debug/dwarf: add Reader.SeekPC and Data.Ranges

These new methods help find the compilation unit to pass to the
LineReader method in order to find the line information for a PC.
The Ranges method also helps identify the specific function for a PC,
needed to determine the function name.

This uses the .debug.ranges section if necessary, and changes the object
file format packages to pass in the section contents if available.

Change-Id: I5ebc3d27faaf1a126ffb17a1e6027efdf64af836
Reviewed-on: https://go-review.googlesource.com/20769
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/debug/dwarf/entry.go

@@ -636,3 +636,123 @@ func (r *Reader) clone() typeReader {
 func (r *Reader) offset() Offset {
 	return r.b.off
 }
+
+// SeekPC returns the Entry for the compilation unit that includes pc,
+// and positions the reader to read the children of that unit.  If pc
+// is not covered by any unit, SeekPC returns ErrUnknownPC and the
+// position of the reader is undefined.
+//
+// Because compilation units can describe multiple regions of the
+// executable, in the worst case SeekPC must search through all the
+// ranges in all the compilation units. Each call to SeekPC starts the
+// search at the compilation unit of the last call, so in general
+// looking up a series of PCs will be faster if they are sorted. If
+// the caller wishes to do repeated fast PC lookups, it should build
+// an appropriate index using the Ranges method.
+func (r *Reader) SeekPC(pc uint64) (*Entry, error) {
+	unit := r.unit
+	for i := 0; i < len(r.d.unit); i++ {
+		if unit >= len(r.d.unit) {
+			unit = 0
+		}
+		r.err = nil
+		r.lastChildren = false
+		r.unit = unit
+		u := &r.d.unit[unit]
+		r.b = makeBuf(r.d, u, "info", u.off, u.data)
+		e, err := r.Next()
+		if err != nil {
+			return nil, err
+		}
+		ranges, err := r.d.Ranges(e)
+		if err != nil {
+			return nil, err
+		}
+		for _, pcs := range ranges {
+			if pcs[0] <= pc && pc < pcs[1] {
+				return e, nil
+			}
+		}
+		unit++
+	}
+	return nil, ErrUnknownPC
+}
+
+// Ranges returns the PC ranges covered by e, a slice of [low,high) pairs.
+// Only some entry types, such as TagCompileUnit or TagSubprogram, have PC
+// ranges; for others, this will return nil with no error.
+func (d *Data) Ranges(e *Entry) ([][2]uint64, error) {
+	var ret [][2]uint64
+
+	low, lowOK := e.Val(AttrLowpc).(uint64)
+
+	var high uint64
+	var highOK bool
+	highField := e.AttrField(AttrHighpc)
+	if highField != nil {
+		switch highField.Class {
+		case ClassAddress:
+			high, highOK = highField.Val.(uint64)
+		case ClassConstant:
+			off, ok := highField.Val.(int64)
+			if ok {
+				high = low + uint64(off)
+				highOK = true
+			}
+		}
+	}
+
+	if lowOK && highOK {
+		ret = append(ret, [2]uint64{low, high})
+	}
+
+	ranges, rangesOK := e.Val(AttrRanges).(int64)
+	if rangesOK && d.ranges != nil {
+		// The initial base address is the lowpc attribute
+		// of the enclosing compilation unit.
+		// Although DWARF specifies the lowpc attribute,
+		// comments in gdb/dwarf2read.c say that some versions
+		// of GCC use the entrypc attribute, so we check that too.
+		var cu *Entry
+		if e.Tag == TagCompileUnit {
+			cu = e
+		} else {
+			i := d.offsetToUnit(e.Offset)
+			if i == -1 {
+				return nil, errors.New("no unit for entry")
+			}
+			u := &d.unit[i]
+			b := makeBuf(d, u, "info", u.off, u.data)
+			cu = b.entry(u.atable, u.base)
+			if b.err != nil {
+				return nil, b.err
+			}
+		}
+
+		var base uint64
+		if cuEntry, cuEntryOK := cu.Val(AttrEntrypc).(uint64); cuEntryOK {
+			base = cuEntry
+		} else if cuLow, cuLowOK := cu.Val(AttrLowpc).(uint64); cuLowOK {
+			base = cuLow
+		}
+
+		u := &d.unit[d.offsetToUnit(e.Offset)]
+		buf := makeBuf(d, u, "ranges", Offset(ranges), d.ranges[ranges:])
+		for len(buf.data) > 0 {
+			low = buf.addr()
+			high = buf.addr()
+
+			if low == 0 && high == 0 {
+				break
+			}
+
+			if low == ^uint64(0)>>uint((8-u.addrsize())*8) {
+				base = high
+			} else {
+				ret = append(ret, [2]uint64{base + low, base + high})
+			}
+		}
+	}
+
+	return ret, nil
+}

