[dev.debug] cmd/compile: better DWARF with optimizations on

Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.

After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.

There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:

  type myint int
  var a int
  b := myint(int)
and
  b := (*uint64)(unsafe.Pointer(a))

don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.

Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.

A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.

TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.

go build -toolexec 'toolstash -cmp' -a std succeeds.

With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after:  06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean  /tmp/220352263 /tmp/621364410
completed   15 of   15, estimated time remaining 0s (eta 3:52PM)
name        old time/op       new time/op       delta
Template          199ms ± 3%        198ms ± 2%     ~     (p=0.400 n=15+14)
Unicode          96.6ms ± 5%       96.4ms ± 5%     ~     (p=0.838 n=15+15)
GoTypes           653ms ± 2%        647ms ± 2%     ~     (p=0.102 n=15+14)
Flate             133ms ± 6%        129ms ± 3%   -2.62%  (p=0.041 n=15+15)
GoParser          164ms ± 5%        159ms ± 3%   -3.05%  (p=0.000 n=15+15)
Reflect           428ms ± 4%        422ms ± 3%     ~     (p=0.156 n=15+13)
Tar               123ms ±10%        124ms ± 8%     ~     (p=0.461 n=15+15)
XML               228ms ± 3%        224ms ± 3%   -1.57%  (p=0.045 n=15+15)
[Geo mean]        206ms             377ms       +82.86%

name        old user-time/op  new user-time/op  delta
Template          292ms ±10%        301ms ±12%     ~     (p=0.189 n=15+15)
Unicode           166ms ±37%        158ms ±14%     ~     (p=0.418 n=15+14)
GoTypes           962ms ± 6%        963ms ± 7%     ~     (p=0.976 n=15+15)
Flate             207ms ±19%        200ms ±14%     ~     (p=0.345 n=14+15)
GoParser          246ms ±22%        240ms ±15%     ~     (p=0.587 n=15+15)
Reflect           611ms ±13%        587ms ±14%     ~     (p=0.085 n=15+13)
Tar               211ms ±12%        217ms ±14%     ~     (p=0.355 n=14+15)
XML               335ms ±15%        320ms ±18%     ~     (p=0.169 n=15+15)
[Geo mean]        317ms             583ms       +83.72%

name        old alloc/op      new alloc/op      delta
Template         40.2MB ± 0%       40.2MB ± 0%   -0.15%  (p=0.000 n=14+15)
Unicode          29.2MB ± 0%       29.3MB ± 0%     ~     (p=0.624 n=15+15)
GoTypes           114MB ± 0%        114MB ± 0%   -0.15%  (p=0.000 n=15+14)
Flate            25.7MB ± 0%       25.6MB ± 0%   -0.18%  (p=0.000 n=13+15)
GoParser         32.2MB ± 0%       32.2MB ± 0%   -0.14%  (p=0.003 n=15+15)
Reflect          77.8MB ± 0%       77.9MB ± 0%     ~     (p=0.061 n=15+15)
Tar              27.1MB ± 0%       27.0MB ± 0%   -0.11%  (p=0.029 n=15+15)
XML              42.7MB ± 0%       42.5MB ± 0%   -0.29%  (p=0.000 n=15+15)
[Geo mean]       42.1MB            75.0MB       +78.05%

name        old allocs/op     new allocs/op     delta
Template           402k ± 1%         398k ± 0%   -0.91%  (p=0.000 n=15+15)
Unicode            344k ± 1%         344k ± 0%     ~     (p=0.715 n=15+14)
GoTypes           1.18M ± 0%        1.17M ± 0%   -0.91%  (p=0.000 n=15+14)
Flate              243k ± 0%         240k ± 1%   -1.05%  (p=0.000 n=13+15)
GoParser           327k ± 1%         324k ± 1%   -0.96%  (p=0.000 n=15+15)
Reflect            984k ± 1%         982k ± 0%     ~     (p=0.050 n=15+15)
Tar                261k ± 1%         259k ± 1%   -0.77%  (p=0.000 n=15+15)
XML                411k ± 0%         404k ± 1%   -1.55%  (p=0.000 n=15+15)
[Geo mean]         439k              755k       +72.01%

name        old text-bytes    new text-bytes    delta
HelloSize         694kB ± 0%        694kB ± 0%   -0.00%  (p=0.000 n=15+15)

name        old data-bytes    new data-bytes    delta
HelloSize        5.55kB ± 0%       5.55kB ± 0%     ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize         133kB ± 0%        133kB ± 0%     ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.04MB ± 0%       1.04MB ± 0%     ~     (all equal)

Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

src/cmd/compile/fmt_test.go

@@ -571,9 +571,16 @@ var knownFormats = map[string]string{
 	"*cmd/compile/internal/ssa.Block %s":              "",
 	"*cmd/compile/internal/ssa.Block %v":              "",
 	"*cmd/compile/internal/ssa.Func %s":               "",
+	"*cmd/compile/internal/ssa.Func %v":               "",
+	"*cmd/compile/internal/ssa.FuncDebug %v":          "",
+	"*cmd/compile/internal/ssa.LocalSlot %+v":         "",
+	"*cmd/compile/internal/ssa.LocalSlot %v":          "",
+	"*cmd/compile/internal/ssa.Register %v":           "",
 	"*cmd/compile/internal/ssa.SparseTreeNode %v":     "",
 	"*cmd/compile/internal/ssa.Value %s":              "",
 	"*cmd/compile/internal/ssa.Value %v":              "",
+	"*cmd/compile/internal/ssa.VarLoc %+v":            "",
+	"*cmd/compile/internal/ssa.VarLoc %v":             "",
 	"*cmd/compile/internal/ssa.sparseTreeMapEntry %v": "",
 	"*cmd/compile/internal/types.Field %p":            "",
 	"*cmd/compile/internal/types.Field %v":            "",
@@ -592,6 +599,7 @@ var knownFormats = map[string]string{
 	"*cmd/compile/internal/types.Type %p":             "",
 	"*cmd/compile/internal/types.Type %s":             "",
 	"*cmd/compile/internal/types.Type %v":             "",
+	"*cmd/internal/dwarf.Location %#v":                "",
 	"*cmd/internal/obj.Addr %v":                       "",
 	"*cmd/internal/obj.LSym %v":                       "",
 	"*cmd/internal/obj.Prog %s":                       "",
@@ -600,17 +608,21 @@ var knownFormats = map[string]string{
 	"[16]byte %x":                                     "",
 	"[]*cmd/compile/internal/gc.Node %v":              "",
 	"[]*cmd/compile/internal/gc.Sig %#v":              "",
+	"[]*cmd/compile/internal/ssa.Block %+v":           "",
 	"[]*cmd/compile/internal/ssa.Value %v":            "",
+	"[][]cmd/compile/internal/ssa.SlotID %v":          "",
 	"[]byte %s":                                       "",
 	"[]byte %x":                                       "",
 	"[]cmd/compile/internal/ssa.Edge %v":              "",
 	"[]cmd/compile/internal/ssa.ID %v":                "",
+	"[]cmd/compile/internal/ssa.VarLocList %v":        "",
 	"[]string %v":                                     "",
 	"bool %v":                                         "",
 	"byte %08b":                                       "",
 	"byte %c":                                         "",
 	"cmd/compile/internal/arm.shift %d":               "",
 	"cmd/compile/internal/gc.Class %d":                "",
+	"cmd/compile/internal/gc.Class %v":                "",
 	"cmd/compile/internal/gc.Ctype %d":                "",
 	"cmd/compile/internal/gc.Ctype %v":                "",
 	"cmd/compile/internal/gc.Level %d":                "",
@@ -630,11 +642,13 @@ var knownFormats = map[string]string{
 	"cmd/compile/internal/ssa.Edge %v":                "",
 	"cmd/compile/internal/ssa.GCNode %v":              "",
 	"cmd/compile/internal/ssa.ID %d":                  "",
+	"cmd/compile/internal/ssa.ID %v":                  "",
 	"cmd/compile/internal/ssa.LocalSlot %v":           "",
 	"cmd/compile/internal/ssa.Location %v":            "",
 	"cmd/compile/internal/ssa.Op %s":                  "",
 	"cmd/compile/internal/ssa.Op %v":                  "",
 	"cmd/compile/internal/ssa.ValAndOff %s":           "",
+	"cmd/compile/internal/ssa.VarLocList %v":          "",
 	"cmd/compile/internal/ssa.rbrank %d":              "",
 	"cmd/compile/internal/ssa.regMask %d":             "",
 	"cmd/compile/internal/ssa.register %d":            "",
@@ -648,6 +662,7 @@ var knownFormats = map[string]string{
 	"cmd/compile/internal/types.EType %d":             "",
 	"cmd/compile/internal/types.EType %s":             "",
 	"cmd/compile/internal/types.EType %v":             "",
+	"cmd/internal/dwarf.Location %#v":                 "",
 	"cmd/internal/src.Pos %s":                         "",
 	"cmd/internal/src.Pos %v":                         "",
 	"error %v":                                        "",

src/cmd/compile/internal/gc/main.go

@@ -44,6 +44,7 @@ var (
 	Debug_vlog         bool
 	Debug_wb           int
 	Debug_pctab        string
+	Debug_locationlist int
 )
 
 // Debug arguments.
@@ -69,6 +70,7 @@ var debugtab = []struct {
 	{"wb", "print information about write barriers", &Debug_wb},
 	{"export", "print export data", &Debug_export},
 	{"pctab", "print named pc-value table", &Debug_pctab},
+	{"locationlists", "print information about DWARF location list creation", &Debug_locationlist},
 }
 
 const debugHelpHeader = `usage: -d arg[,arg]* and arg is <key>[=<value>]
@@ -192,6 +194,7 @@ func Main(archInit func(*Arch)) {
 	flag.BoolVar(&pure_go, "complete", false, "compiling complete package (no C or assembly)")
 	flag.StringVar(&debugstr, "d", "", "print debug information about items in `list`; try -d help")
 	flag.BoolVar(&flagDWARF, "dwarf", true, "generate DWARF symbols")
+	flag.BoolVar(&Ctxt.Flag_locationlists, "dwarflocationlists", false, "add location lists to DWARF in optimized mode")
 	objabi.Flagcount("e", "no limit on number of errors reported", &Debug['e'])
 	objabi.Flagcount("f", "debug stack frames", &Debug['f'])
 	objabi.Flagcount("h", "halt on error", &Debug['h'])
@@ -298,6 +301,9 @@ func Main(archInit func(*Arch)) {
 	if nBackendWorkers > 1 && !concurrentBackendAllowed() {
 		log.Fatalf("cannot use concurrent backend compilation with provided flags; invoked as %v", os.Args)
 	}
+	if Ctxt.Flag_locationlists && len(Ctxt.Arch.DWARFRegisters) == 0 {
+		log.Fatalf("location lists requested but register mapping not available on %v", Ctxt.Arch.Name)
+	}
 
 	// parse -d argument
 	if debugstr != "" {
@@ -383,7 +389,7 @@ func Main(archInit func(*Arch)) {
 		Debug['l'] = 1 - Debug['l']
 	}
 
-	trackScopes = flagDWARF && Debug['l'] == 0 && Debug['N'] != 0
+	trackScopes = flagDWARF && ((Debug['l'] == 0 && Debug['N'] != 0) || Ctxt.Flag_locationlists)
 
