MOVW	12(R13), R0
	MOVW	$1234, R1
	CMP	R0, R1
	BEQ	2(PC)
	BL	runtime·abort(SB)

	MOVW	0(R13), R8    // m
	MOVW	4(R13), R0    // g

	CMP	$0, R8
	BEQ	nog
	CMP	$0, R0
	BEQ	nog

	MOVW	R0, g
	MOVW	R8, g_m(g)

	// paranoia; check they are not nil
	MOVW	0(R8), R0
	MOVW	0(g), R0

	BL	runtime·emptyfunc(SB)	// fault if stack check is wrong

	{"R3", "R3"},
	{"R4", "R4"},
	{"R(4)", "R4"},
	{"R5", "R5"},
	{"R6", "R6"},
	{"R7", "R7"},
	{"R8", "R8"},
	{"[R0,R1,g,R15]", "[R0,R1,g,R15]"},
	{"[R0-R7]", "[R0,R1,R2,R3,R4,R5,R6,R7]"},
	{"[R(0)-R(7)]", "[R0,R1,R2,R3,R4,R5,R6,R7]"},
	{"[R0]", "[R0]"},
	{"[R1-R12]", "[R1,R2,R3,R4,R5,R6,R7,R8,R9,g,R11,R12]"},
	{"armCAS64(SB)", "armCAS64(SB)"},
	{"asmcgocall<>(SB)", "asmcgocall<>(SB)"},
	{"c+28(FP)", "c+28(FP)"},

	MOVV	$~3, R6
	AND	R4, R6
	// R7 = ((ptr & 3) * 8)
	AND	$3, R4, R7
	SLLV	$3, R7
	// Shift val for aligned ptr. R5 = val << R7 | ^(0xFF << R7)
	MOVV	$0xFF, R8
	SLLV	R7, R5
	SLLV	R7, R8
	NOR	R0, R8
	OR	R8, R5

	DBAR
	LL	(R6), R7
	AND	R5, R7
	SC	R7, (R6)
	BEQ	R7, -4(PC)
	DBAR
	RET

// func Or(addr *uint32, v uint32)
TEXT ·Or(SB), NOSPLIT, $0-12

	VPDPWSSD 15(R8)(R14*4), Y2, K3, Y14                // 62126d2b52b4b00f000000
	VPDPWSSD -7(CX)(DX*4), Y2, K3, Y14                 // 62726d2b52b491f9ffffff
	VPDPWSSD Y8, Y22, K3, Y14                          // 62524d2352f0
	VPDPWSSD Y9, Y22, K3, Y14                          // 62524d2352f1
	VPDPWSSD Y22, Y22, K3, Y14                         // 62324d2352f6

//

// func rawVforkSyscall(trap, a1, a2, a3 uintptr) (r1, err uintptr)
TEXT ·rawVforkSyscall(SB),NOSPLIT|NOFRAME,$0-48
	MOVD	a1+8(FP), R3
	MOVD	a2+16(FP), R4
	MOVD	a3+24(FP), R5
	MOVD	R0, R6
	MOVD	R0, R7
	MOVD	R0, R8
	MOVD	trap+0(FP), R9	// syscall entry
	SYSCALL R9
	BVC	ok
	MOVD	$-1, R4
	MOVD	R4, r1+32(FP)	// r1
	MOVD	R3, err+40(FP)	// errno
	RET
ok:
	MOVD	R3, r1+32(FP)	// r1
	MOVD	R0, err+40(FP)	// errno
	RET

TEXT ·Syscall6(SB),NOSPLIT,$20-40
	MOVW	num+0(FP), R2	// syscall entry
	MOVW	a1+4(FP), R4
	MOVW	a2+8(FP), R5
	MOVW	a3+12(FP), R6
	MOVW	a4+16(FP), R7
	MOVW	a5+20(FP), R8
	MOVW	a6+24(FP), R9
	MOVW	R8, 16(R29)
	MOVW	R9, 20(R29)
	MOVW	R0, R3	// reset R3 to 0 as 1-ret SYSCALL keeps it
	SYSCALL
	BEQ	R7, ok
	MOVW	$-1, R1
	MOVW	R1, r1+28(FP)
	MOVW	R0, r2+32(FP)
	MOVW	R2, errno+36(FP)
	RET
ok:

	print("r2    ", hex(u.r2), "\n")
	print("r3    ", hex(u.r3), "\n")
	print("r4    ", hex(u.r4), "\n")
	print("r5    ", hex(u.r5), "\n")
	print("r6    ", hex(u.r6), "\n")
	print("r7    ", hex(u.r7), "\n")
	print("r8    ", hex(u.r8), "\n")
	print("r9    ", hex(u.r9), "\n")
	print("r10   ", hex(u.r10), "\n")
	print("r11   ", hex(u.r11), "\n")
	print("r12   ", hex(u.r12), "\n")
	print("sp    ", hex(u.sp), "\n")
	print("link  ", hex(u.link), "\n")

TEXT ·Syscall6<ABIInternal>(SB),NOSPLIT,$0-80
	MOVD	R3, R10	// Move syscall number to R10. SYSCALL will move it R0, and restore R0.
	MOVD	R4, R3
	MOVD	R5, R4
	MOVD	R6, R5
	MOVD	R7, R6
	MOVD	R8, R7
	MOVD	R9, R8
	SYSCALL	R10
	MOVD	$-1, R6
	ISEL	CR0SO, R3, R0, R5 // errno = (error) ? R3 : 0
	ISEL	CR0SO, R6, R3, R3 // r1 = (error) ? -1 : 0
	MOVD	$0, R4            // r2 is not used on linux/ppc64

	print("cx    ", hex(u.cx), "\n")
	print("dx    ", hex(u.dx), "\n")
	print("di    ", hex(u.di), "\n")
	print("si    ", hex(u.si), "\n")
	print("bp    ", hex(u.bp), "\n")
	print("sp    ", hex(u.sp), "\n")
	print("r8    ", hex(u.r8), "\n")
	print("r9    ", hex(u.r9), "\n")
	print("r10   ", hex(u.r10), "\n")
	print("r11   ", hex(u.r11), "\n")
	print("r12   ", hex(u.r12), "\n")
	print("r13   ", hex(u.r13), "\n")
	print("r14   ", hex(u.r14), "\n")

	MOVW	m_gsignal(R8), R4
	CMP	g, R4
	BNE	3(PC)
	BL	runtime·badmorestackgsignal(SB)
	B	runtime·abort(SB)

	// Called from f.
	// Set m->morebuf to f's caller.
	MOVW	R3, (m_morebuf+gobuf_pc)(R8)	// f's caller's PC
	MOVW	R13, (m_morebuf+gobuf_sp)(R8)	// f's caller's SP
	MOVW	g, (m_morebuf+gobuf_g)(R8)

	// Call newstack on m->g0's stack.
	MOVW	m_g0(R8), R0
	BL	setg<>(SB)