GC_HAVE_BUILTIN_BACKTRACE):
Check absence of GC_MISSING_EXECINFO_H instead of __UCLIBC__.
Ivan Maidanski [Wed, 20 Jul 2016 22:26:58 +0000 (01:26 +0300)]
Re-implement GC_finalized_malloc using GC_malloc_kind
(code refactoring)
* fnlz_mlc.c: Do not test THREAD_LOCAL_ALLOC; do not include
thread_local_alloc.h.
* fnlz_mlc.c (GC_finalized_objfreelist, GC_core_finalized_malloc):
Remove.
* fnlz_mlc.c [ENABLE_DISCLAIM]: Include gc_inline.h (to declare
GC_malloc_kind).
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_init_finalized_malloc): Do not use
GC_finalized_objfreelist.
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_finalized_malloc): Call
GC_malloc_kind (instead of own implementation mostly duplicating the
algorithm of GC_malloc_kind[_global]); change type of "op" local
variable from ptr_t to word*.
* include/private/thread_local_alloc.h [ENABLE_DISCLAIM]
(THREAD_FREELISTS_KINDS): Increase value by 1.
* include/private/thread_local_alloc.h
(thread_local_freelists.finalized_freelists,
GC_finalized_objfreelist): Remove.
* thread_local_alloc.c (GC_finalized_objfreelist): Remove.
* thread_local_alloc.c (GC_init_thread_local, GC_destroy_thread_local,
GC_mark_thread_local_fls_for, GC_check_tls_for): Do not access
finalized_freelists.
Ivan Maidanski [Tue, 19 Jul 2016 22:04:17 +0000 (01:04 +0300)]
Fix integer shift undefined behavior in GC_init_explicit_typing
* typd_mlc.c (GC_init_explicit_typing): Avoid left-shift by WORDSZ
(which is an undefined behavior), initialize GC_bm_table[0] to
GC_DS_BITMAP explicitly.
Ivan Maidanski [Fri, 1 Jul 2016 07:30:57 +0000 (10:30 +0300)]
Replace (fix) 'objs' acronym in comments with 'objects word
* include/gc.h (GC_PROTECTS_POINTER_HEAP): Replace "objs" word to
"objects" in comment.
* include/private/gc_priv.h (GC_arrays._aobjfreelist,
GC_arrays._uobjfreelist, GC_arrays._auobjfreelist, GC_aobjfreelist):
Likewise.
* mark.c (INITIAL_MARK_STACK_SIZE): Likewise.
* include/private/gc_priv.h (GC_arrays._uobjfreelist): Add missing dot
to delimit the end of a sentence in comment.
Ivan Maidanski [Thu, 30 Jun 2016 07:20:53 +0000 (10:20 +0300)]
Revert "Refactoring of GC_Xobjfreelist" partially
This reverts commit 41871b970c8fd9704835c1a221a8f9f1deea707a partially
(GC_malloc_kind_global, GC_generic_malloc_uncollectable,
GC_destroy_thread_local are not reverted).
Ivan Maidanski [Fri, 24 Jun 2016 20:01:53 +0000 (23:01 +0300)]
Fix unchecked fork() result in gctest (Unix, Cygwin)
Fail gctest if forked process failed.
* tests/test.c [!NO_TEST_HANDLE_FORK]: Include sys/types.h, sys/wait.h.
* tests/test.c (run_one_test) [!NO_TEST_HANDLE_FORK]: Declare pid,
wstatus local variables; FAIL if fork() returns -1; call waitpid() for
the child process; FAIL if the child process failed.
Ivan Maidanski [Sat, 30 Apr 2016 22:07:53 +0000 (01:07 +0300)]
Fix GC_new_kind in case of client-defined PREDEFINED_KINDS
There are 2 types of kinds - "pre-allocated" ones (with the number less
than PREDEFINED_KINDS) which have pre-allocated free lists and object
allocation for these kinds is dealt by GC_malloc_kind typically,
and "custom" kinds which require the client to supply the free list
(e.g., allocated by GC_new_free_list[_inner]) and arrange own object
allocation procedure.
By design, GC_new_kind[_inner] creates new "custom" kind.
* mark.c (GC_N_KINDS_INITIAL_VALUE): Simplify and comment out (for now).
* mark.c (GC_n_kinds): Initialize to PREDEFINED_KINDS (instead of
GC_N_KINDS_INITIAL_VALUE); add TODO item.
