windows/svc: align stack in servicemain before calling syscall

I noticed that we call syscall in servicemain without aligning stack.
That is against Windows rules, so align the stack as required.

I tried running this code with specifically non-aligned stack (I aligned
stack, and then subtracted 1 from SP) on my Windows 10 to test this
change. But it makes no difference on my Windows 10 PC - I built and run
golang.org/x/sys/windows/svc/example, and it runs successfully
regardless of stack alignment. But alignment might make difference on
other computers.

Maybe fixes golang/go#40160

Change-Id: I351f7f730fba4aa6dc409a79de4ad737b4a0a7d4
Reviewed-on: https://go-review.googlesource.com/c/sys/+/246317
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

windows/svc/sys_amd64.s

@@ -6,11 +6,15 @@
 
 // func servicemain(argc uint32, argv **uint16)
 TEXT ·servicemain(SB),7,$0
+	MOVQ	SP, AX
+	ANDQ	$~15, SP	// alignment as per Windows requirement
+	SUBQ	$48, SP		// room for SP and 4 args as per Windows requirement
+				// plus one extra word to keep stack 16 bytes aligned
+	MOVQ	AX, 32(SP)
+
 	MOVL	CX, ·sArgc(SB)
 	MOVQ	DX, ·sArgv(SB)
 
-	SUBQ	$32, SP		// stack for the first 4 syscall params
-
 	MOVQ	·sName(SB), CX
 	MOVQ	$·servicectlhandler(SB), DX
 	// BUG(pastarmovj): Figure out a way to pass in context in R8.
@@ -32,7 +36,7 @@ TEXT ·servicemain(SB),7,$0
 	CALL	AX
 
 exit:
-	ADDQ	$32, SP
+	MOVQ	32(SP), SP
 	RET
 
 // I do not know why, but this seems to be the only way to call