Enable stdio locking under Wasm Workers (#26489)

This change removes the use of `pthread_self` from musl's internal file
locking mechanism, while enabled libc's internal threading state in once
the first worker is started. The test changed here come from #18201

This is needed as part of my effort to remove all pthread symbols from
the Wasm Workers builds.

Author: sbc100

system/lib/libc/emscripten_internal.h

@@ -10,6 +10,8 @@
  * for `tools/gen_sig_info.py` to work.   This file contains declarations for
  * functions that are not declared in any other public or private header.
  */
+#ifndef __EMSCRIPTEN_INTERNAL_H__
+#define __EMSCRIPTEN_INTERNAL_H__
 
 #include <emscripten/em_macros.h>
 #include <emscripten/proxying.h>
@@ -148,3 +150,5 @@ int _poll_js(void* fds, int nfds, int timeout, void* ctx, void* arg);
 #ifdef __cplusplus
 }
 #endif
+
+#endif /* __EMSCRIPTEN_INTERNAL_H__ */

system/lib/libc/musl/src/internal/pthread_impl.h

@@ -275,4 +275,13 @@ extern hidden unsigned __default_guardsize;
 
 #define __ATTRP_C11_THREAD ((void*)(uintptr_t)-1)
 
+#ifdef __EMSCRIPTEN_SHARED_MEMORY__
+pid_t gettid(void);
+// Unlike `__pthread_self()->tid, `gettid` works under both wasm workers and
+// pthreads.
+#define CURRENT_THREAD_ID gettid()
+#else
+#define CURRENT_THREAD_ID __pthread_self()->tid
+#endif
+
 #endif

system/lib/libc/musl/src/internal/stdio_impl.h

@@ -6,7 +6,7 @@
 
 #define UNGET 8
 
-#if defined(__EMSCRIPTEN__) && !defined(_REENTRANT)
+#if defined(__EMSCRIPTEN__) && !defined(__EMSCRIPTEN_SHARED_MEMORY__)
 #define FFINALLOCK(f)
 #define FLOCK(f)
 #define FUNLOCK(f)

system/lib/libc/musl/src/linux/gettid.c

@@ -2,9 +2,7 @@
 #include <unistd.h>
 #include "pthread_impl.h"
 
-#ifdef __EMSCRIPTEN_WASM_WORKERS__
-#include <emscripten/wasm_worker.h>
-#elif defined(__EMSCRIPTEN_PTHREADS__)
+#ifdef __EMSCRIPTEN__
 weak int emscripten_wasm_worker_self_id();
 #endif
 
@@ -16,7 +14,14 @@ pid_t gettid(void)
 {
 #ifdef __EMSCRIPTEN_WASM_WORKERS__
 	// Offset the worker ID by 1 so we never return 0 from this function.
-	return emscripten_wasm_worker_self_id() + 1;
+	// Strangly we cannot assume the existence of emscripten_wasm_worker_self_id
+	// here because libc-ww is also used for `-sSHARED_MEMORY` builds (without
+	// libwasm_workers linked in.
+	if (emscripten_wasm_worker_self_id) {
+		return emscripten_wasm_worker_self_id() + 1;
+	} else {
+		return 42;
+	}
 #else
 #if defined(__EMSCRIPTEN_PTHREADS__)
 	// The pthread-variant of libc can also be used alongside wasm workers.

system/lib/libc/musl/src/stdio/__lockfile.c

@@ -3,8 +3,8 @@
 
 int __lockfile(FILE *f)
 {
-#if defined(__EMSCRIPTEN_PTHREADS__)
-	int owner = f->lock, tid = __pthread_self()->tid;
+#ifdef __EMSCRIPTEN_SHARED_MEMORY__
+	int owner = f->lock, tid = CURRENT_THREAD_ID;
 	if ((owner & ~MAYBE_WAITERS) == tid)
 		return 0;
 	owner = a_cas(&f->lock, 0, tid);
@@ -20,7 +20,7 @@ int __lockfile(FILE *f)
 
