runtime: lock output in print/println

This ensures that calls to print/println happening in different threads
are not interleaved. It's a task.PMutex, so this should only change
things when threading is used.

This matches the Go compiler, which does the same thing:
https://godbolt.org/z/na5KzE7en

The locks are not recursive, which means that we need to be careful to
not call `print` or `println` inside a runtime.print* implementation,
inside putchar (recursively), and inside signal handlers. Making them
recursive might be useful to do in the future, but it's not really
necessary.

7 files changed, 90 insertions, 24 deletions

diff --git a/compiler/compiler.go b/compiler/compiler.go
index 8b5f0d1cb..28ec312dd 100644
--- a/compiler/compiler.go
+++ b/compiler/compiler.go
@@ -1686,6 +1686,7 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		b.createRuntimeInvoke("_panic", argValues, "")
 		return llvm.Value{}, nil
 	case "print", "println":
+		b.createRuntimeCall("printlock", nil, "")
 		for i, value := range argValues {
 			if i >= 1 && callName == "println" {
 				b.createRuntimeCall("printspace", nil, "")
@@ -1746,6 +1747,7 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		if callName == "println" {
 			b.createRuntimeCall("printnl", nil, "")
 		}
+		b.createRuntimeCall("printunlock", nil, "")
 		return llvm.Value{}, nil // print() or println() returns void
 	case "real":
 		cplx := argValues[0]
diff --git a/compiler/testdata/defer-cortex-m-qemu.ll b/compiler/testdata/defer-cortex-m-qemu.ll
index cdcfd0ea4..f1fbad7f8 100644
--- a/compiler/testdata/defer-cortex-m-qemu.ll
+++ b/compiler/testdata/defer-cortex-m-qemu.ll
@@ -122,12 +122,18 @@ declare void @runtime.destroyDeferFrame(ptr dereferenceable_or_null(24), ptr) #2
 ; Function Attrs: nounwind
 define internal void @"main.deferSimple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
+declare void @runtime.printlock(ptr) #2
+
 declare void @runtime.printint32(i32, ptr) #2
 
+declare void @runtime.printunlock(ptr) #2
+
 ; Function Attrs: nounwind
 define hidden void @main.deferMultiple(ptr %context) unnamed_addr #1 {
 entry:
@@ -250,14 +256,18 @@ rundefers.end7:                                   ; preds = %rundefers.loophead1
 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$2"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 5, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
diff --git a/compiler/testdata/goroutine-cortex-m-qemu-tasks.ll b/compiler/testdata/goroutine-cortex-m-qemu-tasks.ll
index a57bb20f3..819f01adb 100644
--- a/compiler/testdata/goroutine-cortex-m-qemu-tasks.ll
+++ b/compiler/testdata/goroutine-cortex-m-qemu-tasks.ll
@@ -70,7 +70,9 @@ entry:
   %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
   call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
   %2 = load i32, ptr %n, align 4
+  call void @runtime.printlock(ptr undef) #9
   call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #9
   ret void
 }
 
@@ -91,8 +93,12 @@ entry:
   ret void
 }
 
+declare void @runtime.printlock(ptr) #2
+
 declare void @runtime.printint32(i32, ptr) #2
 
+declare void @runtime.printunlock(ptr) #2
+
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
 entry:
diff --git a/compiler/testdata/goroutine-wasm-asyncify.ll b/compiler/testdata/goroutine-wasm-asyncify.ll
index c4af76037..87c7381f2 100644
--- a/compiler/testdata/goroutine-wasm-asyncify.ll
+++ b/compiler/testdata/goroutine-wasm-asyncify.ll
@@ -76,7 +76,9 @@ entry:
   store ptr %n, ptr %1, align 4
   call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
   %2 = load i32, ptr %n, align 4
+  call void @runtime.printlock(ptr undef) #9
   call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #9
   ret void
 }
 
@@ -98,8 +100,12 @@ entry:
   unreachable
 }
 
+declare void @runtime.printlock(ptr) #1
+
 declare void @runtime.printint32(i32, ptr) #1
 
+declare void @runtime.printunlock(ptr) #1
+
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #2 {
 entry:
diff --git a/src/runtime/print.go b/src/runtime/print.go
index ef9117ff3..a4de46025 100644
--- a/src/runtime/print.go
+++ b/src/runtime/print.go
@@ -1,6 +1,7 @@
 package runtime
 
 import (
+	"internal/task"
 	"unsafe"
 )
 
