mach_absolute_time.s

// Source from the xnu kernel used in macOS
// source https://opensource.apple.com/source/xnu/xnu-3248.60.10/libsyscall/wrappers/mach_absolute_time.s
.globl	_mach_absolute_time
_mach_absolute_time:
	pushq	%rbp				// set up a frame for backtraces
	movq	%rsp,%rbp
	movq	$(_COMM_PAGE_TIME_DATA_START),%rsi
1:
	movl	_NT_GENERATION(%rsi),%r8d	// get generation
	testl	%r8d,%r8d			// if 0, data is being changed...
	jz      1b				// ...so loop until stable
	lfence
	rdtsc					// edx:eax := tsc
	lfence
	shlq	$32,%rdx			// rax := ((edx << 32) | eax), ie 64-bit tsc
	orq     %rdx,%rax

    /*
     * Prior to supporting "slow" processors, xnu always set _NT_SHIFT to 32.
     * Now it defaults to 0, unless the processor is slow.  In order to maintain
     * compatibility with both old and new versions of xnu, we mask the shift
     * down to 0x1F, which maps the old default (32) into the new default (0).
     */
	movl    _NT_SHIFT(%rsi),%ecx
	andl    $0x1F,%ecx			// *** remove this line once 10.9 is GM ***
	subq	_NT_TSC_BASE(%rsi), %rax	// rax := (tsc - base_tsc)
	shlq    %cl,%rax			// rax := (tsc - base_tsc) << NT_SHIFT
	movl	_NT_SCALE(%rsi),%ecx
	mulq	%rcx				// rdx:rax := ((tsc - base_tsc)<<shift) * scale
	shrdq	$32,%rdx,%rax			// divide by 2**32
	addq	_NT_NS_BASE(%rsi),%rax		// (((tsc - base_tsc) * scale) >> 32) + ns_base
	
	cmpl	_NT_GENERATION(%rsi),%r8d	// did the data change during computation?
	jne     1b
	popq	%rbp
	ret

time.c

#include <stdio.h>
#include <mach/mach_time.h>

int main()
{
	uint64_t t = mach_absolute_time();
	printf("%lld\n", (long long) t);
	return (0);
}

time_amd64-v2.s

TEXT ·mach_absolute_tim(SB), NOSPLIT, $32
	MOVQ $0x7fffffe00000, SI // comm page base

timeloop:
	MOVL  nt_generation(SI), R8
	TESTL R8, R8
	JZ    timeloop
	RDTSC
	SHLQ  $32, DX
	ORQ   DX, AX
	MOVL nt_shift(SI), CX
	SUBQ nt_tsc_base(SI), AX
	SHLQ CX, AX
	MOVL nt_scale(SI), CX
	MULQ CX
	SHRQ $32, AX:DX
	ADDQ nt_ns_base(SI), AX
	CMPL nt_generation(SI), R8
	JNE  timeloop
	MOVQ AX, ret+0(FP)
	RET

time_amd64.s

// Implementation of mach_absolute_time in Go AST
// based on https://golang.org/src/runtime/sys_darwin_amd64.s

#include "textflag.h"

// OS X comm page time offsets
// http://www.opensource.apple.com/source/xnu/xnu-1699.26.8/osfmk/i386/cpu_capabilities.h
#define	nt_tsc_base	0x50
#define	nt_scale	0x58
#define	nt_shift	0x5c
#define	nt_ns_base	0x60
#define	nt_generation	0x68

TEXT ·mach_absolute_time(SB),NOSPLIT,$0-8
	MOVQ	$0x7fffffe00000, BP	/* comm page base */
timeloop:
	MOVQ	nt_generation(BP), R8
	TESTL	R8, R8
	JZ	timeloop
	RDTSC
	SHLQ	$32, DX
	ORQ	DX, AX
	MOVL	nt_shift(BP), R9
	ANDL	$0x1F, R9
	SUBQ	nt_tsc_base(BP), AX
	SHLQ	CX, AX
	MOVL	nt_scale(BP), R9
	MULQ	R10
	SHRQ	$32, AX:DX
	ADDQ	nt_ns_base(BP), AX
	CMPL	nt_generation(BP), R8
	JNE	timeloop
	RET