internal/runtime: fix assembly for spectre retpoline instrumentation

In the last year we added two CALLs whose targets are loaded from
memory. Change them to call from a register so that the instrumentation
for spectre mitigations works.

This change also adds a smoke test for the spectre build flags.

For #77420.

Change-Id: I35ec723449ff6a712bcce3276bf1df3fa932bddc
Cq-Include-Trybots: luci.golang.try:gotip-linux-amd64-longtest
Reviewed-on: https://go-review.googlesource.com/c/go/+/741541
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>

src/cmd/dist/test.go

@@ -789,7 +789,7 @@ func (t *tester) registerTests() {
 	if !t.compileOnly && !t.short {
 		t.registerTest("GODEBUG=gcstoptheworld=2 archive/zip",
 			&goTest{
-				variant: "runtime:gcstoptheworld2",
+				variant: "gcstoptheworld2",
 				timeout: 300 * time.Second,
 				short:   true,
 				env:     []string{"GODEBUG=gcstoptheworld=2"},
@@ -797,7 +797,7 @@ func (t *tester) registerTests() {
 			})
 		t.registerTest("GODEBUG=gccheckmark=1 runtime",
 			&goTest{
-				variant: "runtime:gccheckmark",
+				variant: "gccheckmark",
 				timeout: 300 * time.Second,
 				short:   true,
 				env:     []string{"GODEBUG=gccheckmark=1"},
@@ -805,6 +805,22 @@ func (t *tester) registerTests() {
 			})
 	}
 
+	// Spectre mitigation smoke test.
+	if goos == "linux" && goarch == "amd64" {
+		// Pick a bunch of packages known to have some assembly.
+		pkgs := []string{"internal/runtime/...", "reflect", "crypto/..."}
+		if !t.short {
+			pkgs = append(pkgs, "runtime")
+		}
+		t.registerTest("spectre",
+			&goTest{
+				variant: "spectre",
+				short:   true,
+				env:     []string{"GOFLAGS=-gcflags=all=-spectre=all -asmflags=all=-spectre=all"},
+				pkgs:    pkgs,
+			})
+	}
+
 	// morestack tests. We only run these in long-test mode
 	// (with GO_TEST_SHORT=0) because the runtime test is
 	// already quite long and mayMoreStackMove makes it about

src/internal/runtime/gc/scan/scan_amd64.s

@@ -12,7 +12,8 @@ TEXT ·ExpandAVX512(SB), NOSPLIT, $0-24
 
 	// Call the expander for this size class
 	LEAQ ·gcExpandersAVX512(SB), BX
-	CALL (BX)(CX*8)
+	MOVQ (BX)(CX*8), DX // Move to register first so -spectre works
+	CALL DX
 
 	MOVQ unpacked+16(FP), DI // Expanded output bitmap pointer
 	VMOVDQU64 Z1, 0(DI)
@@ -25,7 +26,8 @@ TEXT ·scanSpanPackedAVX512(SB), NOSPLIT, $256-44
 	MOVQ objMarks+16(FP), AX
 	MOVQ sizeClass+24(FP), CX
 	LEAQ ·gcExpandersAVX512(SB), BX
-	CALL (BX)(CX*8)
+	MOVQ (BX)(CX*8), DX // Move to register first so -spectre works
+	CALL DX
 
 	// Z3+Z4 = Load the pointer mask
 	MOVQ ptrMask+32(FP), AX

src/internal/runtime/startlinetest/func_amd64.s

@@ -24,5 +24,6 @@ TEXT	·AsmFunc<ABIInternal>(SB),NOSPLIT,$8-0
 	NO_LOCAL_POINTERS
 	MOVQ	$0, AX // wantInlined
 	MOVQ	·CallerStartLine(SB), DX
-	CALL	(DX)
+	MOVQ	(DX), DX // Move to a register first for -spectre
+	CALL	DX
 	RET