runtime: clean up system calls during cgo callback init

During a cgocallback, the runtime calls needm to get an m.
The calls made during needm cannot themselves assume that
there is an m or a g (which is attached to the m).

In the old days of making direct system calls, the only thing
you had to do for such functions was mark them //go:nosplit,
to avoid the use of g in the stack split prologue.

But now, on operating systems that make system calls through
shared libraries and use code that saves state in the g or m
before doing so, it's not safe to assume g exists. In fact, it is
not even safe to call getg(), because it might fault deferencing
the TLS storage to find the g pointer (that storage may not be
initialized yet, at least on Windows, and perhaps on other systems
in the future).

The specific routines that are problematic are usleep and osyield,
which are called during lock contention in lockextra, called
from needm.

All this is rather subtle and hidden, so in addition to fixing the
problem on Windows, this CL makes the fact of not running on
a g much clearer by introducing variants usleep_no_g and
osyield_no_g whose names should make clear that there is no g.
And then we can remove the various sketchy getg() == nil checks
in the existing routines.

As part of this cleanup, this CL also deletes onosstack on Windows.
onosstack is from back when the runtime was implemented in C.
It predates systemstack but does essentially the same thing.
Instead of having two different copies of this code, we can use
systemstack consistently. This way we need not port onosstack
to each architecture.

This CL is part of a stack adding windows/arm64
support (#36439), intended to land in the Go 1.17 cycle.
This CL is, however, not windows/arm64-specific.
It is cleanup meant to make the port (and future ports) easier.

Change-Id: I3352de1fd0a3c26267c6e209063e6e86abd26187
Reviewed-on: https://go-review.googlesource.com/c/go/+/288793
Trust: Russ Cox <rsc@golang.org>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jason A. Donenfeld <Jason@zx2c4.com>

src/runtime/asm_386.s

@@ -621,6 +621,22 @@ TEXT gosave<>(SB),NOSPLIT,$0
 	POPL	AX
 	RET
 
+// func asmcgocall_no_g(fn, arg unsafe.Pointer)
+// Call fn(arg) aligned appropriately for the gcc ABI.
+// Called on a system stack, and there may be no g yet (during needm).
+TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-8
+	MOVL	fn+0(FP), AX
+	MOVL	arg+4(FP), BX
+	MOVL	SP, DX
+	SUBL	$32, SP
+	ANDL	$~15, SP	// alignment, perhaps unnecessary
+	MOVL	DX, 8(SP)	// save old SP
+	MOVL	BX, 0(SP)	// first argument in x86-32 ABI
+	CALL	AX
+	MOVL	8(SP), DX
+	MOVL	DX, SP
+	RET
+
 // func asmcgocall(fn, arg unsafe.Pointer) int32
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.

src/runtime/asm_amd64.s

@@ -679,6 +679,23 @@ TEXT gosave<>(SB),NOSPLIT,$0
 	CALL	runtime·badctxt(SB)
 	RET
 
+// func asmcgocall_no_g(fn, arg unsafe.Pointer)
+// Call fn(arg) aligned appropriately for the gcc ABI.
+// Called on a system stack, and there may be no g yet (during needm).
+TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-16
+	MOVQ	fn+0(FP), AX
+	MOVQ	arg+8(FP), BX
+	MOVQ	SP, DX
+	SUBQ	$32, SP
+	ANDQ	$~15, SP	// alignment
+	MOVQ	DX, 8(SP)
+	MOVQ	BX, DI		// DI = first argument in AMD64 ABI
+	MOVQ	BX, CX		// CX = first argument in Win64
+	CALL	AX
+	MOVQ	8(SP), DX
+	MOVQ	DX, SP
+	RET
+
 // func asmcgocall(fn, arg unsafe.Pointer) int32
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.

src/runtime/asm_arm.s

@@ -552,6 +552,21 @@ TEXT gosave<>(SB),NOSPLIT|NOFRAME,$0
 	CALL	runtime·badctxt(SB)
 	RET
 
