From eee6f9f825596a769ff163e098d2656cbed09d4f Mon Sep 17 00:00:00 2001
From: eric fang <eric.fang@arm.com>
Date: Thu, 23 Dec 2021 06:53:50 +0000
Subject: [PATCH] runtime: unify C->Go ABI transitions on arm64

There are several of places that save and restore the C callee-saved registers,
the operation is the same everywhere, so this CL defines several macros
to do this, which will help reduce code redundancy and unify the operation.

This CL also replaced consecutive MOVD instructions with STP and LDP instructions
in several places where these macros do not apply.

Change-Id: I815f39fe484a9ab9b6bd157dfcbc8ad99c1420fe
Reviewed-on: https://go-review.googlesource.com/c/go/+/374397
Trust: Eric Fang <eric.fang@arm.com>
Run-TryBot: Eric Fang <eric.fang@arm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
---
 src/runtime/asm_arm64.s         | 178 +++++++++++---------------------
 src/runtime/cgo/abi_arm64.h     |  43 ++++++++
 src/runtime/cgo/asm_arm64.s     |  55 ++--------
 src/runtime/race_arm64.s        |  22 ++--
 src/runtime/rt0_darwin_arm64.s  |  43 ++------
 src/runtime/rt0_freebsd_arm64.s |  41 +-------
 src/runtime/rt0_linux_arm64.s   |  41 +-------
 src/runtime/rt0_netbsd_arm64.s  |  41 +-------
 src/runtime/rt0_openbsd_arm64.s |  41 +-------
 src/runtime/sys_darwin_arm64.s  |  87 ++--------------
 src/runtime/sys_freebsd_arm64.s |  45 ++------
 src/runtime/sys_linux_arm64.s   |  91 ++--------------
 src/runtime/sys_netbsd_arm64.s  |  45 ++------
 src/runtime/sys_openbsd_arm64.s |  87 ++--------------
 src/runtime/sys_windows_arm64.s |  53 +++-------
 15 files changed, 205 insertions(+), 708 deletions(-)
 create mode 100644 src/runtime/cgo/abi_arm64.h

diff --git a/src/runtime/asm_arm64.s b/src/runtime/asm_arm64.s
index 62deb070aa2..9ef7346e002 100644
--- a/src/runtime/asm_arm64.s
+++ b/src/runtime/asm_arm64.s
@@ -311,74 +311,42 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 
 // spillArgs stores return values from registers to a *internal/abi.RegArgs in R20.
 TEXT ·spillArgs(SB),NOSPLIT,$0-0
-	MOVD	R0, (0*8)(R20)
-	MOVD	R1, (1*8)(R20)
-	MOVD	R2, (2*8)(R20)
-	MOVD	R3, (3*8)(R20)
-	MOVD	R4, (4*8)(R20)
-	MOVD	R5, (5*8)(R20)
-	MOVD	R6, (6*8)(R20)
-	MOVD	R7, (7*8)(R20)
-	MOVD	R8, (8*8)(R20)
-	MOVD	R9, (9*8)(R20)
-	MOVD	R10, (10*8)(R20)
-	MOVD	R11, (11*8)(R20)
-	MOVD	R12, (12*8)(R20)
-	MOVD	R13, (13*8)(R20)
-	MOVD	R14, (14*8)(R20)
-	MOVD	R15, (15*8)(R20)
-	FMOVD	F0, (16*8)(R20)
-	FMOVD	F1, (17*8)(R20)
-	FMOVD	F2, (18*8)(R20)
-	FMOVD	F3, (19*8)(R20)
-	FMOVD	F4, (20*8)(R20)
-	FMOVD	F5, (21*8)(R20)
-	FMOVD	F6, (22*8)(R20)
-	FMOVD	F7, (23*8)(R20)
-	FMOVD	F8, (24*8)(R20)
-	FMOVD	F9, (25*8)(R20)
-	FMOVD	F10, (26*8)(R20)
-	FMOVD	F11, (27*8)(R20)
-	FMOVD	F12, (28*8)(R20)
-	FMOVD	F13, (29*8)(R20)
-	FMOVD	F14, (30*8)(R20)
-	FMOVD	F15, (31*8)(R20)
+	STP	(R0, R1), (0*8)(R20)
+	STP	(R2, R3), (2*8)(R20)
+	STP	(R4, R5), (4*8)(R20)
+	STP	(R6, R7), (6*8)(R20)
+	STP	(R8, R9), (8*8)(R20)
+	STP	(R10, R11), (10*8)(R20)
+	STP	(R12, R13), (12*8)(R20)
+	STP	(R14, R15), (14*8)(R20)
+	FSTPD	(F0, F1), (16*8)(R20)
+	FSTPD	(F2, F3), (18*8)(R20)
+	FSTPD	(F4, F5), (20*8)(R20)
+	FSTPD	(F6, F7), (22*8)(R20)
+	FSTPD	(F8, F9), (24*8)(R20)
+	FSTPD	(F10, F11), (26*8)(R20)
+	FSTPD	(F12, F13), (28*8)(R20)
+	FSTPD	(F14, F15), (30*8)(R20)
 	RET
 
 // unspillArgs loads args into registers from a *internal/abi.RegArgs in R20.
 TEXT ·unspillArgs(SB),NOSPLIT,$0-0
-	MOVD	(0*8)(R20), R0
-	MOVD	(1*8)(R20), R1
-	MOVD	(2*8)(R20), R2
-	MOVD	(3*8)(R20), R3
-	MOVD	(4*8)(R20), R4
-	MOVD	(5*8)(R20), R5
-	MOVD	(6*8)(R20), R6
-	MOVD	(7*8)(R20), R7
-	MOVD	(8*8)(R20), R8
-	MOVD	(9*8)(R20), R9
-	MOVD	(10*8)(R20), R10
-	MOVD	(11*8)(R20), R11
-	MOVD	(12*8)(R20), R12
-	MOVD	(13*8)(R20), R13
-	MOVD	(14*8)(R20), R14
-	MOVD	(15*8)(R20), R15
-	FMOVD	(16*8)(R20), F0
-	FMOVD	(17*8)(R20), F1
-	FMOVD	(18*8)(R20), F2
-	FMOVD	(19*8)(R20), F3
-	FMOVD	(20*8)(R20), F4
-	FMOVD	(21*8)(R20), F5
-	FMOVD	(22*8)(R20), F6
-	FMOVD	(23*8)(R20), F7
-	FMOVD	(24*8)(R20), F8
-	FMOVD	(25*8)(R20), F9
-	FMOVD	(26*8)(R20), F10
-	FMOVD	(27*8)(R20), F11
-	FMOVD	(28*8)(R20), F12
-	FMOVD	(29*8)(R20), F13
-	FMOVD	(30*8)(R20), F14
-	FMOVD	(31*8)(R20), F15
+	LDP	(0*8)(R20), (R0, R1)
+	LDP	(2*8)(R20), (R2, R3)
+	LDP	(4*8)(R20), (R4, R5)
+	LDP	(6*8)(R20), (R6, R7)
+	LDP	(8*8)(R20), (R8, R9)
+	LDP	(10*8)(R20), (R10, R11)
+	LDP	(12*8)(R20), (R12, R13)
+	LDP	(14*8)(R20), (R14, R15)
+	FLDPD	(16*8)(R20), (F0, F1)
+	FLDPD	(18*8)(R20), (F2, F3)
+	FLDPD	(20*8)(R20), (F4, F5)
+	FLDPD	(22*8)(R20), (F6, F7)
+	FLDPD	(24*8)(R20), (F8, F9)
+	FLDPD	(26*8)(R20), (F10, F11)
+	FLDPD	(28*8)(R20), (F12, F13)
+	FLDPD	(30*8)(R20), (F14, F15)
 	RET
 
