5 files changed, 58 insertions, 26 deletions

diff --git a/libc-top-half/musl/arch/wasm32/atomic_arch.h b/libc-top-half/musl/arch/wasm32/atomic_arch.h
index dd9428c..c24ce2d 100644
--- a/libc-top-half/musl/arch/wasm32/atomic_arch.h
+++ b/libc-top-half/musl/arch/wasm32/atomic_arch.h
@@ -1,4 +1,3 @@
-#define a_barrier() (__sync_synchronize())
 #define a_cas(p, t, s) (__sync_val_compare_and_swap((p), (t), (s)))
 #define a_crash() (__builtin_trap())
 #define a_clz_32 __builtin_clz
diff --git a/libc-top-half/musl/include/pthread.h b/libc-top-half/musl/include/pthread.h
index b14fe82..05101e8 100644
--- a/libc-top-half/musl/include/pthread.h
+++ b/libc-top-half/musl/include/pthread.h
@@ -55,15 +55,9 @@ extern "C" {
 #define PTHREAD_PROCESS_SHARED 1
 
 
-#if defined(__wasilibc_unmodified_upstream) || defined(_REENTRANT)
 #define PTHREAD_MUTEX_INITIALIZER {{{0}}}
 #define PTHREAD_RWLOCK_INITIALIZER {{{0}}}
 #define PTHREAD_COND_INITIALIZER {{{0}}}
-#else
-#define PTHREAD_MUTEX_INITIALIZER 0
-#define PTHREAD_RWLOCK_INITIALIZER 0
-#define PTHREAD_COND_INITIALIZER 0
-#endif
 #define PTHREAD_ONCE_INIT 0
 
 
diff --git a/libc-top-half/musl/src/env/__init_tls.c b/libc-top-half/musl/src/env/__init_tls.c
index ece8d24..c3e407c 100644
--- a/libc-top-half/musl/src/env/__init_tls.c
+++ b/libc-top-half/musl/src/env/__init_tls.c
@@ -47,10 +47,9 @@ static inline void setup_default_stack_size()
 		stack_size = sp > &__global_base ? &__heap_base - &__data_end : (ptrdiff_t)&__global_base;
 	}
 
-	if (stack_size > __default_stacksize)
-		__default_stacksize =
-			stack_size < DEFAULT_STACK_MAX ?
-			stack_size : DEFAULT_STACK_MAX;
+	__default_stacksize =
+		stack_size < DEFAULT_STACK_MAX ?
+		stack_size : DEFAULT_STACK_MAX;
 }
 
 void __wasi_init_tp() {
diff --git a/libc-top-half/musl/src/thread/pthread_create.c b/libc-top-half/musl/src/thread/pthread_create.c
index 676e2cc..5de9f5a 100644
--- a/libc-top-half/musl/src/thread/pthread_create.c
+++ b/libc-top-half/musl/src/thread/pthread_create.c
@@ -60,6 +60,17 @@ void __tl_sync(pthread_t td)
 	if (tl_lock_waiters) __wake(&__thread_list_lock, 1, 0);
 }
 
+#ifndef __wasilibc_unmodified_upstream
+static void *map_base_deferred_free;
+
+static void process_map_base_deferred_free()
+{
+	/* called with __tl_lock held */
+	free(map_base_deferred_free);
+	map_base_deferred_free = NULL;
+}
+#endif
+
 #ifdef __wasilibc_unmodified_upstream
 _Noreturn void __pthread_exit(void *result)
 #else
@@ -164,14 +175,6 @@ static void __pthread_exit(void *result)
 	self->prev->next = self->next;
 	self->prev = self->next = self;
 
-#ifndef __wasilibc_unmodified_upstream
-	/* On Linux, the thread is created with CLONE_CHILD_CLEARTID,
-	 * and this lock will unlock by kernel when this thread terminates.
-	 * So we should unlock it here in WebAssembly.
-	 * See also set_tid_address(2) */
-	__tl_unlock();
-#endif
-
 #ifdef __wasilibc_unmodified_upstream
 	if (state==DT_DETACHED && self->map_base) {
 		/* Detached threads must block even implementation-internal
@@ -190,10 +193,17 @@ static void __pthread_exit(void *result)
 	}
 #else
 	if (state==DT_DETACHED && self->map_base) {
-		// __syscall(SYS_exit) would unlock the thread, list
-		// do it manually here
-		__tl_unlock();
-		free(self->map_base);
+		/* As we use malloc/free which is considerably more complex
+		 * than mmap/munmap to call and can even require a valid
+		 * thread context, it's difficult to implement __unmapself.
+		 *
+		 * Here we take an alternative approach which simply defers
+		 * the deallocation. An obvious downside of this approach is
+		 * that it keeps the stack longer. (possibly forever.)
+		 * To avoid wasting too much memory, we only defer a single
+		 * item at most. */
+		process_map_base_deferred_free();
+		map_base_deferred_free = self->map_base;
 		// Can't use `exit()` here, because it is too high level
 		return;
 	}
@@ -212,10 +222,15 @@ static void __pthread_exit(void *result)
 #ifdef __wasilibc_unmodified_upstream
 	for (;;) __syscall(SYS_exit, 0);
 #else
-	// __syscall(SYS_exit) would unlock the thread, list
-	// do it manually here
-	__tl_unlock();
 	// Can't use `exit()` here, because it is too high level
+
+	/* On Linux, the thread is created with CLONE_CHILD_CLEARTID,
+	 * and the lock (__thread_list_lock) will be unlocked by kernel when
+	 * this thread terminates.
+	 * See also set_tid_address(2)
+	 *
+	 * In WebAssembly, we leave it to wasi_thread_start instead.
+	 */
 #endif
 }
 
@@ -430,6 +445,14 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
 			if (map == MAP_FAILED) goto fail;
 		}
 #else
+		/* Process the deferred free request if any before
+		 * allocationg a new one. Hopefully it enables a reuse of the memory.
+		 *
+		 * Note: We can't perform a simple "handoff" becasue allocation
+		 * sizes might be different. (eg. the stack size might differ) */
+		__tl_lock();
+		process_map_base_deferred_free();
+		__tl_unlock();
 		map = malloc(size);
 		if (!map) goto fail;
 #endif
diff --git a/libc-top-half/musl/src/thread/wasm32/wasi_thread_start.s b/libc-top-half/musl/src/thread/wasm32/wasi_thread_start.s
index 0fe9854..7a480b8 100644
--- a/libc-top-half/musl/src/thread/wasm32/wasi_thread_start.s
+++ b/libc-top-half/musl/src/thread/wasm32/wasi_thread_start.s
@@ -28,4 +28,21 @@ wasi_thread_start:
 	local.get   1  # start_arg
 	call __wasi_thread_start_C
 
+	# Unlock thread list. (as CLONE_CHILD_CLEARTID would do for Linux)
+	#
+	# Note: once we unlock the thread list, our "map_base" can be freed
+	# by a joining thread. It's safe as we are in ASM and no longer use
+	# our C stack or pthread_t. It's impossible to do this safely in C
+	# because there is no way to tell the C compiler not to use C stack.
+	i32.const   __thread_list_lock
+	i32.const   0
+	i32.atomic.store 0
+	# As an optimization, we can check tl_lock_waiters here.
+	# But for now, simply wake up unconditionally as
+	# CLONE_CHILD_CLEARTID does.
+	i32.const   __thread_list_lock
+	i32.const   1
+	memory.atomic.notify 0
+	drop
+
 	end_function