 void __unlockfile(FILE *f)
 {
-#if defined(__EMSCRIPTEN_PTHREADS__)
+#ifdef __EMSCRIPTEN_SHARED_MEMORY__
 	if (a_swap(&f->lock, 0) & MAYBE_WAITERS)
 		__wake(&f->lock, 1, 1);
 #endif

system/lib/libc/musl/src/stdio/getc.h

@@ -16,7 +16,7 @@ static int locking_getc(FILE *f)
 static inline int do_getc(FILE *f)
 {
 	int l = f->lock;
-	if (l < 0 || l && (l & ~MAYBE_WAITERS) == __pthread_self()->tid)
+	if (l < 0 || l && (l & ~MAYBE_WAITERS) == CURRENT_THREAD_ID)
 		return getc_unlocked(f);
 	return locking_getc(f);
 }

system/lib/libc/musl/src/stdio/putc.h

@@ -16,7 +16,7 @@ static int locking_putc(int c, FILE *f)
 static inline int do_putc(int c, FILE *f)
 {
 	int l = f->lock;
-	if (l < 0 || l && (l & ~MAYBE_WAITERS) == __pthread_self()->tid)
+	if (l < 0 || l && (l & ~MAYBE_WAITERS) == CURRENT_THREAD_ID)
 		return putc_unlocked(c, f);
 	return locking_putc(c, f);
 }

system/lib/wasm_worker/library_wasm_worker.c

@@ -1,3 +1,20 @@
+/*
+ * Copyright 2022 The Emscripten Authors.  All rights reserved.
+ * Emscripten is available under two separate licenses, the MIT license and the
+ * University of Illinois/NCSA Open Source License.  Both these licenses can be
+ * found in the LICENSE file.
+ */
+
+// libc files are compiled as -std=c99 which doesn't normally declare
+// max_align_t.
+#if __STDC_VERSION__ < 201112L
+#define __NEED_max_align_t
+#endif
+
+#include "libc.h"
+#include "stdio_impl.h"
+#include "emscripten_internal.h"
+
 #include <assert.h>
 #include <emscripten/wasm_worker.h>
 #include <emscripten/threading.h>
@@ -7,8 +24,6 @@
 #include <malloc.h>
 #include <sys/param.h> // For MAX()
 
-#include "emscripten_internal.h"
-
 #ifndef __EMSCRIPTEN_WASM_WORKERS__
 #error __EMSCRIPTEN_WASM_WORKERS__ should be defined when building this file!
 #endif
@@ -33,6 +48,15 @@ static void emscripten_wasm_worker_main_thread_initialize() {
   *sbrk_ptr += ROUND_UP(__builtin_wasm_tls_size(), SBRK_ALIGN);
 }
 
+static FILE *volatile dummy_file = 0;
+weak_alias(dummy_file, __stdin_used);
+weak_alias(dummy_file, __stdout_used);
+weak_alias(dummy_file, __stderr_used);
+
+static void init_file_lock(FILE *f) {
+  if (f && f->lock<0) f->lock = 0;
+}
+
 emscripten_wasm_worker_t emscripten_create_wasm_worker(void *stackPlusTLSAddress, size_t stackPlusTLSSize) {
   assert(stackPlusTLSAddress != 0);
   assert((uintptr_t)stackPlusTLSAddress % STACK_ALIGN == 0);
@@ -61,6 +85,22 @@ emscripten_wasm_worker_t emscripten_create_wasm_worker(void *stackPlusTLSAddress
     stackPlusTLSAddress = (void*)tlsBase;
     stackPlusTLSSize -= padding;
   }
+
+  if (!libc.threaded) {
+    for (FILE *f=*__ofl_lock(); f; f=f->next) {
+      init_file_lock(f);
+    }
+    __ofl_unlock();
+    init_file_lock(__stdin_used);
+    init_file_lock(__stdout_used);
+    init_file_lock(__stderr_used);
+    libc.threaded = 1;
+  }
+
+  // Unlike with ptheads, wasm workers never really exit and so this counter
+  // only going one way here.
+  if (!libc.threads_minus_1++) libc.need_locks = 1;
+
   return _emscripten_create_wasm_worker(stackPlusTLSAddress, stackPlusTLSSize);
 }
 

