cmd/link,runtime: move syscalls to libc on iOS

This CL is the darwin/arm and darwin/arm64 equivalent to CL 108679,
110215, 110437, 110438, 111258, 110655.

Updates #17490

Change-Id: Ia95b27b38f9c3535012c566f17a44b4ed26b9db6
Reviewed-on: https://go-review.googlesource.com/111015
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/link/internal/arm/asm.go

@@ -229,6 +229,10 @@ func adddynrel(ctxt *ld.Link, s *sym.Symbol, r *sym.Reloc) bool {
 
 	switch r.Type {
 	case objabi.R_CALLARM:
+		if ctxt.LinkMode == ld.LinkExternal {
+			// External linker will do this relocation.
+			return true
+		}
 		addpltsym(ctxt, targ)
 		r.Sym = ctxt.Syms.Lookup(".plt", 0)
 		r.Add = int64(targ.Plt)

src/cmd/link/internal/arm64/asm.go

@@ -92,8 +92,25 @@ func gentext(ctxt *ld.Link) {
 	initarray_entry.AddAddr(ctxt.Arch, initfunc)
 }
 
+// adddynrel implements just enough to support external linking to
+// the system libc functions used by the runtime.
 func adddynrel(ctxt *ld.Link, s *sym.Symbol, r *sym.Reloc) bool {
-	log.Fatalf("adddynrel not implemented")
+	targ := r.Sym
+
+	switch r.Type {
+	case objabi.R_CALL,
+		objabi.R_PCREL,
+		objabi.R_CALLARM64:
+		if targ.Type != sym.SDYNIMPORT {
+			// nothing to do, the relocation will be laid out in reloc
+			return true
+		}
+		if ctxt.LinkMode == ld.LinkExternal {
+			// External linker will do this relocation.
+			return true
+		}
+	}
+	log.Fatalf("adddynrel not implemented for relocation type %d (%s)", r.Type, sym.RelocName(ctxt.Arch, r.Type))
 	return false
 }
 

src/cmd/vet/all/whitelist/darwin_arm.txt

@@ -8,5 +8,4 @@ runtime/sys_darwin_arm.s: [arm] sigfwd: use of unnamed argument 0(FP); offset 0
 
 // Ok.
 
-runtime/sys_darwin_arm.s: [arm] bsdthread_start: function bsdthread_start missing Go declaration
 runtime/asm_arm.s: [arm] sigreturn: function sigreturn missing Go declaration

src/cmd/vet/all/whitelist/darwin_arm64.txt

@@ -2,7 +2,4 @@
 
 runtime/sys_darwin_arm64.s: [arm64] sigtramp: 24(RSP) should be infostyle+8(FP)
 runtime/sys_darwin_arm64.s: [arm64] sigtramp: 24(RSP) should be infostyle+8(FP)
-runtime/sys_darwin_arm64.s: [arm64] bsdthread_create: RET without writing to 4-byte ret+24(FP)
-runtime/sys_darwin_arm64.s: [arm64] bsdthread_start: function bsdthread_start missing Go declaration
-runtime/sys_darwin_arm64.s: [arm64] bsdthread_register: RET without writing to 4-byte ret+0(FP)
 runtime/asm_arm64.s: [arm64] sigreturn: function sigreturn missing Go declaration

src/runtime/asm_arm.s

@@ -588,9 +588,12 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already.
 	MOVW	g_m(g), R8
+	MOVW	m_gsignal(R8), R3
+	CMP	R3, g
+	BEQ	noswitch
 	MOVW	m_g0(R8), R3
 	CMP	R3, g
-	BEQ	g0
+	BEQ	noswitch
 	BL	gosave<>(SB)
 	MOVW	R0, R5
 	MOVW	R3, R0
@@ -599,7 +602,7 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	MOVW	(g_sched+gobuf_sp)(g), R13
 
 	// Now on a scheduling stack (a pthread-created stack).
-g0:
+noswitch:
 	SUB	$24, R13
 	BIC	$0x7, R13	// alignment for gcc ABI
 	MOVW	R4, 20(R13) // save old g

src/runtime/asm_arm64.s

@@ -869,9 +869,12 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already.
 	MOVD	g_m(g), R8
+	MOVD	m_gsignal(R8), R3
+	CMP	R3, g
+	BEQ	noswitch
 	MOVD	m_g0(R8), R3
 	CMP	R3, g
-	BEQ	g0
+	BEQ	noswitch
 	MOVD	R0, R9	// gosave<> and save_g might clobber R0
 	BL	gosave<>(SB)
 	MOVD	R3, g
@@ -881,11 +884,12 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	MOVD	R9, R0
 