+// func asmcgocall_no_g(fn, arg unsafe.Pointer)
+// Call fn(arg) aligned appropriately for the gcc ABI.
+// Called on a system stack, and there may be no g yet (during needm).
+TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-8
+	MOVW	fn+0(FP), R1
+	MOVW	arg+4(FP), R0
+	MOVW	R13, R2
+	SUB	$32, R13
+	BIC	$0x7, R13	// alignment for gcc ABI
+	MOVW	R2, 8(R13)
+	BL	(R1)
+	MOVW	8(R13), R2
+	MOVW	R2, R13
+	RET
+
 // func asmcgocall(fn, arg unsafe.Pointer) int32
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.

src/runtime/asm_arm64.s

@@ -873,6 +873,17 @@ TEXT gosave<>(SB),NOSPLIT|NOFRAME,$0
 	CALL	runtime·badctxt(SB)
 	RET
 
+// func asmcgocall_no_g(fn, arg unsafe.Pointer)
+// Call fn(arg) aligned appropriately for the gcc ABI.
+// Called on a system stack, and there may be no g yet (during needm).
+TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-16
+	MOVD	fn+0(FP), R1
+	MOVD	arg+8(FP), R0
+	SUB	$16, RSP	// skip over saved frame pointer below RSP
+	BL	(R1)
+	ADD	$16, RSP	// skip over saved frame pointer below RSP
+	RET
+
 // func asmcgocall(fn, arg unsafe.Pointer) int32
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.

src/runtime/asm_mips64x.s

@@ -413,6 +413,15 @@ TEXT gosave<>(SB),NOSPLIT|NOFRAME,$0
 	JAL	runtime·badctxt(SB)
 	RET
 
+// func asmcgocall_no_g(fn, arg unsafe.Pointer)
+// Call fn(arg) aligned appropriately for the gcc ABI.
+// Called on a system stack, and there may be no g yet (during needm).
+TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-16
+	MOVV	fn+0(FP), R25
+	MOVV	arg+8(FP), R4
+	JAL	(R25)
+	RET
+
 // func asmcgocall(fn, arg unsafe.Pointer) int32
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.

src/runtime/os2_aix.go

@@ -527,22 +527,19 @@ func internal_cpu_getsystemcfg(label uint) uint {
 func usleep1(us uint32)
 
 //go:nosplit
-func usleep(us uint32) {
+func usleep_no_g(us uint32) {
-	_g_ := getg()
-
-	// Check the validity of m because we might be called in cgo callback
-	// path early enough where there isn't a g or a m available yet.
-	if _g_ != nil && _g_.m != nil {
-		r, err := syscall1(&libc_usleep, uintptr(us))
-		if int32(r) == -1 {
-			println("syscall usleep failed: ", hex(err))
-			throw("syscall usleep")
-		}
-		return
-	}
 	usleep1(us)
 }
 
+//go:nosplit
+func usleep(us uint32) {
+	r, err := syscall1(&libc_usleep, uintptr(us))
+	if int32(r) == -1 {
+		println("syscall usleep failed: ", hex(err))
+		throw("syscall usleep")
+	}
+}
+
 //go:nosplit
 func clock_gettime(clockid int32, tp *timespec) int32 {
 	r, _ := syscall2(&libc_clock_gettime, uintptr(clockid), uintptr(unsafe.Pointer(tp)))
@@ -611,22 +608,19 @@ func raiseproc(sig uint32) {
 func osyield1()
 
 //go:nosplit
-func osyield() {
+func osyield_no_g() {
-	_g_ := getg()
-
-	// Check the validity of m because it might be called during a cgo
-	// callback early enough where m isn't available yet.
-	if _g_ != nil && _g_.m != nil {
-		r, err := syscall0(&libc_sched_yield)
-		if int32(r) == -1 {
-			println("syscall osyield failed: ", hex(err))
-			throw("syscall osyield")
-		}
-		return
-	}
 	osyield1()
 }
 