 // reflectcall: call a function with the given argument list
@@ -480,10 +448,8 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 // arguments in registers.
 TEXT callRet<>(SB), NOSPLIT, $48-0
 	NO_LOCAL_POINTERS
-	MOVD	R7, 8(RSP)
-	MOVD	R3, 16(RSP)
-	MOVD	R5, 24(RSP)
-	MOVD	R4, 32(RSP)
+	STP	(R7, R3), 8(RSP)
+	STP	(R5, R4), 24(RSP)
 	MOVD	R20, 40(RSP)
 	BL	runtime·reflectcallmove(SB)
 	RET
@@ -1215,8 +1181,7 @@ TEXT ·checkASM(SB),NOSPLIT,$0-1
 // calls to it directly and it does not use the stack-based Go ABI.
 TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$200
 	// Save the registers clobbered by the fast path.
-	MOVD	R0, 184(RSP)
-	MOVD	R1, 192(RSP)
+	STP	(R0, R1), 184(RSP)
 	MOVD	g_m(g), R0
 	MOVD	m_p(R0), R0
 	MOVD	(p_wbBuf+wbBuf_next)(R0), R1
@@ -1232,8 +1197,7 @@ TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$200
 	// Is the buffer full? (flags set in CMP above)
 	BEQ	flush
 ret:
-	MOVD	184(RSP), R0
-	MOVD	192(RSP), R1
+	LDP	184(RSP), (R0, R1)
 	// Do the write.
 	MOVD	R3, (R2)
 	RET
@@ -1241,32 +1205,20 @@ ret:
 flush:
 	// Save all general purpose registers since these could be
 	// clobbered by wbBufFlush and were not saved by the caller.
-	MOVD	R2, 8(RSP)	// Also first argument to wbBufFlush
-	MOVD	R3, 16(RSP)	// Also second argument to wbBufFlush
-	// R0 already saved
-	// R1 already saved
-	MOVD	R4, 24(RSP)
-	MOVD	R5, 32(RSP)
-	MOVD	R6, 40(RSP)
-	MOVD	R7, 48(RSP)
-	MOVD	R8, 56(RSP)
-	MOVD	R9, 64(RSP)
-	MOVD	R10, 72(RSP)
-	MOVD	R11, 80(RSP)
-	MOVD	R12, 88(RSP)
-	MOVD	R13, 96(RSP)
-	MOVD	R14, 104(RSP)
-	MOVD	R15, 112(RSP)
+	// R0 and R1 already saved
+	STP	(R2, R3), 1*8(RSP)	// Also first and second arguments to wbBufFlush
+	STP	(R4, R5), 3*8(RSP)
+	STP	(R6, R7), 5*8(RSP)
+	STP	(R8, R9), 7*8(RSP)
+	STP	(R10, R11), 9*8(RSP)
+	STP	(R12, R13), 11*8(RSP)
+	STP	(R14, R15), 13*8(RSP)
 	// R16, R17 may be clobbered by linker trampoline
 	// R18 is unused.
-	MOVD	R19, 120(RSP)
-	MOVD	R20, 128(RSP)
-	MOVD	R21, 136(RSP)
-	MOVD	R22, 144(RSP)
-	MOVD	R23, 152(RSP)
-	MOVD	R24, 160(RSP)
-	MOVD	R25, 168(RSP)
-	MOVD	R26, 176(RSP)
+	STP	(R19, R20), 15*8(RSP)
+	STP	(R21, R22), 17*8(RSP)
+	STP	(R23, R24), 19*8(RSP)
+	STP	(R25, R26), 21*8(RSP)
 	// R27 is temp register.
 	// R28 is g.
 	// R29 is frame pointer (unused).
@@ -1275,29 +1227,17 @@ flush:
 
 	// This takes arguments R2 and R3.
 	CALL	runtime·wbBufFlush(SB)
-
-	MOVD	8(RSP), R2
-	MOVD	16(RSP), R3
-	MOVD	24(RSP), R4
-	MOVD	32(RSP), R5
-	MOVD	40(RSP), R6
-	MOVD	48(RSP), R7
-	MOVD	56(RSP), R8
-	MOVD	64(RSP), R9
-	MOVD	72(RSP), R10
-	MOVD	80(RSP), R11
-	MOVD	88(RSP), R12
-	MOVD	96(RSP), R13
-	MOVD	104(RSP), R14
-	MOVD	112(RSP), R15
-	MOVD	120(RSP), R19
-	MOVD	128(RSP), R20
-	MOVD	136(RSP), R21
-	MOVD	144(RSP), R22
-	MOVD	152(RSP), R23
-	MOVD	160(RSP), R24
-	MOVD	168(RSP), R25
-	MOVD	176(RSP), R26
+	LDP	1*8(RSP), (R2, R3)
+	LDP	3*8(RSP), (R4, R5)
+	LDP	5*8(RSP), (R6, R7)
+	LDP	7*8(RSP), (R8, R9)
+	LDP	9*8(RSP), (R10, R11)
+	LDP	11*8(RSP), (R12, R13)
+	LDP	13*8(RSP), (R14, R15)
+	LDP	15*8(RSP), (R19, R20)
+	LDP	17*8(RSP), (R21, R22)
+	LDP	19*8(RSP), (R23, R24)
+	LDP	21*8(RSP), (R25, R26)
 	JMP	ret
 
 // Note: these functions use a special calling convention to save generated code space.
diff --git a/src/runtime/cgo/abi_arm64.h b/src/runtime/cgo/abi_arm64.h
new file mode 100644
index 00000000000..e2b5e6d0be9
--- /dev/null
+++ b/src/runtime/cgo/abi_arm64.h
@@ -0,0 +1,43 @@
+// Copyright 2021 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Macros for transitioning from the host ABI to Go ABI0.
+//
+// These macros save and restore the callee-saved registers
+// from the stack, but they don't adjust stack pointer, so
+// the user should prepare stack space in advance.
+// SAVE_R19_TO_R28(offset) saves R19 ~ R28 to the stack space
+// of ((offset)+0*8)(RSP) ~ ((offset)+9*8)(RSP).
+//
+// SAVE_F8_TO_F15(offset) saves F8 ~ F15 to the stack space
+// of ((offset)+0*8)(RSP) ~ ((offset)+7*8)(RSP).
+//
+// R29 is not saved because Go will save and restore it.
+
+#define SAVE_R19_TO_R28(offset) \
+	STP	(R19, R20), ((offset)+0*8)(RSP) \
+	STP	(R21, R22), ((offset)+2*8)(RSP) \
+	STP	(R23, R24), ((offset)+4*8)(RSP) \
+	STP	(R25, R26), ((offset)+6*8)(RSP) \
+	STP	(R27, g), ((offset)+8*8)(RSP)
+
+#define RESTORE_R19_TO_R28(offset) \
+	LDP	((offset)+0*8)(RSP), (R19, R20) \
+	LDP	((offset)+2*8)(RSP), (R21, R22) \
+	LDP	((offset)+4*8)(RSP), (R23, R24) \
+	LDP	((offset)+6*8)(RSP), (R25, R26) \
+	LDP	((offset)+8*8)(RSP), (R27, g) /* R28 */
+
+#define SAVE_F8_TO_F15(offset) \
+	FSTPD	(F8, F9), ((offset)+0*8)(RSP) \
+	FSTPD	(F10, F11), ((offset)+2*8)(RSP) \
+	FSTPD	(F12, F13), ((offset)+4*8)(RSP) \
+	FSTPD	(F14, F15), ((offset)+6*8)(RSP)
+
+#define RESTORE_F8_TO_F15(offset) \
+	FLDPD	((offset)+0*8)(RSP), (F8, F9) \
+	FLDPD	((offset)+2*8)(RSP), (F10, F11) \
+	FLDPD	((offset)+4*8)(RSP), (F12, F13) \
+	FLDPD	((offset)+6*8)(RSP), (F14, F15)
+
diff --git a/src/runtime/cgo/asm_arm64.s b/src/runtime/cgo/asm_arm64.s
index 1cb25cf89ef..e808dedcfc5 100644
--- a/src/runtime/cgo/asm_arm64.s
+++ b/src/runtime/cgo/asm_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "abi_arm64.h"
 
