compiler: work around AVR atomics bugs

The AVR backend has several critical atomics bugs.
This change invokes libcalls for all atomic operations on AVR.
Now `testdata/atomic.go` compiles and runs correctly.

Author: niaow

compiler/atomic.go

@@ -1,6 +1,7 @@
 package compiler
 
 import (
+	"fmt"
 	"strings"
 
 	"golang.org/x/tools/go/ssa"
@@ -16,6 +17,20 @@ func (b *builder) createAtomicOp(call *ssa.CallCommon) (llvm.Value, bool) {
 	case "AddInt32", "AddInt64", "AddUint32", "AddUint64", "AddUintptr":
 		ptr := b.getValue(call.Args[0])
 		val := b.getValue(call.Args[1])
+		if strings.HasPrefix(b.Triple, "avr") {
+			// AtomicRMW does not work on AVR as intended:
+			// - There are some register allocation issues (fixed by https://reviews.llvm.org/D97127 which is not yet in a usable LLVM release)
+			// - The result is the new value instead of the old value
+			vType := val.Type()
+			name := fmt.Sprintf("__sync_fetch_and_add_%d", vType.IntTypeWidth()/8)
+			fn := b.mod.NamedFunction(name)
+			if fn.IsNil() {
+				fn = llvm.AddFunction(b.mod, name, llvm.FunctionType(vType, []llvm.Type{ptr.Type(), vType}, false))
+			}
+			oldVal := b.createCall(fn, []llvm.Value{ptr, val}, "")
+			// Return the new value, not the original value returned.
+			return b.CreateAdd(oldVal, val, ""), true
+		}
 		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpAdd, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
 		// Return the new value, not the original value returned by atomicrmw.
 		return b.CreateAdd(oldVal, val, ""), true
@@ -73,6 +88,23 @@ func (b *builder) createAtomicOp(call *ssa.CallCommon) (llvm.Value, bool) {
 	case "StoreInt32", "StoreInt64", "StoreUint32", "StoreUint64", "StoreUintptr", "StorePointer":
 		ptr := b.getValue(call.Args[0])
 		val := b.getValue(call.Args[1])
+		if strings.HasPrefix(b.Triple, "avr") {
+			// SelectionDAGBuilder is currently missing the "are unaligned atomics allowed" check for stores.
+			vType := val.Type()
+			isPointer := vType.TypeKind() == llvm.PointerTypeKind
+			if isPointer {
+				// libcalls only supports integers, so cast to an integer.
+				vType = b.uintptrType
+				val = b.CreatePtrToInt(val, vType, "")
+				ptr = b.CreateBitCast(ptr, llvm.PointerType(vType, 0), "")
+			}
+			name := fmt.Sprintf("__atomic_store_%d", vType.IntTypeWidth()/8)
+			fn := b.mod.NamedFunction(name)
+			if fn.IsNil() {
+				fn = llvm.AddFunction(b.mod, name, llvm.FunctionType(vType, []llvm.Type{ptr.Type(), vType, b.uintptrType}, false))
+			}
+			return b.createCall(fn, []llvm.Value{ptr, val, llvm.ConstInt(b.uintptrType, 5, false)}, ""), true
+		}
 		store := b.CreateStore(val, ptr)
 		store.SetOrdering(llvm.AtomicOrderingSequentiallyConsistent)
 		store.SetAlignment(b.targetData.PrefTypeAlignment(val.Type())) // required

main_test.go

@@ -140,7 +140,8 @@ func TestBuild(t *testing.T) {
 		for _, t := range tests {
 			switch t {
 			case "atomic.go":
-				// Not supported due to unaligned atomic accesses.
+				// Requires GCC 11.2.0 or above for interface comparison.
+				// https://github.com/gcc-mirror/gcc/commit/f30dd607669212de135dec1f1d8a93b8954c327c
 
 			case "reflect.go":
 				// Reflect tests do not work due to type code issues.

src/runtime/arch_avr.go

@@ -1,7 +1,10 @@
+//go:build avr
 // +build avr
 
 package runtime
 
+import "runtime/interrupt"
+
 const GOARCH = "arm" // avr pretends to be arm
 
 // The bitness of the CPU (e.g. 8, 32, 64).
@@ -16,3 +19,19 @@ func align(ptr uintptr) uintptr {
 func getCurrentStackPointer() uintptr {
 	return uintptr(stacksave())
 }
+
+// The safest thing to do here would just be to disable interrupts for
+// procPin/procUnpin. Note that a global variable is safe in this case, as any
+// access to procPinnedMask will happen with interrupts disabled.
+
+var procPinnedMask interrupt.State
+
+//go:linkname procPin sync/atomic.runtime_procPin
+func procPin() {
+	procPinnedMask = interrupt.Disable()
+}
+
+//go:linkname procUnpin sync/atomic.runtime_procUnpin
+func procUnpin() {
+	interrupt.Restore(procPinnedMask)
+}