 	Widthptr = thearch.LinkArch.PtrSize
 	Widthreg = thearch.LinkArch.RegSize

src/cmd/compile/internal/gc/sizeof_test.go

@@ -22,7 +22,7 @@ func TestSizeof(t *testing.T) {
 		_32bit uintptr     // size on 32bit platforms
 		_64bit uintptr     // size on 64bit platforms
 	}{
-		{Func{}, 124, 216},
+		{Func{}, 128, 224},
 		{Name{}, 36, 56},
 		{Param{}, 28, 56},
 		{Node{}, 76, 128},

src/cmd/compile/internal/gc/ssa.go

@@ -4384,6 +4384,7 @@ func genssa(f *ssa.Func, pp *Progs) {
 	s.pp = pp
 	var progToValue map[*obj.Prog]*ssa.Value
 	var progToBlock map[*obj.Prog]*ssa.Block
+	var valueToProg []*obj.Prog
 	var logProgs = e.log
 	if logProgs {
 		progToValue = make(map[*obj.Prog]*ssa.Value, f.NumValues())
@@ -4398,6 +4399,11 @@ func genssa(f *ssa.Func, pp *Progs) {
 
 	s.ScratchFpMem = e.scratchFpMem
 
+	logLocationLists := Debug_locationlist != 0
+	if Ctxt.Flag_locationlists {
+		e.curfn.Func.DebugInfo = ssa.BuildFuncDebug(f, logLocationLists)
+		valueToProg = make([]*obj.Prog, f.NumValues())
+	}
 	// Emit basic blocks
 	for i, b := range f.Blocks {
 		s.bstart[b.ID] = s.pp.next
@@ -4438,12 +4444,16 @@ func genssa(f *ssa.Func, pp *Progs) {
 				}
 			case ssa.OpPhi:
 				CheckLoweredPhi(v)
-
+			case ssa.OpRegKill:
+				// nothing to do
 			default:
 				// let the backend handle it
 				thearch.SSAGenValue(&s, v)
 			}
 
+			if Ctxt.Flag_locationlists {
+				valueToProg[v.ID] = x
+			}
 			if logProgs {
 				for ; x != s.pp.next; x = x.Link {
 					progToValue[x] = v
@@ -4469,6 +4479,22 @@ func genssa(f *ssa.Func, pp *Progs) {
 		}
 	}
 
+	if Ctxt.Flag_locationlists {
+		for _, locList := range e.curfn.Func.DebugInfo.Variables {
+			for _, loc := range locList.Locations {
+				loc.StartProg = valueToProg[loc.Start.ID]
+				if loc.End == nil {
+					Fatalf("empty loc %v compiling %v", loc, f.Name)
+				}
+				loc.EndProg = valueToProg[loc.End.ID]
+				if !logLocationLists {
+					loc.Start = nil
+					loc.End = nil
+				}
+			}
+		}
+	}
+
 	// Resolve branches
 	for _, br := range s.Branches {
 		br.P.To.Val = s.bstart[br.B.ID]

src/cmd/compile/internal/gc/syntax.go

@@ -7,6 +7,7 @@
 package gc
 
 import (
+	"cmd/compile/internal/ssa"
 	"cmd/compile/internal/syntax"
 	"cmd/compile/internal/types"
 	"cmd/internal/obj"
@@ -369,6 +370,7 @@ type Func struct {
 	Closgen    int
 	Outerfunc  *Node // outer function (for closure)
 	FieldTrack map[*types.Sym]struct{}
+	DebugInfo  *ssa.FuncDebug
 	Ntype      *Node // signature
 	Top        int   // top context (Ecall, Eproc, etc)
 	Closure    *Node // OCLOSURE <-> ODCLFUNC

src/cmd/compile/internal/ssa/cache.go

@@ -14,6 +14,11 @@ type Cache struct {
 	blocks [200]Block
 	locs   [2000]Location
 
+	// Storage for DWARF variable locations. Lazily allocated
+	// since location lists are off by default.
+	varLocs   []VarLoc
+	curVarLoc int
+
 	// Reusable stackAllocState.
 	// See stackalloc.go's {new,put}StackAllocState.
 	stackAllocState *stackAllocState
@@ -38,4 +43,21 @@ func (c *Cache) Reset() {
 	for i := range xl {
 		xl[i] = nil
 	}
+	xvl := c.varLocs[:c.curVarLoc]
+	for i := range xvl {
+		xvl[i] = VarLoc{}
+	}
+	c.curVarLoc = 0
+}
+
+func (c *Cache) NewVarLoc() *VarLoc {
+	if c.varLocs == nil {
+		c.varLocs = make([]VarLoc, 4000)
+	}
+	if c.curVarLoc == len(c.varLocs) {
+		return &VarLoc{}
+	}
+	vl := &c.varLocs[c.curVarLoc]
+	c.curVarLoc++
+	return vl
 }

src/cmd/compile/internal/ssa/gen/genericOps.go

@@ -417,6 +417,7 @@ var genericOps = []opData{
 	{name: "VarKill", argLength: 1, aux: "Sym", symEffect: "None"},            // aux is a *gc.Node of a variable that is known to be dead.  arg0=mem, returns mem
 	{name: "VarLive", argLength: 1, aux: "Sym", symEffect: "None"},            // aux is a *gc.Node of a variable that must be kept live.  arg0=mem, returns mem
 	{name: "KeepAlive", argLength: 2, typ: "Mem"},                             // arg[0] is a value that must be kept alive until this mark.  arg[1]=mem, returns mem
+	{name: "RegKill"},                                                         // regalloc has determined that the value in this register is dead
 
 	// Ops for breaking 64-bit operations on 32-bit architectures
 	{name: "Int64Make", argLength: 2, typ: "UInt64"}, // arg0=hi, arg1=lo

src/cmd/compile/internal/ssa/html.go

@@ -11,6 +11,7 @@ import (
 	"html"
 	"io"
 	"os"
+	"strings"
 )
 
 type HTMLWriter struct {
@@ -362,6 +363,18 @@ func (v *Value) LongHTML() string {
 	if int(v.ID) < len(r) && r[v.ID] != nil {
 		s += " : " + html.EscapeString(r[v.ID].Name())
 	}
+	var names []string
+	for name, values := range v.Block.Func.NamedValues {
+		for _, value := range values {
+			if value == v {
+				names = append(names, name.Name())
+				break // drop duplicates.
+			}
+		}
+	}
+	if len(names) != 0 {
+		s += " (" + strings.Join(names, ", ") + ")"
+	}
 	s += "</span>"
 	return s
 }

src/cmd/compile/internal/ssa/location.go

@@ -26,6 +26,12 @@ func (r *Register) Name() string {
 	return r.name
 }
 
+// ObjNum returns the register number from cmd/internal/obj/$ARCH that
+// corresponds to this register.
+func (r *Register) ObjNum() int16 {
+	return r.objNum
+}
+
 // A LocalSlot is a location in the stack frame, which identifies and stores
 // part or all of a PPARAM, PPARAMOUT, or PAUTO ONAME node.
 // It can represent a whole variable, part of a larger stack slot, or part of a

src/cmd/compile/internal/ssa/opGen.go

@@ -1897,6 +1897,7 @@ const (
 	OpVarKill
 	OpVarLive
 	OpKeepAlive
+	OpRegKill
 	OpInt64Make
 	OpInt64Hi
 	OpInt64Lo
@@ -22497,6 +22498,11 @@ var opcodeTable = [...]opInfo{
 		argLen:  2,
 		generic: true,
 	},
+	{
+		name:    "RegKill",
+		argLen:  0,
+		generic: true,
+	},
 	{
 		name:    "Int64Make",
 		argLen:  2,

src/cmd/compile/internal/ssa/regalloc.go

@@ -242,6 +242,9 @@ type regAllocState struct {
 	// current state of each (preregalloc) Value
 	values []valState
 