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
@@ -14,57 +15,23 @@ TEXT crosscall2(SB),NOSPLIT|NOFRAME,$0
 	 *  push 3 args for fn (R0, R1, R3), skipping R2.
 	 * Also note that at procedure entry in gc world, 8(RSP) will be the
 	 *  first arg.
-	 * TODO(minux): use LDP/STP here if it matters.
 	 */
 	SUB	$(8*24), RSP
-	MOVD	R0, (8*1)(RSP)
-	MOVD	R1, (8*2)(RSP)
+	STP	(R0, R1), (8*1)(RSP)
 	MOVD	R3, (8*3)(RSP)
-	MOVD	R19, (8*4)(RSP)
-	MOVD	R20, (8*5)(RSP)
-	MOVD	R21, (8*6)(RSP)
-	MOVD	R22, (8*7)(RSP)
-	MOVD	R23, (8*8)(RSP)
-	MOVD	R24, (8*9)(RSP)
-	MOVD	R25, (8*10)(RSP)
-	MOVD	R26, (8*11)(RSP)
-	MOVD	R27, (8*12)(RSP)
-	MOVD	g, (8*13)(RSP)
-	MOVD	R29, (8*14)(RSP)
-	MOVD	R30, (8*15)(RSP)
-	FMOVD	F8, (8*16)(RSP)
-	FMOVD	F9, (8*17)(RSP)
-	FMOVD	F10, (8*18)(RSP)
-	FMOVD	F11, (8*19)(RSP)
-	FMOVD	F12, (8*20)(RSP)
-	FMOVD	F13, (8*21)(RSP)
-	FMOVD	F14, (8*22)(RSP)
-	FMOVD	F15, (8*23)(RSP)
+
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
+	STP	(R29, R30), (8*22)(RSP)
+
 
 	// Initialize Go ABI environment
 	BL	runtime·load_g(SB)
-
 	BL	runtime·cgocallback(SB)
 
-	MOVD	(8*4)(RSP), R19
-	MOVD	(8*5)(RSP), R20
-	MOVD	(8*6)(RSP), R21
-	MOVD	(8*7)(RSP), R22
-	MOVD	(8*8)(RSP), R23
-	MOVD	(8*9)(RSP), R24
-	MOVD	(8*10)(RSP), R25
-	MOVD	(8*11)(RSP), R26
-	MOVD	(8*12)(RSP), R27
-	MOVD	(8*13)(RSP), g
-	MOVD	(8*14)(RSP), R29
-	MOVD	(8*15)(RSP), R30
-	FMOVD	(8*16)(RSP), F8
-	FMOVD	(8*17)(RSP), F9
-	FMOVD	(8*18)(RSP), F10
-	FMOVD	(8*19)(RSP), F11
-	FMOVD	(8*20)(RSP), F12
-	FMOVD	(8*21)(RSP), F13
-	FMOVD	(8*22)(RSP), F14
-	FMOVD	(8*23)(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
+	LDP	(8*22)(RSP), (R29, R30)
+
 	ADD	$(8*24), RSP
 	RET
diff --git a/src/runtime/race_arm64.s b/src/runtime/race_arm64.s
index 8c0dd25f0ba..edbb3b12c73 100644
--- a/src/runtime/race_arm64.s
+++ b/src/runtime/race_arm64.s
@@ -8,6 +8,7 @@
 #include "funcdata.h"
 #include "textflag.h"
 #include "tls_arm64.h"
+#include "cgo/abi_arm64.h"
 
 // The following thunks allow calling the gcc-compiled race runtime directly
 // from Go code without going all the way through cgo.
@@ -450,13 +451,12 @@ TEXT	runtime·racecallbackthunk(SB), NOSPLIT|NOFRAME, $0
 rest:
 	// Save callee-saved registers (Go code won't respect that).
 	// 8(RSP) and 16(RSP) are for args passed through racecallback
-	SUB	$112, RSP
+	SUB	$176, RSP
 	MOVD	LR, 0(RSP)
-	STP	(R19, R20), 24(RSP)
-	STP	(R21, R22), 40(RSP)
-	STP	(R23, R24), 56(RSP)
-	STP	(R25, R26), 72(RSP)
-	STP	(R27,   g), 88(RSP)
+
+	SAVE_R19_TO_R28(8*3)
+	SAVE_F8_TO_F15(8*13)
+	MOVD	R29, (8*21)(RSP)
 	// Set g = g0.
 	// load_g will clobber R0, Save R0
 	MOVD	R0, R13
@@ -479,12 +479,10 @@ rest:
 ret:
 	// Restore callee-saved registers.
 	MOVD	0(RSP), LR
-	LDP	24(RSP), (R19, R20)
-	LDP	40(RSP), (R21, R22)
-	LDP	56(RSP), (R23, R24)
-	LDP	72(RSP), (R25, R26)
-	LDP	88(RSP), (R27,   g)
-	ADD	$112, RSP
+	MOVD	(8*21)(RSP), R29
+	RESTORE_F8_TO_F15(8*13)
+	RESTORE_R19_TO_R28(8*3)
+	ADD	$176, RSP
 	JMP	(LR)
 
 noswitch:
diff --git a/src/runtime/rt0_darwin_arm64.s b/src/runtime/rt0_darwin_arm64.s
index 00403612154..697104ac641 100644
--- a/src/runtime/rt0_darwin_arm64.s
+++ b/src/runtime/rt0_darwin_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 TEXT _rt0_arm64_darwin(SB),NOSPLIT|NOFRAME,$0
 	MOVD	$runtime·rt0_go(SB), R2
@@ -18,26 +19,10 @@ exit:
 //
 // Note that all currently shipping darwin/arm64 platforms require
 // cgo and do not support c-shared.
-TEXT _rt0_arm64_darwin_lib(SB),NOSPLIT,$168
+TEXT _rt0_arm64_darwin_lib(SB),NOSPLIT,$152
 	// Preserve callee-save registers.
-	MOVD R19, 24(RSP)
-	MOVD R20, 32(RSP)
-	MOVD R21, 40(RSP)
-	MOVD R22, 48(RSP)
-	MOVD R23, 56(RSP)
-	MOVD R24, 64(RSP)
-	MOVD R25, 72(RSP)
-	MOVD R26, 80(RSP)
-	MOVD R27, 88(RSP)
-	MOVD g, 96(RSP)
-	FMOVD F8, 104(RSP)
-	FMOVD F9, 112(RSP)
-	FMOVD F10, 120(RSP)
-	FMOVD F11, 128(RSP)
-	FMOVD F12, 136(RSP)
-	FMOVD F13, 144(RSP)
-	FMOVD F14, 152(RSP)
-	FMOVD F15, 160(RSP)
+	SAVE_R19_TO_R28(8)
+	SAVE_F8_TO_F15(88)
 
