runtime, sycall/js: add support for callbacks from JavaScript

This commit adds support for JavaScript callbacks back into
WebAssembly. This is experimental API, just like the rest of the
syscall/js package. The time package now also uses this mechanism
to properly support timers without resorting to a busy loop.

JavaScript code can call into the same entry point multiple times.
The new RUN register is used to keep track of the program's
run state. Possible values are: starting, running, paused and exited.
If no goroutine is ready any more, the scheduler can put the
program into the "paused" state and the WebAssembly code will
stop running. When a callback occurs, the JavaScript code puts
the callback data into a queue and then calls into WebAssembly
to allow the Go code to continue running.

Updates golang#18892
Updates golang#25506

Change-Id: Ib8701cfa0536d10d69bd541c85b0e2a754eb54fb

misc/wasm/wasm_exec.js

@@ -120,6 +120,7 @@
 				go: {
 					// func wasmExit(code int32)
 					"runtime.wasmExit": (sp) => {
+						this.exited = true;
 						this.exit(mem().getInt32(sp + 8, true));
 					},
 
@@ -143,6 +144,11 @@
 						mem().setInt32(sp + 16, (msec % 1000) * 1000000, true);
 					},
 
+					// func scheduleCallback(delay int64)
+					"runtime.scheduleCallback": (sp) => {
+						setTimeout(() => { this._resolveCallbackPromise(); }, getInt64(sp + 8));
+					},
+
 					// func getRandomData(r []byte)
 					"runtime.getRandomData": (sp) => {
 						crypto.getRandomValues(loadSlice(sp + 8));
@@ -270,7 +276,19 @@
 
 		async run(instance) {
 			this._inst = instance;
-			this._values = [undefined, null, global, this._inst.exports.mem]; // TODO: garbage collection
+			this._values = [ // TODO: garbage collection
+				undefined,
+				null,
+				global,
+				this._inst.exports.mem,
+				() => {
+					if (this.exited) {
+						throw new Error('bad callback: Go program has already exited (hint: use "select {}")');
+					}
+					setTimeout(this._resolveCallbackPromise, 0); // make sure it is asynchronous
+				},
+			];
+			this.exited = false;
 
 			const mem = new DataView(this._inst.exports.mem.buffer)
 
@@ -304,7 +322,16 @@
 				offset += 8;
 			});
 
-			this._inst.exports.run(argc, argv);
+			while (true) {
+				const callbackPromise = new Promise((resolve) => {
+					this._resolveCallbackPromise = resolve;
+				});
+				this._inst.exports.run(argc, argv);
+				if (this.exited) {
+					break;
+				}
+				await callbackPromise;
+			}
 		}
 	}
 
@@ -319,6 +346,12 @@
 		go.env = process.env;
 		go.exit = process.exit;
 		WebAssembly.instantiate(fs.readFileSync(process.argv[2]), go.importObject).then((result) => {
+			process.on("exit", () => { // Node.js exits if no callback is pending
+				if (!go.exited) {
+					console.error("error: all goroutines asleep and no JavaScript callback pending - deadlock!");
+					process.exit(1);
+				}
+			});
 			return go.run(result.instance);
 		}).catch((err) => {
 			console.error(err);

src/cmd/internal/obj/wasm/a.out.go

@@ -219,6 +219,8 @@ const (
 	// However, it is not allowed to switch goroutines while inside of an ACALLNORESUME call.
 	ACALLNORESUME
 
+	ARETUNWIND
+
 	AMOVB
 	AMOVH
 	AMOVW
@@ -244,6 +246,7 @@ const (
 	REG_RET1
 	REG_RET2
 	REG_RET3
+	REG_RUN
 
 	// locals
 	REG_R0

src/cmd/internal/obj/wasm/anames.go

@@ -180,6 +180,7 @@ var Anames = []string{
 	"F64ReinterpretI64",
 	"RESUMEPOINT",
 	"CALLNORESUME",
+	"RETUNWIND",
 	"MOVB",
 	"MOVH",
 	"MOVW",

src/cmd/internal/obj/wasm/wasmobj.go

@@ -25,6 +25,7 @@ var Register = map[string]int16{
 	"RET1": REG_RET1,
 	"RET2": REG_RET2,
 	"RET3": REG_RET3,
+	"RUN":  REG_RUN,
 
