runtime: fast clock_gettime call on FreeBSD

Use AT_TIMEKEEP ELF aux entry to access a kernel mapped ring of timehands structs.
The timehands are updated by the kernel periodically, but for accurate measure the
timecounter still needs to be queried.
Currently the fast path is used only when kern.timecounter.hardware==TSC-low
or kern.timecounter.hardware=='ARM MPCore Timecounter',
other timecounters revert back to regular system call.

TODO: add support for HPET timecounter on 386/amd64.

Change-Id: I321ca4e92be63ba21a2574b758ef5c1e729086ad
Reviewed-on: https://go-review.googlesource.com/93156
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/runtime/defs_freebsd.go

@@ -31,6 +31,7 @@ package runtime
 #include <sys/sysctl.h>
 #include <sys/cpuset.h>
 #include <sys/param.h>
+#include <sys/vdso.h>
 */
 import "C"
 
@@ -62,6 +63,7 @@ const (
 	SA_ONSTACK = C.SA_ONSTACK
 
 	CLOCK_MONOTONIC = C.CLOCK_MONOTONIC
+	CLOCK_REALTIME  = C.CLOCK_REALTIME
 
 	UMTX_OP_WAIT_UINT         = C.UMTX_OP_WAIT_UINT
 	UMTX_OP_WAIT_UINT_PRIVATE = C.UMTX_OP_WAIT_UINT_PRIVATE
@@ -147,3 +149,14 @@ type Itimerval C.struct_itimerval
 type Umtx_time C.struct__umtx_time
 
 type Kevent C.struct_kevent
+
+type bintime C.struct_bintime
+type vdsoTimehands C.struct_vdso_timehands
+type vdsoTimekeep C.struct_vdso_timekeep
+
+const (
+	_VDSO_TK_VER_CURR = C.VDSO_TK_VER_CURR
+
+	vdsoTimehandsSize = C.sizeof_struct_vdso_timehands
+	vdsoTimekeepSize  = C.sizeof_struct_vdso_timekeep
+)

src/runtime/defs_freebsd_386.go

@@ -32,6 +32,7 @@ const (
 	_SA_ONSTACK = 0x1
 
 	_CLOCK_MONOTONIC = 0x4
+	_CLOCK_REALTIME  = 0x0
 
 	_UMTX_OP_WAIT_UINT         = 0xb
 	_UMTX_OP_WAIT_UINT_PRIVATE = 0xf
@@ -221,3 +222,34 @@ type keventt struct {
 	data   int32
 	udata  *byte
 }
+
+type bintime struct {
+	sec  int32
+	frac uint64
+}
+
+type vdsoTimehands struct {
+	algo         uint32
+	gen          uint32
+	scale        uint64
+	offset_count uint32
+	counter_mask uint32
+	offset       bintime
+	boottime     bintime
+	x86_shift    uint32
+	x86_hpet_idx uint32
+	res          [6]uint32
+}
+
+type vdsoTimekeep struct {
+	ver     uint32
+	enabled uint32
+	current uint32
+}
+
+const (
+	_VDSO_TK_VER_CURR = 0x1
+
+	vdsoTimehandsSize = 0x50
+	vdsoTimekeepSize  = 0xc
+)

src/runtime/defs_freebsd_amd64.go

@@ -32,6 +32,7 @@ const (
 	_SA_ONSTACK = 0x1
 
 	_CLOCK_MONOTONIC = 0x4
+	_CLOCK_REALTIME  = 0x0
 
 	_UMTX_OP_WAIT_UINT         = 0xb
 	_UMTX_OP_WAIT_UINT_PRIVATE = 0xf
@@ -231,3 +232,35 @@ type keventt struct {
 	data   int64
 	udata  *byte
 }
+
+type bintime struct {
+	sec  int64
+	frac uint64
+}
+
+type vdsoTimehands struct {
+	algo         uint32
+	gen          uint32
+	scale        uint64
+	offset_count uint32
+	counter_mask uint32
+	offset       bintime
+	boottime     bintime
+	x86_shift    uint32
+	x86_hpet_idx uint32
+	res          [6]uint32
+}
+
+type vdsoTimekeep struct {
+	ver       uint32
+	enabled   uint32
+	current   uint32
+	pad_cgo_0 [4]byte
+}
+
+const (
+	_VDSO_TK_VER_CURR = 0x1
+
+	vdsoTimehandsSize = 0x58
+	vdsoTimekeepSize  = 0x10
+)