 	MOVD  R0, _rt0_arm64_darwin_lib_argc<>(SB)
 	MOVD  R1, _rt0_arm64_darwin_lib_argv<>(SB)
@@ -57,24 +42,8 @@ TEXT _rt0_arm64_darwin_lib(SB),NOSPLIT,$168
 	ADD   $16, RSP
 
 	// Restore callee-save registers.
-	MOVD 24(RSP), R19
-	MOVD 32(RSP), R20
-	MOVD 40(RSP), R21
-	MOVD 48(RSP), R22
-	MOVD 56(RSP), R23
-	MOVD 64(RSP), R24
-	MOVD 72(RSP), R25
-	MOVD 80(RSP), R26
-	MOVD 88(RSP), R27
-	MOVD 96(RSP), g
-	FMOVD 104(RSP), F8
-	FMOVD 112(RSP), F9
-	FMOVD 120(RSP), F10
-	FMOVD 128(RSP), F11
-	FMOVD 136(RSP), F12
-	FMOVD 144(RSP), F13
-	FMOVD 152(RSP), F14
-	FMOVD 160(RSP), F15
+	RESTORE_R19_TO_R28(8)
+	RESTORE_F8_TO_F15(88)
 
 	RET
 
diff --git a/src/runtime/rt0_freebsd_arm64.s b/src/runtime/rt0_freebsd_arm64.s
index a938d98262e..e517ae059dc 100644
--- a/src/runtime/rt0_freebsd_arm64.s
+++ b/src/runtime/rt0_freebsd_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 // On FreeBSD argc/argv are passed in R0, not RSP
 TEXT _rt0_arm64_freebsd(SB),NOSPLIT|NOFRAME,$0
@@ -14,24 +15,8 @@ TEXT _rt0_arm64_freebsd(SB),NOSPLIT|NOFRAME,$0
 // library is loaded.
 TEXT _rt0_arm64_freebsd_lib(SB),NOSPLIT,$184
 	// Preserve callee-save registers.
-	MOVD R19, 24(RSP)
-	MOVD R20, 32(RSP)
-	MOVD R21, 40(RSP)
-	MOVD R22, 48(RSP)
-	MOVD R23, 56(RSP)
-	MOVD R24, 64(RSP)
-	MOVD R25, 72(RSP)
-	MOVD R26, 80(RSP)
-	MOVD R27, 88(RSP)
-	FMOVD F8, 96(RSP)
-	FMOVD F9, 104(RSP)
-	FMOVD F10, 112(RSP)
-	FMOVD F11, 120(RSP)
-	FMOVD F12, 128(RSP)
-	FMOVD F13, 136(RSP)
-	FMOVD F14, 144(RSP)
-	FMOVD F15, 152(RSP)
-	MOVD g, 160(RSP)
+	SAVE_R19_TO_R28(24)
+	SAVE_F8_TO_F15(104)
 
 	// Initialize g as null in case of using g later e.g. sigaction in cgo_sigaction.go
 	MOVD	ZR, g
@@ -63,24 +48,8 @@ nocgo:
 
 restore:
 	// Restore callee-save registers.
-	MOVD 24(RSP), R19
-	MOVD 32(RSP), R20
-	MOVD 40(RSP), R21
-	MOVD 48(RSP), R22
-	MOVD 56(RSP), R23
-	MOVD 64(RSP), R24
-	MOVD 72(RSP), R25
-	MOVD 80(RSP), R26
-	MOVD 88(RSP), R27
-	FMOVD 96(RSP), F8
-	FMOVD 104(RSP), F9
-	FMOVD 112(RSP), F10
-	FMOVD 120(RSP), F11
-	FMOVD 128(RSP), F12
-	FMOVD 136(RSP), F13
-	FMOVD 144(RSP), F14
-	FMOVD 152(RSP), F15
-	MOVD 160(RSP), g
+	RESTORE_R19_TO_R28(24)
+	RESTORE_F8_TO_F15(104)
 	RET
 
 TEXT _rt0_arm64_freebsd_lib_go(SB),NOSPLIT,$0
diff --git a/src/runtime/rt0_linux_arm64.s b/src/runtime/rt0_linux_arm64.s
index f48a8d61902..0eb8fc2f481 100644
--- a/src/runtime/rt0_linux_arm64.s
+++ b/src/runtime/rt0_linux_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 TEXT _rt0_arm64_linux(SB),NOSPLIT|NOFRAME,$0
 	MOVD	0(RSP), R0	// argc
@@ -13,24 +14,8 @@ TEXT _rt0_arm64_linux(SB),NOSPLIT|NOFRAME,$0
 // library is loaded.
 TEXT _rt0_arm64_linux_lib(SB),NOSPLIT,$184
 	// Preserve callee-save registers.
-	MOVD R19, 24(RSP)
-	MOVD R20, 32(RSP)
-	MOVD R21, 40(RSP)
-	MOVD R22, 48(RSP)
-	MOVD R23, 56(RSP)
-	MOVD R24, 64(RSP)
-	MOVD R25, 72(RSP)
-	MOVD R26, 80(RSP)
-	MOVD R27, 88(RSP)
-	FMOVD F8, 96(RSP)
-	FMOVD F9, 104(RSP)
-	FMOVD F10, 112(RSP)
-	FMOVD F11, 120(RSP)
-	FMOVD F12, 128(RSP)
-	FMOVD F13, 136(RSP)
-	FMOVD F14, 144(RSP)
-	FMOVD F15, 152(RSP)
-	MOVD g, 160(RSP)
+	SAVE_R19_TO_R28(24)
+	SAVE_F8_TO_F15(104)
 
 	// Initialize g as null in case of using g later e.g. sigaction in cgo_sigaction.go
 	MOVD	ZR, g
@@ -62,24 +47,8 @@ nocgo:
 
 restore:
 	// Restore callee-save registers.
-	MOVD 24(RSP), R19
-	MOVD 32(RSP), R20
-	MOVD 40(RSP), R21
-	MOVD 48(RSP), R22
-	MOVD 56(RSP), R23
-	MOVD 64(RSP), R24
-	MOVD 72(RSP), R25
-	MOVD 80(RSP), R26
-	MOVD 88(RSP), R27
-	FMOVD 96(RSP), F8
-	FMOVD 104(RSP), F9
-	FMOVD 112(RSP), F10
-	FMOVD 120(RSP), F11
-	FMOVD 128(RSP), F12
-	FMOVD 136(RSP), F13
-	FMOVD 144(RSP), F14
-	FMOVD 152(RSP), F15
-	MOVD 160(RSP), g
+	RESTORE_R19_TO_R28(24)
+	RESTORE_F8_TO_F15(104)
 	RET
 
