Include implemenation of musl's __lock/__unlock functions. NFC (#16129)

The functions `__lock` and `__unlock` are never called directly within
musl but always via the `LOCK()` and `UNLOCK()` macros.

They are implemented internally using an atomic integer along with the
`__wake` and `__futexwait` internal function which emscripten already
supports.

Not to be confused with `__lockfile/__unlockfile` which are used to
implement the `FLOCK()/FUNLOCK()` macros (which we already support and
have enabled).

In theory this could help remove data races and crashed within musl.
In practice many of the uses of `LOCK/UNLOCK` are in places not compiled
into emscripten.  However I noticed at least these 3 that we do depend
on and they do look useful/valid:

- stdio/ofl.c: Open file lock used in fclose.c/fflush.c/__stdio_exit.c (We
  don't use __stdio_exit.c under emscripten)
- random/srandom: Used as a mutex for the prng generator
- dirent/rewinddir/seekdir: Used a mutex on an open directory handle

Author: sbc100

system/lib/libc/extras.c

@@ -389,6 +389,10 @@ int sem_destroy(sem_t *sem) {
 
 void __wait(volatile int *addr, volatile int *waiters, int val, int priv) {}
 
+void __lock(void* ptr) {}
+
+void __unlock(void* ptr) {}
+
 // When pthreads is not enabled, we can't use the Atomics futex api to do
 // proper sleeps, so simulate a busy spin wait loop instead.
 void emscripten_thread_sleep(double msecs) {

system/lib/pthread/library_pthread_stub.c

@@ -144,9 +144,17 @@ int __pthread_create(pthread_t* restrict res,
 #endif
 
   //printf("start __pthread_create: %p\n", self);
+
+  // Set libc.need_locks before calling __pthread_create_js since
+  // by the time it returns the thread could be running and we
+  // want libc.need_locks to be set from the moment it starts.
+  if (!libc.threads_minus_1++) libc.need_locks = 1;
+
   int rtn = __pthread_create_js(new, attrp, entry, arg);
-  if (rtn != 0)
+  if (rtn != 0) {
+    if (!--libc.threads_minus_1) libc.need_locks = 0;
     return rtn;
+  }
 
   // TODO(sbc): Implement circular list of threads
   /*
@@ -209,6 +217,8 @@ void _emscripten_thread_exit(void* result) {
 
   free_tls_data();
 
+  if (!--libc.threads_minus_1) libc.need_locks = 0;
+
   // TODO(sbc): Implement circular list of threads
   /*
   __tl_lock();

system/lib/pthread/pthread_create.c

@@ -21,7 +21,8 @@ $_emscripten_thread_init
 $_main_thread
 $a_cas
 $a_dec
-$a_inc.1
+$a_fetch_add
+$a_inc
 $a_swap
 $add
 $dispose_chunk

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.funcs

@@ -817,7 +817,7 @@ def get_files(self):
 
     if self.is_mt:
       ignore += [
-        'clone.c', '__lock.c',
+        'clone.c',
         'pthread_create.c',
         'pthread_kill.c', 'pthread_sigmask.c',
         '__set_thread_area.c', 'synccall.c',
@@ -961,7 +961,6 @@ def get_files(self):
         path='system/lib/libc',
         filenames=[
           'dynlink.c',
-          'extras.c',
           'wasi-helpers.c',
           'emscripten_get_heap_size.c',
           'raise.c',