+	// names associated with each Value
+	valueNames [][]LocalSlot
+
 	// ID of SP, SB values
 	sp, sb ID
 
@@ -300,6 +303,13 @@ type startReg struct {
 
 // freeReg frees up register r. Any current user of r is kicked out.
 func (s *regAllocState) freeReg(r register) {
+	s.freeOrResetReg(r, false)
+}
+
+// freeOrResetReg frees up register r. Any current user of r is kicked out.
+// resetting indicates that the operation is only for bookkeeping,
+// e.g. when clearing out state upon entry to a new block.
+func (s *regAllocState) freeOrResetReg(r register, resetting bool) {
 	v := s.regs[r].v
 	if v == nil {
 		s.f.Fatalf("tried to free an already free register %d\n", r)
@@ -309,6 +319,16 @@ func (s *regAllocState) freeReg(r register) {
 	if s.f.pass.debug > regDebug {
 		fmt.Printf("freeReg %s (dump %s/%s)\n", s.registers[r].Name(), v, s.regs[r].c)
 	}
+	if !resetting && s.f.Config.ctxt.Flag_locationlists && len(s.valueNames[v.ID]) != 0 {
+		kill := s.curBlock.NewValue0(src.NoXPos, OpRegKill, types.TypeVoid)
+		for int(kill.ID) >= len(s.orig) {
+			s.orig = append(s.orig, nil)
+		}
+		for _, name := range s.valueNames[v.ID] {
+			s.f.NamedValues[name] = append(s.f.NamedValues[name], kill)
+		}
+		s.f.setHome(kill, &s.registers[r])
+	}
 	s.regs[r] = regState{}
 	s.values[v.ID].regs &^= regMask(1) << r
 	s.used &^= regMask(1) << r
@@ -599,6 +619,17 @@ func (s *regAllocState) init(f *Func) {
 	s.values = make([]valState, f.NumValues())
 	s.orig = make([]*Value, f.NumValues())
 	s.copies = make(map[*Value]bool)
+	if s.f.Config.ctxt.Flag_locationlists {
+		s.valueNames = make([][]LocalSlot, f.NumValues())
+		for slot, values := range f.NamedValues {
+			if isSynthetic(&slot) {
+				continue
+			}
+			for _, value := range values {
+				s.valueNames[value.ID] = append(s.valueNames[value.ID], slot)
+			}
+		}
+	}
 	for _, b := range f.Blocks {
 		for _, v := range b.Values {
 			if !v.Type.IsMemory() && !v.Type.IsVoid() && !v.Type.IsFlags() && !v.Type.IsTuple() {
@@ -692,7 +723,9 @@ func (s *regAllocState) liveAfterCurrentInstruction(v *Value) bool {
 
 // Sets the state of the registers to that encoded in regs.
 func (s *regAllocState) setState(regs []endReg) {
-	s.freeRegs(s.used)
+	for s.used != 0 {
+		s.freeOrResetReg(pickReg(s.used), true)
+	}
 	for _, x := range regs {
 		s.assignReg(x.r, x.v, x.c)
 	}
@@ -735,6 +768,9 @@ func (s *regAllocState) regalloc(f *Func) {
 	}
 
 	for _, b := range f.Blocks {
+		if s.f.pass.debug > regDebug {
+			fmt.Printf("Begin processing block %v\n", b)
+		}
 		s.curBlock = b
 
 		// Initialize regValLiveSet and uses fields for this block.
@@ -830,9 +866,6 @@ func (s *regAllocState) regalloc(f *Func) {
 			// This is the complicated case. We have more than one predecessor,
 			// which means we may have Phi ops.
 
-			// Copy phi ops into new schedule.
-			b.Values = append(b.Values, phis...)
-
 			// Start with the final register state of the primary predecessor
 			idx := s.primary[b.ID]
 			if idx < 0 {
@@ -910,6 +943,9 @@ func (s *regAllocState) regalloc(f *Func) {
 				}
 			}
 