 	// Now on a scheduling stack (a pthread-created stack).
-g0:
+noswitch:
 	// Save room for two of our pointers /*, plus 32 bytes of callee
 	// save area that lives on the caller stack. */
 	MOVD	RSP, R13
 	SUB	$16, R13
+	BIC	$0xf, R13	// alignment for gcc ABI
 	MOVD	R13, RSP
 	MOVD	R4, 0(RSP)	// save old g on stack
 	MOVD	(g_stack+stack_hi)(R4), R4

src/runtime/defs_darwin_arm.go

@@ -258,3 +258,8 @@ type pthreadattr struct {
 	X__sig    int32
 	X__opaque [36]int8
 }
+
+type machTimebaseInfo struct {
+	numer uint32
+	denom uint32
+}

src/runtime/defs_darwin_arm64.go

@@ -261,3 +261,8 @@ type pthreadattr struct {
 	X__sig    int64
 	X__opaque [56]int8
 }
+
+type machTimebaseInfo struct {
+	numer uint32
+	denom uint32
+}

src/runtime/mmap.go

@@ -9,8 +9,7 @@
 // +build !linux !amd64
 // +build !linux !arm64
 // +build !js
-// +build !darwin !amd64
-// +build !darwin !386
+// +build !darwin
 
 package runtime
 

src/runtime/os_darwin.go

@@ -2,8 +2,6 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build 386 amd64
-
 package runtime
 
 import "unsafe"

src/runtime/proc.go

@@ -1201,7 +1201,7 @@ func mstart() {
 	mstart1()
 
 	// Exit this thread.
-	if GOOS == "windows" || GOOS == "solaris" || GOOS == "plan9" || (GOOS == "darwin" && (GOARCH == "amd64" || GOARCH == "386")) {
+	if GOOS == "windows" || GOOS == "solaris" || GOOS == "plan9" || GOOS == "darwin" {
 		// Window, Solaris, Darwin and Plan 9 always system-allocate
 		// the stack, but put it in _g_.stack before mstart,
 		// so the logic above hasn't set osStack yet.
@@ -1525,7 +1525,7 @@ func allocm(_p_ *p, fn func()) *m {
 
 	// In case of cgo or Solaris or Darwin, pthread_create will make us a stack.
 	// Windows and Plan 9 will layout sched stack on OS stack.
-	if iscgo || GOOS == "solaris" || GOOS == "windows" || GOOS == "plan9" || (GOOS == "darwin" && (GOARCH == "386" || GOARCH == "amd64")) {
+	if iscgo || GOOS == "solaris" || GOOS == "windows" || GOOS == "plan9" || GOOS == "darwin" {
 		mp.g0 = malg(-1)
 	} else {
 		mp.g0 = malg(8192 * sys.StackGuardMultiplier)

src/runtime/stubs2.go

@@ -7,8 +7,7 @@
 // +build !windows
 // +build !nacl
 // +build !js
-// +build !darwin !amd64
-// +build !darwin !386
+// +build !darwin
 
 package runtime
 

src/runtime/stubs4.go

-// Copyright 2018 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.
-
-// +build darwin,arm darwin,arm64
-
-package runtime
-
-func nanotime() int64

src/runtime/sys_darwin.go

@@ -2,8 +2,6 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build darwin,386 darwin,amd64
-
 package runtime
 
 import "unsafe"

src/runtime/sys_darwin_arm.s

@@ -11,27 +11,14 @@
 #include "textflag.h"
 
 // Copied from /usr/include/sys/syscall.h
-#define	SYS_exit           1
-#define	SYS_read           3
-#define	SYS_write          4
-#define	SYS_open           5
-#define	SYS_close          6
-#define	SYS_mmap           197
-#define	SYS_munmap         73
-#define	SYS_madvise        75
 #define	SYS_gettimeofday   116
 #define	SYS_kill           37
 #define	SYS_getpid         20
-#define	SYS___pthread_kill 328
 #define	SYS_pthread_sigmask 329
 #define	SYS_setitimer      83
 #define	SYS___sysctl       202
 #define	SYS_sigaction      46
 #define	SYS_sigreturn      184
-#define	SYS_select         93
-#define	SYS_bsdthread_register 366
-#define	SYS_bsdthread_create 360
-#define	SYS_bsdthread_terminate 361
 #define	SYS_kqueue         362
 #define	SYS_kevent         363
 #define	SYS_fcntl          92
@@ -41,85 +28,39 @@ TEXT notok<>(SB),NOSPLIT,$0
 	MOVW	R8, (R8)
 	B		0(PC)
 
-TEXT runtime·open(SB),NOSPLIT,$0
-	MOVW	name+0(FP), R0
-	MOVW	mode+4(FP), R1
-	MOVW	perm+8(FP), R2
-	MOVW	$SYS_open, R12
-	SWI	$0x80
-	MOVW.CS	$-1, R0
-	MOVW	R0, ret+12(FP)
+TEXT runtime·open_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 mode
+	MOVW	8(R0), R2	// arg 3 perm
+	MOVW	0(R0), R0	// arg 1 name
+	BL	libc_open(SB)
 	RET
 
