4 files changed, 42 insertions, 34 deletions
diff --git a/pkg/atomicbitops/aligned_32bit_unsafe.go b/pkg/atomicbitops/aligned_32bit_unsafe.go
index a17d317cc..0e4765c48 100644
--- a/pkg/atomicbitops/aligned_32bit_unsafe.go
+++ b/pkg/atomicbitops/aligned_32bit_unsafe.go
@@ -39,9 +39,9 @@ type AlignedAtomicInt64 struct {
 }
 
 func (aa *AlignedAtomicInt64) ptr() *int64 {
-	// 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.
+	// On 32-bit systems, aa.value 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))
 }
 
@@ -77,9 +77,9 @@ type AlignedAtomicUint64 struct {
 }
 
 func (aa *AlignedAtomicUint64) ptr() *uint64 {
-	// 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.
+	// On 32-bit systems, aa.value 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))
 }
 
diff --git a/pkg/atomicbitops/atomicbitops_amd64.s b/pkg/atomicbitops/atomicbitops_amd64.s
index 54c887ee5..6b9a67adc 100644
--- a/pkg/atomicbitops/atomicbitops_amd64.s
+++ b/pkg/atomicbitops/atomicbitops_amd64.s
@@ -16,28 +16,28 @@
 
 #include "textflag.h"
 
-TEXT ·AndUint32(SB),$0-12
-  MOVQ  addr+0(FP), BP
+TEXT ·AndUint32(SB),NOSPLIT,$0-12
+  MOVQ  addr+0(FP), BX
   MOVL  val+8(FP), AX
   LOCK
-  ANDL   AX, 0(BP)
+  ANDL   AX, 0(BX)
   RET
 
-TEXT ·OrUint32(SB),$0-12
-  MOVQ  addr+0(FP), BP
+TEXT ·OrUint32(SB),NOSPLIT,$0-12
+  MOVQ  addr+0(FP), BX
   MOVL  val+8(FP), AX
   LOCK
-  ORL   AX, 0(BP)
+  ORL   AX, 0(BX)
   RET
 
-TEXT ·XorUint32(SB),$0-12
-  MOVQ  addr+0(FP), BP
+TEXT ·XorUint32(SB),NOSPLIT,$0-12
+  MOVQ  addr+0(FP), BX
   MOVL  val+8(FP), AX
   LOCK
-  XORL   AX, 0(BP)
+  XORL   AX, 0(BX)
   RET
 
-TEXT ·CompareAndSwapUint32(SB),$0-20
+TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20
   MOVQ  addr+0(FP), DI
   MOVL  old+8(FP), AX
   MOVL  new+12(FP), DX
@@ -46,28 +46,28 @@ TEXT ·CompareAndSwapUint32(SB),$0-20
   MOVL  AX, ret+16(FP)
   RET
 
-TEXT ·AndUint64(SB),$0-16
-  MOVQ  addr+0(FP), BP
+TEXT ·AndUint64(SB),NOSPLIT,$0-16
+  MOVQ  addr+0(FP), BX
   MOVQ  val+8(FP), AX
   LOCK
-  ANDQ   AX, 0(BP)
+  ANDQ   AX, 0(BX)
   RET
 
-TEXT ·OrUint64(SB),$0-16
-  MOVQ  addr+0(FP), BP
+TEXT ·OrUint64(SB),NOSPLIT,$0-16
+  MOVQ  addr+0(FP), BX
   MOVQ  val+8(FP), AX
   LOCK
-  ORQ   AX, 0(BP)
+  ORQ   AX, 0(BX)
   RET
 
-TEXT ·XorUint64(SB),$0-16
-  MOVQ  addr+0(FP), BP
+TEXT ·XorUint64(SB),NOSPLIT,$0-16
+  MOVQ  addr+0(FP), BX
   MOVQ  val+8(FP), AX
   LOCK
-  XORQ   AX, 0(BP)
+  XORQ   AX, 0(BX)
   RET
 
-TEXT ·CompareAndSwapUint64(SB),$0-32
+TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32
   MOVQ  addr+0(FP), DI
   MOVQ  old+8(FP), AX
   MOVQ  new+16(FP), DX
diff --git a/pkg/atomicbitops/atomicbitops_arm64.s b/pkg/atomicbitops/atomicbitops_arm64.s
index 5c780851b..644a6bca5 100644
--- a/pkg/atomicbitops/atomicbitops_arm64.s
+++ b/pkg/atomicbitops/atomicbitops_arm64.s
@@ -16,7 +16,7 @@
 