src/debug/dwarf/entry_test.go

@@ -6,6 +6,7 @@ package dwarf_test
 
 import (
 	. "debug/dwarf"
+	"reflect"
 	"testing"
 )
 
@@ -34,3 +35,103 @@ func TestSplit(t *testing.T) {
 		t.Fatalf("bad class: have %s, want %s", f.Class, ClassUnknown)
 	}
 }
+
+// wantRange maps from a PC to the ranges of the compilation unit
+// containing that PC.
+type wantRange struct {
+	pc     uint64
+	ranges [][2]uint64
+}
+
+func TestReaderSeek(t *testing.T) {
+	want := []wantRange{
+		{0x40059d, [][2]uint64{{0x40059d, 0x400601}}},
+		{0x400600, [][2]uint64{{0x40059d, 0x400601}}},
+		{0x400601, [][2]uint64{{0x400601, 0x400611}}},
+		{0x4005f0, [][2]uint64{{0x40059d, 0x400601}}}, // loop test
+		{0x10, nil},
+		{0x400611, nil},
+	}
+	testRanges(t, "testdata/line-gcc.elf", want)
+}
+
+func TestRangesSection(t *testing.T) {
+	want := []wantRange{
+		{0x400500, [][2]uint64{{0x400500, 0x400549}, {0x400400, 0x400408}}},
+		{0x400400, [][2]uint64{{0x400500, 0x400549}, {0x400400, 0x400408}}},
+		{0x400548, [][2]uint64{{0x400500, 0x400549}, {0x400400, 0x400408}}},
+		{0x400407, [][2]uint64{{0x400500, 0x400549}, {0x400400, 0x400408}}},
+		{0x400408, nil},
+		{0x400449, nil},
+		{0x4003ff, nil},
+	}
+	testRanges(t, "testdata/ranges.elf", want)
+}
+
+func testRanges(t *testing.T, name string, want []wantRange) {
+	d := elfData(t, name)
+	r := d.Reader()
+	for _, w := range want {
+		entry, err := r.SeekPC(w.pc)
+		if err != nil {
+			if w.ranges != nil {
+				t.Errorf("%s: missing Entry for %#x", name, w.pc)
+			}
+			if err != ErrUnknownPC {
+				t.Errorf("%s: expected ErrUnknownPC for %#x, got %v", name, w.pc, err)
+			}
+			continue
+		}
+
+		ranges, err := d.Ranges(entry)
+		if err != nil {
+			t.Errorf("%s: %v", name, err)
+			continue
+		}
+		if !reflect.DeepEqual(ranges, w.ranges) {
+			t.Errorf("%s: for %#x got %x, expected %x", name, w.pc, ranges, w.ranges)
+		}
+	}
+}
+
+func TestReaderRanges(t *testing.T) {
+	d := elfData(t, "testdata/line-gcc.elf")
+
+	subprograms := []struct {
+		name   string
+		ranges [][2]uint64
+	}{
+		{"f1", [][2]uint64{{0x40059d, 0x4005e7}}},
+		{"main", [][2]uint64{{0x4005e7, 0x400601}}},
+		{"f2", [][2]uint64{{0x400601, 0x400611}}},
+	}
+
+	r := d.Reader()
+	i := 0
+	for entry, err := r.Next(); entry != nil && err == nil; entry, err = r.Next() {
+		if entry.Tag != TagSubprogram {
+			continue
+		}
+
+		if i > len(subprograms) {
+			t.Fatalf("too many subprograms (expected at most %d)", i)
+		}
+
+		if got := entry.Val(AttrName).(string); got != subprograms[i].name {
+			t.Errorf("subprogram %d name is %s, expected %s", i, got, subprograms[i].name)
+		}
+		ranges, err := d.Ranges(entry)
+		if err != nil {
+			t.Errorf("subprogram %d: %v", i, err)
+			continue
+		}
+		if !reflect.DeepEqual(ranges, subprograms[i].ranges) {
+			t.Errorf("subprogram %d ranges are %x, expected %x", i, ranges, subprograms[i].ranges)
+		}
+		i++
+	}
+
+	if i < len(subprograms) {
+		t.Errorf("saw only %d subprograms, expected %d", i, len(subprograms))
+	}
+}

