(code refactoring)
* include/private/gc_priv.h (GC_MACH_THREAD_STATE,
GC_MACH_THREAD_STATE_COUNT): Remove explicit definition to
PPC_THREAD_STATE[_COUNT], ARM_THREAD_STATE[_COUNT] (as the correct
values are defined via MACHINE_THREAD_STATE[_COUNT] macro).
* include/private/gc_priv.h (GC_MACH_THREAD_STATE,
GC_MACH_THREAD_STATE_COUNT): Define to ARM_THREAD_STATE[_COUNT],
respectively, instead of mistyped ARM_MACHINE_THREAD_STATE[_COUNT]
(if DARWIN and ARM32).
Since __clang_major__/__clang_minor__ etc. are vendor dependent values,
we cannot implement the feature detection based on it.
For example, Apple clang versioning is different from the FreeBSD clang.
(At this time, Apple clang version is "6.0 (clang-600.0.51)" and
__clang_major__ is 6.)
Instead of this, we can use the clang feature detection macro,
__has_attribute.
* include/gc_config_macros.h (GC_ATTR_ALLOC_SIZE): Replace predefined
__clang_major/minor__ testing with __has_attribute() one (in case of
clang).
* include/private/darwin_semaphore.h (sem_init): Destroy sem->mutex if
sem->cond initialization failed.
* include/private/darwin_semaphore.h (sem_post): Ignore
pthread_mutex_unlock result in case of pthread_cond_signal.
* include/private/darwin_semaphore.h (sem_wait): Unlock mutex and
return error (-1) if pthread_cond_wait failed.
* include/private/darwin_semaphore.h (sem_init): If pshared then
return -1 (with the appropriate errno code set) instead of ABORT.
* include/private/darwin_semaphore.h (sem_init, sem_post, sem_wait):
Treat non-zero value returned by pthread functions as error (instead
of only negative values).
* include/private/darwin_semaphore.h (sem_init, sem_post, sem_wait):
Reformat code.
* misc.c (GC_init): Abort (with the appropriate message) if
pthread_mutex[attr]_init failed (SN_TARGET_PS3 case only).
* specific.c (GC_key_create_inner): If pthread_mutex_init failed then
return its error code.
(and do not call it if no parallel marker)
* include/private/pthread_support.h (GC_setup_mark_lock): Remove
prototype.
* misc.c (GC_init): Remove GC_setup_mark_lock call (move it to
GC_thr_init).
* pthread_support.c (setup_mark_lock): Rename from GC_setup_mark_lock;
add prototype (only if PARALLEL_MARK).
* pthread_support.c (GC_thr_init): Call setup_mark_lock() unless
available_markers_m1 is 0.
store a thread's state when compiling against the iOS 7/8 SDK. Without this
patch GC_stack_range_for() crashes when running a 32-bit app on iOS 64-bit.
The old code passed an arm_thread_state_t and ARM_THREAD_STATE to
thread_get_state(). ARM_THREAD_STATE is the same as ARM_UNIFIED_THREAD_STATE
on iOS 7/8 and thread_get_state() actually expects an
arm_unified_thread_state_t. On iOS 32-bit it looks like thread_get_state()
only touches the first bytes corresponding to the size of arm_thread_state_t
so no crash there. On iOS 64-bit however it seems thread_get_state() writes to
the full arm_unified_thread_state_t which meant it would overflow the stack
allocated struct passed to it and mess up other values on the stack leading to
a crash later on.
* configure.ac (HAVE_LIBC_VERSION_H, HAVE_GNU_GET_LIBC_VERSION): Remove
(revert change in previous commit).
* include/private/gcconfig.h (GLIBC_2_19_TSX_BUG): New macro defined
for Linux/x86_64 (if Glibc used) to workaround a bug in Glibc lock
elision implementation.
* pthread_support.c: Move include of gnu/libc-version.h to gcconfig.h
(used to check whether lock elision workaround needed).
* misc.c (GC_init): Reformat code.
* pthread_support.c (mark_mutex): Initialize (to
PTHREAD_MUTEX_INITIALIZER) even lock elision workaround is needed
(revert change in previous commit).