-TEXT runtime·closefd(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVW	$SYS_close, R12
-	SWI	$0x80
-	MOVW.CS	$-1, R0
-	MOVW	R0, ret+4(FP)
+TEXT runtime·close_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 fd
+	BL	libc_close(SB)
 	RET
 
-TEXT runtime·write(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVW	p+4(FP), R1
-	MOVW	n+8(FP), R2
-	MOVW	$SYS_write, R12
-	SWI	$0x80
-	MOVW.CS	$-1, R0
-	MOVW	R0, ret+12(FP)
+TEXT runtime·write_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 buf
+	MOVW	8(R0), R2	// arg 3 count
+	MOVW	0(R0), R0	// arg 1 fd
+	BL	libc_write(SB)
 	RET
 
-TEXT runtime·read(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVW	p+4(FP), R1
-	MOVW	n+8(FP), R2
-	MOVW	$SYS_read, R12
-	SWI	$0x80
-	MOVW.CS	$-1, R0
-	MOVW	R0, ret+12(FP)
+TEXT runtime·read_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 buf
+	MOVW	8(R0), R2	// arg 3 count
+	MOVW	0(R0), R0	// arg 1 fd
+	BL	libc_read(SB)
 	RET
 
-TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
-	MOVW	code+0(FP), R0
-	MOVW	$SYS_exit, R12
-	SWI	$0x80
+TEXT runtime·exit_trampoline(SB),NOSPLIT|NOFRAME,$0
+	MOVW	0(R0), R0	// arg 0 code
+	BL libc_exit(SB)
 	MOVW	$1234, R0
 	MOVW	$1002, R1
 	MOVW	R0, (R1)	// fail hard
 
-// Exit this OS thread (like pthread_exit, which eventually
-// calls __bsdthread_terminate).
-TEXT exit1<>(SB),NOSPLIT,$0
-	// Because of exitThread below, this must not use the stack.
-	// __bsdthread_terminate takes 4 word-size arguments.
-	// Set them all to 0. (None are an exit status.)
-	MOVW	$0, R0
-	MOVW	$0, R1
-	MOVW	$0, R2
-	MOVW	$0, R3
-	MOVW	$SYS_bsdthread_terminate, R12
-	SWI	$0x80
-	MOVW	$1234, R0
-	MOVW	$1003, R1
-	MOVW	R0, (R1)	// fail hard
-
-// func exitThread(wait *uint32)
-TEXT runtime·exitThread(SB),NOSPLIT,$0-4
-	MOVW	wait+0(FP), R0
-	// We're done using the stack.
-	MOVW	$0, R2
-storeloop:
-	LDREX	(R0), R4          // loads R4
-	STREX	R2, (R0), R1      // stores R2
-	CMP	$0, R1
-	BNE	storeloop
-	JMP	exit1<>(SB)
-
-TEXT runtime·raise(SB),NOSPLIT,$0
-	// Ideally we'd send the signal to the current thread,
-	// not the whole process, but that's too hard on OS X.
-	JMP	runtime·raiseproc(SB)
-
 TEXT runtime·raiseproc(SB),NOSPLIT,$24
 	MOVW	$SYS_getpid, R12
 	SWI	$0x80
@@ -130,41 +71,49 @@ TEXT runtime·raiseproc(SB),NOSPLIT,$24
 	SWI $0x80
 	RET
 
-TEXT runtime·mmap(SB),NOSPLIT,$0
-	MOVW	addr+0(FP), R0
-	MOVW	n+4(FP), R1
-	MOVW	prot+8(FP), R2
-	MOVW	flags+12(FP), R3
-	MOVW	fd+16(FP), R4
-	MOVW	off+20(FP), R5
-	MOVW	$0, R6 // off_t is uint64_t
-	MOVW	$SYS_mmap, R12
-	SWI	$0x80
+TEXT runtime·mmap_trampoline(SB),NOSPLIT,$0
+	MOVW	R0, R8
+	MOVW	0(R8), R0	// arg 1 addr
+	MOVW	4(R8), R1	// arg 2 len
+	MOVW	8(R8), R2	// arg 3 prot
+	MOVW	12(R8), R3	// arg 4 flags
+	MOVW	16(R8), R4	// arg 5 fid
+	MOVW	20(R8), R5	// arg 6 offset
+	MOVW	$0, R6	// off_t is uint64_t
+	// Only R0-R3 are used for arguments, the rest
+	// go on the stack.
+	MOVM.DB.W [R4-R6], (R13)
+	BL	libc_mmap(SB)
+	ADD $12, R13
 	MOVW	$0, R1
-	BCC     ok
-	MOVW	R1, p+24(FP)
-	MOVW	R0, err+28(FP)
-	RET
+	MOVW	$-1, R2
+	CMP	R0, R2
+	BNE ok
+	BL	libc_error(SB)
+	MOVW	(R0), R1
+	MOVW	$0, R0
 ok:
-	MOVW	R0, p+24(FP)
-	MOVW	R1, err+28(FP)
+	MOVW	R0, 24(R8)	// ret 1 addr
+	MOVW	R1, 28(R8)	// ret 2 err
 	RET
 
