runtime: use libc for nanotime on Darwin

Use mach_absolute_time and mach_timebase_info to get nanosecond-level
timing information from libc on Darwin.

The conversion code from Apple's arbitrary time unit to nanoseconds is
really annoying.  It would be nice if we could replace the internal
runtime "time" with arbitrary units and put the conversion to nanoseconds
only in the places that really need it (so it isn't in every nanotime call).

It's especially annoying because numer==denom==1 for all the machines
I tried.  Makes it hard to test the conversion code :(

Update #17490

Change-Id: I6c5d602a802f5c24e35184e33d5e8194aa7afa86
Reviewed-on: https://go-review.googlesource.com/110655
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/runtime/asm_386.s

@@ -704,6 +704,8 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	MOVL	g(CX), DI
 	CMPL	SI, DI
 	JEQ	noswitch
+	CMPL	DI, m_gsignal(BP)
+	JEQ	noswitch
 	CALL	gosave<>(SB)
 	get_tls(CX)
 	MOVL	SI, g(CX)

src/runtime/defs_darwin.go

@@ -17,6 +17,7 @@ package runtime
 #define __DARWIN_UNIX03 0
 #include <mach/mach.h>
 #include <mach/message.h>
+#include <mach/mach_time.h>
 #include <sys/types.h>
 #include <sys/time.h>
 #include <errno.h>
@@ -184,3 +185,5 @@ type Kevent C.struct_kevent
 
 type Pthread C.pthread_t
 type PthreadAttr C.pthread_attr_t
+
+type MachTimebaseInfo C.mach_timebase_info_data_t

src/runtime/defs_darwin_386.go

@@ -395,3 +395,7 @@ type pthreadattr struct {
 	X__sig    int32
 	X__opaque [36]int8
 }
+type machTimebaseInfo struct {
+	numer uint32
+	denom uint32
+}

src/runtime/defs_darwin_amd64.go

@@ -398,3 +398,7 @@ type pthreadattr struct {
 	X__sig    int64
 	X__opaque [56]int8
 }
+type machTimebaseInfo struct {
+	numer uint32
+	denom uint32
+}

src/runtime/stubs3.go

@@ -7,6 +7,7 @@
 // +build !windows
 // +build !nacl
 // +build !freebsd
+// +build !darwin
 
 package runtime
 

src/runtime/stubs4.go

@@ -0,0 +1,9 @@
+// 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

@@ -124,6 +124,28 @@ func open(name *byte, mode, perm int32) (ret int32) {
 }
 func open_trampoline()
 
+//go:nosplit
+//go:cgo_unsafe_args
+func nanotime() int64 {
+	var r struct {
+		t            int64  // raw timer
+		numer, denom uint32 // conversion factors. nanoseconds = t * numer / denom.
+	}
+	asmcgocall(unsafe.Pointer(funcPC(nanotime_trampoline)), unsafe.Pointer(&r))
+	// Note: Apple seems unconcerned about overflow here. See
+	// https://developer.apple.com/library/content/qa/qa1398/_index.html
+	// Note also, numer == denom == 1 is common.
+	t := r.t
+	if r.numer != 1 {
+		t *= int64(r.numer)
+	}
+	if r.denom != 1 {
+		t /= int64(r.denom)
+	}
+	return t
+}
+func nanotime_trampoline()
+
 // Not used on Darwin, but must be defined.
 func exitThread(wait *uint32) {
 }
@@ -150,6 +172,9 @@ func exitThread(wait *uint32) {
 //go:cgo_import_dynamic libc_error __error "/usr/lib/libSystem.B.dylib"
 //go:cgo_import_dynamic libc_usleep usleep "/usr/lib/libSystem.B.dylib"
 
+//go:cgo_import_dynamic libc_mach_timebase_info mach_timebase_info "/usr/lib/libSystem.B.dylib"
+//go:cgo_import_dynamic libc_mach_absolute_time mach_absolute_time "/usr/lib/libSystem.B.dylib"
+
 // Magic incantation to get libSystem actually dynamically linked.
 // TODO: Why does the code require this?  See cmd/compile/internal/ld/go.go:210
 //go:cgo_import_dynamic _ _ "/usr/lib/libSystem.B.dylib"

src/runtime/sys_darwin_386.s

@@ -311,13 +311,37 @@ TEXT time·now(SB),NOSPLIT,$0-20
 	MOVL	DX, nsec+8(FP)
 	RET
 