+//go:nosplit
+func osyield() {
+	r, err := syscall0(&libc_sched_yield)
+	if int32(r) == -1 {
+		println("syscall osyield failed: ", hex(err))
+		throw("syscall osyield")
+	}
+}
+
 //go:nosplit
 func sysconf(name int32) uintptr {
 	r, _ := syscall1(&libc_sysconf, uintptr(name))

src/runtime/os3_solaris.go

@@ -521,6 +521,11 @@ func sysconf(name int32) int64 {
 
 func usleep1(usec uint32)
 
+//go:nosplit
+func usleep_no_g(µs uint32) {
+	usleep1(µs)
+}
+
 //go:nosplit
 func usleep(µs uint32) {
 	usleep1(µs)
@@ -569,18 +574,15 @@ func setNonblock(fd int32) {
 func osyield1()
 
 //go:nosplit
-func osyield() {
+func osyield_no_g() {
-	_g_ := getg()
-
-	// Check the validity of m because we might be called in cgo callback
-	// path early enough where there isn't a m available yet.
-	if _g_ != nil && _g_.m != nil {
-		sysvicall0(&libc_sched_yield)
-		return
-	}
 	osyield1()
 }
 
+//go:nosplit
+func osyield() {
+	sysvicall0(&libc_sched_yield)
+}
+
 //go:linkname executablePath os.executablePath
 var executablePath string
 

src/runtime/os_darwin.go

@@ -330,6 +330,11 @@ func unminit() {
 func mdestroy(mp *m) {
 }
 
+//go:nosplit
+func osyield_no_g() {
+	usleep_no_g(1)
+}
+
 //go:nosplit
 func osyield() {
 	usleep(1)

src/runtime/os_dragonfly.go

@@ -51,6 +51,11 @@ func sys_umtx_wakeup(addr *uint32, val int32) int32
 
 func osyield()
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 func kqueue() int32
 
 //go:noescape

src/runtime/os_freebsd.go

@@ -36,6 +36,11 @@ func sys_umtx_op(addr *uint32, mode int32, val uint32, uaddr1 uintptr, ut *umtx_
 
 func osyield()
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 func kqueue() int32
 
 //go:noescape

src/runtime/os_js.go

@@ -30,12 +30,22 @@ func wasmWrite(fd uintptr, p unsafe.Pointer, n int32)
 
 func usleep(usec uint32)
 
+//go:nosplit
+func usleep_no_g(usec uint32) {
+	usleep(usec)
+}
+
 func exitThread(wait *uint32)
 
 type mOS struct{}
 
 func osyield()
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 const _SIGSEGV = 0xb
 
 func sigpanic() {

src/runtime/os_linux.go

@@ -410,6 +410,11 @@ func raiseproc(sig uint32)
 func sched_getaffinity(pid, len uintptr, buf *byte) int32
 func osyield()
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 func pipe() (r, w int32, errno int32)
 func pipe2(flags int32) (r, w int32, errno int32)
 func setNonblock(fd int32)

src/runtime/os_netbsd.go

@@ -67,6 +67,11 @@ func lwp_self() int32
 
 func osyield()
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 func kqueue() int32
 
 //go:noescape

src/runtime/os_openbsd_syscall1.go

@@ -13,3 +13,8 @@ func thrsleep(ident uintptr, clock_id int32, tsp *timespec, lock uintptr, abort
 func thrwakeup(ident uintptr, n int32) int32
 
 func osyield()
+
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}

src/runtime/os_openbsd_syscall2.go

@@ -32,6 +32,11 @@ func closefd(fd int32) int32
 func exit(code int32)
 func usleep(usec uint32)
 