src/runtime/defs_freebsd_arm.go

@@ -32,6 +32,7 @@ const (
 	_SA_ONSTACK = 0x1
 
 	_CLOCK_MONOTONIC = 0x4
+	_CLOCK_REALTIME  = 0x0
 
 	_UMTX_OP_WAIT_UINT         = 0xb
 	_UMTX_OP_WAIT_UINT_PRIVATE = 0xf
@@ -194,3 +195,34 @@ type keventt struct {
 	data   int32
 	udata  *byte
 }
+
+type bintime struct {
+	sec  int64
+	frac uint64
+}
+
+type vdsoTimehands struct {
+	algo         uint32
+	gen          uint32
+	scale        uint64
+	offset_count uint32
+	counter_mask uint32
+	offset       bintime
+	boottime     bintime
+	physical     uint32
+	res          [7]uint32
+}
+
+type vdsoTimekeep struct {
+	ver       uint32
+	enabled   uint32
+	current   uint32
+	pad_cgo_0 [4]byte
+}
+
+const (
+	_VDSO_TK_VER_CURR = 0x1
+
+	vdsoTimehandsSize = 0x58
+	vdsoTimekeepSize  = 0x10
+)

src/runtime/os_freebsd.go

@@ -380,6 +380,7 @@ func sysargs(argc int32, argv **byte) {
 const (
 	_AT_NULL     = 0  // Terminates the vector
 	_AT_PAGESZ   = 6  // Page size in bytes
+	_AT_TIMEKEEP = 22 // Pointer to timehands.
 	_AT_HWCAP    = 26 // CPU feature flags
 )
 
@@ -390,6 +391,8 @@ func sysauxv(auxv []uintptr) {
 		// _AT_NCPUS from auxv shouldn't be used due to golang.org/issue/15206
 		case _AT_PAGESZ:
 			physPageSize = val
+		case _AT_TIMEKEEP:
+			timekeepSharedPage = (*vdsoTimekeep)(unsafe.Pointer(val))
 		}
 
 		archauxv(tag, val)

src/runtime/stubs2.go

@@ -15,7 +15,6 @@ func read(fd int32, p unsafe.Pointer, n int32) int32
 func closefd(fd int32) int32
 
 func exit(code int32)
-func nanotime() int64
 func usleep(usec uint32)
 
 //go:noescape

src/runtime/stubs3.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 !plan9
+// +build !solaris
+// +build !windows
+// +build !nacl
+// +build !freebsd
+
+package runtime
+
+func nanotime() int64

src/runtime/sys_freebsd_386.s

@@ -171,8 +171,8 @@ TEXT runtime·setitimer(SB), NOSPLIT, $-4
 	INT	$0x80
 	RET
 
-// func walltime() (sec int64, nsec int32)
-TEXT runtime·walltime(SB), NOSPLIT, $32
+// func fallback_walltime() (sec int64, nsec int32)
+TEXT runtime·fallback_walltime(SB), NOSPLIT, $32-12
 	MOVL	$232, AX // clock_gettime
 	LEAL	12(SP), BX
 	MOVL	$0, 4(SP)	// CLOCK_REALTIME
@@ -187,13 +187,10 @@ TEXT runtime·walltime(SB), NOSPLIT, $32
 	MOVL	BX, nsec+8(FP)
 	RET
 
-// int64 nanotime(void) so really
-// void nanotime(int64 *nsec)
-TEXT runtime·nanotime(SB), NOSPLIT, $32
+// func fallback_nanotime() int64
+TEXT runtime·fallback_nanotime(SB), NOSPLIT, $32-8
 	MOVL	$232, AX
 	LEAL	12(SP), BX
-	// We can use CLOCK_MONOTONIC_FAST here when we drop
-	// support for FreeBSD 8-STABLE.
 	MOVL	$4, 4(SP)	// CLOCK_MONOTONIC
 	MOVL	BX, 8(SP)
 	INT	$0x80

src/runtime/sys_freebsd_amd64.s

@@ -139,8 +139,8 @@ TEXT runtime·setitimer(SB), NOSPLIT, $-8
 	SYSCALL
 	RET
 
-// func walltime() (sec int64, nsec int32)
-TEXT runtime·walltime(SB), NOSPLIT, $32
+// func fallback_walltime() (sec int64, nsec int32)
+TEXT runtime·fallback_walltime(SB), NOSPLIT, $32-12
 	MOVL	$232, AX	// clock_gettime
 	MOVQ	$0, DI		// CLOCK_REALTIME
 	LEAQ	8(SP), SI
@@ -153,10 +153,8 @@ TEXT runtime·walltime(SB), NOSPLIT, $32
 	MOVL	DX, nsec+8(FP)
 	RET
 