-TEXT runtime·munmap(SB),NOSPLIT,$0
-	MOVW	addr+0(FP), R0
-	MOVW	n+4(FP), R1
-	MOVW	$SYS_munmap, R12
-	SWI	$0x80
-	BL.CS	notok<>(SB)
+TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 len
+	MOVW	0(R0), R0	// arg 1 addr
+	BL libc_munmap(SB)
+	MOVW	$-1, R2
+	CMP	R0, R2
+	BL.EQ	notok<>(SB)
 	RET
 
-TEXT runtime·madvise(SB),NOSPLIT,$0
-	MOVW	addr+0(FP), R0
-	MOVW	n+4(FP), R1
-	MOVW	flags+8(FP), R2
-	MOVW	$SYS_madvise, R12
-	SWI	$0x80
-	BL.CS	notok<>(SB)
+TEXT runtime·madvise_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 len
+	MOVW	8(R0), R2	// arg 3 advice
+	MOVW	0(R0), R0	// arg 1 addr
+	BL	libc_madvise(SB)
+	MOVW	$-1, R2
+	CMP	R0, R2
+	BL.EQ	notok<>(SB)
 	RET
 
 TEXT runtime·setitimer(SB),NOSPLIT,$0
@@ -195,28 +144,36 @@ inreg:
 	MOVW	R2, nsec+8(FP)
 	RET
 
-TEXT runtime·nanotime(SB),NOSPLIT,$32
-	MOVW	$8(R13), R0  // timeval
-	MOVW	$0, R1  // zone
-	MOVW	$0, R2	// see issue 16570
-	MOVW	$SYS_gettimeofday, R12
-	SWI	$0x80 // Note: R0 is tv_sec, R1 is tv_usec
-	CMP	$0, R0
-	BNE	inreg
-	MOVW	8(R13), R0
-	MOVW	12(R13), R1
-inreg:
-	MOVW    R1, R2
-	MOVW	$1000000000, R3
-	MULLU	R0, R3, (R1, R0)
-	MOVW	$1000, R3
-	MOVW	$0, R4
-	MUL	R3, R2
-	ADD.S	R2, R0
-	ADC	R4, R1
+GLOBL timebase<>(SB),NOPTR,$(machTimebaseInfo__size)
+
+TEXT runtime·nanotime_trampoline(SB),NOSPLIT,$0
+	MOVW	R0, R8
+	BL	libc_mach_absolute_time(SB)
+	MOVW	R0, 0(R8)
+	MOVW	R1, 4(R8)
+	MOVW	timebase<>+machTimebaseInfo_numer(SB), R6
+	MOVW	$timebase<>+machTimebaseInfo_denom(SB), R5
+	MOVW	(R5), R7
+	DMB	MB_ISH	// memory barrier for atomic read
+	CMP	$0, R7
+	BNE	initialized
+
+	SUB	$(machTimebaseInfo__size+7)/8*8, R13
+	MOVW	R13, R0
+	BL	libc_mach_timebase_info(SB)
+	MOVW	machTimebaseInfo_numer(R13), R6
+	MOVW	machTimebaseInfo_denom(R13), R7
+	ADD	$(machTimebaseInfo__size+7)/8*8, R13
+
+	MOVW	R6, timebase<>+machTimebaseInfo_numer(SB)
+	MOVW	$timebase<>+machTimebaseInfo_denom(SB), R5
+	DMB	MB_ISH	// memory barrier for atomic write
+	MOVW	R7, (R5)
+	DMB	MB_ISH
 
-	MOVW	R0, ret_lo+0(FP)
-	MOVW	R1, ret_hi+4(FP)
+initialized:
+	MOVW	R6, 8(R8)
+	MOVW	R7, 12(R8)
 	RET
 
 TEXT runtime·sigfwd(SB),NOSPLIT,$0-16
@@ -318,20 +275,9 @@ TEXT runtime·sigaction(SB),NOSPLIT,$0
 	SWI	$0x80
 	RET
 
-TEXT runtime·usleep(SB),NOSPLIT,$12
-	MOVW	usec+0(FP), R0
-	CALL	runtime·usplitR0(SB)
-	MOVW	R0, a-12(SP)
-	MOVW	R1, b-8(SP)
-
-	// select(0, 0, 0, 0, &tv)
-	MOVW	$0, R0
-	MOVW	$0, R1
-	MOVW	$0, R2
-	MOVW	$0, R3
-	MOVW	$a-12(SP), R4
-	MOVW	$SYS_select, R12
-	SWI	$0x80
+TEXT runtime·usleep_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 usec
+	BL libc_usleep(SB)
 	RET
 
