Reduce overhead of AlignedAtomic types

AlignedAtomicUint64 is 15 bytes and it takes 16 bytes in structures.  On
32-bit systems, variables and structure fields is guaranteed to be
32-bit aligned and this means that we need only 12 bytes to find 8
contiguous bytes.

3 files changed, 51 insertions, 11 deletions

diff --git a/pkg/atomicbitops/BUILD b/pkg/atomicbitops/BUILD
index 11072d4de..02c0e52b9 100644
--- a/pkg/atomicbitops/BUILD
+++ b/pkg/atomicbitops/BUILD
@@ -18,7 +18,10 @@ go_library(
 go_test(
     name = "atomicbitops_test",
     size = "small",
-    srcs = ["atomicbitops_test.go"],
+    srcs = [
+        "aligned_test.go",
+        "atomicbitops_test.go",
+    ],
     library = ":atomicbitops",
     deps = ["//pkg/sync"],
 )
diff --git a/pkg/atomicbitops/aligned_32bit_unsafe.go b/pkg/atomicbitops/aligned_32bit_unsafe.go
index 383f81ff2..a17d317cc 100644
--- a/pkg/atomicbitops/aligned_32bit_unsafe.go
+++ b/pkg/atomicbitops/aligned_32bit_unsafe.go
@@ -34,14 +34,15 @@ import (
 //
 // +stateify savable
 type AlignedAtomicInt64 struct {
-	value [15]byte
+	value   int64
+	value32 int32
 }
 
 func (aa *AlignedAtomicInt64) ptr() *int64 {
-	// In the 15-byte aa.value, there are guaranteed to be 8 contiguous
-	// bytes with 64-bit alignment. We find an address in this range by
-	// adding 7, then clear the 3 least significant bits to get its start.
-	return (*int64)(unsafe.Pointer((uintptr(unsafe.Pointer(&aa.value[0])) + 7) &^ 7))
+	// On 32-bit systems, aa.value is is guaranteed to be 32-bit aligned.
+	// It means that in the 12-byte aa.value, there are guaranteed to be 8
+	// contiguous bytes with 64-bit alignment.
+	return (*int64)(unsafe.Pointer((uintptr(unsafe.Pointer(&aa.value)) + 4) &^ 7))
 }
 
 // Load is analagous to atomic.LoadInt64.
@@ -71,14 +72,15 @@ func (aa *AlignedAtomicInt64) Add(v int64) int64 {
 //
 // +stateify savable
 type AlignedAtomicUint64 struct {
-	value [15]byte
+	value   uint64
+	value32 uint32
 }
 
 func (aa *AlignedAtomicUint64) ptr() *uint64 {
-	// In the 15-byte aa.value, there are guaranteed to be 8 contiguous
-	// bytes with 64-bit alignment. We find an address in this range by
-	// adding 7, then clear the 3 least significant bits to get its start.
-	return (*uint64)(unsafe.Pointer((uintptr(unsafe.Pointer(&aa.value[0])) + 7) &^ 7))
+	// On 32-bit systems, aa.value is is guaranteed to be 32-bit aligned.
+	// It means that in the 12-byte aa.value, there are guaranteed to be 8
+	// contiguous bytes with 64-bit alignment.
+	return (*uint64)(unsafe.Pointer((uintptr(unsafe.Pointer(&aa.value)) + 4) &^ 7))
 }
 
 // Load is analagous to atomic.LoadUint64.
diff --git a/pkg/atomicbitops/aligned_test.go b/pkg/atomicbitops/aligned_test.go
new file mode 100644
index 000000000..e7123d2b8
--- /dev/null
+++ b/pkg/atomicbitops/aligned_test.go
@@ -0,0 +1,35 @@
+// Copyright 2021 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package atomicbitops
+
+import (
+	"testing"
+)
+
+func TestAtomiciInt64(t *testing.T) {
+	v := struct {
+		v8  int8
+		v64 AlignedAtomicInt64
+	}{}
+	v.v64.Add(1)
+}
+
+func TestAtomicUint64(t *testing.T) {
+	v := struct {
+		v8  uint8
+		v64 AlignedAtomicUint64
+	}{}
+	v.v64.Add(1)
+}