+			// Copy phi ops into new schedule.
+			b.Values = append(b.Values, phis...)
+
 			// Third pass - pick registers for phis whose inputs
 			// were not in a register.
 			for i, v := range phis {
@@ -1005,7 +1041,7 @@ func (s *regAllocState) regalloc(f *Func) {
 			pidx := e.i
 			for _, v := range succ.Values {
 				if v.Op != OpPhi {
-					break
+					continue
 				}
 				if !s.values[v.ID].needReg {
 					continue
@@ -1565,6 +1601,9 @@ func (s *regAllocState) placeSpills() {
 	for _, b := range f.Blocks {
 		var m regMask
 		for _, v := range b.Values {
+			if v.Op == OpRegKill {
+				continue
+			}
 			if v.Op != OpPhi {
 				break
 			}
@@ -1675,7 +1714,7 @@ func (s *regAllocState) placeSpills() {
 	for _, b := range f.Blocks {
 		nphi := 0
 		for _, v := range b.Values {
-			if v.Op != OpPhi {
+			if v.Op != OpRegKill && v.Op != OpPhi {
 				break
 			}
 			nphi++
@@ -1800,6 +1839,9 @@ func (e *edgeState) setup(idx int, srcReg []endReg, dstReg []startReg, stacklive
 	}
 	// Phis need their args to end up in a specific location.
 	for _, v := range e.b.Values {
+		if v.Op == OpRegKill {
+			continue
+		}
 		if v.Op != OpPhi {
 			break
 		}
@@ -1878,6 +1920,7 @@ func (e *edgeState) process() {
 		if e.s.f.pass.debug > regDebug {
 			fmt.Printf("breaking cycle with v%d in %s:%s\n", vid, loc.Name(), c)
 		}
+		e.erase(r)
 		if _, isReg := loc.(*Register); isReg {
 			c = e.p.NewValue1(d.pos, OpCopy, c.Type, c)
 		} else {
@@ -1943,6 +1986,18 @@ func (e *edgeState) processDest(loc Location, vid ID, splice **Value, pos src.XP
 		}
 	}
 	_, dstReg := loc.(*Register)
+
+	// Pre-clobber destination. This avoids the
+	// following situation:
+	//   - v is currently held in R0 and stacktmp0.
+	//   - We want to copy stacktmp1 to stacktmp0.
+	//   - We choose R0 as the temporary register.
+	// During the copy, both R0 and stacktmp0 are
+	// clobbered, losing both copies of v. Oops!
+	// Erasing the destination early means R0 will not
+	// be chosen as the temp register, as it will then
+	// be the last copy of v.
+	e.erase(loc)
 	var x *Value
 	if c == nil {
 		if !e.s.values[vid].rematerializeable {
@@ -1953,8 +2008,8 @@ func (e *edgeState) processDest(loc Location, vid ID, splice **Value, pos src.XP
 		} else {
 			// Rematerialize into stack slot. Need a free
 			// register to accomplish this.
-			e.erase(loc) // see pre-clobber comment below
 			r := e.findRegFor(v.Type)
+			e.erase(r)
 			x = v.copyIntoNoXPos(e.p)
 			e.set(r, vid, x, false, pos)
 			// Make sure we spill with the size of the slot, not the
@@ -1976,20 +2031,8 @@ func (e *edgeState) processDest(loc Location, vid ID, splice **Value, pos src.XP
 				x = e.p.NewValue1(pos, OpLoadReg, c.Type, c)
 			} else {
 				// mem->mem. Use temp register.
-
-				// Pre-clobber destination. This avoids the
-				// following situation:
-				//   - v is currently held in R0 and stacktmp0.
-				//   - We want to copy stacktmp1 to stacktmp0.
-				//   - We choose R0 as the temporary register.
-				// During the copy, both R0 and stacktmp0 are
-				// clobbered, losing both copies of v. Oops!
-				// Erasing the destination early means R0 will not
-				// be chosen as the temp register, as it will then
-				// be the last copy of v.
-				e.erase(loc)
-
 				r := e.findRegFor(c.Type)
+				e.erase(r)
 				t := e.p.NewValue1(pos, OpLoadReg, c.Type, c)
 				e.set(r, vid, t, false, pos)
 				x = e.p.NewValue1(pos, OpStoreReg, loc.(LocalSlot).Type, t)
@@ -2008,7 +2051,6 @@ func (e *edgeState) processDest(loc Location, vid ID, splice **Value, pos src.XP
 // set changes the contents of location loc to hold the given value and its cached representative.
 func (e *edgeState) set(loc Location, vid ID, c *Value, final bool, pos src.XPos) {
 	e.s.f.setHome(c, loc)
-	e.erase(loc)
 	e.contents[loc] = contentRecord{vid, c, final, pos}
 	a := e.cache[vid]
 	if len(a) == 0 {
@@ -2059,6 +2101,16 @@ func (e *edgeState) erase(loc Location) {
 				fmt.Printf("v%d no longer available in %s:%s\n", vid, loc.Name(), c)
 			}
 			a[i], a = a[len(a)-1], a[:len(a)-1]
+			if e.s.f.Config.ctxt.Flag_locationlists {
+				if _, isReg := loc.(*Register); isReg && int(c.ID) < len(e.s.valueNames) && len(e.s.valueNames[c.ID]) != 0 {
+					kill := e.p.NewValue0(src.NoXPos, OpRegKill, types.TypeVoid)
+					e.s.f.setHome(kill, loc)
+					for _, name := range e.s.valueNames[c.ID] {
+						e.s.f.NamedValues[name] = append(e.s.f.NamedValues[name], kill)
+					}
+				}
+			}
+
 			break
 		}
 	}

src/cmd/compile/internal/ssa/value.go

@@ -10,6 +10,7 @@ import (
 	"cmd/internal/src"
 	"fmt"
 	"math"
+	"strings"
 )
 
 // A Value represents a value in the SSA representation of the program.
@@ -98,7 +99,7 @@ func (v *Value) AuxValAndOff() ValAndOff {
 	return ValAndOff(v.AuxInt)
 }
 
-// long form print.  v# = opcode <type> [aux] args [: reg]
+// long form print.  v# = opcode <type> [aux] args [: reg] (names)
 func (v *Value) LongString() string {
 	s := fmt.Sprintf("v%d = %s", v.ID, v.Op)
 	s += " <" + v.Type.String() + ">"
@@ -110,6 +111,18 @@ func (v *Value) LongString() string {
 	if int(v.ID) < len(r) && r[v.ID] != nil {
 		s += " : " + r[v.ID].Name()
 	}
+	var names []string
+	for name, values := range v.Block.Func.NamedValues {
+		for _, value := range values {
+			if value == v {
+				names = append(names, name.Name())
+				break // drop duplicates.
+			}
+		}
+	}
+	if len(names) != 0 {
+		s += " (" + strings.Join(names, ", ") + ")"
+	}
 	return s
 }
 

src/cmd/internal/dwarf/dwarf.go

@@ -15,6 +15,9 @@ import (
 // InfoPrefix is the prefix for all the symbols containing DWARF info entries.
 const InfoPrefix = "go.info."
 
+// RangePrefix is the prefix for all the symbols containing DWARF location lists.
+const LocPrefix = "go.loc."
+
 // RangePrefix is the prefix for all the symbols containing DWARF range lists.
 const RangePrefix = "go.range."
 