 TEXT ·publicationBarrier(SB),NOSPLIT|NOFRAME,$0-0
@@ -355,65 +301,6 @@ sysctl_ret:
 	MOVW	R0, ret+24(FP)
 	RET
 
-// Thread related functions
-// 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+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	$SYS_bsdthread_create, R12
-	SWI	$0x80
-	BCC		create_ret
-	RSB 	$0, R0, R0
-	MOVW	R0, ret+12(FP)
-	RET
-create_ret:
-	MOVW	$0, R0
-	MOVW	R0, ret+12(FP)
-	RET
-
-// The thread that bsdthread_create creates starts executing here,
-// because we registered this function using bsdthread_register
-// at startup.
-//	R0 = "pthread"
-//	R1 = mach thread port
-//	R2 = "func" (= fn)
-//	R3 = "arg" (= m)
-//	R4 = stack
-//	R5 = flags (= 0)
-// XXX: how to deal with R4/SP? ref: Libc-594.9.1/arm/pthreads/thread_start.s
-TEXT runtime·bsdthread_start(SB),NOSPLIT,$0
-	MOVW    R1, m_procid(R3) // thread port is m->procid
-	MOVW	m_g0(R3), g
-	MOVW	R3, g_m(g)
-	// ARM don't have runtime·stackcheck(SB)
-	// disable runfast mode of vfp
-	EOR     R12, R12
-	WORD    $0xeee1ca10 // fmxr	fpscr, ip
-	BL      (R2) // fn
-	BL      exit1<>(SB)
-	RET
-
-// int32 bsdthread_register(void)
-// registers callbacks for threadstart (see bsdthread_create above
-// and wqthread and pthsize (not used).  returns 0 on success.
-TEXT runtime·bsdthread_register(SB),NOSPLIT,$0
-	MOVW	$runtime·bsdthread_start(SB), R0	// threadstart
-	MOVW	$0, R1	// wqthread, not used by us
-	MOVW	$0, R2	// pthsize, not used by us
-	MOVW	$0, R3 	// dummy_value [sic]
-	MOVW	$0, R4	// targetconc_ptr
-	MOVW	$0, R5	// dispatchqueue_offset
-	MOVW	$SYS_bsdthread_register, R12	// bsdthread_register
-	SWI	$0x80
-	MOVW	R0, ret+0(FP)
-	RET
-
 // uint32 mach_msg_trap(void*, uint32, uint32, uint32, uint32, uint32, uint32)
 TEXT runtime·mach_msg_trap(SB),NOSPLIT,$0
 	MOVW    h+0(FP), R0
@@ -519,3 +406,38 @@ TEXT runtime·closeonexec(SB),NOSPLIT,$0
 // to do the stack switch ourselves.
 TEXT runtime·sigaltstack(SB),NOSPLIT,$0
 	RET
+
+// Thread related functions
+// Note: On darwin/arm, the runtime always use runtime/cgo to
+// create threads, so all thread related functions will just exit with a
+// unique status.
+
+TEXT runtime·mstart_stub(SB),NOSPLIT,$0
+	MOVW	$44, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_init_trampoline(SB),NOSPLIT,$0
+	MOVW	$45, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_setstacksize_trampoline(SB),NOSPLIT,$0
+	MOVW	$46, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_setdetachstate_trampoline(SB),NOSPLIT,$0
+	MOVW	$47, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_create_trampoline(SB),NOSPLIT,$0
+	MOVW	$48, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·raise_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 sig
+	BL	libc_raise(SB)
+	RET

src/runtime/sys_darwin_arm64.s

@@ -11,27 +11,14 @@
 #include "textflag.h"
 
 // Copied from /usr/include/sys/syscall.h
-#define	SYS_exit           1
-#define	SYS_read           3
-#define	SYS_write          4
-#define	SYS_open           5
-#define	SYS_close          6
-#define	SYS_mmap           197
-#define	SYS_munmap         73
-#define	SYS_madvise        75
 #define	SYS_gettimeofday   116
 #define	SYS_kill           37
 #define	SYS_getpid         20
-#define	SYS___pthread_kill 328
 #define	SYS_pthread_sigmask 329
 #define	SYS_setitimer      83
 #define	SYS___sysctl       202
 #define	SYS_sigaction      46
 #define	SYS_sigreturn      184
-#define	SYS_select         93
-#define	SYS_bsdthread_register 366
-#define	SYS_bsdthread_create 360
-#define	SYS_bsdthread_terminate 361
 #define	SYS_kqueue         362
 #define	SYS_kevent         363
 #define	SYS_fcntl          92
@@ -41,81 +28,39 @@ TEXT notok<>(SB),NOSPLIT,$0
 	MOVD	R8, (R8)
 	B	0(PC)
 