 TEXT _rt0_arm64_linux_lib_go(SB),NOSPLIT,$0
diff --git a/src/runtime/rt0_netbsd_arm64.s b/src/runtime/rt0_netbsd_arm64.s
index 2f3b5a5a87a..691a8e4be77 100644
--- a/src/runtime/rt0_netbsd_arm64.s
+++ b/src/runtime/rt0_netbsd_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 TEXT _rt0_arm64_netbsd(SB),NOSPLIT|NOFRAME,$0
 	MOVD	0(RSP), R0	// argc
@@ -13,24 +14,8 @@ TEXT _rt0_arm64_netbsd(SB),NOSPLIT|NOFRAME,$0
 // library is loaded.
 TEXT _rt0_arm64_netbsd_lib(SB),NOSPLIT,$184
 	// Preserve callee-save registers.
-	MOVD R19, 24(RSP)
-	MOVD R20, 32(RSP)
-	MOVD R21, 40(RSP)
-	MOVD R22, 48(RSP)
-	MOVD R23, 56(RSP)
-	MOVD R24, 64(RSP)
-	MOVD R25, 72(RSP)
-	MOVD R26, 80(RSP)
-	MOVD R27, 88(RSP)
-	FMOVD F8, 96(RSP)
-	FMOVD F9, 104(RSP)
-	FMOVD F10, 112(RSP)
-	FMOVD F11, 120(RSP)
-	FMOVD F12, 128(RSP)
-	FMOVD F13, 136(RSP)
-	FMOVD F14, 144(RSP)
-	FMOVD F15, 152(RSP)
-	MOVD g, 160(RSP)
+	SAVE_R19_TO_R28(24)
+	SAVE_F8_TO_F15(104)
 
 	// Initialize g as null in case of using g later e.g. sigaction in cgo_sigaction.go
 	MOVD	ZR, g
@@ -62,24 +47,8 @@ nocgo:
 
 restore:
 	// Restore callee-save registers.
-	MOVD 24(RSP), R19
-	MOVD 32(RSP), R20
-	MOVD 40(RSP), R21
-	MOVD 48(RSP), R22
-	MOVD 56(RSP), R23
-	MOVD 64(RSP), R24
-	MOVD 72(RSP), R25
-	MOVD 80(RSP), R26
-	MOVD 88(RSP), R27
-	FMOVD 96(RSP), F8
-	FMOVD 104(RSP), F9
-	FMOVD 112(RSP), F10
-	FMOVD 120(RSP), F11
-	FMOVD 128(RSP), F12
-	FMOVD 136(RSP), F13
-	FMOVD 144(RSP), F14
-	FMOVD 152(RSP), F15
-	MOVD 160(RSP), g
+	RESTORE_R19_TO_R28(24)
+	RESTORE_F8_TO_F15(104)
 	RET
 
 TEXT _rt0_arm64_netbsd_lib_go(SB),NOSPLIT,$0
diff --git a/src/runtime/rt0_openbsd_arm64.s b/src/runtime/rt0_openbsd_arm64.s
index 722fab61292..49d49b34ac4 100644
--- a/src/runtime/rt0_openbsd_arm64.s
+++ b/src/runtime/rt0_openbsd_arm64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 // See comment in runtime/sys_openbsd_arm64.s re this construction.
 #define	INVOKE_SYSCALL	\
@@ -19,24 +20,8 @@ TEXT _rt0_arm64_openbsd(SB),NOSPLIT|NOFRAME,$0
 // library is loaded.
 TEXT _rt0_arm64_openbsd_lib(SB),NOSPLIT,$184
 	// Preserve callee-save registers.
-	MOVD R19, 24(RSP)
-	MOVD R20, 32(RSP)
-	MOVD R21, 40(RSP)
-	MOVD R22, 48(RSP)
-	MOVD R23, 56(RSP)
-	MOVD R24, 64(RSP)
-	MOVD R25, 72(RSP)
-	MOVD R26, 80(RSP)
-	MOVD R27, 88(RSP)
-	FMOVD F8, 96(RSP)
-	FMOVD F9, 104(RSP)
-	FMOVD F10, 112(RSP)
-	FMOVD F11, 120(RSP)
-	FMOVD F12, 128(RSP)
-	FMOVD F13, 136(RSP)
-	FMOVD F14, 144(RSP)
-	FMOVD F15, 152(RSP)
-	MOVD g, 160(RSP)
+	SAVE_R19_TO_R28(24)
+	SAVE_F8_TO_F15(104)
 
 	// Initialize g as null in case of using g later e.g. sigaction in cgo_sigaction.go
 	MOVD	ZR, g
@@ -68,24 +53,8 @@ nocgo:
 
 restore:
 	// Restore callee-save registers.
-	MOVD 24(RSP), R19
-	MOVD 32(RSP), R20
-	MOVD 40(RSP), R21
-	MOVD 48(RSP), R22
-	MOVD 56(RSP), R23
-	MOVD 64(RSP), R24
-	MOVD 72(RSP), R25
-	MOVD 80(RSP), R26
-	MOVD 88(RSP), R27
-	FMOVD 96(RSP), F8
-	FMOVD 104(RSP), F9
-	FMOVD 112(RSP), F10
-	FMOVD 120(RSP), F11
-	FMOVD 128(RSP), F12
-	FMOVD 136(RSP), F13
-	FMOVD 144(RSP), F14
-	FMOVD 152(RSP), F15
-	MOVD 160(RSP), g
+	RESTORE_R19_TO_R28(24)
+	RESTORE_F8_TO_F15(104)
 	RET
 
 TEXT _rt0_arm64_openbsd_lib_go(SB),NOSPLIT,$0
diff --git a/src/runtime/sys_darwin_arm64.s b/src/runtime/sys_darwin_arm64.s
index e57ac53e108..dc7c9bffa83 100644
--- a/src/runtime/sys_darwin_arm64.s
+++ b/src/runtime/sys_darwin_arm64.s
@@ -9,6 +9,7 @@
 #include "go_asm.h"
 #include "go_tls.h"
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 #define CLOCK_REALTIME		0
 
@@ -175,28 +176,11 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 	BL	(R11)
 	RET
 
-TEXT runtime·sigtramp(SB),NOSPLIT,$192
+TEXT runtime·sigtramp(SB),NOSPLIT,$176
 	// Save callee-save registers in the case of signal forwarding.
 	// Please refer to https://golang.org/issue/31827 .
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 
 	// Save arguments.
 	MOVW	R0, (8*1)(RSP)	// sig
@@ -249,25 +233,8 @@ nog:
 #endif
 
 	// Restore callee-save registers.
-	MOVD	(8*4)(RSP), R19
-	MOVD	(8*5)(RSP), R20
-	MOVD	(8*6)(RSP), R21
-	MOVD	(8*7)(RSP), R22
-	MOVD	(8*8)(RSP), R23
-	MOVD	(8*9)(RSP), R24
-	MOVD	(8*10)(RSP), R25
-	MOVD	(8*11)(RSP), R26
-	MOVD	(8*12)(RSP), R27
-	MOVD	(8*13)(RSP), g
-	MOVD	(8*14)(RSP), R29
-	FMOVD	(8*15)(RSP), F8
-	FMOVD	(8*16)(RSP), F9
-	FMOVD	(8*17)(RSP), F10
-	FMOVD	(8*18)(RSP), F11
-	FMOVD	(8*19)(RSP), F12
-	FMOVD	(8*20)(RSP), F13
-	FMOVD	(8*21)(RSP), F14
-	FMOVD	(8*22)(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 
 	RET
 
@@ -376,25 +343,8 @@ TEXT runtime·mstart_stub(SB),NOSPLIT,$160
 	// We are already on m's g0 stack.
 
