This commit is contained in:
Michael McLoughlin
2018-11-21 13:02:18 -06:00
parent cb259ce43b
commit 59e6af7d36
10 changed files with 553 additions and 2 deletions

34
internal/gen/gen.go Normal file
View File

@@ -0,0 +1,34 @@
package gen
import (
"fmt"
"io"
"github.com/mmcloughlin/avo/internal/inst"
)
type Interface interface {
Generate(io.Writer, []*inst.Instruction) error
}
type Func func(io.Writer, []*inst.Instruction) error
func (f Func) Generate(w io.Writer, is []*inst.Instruction) error {
return f(w, is)
}
type printer struct {
w io.Writer
err error
}
func (p *printer) printf(format string, args ...interface{}) {
if p.err != nil {
return
}
_, p.err = fmt.Fprintf(p.w, format, args...)
}
func (p *printer) Err() error {
return p.err
}

127
internal/gen/loadertest.go Normal file
View File

@@ -0,0 +1,127 @@
package gen
import (
"io"
"strings"
"github.com/mmcloughlin/avo/internal/inst"
)
type LoaderTest struct{}
func (l LoaderTest) Generate(w io.Writer, is []*inst.Instruction) error {
p := &printer{w: w}
p.printf("TEXT loadertest(SB), 0, $0\n")
for _, i := range is {
p.printf("\t// %s %s\n", i.Opcode, i.Summary)
for _, f := range i.Forms {
as := args(f.Operands)
p.printf("\t// %#v\n", f.Operands)
if as == nil {
p.printf("\t// skip\n")
continue
}
p.printf("\t%s\t%s\n", i.Opcode, strings.Join(as, ", "))
}
p.printf("\n")
}
p.printf("\tRET\n")
return p.Err()
}
func args(ops []inst.Operand) []string {
as := make([]string, len(ops))
for i, op := range ops {
a := arg(op.Type)
if a == "" {
return nil
}
as[i] = a
}
return as
}
// arg generates an argument for an operand of the given type.
func arg(t string) string {
m := map[string]string{
// <xs:enumeration value="1" />
// <xs:enumeration value="3" />
// <xs:enumeration value="imm4" />
// <xs:enumeration value="imm8" />
// <xs:enumeration value="imm16" />
// <xs:enumeration value="imm32" />
// <xs:enumeration value="imm64" />
// <xs:enumeration value="al" />
// <xs:enumeration value="cl" />
// <xs:enumeration value="r8" />
// <xs:enumeration value="r8l" />
// <xs:enumeration value="ax" />
// <xs:enumeration value="r16" />
// <xs:enumeration value="r16l" />
// <xs:enumeration value="eax" />
// <xs:enumeration value="r32" />
// <xs:enumeration value="r32l" />
// <xs:enumeration value="rax" />
// <xs:enumeration value="r64" />
"r64": "R8",
// <xs:enumeration value="mm" />
// <xs:enumeration value="xmm0" />
// <xs:enumeration value="xmm" />
// <xs:enumeration value="xmm{k}" />
// <xs:enumeration value="xmm{k}{z}" />
// <xs:enumeration value="ymm" />
// <xs:enumeration value="ymm{k}" />
// <xs:enumeration value="ymm{k}{z}" />
// <xs:enumeration value="zmm" />
// <xs:enumeration value="zmm{k}" />
// <xs:enumeration value="zmm{k}{z}" />
// <xs:enumeration value="k" />
// <xs:enumeration value="k{k}" />
// <xs:enumeration value="moffs32" />
// <xs:enumeration value="moffs64" />
// <xs:enumeration value="m" />
// <xs:enumeration value="m8" />
// <xs:enumeration value="m16" />
// <xs:enumeration value="m16{k}{z}" />
// <xs:enumeration value="m32" />
// <xs:enumeration value="m32{k}" />
// <xs:enumeration value="m32{k}{z}" />
// <xs:enumeration value="m64" />
// <xs:enumeration value="m64{k}" />
// <xs:enumeration value="m64{k}{z}" />
// <xs:enumeration value="m128" />
// <xs:enumeration value="m128{k}{z}" />
// <xs:enumeration value="m256" />
// <xs:enumeration value="m256{k}{z}" />
// <xs:enumeration value="m512" />
// <xs:enumeration value="m512{k}{z}" />
// <xs:enumeration value="m64/m32bcst" />
// <xs:enumeration value="m128/m32bcst" />
// <xs:enumeration value="m256/m32bcst" />
// <xs:enumeration value="m512/m32bcst" />
// <xs:enumeration value="m128/m64bcst" />
// <xs:enumeration value="m256/m64bcst" />
// <xs:enumeration value="m512/m64bcst" />
// <xs:enumeration value="vm32x" />
// <xs:enumeration value="vm32x{k}" />
// <xs:enumeration value="vm64x" />
// <xs:enumeration value="vm64x{k}" />
// <xs:enumeration value="vm32y" />
// <xs:enumeration value="vm32y{k}" />
// <xs:enumeration value="vm64y" />
// <xs:enumeration value="vm64y{k}" />
// <xs:enumeration value="vm32z" />
// <xs:enumeration value="vm32z{k}" />
// <xs:enumeration value="vm64z" />
// <xs:enumeration value="vm64z{k}" />
// <xs:enumeration value="rel8" />
// <xs:enumeration value="rel32" />
// <xs:enumeration value="{er}" />
// <xs:enumeration value="{sae}" />
}
return m[t]
}