+//go:nosplit
+func usleep_no_g(usec uint32) {
+	usleep(usec)
+}
+
 // write calls the write system call.
 // It returns a non-negative number of bytes written or a negative errno value.
 //go:noescape

src/runtime/os_plan9.go

@@ -339,6 +339,11 @@ func osyield() {
 	sleep(0)
 }
 
+//go:nosplit
+func osyield_no_g() {
+	osyield()
+}
+
 //go:nosplit
 func usleep(µs uint32) {
 	ms := int32(µs / 1000)
@@ -348,6 +353,11 @@ func usleep(µs uint32) {
 	sleep(ms)
 }
 
+//go:nosplit
+func usleep_no_g(usec uint32) {
+	usleep(usec)
+}
+
 //go:nosplit
 func nanotime1() int64 {
 	var scratch int64

src/runtime/os_windows.go

@@ -461,15 +461,12 @@ func initHighResTimer() {
 	h := createHighResTimer()
 	if h != 0 {
 		haveHighResTimer = true
-		usleep2Addr = unsafe.Pointer(funcPC(usleep2HighRes))
 		stdcall1(_CloseHandle, h)
 	}
 }
 
 func osinit() {
 	asmstdcallAddr = unsafe.Pointer(funcPC(asmstdcall))
-	usleep2Addr = unsafe.Pointer(funcPC(usleep2))
-	switchtothreadAddr = unsafe.Pointer(funcPC(switchtothread))
 
 	setBadSignalMsg()
 
@@ -1061,26 +1058,39 @@ func stdcall7(fn stdFunction, a0, a1, a2, a3, a4, a5, a6 uintptr) uintptr {
 	return stdcall(fn)
 }
 
-// In sys_windows_386.s and sys_windows_amd64.s.
+// These must run on the system stack only.
-func onosstack(fn unsafe.Pointer, arg uint32)
+func usleep2(dt int32)
-
+func usleep2HighRes(dt int32)
-// These are not callable functions. They should only be called via onosstack.
-func usleep2(usec uint32)
-func usleep2HighRes(usec uint32)
 func switchtothread()
 
-var usleep2Addr unsafe.Pointer
+//go:nosplit
-var switchtothreadAddr unsafe.Pointer
+func osyield_no_g() {
+	switchtothread()
+}
 
 //go:nosplit
 func osyield() {
-	onosstack(switchtothreadAddr, 0)
+	systemstack(switchtothread)
+}
+
+//go:nosplit
+func usleep_no_g(us uint32) {
+	dt := -10 * int32(us) // relative sleep (negative), 100ns units
+	usleep2(dt)
 }
 
 //go:nosplit
 func usleep(us uint32) {
-	// Have 1us units; want 100ns units.
+	systemstack(func() {
-	onosstack(usleep2Addr, 10*us)
+		dt := -10 * int32(us) // relative sleep (negative), 100ns units
+		// If the high-res timer is available and its handle has been allocated for this m, use it.
+		// Otherwise fall back to the low-res one, which doesn't need a handle.
+		if haveHighResTimer && getg().m.highResTimer != 0 {
+			usleep2HighRes(dt)
+		} else {
+			usleep2(dt)
+		}
+	})
 }
 
 func ctrlhandler1(_type uint32) uint32 {

src/runtime/proc.go

@@ -2012,7 +2012,7 @@ func lockextra(nilokay bool) *m {
 	for {
 		old := atomic.Loaduintptr(&extram)
 		if old == locked {
-			osyield()
+			osyield_no_g()
 			continue
 		}
 		if old == 0 && !nilokay {
@@ -2023,13 +2023,13 @@ func lockextra(nilokay bool) *m {
 				atomic.Xadd(&extraMWaiters, 1)
 				incr = true
 			}
-			usleep(1)
+			usleep_no_g(1)
 			continue
 		}
 		if atomic.Casuintptr(&extram, old, locked) {
 			return (*m)(unsafe.Pointer(old))
 		}
-		osyield()
+		osyield_no_g()
 		continue
 	}
 }

src/runtime/stubs2.go

@@ -23,6 +23,11 @@ func closefd(fd int32) int32
 func exit(code int32)
 func usleep(usec uint32)
 