-// func nanotime() int64
-TEXT runtime·nanotime(SB),NOSPLIT,$0
-	CALL	runtime·now(SB)
-	SUBL	runtime·startNano(SB), AX
-	SBBL	runtime·startNano+4(SB), DX
-	MOVL	AX, ret_lo+0(FP)
-	MOVL	DX, ret_hi+4(FP)
+GLOBL timebase<>(SB),NOPTR,$(machTimebaseInfo__size)
+
+TEXT runtime·nanotime_trampoline(SB),NOSPLIT,$0
+	PUSHL	BP
+	MOVL	SP, BP
+	SUBL	$8+(machTimebaseInfo__size+15)/16*16, SP
+	CALL	libc_mach_absolute_time(SB)
+	MOVL	16+(machTimebaseInfo__size+15)/16*16(SP), CX
+	MOVL	AX, 0(CX)
+	MOVL	DX, 4(CX)
+	MOVL	timebase<>+machTimebaseInfo_denom(SB), DI // atomic read
+	MOVL	timebase<>+machTimebaseInfo_numer(SB), SI
+	TESTL	DI, DI
+	JNE	initialized
+
+	LEAL	4(SP), AX
+	MOVL	AX, 0(SP)
+	CALL	libc_mach_timebase_info(SB)
+	MOVL	4+machTimebaseInfo_numer(SP), SI
+	MOVL	4+machTimebaseInfo_denom(SP), DI
+
+	MOVL	SI, timebase<>+machTimebaseInfo_numer(SB)
+	MOVL	DI, AX
+	XCHGL	AX, timebase<>+machTimebaseInfo_denom(SB) // atomic write
+	MOVL	16+(machTimebaseInfo__size+15)/16*16(SP), CX
+
+initialized:
+	MOVL	SI, 8(CX)
+	MOVL	DI, 12(CX)
+	MOVL	BP, SP
+	POPL	BP
 	RET
 
 TEXT runtime·sigprocmask(SB),NOSPLIT,$0

src/runtime/sys_darwin_amd64.s

@@ -110,33 +110,35 @@ TEXT runtime·madvise_trampoline(SB), NOSPLIT, $0
 #define	v17_gtod_scale		0xe8
 #define	v17_gtod_tkspersec	0xf0
 
-TEXT runtime·nanotime(SB),NOSPLIT,$0-8
-	MOVQ	$0x7fffffe00000, BP	/* comm page base */
-	// Loop trying to take a consistent snapshot
-	// of the time parameters.
-timeloop:
-	MOVL	nt_generation(BP), R9
-	TESTL	R9, R9
-	JZ	timeloop
-	RDTSC
-	MOVQ	nt_tsc_base(BP), R10
-	MOVL	nt_scale(BP), R11
-	MOVQ	nt_ns_base(BP), R12
-	CMPL	nt_generation(BP), R9
-	JNE	timeloop
+GLOBL timebase<>(SB),NOPTR,$(machTimebaseInfo__size)
 
-	// Gathered all the data we need. Compute monotonic time:
-	//	((tsc - nt_tsc_base) * nt_scale) >> 32 + nt_ns_base
-	// The multiply and shift extracts the top 64 bits of the 96-bit product.
-	SHLQ	$32, DX
-	ADDQ	DX, AX
-	SUBQ	R10, AX
-	MULQ	R11
-	SHRQ	$32, AX:DX
-	ADDQ	R12, AX
-	MOVQ	runtime·startNano(SB), CX
-	SUBQ	CX, AX
-	MOVQ	AX, ret+0(FP)
+TEXT runtime·nanotime_trampoline(SB),NOSPLIT,$0
+	PUSHQ	BP
+	MOVQ	SP, BP
+	MOVQ	DI, BX
+	CALL	libc_mach_absolute_time(SB)
+	MOVQ	AX, 0(BX)
+	MOVL	timebase<>+machTimebaseInfo_numer(SB), SI
+	MOVL	timebase<>+machTimebaseInfo_denom(SB), DI // atomic read
+	TESTL	DI, DI
+	JNE	initialized
+
+	SUBQ	$(machTimebaseInfo__size+15)/16*16, SP
+	MOVQ	SP, DI
+	CALL	libc_mach_timebase_info(SB)
+	MOVL	machTimebaseInfo_numer(SP), SI
+	MOVL	machTimebaseInfo_denom(SP), DI
+	ADDQ	$(machTimebaseInfo__size+15)/16*16, SP
+
+	MOVL	SI, timebase<>+machTimebaseInfo_numer(SB)
+	MOVL	DI, AX
+	XCHGL	AX, timebase<>+machTimebaseInfo_denom(SB) // atomic write
+
+initialized:
+	MOVL	SI, 8(BX)
+	MOVL	DI, 12(BX)
+	MOVQ	BP, SP
+	POPQ	BP
 	RET
 
 TEXT time·now(SB), NOSPLIT, $32-24