WASIp3 Cooperative Threading (#799)

Currently a draft while the LLVM patches merge and I get all of this
ready for review

---------

Co-authored-by: Alex Crichton <alex@alexcrichton.com>

Author: TartanLlama

CMakeLists.txt

@@ -26,6 +26,15 @@ if (CMAKE_HOST_WIN32)
   set(CMAKE_INCLUDE_SYSTEM_FLAG_C "-isystem ")
 endif()
 
+set(TARGET_TRIPLE "wasm32-wasip1" CACHE STRING "WASI target to test")
+set(CMAKE_C_COMPILER_TARGET ${TARGET_TRIPLE})
+
+# We're building libc, and the compiler being used may not have a wasm sysroot
+# for libc, so skip this check. Otherwise the `project` call below attempts a
+# link to see if the compiler is working and if that fails it doesn't mean it
+# doesn't work. Let the build keep going to see what happens.
+set(CMAKE_C_COMPILER_WORKS 1)
+
 project(wasi-libc LANGUAGES C ASM)
 
 if(NOT CMAKE_C_COMPILER_ID MATCHES Clang)
@@ -36,7 +45,6 @@ message(STATUS "Found executable for `nm`: ${CMAKE_NM}")
 message(STATUS "Found executable for `ar`: ${CMAKE_AR}")
 message(STATUS "Found executable for `ranlib`: ${CMAKE_RANLIB}")
 
-set(TARGET_TRIPLE "wasm32-wasip1" CACHE STRING "WASI target to test")
 # Note: thin LTO here is just for experimentation. It has known issues:
 # - https://github.com/llvm/llvm-project/issues/91700
 # - https://github.com/llvm/llvm-project/issues/91711
@@ -51,6 +59,7 @@ option(BUILD_SHARED "Whether or not to build shared libraries" ON)
 option(CHECK_SYMBOLS "Whether or not to check the exported symbols of libc.a" OFF)
 set(WASI_SDK_VERSION "" CACHE STRING "Version information for wasi-sdk to embed in headers")
 option(ENABLE_WERROR "Whether to compile with `-Werror`" OFF)
+option(ENABLE_COOP_THREADS "Whether to compile with cooperative threads support" OFF)
 
 if(TARGET_TRIPLE MATCHES "-threads$")
   set(WASI p1)
@@ -86,10 +95,10 @@ else()
 endif()
 
 if(TARGET_TRIPLE MATCHES "-threads$")
-  set(THREADS ON)
   add_compile_options(-mthread-model posix -pthread -ftls-model=local-exec -matomics)
+elseif(ENABLE_COOP_THREADS)
+  add_compile_options(-mthread-model posix -pthread -ftls-model=local-exec)
 else()
-  set(THREADS OFF)
   add_compile_options(-mthread-model single)
 endif()
 
@@ -209,6 +218,28 @@ if (ENABLE_WERROR)
   add_compile_options(-Werror)
 endif()
 
+# =============================================================================
+# Detection of a libcall thread context for wasip3
+
+include(CheckSourceCompiles)
+set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+check_source_compiles(C [[
+  #ifndef __wasm_libcall_thread_context__
+  #error "missing define"
+  #endif
+]] HAVE_WASM_LIBCALL_THREAD_CONTEXT)
+
+if(ENABLE_COOP_THREADS)
+  if(NOT WASI STREQUAL "p3")
+    message(FATAL_ERROR "Cooperative threads are only supported on WASIp3")
+  endif()
+  if (NOT HAVE_WASM_LIBCALL_THREAD_CONTEXT)
+    message(FATAL_ERROR "Cooperative threads require clang 23 or later")
+  endif()
+  set(__wasi_cooperative_threads__ ON)
+else()
+endif()
+
 # =============================================================================
 # Helper functions for adding libraries.
 