-TEXT runtime·nanotime(SB), NOSPLIT, $32
+TEXT runtime·fallback_nanotime(SB), NOSPLIT, $32-8
 	MOVL	$232, AX
-	// We can use CLOCK_MONOTONIC_FAST here when we drop
-	// support for FreeBSD 8-STABLE.
 	MOVQ	$4, DI		// CLOCK_MONOTONIC
 	LEAQ	8(SP), SI
 	SYSCALL

src/runtime/sys_freebsd_arm.s

@@ -168,8 +168,8 @@ TEXT runtime·setitimer(SB), NOSPLIT|NOFRAME, $0
 	SWI $0
 	RET
 
-// func walltime() (sec int64, nsec int32)
-TEXT runtime·walltime(SB), NOSPLIT, $32
+// func fallback_walltime() (sec int64, nsec int32)
+TEXT runtime·fallback_walltime(SB), NOSPLIT, $32-12
 	MOVW $0, R0 // CLOCK_REALTIME
 	MOVW $8(R13), R1
 	MOVW $SYS_clock_gettime, R7
@@ -184,11 +184,8 @@ TEXT runtime·walltime(SB), NOSPLIT, $32
 	MOVW R2, nsec+8(FP)
 	RET
 
-// int64 nanotime(void) so really
-// void nanotime(int64 *nsec)
-TEXT runtime·nanotime(SB), NOSPLIT, $32
-	// We can use CLOCK_MONOTONIC_FAST here when we drop
-	// support for FreeBSD 8-STABLE.
+// func fallback_nanotime() int64
+TEXT runtime·fallback_nanotime(SB), NOSPLIT, $32
 	MOVW $4, R0 // CLOCK_MONOTONIC
 	MOVW $8(R13), R1
 	MOVW $SYS_clock_gettime, R7
@@ -395,3 +392,26 @@ TEXT runtime·cpuset_getaffinity(SB), NOSPLIT, $0-28
 	SUB	$20, R13
 	MOVW	R0, ret+24(FP)
 	RET
+
+// func getCntxct(physical bool) uint32
+TEXT runtime·getCntxct(SB),NOSPLIT|NOFRAME,$0-8
+	MOVB	runtime·goarm(SB), R11
+	CMP	$7, R11
+	BLT	2(PC)
+	DMB
+
+	MOVB	physical+0(FP), R0
+	CMP	$1, R0
+	B.NE	3(PC)
+
+	// get CNTPCT (Physical Count Register) into R0(low) R1(high)
+	// mrrc    15, 0, r0, r1, cr14
+	WORD	$0xec510f0e
+	B	2(PC)
+
+	// get CNTVCT (Virtual Count Register) into R0(low) R1(high)
+	// mrrc    15, 1, r0, r1, cr14
+	WORD	$0xec510f1e
+
+	MOVW	R0, ret+4(FP)
+	RET

src/runtime/timestub.go

@@ -12,8 +12,6 @@ package runtime
 
 import _ "unsafe" // for go:linkname
 
-func walltime() (sec int64, nsec int32)
-
 //go:linkname time_now time.now
 func time_now() (sec int64, nsec int32, mono int64) {
 	sec, nsec = walltime()

src/runtime/timestub2.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 !amd64,!386
+// +build !windows
+// +build !freebsd
+
+package runtime
+
+func walltime() (sec int64, nsec int32)