-TEXT runtime·open(SB),NOSPLIT,$0
-	MOVD	name+0(FP), R0
-	MOVW	mode+8(FP), R1
-	MOVW	perm+12(FP), R2
-	MOVD	$SYS_open, R16
-	SVC	$0x80
-	CSINV	LO, R0, ZR, R0
-	MOVW	R0, ret+16(FP)
+TEXT runtime·open_trampoline(SB),NOSPLIT,$0
+	MOVW	8(R0), R1	// arg 2 flags
+	MOVW	12(R0), R2	// arg 3 mode
+	MOVD	0(R0), R0	// arg 1 pathname
+	BL libc_open(SB)
 	RET
 
-TEXT runtime·closefd(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVW	$SYS_close, R16
-	SVC	$0x80
-	CSINV	LO, R0, ZR, R0
-	MOVW	R0, ret+8(FP)
+TEXT runtime·close_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 fd
+	BL	libc_close(SB)
 	RET
 
-TEXT runtime·write(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVD	p+8(FP), R1
-	MOVW	n+16(FP), R2
-	MOVW	$SYS_write, R16
-	SVC	$0x80
-	CSINV	LO, R0, ZR, R0
-	MOVW	R0, ret+24(FP)
+TEXT runtime·write_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 buf
+	MOVW	16(R0), R2	// arg 3 count
+	MOVW	0(R0), R0	// arg 1 fd
+	BL	libc_write(SB)
 	RET
 
-TEXT runtime·read(SB),NOSPLIT,$0
-	MOVW	fd+0(FP), R0
-	MOVD	p+8(FP), R1
-	MOVW	n+16(FP), R2
-	MOVW	$SYS_read, R16
-	SVC	$0x80
-	CSINV	LO, R0, ZR, R0
-	MOVW	R0, ret+24(FP)
+TEXT runtime·read_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 buf
+	MOVW	16(R0), R2	// arg 3 count
+	MOVW	0(R0), R0	// arg 1 fd
+	BL libc_read(SB)
 	RET
 
-TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
-	MOVW	code+0(FP), R0
-	MOVW	$SYS_exit, R16
-	SVC	$0x80
+TEXT runtime·exit_trampoline(SB),NOSPLIT|NOFRAME,$0
+	MOVW	0(R0), R0
+	BL	libc_exit(SB)
 	MOVD	$1234, R0
 	MOVD	$1002, R1
 	MOVD	R0, (R1)	// fail hard
 
-// Exit this OS thread (like pthread_exit, which eventually
-// calls __bsdthread_terminate).
-TEXT exit1<>(SB),NOSPLIT,$0
-	// Because of exitThread below, this must not use the stack.
-	// __bsdthread_terminate takes 4 word-size arguments.
-	// Set them all to 0. (None are an exit status.)
-	MOVW	$0, R0
-	MOVW	$0, R1
-	MOVW	$0, R2
-	MOVW	$0, R3
-	MOVW	$SYS_bsdthread_terminate, R16
-	SVC	$0x80
-	MOVD	$1234, R0
-	MOVD	$1003, R1
-	MOVD	R0, (R1)	// fail hard
-
-// func exitThread(wait *uint32)
-TEXT runtime·exitThread(SB),NOSPLIT,$0-8
-	MOVD	wait+0(FP), R0
-	// We're done using the stack.
-	MOVW	$0, R1
-	STLRW	R1, (R0)
-	JMP	exit1<>(SB)
-
-TEXT runtime·raise(SB),NOSPLIT,$0
-	// Ideally we'd send the signal to the current thread,
-	// not the whole process, but that's too hard on OS X.
-	JMP	runtime·raiseproc(SB)
-
 TEXT runtime·raiseproc(SB),NOSPLIT,$0
 	MOVW	$SYS_getpid, R16
 	SVC	$0x80
@@ -126,41 +71,41 @@ TEXT runtime·raiseproc(SB),NOSPLIT,$0
 	SVC	$0x80
 	RET
 