src/debug/dwarf/testdata/ranges.c

@@ -0,0 +1,25 @@
+// gcc -g -O2 -freorder-blocks-and-partition
+
+const char *arr[10000];
+const char *hot = "hot";
+const char *cold = "cold";
+
+__attribute__((noinline))
+void fn(int path) {
+	int i;
+
+	if (path) {
+		for (i = 0; i < sizeof arr / sizeof arr[0]; i++) {
+			arr[i] = hot;
+		}
+	} else {
+		for (i = 0; i < sizeof arr / sizeof arr[0]; i++) {
+			arr[i] = cold;
+		}
+	}
+}
+
+int main(int argc, char *argv[]) {
+	fn(argc);
+	return 0;
+}

src/debug/elf/file.go

@@ -958,7 +958,7 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 	// There are many other DWARF sections, but these
 	// are the ones the debug/dwarf package uses.
 	// Don't bother loading others.
-	var dat = map[string][]byte{"abbrev": nil, "info": nil, "str": nil, "line": nil}
+	var dat = map[string][]byte{"abbrev": nil, "info": nil, "str": nil, "line": nil, "ranges": nil}
 	for i, s := range f.Sections {
 		suffix := ""
 		switch {
@@ -979,7 +979,7 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 		dat[suffix] = b
 	}
 
-	d, err := dwarf.New(dat["abbrev"], nil, nil, dat["info"], dat["line"], nil, nil, dat["str"])
+	d, err := dwarf.New(dat["abbrev"], nil, nil, dat["info"], dat["line"], nil, dat["ranges"], dat["str"])
 	if err != nil {
 		return nil, err
 	}

src/debug/macho/file.go

@@ -474,7 +474,7 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 	// There are many other DWARF sections, but these
 	// are the ones the debug/dwarf package uses.
 	// Don't bother loading others.
-	var names = [...]string{"abbrev", "info", "line", "str"}
+	var names = [...]string{"abbrev", "info", "line", "ranges", "str"}
 	var dat [len(names)][]byte
 	for i, name := range names {
 		name = "__debug_" + name
@@ -489,8 +489,8 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 		dat[i] = b
 	}
 
-	abbrev, info, line, str := dat[0], dat[1], dat[2], dat[3]
-	return dwarf.New(abbrev, nil, nil, info, line, nil, nil, str)
+	abbrev, info, line, ranges, str := dat[0], dat[1], dat[2], dat[3], dat[4]
+	return dwarf.New(abbrev, nil, nil, info, line, nil, ranges, str)
 }
 
 // ImportedSymbols returns the names of all symbols

src/debug/pe/file.go

@@ -298,7 +298,7 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 	// There are many other DWARF sections, but these
 	// are the ones the debug/dwarf package uses.
 	// Don't bother loading others.
-	var names = [...]string{"abbrev", "info", "line", "str"}
+	var names = [...]string{"abbrev", "info", "line", "ranges", "str"}
 	var dat [len(names)][]byte
 	for i, name := range names {
 		name = ".debug_" + name
@@ -316,8 +316,8 @@ func (f *File) DWARF() (*dwarf.Data, error) {
 		dat[i] = b
 	}
 
-	abbrev, info, line, str := dat[0], dat[1], dat[2], dat[3]
-	return dwarf.New(abbrev, nil, nil, info, line, nil, nil, str)
+	abbrev, info, line, ranges, str := dat[0], dat[1], dat[2], dat[3], dat[4]
+	return dwarf.New(abbrev, nil, nil, info, line, nil, ranges, str)
 }
 
 // ImportedSymbols returns the names of all symbols