Disable aux stack allocations for threads spawned by wasi_thread_start (#1867)

This syscall doesn't need allocating stack or TLS and it's expected from the application
to do that instead. E.g. WASI-libc already does this for `pthread_create`.

Also fix some of the examples to allocate memory for stack and not use stack before
the stack pointer is set to a correct value.

samples/wasi-threads/wasm-apps/CMakeLists.txt

@@ -13,12 +13,13 @@ endif ()
 
 set (CMAKE_SYSROOT                  "${WASI_SYSROOT}")
 set (CMAKE_C_COMPILER               "${WASI_SDK_DIR}/bin/clang")
+set (CMAKE_ASM_COMPILER               "${WASI_SDK_DIR}/bin/clang")
 set (CMAKE_C_COMPILER_TARGET        "wasm32-wasi")
 
 function (compile_sample SOURCE_FILE)
   get_filename_component (FILE_NAME ${SOURCE_FILE} NAME_WLE)
   set (WASM_MODULE ${FILE_NAME}.wasm)
-  add_executable (${WASM_MODULE} ${SOURCE_FILE})
+  add_executable (${WASM_MODULE} ${SOURCE_FILE} ${ARGN})
 
   target_compile_options (${WASM_MODULE} PRIVATE
     -pthread -ftls-model=local-exec)
@@ -34,5 +35,5 @@ function (compile_sample SOURCE_FILE)
   )
 endfunction ()
 
-compile_sample(no_pthread.c)
-compile_sample(exception_propagation.c)
+compile_sample(no_pthread.c wasi_thread_start.S)
+compile_sample(exception_propagation.c wasi_thread_start.S)

samples/wasi-threads/wasm-apps/exception_propagation.c

@@ -9,15 +9,21 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <assert.h>
-#include <wasi/api.h>
 #include <semaphore.h>
 #include <stdbool.h>
 #include <unistd.h>
 
+#include "wasi_thread_start.h"
+
 #define TIMEOUT_SECONDS 10
 #define NUM_THREADS 3
 static sem_t sem;
 
+typedef struct {
+    start_args_t base;
+    bool throw_exception;
+} shared_t;
+
 void
 run_long_task()
 {
@@ -26,12 +32,12 @@ run_long_task()
         sleep(1);
 }
 
-__attribute__((export_name("wasi_thread_start"))) void
-wasi_thread_start(int thread_id, int *start_arg)
+void
+__wasi_thread_start_C(int thread_id, int *start_arg)
 {
-    bool has_to_throw_exception = (bool)start_arg;
+    shared_t *data = (shared_t *)start_arg;
 
-    if (has_to_throw_exception) {
+    if (data->throw_exception) {
         // Wait for all other threads (including main thread) to be ready
         printf("Waiting before throwing exception\n");
         for (int i = 0; i < NUM_THREADS; i++)
@@ -52,26 +58,36 @@ wasi_thread_start(int thread_id, int *start_arg)
 int
 main(int argc, char **argv)
 {
-    int thread_id = -1;
+    int thread_id = -1, i;
+    shared_t data[NUM_THREADS] = { 0 };
+
     if (sem_init(&sem, 0, 0) != 0) {
         printf("Failed to init semaphore\n");
         return EXIT_FAILURE;
     }
 
-    // Create a thread that throws an exception
-    thread_id = __wasi_thread_spawn((void *)true);
-    if (thread_id < 0) {
-        printf("Failed to create thread: %d\n", thread_id);
-        return EXIT_FAILURE;
+    for (i = 0; i < NUM_THREADS; i++) {
+        // No graceful memory free to simplify the example
+        if (!start_args_init(&data[i].base)) {
+            printf("Failed to allocate thread's stack\n");
+            return EXIT_FAILURE;
+        }
     }
 