 #include "textflag.h"
 
-TEXT ·AndUint32(SB),$0-12
+TEXT ·AndUint32(SB),NOSPLIT,$0-12
   MOVD    ptr+0(FP), R0
   MOVW    val+8(FP), R1
 again:
@@ -26,7 +26,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·OrUint32(SB),$0-12
+TEXT ·OrUint32(SB),NOSPLIT,$0-12
   MOVD    ptr+0(FP), R0
   MOVW    val+8(FP), R1
 again:
@@ -36,7 +36,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·XorUint32(SB),$0-12
+TEXT ·XorUint32(SB),NOSPLIT,$0-12
   MOVD    ptr+0(FP), R0
   MOVW    val+8(FP), R1
 again:
@@ -46,7 +46,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·CompareAndSwapUint32(SB),$0-20
+TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20
   MOVD addr+0(FP), R0
   MOVW old+8(FP), R1
   MOVW new+12(FP), R2
@@ -60,7 +60,7 @@ done:
   MOVW R3, prev+16(FP)
   RET
 
-TEXT ·AndUint64(SB),$0-16
+TEXT ·AndUint64(SB),NOSPLIT,$0-16
   MOVD    ptr+0(FP), R0
   MOVD    val+8(FP), R1
 again:
@@ -70,7 +70,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·OrUint64(SB),$0-16
+TEXT ·OrUint64(SB),NOSPLIT,$0-16
   MOVD    ptr+0(FP), R0
   MOVD    val+8(FP), R1
 again:
@@ -80,7 +80,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·XorUint64(SB),$0-16
+TEXT ·XorUint64(SB),NOSPLIT,$0-16
   MOVD    ptr+0(FP), R0
   MOVD    val+8(FP), R1
 again:
@@ -90,7 +90,7 @@ again:
   CBNZ    R3, again
   RET
 
-TEXT ·CompareAndSwapUint64(SB),$0-32
+TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32
   MOVD addr+0(FP), R0
   MOVD old+8(FP), R1
   MOVD new+16(FP), R2
diff --git a/pkg/atomicbitops/atomicbitops_noasm.go b/pkg/atomicbitops/atomicbitops_noasm.go
index 474c0c815..af6b1362e 100644
--- a/pkg/atomicbitops/atomicbitops_noasm.go
+++ b/pkg/atomicbitops/atomicbitops_noasm.go
@@ -21,6 +21,7 @@ import (
 	"sync/atomic"
 )
 
+//go:nosplit
 func AndUint32(addr *uint32, val uint32) {
 	for {
 		o := atomic.LoadUint32(addr)
@@ -31,6 +32,7 @@ func AndUint32(addr *uint32, val uint32) {
 	}
 }
 
+//go:nosplit
 func OrUint32(addr *uint32, val uint32) {
 	for {
 		o := atomic.LoadUint32(addr)
@@ -41,6 +43,7 @@ func OrUint32(addr *uint32, val uint32) {
 	}
 }
 
+//go:nosplit
 func XorUint32(addr *uint32, val uint32) {
 	for {
 		o := atomic.LoadUint32(addr)
@@ -51,6 +54,7 @@ func XorUint32(addr *uint32, val uint32) {
 	}
 }
 
+//go:nosplit
 func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) {
 	for {
 		prev = atomic.LoadUint32(addr)
@@ -63,6 +67,7 @@ func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) {
 	}
 }
 
+//go:nosplit
 func AndUint64(addr *uint64, val uint64) {
 	for {
 		o := atomic.LoadUint64(addr)
@@ -73,6 +78,7 @@ func AndUint64(addr *uint64, val uint64) {
 	}
 }
 
+//go:nosplit
 func OrUint64(addr *uint64, val uint64) {
 	for {
 		o := atomic.LoadUint64(addr)
@@ -83,6 +89,7 @@ func OrUint64(addr *uint64, val uint64) {
 	}
 }
 
+//go:nosplit
 func XorUint64(addr *uint64, val uint64) {
 	for {
 		o := atomic.LoadUint64(addr)
@@ -93,6 +100,7 @@ func XorUint64(addr *uint64, val uint64) {
 	}
 }
 
+//go:nosplit
 func CompareAndSwapUint64(addr *uint64, old, new uint64) (prev uint64) {
 	for {
 		prev = atomic.LoadUint64(addr)