* pthread_support.c (parse_version): New static function (defined only
if GLIBC_2_19_TSX_BUG).
* pthread_support.c (GC_setup_mark_lock): Use parse_version to check
target Glibc version properly; do not reinitialize mutex unless
workaround needed; call ABORT (with the appropriate message) in case
of a failure in pthread_mutexattr_init/settype, pthread_mutex_init.
glibc 2.19 on Linux x86-64 platforms includes support for lock elision,
by using Intel's TSX support when it is available. Without modifying an
application this converts suitable critical sections that use mutex into
transactional memory critical sections. See http://lwn.net/Articles/534758/
If a problem occurs that means that transactional memory can't be used, such
as a system call or buffer overflow, the pthreads implementation will catch
this error and retry the critical section using a normal mutex.
I noticed that since upgrading glibc that programs using Boehm GC crash, one
of these crashes was an assertion that the owner field of a mutex was
invalid. The assertion was generated by the pthreads implementation.
I believe that there is a bug in glibc that when a mutex cannot be used
safely for transactions that some series of events causes it's owner field
to be set incorrectly (or cleared when it shouldn't be).
I've found that I can work around this problem by having Boehm GC use an
error checking mutex, which I believe doesn't use lock elision and in my
testing doesn't crash.
XXX: This work-around mostly works except for linking the feature detection
in configure.ac to the conditional compilation in pthread_support.c as there
isn't an obvious way to make it work for automake and Makefile.direct.
Could I have some help updating the build system please?
include/private/pthread_support.h:
pthread_support.c:
Define GC_setup_mark_lock() This procedure creates the lock specifying a
pthread_mutexattr_t structure. This is used to disable lock elision on
Linux with glibc 2.19 or greater.
configure.ac:
If we're using Linux then check for the gnu extensions required to
identify the version of glibc at runtime.
misc.c:
Call GC_setup_mark_lock() when initialising the collector.
Winpthreads is a different pthread implementation for MinGW-w64.
This patch redefines GC_WIN32_PTHREADS to mean either pthreads-win32
or winpthreads.
* configure.ac (GC_WIN32_PTHREADS): Improve description to cover
"winpthreads" library.
* doc/README.macros (GC_WIN32_PTHREADS): Likewise.
* doc/README.win32: Likewise.
* include/gc_config_macros.h (GC_WIN32_THREADS): Mention "winpthreads"
library in comment.
* win32_threads.c (GC_pthread_join): Likewise.
* include/private/gc_locks.h (NUMERIC_THREAD_ID, THREAD_EQUAL,
NUMERIC_THREAD_ID_UNIQUE): Define to support winpthreads properly (if
GC_WIN32_PTHREADS).
* win32_threads.c (GC_PTHREAD_PTRVAL): Likewise.
* alloc.c (min_bytes_allocd): Test STACK_NOT_SCANNED macro instead
of __EMSCRIPTEN__ (stack size to scan is zero if STACK_NOT_SCANNED).
* include/private/gcconfig.h (ALIGNMENT): Remove duplicate definition
for _EMSCRIPTEN__.
* include/private/gcconfig.h (STACK_NOT_SCANNED): New macro defined
for __EMSCRIPTEN__ target (in addition to OS_TYPE, CPP_WORDSZ,
ALIGNMENT, DATASTART, DATAEND).
* mach_dep.c (GC_push_regs): Test STACK_NOT_SCANNED macro instead
of __EMSCRIPTEN__ (push nothing if STACK_NOT_SCANNED).
* mark_rts.c (GC_push_roots): Test STACK_NOT_SCANNED macro instead
of __EMSCRIPTEN__ (do not call GC_push_regs_and_stack if
STACK_NOT_SCANNED); mark cold_gc_frame argument as potentially unused.
* misc.c (GC_clear_stack): Test STACK_NOT_SCANNED macro instead
of __EMSCRIPTEN__ (do not clear stack if STACK_NOT_SCANNED).
* misc.c (GC_clear_stack): Reformat code.
Niklas Therning
Ivan Maidanski
Yusuke Suzuki
Jan Alexander Steffens (heftig)
Paul Bone
Manuel A. Fernandez Montecelo
Peter Wang