 	// Save callee-save registers.
-	MOVD	R19, 8(RSP)
-	MOVD	R20, 16(RSP)
-	MOVD	R21, 24(RSP)
-	MOVD	R22, 32(RSP)
-	MOVD	R23, 40(RSP)
-	MOVD	R24, 48(RSP)
-	MOVD	R25, 56(RSP)
-	MOVD	R26, 64(RSP)
-	MOVD	R27, 72(RSP)
-	MOVD	g, 80(RSP)
-	MOVD	R29, 88(RSP)
-	FMOVD	F8, 96(RSP)
-	FMOVD	F9, 104(RSP)
-	FMOVD	F10, 112(RSP)
-	FMOVD	F11, 120(RSP)
-	FMOVD	F12, 128(RSP)
-	FMOVD	F13, 136(RSP)
-	FMOVD	F14, 144(RSP)
-	FMOVD	F15, 152(RSP)
+	SAVE_R19_TO_R28(8)
+	SAVE_F8_TO_F15(88)
 
 	MOVD	m_g0(R0), g
 	BL	·save_g(SB)
@@ -402,25 +352,8 @@ TEXT runtime·mstart_stub(SB),NOSPLIT,$160
 	BL	runtime·mstart(SB)
 
 	// Restore callee-save registers.
-	MOVD	8(RSP), R19
-	MOVD	16(RSP), R20
-	MOVD	24(RSP), R21
-	MOVD	32(RSP), R22
-	MOVD	40(RSP), R23
-	MOVD	48(RSP), R24
-	MOVD	56(RSP), R25
-	MOVD	64(RSP), R26
-	MOVD	72(RSP), R27
-	MOVD	80(RSP), g
-	MOVD	88(RSP), R29
-	FMOVD	96(RSP), F8
-	FMOVD	104(RSP), F9
-	FMOVD	112(RSP), F10
-	FMOVD	120(RSP), F11
-	FMOVD	128(RSP), F12
-	FMOVD	136(RSP), F13
-	FMOVD	144(RSP), F14
-	FMOVD	152(RSP), F15
+	RESTORE_R19_TO_R28(8)
+	RESTORE_F8_TO_F15(88)
 
 	// Go is all done with this OS thread.
 	// Tell pthread everything is ok (we never join with this thread, so
diff --git a/src/runtime/sys_freebsd_arm64.s b/src/runtime/sys_freebsd_arm64.s
index 59adf4e5f34..7b05fb09004 100644
--- a/src/runtime/sys_freebsd_arm64.s
+++ b/src/runtime/sys_freebsd_arm64.s
@@ -10,6 +10,7 @@
 #include "go_asm.h"
 #include "go_tls.h"
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 #define CLOCK_REALTIME		0
 #define CLOCK_MONOTONIC		4
@@ -279,28 +280,11 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 	RET
 
 // func sigtramp()
-TEXT runtime·sigtramp(SB),NOSPLIT,$192
+TEXT runtime·sigtramp(SB),NOSPLIT,$176
 	// Save callee-save registers in the case of signal forwarding.
 	// Please refer to https://golang.org/issue/31827 .
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 
 	// this might be called in external code context,
 	// where g is not set.
@@ -317,25 +301,8 @@ TEXT runtime·sigtramp(SB),NOSPLIT,$192
 	BL	(R0)
 
 	// Restore callee-save registers.
-	MOVD	8*4(RSP), R19
-	MOVD	8*5(RSP), R20
-	MOVD	8*6(RSP), R21
-	MOVD	8*7(RSP), R22
-	MOVD	8*8(RSP), R23
-	MOVD	8*9(RSP), R24
-	MOVD	8*10(RSP), R25
-	MOVD	8*11(RSP), R26
-	MOVD	8*12(RSP), R27
-	MOVD	8*13(RSP), g
-	MOVD	8*14(RSP), R29
-	FMOVD	8*15(RSP), F8
-	FMOVD	8*16(RSP), F9
-	FMOVD	8*17(RSP), F10
-	FMOVD	8*18(RSP), F11
-	FMOVD	8*19(RSP), F12
-	FMOVD	8*20(RSP), F13
-	FMOVD	8*21(RSP), F14
-	FMOVD	8*22(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 
 	RET
 
diff --git a/src/runtime/sys_linux_arm64.s b/src/runtime/sys_linux_arm64.s
index ca362ed5520..36ac014f03f 100644
--- a/src/runtime/sys_linux_arm64.s
+++ b/src/runtime/sys_linux_arm64.s
@@ -9,6 +9,7 @@
 #include "go_asm.h"
 #include "go_tls.h"
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 #define AT_FDCWD -100
 
@@ -444,28 +445,11 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 	RET
 
 // Called from c-abi, R0: sig, R1: info, R2: cxt
-TEXT runtime·sigtramp(SB),NOSPLIT,$192
+TEXT runtime·sigtramp(SB),NOSPLIT,$176
 	// Save callee-save registers in the case of signal forwarding.
 	// Please refer to https://golang.org/issue/31827 .
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 
 	// this might be called in external code context,
 	// where g is not set.
@@ -481,52 +465,16 @@ TEXT runtime·sigtramp(SB),NOSPLIT,$192
 	BL	(R0)
 
 	// Restore callee-save registers.
-	MOVD	8*4(RSP), R19
-	MOVD	8*5(RSP), R20
-	MOVD	8*6(RSP), R21
-	MOVD	8*7(RSP), R22
-	MOVD	8*8(RSP), R23
-	MOVD	8*9(RSP), R24
-	MOVD	8*10(RSP), R25
-	MOVD	8*11(RSP), R26
-	MOVD	8*12(RSP), R27
-	MOVD	8*13(RSP), g
-	MOVD	8*14(RSP), R29
-	FMOVD	8*15(RSP), F8
-	FMOVD	8*16(RSP), F9
-	FMOVD	8*17(RSP), F10
-	FMOVD	8*18(RSP), F11
-	FMOVD	8*19(RSP), F12
-	FMOVD	8*20(RSP), F13
-	FMOVD	8*21(RSP), F14
-	FMOVD	8*22(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 
 	RET
 
 // Called from c-abi, R0: sig, R1: info, R2: cxt
-TEXT runtime·sigprofNonGoWrapper<>(SB),NOSPLIT,$192
-	// TODO(eric): In multiple places we need to save and restore the
-	// callee-saved registers, we can define a macro for this.
+TEXT runtime·sigprofNonGoWrapper<>(SB),NOSPLIT,$176
 	// Save callee-save registers because it's a callback from c code.
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 
 	MOVW	R0, 8(RSP)	// sig
 	MOVD	R1, 16(RSP)	// info
@@ -534,25 +482,8 @@ TEXT runtime·sigprofNonGoWrapper<>(SB),NOSPLIT,$192
 	CALL	runtime·sigprofNonGo(SB)
 
 	// Restore callee-save registers.
-	MOVD	8*4(RSP), R19
-	MOVD	8*5(RSP), R20
-	MOVD	8*6(RSP), R21
-	MOVD	8*7(RSP), R22
-	MOVD	8*8(RSP), R23
-	MOVD	8*9(RSP), R24
-	MOVD	8*10(RSP), R25
-	MOVD	8*11(RSP), R26
-	MOVD	8*12(RSP), R27
-	MOVD	8*13(RSP), g
-	MOVD	8*14(RSP), R29
-	FMOVD	8*15(RSP), F8
-	FMOVD	8*16(RSP), F9
-	FMOVD	8*17(RSP), F10
-	FMOVD	8*18(RSP), F11
-	FMOVD	8*19(RSP), F12
-	FMOVD	8*20(RSP), F13
-	FMOVD	8*21(RSP), F14
-	FMOVD	8*22(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 	RET
 