View File

@@ -2,6 +2,7 @@ package inst
type Instruction struct {
Opcode string
Summary string
Forms []Form
}
@@ -22,3 +23,33 @@ const (
W Action = 0x2
RW Action = R | W
)
func ActionFromReadWrite(r, w bool) Action {
var a Action
if r {
a |= R
}
if w {
a |= W
}
return a
}
func (a Action) Read() bool {
return (a & R) != 0
}
func (a Action) Write() bool {
return (a & W) != 0
}
func (a Action) String() string {
s := ""
if a.Read() {
s += "r"
}
if a.Write() {
s += "w"
}
return s
}

20
internal/load/ifind.sh Executable file
View File

@@ -0,0 +1,20 @@
#!/bin/bash
inst=$1
eval $(go env)
section() {
echo '------------------------------------------'
echo $1
echo '------------------------------------------'
}
section 'stdlib cmd/'
grep -iR ${inst} ${GOROOT}/src/cmd/{asm,internal/obj/x86}
section 'x/arch/x86 repo'
grep -iR ${inst} ${GOPATH}/src/golang.org/x/arch/x86/
section '*_amd64.s files in stdlib'
find ${GOROOT}/src -name '*_amd64.s' | xargs grep -i ${inst}

181
internal/load/load.go Normal file
View File

@@ -0,0 +1,181 @@
package load
import (
"path/filepath"
"strconv"
"strings"
"github.com/mmcloughlin/avo/internal/inst"
"github.com/mmcloughlin/avo/internal/opcodescsv"
"github.com/mmcloughlin/avo/internal/opcodesxml"
)
const (
defaultCSVName = "x86.v0.2.csv"
defaultOpcodesXMLName = "x86_64.xml"
)
type Loader struct {
X86CSVPath string
OpcodesXMLPath string
alias map[opcodescsv.Alias]string
}
func NewLoaderFromDataDir(dir string) *Loader {
return &Loader{
X86CSVPath: filepath.Join(dir, defaultCSVName),
OpcodesXMLPath: filepath.Join(dir, defaultOpcodesXMLName),
}
}
func (l *Loader) Load() ([]*inst.Instruction, error) {
if err := l.init(); err != nil {
return nil, err
}
// Load Opcodes XML file.
iset, err := opcodesxml.ReadFile(l.OpcodesXMLPath)
if err != nil {
return nil, err
}
// Load opcodes XML data, grouped by Go opcode.
im := map[string]*inst.Instruction{}
for _, i := range iset.Instructions {
for _, f := range i.Forms {
if !l.include(f) {
continue
}
opcode := l.goname(f)
if im[opcode] == nil {
im[opcode] = &inst.Instruction{
Opcode: opcode,
Summary: i.Summary,
}
}
im[opcode].Forms = append(im[opcode].Forms, l.form(f))
}
}
// Convert to a slice to return.
is := make([]*inst.Instruction, 0, len(im))
for _, i := range im {
is = append(is, i)
}
return is, nil
}
func (l *Loader) init() error {
icsv, err := opcodescsv.ReadFile(l.X86CSVPath)
if err != nil {
return err
}
l.alias, err = opcodescsv.BuildAliasMap(icsv)
if err != nil {
return err
}
return nil
}
// include decides whether to include the instruction form in the avo listing.
// This discards some opcodes that are not supported in Go.
func (l Loader) include(f opcodesxml.Form) bool {
// Exclude certain ISAs simply not present in Go.
for _, isa := range f.ISA {
switch isa.ID {
case "TBM", "CLZERO", "MONITORX", "FEMMS":
return false
}
}
// x86 csv contains a number of CMOV* instructions which are actually not valid
// Go instructions. The valid Go forms should have different opcodes from GNU.
// Therefore a decent "heuristic" is CMOV* instructions that do not have
// aliases.
if strings.HasPrefix(f.GASName, "cmov") && l.lookupAlias(f) == "" {
return false
}
// Some specific exclusions.
switch f.GASName {
case "callq":
return false
}
return true
}
func (l Loader) lookupAlias(f opcodesxml.Form) string {
a := opcodescsv.Alias{Opcode: f.GASName, DataSize: datasize(f)}
return l.alias[a]
}
func (l Loader) goname(f opcodesxml.Form) string {
// Use go opcode from Opcodes XML where available.
if f.GoName != "" {
return f.GoName
}
// Return alias if available.
if a := l.lookupAlias(f); a != "" {
return a
}
n := strings.ToUpper(f.GASName)
// Some need data sizes added to them.
// TODO(mbm): is there a better way of determining which ones these are?
s := datasize(f)
suffix := map[int]string{16: "W", 32: "L", 64: "Q"}
switch n {
case "RDRAND", "RDSEED":
n += suffix[s]
}
return n
}
func (l Loader) form(f opcodesxml.Form) inst.Form {
return inst.Form{
Operands: operands(f.Operands),
}
}
// operands maps Opcodes XML operands to avo format.
func operands(ops []opcodesxml.Operand) []inst.Operand {
r := make([]inst.Operand, 0, len(ops))
for _, op := range ops {
r = append(r, operand(op))
}
return r
}
// operand maps an Opcodes XML operand to avo format.
func operand(op opcodesxml.Operand) inst.Operand {
return inst.Operand{
Type: op.Type,
Action: inst.ActionFromReadWrite(op.Input, op.Output),
}
}
// datasize (intelligently) guesses the datasize of an instruction form.
func datasize(f opcodesxml.Form) int {
for _, op := range f.Operands {
if !op.Output {
continue
}
for s := 8; s <= 64; s *= 2 {
if strings.HasSuffix(op.Type, strconv.Itoa(s)) {
return s
}
}
}
return 0
}