@@ -23,13 +26,31 @@ type Sym interface {
 	Len() int64
 }
 
+// A Location represents a variable's location at a particular PC range.
+// It becomes a location list entry in the DWARF.
+type Location struct {
+	StartPC, EndPC int64
+	Pieces         []Piece
+}
+
+// A Piece represents the location of a particular part of a variable.
+// It becomes part of a location list entry (a DW_OP_piece) in the DWARF.
+type Piece struct {
+	Length      int64
+	StackOffset int32
+	RegNum      int16
+	Missing     bool
+	OnStack     bool // if true, RegNum is unset.
+}
+
 // A Var represents a local variable or a function parameter.
 type Var struct {
-	Name        string
-	Abbrev      int // Either DW_ABRV_AUTO or DW_ABRV_PARAM
-	StackOffset int32
-	Scope       int32
-	Type        Sym
+	Name         string
+	Abbrev       int // Either DW_ABRV_AUTO or DW_ABRV_PARAM
+	StackOffset  int32
+	LocationList []Location
+	Scope        int32
+	Type         Sym
 }
 
 // A Scope represents a lexical scope. All variables declared within a
@@ -205,7 +226,7 @@ const (
 )
 
 // Index into the abbrevs table below.
-// Keep in sync with ispubname() and ispubtype() below.
+// Keep in sync with ispubname() and ispubtype() in ld/dwarf.go.
 // ispubtype considers >= NULLTYPE public
 const (
 	DW_ABRV_NULL = iota
@@ -709,31 +730,30 @@ func HasChildren(die *DWDie) bool {
 
 // PutFunc writes a DIE for a function to s.
 // It also writes child DIEs for each variable in vars.
-func PutFunc(ctxt Context, s, ranges Sym, name string, external bool, startPC Sym, size int64, scopes []Scope) error {
-	Uleb128put(ctxt, s, DW_ABRV_FUNCTION)
-	putattr(ctxt, s, DW_ABRV_FUNCTION, DW_FORM_string, DW_CLS_STRING, int64(len(name)), name)
-	putattr(ctxt, s, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, 0, startPC)
-	putattr(ctxt, s, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, size, startPC)
-	putattr(ctxt, s, DW_ABRV_FUNCTION, DW_FORM_block1, DW_CLS_BLOCK, 1, []byte{DW_OP_call_frame_cfa})
+func PutFunc(ctxt Context, info, loc, ranges Sym, name string, external bool, startPC Sym, size int64, scopes []Scope) error {
+	Uleb128put(ctxt, info, DW_ABRV_FUNCTION)
+	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_string, DW_CLS_STRING, int64(len(name)), name)
+	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, 0, startPC)
+	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, size, startPC)
+	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_block1, DW_CLS_BLOCK, 1, []byte{DW_OP_call_frame_cfa})
 	var ev int64
 	if external {
 		ev = 1
 	}
-	putattr(ctxt, s, DW_ABRV_FUNCTION, DW_FORM_flag, DW_CLS_FLAG, ev, 0)
+	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_flag, DW_CLS_FLAG, ev, 0)
 	if len(scopes) > 0 {
 		var encbuf [20]byte
-		if putscope(ctxt, s, ranges, startPC, 0, scopes, encbuf[:0]) < int32(len(scopes)) {
+		if putscope(ctxt, info, loc, ranges, startPC, 0, scopes, encbuf[:0]) < int32(len(scopes)) {
 			return errors.New("multiple toplevel scopes")
 		}
 	}
-
-	Uleb128put(ctxt, s, 0)
+	Uleb128put(ctxt, info, 0)
 	return nil
 }
 
-func putscope(ctxt Context, s, ranges Sym, startPC Sym, curscope int32, scopes []Scope, encbuf []byte) int32 {
+func putscope(ctxt Context, info, loc, ranges, startPC Sym, curscope int32, scopes []Scope, encbuf []byte) int32 {
 	for _, v := range scopes[curscope].Vars {
-		putvar(ctxt, s, v, encbuf)
+		putvar(ctxt, info, loc, v, startPC, encbuf)
 	}
 	this := curscope
 	curscope++
@@ -744,12 +764,12 @@ func putscope(ctxt Context, s, ranges Sym, startPC Sym, curscope int32, scopes [
 		}
 
 		if len(scope.Ranges) == 1 {
-			Uleb128put(ctxt, s, DW_ABRV_LEXICAL_BLOCK_SIMPLE)
-			putattr(ctxt, s, DW_ABRV_LEXICAL_BLOCK_SIMPLE, DW_FORM_addr, DW_CLS_ADDRESS, scope.Ranges[0].Start, startPC)
-			putattr(ctxt, s, DW_ABRV_LEXICAL_BLOCK_SIMPLE, DW_FORM_addr, DW_CLS_ADDRESS, scope.Ranges[0].End, startPC)
+			Uleb128put(ctxt, info, DW_ABRV_LEXICAL_BLOCK_SIMPLE)
+			putattr(ctxt, info, DW_ABRV_LEXICAL_BLOCK_SIMPLE, DW_FORM_addr, DW_CLS_ADDRESS, scope.Ranges[0].Start, startPC)
+			putattr(ctxt, info, DW_ABRV_LEXICAL_BLOCK_SIMPLE, DW_FORM_addr, DW_CLS_ADDRESS, scope.Ranges[0].End, startPC)
 		} else {
-			Uleb128put(ctxt, s, DW_ABRV_LEXICAL_BLOCK_RANGES)
-			putattr(ctxt, s, DW_ABRV_LEXICAL_BLOCK_RANGES, DW_FORM_sec_offset, DW_CLS_PTR, ranges.Len(), ranges)
+			Uleb128put(ctxt, info, DW_ABRV_LEXICAL_BLOCK_RANGES)
+			putattr(ctxt, info, DW_ABRV_LEXICAL_BLOCK_RANGES, DW_FORM_sec_offset, DW_CLS_PTR, ranges.Len(), ranges)
 
 			ctxt.AddAddress(ranges, nil, -1)
 			ctxt.AddAddress(ranges, startPC, 0)
@@ -761,26 +781,66 @@ func putscope(ctxt Context, s, ranges Sym, startPC Sym, curscope int32, scopes [
 			ctxt.AddAddress(ranges, nil, 0)
 		}
 