+//go:nosplit
+func usleep_no_g(usec uint32) {
+	usleep(usec)
+}
+
 // write calls the write system call.
 // It returns a non-negative number of bytes written or a negative errno value.
 //go:noescape

src/runtime/stubs_386.go

@@ -15,3 +15,6 @@ func stackcheck()
 // Called from assembly only; declared for go vet.
 func setldt(slot uintptr, base unsafe.Pointer, size uintptr)
 func emptyfunc()
+
+//go:noescape
+func asmcgocall_no_g(fn, arg unsafe.Pointer)

src/runtime/stubs_amd64.go

@@ -4,6 +4,8 @@
 
 package runtime
 
+import "unsafe"
+
 // Called from compiled code; declared for vet; do NOT call from Go.
 func gcWriteBarrierCX()
 func gcWriteBarrierDX()
@@ -35,3 +37,6 @@ func retpolineR12()
 func retpolineR13()
 func retpolineR14()
 func retpolineR15()
+
+//go:noescape
+func asmcgocall_no_g(fn, arg unsafe.Pointer)

src/runtime/stubs_arm.go

@@ -4,6 +4,8 @@
 
 package runtime
 
+import "unsafe"
+
 // Called from compiler-generated code; declared for go vet.
 func udiv()
 func _div()
@@ -18,3 +20,6 @@ func save_g()
 func emptyfunc()
 func _initcgo()
 func read_tls_fallback()
+
+//go:noescape
+func asmcgocall_no_g(fn, arg unsafe.Pointer)

src/runtime/stubs_arm64.go

@@ -4,6 +4,11 @@
 
 package runtime
 
+import "unsafe"
+
 // Called from assembly only; declared for go vet.
 func load_g()
 func save_g()
+
+//go:noescape
+func asmcgocall_no_g(fn, arg unsafe.Pointer)

src/runtime/stubs_mips64x.go

@@ -6,6 +6,11 @@
 
 package runtime
 
+import "unsafe"
+
 // Called from assembly only; declared for go vet.
 func load_g()
 func save_g()
+
+//go:noescape
+func asmcgocall_no_g(fn, arg unsafe.Pointer)

src/runtime/sys_darwin.go

@@ -227,6 +227,12 @@ func usleep(usec uint32) {
 }
 func usleep_trampoline()
 
+//go:nosplit
+//go:cgo_unsafe_args
+func usleep_no_g(usec uint32) {
+	asmcgocall_no_g(unsafe.Pointer(funcPC(usleep_trampoline)), unsafe.Pointer(&usec))
+}
+
 //go:nosplit
 //go:cgo_unsafe_args
 func write1(fd uintptr, p unsafe.Pointer, n int32) int32 {

src/runtime/sys_openbsd1.go

@@ -27,6 +27,11 @@ func osyield() {
 }
 func sched_yield_trampoline()
 
+//go:nosplit
+func osyield_no_g() {
+	asmcgocall_no_g(unsafe.Pointer(funcPC(sched_yield_trampoline)), unsafe.Pointer(nil))
+}
+
 //go:cgo_import_dynamic libc_thrsleep __thrsleep "libc.so"
 //go:cgo_import_dynamic libc_thrwakeup __thrwakeup "libc.so"
 //go:cgo_import_dynamic libc_sched_yield sched_yield "libc.so"

src/runtime/sys_openbsd2.go

@@ -128,6 +128,12 @@ func usleep(usec uint32) {
 }
 func usleep_trampoline()
 
+//go:nosplit
+//go:cgo_unsafe_args
+func usleep_no_g(usec uint32) {
+	asmcgocall_no_g(unsafe.Pointer(funcPC(usleep_trampoline)), unsafe.Pointer(&usec))
+}
+
 //go:nosplit
 //go:cgo_unsafe_args
 func sysctl(mib *uint32, miblen uint32, out *byte, size *uintptr, dst *byte, ndst uintptr) int32 {

src/runtime/sys_windows_386.s

@@ -347,60 +347,11 @@ TEXT runtime·setldt(SB),NOSPLIT,$0
 	MOVL	CX, 0x14(FS)
 	RET
 
