runtime/internal/atomic: add early nil check on ARM

If nil, fault before taking the lock or calling into the kernel.

Change-Id: I013d78a5f9233c2a9197660025f679940655d384
Reviewed-on: https://go-review.googlesource.com/93636
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

src/runtime/internal/atomic/atomic_arm.go

@@ -109,6 +109,7 @@ func Cas64(addr *uint64, old, new uint64) bool {
 	if uintptr(unsafe.Pointer(addr))&7 != 0 {
 		*(*int)(nil) = 0 // crash on unaligned uint64
 	}
+	_ = *addr // if nil, fault before taking the lock
 	var ok bool
 	addrLock(addr).lock()
 	if *addr == old {
@@ -124,6 +125,7 @@ func Xadd64(addr *uint64, delta int64) uint64 {
 	if uintptr(unsafe.Pointer(addr))&7 != 0 {
 		*(*int)(nil) = 0 // crash on unaligned uint64
 	}
+	_ = *addr // if nil, fault before taking the lock
 	var r uint64
 	addrLock(addr).lock()
 	r = *addr + uint64(delta)
@@ -137,6 +139,7 @@ func Xchg64(addr *uint64, v uint64) uint64 {
 	if uintptr(unsafe.Pointer(addr))&7 != 0 {
 		*(*int)(nil) = 0 // crash on unaligned uint64
 	}
+	_ = *addr // if nil, fault before taking the lock
 	var r uint64
 	addrLock(addr).lock()
 	r = *addr
@@ -150,6 +153,7 @@ func Load64(addr *uint64) uint64 {
 	if uintptr(unsafe.Pointer(addr))&7 != 0 {
 		*(*int)(nil) = 0 // crash on unaligned uint64
 	}
+	_ = *addr // if nil, fault before taking the lock
 	var r uint64
 	addrLock(addr).lock()
 	r = *addr
@@ -162,6 +166,7 @@ func Store64(addr *uint64, v uint64) {
 	if uintptr(unsafe.Pointer(addr))&7 != 0 {
 		*(*int)(nil) = 0 // crash on unaligned uint64
 	}
+	_ = *addr // if nil, fault before taking the lock
 	addrLock(addr).lock()
 	*addr = v
 	addrLock(addr).unlock()

src/runtime/internal/atomic/sys_linux_arm.s

@@ -11,6 +11,9 @@ TEXT cas<>(SB),NOSPLIT,$0
 
 TEXT runtime∕internal∕atomic·Cas(SB),NOSPLIT,$0
 	MOVW	ptr+0(FP), R2
+	// trigger potential paging fault here,
+	// because we don't know how to traceback through __kuser_cmpxchg
+	MOVW    (R2), R0
 	MOVW	old+4(FP), R0
 loop:
 	MOVW	new+8(FP), R1