@@ -259,7 +290,7 @@ function(add_internal_shared_library name)
   add_library(${name} SHARED EXCLUDE_FROM_ALL ${ARGN})
   set_pic(${name})
   target_link_libraries(${name} PUBLIC musl-top-half-interface)
-  add_dependencies(${name} sysroot-c builtins)
+  add_dependencies(${name} sysroot-c builtins sysroot-empty-pthread)
   target_compile_options(${name} PRIVATE -fvisibility=default)
   set_target_properties(${name} PROPERTIES NO_SONAME 1)
   clang_format_target(${name})

cmake/check-symbols.cmake

@@ -22,10 +22,12 @@ add_custom_command(
     ${CMAKE_CURRENT_LIST_DIR}/scripts/run-check-symbols.cmake
 )
 
-if(THREADS)
-  set(expected "${CMAKE_SOURCE_DIR}/expected/wasm32-wasip1-threads")
-elseif(WASI STREQUAL "p1")
-  set(expected "${CMAKE_SOURCE_DIR}/expected/wasm32-wasip1")
+if(WASI STREQUAL "p1")
+  if(TARGET_TRIPLE MATCHES "threads")
+    set(expected "${CMAKE_SOURCE_DIR}/expected/wasm32-wasip1-threads")
+  else()
+    set(expected "${CMAKE_SOURCE_DIR}/expected/wasm32-wasip1")
+  endif()
 elseif(WASI STREQUAL "p2")
   set(expected "${CMAKE_SOURCE_DIR}/expected/wasm32-wasip2")
 elseif(WASI STREQUAL "p3")

cmake/clang-format.cmake

@@ -21,7 +21,8 @@ set(formatted_sources)
 
 function(clang_format_file file)
   if (file MATCHES "__generated" OR
-      file MATCHES "wasip.\.c$")  # Skip auto-generated files
+      file MATCHES "wasip.\.c$" OR # Skip auto-generated files
+      file MATCHES "\.(s|S)$")     # Skip assembly files
     return()
   endif()
   cmake_path(ABSOLUTE_PATH file BASE_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} OUTPUT_VARIABLE src)
@@ -30,7 +31,7 @@ function(clang_format_file file)
   # excludes upstream projects like cloudlibc/musl, malloc implementations, etc.
   if (NOT EXISTS ${src}
     OR ${src} MATCHES "cloudlibc"
-    OR ${src} MATCHES "libc-top-half"
+    OR (${src} MATCHES "libc-top-half" AND NOT ${src} MATCHES "libc-top-half/musl/src/thread/coop-threads")
     OR ${src} MATCHES "fts/musl-fts"
     OR ${src} MATCHES "dlmalloc"
     OR ${src} MATCHES "emmalloc"

cmake/wasm-component-ld.cmake

@@ -10,7 +10,7 @@ if (NOT WASM_COMPONENT_LD_EXECUTABLE)
   ba_download(
     wasm-component-ld
     "https://github.com/bytecodealliance/wasm-component-ld"
-    "v0.5.21"
+    "v0.5.25"
   )
   ExternalProject_Get_Property(wasm-component-ld SOURCE_DIR)
   set(WASM_COMPONENT_LD_EXECUTABLE "${SOURCE_DIR}/wasm-component-ld")

expected/wasm32-wasip1-threads/defined-symbols.txt

@@ -13,10 +13,9 @@ __SIG_ERR
 __SIG_IGN
 __acquire_ptc
 __aio_close
+__allocate_tls
 __asctime_r
 __assert_fail
-__at_quick_exit_lockptr
-__atexit_lockptr
 __c_dot_utf8
 __c_dot_utf8_locale
 __c_locale
@@ -81,6 +80,7 @@ __ftello
 __ftello_unlocked
 __funcs_on_exit
 __funcs_on_quick_exit
+__futexwait
 __futimesat
 __fwritable
 __fwritex
@@ -92,6 +92,7 @@ __getopt_msg
 __gmtime_r
 __hwcap
 __inet_aton
+__inhibit_ptc
 __init_ssp
 __init_tp
 __intscan
@@ -148,7 +149,6 @@ __libc_free
 __libc_malloc
 __loc_is_allocated
 __locale_lock
-__locale_lockptr
 __localtime_r
 __lock
 __lockfile
@@ -219,7 +219,6 @@ __pthread_tsd_size
 __putenv
 __qsort_r
 __rand48_step
-__random_lockptr
 __reallocarray
 __register_locked_file
 __release_ptc
@@ -248,7 +247,6 @@ __stdin_used
 __stdio_close
 __stdio_exit
 __stdio_exit_needed
-__stdio_ofl_lockptr
 __stdio_read
 __stdio_seek
 __stdio_write
@@ -308,6 +306,7 @@ __unlockfile
 __uselocale
 __utc
 __wait
+__wake
 __wasi_args_get
 __wasi_args_sizes_get
 __wasi_clock_res_get
@@ -359,7 +358,6 @@ __wasi_thread_start_C
 __wasilibc_access
 __wasilibc_cwd
 __wasilibc_cwd_lock
-__wasilibc_cwd_unlock
 __wasilibc_deinitialize_environ
 __wasilibc_dttoif
 __wasilibc_enable_futex_busywait_on_current_thread
@@ -946,6 +944,9 @@ modff
 modfl
 mprotect
 mrand48
+mtx_destroy
+mtx_init
+mtx_timedlock
 munmap
 nan
 nanf