-// onosstack calls fn on OS stack.
+// Runs on OS stack.
-// func onosstack(fn unsafe.Pointer, arg uint32)
+// duration (in -100ns units) is in dt+0(FP).
-TEXT runtime·onosstack(SB),NOSPLIT,$0
+// g may be nil.
-	MOVL	fn+0(FP), AX		// to hide from 8l
+TEXT runtime·usleep2(SB),NOSPLIT,$20-4
-	MOVL	arg+4(FP), BX
+	MOVL	dt+0(FP), BX
-
-	// Execute call on m->g0 stack, in case we are not actually
-	// calling a system call wrapper, like when running under WINE.
-	get_tls(CX)
-	CMPL	CX, $0
-	JNE	3(PC)
-	// Not a Go-managed thread. Do not switch stack.
-	CALL	AX
-	RET
-
-	MOVL	g(CX), BP
-	MOVL	g_m(BP), BP
-
-	// leave pc/sp for cpu profiler
-	MOVL	(SP), SI
-	MOVL	SI, m_libcallpc(BP)
-	MOVL	g(CX), SI
-	MOVL	SI, m_libcallg(BP)
-	// sp must be the last, because once async cpu profiler finds
-	// all three values to be non-zero, it will use them
-	LEAL	fn+0(FP), SI
-	MOVL	SI, m_libcallsp(BP)
-
-	MOVL	m_g0(BP), SI
-	CMPL	g(CX), SI
-	JNE	switch
-	// executing on m->g0 already
-	CALL	AX
-	JMP	ret
-
-switch:
-	// Switch to m->g0 stack and back.
-	MOVL	(g_sched+gobuf_sp)(SI), SI
-	MOVL	SP, -4(SI)
-	LEAL	-4(SI), SP
-	CALL	AX
-	MOVL	0(SP), SP
-
-ret:
-	get_tls(CX)
-	MOVL	g(CX), BP
-	MOVL	g_m(BP), BP
-	MOVL	$0, m_libcallsp(BP)
-	RET
-
-// Runs on OS stack. duration (in 100ns units) is in BX.
-TEXT runtime·usleep2(SB),NOSPLIT,$20
-	// Want negative 100ns units.
-	NEGL	BX
 	MOVL	$-1, hi-4(SP)
 	MOVL	BX, lo-8(SP)
 	LEAL	lo-8(SP), BX
@@ -413,17 +364,15 @@ TEXT runtime·usleep2(SB),NOSPLIT,$20
 	MOVL	BP, SP
 	RET
 
-// Runs on OS stack. duration (in 100ns units) is in BX.
+// Runs on OS stack.
-TEXT runtime·usleep2HighRes(SB),NOSPLIT,$36
+// duration (in -100ns units) is in dt+0(FP).
-	get_tls(CX)
+// g is valid.
-	CMPL	CX, $0
+TEXT runtime·usleep2HighRes(SB),NOSPLIT,$36-4
-	JE	gisnotset
+	MOVL	dt+0(FP), BX
-
-	// Want negative 100ns units.
-	NEGL	BX
 	MOVL	$-1, hi-4(SP)
 	MOVL	BX, lo-8(SP)
 
+	get_tls(CX)
 	MOVL	g(CX), CX
 	MOVL	g_m(CX), CX
 	MOVL	(m_mOS+mOS_highResTimer)(CX), CX
@@ -452,12 +401,6 @@ TEXT runtime·usleep2HighRes(SB),NOSPLIT,$36
 
 	RET
 
-gisnotset:
-	// TLS is not configured. Call usleep2 instead.
-	MOVL	$runtime·usleep2(SB), AX
-	CALL	AX
-	RET
-
 // Runs on OS stack.
 TEXT runtime·switchtothread(SB),NOSPLIT,$0
 	MOVL	SP, BP

src/runtime/sys_windows_amd64.s

@@ -388,61 +388,16 @@ TEXT runtime·settls(SB),NOSPLIT,$0
 	MOVQ	DI, 0x28(GS)
 	RET
 
