runtime: get s390x vector facility availability from AT_HWCAP

This is a more robust method for obtaining the availability of vx.
Since this variable may be checked frequently I've also now
padded it so that it will be in its own cache line.

I've kept the other check (in hash/crc32) the same for now until
I can figure out the best way to update it.

Updates #15403.

Change-Id: I74eed651afc6f6a9c5fa3b88fa6a2b0c9ecf5875
Reviewed-on: https://go-review.googlesource.com/31149Reviewed-by: Austin Clements <austin@google.com>

src/runtime/asm_s390x.s

@@ -7,44 +7,6 @@
 #include "funcdata.h"
 #include "textflag.h"
 
-// Indicate the status of vector facility
-// -1: 	init value
-// 0:	vector not installed
-// 1:	vector installed and enabled
-// 2:	vector installed but not enabled
-
-DATA runtime·vectorfacility+0x00(SB)/4, $-1
-GLOBL runtime·vectorfacility(SB), NOPTR, $4
-
-TEXT runtime·checkvectorfacility(SB),NOSPLIT,$32-0
-	MOVD    $2, R0
-	MOVD	R1, tmp-32(SP)
-	MOVD    $x-24(SP), R1
-	XC	$24, 0(R1), 0(R1)
-//      STFLE   0(R1)
-	WORD    $0xB2B01000
-	MOVBZ   z-8(SP), R1
-	AND     $0x40, R1
-	BNE     vectorinstalled
-	MOVB    $0, runtime·vectorfacility(SB) //Vector not installed
-	MOVD	tmp-32(SP), R1
-	MOVD    $0, R0
-	RET
-vectorinstalled:
-	// check if the vector instruction has been enabled
-	VLEIB   $0, $0xF, V16
-	VLGVB   $0, V16, R0
-	CMPBEQ  R0, $0xF, vectorenabled
-	MOVB    $2, runtime·vectorfacility(SB) //Vector installed but not enabled
-	MOVD    tmp-32(SP), R1
-	MOVD    $0, R0
-	RET
-vectorenabled:
-	MOVB    $1, runtime·vectorfacility(SB) //Vector installed and enabled
-	MOVD    tmp-32(SP), R1
-	MOVD    $0, R0
-	RET
-
 TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	// R2 = argc; R3 = argv; R11 = temp; R13 = g; R15 = stack pointer
 	// C TLS base pointer in AR0:AR1
@@ -888,14 +850,14 @@ TEXT runtime·fastrand(SB), NOSPLIT, $0-4
 	MOVW	R3, ret+0(FP)
 	RET
 
-TEXT bytes·IndexByte(SB),NOSPLIT,$0-40
+TEXT bytes·IndexByte(SB),NOSPLIT|NOFRAME,$0-40
 	MOVD	s+0(FP), R3     // s => R3
 	MOVD	s_len+8(FP), R4 // s_len => R4
 	MOVBZ	c+24(FP), R5    // c => R5
 	MOVD	$ret+32(FP), R2 // &ret => R9
 	BR	runtime·indexbytebody(SB)
 
-TEXT strings·IndexByte(SB),NOSPLIT,$0-32
+TEXT strings·IndexByte(SB),NOSPLIT|NOFRAME,$0-32
 	MOVD	s+0(FP), R3     // s => R3
 	MOVD	s_len+8(FP), R4 // s_len => R4
 	MOVBZ	c+16(FP), R5    // c => R5
@@ -907,7 +869,7 @@ TEXT strings·IndexByte(SB),NOSPLIT,$0-32
 // R4: s_len
 // R5: c -- byte sought
 // R2: &ret -- address to put index into
-TEXT runtime·indexbytebody(SB),NOSPLIT,$0
+TEXT runtime·indexbytebody(SB),NOSPLIT|NOFRAME,$0
 	CMPBEQ	R4, $0, notfound
 	MOVD	R3, R6          // store base for later
 	ADD	R3, R4, R8      // the address after the end of the string
@@ -931,12 +893,10 @@ notfound:
 	RET
 
 large:
-	MOVB	runtime·vectorfacility(SB), R1
-	CMPBEQ	R1, $-1, checkvector	// vectorfacility = -1, vector not checked yet
-vectorchecked:
-	CMPBEQ	R1, $1, vectorimpl      // vectorfacility = 1, vector supported
+	MOVBZ	·cpu+facilities_hasVX(SB), R1
+	CMPBNE	R1, $0, vectorimpl
 
-srstimpl:                       // vectorfacility != 1, not support or enable vector
+srstimpl:                       // no vector facility
 	MOVBZ	R5, R0          // c needs to be in R0, leave until last minute as currently R0 is expected to be 0
 srstloop:
 	WORD	$0xB25E0083     // srst %r8, %r3 (search the range [R3, R8))
@@ -989,11 +949,6 @@ notalignedloop:
 	CMPBNE	R7, R5, notalignedloop
 	BR	found
 
-checkvector:
-	CALL	runtime·checkvectorfacility(SB)
-	MOVB    runtime·vectorfacility(SB), R1
-	BR	vectorchecked
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVW	$0, R3
 	RET

src/runtime/os_linux.go

@@ -182,6 +182,7 @@ var failthreadcreate = []byte("runtime: failed to create new OS thread\n")
 const (
 	_AT_NULL   = 0  // End of vector
 	_AT_PAGESZ = 6  // System physical page size
+	_AT_HWCAP  = 16 // hardware capability bit vector
 	_AT_RANDOM = 25 // introduced in 2.6.29
 )
 

src/runtime/os_linux_arm.go

@@ -8,7 +8,6 @@ import "unsafe"
 
 const (
 	_AT_PLATFORM = 15 //  introduced in at least 2.6.11
-	_AT_HWCAP    = 16 // introduced in at least 2.6.11
 
 	_HWCAP_VFP   = 1 << 6  // introduced in at least 2.6.11
 	_HWCAP_VFPv3 = 1 << 13 // introduced in 2.6.30

src/runtime/os_linux_noauxv.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !amd64,!arm,!arm64,!mips64,!mips64le
+// +build !amd64,!arm,!arm64,!mips64,!mips64le,!s390x
 
 package runtime
 

src/runtime/os_linux_s390x.go

+// Copyright 2016 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
+
+import (
+	"runtime/internal/sys"
+)
+
+const (
+	// bit masks taken from bits/hwcap.h
+	_HWCAP_S390_VX = 2048 // vector facility
+)
+
+// facilities is padded to avoid false sharing.
+type facilities struct {
+	_     [sys.CacheLineSize]byte
+	hasVX bool // vector facility
+	_     [sys.CacheLineSize]byte
+}
+
+// cpu indicates the availability of s390x facilities that can be used in
+// Go assembly but are optional on models supported by Go.
+var cpu facilities
+
+func archauxv(tag, val uintptr) {
+	switch tag {
+	case _AT_HWCAP: // CPU capability bit flags
+		cpu.hasVX = val&_HWCAP_S390_VX != 0
+	}
+}