Ivan Maidanski [Tue, 21 Jun 2016 20:38:23 +0000 (23:38 +0300)]
Fix assertion violation in GC_wait_builder called from start_mark_threads
(fix commit 0ca6d3f)
* include/private/gc_priv.h [PARALLEL_MARK]
(GC_start_mark_threads_inner): Declare even if no CAN_HANDLE_FORK.
* misc.c [THREADS] (GC_start_mark_threads): Define also for the case
of PARALLEL_MARK and CAN_HANDLE_FORK - call GC_start_mark_threads_inner
surrounded with DISABLE/RESTORE_CANCEL.
* pthread_support.c [PARALLEL_MARK] (GC_start_mark_threads_inner):
Always define as GC_INNER.
* win32_threads.c [GC_PTHREADS_PARAMARK] (GC_start_mark_threads_inner):
Likewise.
Ivan Maidanski [Tue, 21 Jun 2016 06:48:21 +0000 (09:48 +0300)]
Fix GC_suspend_thread for terminated threads
* pthread_stop_world.c (GC_suspend_thread): Do not clear
SUSPENDED_EXT flag in case of RAISE_SIGNAL() failure, add assertion
about FINISHED (in case of ESRCH), update comment.
* pthread_stop_world.c (GC_register_my_thread): Add assertion
that SUSPENDED_EXT flag is not set if the thread is registered from
a thread key destructor.
Ivan Maidanski [Mon, 20 Jun 2016 08:38:50 +0000 (11:38 +0300)]
Fix deadlock (and double lock) in explicit thread suspend/resume
* pthread_stop_world.c (GC_suspend_handler_inner)
[GC_ENABLE_SUSPEND_THREAD]: If SUSPENDED_EXT flag then set
stop_info.stack_ptr, call sem_post(suspend_ack_sem), and call
suspend_self_inner instead of GC_do_blocking(suspend_self_inner).
* pthread_stop_world.c [GC_ENABLE_SUSPEND_THREAD] (GC_suspend_thread):
No-op if already suspended; UNLOCK before GC_do_blocking (if
self-suspend); add TODO about GC_retry_signals; clear SUSPENDED_EXT
flag if RAISE_SIGNAL failed with ESRCH code; sem_wait(suspend_ack_sem)
to let the suspend handler to lookup the thread and store stack_ptr
(and save registers if needed).
* pthread_stop_world.c (GC_suspend_all, GC_start_world): Skip threads
with SUSPENDED_EXT flag.
Peter Wang [Thu, 16 Jun 2016 09:16:05 +0000 (12:16 +0300)]
Fix STACKBOTTOM for Solaris 11/x86
* include/private/gcconfig.h [I386 && SOLARIS] (STACKBOTTOM,
HEURISTIC2): Define conditionally depening on USERLIMIT - use
HEURISTIC2 if USERLIMIT is undefined (workaround similar to that for
Solaris/sparc and Solaris/x64); include sys/vmparam.h instead of
sys/vm.h; update comment.
Adrian Pop [Fri, 10 Jun 2016 17:39:18 +0000 (20:39 +0300)]
Workaround a bug in winpthreads causing parallel marks deadlock (MinGW)
The solution is to use Win32 threads for parallel markers while using
winpthreads to provide GC-aware pthread_create/join/detach and fork.
See details in GitHub issue #81 and issue #119.
* win32_threads.c (GC_PTHREADS_PARAMARK): Do not define if
GC_PTHREADS, PARALLEL_MARK and __MINGW32__ (to workaround a deadlock
in do_sema_b_wait() of winpthreads).
Paul Bone [Fri, 10 Jun 2016 08:11:47 +0000 (11:11 +0300)]
Dump the block information in CSV format
This makes it easy to copy this information into a new file and
process it with tools that understand CSV, e.g. to create histograms
of block utilization.
* reclaim.c (GC_print_block_descr): add n_objs local variable;
separate printed values with a comma; print also size of block in
object units; refactor code for computing total_bytes field.
* reclaim.c (GC_print_block_list): Separate columns in printed table
header with a comma; append "#objs" column to the table.
Ivan Maidanski [Mon, 30 May 2016 19:46:59 +0000 (22:46 +0300)]
Refactoring of android_thread_kill/pthread_kill calls
* pthread_stop_world.c (android_thread_kill) [USE_TKILL_ON_ANDROID]:
Move definition upper (to be before its first use); remove forward
declaration.