-// func onosstack(fn unsafe.Pointer, arg uint32)
+// Runs on OS stack.
-TEXT runtime·onosstack(SB),NOSPLIT,$0
+// duration (in -100ns units) is in dt+0(FP).
-	MOVQ	fn+0(FP), AX		// to hide from 6l
+// g may be nil.
-	MOVL	arg+8(FP), BX
-
-	// Execute call on m->g0 stack, in case we are not actually
-	// calling a system call wrapper, like when running under WINE.
-	get_tls(R15)
-	CMPQ	R15, $0
-	JNE	3(PC)
-	// Not a Go-managed thread. Do not switch stack.
-	CALL	AX
-	RET
-
-	MOVQ	g(R15), R13
-	MOVQ	g_m(R13), R13
-
-	// leave pc/sp for cpu profiler
-	MOVQ	(SP), R12
-	MOVQ	R12, m_libcallpc(R13)
-	MOVQ	g(R15), R12
-	MOVQ	R12, m_libcallg(R13)
-	// sp must be the last, because once async cpu profiler finds
-	// all three values to be non-zero, it will use them
-	LEAQ	fn+0(FP), R12
-	MOVQ	R12, m_libcallsp(R13)
-
-	MOVQ	m_g0(R13), R14
-	CMPQ	g(R15), R14
-	JNE	switch
-	// executing on m->g0 already
-	CALL	AX
-	JMP	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:
-	MOVQ	$0, m_libcallsp(R13)
-	RET
-
-// Runs on OS stack. duration (in 100ns units) is in BX.
 // The function leaves room for 4 syscall parameters
 // (as per windows amd64 calling convention).
-TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$48
+TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$48-4
+	MOVLQSX	dt+0(FP), BX
 	MOVQ	SP, AX
 	ANDQ	$~15, SP	// alignment as per Windows requirement
 	MOVQ	AX, 40(SP)
-	// Want negative 100ns units.
-	NEGQ	BX
 	LEAQ	32(SP), R8  // ptime
 	MOVQ	BX, (R8)
 	MOVQ	$-1, CX // handle
@@ -452,11 +407,11 @@ TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$48
 	MOVQ	40(SP), SP
 	RET
 
-// Runs on OS stack. duration (in 100ns units) is in BX.
+// Runs on OS stack. duration (in -100ns units) is in dt+0(FP).
-TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72
+// g is valid.
+TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72-4
+	MOVLQSX	dt+0(FP), BX
 	get_tls(CX)
-	CMPQ	CX, $0
-	JE	gisnotset
 
 	MOVQ	SP, AX
 	ANDQ	$~15, SP	// alignment as per Windows requirement
@@ -466,8 +421,6 @@ TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72
 	MOVQ	g_m(CX), CX
 	MOVQ	(m_mOS+mOS_highResTimer)(CX), CX	// hTimer
 	MOVQ	CX, 48(SP)				// save hTimer for later
-	// Want negative 100ns units.
-	NEGQ	BX
 	LEAQ	56(SP), DX				// lpDueTime
 	MOVQ	BX, (DX)
 	MOVQ	$0, R8					// lPeriod
@@ -487,12 +440,6 @@ TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72
 	MOVQ	64(SP), SP
 	RET
 
-gisnotset:
-	// TLS is not configured. Call usleep2 instead.
-	MOVQ	$runtime·usleep2(SB), AX
-	CALL	AX
-	RET
-
 // Runs on OS stack.
 TEXT runtime·switchtothread(SB),NOSPLIT|NOFRAME,$0
 	MOVQ	SP, AX

src/runtime/sys_windows_arm.s

@@ -377,79 +377,11 @@ TEXT runtime·tstart_stdcall(SB),NOSPLIT|NOFRAME,$0
 	MOVW	$0, R0
 	MOVM.IA.W (R13), [R4-R11, R15]		// pop {r4-r11, pc}
 
