Define __stack_base and __stack_limit globals in debug mode for s…

…tack overflow detection

That enables VMs to implement stack overflow detection or using passes like https://github.com/WebAssembly/binaryen/blob/main/src/passes/StackCheck.cpp

Makefile

@@ -269,7 +269,7 @@ LIBC_TOP_HALF_MUSL_SOURCES += \
         thread/sem_timedwait.c \
         thread/sem_trywait.c \
         thread/sem_wait.c \
-        thread/wasm32/wasi_thread_start.s \
+        thread/wasm32/wasi_thread_start.S \
     )
 endif
 
@@ -341,7 +341,7 @@ CFLAGS += -isystem "$(SYSROOT_INC)"
 # These variables describe the locations of various files and directories in
 # the build tree.
 objs = $(patsubst $(CURDIR)/%.c,$(OBJDIR)/%.o,$(1))
-asmobjs = $(patsubst $(CURDIR)/%.s,$(OBJDIR)/%.o,$(1))
+asmobjs = $(patsubst $(CURDIR)/%.S,$(OBJDIR)/%.o,$(1))
 DLMALLOC_OBJS = $(call objs,$(DLMALLOC_SOURCES))
 EMMALLOC_OBJS = $(call objs,$(EMMALLOC_SOURCES))
 LIBC_BOTTOM_HALF_ALL_OBJS = $(call objs,$(LIBC_BOTTOM_HALF_ALL_SOURCES))
@@ -520,7 +520,7 @@ $(OBJDIR)/%.o: $(CURDIR)/%.c include_dirs
 	@mkdir -p "$(@D)"
 	$(CC) $(CFLAGS) -MD -MP -o $@ -c $<
 
-$(OBJDIR)/%.o: $(CURDIR)/%.s include_dirs
+$(OBJDIR)/%.o: $(CURDIR)/%.S include_dirs
 	@mkdir -p "$(@D)"
 	$(CC) $(ASMFLAGS) -o $@ -c $<
 

libc-top-half/musl/src/thread/pthread_create.c

@@ -239,11 +239,14 @@ struct start_args {
 	unsigned long sig_mask[_NSIG/8/sizeof(long)];
 #else
 	/*
-	 * Note: the offset of the "stack" and "tls_base" members
+	 * Note: the offset of the "stack", "stack_limit" and "tls_base" members
 	 * in this structure is hardcoded in wasi_thread_start.
 	 */
 	void *stack;
 	void *tls_base;
+#ifndef NDEBUG
+	void *stack_limit;
+#endif
 	void *(*start_func)(void *);
 	void *start_arg;
 #endif
@@ -501,7 +504,10 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
 	/* Correct the stack size */
 	new->stack_size = stack - stack_limit;
 
-	args->stack = new->stack; /* just for convenience of asm trampoline */
+	/* just for convenience of asm trampoline */
+	args->stack = new->stack;
+	args->stack_limit = stack_limit;
+
 	args->start_func = entry;
 	args->start_arg = arg;
 	args->tls_base = (void*)new_tls_base;

libc-top-half/musl/src/thread/wasm32/wasi_thread_start.S

@@ -0,0 +1,64 @@
+#define PTR i32
+#define PTRSIZE 4
+# Preprocessor's define directive doesn't resolve math expressions,
+# so we hardcode the result of PTRSIZE * 2 operation
+#define DOUBLE_PTRSIZE 8
+
+	.text
+
+	.export_name	wasi_thread_start, wasi_thread_start
+
+	.globaltype	__stack_pointer, PTR
+	.globaltype	__tls_base, PTR
+
+#ifndef NDEBUG
+	.globaltype	__stack_base, PTR
+	.globaltype	__stack_limit, PTR
+#endif
+
+	.functype	__wasi_thread_start_C (i32, PTR) -> ()
+
+	.hidden	wasi_thread_start
+	.globl	wasi_thread_start
+	.type	wasi_thread_start,@function
+
+wasi_thread_start:
+	.functype	wasi_thread_start (i32, PTR) -> ()
+
+	# Set up the minimum C environment.
+	# Note: offsetof(start_arg, stack) == 0
+	local.get   1  # start_arg
+	PTR.load    0  # stack
+	global.set  __stack_pointer
+
+	local.get   1        # start_arg
+	PTR.load    PTRSIZE  # tls_base
+	global.set  __tls_base
+
+#ifndef NDEBUG
+	# configure __stack_base and __stack_limit in debug mode
+	# to allow for stack overflow detection
+	local.get   1  # start_arg
+	PTR.load    0  # stack
+	global.set  __stack_base
+
+	local.get   1               # start_arg
+	PTR.load    DOUBLE_PTRSIZE  # stack_limit
+	global.set __stack_limit
+#endif
+
+	# Make the C function do the rest of work.
+	local.get   0  # tid
+	local.get   1  # start_arg
+	call __wasi_thread_start_C
+
+	end_function
+
+#ifndef NDEBUG
+.section .data,"",@
+	__stack_base:
+	.size __stack_base, PTRSIZE
+
+	__stack_limit:
+	.size __stack_limit, PTRSIZE
+#endif