cpu: support reading arm64 CPU feature registers

This allows to detect ARM64 CPU features on non-Linux systems. On Linux,
this is used in case /proc/self/auxv cannot be read.

Change-Id: I67d55e989f2beda0c05a97ca5e55781840a8e01c
Reviewed-on: https://go-review.googlesource.com/c/sys/+/209478
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cpu/cpu_arm64.go

@@ -0,0 +1,134 @@
+// Copyright 2019 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 cpu
+
+import "runtime"
+
+const cacheLineSize = 64
+
+func init() {
+	switch runtime.GOOS {
+	case "darwin":
+		// iOS does not seem to allow reading these registers
+	case "android", "linux":
+		doinit()
+	default:
+		readARM64Registers()
+	}
+}
+
+func readARM64Registers() {
+	Initialized = true
+
+	// ID_AA64ISAR0_EL1
+	isar0 := getisar0()
+
+	switch extractBits(isar0, 4, 7) {
+	case 1:
+		ARM64.HasAES = true
+	case 2:
+		ARM64.HasAES = true
+		ARM64.HasPMULL = true
+	}
+
+	switch extractBits(isar0, 8, 11) {
+	case 1:
+		ARM64.HasSHA1 = true
+	}
+
+	switch extractBits(isar0, 12, 15) {
+	case 1:
+		ARM64.HasSHA2 = true
+	case 2:
+		ARM64.HasSHA2 = true
+		ARM64.HasSHA512 = true
+	}
+
+	switch extractBits(isar0, 16, 19) {
+	case 1:
+		ARM64.HasCRC32 = true
+	}
+
+	switch extractBits(isar0, 20, 23) {
+	case 2:
+		ARM64.HasATOMICS = true
+	}
+
+	switch extractBits(isar0, 28, 31) {
+	case 1:
+		ARM64.HasASIMDRDM = true
+	}
+
+	switch extractBits(isar0, 32, 35) {
+	case 1:
+		ARM64.HasSHA3 = true
+	}
+
+	switch extractBits(isar0, 36, 39) {
+	case 1:
+		ARM64.HasSM3 = true
+	}
+
+	switch extractBits(isar0, 40, 43) {
+	case 1:
+		ARM64.HasSM4 = true
+	}
+
+	switch extractBits(isar0, 44, 47) {
+	case 1:
+		ARM64.HasASIMDDP = true
+	}
+
+	// ID_AA64ISAR1_EL1
+	isar1 := getisar1()
+
+	switch extractBits(isar1, 0, 3) {
+	case 1:
+		ARM64.HasDCPOP = true
+	}
+
+	switch extractBits(isar1, 12, 15) {
+	case 1:
+		ARM64.HasJSCVT = true
+	}
+
+	switch extractBits(isar1, 16, 19) {
+	case 1:
+		ARM64.HasFCMA = true
+	}
+
+	switch extractBits(isar1, 20, 23) {
+	case 1:
+		ARM64.HasLRCPC = true
+	}
+
+	// ID_AA64PFR0_EL1
+	pfr0 := getpfr0()
+
+	switch extractBits(pfr0, 16, 19) {
+	case 0:
+		ARM64.HasFP = true
+	case 1:
+		ARM64.HasFP = true
+		ARM64.HasFPHP = true
+	}
+
+	switch extractBits(pfr0, 20, 23) {
+	case 0:
+		ARM64.HasASIMD = true
+	case 1:
+		ARM64.HasASIMD = true
+		ARM64.HasASIMDHP = true
+	}
+
+	switch extractBits(pfr0, 32, 35) {
+	case 1:
+		ARM64.HasSVE = true
+	}
+}
+
+func extractBits(data uint64, start, end uint) uint {
+	return (uint)(data>>start) & ((1 << (end - start + 1)) - 1)
+}

cpu/cpu_arm64.s

@@ -0,0 +1,31 @@
+// Copyright 2019 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 !gccgo
+
+#include "textflag.h"
+
+// func getisar0() uint64
+TEXT ·getisar0(SB),NOSPLIT,$0-8
+	// get Instruction Set Attributes 0 into x0
+	// mrs x0, ID_AA64ISAR0_EL1 = d5380600
+	WORD	$0xd5380600
+	MOVD	R0, ret+0(FP)
+	RET
+
+// func getisar1() uint64
+TEXT ·getisar1(SB),NOSPLIT,$0-8
+	// get Instruction Set Attributes 1 into x0
+	// mrs x0, ID_AA64ISAR1_EL1 = d5380620
+	WORD	$0xd5380620
+	MOVD	R0, ret+0(FP)
+	RET
+
+// func getpfr0() uint64
+TEXT ·getpfr0(SB),NOSPLIT,$0-8
+	// get Processor Feature Register 0 into x0
+	// mrs x0, ID_AA64PFR0_EL1 = d5380400
+	WORD	$0xd5380400
+	MOVD	R0, ret+0(FP)
+	RET

cpu/cpu_gc_arm64.go

@@ -0,0 +1,11 @@
+// Copyright 2019 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 !gccgo
+
+package cpu
+
+func getisar0() uint64
+func getisar1() uint64
+func getpfr0() uint64