 // Called from c-abi, R0: sig, R1: info, R2: cxt
diff --git a/src/runtime/sys_netbsd_arm64.s b/src/runtime/sys_netbsd_arm64.s
index f7cce57c2d6..32e6740c520 100644
--- a/src/runtime/sys_netbsd_arm64.s
+++ b/src/runtime/sys_netbsd_arm64.s
@@ -9,6 +9,7 @@
 #include "go_asm.h"
 #include "go_tls.h"
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 #define CLOCK_REALTIME		0
 #define CLOCK_MONOTONIC		3
@@ -294,34 +295,17 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 	BL	(R11)
 	RET
 
-TEXT runtime·sigtramp(SB),NOSPLIT,$192
+TEXT runtime·sigtramp(SB),NOSPLIT,$176
 	// Save callee-save registers in the case of signal forwarding.
 	// Please refer to https://golang.org/issue/31827 .
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 	// Unclobber g for now (kernel uses it as ucontext ptr)
 	// See https://github.com/golang/go/issues/30824#issuecomment-492772426
 	// This is only correct in the non-cgo case.
 	// XXX should use lwp_getprivate as suggested.
 	// 8*36 is ucontext.uc_mcontext.__gregs[_REG_X28]
 	MOVD	8*36(g), g
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
 
 	// this might be called in external code context,
 	// where g is not set.
@@ -338,25 +322,8 @@ TEXT runtime·sigtramp(SB),NOSPLIT,$192
 	BL	runtime·sigtrampgo(SB)
 
 	// Restore callee-save registers.
-	MOVD	8*4(RSP), R19
-	MOVD	8*5(RSP), R20
-	MOVD	8*6(RSP), R21
-	MOVD	8*7(RSP), R22
-	MOVD	8*8(RSP), R23
-	MOVD	8*9(RSP), R24
-	MOVD	8*10(RSP), R25
-	MOVD	8*11(RSP), R26
-	MOVD	8*12(RSP), R27
-	MOVD	8*13(RSP), g
-	MOVD	8*14(RSP), R29
-	FMOVD	8*15(RSP), F8
-	FMOVD	8*16(RSP), F9
-	FMOVD	8*17(RSP), F10
-	FMOVD	8*18(RSP), F11
-	FMOVD	8*19(RSP), F12
-	FMOVD	8*20(RSP), F13
-	FMOVD	8*21(RSP), F14
-	FMOVD	8*22(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 
 	RET
 
diff --git a/src/runtime/sys_openbsd_arm64.s b/src/runtime/sys_openbsd_arm64.s
index 3fa7e1ede25..7c1886e0a04 100644
--- a/src/runtime/sys_openbsd_arm64.s
+++ b/src/runtime/sys_openbsd_arm64.s
@@ -11,6 +11,7 @@
 #include "go_asm.h"
 #include "go_tls.h"
 #include "textflag.h"
+#include "cgo/abi_arm64.h"
 
 #define CLOCK_REALTIME	$0
 #define	CLOCK_MONOTONIC	$3
@@ -18,30 +19,13 @@
 // mstart_stub is the first function executed on a new thread started by pthread_create.
 // It just does some low-level setup and then calls mstart.
 // Note: called with the C calling convention.
-TEXT runtime·mstart_stub(SB),NOSPLIT,$160
+TEXT runtime·mstart_stub(SB),NOSPLIT,$144
 	// R0 points to the m.
 	// We are already on m's g0 stack.
 
 	// Save callee-save registers.
-	MOVD	R19, 8(RSP)
-	MOVD	R20, 16(RSP)
-	MOVD	R21, 24(RSP)
-	MOVD	R22, 32(RSP)
-	MOVD	R23, 40(RSP)
-	MOVD	R24, 48(RSP)
-	MOVD	R25, 56(RSP)
-	MOVD	R26, 64(RSP)
-	MOVD	R27, 72(RSP)
-	MOVD	g, 80(RSP)
-	MOVD	R29, 88(RSP)
-	FMOVD	F8, 96(RSP)
-	FMOVD	F9, 104(RSP)
-	FMOVD	F10, 112(RSP)
-	FMOVD	F11, 120(RSP)
-	FMOVD	F12, 128(RSP)
-	FMOVD	F13, 136(RSP)
-	FMOVD	F14, 144(RSP)
-	FMOVD	F15, 152(RSP)
+	SAVE_R19_TO_R28(8)
+	SAVE_F8_TO_F15(88)
 
 	MOVD    m_g0(R0), g
 	BL	runtime·save_g(SB)
@@ -49,25 +33,8 @@ TEXT runtime·mstart_stub(SB),NOSPLIT,$160
 	BL	runtime·mstart(SB)
 
 	// Restore callee-save registers.
-	MOVD	8(RSP), R19
-	MOVD	16(RSP), R20
-	MOVD	24(RSP), R21
-	MOVD	32(RSP), R22
-	MOVD	40(RSP), R23
-	MOVD	48(RSP), R24
-	MOVD	56(RSP), R25
-	MOVD	64(RSP), R26
-	MOVD	72(RSP), R27
-	MOVD	80(RSP), g
-	MOVD	88(RSP), R29
-	FMOVD	96(RSP), F8
-	FMOVD	104(RSP), F9
-	FMOVD	112(RSP), F10
-	FMOVD	120(RSP), F11
-	FMOVD	128(RSP), F12
-	FMOVD	136(RSP), F13
-	FMOVD	144(RSP), F14
-	FMOVD	152(RSP), F15
+	RESTORE_R19_TO_R28(8)
+	RESTORE_F8_TO_F15(88)
 
 	// Go is all done with this OS thread.
 	// Tell pthread everything is ok (we never join with this thread, so
@@ -87,25 +54,8 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 TEXT runtime·sigtramp(SB),NOSPLIT,$192
 	// Save callee-save registers in the case of signal forwarding.
 	// Please refer to https://golang.org/issue/31827 .
-	MOVD	R19, 8*4(RSP)
-	MOVD	R20, 8*5(RSP)
-	MOVD	R21, 8*6(RSP)
-	MOVD	R22, 8*7(RSP)
-	MOVD	R23, 8*8(RSP)
-	MOVD	R24, 8*9(RSP)
-	MOVD	R25, 8*10(RSP)
-	MOVD	R26, 8*11(RSP)
-	MOVD	R27, 8*12(RSP)
-	MOVD	g, 8*13(RSP)
-	MOVD	R29, 8*14(RSP)
-	FMOVD	F8, 8*15(RSP)
-	FMOVD	F9, 8*16(RSP)
-	FMOVD	F10, 8*17(RSP)
-	FMOVD	F11, 8*18(RSP)
-	FMOVD	F12, 8*19(RSP)
-	FMOVD	F13, 8*20(RSP)
-	FMOVD	F14, 8*21(RSP)
-	FMOVD	F15, 8*22(RSP)
+	SAVE_R19_TO_R28(8*4)
+	SAVE_F8_TO_F15(8*14)
 
 	// If called from an external code context, g will not be set.
 	// Save R0, since runtime·load_g will clobber it.
@@ -117,25 +67,8 @@ TEXT runtime·sigtramp(SB),NOSPLIT,$192
 	BL	runtime·sigtrampgo(SB)
 
