runtime: use common library entry point on windows amd64/386

Windows can reuse the common library entry point instead of implementing
a its own version. Note that windows/arm64 already uses the common one.

Change-Id: I1a27bbec04bfd1d58a136638bafcdc0583bd106f
Reviewed-on: https://go-review.googlesource.com/c/go/+/706235
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>

src/runtime/asm_amd64.s

@@ -37,7 +37,12 @@ TEXT _rt0_amd64_lib(SB),NOSPLIT|NOFRAME,$0
 	MOVQ	SI, _rt0_amd64_lib_argv<>(SB)
 
 	// Synchronous initialization.
+#ifndef GOOS_windows
+	// Avoid calling it on Windows because it is not used
+	// and it would crash the application due to the autogenerated
+	// ABI wrapper trying to access a non-existent TLS slot.
 	CALL	runtime·libpreinit(SB)
+#endif
 
 	// Create a new thread to finish Go runtime initialization.
 	MOVQ	_cgo_sys_thread_create(SB), AX
@@ -45,12 +50,23 @@ TEXT _rt0_amd64_lib(SB),NOSPLIT|NOFRAME,$0
 	JZ	nocgo
 
 	// We're calling back to C.
-	// Align stack per ELF ABI requirements.
+	// Align stack per C ABI requirements.
 	MOVQ	SP, BX  // Callee-save in C ABI
 	ANDQ	$~15, SP
 	MOVQ	$_rt0_amd64_lib_go(SB), DI
 	MOVQ	$0, SI
+#ifdef GOOS_windows
+	// For Windows ABI
+	MOVQ	DI, CX
+	MOVQ	SI, DX
+	// Leave space for four words on the stack as required
+	// by the Windows amd64 calling convention.
+	ADJSP	$32
+#endif
 	CALL	AX
+#ifdef GOOS_windows
+	ADJSP	$-32 // just to make the assembler not complain about unbalanced stack
+#endif
 	MOVQ	BX, SP
 	JMP	restore
 

src/runtime/rt0_windows_386.s

@@ -12,33 +12,8 @@ TEXT _rt0_386_windows(SB),NOSPLIT,$0
 // library is loaded. For static libraries it is called when the
 // final executable starts, during the C runtime initialization
 // phase.
-TEXT _rt0_386_windows_lib(SB),NOSPLIT,$0x1C
+TEXT _rt0_386_windows_lib(SB),NOSPLIT,$0
-	MOVL	BP, 0x08(SP)
+	JMP	_rt0_386_lib(SB)
-	MOVL	BX, 0x0C(SP)
-	MOVL	AX, 0x10(SP)
-	MOVL  CX, 0x14(SP)
-	MOVL  DX, 0x18(SP)
-
-	// Create a new thread to do the runtime initialization and return.
-	MOVL	_cgo_sys_thread_create(SB), AX
-	MOVL	$_rt0_386_windows_lib_go(SB), 0x00(SP)
-	MOVL	$0, 0x04(SP)
-
-	 // Top two items on the stack are passed to _cgo_sys_thread_create
-	 // as parameters. This is the calling convention on 32-bit Windows.
-	CALL	AX
-
-	MOVL	0x08(SP), BP
-	MOVL	0x0C(SP), BX
-	MOVL	0x10(SP), AX
-	MOVL	0x14(SP), CX
-	MOVL	0x18(SP), DX
-	RET
-
-TEXT _rt0_386_windows_lib_go(SB),NOSPLIT,$0
-	PUSHL	$0
-	PUSHL	$0
-	JMP	runtime·rt0_go(SB)
 
 TEXT _main(SB),NOSPLIT,$0
 	// Remove the return address from the stack.

src/runtime/rt0_windows_amd64.s

@@ -6,31 +6,10 @@
 #include "go_tls.h"
 #include "textflag.h"
 
-TEXT _rt0_amd64_windows(SB),NOSPLIT|NOFRAME,$-8
+TEXT _rt0_amd64_windows(SB),NOSPLIT,$0
 	JMP	_rt0_amd64(SB)
 
 // When building with -buildmode=(c-shared or c-archive), this
-// symbol is called. For dynamic libraries it is called when the
+// symbol is called.
-// library is loaded. For static libraries it is called when the
+TEXT _rt0_amd64_windows_lib(SB),NOSPLIT,$0
-// final executable starts, during the C runtime initialization
+	JMP	_rt0_amd64_lib(SB)
-// phase.
-// Leave space for four pointers on the stack as required
-// by the Windows amd64 calling convention.
-TEXT _rt0_amd64_windows_lib(SB),NOSPLIT|NOFRAME,$40
-	// Create a new thread to do the runtime initialization and return.
-	MOVQ	BX, 32(SP) // callee-saved, preserved across the CALL
-	MOVQ	SP, BX
-	ANDQ	$~15, SP // alignment as per Windows requirement
-	MOVQ	_cgo_sys_thread_create(SB), AX
-	MOVQ	$_rt0_amd64_windows_lib_go(SB), CX
-	MOVQ	$0, DX
-	CALL	AX
-	MOVQ	BX, SP
-	MOVQ	32(SP), BX
-	RET
-
-TEXT _rt0_amd64_windows_lib_go(SB),NOSPLIT|NOFRAME,$0
-	MOVQ  $0, DI
-	MOVQ	$0, SI
-	MOVQ	$runtime·rt0_go(SB), AX
-	JMP	AX