cpu/cpu_gccgo_arm64.go

@@ -0,0 +1,11 @@
+// Copyright 2019 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 gccgo
+
+package cpu
+
+func getisar0() uint64 { return 0 }
+func getisar1() uint64 { return 0 }
+func getpfr0() uint64  { return 0 }

cpu/cpu_linux.go

@@ -2,58 +2,14 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !amd64,!amd64p32,!386
+// +build !386,!amd64,!amd64p32,!arm64
 
 package cpu
 
-import (
-	"io/ioutil"
-)
-
-const (
-	_AT_HWCAP  = 16
-	_AT_HWCAP2 = 26
-
-	procAuxv = "/proc/self/auxv"
-
-	uintSize = int(32 << (^uint(0) >> 63))
-)
-
-// For those platforms don't have a 'cpuid' equivalent we use HWCAP/HWCAP2
-// These are initialized in cpu_$GOARCH.go
-// and should not be changed after they are initialized.
-var hwCap uint
-var hwCap2 uint
-
 func init() {
-	buf, err := ioutil.ReadFile(procAuxv)
-	if err != nil {
-		// e.g. on android /proc/self/auxv is not accessible, so silently
-		// ignore the error and leave Initialized = false
+	if err := readHWCAP(); err != nil {
 		return
 	}
-
-	bo := hostByteOrder()
-	for len(buf) >= 2*(uintSize/8) {
-		var tag, val uint
-		switch uintSize {
-		case 32:
-			tag = uint(bo.Uint32(buf[0:]))
-			val = uint(bo.Uint32(buf[4:]))
-			buf = buf[8:]
-		case 64:
-			tag = uint(bo.Uint64(buf[0:]))
-			val = uint(bo.Uint64(buf[8:]))
-			buf = buf[16:]
-		}
-		switch tag {
-		case _AT_HWCAP:
-			hwCap = val
-		case _AT_HWCAP2:
-			hwCap2 = val
-		}
-	}
 	doinit()
-
 	Initialized = true
 }

cpu/cpu_linux_arm64.go

@@ -4,8 +4,6 @@
 
 package cpu
 
-const cacheLineSize = 64
-
 // HWCAP/HWCAP2 bits. These are exposed by Linux.
 const (
 	hwcap_FP       = 1 << 0
@@ -35,6 +33,12 @@ const (
 )
 
 func doinit() {
+	if err := readHWCAP(); err != nil {
+		// failed to read /proc/self/auxv, try reading registers directly
+		readARM64Registers()
+		return
+	}
+
 	// HWCAP feature bits
 	ARM64.HasFP = isSet(hwCap, hwcap_FP)
 	ARM64.HasASIMD = isSet(hwCap, hwcap_ASIMD)

cpu/cpu_other_arm64.go

@@ -6,4 +6,4 @@
 
 package cpu
 
-const cacheLineSize = 64
+func doinit() {}

cpu/cpu_test.go

@@ -31,7 +31,7 @@ func TestAVX2hasAVX(t *testing.T) {
 }
 
 func TestARM64minimalFeatures(t *testing.T) {
-	if runtime.GOARCH != "arm64" || runtime.GOOS != "linux" {
+	if runtime.GOARCH != "arm64" || runtime.GOOS == "darwin" {
 		return
 	}
 	if !cpu.ARM64.HasASIMD {

cpu/hwcap_linux.go

@@ -0,0 +1,56 @@
+// Copyright 2019 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 cpu
+
+import (
+	"io/ioutil"
+)
+
+const (
+	_AT_HWCAP  = 16
+	_AT_HWCAP2 = 26
+
+	procAuxv = "/proc/self/auxv"
+
+	uintSize = int(32 << (^uint(0) >> 63))
+)
+
+// For those platforms don't have a 'cpuid' equivalent we use HWCAP/HWCAP2
+// These are initialized in cpu_$GOARCH.go
+// and should not be changed after they are initialized.
+var hwCap uint
+var hwCap2 uint
+
+func readHWCAP() error {
+	buf, err := ioutil.ReadFile(procAuxv)
+	if err != nil {
+		// e.g. on android /proc/self/auxv is not accessible, so silently
+		// ignore the error and leave Initialized = false. On some
+		// architectures (e.g. arm64) doinit() implements a fallback
+		// readout and will set Initialized = true again.
+		return err
+	}
+	bo := hostByteOrder()
+	for len(buf) >= 2*(uintSize/8) {
+		var tag, val uint
+		switch uintSize {
+		case 32:
+			tag = uint(bo.Uint32(buf[0:]))
+			val = uint(bo.Uint32(buf[4:]))
+			buf = buf[8:]
+		case 64:
+			tag = uint(bo.Uint64(buf[0:]))
+			val = uint(bo.Uint64(buf[8:]))
+			buf = buf[16:]
+		}
+		switch tag {
+		case _AT_HWCAP:
+			hwCap = val
+		case _AT_HWCAP2:
+			hwCap2 = val
+		}
+	}
+	return nil
+}