test/codesize/hello_wasm_worker_wasm.expected.js

@@ -1,75 +1,80 @@
-var c = Module, d = "em-ww" == globalThis.name, e, f, y, z, l, A, t;
+var c = Module, d = "em-ww" == globalThis.name, e = !!globalThis.WorkerGlobalScope, f, g, C, r, D, n, E, w;
 
 d && (onmessage = a => {
     onmessage = null;
-    e = a = a.data;
-    f = a.l;
-    g();
-    c ||= {};
-    c.wasm = a.j;
+    f = a = a.data;
+    g = a.o;
     h();
-    a.j = a.l = 0;
+    c ||= {};
+    c.wasm = a.m;
+    k();
+    a.m = a.o = 0;
 });
 
-function g() {}
+function h() {}
 
-d || (f = c.mem || new WebAssembly.Memory({
+d || (g = c.mem || new WebAssembly.Memory({
     initial: 256,
     maximum: 256,
     shared: !0
-}), g());
+}), h());
 
-var k = [], n = a => {
+var l = [], p = a => {
     a = a.data;
     let b = a._wsc;
-    b && l.get(b)(...a.x);
-}, p = a => {
-    k.push(a);
-}, q = {}, r = 1, u = (a, b) => {
-    let m = q[r] = new Worker(c.js, {
+    b && n.get(b)(...a.x);
+}, q = a => {
+    l.push(a);
+}, t = () => {
+    r(0, !e, !d, e && 1);
+}, u = {}, v = 1, x = (a, b) => {
+    let m = u[v] = new Worker(c.js, {
         name: "em-ww"
     });
     m.postMessage({
-        s: r,
-        j: t,
-        l: f,
-        m: a,
-        o: b
+        v: v,
+        m: w,
+        o: g,
+        s: a,
+        u: b
     });
-    m.onmessage = n;
-    return r++;
-}, v = () => !1, w = (a, b) => {
-    q[a].postMessage({
+    m.onmessage = p;
+    return v++;
+}, y = () => performance.now(), z = () => !1, A = (a, b) => {
+    u[a].postMessage({
         _wsc: b,
         x: []
     });
 };
 
-d && (q[0] = globalThis, addEventListener("message", p));
+d && (u[0] = globalThis, addEventListener("message", q));
 
-function x() {
+function B() {
     console.log("Hello from wasm worker!");
 }
 
-function h() {
-    A = {
-        b: u,
-        c: v,
-        d: w,
-        e: x,
-        a: f
+function k() {
+    E = {
+        e: t,
+        c: x,
+        b: y,
+        d: z,
+        f: A,
+        g: B,
+        a: g
     };
     WebAssembly.instantiate(c.wasm, {
-        a: A
+        a: E
     }).then((a => {
         var b = (a.instance || a).exports;
-        t = a.module || c.wasm;
-        y = b.g;
-        z = b.i;
-        l = b.h;
-        d ? (z(e.s, e.m, e.o), removeEventListener("message", p), k = k.forEach(n), addEventListener("message", n)) : b.f();
-        d || y();
+        w = a.module || c.wasm;
+        C = b.i;
+        r = b.k;
+        D = b.l;
+        n = b.j;
+        d ? (D(f.v, f.s, f.u), removeEventListener("message", q), l = l.forEach(p), addEventListener("message", p)) : b.h();
+        d || C();
     }));
 }
 
-d || h();
+d || k();

test/codesize/test_minimal_runtime_code_size_hello_wasm_worker.json

@@ -1,10 +1,10 @@
 {
   "a.html": 515,
   "a.html.gz": 355,
-  "a.js": 859,
-  "a.js.gz": 545,
-  "a.wasm": 1847,
-  "a.wasm.gz": 1071,
-  "total": 3221,
-  "total_gz": 1971
+  "a.js": 956,
+  "a.js.gz": 605,
+  "a.wasm": 2744,
+  "a.wasm.gz": 1530,
+  "total": 4215,
+  "total_gz": 2490
 }