-		curscope = putscope(ctxt, s, ranges, startPC, curscope, scopes, encbuf)
+		curscope = putscope(ctxt, info, loc, ranges, startPC, curscope, scopes, encbuf)
 
-		Uleb128put(ctxt, s, 0)
+		Uleb128put(ctxt, info, 0)
 	}
 	return curscope
 }
 
-func putvar(ctxt Context, s Sym, v *Var, encbuf []byte) {
+func putvar(ctxt Context, info, loc Sym, v *Var, startPC Sym, encbuf []byte) {
 	n := v.Name
 
-	Uleb128put(ctxt, s, int64(v.Abbrev))
-	putattr(ctxt, s, v.Abbrev, DW_FORM_string, DW_CLS_STRING, int64(len(n)), n)
-	loc := append(encbuf[:0], DW_OP_call_frame_cfa)
-	if v.StackOffset != 0 {
-		loc = append(loc, DW_OP_consts)
-		loc = AppendSleb128(loc, int64(v.StackOffset))
-		loc = append(loc, DW_OP_plus)
+	Uleb128put(ctxt, info, int64(v.Abbrev))
+	putattr(ctxt, info, v.Abbrev, DW_FORM_string, DW_CLS_STRING, int64(len(n)), n)
+	if v.Abbrev == DW_ABRV_AUTO_LOCLIST || v.Abbrev == DW_ABRV_PARAM_LOCLIST {
+		putattr(ctxt, info, v.Abbrev, DW_FORM_sec_offset, DW_CLS_PTR, int64(loc.Len()), loc)
+		addLocList(ctxt, loc, startPC, v, encbuf)
+	} else {
+		loc := append(encbuf[:0], DW_OP_call_frame_cfa)
+		if v.StackOffset != 0 {
+			loc = append(loc, DW_OP_consts)
+			loc = AppendSleb128(loc, int64(v.StackOffset))
+			loc = append(loc, DW_OP_plus)
+		}
+		putattr(ctxt, info, v.Abbrev, DW_FORM_block1, DW_CLS_BLOCK, int64(len(loc)), loc)
+	}
+	putattr(ctxt, info, v.Abbrev, DW_FORM_ref_addr, DW_CLS_REFERENCE, 0, v.Type)
+}
+
+func addLocList(ctxt Context, listSym, startPC Sym, v *Var, encbuf []byte) {
+	// Base address entry: max ptr followed by the base address.
+	ctxt.AddInt(listSym, ctxt.PtrSize(), ^0)
+	ctxt.AddAddress(listSym, startPC, 0)
+	for _, entry := range v.LocationList {
+		ctxt.AddInt(listSym, ctxt.PtrSize(), entry.StartPC)
+		ctxt.AddInt(listSym, ctxt.PtrSize(), entry.EndPC)
+		locBuf := encbuf[:0]
+		for _, piece := range entry.Pieces {
+			if !piece.Missing {
+				if piece.OnStack {
+					locBuf = append(locBuf, DW_OP_fbreg)
+					locBuf = AppendSleb128(locBuf, int64(piece.StackOffset))
+				} else {
+					if piece.RegNum < 32 {
+						locBuf = append(locBuf, DW_OP_reg0+byte(piece.RegNum))
+					} else {
+						locBuf = append(locBuf, DW_OP_regx)
+						locBuf = AppendUleb128(locBuf, uint64(piece.RegNum))
+					}
+				}
+			}
+			if len(entry.Pieces) > 1 {
+				locBuf = append(locBuf, DW_OP_piece)
+				locBuf = AppendUleb128(locBuf, uint64(piece.Length))
+			}
+		}
+		ctxt.AddInt(listSym, 2, int64(len(locBuf)))
+		ctxt.AddBytes(listSym, locBuf)
 	}
-	putattr(ctxt, s, v.Abbrev, DW_FORM_block1, DW_CLS_BLOCK, int64(len(loc)), loc)
-	putattr(ctxt, s, v.Abbrev, DW_FORM_ref_addr, DW_CLS_REFERENCE, 0, v.Type)
+	// End list
+	ctxt.AddInt(listSym, ctxt.PtrSize(), 0)
+	ctxt.AddInt(listSym, ctxt.PtrSize(), 0)
 }
 
 // VarsByOffset attaches the methods of sort.Interface to []*Var,

src/cmd/internal/obj/link.go

@@ -330,7 +330,8 @@ type FuncInfo struct {
 	Autom  []*Auto
 	Pcln   Pcln
 
-	dwarfSym       *LSym
+	dwarfInfoSym   *LSym
+	dwarfLocSym    *LSym
 	dwarfRangesSym *LSym
 
 	GCArgs   LSym
@@ -476,25 +477,26 @@ type Pcdata struct {
 // Link holds the context for writing object code from a compiler
 // to be linker input or for reading that input into the linker.
 type Link struct {
-	Headtype      objabi.HeadType
-	Arch          *LinkArch
-	Debugasm      bool
-	Debugvlog     bool
-	Debugpcln     string
-	Flag_shared   bool
-	Flag_dynlink  bool
-	Flag_optimize bool
-	Bso           *bufio.Writer
-	Pathname      string
-	hashmu        sync.Mutex       // protects hash
-	hash          map[string]*LSym // name -> sym mapping
-	statichash    map[string]*LSym // name -> sym mapping for static syms
-	PosTable      src.PosTable
-	InlTree       InlTree // global inlining tree used by gc/inl.go
-	Imports       []string
-	DiagFunc      func(string, ...interface{})
-	DebugInfo     func(fn *LSym, curfn interface{}) []dwarf.Scope // if non-nil, curfn is a *gc.Node
-	Errors        int
+	Headtype           objabi.HeadType
+	Arch               *LinkArch
+	Debugasm           bool
+	Debugvlog          bool
+	Debugpcln          string
+	Flag_shared        bool
+	Flag_dynlink       bool
+	Flag_optimize      bool
+	Flag_locationlists bool
+	Bso                *bufio.Writer
+	Pathname           string
+	hashmu             sync.Mutex       // protects hash
+	hash               map[string]*LSym // name -> sym mapping
+	statichash         map[string]*LSym // name -> sym mapping for static syms
+	PosTable           src.PosTable
+	InlTree            InlTree // global inlining tree used by gc/inl.go
+	Imports            []string
+	DiagFunc           func(string, ...interface{})
+	DebugInfo          func(fn *LSym, curfn interface{}) []dwarf.Scope // if non-nil, curfn is a *gc.Node
+	Errors             int
 