 	// Restore callee-save registers.
-	MOVD	8*4(RSP), R19
-	MOVD	8*5(RSP), R20
-	MOVD	8*6(RSP), R21
-	MOVD	8*7(RSP), R22
-	MOVD	8*8(RSP), R23
-	MOVD	8*9(RSP), R24
-	MOVD	8*10(RSP), R25
-	MOVD	8*11(RSP), R26
-	MOVD	8*12(RSP), R27
-	MOVD	8*13(RSP), g
-	MOVD	8*14(RSP), R29
-	FMOVD	8*15(RSP), F8
-	FMOVD	8*16(RSP), F9
-	FMOVD	8*17(RSP), F10
-	FMOVD	8*18(RSP), F11
-	FMOVD	8*19(RSP), F12
-	FMOVD	8*20(RSP), F13
-	FMOVD	8*21(RSP), F14
-	FMOVD	8*22(RSP), F15
+	RESTORE_R19_TO_R28(8*4)
+	RESTORE_F8_TO_F15(8*14)
 
 	RET
 
diff --git a/src/runtime/sys_windows_arm64.s b/src/runtime/sys_windows_arm64.s
index 87f8f0d2183..7b1514f552b 100644
--- a/src/runtime/sys_windows_arm64.s
+++ b/src/runtime/sys_windows_arm64.s
@@ -7,6 +7,7 @@
 #include "textflag.h"
 #include "funcdata.h"
 #include "time_windows.h"
+#include "cgo/abi_arm64.h"
 
 // Offsets into Thread Environment Block (pointer in R18)
 #define TEB_error 0x68
@@ -128,30 +129,6 @@ TEXT runtime·getlasterror(SB),NOSPLIT|NOFRAME,$0
 	MOVD	R0, ret+0(FP)
 	RET
 
-#define SAVE_R19_TO_R28(offset) \
-	MOVD	R19, savedR19+((offset)+0*8)(SP); \
-	MOVD	R20, savedR20+((offset)+1*8)(SP); \
-	MOVD	R21, savedR21+((offset)+2*8)(SP); \
-	MOVD	R22, savedR22+((offset)+3*8)(SP); \
-	MOVD	R23, savedR23+((offset)+4*8)(SP); \
-	MOVD	R24, savedR24+((offset)+5*8)(SP); \
-	MOVD	R25, savedR25+((offset)+6*8)(SP); \
-	MOVD	R26, savedR26+((offset)+7*8)(SP); \
-	MOVD	R27, savedR27+((offset)+8*8)(SP); \
-	MOVD	g, savedR28+((offset)+9*8)(SP);
-
-#define RESTORE_R19_TO_R28(offset) \
-	MOVD	savedR19+((offset)+0*8)(SP), R19; \
-	MOVD	savedR20+((offset)+1*8)(SP), R20; \
-	MOVD	savedR21+((offset)+2*8)(SP), R21; \
-	MOVD	savedR22+((offset)+3*8)(SP), R22; \
-	MOVD	savedR23+((offset)+4*8)(SP), R23; \
-	MOVD	savedR24+((offset)+5*8)(SP), R24; \
-	MOVD	savedR25+((offset)+6*8)(SP), R25; \
-	MOVD	savedR26+((offset)+7*8)(SP), R26; \
-	MOVD	savedR27+((offset)+8*8)(SP), R27; \
-	MOVD	savedR28+((offset)+9*8)(SP), g; /* R28 */
-
 // Called by Windows as a Vectored Exception Handler (VEH).
 // First argument is pointer to struct containing
 // exception record and context pointers.
@@ -221,7 +198,8 @@ TEXT sigtramp_g0<>(SB),NOSPLIT,$128
 	NO_LOCAL_POINTERS
 
 	// Push C callee-save registers R19-R28. LR, FP already saved.
-	SAVE_R19_TO_R28(-10*8)
+	// These registers will occupy the upper 10 words of the frame.
+	SAVE_R19_TO_R28(8*7)
 
 	MOVD	0(R0), R5	// R5 = ExceptionPointers->ExceptionRecord
 	MOVD	8(R0), R6	// R6 = ExceptionPointers->ContextRecord
@@ -275,7 +253,7 @@ TEXT sigtramp_g0<>(SB),NOSPLIT,$128
 	MOVD	R2, context_pc(R6)
 
 return:
-	RESTORE_R19_TO_R28(-10*8)		// smashes g
+	RESTORE_R19_TO_R28(8*7)		// smashes g
 	RET
 
 // Trampoline to resume execution from exception handler.
@@ -317,18 +295,14 @@ TEXT runtime·callbackasm1(SB),NOSPLIT,$208-0
 	// but we are not called from Go so that space is ours to use,
 	// and we must to be contiguous with the stack arguments.
 	MOVD	$arg0-(7*8)(SP), R14
-	MOVD	R0, (0*8)(R14)
-	MOVD	R1, (1*8)(R14)
-	MOVD	R2, (2*8)(R14)
-	MOVD	R3, (3*8)(R14)
-	MOVD	R4, (4*8)(R14)
-	MOVD	R5, (5*8)(R14)
-	MOVD	R6, (6*8)(R14)
-	MOVD	R7, (7*8)(R14)
+	STP	(R0, R1), (0*8)(R14)
+	STP	(R2, R3), (2*8)(R14)
+	STP	(R4, R5), (4*8)(R14)
+	STP	(R6, R7), (6*8)(R14)
 
 	// Push C callee-save registers R19-R28.
 	// LR, FP already saved.
-	SAVE_R19_TO_R28(-18*8)
+	SAVE_R19_TO_R28(8*9)
 
 	// Create a struct callbackArgs on our stack.
 	MOVD	$cbargs-(18*8+callbackArgs__size)(SP), R13
@@ -342,8 +316,7 @@ TEXT runtime·callbackasm1(SB),NOSPLIT,$208-0
 	MOVD	$·callbackWrap<ABIInternal>(SB), R0	// PC of function to call, cgocallback takes an ABIInternal entry-point
 	MOVD	R13, R1	// frame (&callbackArgs{...})
 	MOVD	$0, R2	// context
-	MOVD	R0, (1*8)(RSP)
-	MOVD	R1, (2*8)(RSP)
+	STP	(R0, R1), (1*8)(RSP)
 	MOVD	R2, (3*8)(RSP)
 	BL	runtime·cgocallback(SB)
 
@@ -351,13 +324,13 @@ TEXT runtime·callbackasm1(SB),NOSPLIT,$208-0
 	MOVD	$cbargs-(18*8+callbackArgs__size)(SP), R13
 	MOVD	callbackArgs_result(R13), R0
 
-	RESTORE_R19_TO_R28(-18*8)
+	RESTORE_R19_TO_R28(8*9)
 
 	RET
 
 // uint32 tstart_stdcall(M *newm);
 TEXT runtime·tstart_stdcall(SB),NOSPLIT,$96-0
-	SAVE_R19_TO_R28(-10*8)
+	SAVE_R19_TO_R28(8*3)
 
 	MOVD	m_g0(R0), g
 	MOVD	R0, g_m(g)
@@ -374,7 +347,7 @@ TEXT runtime·tstart_stdcall(SB),NOSPLIT,$96-0
 	BL	runtime·emptyfunc(SB)	// fault if stack check is wrong
 	BL	runtime·mstart(SB)
 
-	RESTORE_R19_TO_R28(-10*8)
+	RESTORE_R19_TO_R28(8*3)
 
 	// Exit the thread.
 	MOVD	$0, R0