-// onosstack calls fn on OS stack.
+// Runs on OS stack.
-// adapted from asm_arm.s : systemstack
+// duration (in -100ns units) is in dt+0(FP).
-// func onosstack(fn unsafe.Pointer, arg uint32)
+// g may be nil.
-TEXT runtime·onosstack(SB),NOSPLIT,$0
+TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$0-4
-	MOVW	fn+0(FP), R5		// R5 = fn
+	MOVW	dt+0(FP), R0
-	MOVW	arg+4(FP), R6		// R6 = arg
-
-	// This function can be called when there is no g,
-	// for example, when we are handling a callback on a non-go thread.
-	// In this case we're already on the system stack.
-	CMP	$0, g
-	BEQ	noswitch
-
-	MOVW	g_m(g), R1		// R1 = m
-
-	MOVW	m_gsignal(R1), R2	// R2 = gsignal
-	CMP	g, R2
-	B.EQ	noswitch
-
-	MOVW	m_g0(R1), R2		// R2 = g0
-	CMP	g, R2
-	B.EQ	noswitch
-
-	MOVW	m_curg(R1), R3
-	CMP	g, R3
-	B.EQ	switch
-
-	// Bad: g is not gsignal, not g0, not curg. What is it?
-	// Hide call from linker nosplit analysis.
-	MOVW	$runtime·badsystemstack(SB), R0
-	BL	(R0)
-	B	runtime·abort(SB)
-
-switch:
-	// save our state in g->sched. Pretend to
-	// be systemstack_switch if the G stack is scanned.
-	MOVW	$runtime·systemstack_switch(SB), R3
-	ADD	$4, R3, R3 // get past push {lr}
-	MOVW	R3, (g_sched+gobuf_pc)(g)
-	MOVW	R13, (g_sched+gobuf_sp)(g)
-	MOVW	LR, (g_sched+gobuf_lr)(g)
-	MOVW	g, (g_sched+gobuf_g)(g)
-
-	// switch to g0
-	MOVW	R2, g
-	MOVW	(g_sched+gobuf_sp)(R2), R3
-	// make it look like mstart called systemstack on g0, to stop traceback
-	SUB	$4, R3, R3
-	MOVW	$runtime·mstart(SB), R4
-	MOVW	R4, 0(R3)
-	MOVW	R3, R13
-
-	// call target function
-	MOVW	R6, R0		// arg
-	BL	(R5)
-
-	// switch back to g
-	MOVW	g_m(g), R1
-	MOVW	m_curg(R1), g
-	MOVW	(g_sched+gobuf_sp)(g), R13
-	MOVW	$0, R3
-	MOVW	R3, (g_sched+gobuf_sp)(g)
-	RET
-
-noswitch:
-	// Using a tail call here cleans up tracebacks since we won't stop
-	// at an intermediate systemstack.
-	MOVW.P	4(R13), R14	// restore LR
-	MOVW	R6, R0		// arg
-	B	(R5)
-
-// Runs on OS stack. Duration (in 100ns units) is in R0.
-TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$0
 	MOVM.DB.W [R4, R14], (R13)	// push {r4, lr}
 	MOVW	R13, R4			// Save SP
 	SUB	$8, R13			// R13 = R13 - 8
@@ -465,9 +397,11 @@ TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$0
 	MOVW	R4, R13			// Restore SP
 	MOVM.IA.W (R13), [R4, R15]	// pop {R4, pc}
 
-// Runs on OS stack. Duration (in 100ns units) is in R0.
+// Runs on OS stack.
+// duration (in -100ns units) is in dt+0(FP).
+// g is valid.
 // TODO: neeeds to be implemented properly.
-TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$0-4
 	B	runtime·abort(SB)
 
 // Runs on OS stack.