-    // Create two additional threads to test exception propagation
-    thread_id = __wasi_thread_spawn((void *)false);
+    // Create a thread that throws an exception
+    data[0].throw_exception = true;
+    thread_id = __wasi_thread_spawn(&data[0]);
     if (thread_id < 0) {
         printf("Failed to create thread: %d\n", thread_id);
         return EXIT_FAILURE;
     }
-    thread_id = __wasi_thread_spawn((void *)false);
+    // Create two additional threads to test exception propagation
+    thread_id = __wasi_thread_spawn(&data[1]);
+    if (thread_id < 0) {
+        printf("Failed to create thread: %d\n", thread_id);
+        return EXIT_FAILURE;
+    }
+    thread_id = __wasi_thread_spawn(&data[2]);
     if (thread_id < 0) {
         printf("Failed to create thread: %d\n", thread_id);
         return EXIT_FAILURE;

samples/wasi-threads/wasm-apps/no_pthread.c

@@ -9,18 +9,20 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <assert.h>
-#include <wasi/api.h>
+
+#include "wasi_thread_start.h"
 
 static const int64_t SECOND = 1000 * 1000 * 1000;
 
 typedef struct {
+    start_args_t base;
     int th_ready;
     int value;
     int thread_id;
 } shared_t;
 
-__attribute__((export_name("wasi_thread_start"))) void
-wasi_thread_start(int thread_id, int *start_arg)
+void
+__wasi_thread_start_C(int thread_id, int *start_arg)
 {
     shared_t *data = (shared_t *)start_arg;
 
@@ -38,18 +40,26 @@ wasi_thread_start(int thread_id, int *start_arg)
 int
 main(int argc, char **argv)
 {
-    shared_t data = { 0, 52, -1 };
+    shared_t data = { { NULL }, 0, 52, -1 };
     int thread_id;
+    int ret = EXIT_SUCCESS;
+
+    if (!start_args_init(&data.base)) {
+        printf("Stack allocation for thread failed\n");
+        return EXIT_FAILURE;
+    }
 
     thread_id = __wasi_thread_spawn(&data);
     if (thread_id < 0) {
         printf("Failed to create thread: %d\n", thread_id);
-        return EXIT_FAILURE;
+        ret = EXIT_FAILURE;
+        goto final;
     }
 
     if (__builtin_wasm_memory_atomic_wait32(&data.th_ready, 0, SECOND) == 2) {
         printf("Timeout\n");
-        return EXIT_FAILURE;
+        ret = EXIT_FAILURE;
+        goto final;
     }
 
     printf("Thread completed, new value: %d, thread id: %d\n", data.value,
@@ -57,5 +67,8 @@ main(int argc, char **argv)
 
     assert(thread_id == data.thread_id);
 
-    return EXIT_SUCCESS;
+final:
+    start_args_deinit(&data.base);
+
+    return ret;
 }

samples/wasi-threads/wasm-apps/wasi_thread_start.S

@@ -0,0 +1,22 @@
+# A slightly modified copy of the wasi-libc implementation
+# https://github.com/WebAssembly/wasi-libc/pull/376/
+	.globaltype	__stack_pointer, i32
+	.functype	__wasi_thread_start_C (i32, i32) -> ()
+
+	.globl	wasi_thread_start
+
+wasi_thread_start:
+	.functype	wasi_thread_start (i32, i32) -> ()
+
+	# Set up the minimum C environment.
+	# Note: offsetof(start_arg, stack) == 0
+	local.get   1  # start_arg
+	i32.load    0  # stack
+	global.set  __stack_pointer
+
+	# 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

samples/wasi-threads/wasm-apps/wasi_thread_start.h

@@ -0,0 +1,32 @@
+/*
+ * Copyright (C) 2022 Amazon.com Inc. or its affiliates. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+#ifndef WASI_THREAD_START_H
+#define WASI_THREAD_START_H
+
+#define STACK_SIZE 1024
+
+typedef struct {
+    void *stack;
+} start_args_t;
+
+static inline int
+start_args_init(start_args_t *start_args)
+{
+    start_args->stack = malloc(STACK_SIZE);
+    if (!start_args->stack) {
+        return 0;
+    }
+
+    start_args->stack += STACK_SIZE;
+    return 1;
+}
+
+static inline void
+start_args_deinit(start_args_t *start_args)
+{
+    free(start_args->stack - STACK_SIZE);
+}
+
+#endif