Issue #1699 - Part 3: Update and rework jemalloc for compatibility with modern FreeBSD

- Fixes a bootstrapping loop between FreeBSD's pthread implementation and current mozjemalloc code ('pthread_mutex_lock' called in 'malloc_init/malloc_mutex_lock' itself calls 'calloc', resulting in 'malloc_init' being called again).
- Reuses Linux code (adaptive mutexes interface is the same), as well as Solaris' for ncpus, which is actually POSIX-compliant. Adaptive mutex use is used under __FreeBSD__ instead of MOZ_MEMORY_BSD, since this interface is not always present on the other BSDs.
- Creates a simple memory allocator meant to be use during jemalloc bootstrap (malloc_init_hard()). Although protected by "#ifdef __FreeBSD__", it is not FreeBSD-specific: Any POSIX platform could use it.
- Hooks it up so that it is used in place of jemalloc's own routines while malloc_init_hard() is executing or memory for mutexes is allocated. Currently, 'malloc', 'calloc', '*memalign' and 'free' are diverted during init or lock initializations.
- Removes 'malloc_initializing' by making sure the initialization done for multi-threading will be triggered first by one of the 'malloc_lock_init' calls. This initialization happens only at first lock (pthread_mutex_lock), as reading libthr's code confirms (and not in 'pthread_mutex_init'). So, lock and unlock the just created mutex in 'malloc_mutex_init'. If multi-threading initialization in libthr is at some point moved to 'pthread_mutex_init', this will continue to work (and be a no-op).
- Since FreeBSD's code then diverges a bit more from Linux's, separate them, as it also eases code reading.

0 files changed, 0 insertions, 0 deletions