View File

@@ -0,0 +1,31 @@
package load_test
import (
"bytes"
"testing"
"github.com/mmcloughlin/avo/internal/gen"
"github.com/mmcloughlin/avo/internal/inst"
"github.com/mmcloughlin/avo/internal/load"
"github.com/mmcloughlin/avo/internal/test"
)
func Load(t *testing.T) []*inst.Instruction {
t.Helper()
l := load.NewLoaderFromDataDir("testdata")
is, err := l.Load()
if err != nil {
t.Fatal(err)
}
return is
}
func TestAssembles(t *testing.T) {
is := Load(t)
g := &gen.LoaderTest{}
var buf bytes.Buffer
g.Generate(&buf, is)
test.Assembles(t, buf.Bytes())
}

1
internal/load/testdata Symbolic link
View File

@@ -0,0 +1 @@
../data

View File

@@ -0,0 +1,38 @@
package opcodescsv
import (
"strconv"
"strings"
"golang.org/x/arch/x86/x86csv"
)
type Alias struct {
Opcode string
DataSize int
}
// BuildAliasMap constructs a map from AT&T/GNU/Intel to Go syntax.
func BuildAliasMap(is []*x86csv.Inst) (map[Alias]string, error) {
m := map[Alias]string{}
for _, i := range is {
s, err := datasize(i.DataSize)
if err != nil {
return nil, err
}
for _, alt := range []string{i.IntelOpcode(), i.GNUOpcode()} {
if strings.ToUpper(alt) != i.GoOpcode() {
m[Alias{Opcode: strings.ToLower(alt), DataSize: s}] = i.GoOpcode()
}
}
}
return m, nil
}
func datasize(s string) (int, error) {
if s == "" {
return 0, nil
}
return strconv.Atoi(s)
}

19
internal/opcodescsv/io.go Normal file
View File

@@ -0,0 +1,19 @@
package opcodescsv
import (
"os"
"golang.org/x/arch/x86/x86csv"
)
// ReadFile reads the given x86 CSV file.
func ReadFile(filename string) ([]*x86csv.Inst, error) {
f, err := os.Open(filename)
if err != nil {
return nil, err
}
defer f.Close()
r := x86csv.NewReader(f)
return r.ReadAll()
}

69
internal/test/utils.go Normal file
View File

@@ -0,0 +1,69 @@
package test
import (
"io/ioutil"
"os"
"os/exec"
"path/filepath"
"runtime"
"testing"
)
// Assembles asserts that the given assembly code passes the go assembler.
func Assembles(t *testing.T, asm []byte) {
t.Helper()
dir, clean := TempDir(t)
defer clean()
asmfilename := filepath.Join(dir, "asm.s")
if err := ioutil.WriteFile(asmfilename, asm, 0600); err != nil {
t.Fatal(err)
}
objfilename := filepath.Join(dir, "asm.o")
goexec(t, "tool", "asm", "-e", "-o", objfilename, asmfilename)
}
// TempDir creates a temp directory. Returns the path to the directory and a
// cleanup function.
func TempDir(t *testing.T) (string, func()) {
t.Helper()
dir, err := ioutil.TempDir("", "avo")
if err != nil {
t.Fatal(err)
}
return dir, func() {
if err := os.RemoveAll(dir); err != nil {
t.Fatal(err)
}
}
}
// gobin returns a best guess path to the "go" binary.
func gobin() string {
var exeSuffix string
if runtime.GOOS == "windows" {
exeSuffix = ".exe"
}
path := filepath.Join(runtime.GOROOT(), "bin", "go"+exeSuffix)
if _, err := os.Stat(path); err == nil {
return path
}
return "go"
}
// goexec runs a "go" command and checks the output.
func goexec(t *testing.T, arg ...string) {
t.Helper()
cmd := exec.Command(gobin(), arg...)
t.Logf("exec: %s", cmd.Args)
b, err := cmd.CombinedOutput()
t.Logf("output:\n%s\n", string(b))
if err != nil {
t.Fatal(err)
}
}