@@ -8,6 +9,18 @@ type stringer interface {
 	String() string
 }
 
+// Lock to make sure print calls do not interleave.
+// This is a no-op lock on systems that do not have parallelism.
+var printLock task.PMutex
+
+func printlock() {
+	printLock.Lock()
+}
+
+func printunlock() {
+	printLock.Unlock()
+}
+
 //go:nobounds
 func printstring(s string) {
 	for i := 0; i < len(s); i++ {
@@ -293,67 +306,84 @@ func printnl() {
 func printitf(msg interface{}) {
 	switch msg := msg.(type) {
 	case bool:
-		print(msg)
+		printbool(msg)
 	case int:
-		print(msg)
+		switch unsafe.Sizeof(msg) {
+		case 8:
+			printint64(int64(msg))
+		case 4:
+			printint32(int32(msg))
+		}
 	case int8:
-		print(msg)
+		printint8(msg)
 	case int16:
-		print(msg)
+		printint16(msg)
 	case int32:
-		print(msg)
+		printint32(msg)
 	case int64:
-		print(msg)
+		printint64(msg)
 	case uint:
-		print(msg)
+		switch unsafe.Sizeof(msg) {
+		case 8:
+			printuint64(uint64(msg))
+		case 4:
+			printuint32(uint32(msg))
+		}
 	case uint8:
-		print(msg)
+		printuint8(msg)
 	case uint16:
-		print(msg)
+		printuint16(msg)
 	case uint32:
-		print(msg)
+		printuint32(msg)
 	case uint64:
-		print(msg)
+		printuint64(msg)
 	case uintptr:
-		print(msg)
+		printuintptr(msg)
 	case float32:
-		print(msg)
+		printfloat32(msg)
 	case float64:
-		print(msg)
+		printfloat64(msg)
 	case complex64:
-		print(msg)
+		printcomplex64(msg)
 	case complex128:
-		print(msg)
+		printcomplex128(msg)
 	case string:
-		print(msg)
+		printstring(msg)
 	case error:
-		print(msg.Error())
+		printstring(msg.Error())
 	case stringer:
-		print(msg.String())
+		printstring(msg.String())
 	default:
 		// cast to underlying type
 		itf := *(*_interface)(unsafe.Pointer(&msg))
 		putchar('(')
 		printuintptr(uintptr(itf.typecode))
 		putchar(':')
-		print(itf.value)
+		printptr(uintptr(itf.value))
 		putchar(')')
 	}
 }
 
 func printmap(m *hashmap) {
-	print("map[")
+	printstring("map[")
 	if m == nil {
-		print("nil")
+		printstring("nil")
 	} else {
-		print(uint(m.count))
+		switch unsafe.Sizeof(m.count) {
+		case 8:
+			printuint64(uint64(m.count))
+		case 4:
+			printuint32(uint32(m.count))
+		case 2:
+			printuint16(uint16(m.count))
+		}
 	}
 	putchar(']')
 }
 
 func printptr(ptr uintptr) {
 	if ptr == 0 {
-		print("nil")
+		printstring("nil")
 		return
 	}
 	putchar('0')
diff --git a/testdata/print.go b/testdata/print.go
index 7f7f843c4..5156ad58e 100644
--- a/testdata/print.go
+++ b/testdata/print.go
@@ -37,6 +37,12 @@ func main() {
 
 	// print interface
 	println(interface{}(nil))
+	println(interface{}(true))
+	println(interface{}("foobar"))
+	println(interface{}(int64(-3)))
+	println(interface{}(uint64(3)))
+	println(interface{}(int(-3)))
+	println(interface{}(uint(3)))
 
 	// print map
 	println(map[string]int{"three": 3, "five": 5})
diff --git a/testdata/print.txt b/testdata/print.txt
index 116de945d..3a88cf91e 100644
--- a/testdata/print.txt
+++ b/testdata/print.txt
@@ -19,6 +19,12 @@ a b c
 +3.140000e+000
 (+5.000000e+000+1.234500e+000i)
 (0:nil)
+true
+foobar
+-3
+3
+-3
+3
 map[2]
 true false
 [0/0]nil