src/runtime/vdso_freebsd.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 freebsd
+
+package runtime
+
+import (
+	"runtime/internal/atomic"
+	"unsafe"
+)
+
+const _VDSO_TH_NUM = 4 // defined in <sys/vdso.h> #ifdef _KERNEL
+
+var timekeepSharedPage *vdsoTimekeep
+
+//go:nosplit
+func (bt bintime) Add(bt2 bintime) bintime {
+	u := bt.frac
+	bt.frac += bt2.frac
+	if u > bt.frac {
+		bt.sec++
+	}
+	bt.sec += bt2.sec
+	return bt
+}
+
+//go:nosplit
+func (bt bintime) AddX(x uint64) bintime {
+	u := bt.frac
+	bt.frac += x
+	if u > bt.frac {
+		bt.sec++
+	}
+	return bt
+}
+
+var binuptimeDummy uint32
+
+// based on /usr/src/lib/libc/sys/__vdso_gettimeofday.c
+//
+//go:nosplit
+func binuptime(abs bool) (bintime, bool) {
+	var bt bintime
+	timehands := (*[_VDSO_TH_NUM]vdsoTimehands)(add(unsafe.Pointer(timekeepSharedPage), vdsoTimekeepSize))
+	for {
+		if timekeepSharedPage.enabled == 0 {
+			return bt, false
+		}
+
+		curr := atomic.Load(&timekeepSharedPage.current) // atomic_load_acq_32
+		th := &timehands[curr]
+		gen := atomic.Load(&th.gen) // atomic_load_acq_32
+		bt = th.offset
+
+		if tc, ok := th.getTimecounter(); !ok {
+			return bt, false
+		} else {
+			delta := (tc - th.offset_count) & th.counter_mask
+			bt = bt.AddX(th.scale * uint64(delta))
+		}
+		if abs {
+			bt = bt.Add(th.boottime)
+		}
+
+		atomic.Load(&binuptimeDummy) // atomic_thread_fence_acq()
+		if curr == timekeepSharedPage.current && gen != 0 && gen == th.gen {
+			break
+		}
+	}
+	return bt, true
+}
+
+//go:nosplit
+func vdsoClockGettime(clockID int32) (bintime, bool) {
+	if timekeepSharedPage == nil || timekeepSharedPage.ver != _VDSO_TK_VER_CURR {
+		return bintime{}, false
+	}
+	abs := false
+	switch clockID {
+	case _CLOCK_MONOTONIC:
+		/* ok */
+	case _CLOCK_REALTIME:
+		abs = true
+	default:
+		return bintime{}, false
+	}
+
+	return binuptime(abs)
+}
+
+func fallback_nanotime() int64
+func fallback_walltime() (sec int64, nsec int32)
+
+//go:nosplit
+func nanotime() int64 {
+	bt, ok := vdsoClockGettime(_CLOCK_MONOTONIC)
+	if !ok {
+		return fallback_nanotime()
+	}
+	return int64((1e9 * uint64(bt.sec)) + ((1e9 * uint64(bt.frac>>32)) >> 32))
+}
+
+func walltime() (sec int64, nsec int32) {
+	bt, ok := vdsoClockGettime(_CLOCK_REALTIME)
+	if !ok {
+		return fallback_walltime()
+	}
+	return int64(bt.sec), int32((1e9 * uint64(bt.frac>>32)) >> 32)
+}

src/runtime/vdso_freebsd_arm.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.
+
+package runtime
+
+const (
+	_VDSO_TH_ALGO_ARM_GENTIM = 1
+)
+
+func getCntxct(physical bool) uint32
+
+//go:nosplit
+func (th *vdsoTimehands) getTimecounter() (uint32, bool) {
+	switch th.algo {
+	case _VDSO_TH_ALGO_ARM_GENTIM:
+		return getCntxct(th.physical != 0), true
+	default:
+		return 0, false
+	}
+}

src/runtime/vdso_freebsd_x86.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 freebsd
+// +build 386 amd64
+
+package runtime
+
+const (
+	_VDSO_TH_ALGO_X86_TSC = 1
+)
+
+//go:nosplit
+func (th *vdsoTimehands) getTSCTimecounter() uint32 {
+	tsc := cputicks()
+	if th.x86_shift > 0 {
+		tsc >>= th.x86_shift
+	}
+	return uint32(tsc)
+}
+
+//go:nosplit
+func (th *vdsoTimehands) getTimecounter() (uint32, bool) {
+	switch th.algo {
+	case _VDSO_TH_ALGO_X86_TSC:
+		return th.getTSCTimecounter(), true
+	default:
+		return 0, false
+	}
+}