libc-bottom-half/CMakeLists.txt

@@ -179,6 +179,13 @@ if (WASI STREQUAL "p3")
   )
 endif()
 
+if (HAVE_WASM_LIBCALL_THREAD_CONTEXT)
+  list(APPEND bottom_half_sources
+    sources/__libcall_thread_context.S
+    sources/__wasm_init_task.S
+  )
+endif()
+
 # Don't export symbols of generated code in shared libraries, so specifically
 # pass `-fvisibility=hidden`.
 set_source_files_properties(sources/wasip2.c PROPERTIES

libc-bottom-half/cloudlibc/src/libc/sched/sched_yield.c

@@ -7,7 +7,15 @@
 #include <sched.h>
 
 int sched_yield(void) {
-#ifdef __wasip1__
+#ifdef __wasi_cooperative_threads__
+  #ifdef __wasip3__
+    wasip3_thread_yield();
+    return 0;
+  #else
+    #error "Unknown WASI version"
+  #endif
+
+#elif defined(__wasip1__)
   __wasi_errno_t error = __wasi_sched_yield();
   if (error != 0) {
     errno = error;

libc-bottom-half/cloudlibc/src/libc/unistd/pwrite.c

@@ -83,7 +83,7 @@ ssize_t pwrite(int fildes, const void *buf, size_t nbyte, off_t offset) {
     &closed);
   filesystem_stream_u8_drop_writable(writer);
 
-  // Wait for the subtask to resolve now that the writer half is closed and if
+  // Wait for the future to resolve now that the writer half is closed and if
   // we failed to write bytes (0 bytes written) and the result is an error we
   // can return -1.
   filesystem_result_void_error_code_t result;

libc-bottom-half/crt/crt1-command.c

@@ -9,6 +9,14 @@ extern void __wasm_call_ctors(void);
 extern int __main_void(void);
 extern void __wasm_call_dtors(void);
 
+#ifdef __wasm_libcall_thread_context__
+// Force __wasm_init_task and __wasm_init_async_task to be linked in for wasip3
+extern void __wasm_init_task(void);
+extern void __wasm_init_async_task(void);
+__attribute__((used)) void *__wasm_init_task_ref = __wasm_init_task;
+__attribute__((used)) void *__wasm_init_async_task_ref = __wasm_init_async_task;
+#endif
+
 #if defined(__wasip1__)
 __attribute__((export_name("_start"))) void _start(void)
 #elif defined(__wasip2__)

libc-bottom-half/crt/crt1-reactor.c

@@ -4,6 +4,15 @@
 extern void __wasi_init_tp(void);
 extern void __wasm_call_ctors(void);
 
+#ifdef __wasm_libcall_thread_context__
+// Force __wasm_init_task and __wasm_init_async_task to be linked in for wasip3
+extern void __wasm_init_task(void);
+extern void __wasm_init_async_task(void);
+__attribute__((used)) static void *__wasm_init_task_ref = __wasm_init_task;
+__attribute__((used)) static void *__wasm_init_async_task_ref =
+    __wasm_init_async_task;
+#endif
+
 __attribute__((export_name("_initialize"))) void _initialize(void) {
 #if defined(_REENTRANT)
   static volatile atomic_int initialized = 0;