MOVD	$sec+0(FP), R4 	// return parameter caller

	MOVD	R8, m_vdsoPC(R6)
	MOVD	R4, m_vdsoSP(R6)

	MOVD	m_curg(R6), R5
	CMP		g, R5
	BNE		noswitch

	MOVD	m_g0(R6), R4
	MOVD	(g_sched+gobuf_sp)(R4), R15	// Set SP to g0 stack

noswitch:
	SUB		$16, R15		// reserve 2x 8 bytes for parameters
	MOVD	$~7, R4			// align to 8 bytes because of gcc ABI
	AND		R4, R15

	MOV	T0, 32(X2)

	MOV	RA, m_vdsoPC(S3)
	MOV	$ret-8(FP), T1 // caller's SP
	MOV	T1, m_vdsoSP(S3)

	MOV	m_curg(S3), T1
	BNE	g, T1, noswitch

	MOV	m_g0(S3), T1
	MOV	(g_sched+gobuf_sp)(T1), X2

noswitch:
	SUB	$24, X2 // Space for result
	ANDI	$~7, X2 // Align for C code
	MOV	$8(X2), A1

	// Store g on gsignal's stack, see sys_linux_arm64.s for detail
	MOVBU	runtime·iscgo(SB), S4

	MOVV    $ret-8(FP), R11 // caller's SP
	MOVV	R1, m_vdsoPC(R24)
	MOVV	R11, m_vdsoSP(R24)

	MOVV	m_curg(R24), R4
	MOVV	g, R5
	BNE	R4, R5, noswitch

	MOVV	m_g0(R24), R4
	MOVV	(g_sched+gobuf_sp)(R4), R25	// Set SP to g0 stack

noswitch:
	SUBV	$16, R25
	AND	$~15, R25	// Align for C code
	MOVV	R25, R3

	MOVW	$CLOCK_REALTIME, R4
	MOVV	$0(R3), R5

	MOVV	runtime·vdsoClockgettimeSym(SB), R20

	MOVQ	m_g0(R13), R14
	CMPQ	g(R15), R14
	JNE	switch
	// executing on m->g0 already
	CALL	AX
	RET

switch:
	// Switch to m->g0 stack and back.
	MOVQ	(g_sched+gobuf_sp)(R14), R14
	MOVQ	SP, -8(R14)
	LEAQ	-8(R14), SP
	CALL	AX
	MOVQ	0(SP), SP
	RET

noswitch:
	// Not a Go-managed thread. Do not switch stack.
	CALL	AX
	RET

	MOVD	LR, R14
	MOVD	$ret-FIXED_FRAME(FP), R5 // caller's SP
	MOVD	R14, m_vdsoPC(R21)
	MOVD	R5, m_vdsoSP(R21)

	MOVD	m_curg(R21), R6
	CMP	g, R6
	BNE	noswitch

	MOVD	m_g0(R21), R7
	MOVD	(g_sched+gobuf_sp)(R7), R1	// Set SP to g0 stack

noswitch:
	SUB	$16, R1                 // Space for results
	RLDICR	$0, R1, $59, R1         // Align for C code
	MOVD	R12, CTR
	MOVD	R1, R4

	LEAL	sec+0(FP), DX
	MOVL	-4(DX), CX
	MOVL	CX, m_vdsoPC(SI)
	MOVL	DX, m_vdsoSP(SI)

	CMPL	AX, m_curg(SI)	// Only switch if on curg.
	JNE	noswitch

	MOVL	m_g0(SI), DX
	MOVL	(g_sched+gobuf_sp)(DX), SP	// Set SP to g0 stack

noswitch:
	SUBL	$16, SP		// Space for results
	ANDL	$~15, SP	// Align for C code

	// Stack layout, depending on call path:
	//  x(SP)   vDSO            INVOKE_SYSCALL

	MOVD	R3, 16(RSP)

	MOVD	$ret-8(FP), R2 // caller's SP
	MOVD	LR, m_vdsoPC(R21)
	MOVD	R2, m_vdsoSP(R21)

	MOVD	m_curg(R21), R0
	CMP	g, R0
	BNE	noswitch

	MOVD	m_g0(R21), R3
	MOVD	(g_sched+gobuf_sp)(R3), R1	// Set RSP to g0 stack

noswitch:
	SUB	$16, R1
	BIC	$15, R1	// Align for C code
	MOVD	R1, RSP

	MOVW	$CLOCK_REALTIME, R0
	MOVD	runtime·vdsoClockgettimeSym(SB), R2
	CBZ	R2, fallback

	BL	runtime·save_g(SB)		// Save g for callbacks.
	MOVD	R15, R7				// Save SP.
	MOVD	g_m(g), R8			// R8 = thread.
	MOVD	m_g0(R8), R8			// R8 = g0.
	CMPBEQ	R8, g, call			// Already on g0?
	MOVD	(g_sched+gobuf_sp)(R8), R15	// Switch SP to g0.
call:	SUB	$160, R15			// Allocate C frame.
	BL	R1				// Call C code.
	MOVD	R7, R15				// Restore SP.
	RET					// Return to Go.

	MOVQ	g_m(R14), R13
	// Switch to g0 stack.
	MOVQ	SP, R12		// callee-saved, preserved across the CALL
	MOVQ	m_g0(R13), R10
	CMPQ	R10, R14
	JE	call	// already on g0
	MOVQ	(g_sched+gobuf_sp)(R10), SP
call:
	ANDQ	$~15, SP	// alignment for gcc ABI
	CALL	AX
	MOVQ	R12, SP
	// Back to Go world, set special registers.
	// The g register (R14) is preserved in C.
	XORPS	X15, X15
	RET

	// Switch to g0 stack.
	MOVD	RSP, R19	// callee-saved, preserved across the CALL
	MOVD	R30, R20	// callee-saved, preserved across the CALL
	MOVD	m_g0(R10), R11
	CMP	R11, g
	BEQ	call	// already on g0
	MOVD	(g_sched+gobuf_sp)(R11), R12
	MOVD	R12, RSP
call:
	// Decrement SP past where the frame pointer is saved in the Go arm64
	// ABI (one word below the stack pointer) so the race detector library
	// code doesn't clobber it