 	"R0":  REG_R0,
 	"R1":  REG_R1,
@@ -487,7 +488,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 				p = appendp(p, AEnd) // end of Loop
 			}
 
-		case obj.ARET:
+		case obj.ARET, ARETUNWIND:
 			ret := *p
 			p.As = obj.ANOP
 
@@ -528,7 +529,14 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 			p = appendp(p, AI32Add)
 			p = appendp(p, ASet, regAddr(REG_SP))
 
-			// not switching goroutine, return 0
+			if ret.As == ARETUNWIND {
+				// function needs to unwind the WebAssembly stack, return 1
+				p = appendp(p, AI32Const, constAddr(1))
+				p = appendp(p, AReturn)
+				break
+			}
+
+			// not unwinding the WebAssembly stack, return 0
 			p = appendp(p, AI32Const, constAddr(0))
 			p = appendp(p, AReturn)
 		}
@@ -726,7 +734,7 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 			}
 			reg := p.From.Reg
 			switch {
-			case reg >= REG_PC_F && reg <= REG_RET3:
+			case reg >= REG_PC_F && reg <= REG_RUN:
 				w.WriteByte(0x23) // get_global
 				writeUleb128(w, uint64(reg-REG_PC_F))
 			case reg >= REG_R0 && reg <= REG_F15:
@@ -743,7 +751,7 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 			}
 			reg := p.To.Reg
 			switch {
-			case reg >= REG_PC_F && reg <= REG_RET3:
+			case reg >= REG_PC_F && reg <= REG_RUN:
 				w.WriteByte(0x24) // set_global
 				writeUleb128(w, uint64(reg-REG_PC_F))
 			case reg >= REG_R0 && reg <= REG_F15:

src/cmd/link/internal/wasm/asm.go

@@ -304,6 +304,7 @@ func writeGlobalSec(ctxt *ld.Link) {
 		I64, // 6: RET1
 		I64, // 7: RET2
 		I64, // 8: RET3
+		I32, // 9: RUN
 	}
 
 	writeUleb128(ctxt.Out, uint64(len(globalRegs))) // number of globals

src/runtime/lock_futex.go

@@ -229,3 +229,9 @@ func notetsleepg(n *note, ns int64) bool {
 	exitsyscall()
 	return ok
 }
+
+func pauseSchedulerUntilCallback() bool {
+	return false
+}
+
+func checkTimeouts() {}

src/runtime/lock_js.go

@@ -6,14 +6,22 @@
 
 package runtime
 
+import (
+	_ "unsafe"
+)
+
 // js/wasm has no support for threads yet. There is no preemption.
-// Waiting for a mutex or timeout is implemented as a busy loop
-// while allowing other goroutines to run.
+// Waiting for a mutex is as simple as allowing other goroutines
+// to run until the mutex got unlocked.
 
 const (
 	mutex_unlocked = 0
 	mutex_locked   = 1
 
+	note_cleared = 0
+	note_woken   = 1
+	note_timeout = 2
+
 	active_spin     = 4
 	active_spin_cnt = 30
 	passive_spin    = 1
@@ -34,16 +42,28 @@ func unlock(l *mutex) {
 }
 