-TEXT runtime·mmap(SB),NOSPLIT,$0
-	MOVD	addr+0(FP), R0
-	MOVD	n+8(FP), R1
-	MOVW	prot+16(FP), R2
-	MOVW	flags+20(FP), R3
-	MOVW	fd+24(FP), R4
-	MOVW	off+28(FP), R5
-	MOVW	$SYS_mmap, R16
-	SVC	$0x80
-	BCC	ok
-	MOVD	$0, p+32(FP)
-	MOVD	R0, err+40(FP)
-	RET
+TEXT runtime·mmap_trampoline(SB),NOSPLIT,$0
+	MOVD	R0, R19
+	MOVD	0(R19), R0	// arg 1 addr
+	MOVD	8(R19), R1	// arg 2 len
+	MOVW	16(R19), R2	// arg 3 prot
+	MOVW	20(R19), R3	// arg 4 flags
+	MOVW	24(R19), R4	// arg 5 fd
+	MOVW	28(R19), R5	// arg 6 off
+	BL	libc_mmap(SB)
+	MOVD	$0, R1
+	MOVD	$-1, R2
+	CMP	R0, R2
+	BNE	ok
+	BL libc_error(SB)
+	MOVD	(R0), R1
+	MOVD	$0, R0
 ok:
-	MOVD	R0, p+32(FP)
-	MOVD	$0, err+40(FP)
+	MOVD	R0, 32(R19) // ret 1 p
+	MOVD	R1, 40(R19)	// ret 2 err
 	RET
 
-TEXT runtime·munmap(SB),NOSPLIT,$0
-	MOVD	addr+0(FP), R0
-	MOVD	n+8(FP), R1
-	MOVW	$SYS_munmap, R16
-	SVC	$0x80
-	BCC	2(PC)
+TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 len
+	MOVD	0(R0), R0	// arg 1 addr
+	BL	libc_munmap(SB)
+	CMP $0, R0
+	BEQ 2(PC)
 	BL	notok<>(SB)
 	RET
 
-TEXT runtime·madvise(SB),NOSPLIT,$0
-	MOVD	addr+0(FP), R0
-	MOVD	n+8(FP), R1
-	MOVW	flags+16(FP), R2
-	MOVW	$SYS_madvise, R16
-	SVC	$0x80
-	BCC	2(PC)
-	BL	notok<>(SB)
+TEXT runtime·madvise_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 len
+	MOVW	16(R0), R2	// arg 3 advice
+	MOVD	0(R0), R0	// arg 1 addr
+	BL	libc_madvise(SB)
 	RET
 
 TEXT runtime·setitimer(SB),NOSPLIT,$0
@@ -189,25 +134,32 @@ inreg:
 	MOVW	R1, nsec+8(FP)
 	RET
 
-TEXT runtime·nanotime(SB),NOSPLIT,$40
-	MOVD	RSP, R0	// timeval
-	MOVD	R0, R9	// this is how dyld calls gettimeofday
-	MOVW	$0, R1	// zone
-	MOVD	$0, R2	// see issue 16570
-	MOVW	$SYS_gettimeofday, R16
-	SVC	$0x80	// Note: x0 is tv_sec, w1 is tv_usec
-	CMP	$0, R0
-	BNE	inreg
-	MOVD	0(RSP), R0
-	MOVW	8(RSP), R1
-inreg:
-	MOVW	$1000000000, R3
-	MUL	R3, R0
-	MOVW	$1000, R3
-	MUL	R3, R1
-	ADD	R1, R0
+GLOBL timebase<>(SB),NOPTR,$(machTimebaseInfo__size)
+
+TEXT runtime·nanotime_trampoline(SB),NOSPLIT,$40
+	MOVD	R0, R19
+	BL	libc_mach_absolute_time(SB)
+	MOVD	R0, 0(R19)
+	MOVW	timebase<>+machTimebaseInfo_numer(SB), R20
+	MOVD	$timebase<>+machTimebaseInfo_denom(SB), R21
+	LDARW	(R21), R21	// atomic read
+	CMP	$0, R21
+	BNE	initialized
+
+	SUB	$(machTimebaseInfo__size+15)/16*16, RSP
+	MOVD	RSP, R0
+	BL	libc_mach_timebase_info(SB)
+	MOVW	machTimebaseInfo_numer(RSP), R20
+	MOVW	machTimebaseInfo_denom(RSP), R21
+	ADD	$(machTimebaseInfo__size+15)/16*16, RSP
+
+	MOVW	R20, timebase<>+machTimebaseInfo_numer(SB)
+	MOVD	$timebase<>+machTimebaseInfo_denom(SB), R22
+	STLRW	R21, (R22)	// atomic write
 
-	MOVD	R0, ret+0(FP)
+initialized:
+	MOVW	R20, 8(R19)
+	MOVW	R21, 12(R19)
 	RET
 
 TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
@@ -310,24 +262,9 @@ TEXT runtime·sigaction(SB),NOSPLIT,$0
 	BL	notok<>(SB)
 	RET
 
