runtime: fix darwin/386, darwin/arm builds

In cl/8652 I broke darwin/arm and darwin/386 because I removed the *g
parameter, which they both expect and use. This CL adjusts both ports
to look for g0 in m, just as darwin/amd64 does.

Tested on darwin{386,arm,amd64}.

Change-Id: Ia56f3d97e126b40d8bbd2e8f677b008e4a1badad
Reviewed-on: https://go-review.googlesource.com/8666
Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/runtime/sys_darwin_386.s

@@ -327,33 +327,32 @@ TEXT runtime·usleep(SB),NOSPLIT,$32
 	INT	$0x80
 	RET
 
-// func bsdthread_create(stk unsafe.Pointer, mm *m, gg *g, fn uintptr) int32
+// func bsdthread_create(stk, arg unsafe.Pointer, fn uintptr) int32
 // System call args are: func arg stack pthread flags.
 TEXT runtime·bsdthread_create(SB),NOSPLIT,$32
 	MOVL	$360, AX
 	// 0(SP) is where the caller PC would be; kernel skips it
-	MOVL	fn+12(FP), BX
+	MOVL	fn+8(FP), BX
 	MOVL	BX, 4(SP)	// func
-	MOVL	mm+4(FP), BX
+	MOVL	arg+4(FP), BX
 	MOVL	BX, 8(SP)	// arg
 	MOVL	stk+0(FP), BX
 	MOVL	BX, 12(SP)	// stack
-	MOVL	gg+8(FP), BX
-	MOVL	BX, 16(SP)	// pthread
+	MOVL    $0, 16(SP)      // pthread
 	MOVL	$0x1000000, 20(SP)	// flags = PTHREAD_START_CUSTOM
 	INT	$0x80
 	JAE	4(PC)
 	NEGL	AX
-	MOVL	AX, ret+16(FP)
+	MOVL	AX, ret+12(FP)
 	RET
 	MOVL	$0, AX
-	MOVL	AX, ret+16(FP)
+	MOVL	AX, ret+12(FP)
 	RET
 
 // The thread that bsdthread_create creates starts executing here,
 // because we registered this function using bsdthread_register
 // at startup.
-//	AX = "pthread" (= g)
+//	AX = "pthread" (= 0x0)
 //	BX = mach thread port
 //	CX = "func" (= fn)
 //	DX = "arg" (= m)
@@ -380,6 +379,7 @@ TEXT runtime·bsdthread_start(SB),NOSPLIT,$0
 
 	// Now segment is established.  Initialize m, g.
 	get_tls(BP)
+	MOVL    m_g0(DX), AX
 	MOVL	AX, g(BP)
 	MOVL	DX, g_m(AX)
 	MOVL	BX, m_procid(DX)	// m->procid = thread port (for debuggers)

src/runtime/sys_darwin_amd64.s

@@ -308,12 +308,12 @@ TEXT runtime·bsdthread_create(SB),NOSPLIT,$0
 	// Set up arguments to bsdthread_create system call.
 	// The ones in quotes pass through to the thread callback
 	// uninterpreted, so we can put whatever we want there.
-	MOVQ	fn+24(SP), DI	// "func"
-	MOVQ	arg+16(SP), SI	// "arg"
-	MOVQ	stk+8(SP), DX	// stack
-	MOVQ	$0, R10		// "pthread", paranoia
-	MOVQ	$0x01000000, R8	// flags = PTHREAD_START_CUSTOM
-	MOVQ	$0, R9	// paranoia
+	MOVQ	fn+24(SP),   DI
+	MOVQ	arg+16(SP),  SI
+	MOVQ	stk+8(SP),   DX
+	MOVQ	$0x01000000, R8  // flags = PTHREAD_START_CUSTOM
+	MOVQ	$0,          R9  // paranoia
+	MOVQ	$0,          R10 // paranoia, "pthread"
 	MOVQ	$(0x2000000+360), AX	// bsdthread_create
 	SYSCALL
 	JCC 4(PC)

src/runtime/sys_darwin_arm.s

@@ -320,26 +320,25 @@ sysctl_ret:
 	RET
 
 // Thread related functions
-// void bsdthread_create(void *stk, M *m, G *g, void (*fn)(void))
+// func bsdthread_create(stk, arg unsafe.Pointer, fn uintptr) int32
 TEXT runtime·bsdthread_create(SB),NOSPLIT,$0
 	// Set up arguments to bsdthread_create system call.
 	// The ones in quotes pass through to the thread callback
 	// uninterpreted, so we can put whatever we want there.
-	MOVW    fn+12(FP), R0   // "func"
-	MOVW    mm+4(FP), R1 // "arg"
-	MOVW    stk+0(FP), R2 // stack
-	MOVW    gg+8(FP), R3 // "pthread"
+	MOVW    fn+8(FP),    R0 // "func"
+	MOVW    arg+4(FP),   R1 // "arg"
+	MOVW    stk+0(FP),   R2 // stack
 	MOVW	$0x01000000, R4	// flags = PTHREAD_START_CUSTOM
-	MOVW	$0, R5	// paranoia
+	MOVW	$0,          R5 // paranoia
 	MOVW	$SYS_bsdthread_create, R12
 	SWI	$0x80
 	BCC		create_ret
 	RSB 	$0, R0, R0
-	MOVW	R0, ret+16(FP)
+	MOVW	R0, ret+12(FP)
 	RET
 create_ret:
 	MOVW	$0, R0
-	MOVW	R0, ret+16(FP)
+	MOVW	R0, ret+12(FP)
 	RET
 
 // The thread that bsdthread_create creates starts executing here,