 // One-time notifications.
+
+type noteWithTimeout struct {
+	gp       *g
+	deadline int64
+}
+
+var notes = make(map[*note]*g)
+var notesWithTimeout = make(map[*note]noteWithTimeout)
+
 func noteclear(n *note) {
-	n.key = 0
+	n.key = note_cleared
 }
 
 func notewakeup(n *note) {
-	if n.key != 0 {
-		print("notewakeup - double wakeup (", n.key, ")\n")
+	if n.key == note_woken {
 		throw("notewakeup - double wakeup")
 	}
-	n.key = 1
+	cleared := n.key == note_cleared
+	n.key = note_woken
+	if cleared {
+		goready(notes[n], 1)
+	}
 }
 
 func notesleep(n *note) {
@@ -62,14 +82,62 @@ func notetsleepg(n *note, ns int64) bool {
 		throw("notetsleepg on g0")
 	}
 
-	deadline := nanotime() + ns
-	for {
-		if n.key != 0 {
-			return true
+	if ns >= 0 {
+		delay := ns/1000000 + 2
+		if delay > 1<<31-1 {
+			delay = 1<<31 - 1 // cap to max int32
 		}
-		Gosched()
-		if ns >= 0 && nanotime() >= deadline {
-			return false
+
+		notes[n] = gp
+		notesWithTimeout[n] = noteWithTimeout{gp: gp, deadline: nanotime() + ns}
+		scheduleCallback(delay)
+		gopark(nil, nil, waitReasonSleep, traceEvNone, 1)
+		delete(notes, n)
+		delete(notesWithTimeout, n)
+
+		return n.key == note_woken
+	}
+
+	for n.key != note_woken {
+		notes[n] = gp
+		gopark(nil, nil, waitReasonZero, traceEvNone, 1)
+		delete(notes, n)
+	}
+	return true
+}
+
+func checkTimeouts() {
+	now := nanotime()
+	for n, nt := range notesWithTimeout {
+		if n.key == note_cleared && now > nt.deadline {
+			n.key = note_timeout
+			goready(nt.gp, 1)
 		}
 	}
 }
+
+var waitingForCallback []*g
+
+//go:linkname sleepUntilCallback syscall/js.sleepUntilCallback
+func sleepUntilCallback() {
+	waitingForCallback = append(waitingForCallback, getg())
+	gopark(nil, nil, waitReasonZero, traceEvNone, 1)
+}
+
+func pauseSchedulerUntilCallback() bool {
+	if len(waitingForCallback) == 0 {
+		return false
+	}
+
+	pause()
+	checkTimeouts()
+	for _, gp := range waitingForCallback {
+		goready(gp, 1)
+	}
+	waitingForCallback = waitingForCallback[:0]
+	return true
+}
+
+func pause()
+
+func scheduleCallback(delay int64)

src/runtime/lock_sema.go

@@ -282,3 +282,9 @@ func notetsleepg(n *note, ns int64) bool {
 	exitsyscall()
 	return ok
 }
+
+func pauseSchedulerUntilCallback() bool {
+	return false
+}
+
+func checkTimeouts() {}

src/runtime/proc.go

@@ -262,6 +262,7 @@ func forcegchelper() {
 // Gosched yields the processor, allowing other goroutines to run. It does not
 // suspend the current goroutine, so execution resumes automatically.
 func Gosched() {
+	checkTimeouts()
 	mcall(gosched_m)
 }
 
@@ -281,6 +282,9 @@ func goschedguarded() {
 // Reasons should be unique and descriptive.
 // Do not re-use reasons, add new ones.
 func gopark(unlockf func(*g, unsafe.Pointer) bool, lock unsafe.Pointer, reason waitReason, traceEv byte, traceskip int) {
+	if reason != waitReasonSleep {
+		checkTimeouts() // timeouts may expire while two goroutines keep the scheduler busy
+	}
 	mp := acquirem()
 	gp := mp.curg
 	status := readgstatus(gp)
@@ -2315,6 +2319,10 @@ stop:
 		return gp, false
 	}
 
+	if pauseSchedulerUntilCallback() {
+		goto top
+	}
+
 	// Before we drop our P, make a snapshot of the allp slice,
 	// which can change underfoot once we no longer block
 	// safe-points. We don't need to snapshot the contents because