-TEXT runtime·usleep(SB),NOSPLIT,$24
-	MOVW	usec+0(FP), R0
-	MOVW	R0, R1
-	MOVW	$1000000, R2
-	UDIV	R2, R0
-	MUL	R0, R2
-	SUB	R2, R1
-	MOVD	R0, 0(RSP)
-	MOVW	R1, 8(RSP)
-
-	// select(0, 0, 0, 0, &tv)
-	MOVW	$0, R0
-	MOVW	$0, R1
-	MOVW	$0, R2
-	MOVW	$0, R3
-	MOVD	RSP, R4
-	MOVW	$SYS_select, R16
-	SVC	$0x80
+TEXT runtime·usleep_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 usec
+	BL	libc_usleep(SB)
 	RET
 
 TEXT runtime·sysctl(SB),NOSPLIT,$0
@@ -348,42 +285,6 @@ ok:
 	MOVW	R0, ret+48(FP)
 	RET
 
-// Thread related functions
-// Note: On darwin/arm64, it is no longer possible to use bsdthread_register
-// as the libc is always linked in. The runtime must use runtime/cgo to
-// create threads, so all thread related functions will just exit with a
-// unique status.
-// void bsdthread_create(void *stk, M *m, G *g, void (*fn)(void))
-TEXT runtime·bsdthread_create(SB),NOSPLIT,$0
-	MOVD	$44, R0
-	MOVW	$SYS_exit, R16
-	SVC	$0x80
-	RET
-
-// The thread that bsdthread_create creates starts executing here,
-// because we registered this function using bsdthread_register
-// at startup.
-//	R0 = "pthread"
-//	R1 = mach thread port
-//	R2 = "func" (= fn)
-//	R3 = "arg" (= m)
-//	R4 = stack
-//	R5 = flags (= 0)
-TEXT runtime·bsdthread_start(SB),NOSPLIT,$0
-	MOVD	$45, R0
-	MOVW	$SYS_exit, R16
-	SVC	$0x80
-	RET
-
-// int32 bsdthread_register(void)
-// registers callbacks for threadstart (see bsdthread_create above
-// and wqthread and pthsize (not used).  returns 0 on success.
-TEXT runtime·bsdthread_register(SB),NOSPLIT,$0
-	MOVD	$46, R0
-	MOVW	$SYS_exit, R16
-	SVC	$0x80
-	RET
-
 // uint32 mach_msg_trap(void*, uint32, uint32, uint32, uint32, uint32, uint32)
 TEXT runtime·mach_msg_trap(SB),NOSPLIT,$0
 	MOVD	h+0(FP), R0
@@ -491,3 +392,38 @@ TEXT runtime·closeonexec(SB),NOSPLIT,$0
 // to do the stack switch ourselves.
 TEXT runtime·sigaltstack(SB),NOSPLIT,$0
 	RET
+
+// Thread related functions
+// Note: On darwin/arm64, the runtime always use runtime/cgo to
+// create threads, so all thread related functions will just exit with a
+// unique status.
+
+TEXT runtime·mstart_stub(SB),NOSPLIT,$0
+	MOVW	$44, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_init_trampoline(SB),NOSPLIT,$0
+	MOVW	$45, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_setstacksize_trampoline(SB),NOSPLIT,$0
+	MOVW	$46, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_attr_setdetachstate_trampoline(SB),NOSPLIT,$0
+	MOVW	$47, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·pthread_create_trampoline(SB),NOSPLIT,$0
+	MOVW	$48, R0
+	BL	libc_exit(SB)
+	RET
+
+TEXT runtime·raise_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R0	// arg 1 sig
+	BL	libc_raise(SB)
+	RET

src/runtime/vlrt.go

@@ -106,7 +106,6 @@ func _d2v(y *uint64, d float64) {
 
 	*y = uint64(yhi)<<32 | uint64(ylo)
 }
-
 func uint64div(n, d uint64) uint64 {
 	// Check for 32 bit operands
 	if uint32(n>>32) == 0 && uint32(d>>32) == 0 {
@@ -131,6 +130,9 @@ func uint64mod(n, d uint64) uint64 {
 	return r
 }
 
+//go:nosplit
+// nosplit because division is used in syscall context in nanotime on darwin/386
+// and darwin/arm where stack splits are not allowed.
 func int64div(n, d int64) int64 {
 	// Check for 32 bit operands
 	if int64(int32(n)) == n && int64(int32(d)) == d {
@@ -191,6 +193,7 @@ func _mul64by32(lo64 *uint64, a uint64, b uint32) (hi32 uint32)
 //go:noescape
 func _div64by32(a uint64, b uint32, r *uint32) (q uint32)
 
+//go:nosplit
 func dodiv(n, d uint64) (q, r uint64) {
 	if GOARCH == "arm" {
 		// arm doesn't have a division instruction, so
@@ -234,6 +237,7 @@ func dodiv(n, d uint64) (q, r uint64) {
 	return uint64(qhi)<<32 + uint64(qlo), uint64(rlo)
 }
 
+//go:nosplit
 func slowdodiv(n, d uint64) (q, r uint64) {
 	if d == 0 {
 		panicdivide()