internal/runtime/sys: improve DIT assembly

In EnableDIT, if DIT is already enabled, return early instead of
executing MSR and DSB/ISB, since they are not particularly cheap
instructions.

Also, if we have support for the SB (Speculation Barrier) instruction,
use it instead of DSB+ISB when enabling DIT, since SB is cheaper.

Change-Id: I1b3ecbd95ed42bfd10d646125704abf4e80b6d2e
Reviewed-on: https://go-review.googlesource.com/c/go/+/729800
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

src/internal/runtime/sys/dit_arm64.go

@@ -8,10 +8,13 @@ package sys
 
 import (
 	"internal/cpu"
+	"unsafe"
 )
 
 var DITSupported = cpu.ARM64.HasDIT
 
+const offsetARM64HasSB = unsafe.Offsetof(cpu.ARM64.HasSB)
+
 func EnableDIT() bool
 func DITEnabled() bool
 func DisableDIT()

src/internal/runtime/sys/dit_arm64.s

@@ -2,16 +2,21 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+#include "go_asm.h"
 #include "textflag.h"
 
 TEXT ·EnableDIT(SB),$0-1
     MRS DIT, R0
     UBFX $24, R0, $1, R1
     MOVB R1, ret+0(FP)
+    TBNZ $0, R1, ret
     MSR $1, DIT
-    // TODO(roland): the SB instruction is significantly more
-    // performant when available. We should detect its availability
-    // and use it when we can.
+    MOVBU internal∕cpu·ARM64+const_offsetARM64HasSB(SB), R2
+    TBZ $0, R2, sbFallback
+    SB
+ret:
+    RET
+sbFallback:
     DSB $7  // nsh
     ISB $15 // sy
     RET