* pthread_stop_world.c (THREAD_SYSTEM_ID, RAISE_SIGNAL): New macro.
* pthread_stop_world.c (GC_suspend_thread): Remove "result" local
variable
* pthread_stop_world.c (GC_suspend_thread, GC_suspend_all,
GC_start_world): Use RAISE_SIGNAL() instead of pthread_kill and
android_thread_kill.
* pthread_stop_world.c (GC_suspend_all, GC_start_world): Remove
"thread_id" local variable; use THREAD_SYSTEM_ID instead of thread_id.
Ivan Maidanski [Sat, 30 Apr 2016 22:07:53 +0000 (01:07 +0300)]
Fix GC_new_kind in case of client-defined PREDEFINED_KINDS
There are 2 types of kinds - "pre-allocated" ones (with the number less
than PREDEFINED_KINDS) which have pre-allocated free lists and object
allocation for these kinds is dealt by GC_malloc_kind typically,
and "custom" kinds which require the client to supply the free list
(e.g., allocated by GC_new_free_list[_inner]) and arrange own object
allocation procedure.
By design, GC_new_kind[_inner] creates new "custom" kind.
* mark.c (GC_N_KINDS_INITIAL_VALUE): Simplify and comment out (for now).
* mark.c (GC_n_kinds): Initialize to PREDEFINED_KINDS (instead of
GC_N_KINDS_INITIAL_VALUE); add TODO item.
Ivan Maidanski [Tue, 5 Apr 2016 19:08:55 +0000 (22:08 +0300)]
Fix GC_REALLOC to call GC_FREE if new size is zero and pointer is non-NULL
* dbg_mlc.c (GC_debug_realloc): Call GC_debug_free (and return NULL)
if lb is zero and p is non-NULL.
* extra/AmigaOS.c (GC_amiga_realloc): Do not retry and do not call
WARN if new_size_in_bytes is zero.
* include/gc.h (GC_realloc): Refine comment (better document the cases
of NULL and/or zero arguments).
* mallocx.c (GC_realloc): Call GC_free (skip it if IGNORE_FREE) and
return NULL if lb is zero and p is non-NULL.
Ivan Maidanski [Fri, 18 Mar 2016 18:06:53 +0000 (21:06 +0300)]
Fix 'GC_generic_malloc_inner_ignore_off_page not used' compiler warning
(for GC_NO_FINALIZATION case)
* include/private/gc_priv.h (GC_generic_malloc_inner_ignore_off_page):
Do not declare unless DBG_HDRS_ALL or GC_GCJ_SUPPORT, or
!GC_NO_FINALIZATION.
* malloc.c (GC_generic_malloc_inner_ignore_off_page): Do not define
unless DBG_HDRS_ALL or GC_GCJ_SUPPORT, or !GC_NO_FINALIZATION.
Ivan Maidanski [Fri, 4 Mar 2016 22:11:05 +0000 (01:11 +0300)]
start_mark_threads refactoring regarding GC_markers_m1 update (Win32)
(Sync semantics around GC_markers_m1 with that of pthread_support.c.)
(fix commit 95947bd)
* win32_threads.c (GC_start_mark_threads_inner): Update GC_markers_m1
value to i just after loop creating the threads.
Hiroshi Yokota [Sat, 6 Feb 2016 01:25:48 +0000 (10:25 +0900)]
Fix user-defined signals drop by marker threads
When one user thread sets its own signal handler in multi-threaded
code, GC marker thread sometimes steal and drop user signal calls.
Because GC marker thread doesn't knows the user thread's signal
handler, and uses the default signal handler that ignores any signal
call. This becomes to drop the signal call.
This patch inhibits to accept signals by GC marker threads except
GC system signals, and don't drop user signal calls.
This patch comes from Gauche.
* include/private/gcconfig.h (NO_MARKER_SPECIAL_SIGMASK): New macro.
* pthread_support.c (GC_start_mark_threads_inner): Unless
NO_MARKER_SPECIAL_SIGMASK, call pthread_sigmask to block all signals
(except for the ones used for GC suspend/resume if any) before
starting marker threads and restore signals mask afterwards.
* win32_threads.c (GC_start_mark_threads_inner): Likewise.