 	Framepointer_enabled bool
 
@@ -533,9 +535,10 @@ func (ctxt *Link) FixedFrameSize() int64 {
 // LinkArch is the definition of a single architecture.
 type LinkArch struct {
 	*sys.Arch
-	Init       func(*Link)
-	Preprocess func(*Link, *LSym, ProgAlloc)
-	Assemble   func(*Link, *LSym, ProgAlloc)
-	Progedit   func(*Link, *Prog, ProgAlloc)
-	UnaryDst   map[As]bool // Instruction takes one operand, a destination.
+	Init           func(*Link)
+	Preprocess     func(*Link, *LSym, ProgAlloc)
+	Assemble       func(*Link, *LSym, ProgAlloc)
+	Progedit       func(*Link, *Prog, ProgAlloc)
+	UnaryDst       map[As]bool // Instruction takes one operand, a destination.
+	DWARFRegisters map[int16]int16
 }

src/cmd/internal/obj/objfile.go

@@ -465,15 +465,18 @@ func (c dwCtxt) AddSectionOffset(s dwarf.Sym, size int, t interface{}, ofs int64
 }
 
 // dwarfSym returns the DWARF symbols for TEXT symbol.
-func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfRangesSym *LSym) {
+func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfLocSym, dwarfRangesSym *LSym) {
 	if s.Type != objabi.STEXT {
 		ctxt.Diag("dwarfSym of non-TEXT %v", s)
 	}
-	if s.Func.dwarfSym == nil {
-		s.Func.dwarfSym = ctxt.LookupDerived(s, dwarf.InfoPrefix+s.Name)
+	if s.Func.dwarfInfoSym == nil {
+		s.Func.dwarfInfoSym = ctxt.LookupDerived(s, dwarf.InfoPrefix+s.Name)
+		if ctxt.Flag_locationlists {
+			s.Func.dwarfLocSym = ctxt.LookupDerived(s, dwarf.LocPrefix+s.Name)
+		}
 		s.Func.dwarfRangesSym = ctxt.LookupDerived(s, dwarf.RangePrefix+s.Name)
 	}
-	return s.Func.dwarfSym, s.Func.dwarfRangesSym
+	return s.Func.dwarfInfoSym, s.Func.dwarfLocSym, s.Func.dwarfRangesSym
 }
 
 func (s *LSym) Len() int64 {
@@ -483,15 +486,15 @@ func (s *LSym) Len() int64 {
 // populateDWARF fills in the DWARF Debugging Information Entries for TEXT symbol s.
 // The DWARFs symbol must already have been initialized in InitTextSym.
 func (ctxt *Link) populateDWARF(curfn interface{}, s *LSym) {
-	dsym, drsym := ctxt.dwarfSym(s)
-	if dsym.Size != 0 {
+	info, loc, ranges := ctxt.dwarfSym(s)
+	if info.Size != 0 {
 		ctxt.Diag("makeFuncDebugEntry double process %v", s)
 	}
 	var scopes []dwarf.Scope
 	if ctxt.DebugInfo != nil {
 		scopes = ctxt.DebugInfo(s, curfn)
 	}
-	err := dwarf.PutFunc(dwCtxt{ctxt}, dsym, drsym, s.Name, !s.Static(), s, s.Size, scopes)
+	err := dwarf.PutFunc(dwCtxt{ctxt}, info, loc, ranges, s.Name, !s.Static(), s, s.Size, scopes)
 	if err != nil {
 		ctxt.Diag("emitting DWARF for %s failed: %v", s.Name, err)
 	}

src/cmd/internal/obj/plist.go

@@ -136,13 +136,17 @@ func (ctxt *Link) InitTextSym(s *LSym, flag int) {
 	ctxt.Text = append(ctxt.Text, s)
 
 	// Set up DWARF entries for s.
-	dsym, drsym := ctxt.dwarfSym(s)
-	dsym.Type = objabi.SDWARFINFO
-	dsym.Set(AttrDuplicateOK, s.DuplicateOK())
-	drsym.Type = objabi.SDWARFRANGE
-	drsym.Set(AttrDuplicateOK, s.DuplicateOK())
-	ctxt.Data = append(ctxt.Data, dsym)
-	ctxt.Data = append(ctxt.Data, drsym)
+	info, loc, ranges := ctxt.dwarfSym(s)
+	info.Type = objabi.SDWARFINFO
+	info.Set(AttrDuplicateOK, s.DuplicateOK())
+	if loc != nil {
+		loc.Type = objabi.SDWARFLOC
+		loc.Set(AttrDuplicateOK, s.DuplicateOK())
+		ctxt.Data = append(ctxt.Data, loc)
+	}
+	ranges.Type = objabi.SDWARFRANGE
+	ranges.Set(AttrDuplicateOK, s.DuplicateOK())
+	ctxt.Data = append(ctxt.Data, info, ranges)
 
 	// Set up the function's gcargs and gclocals.
 	// They will be filled in later if needed.

src/cmd/internal/obj/x86/a.out.go

@@ -1006,3 +1006,120 @@ const (
 	T_64     = 1 << 6
 	T_GOTYPE = 1 << 7
 )
+
+// https://www.uclibc.org/docs/psABI-x86_64.pdf, figure 3.36
+var AMD64DWARFRegisters = map[int16]int16{
+	REG_AX:  0,
+	REG_DX:  1,
+	REG_CX:  2,
+	REG_BX:  3,
+	REG_SI:  4,
+	REG_DI:  5,
+	REG_BP:  6,
+	REG_SP:  7,
+	REG_R8:  8,
+	REG_R9:  9,
+	REG_R10: 10,
+	REG_R11: 11,
+	REG_R12: 12,
+	REG_R13: 13,
+	REG_R14: 14,
+	REG_R15: 15,
+	// 16 is "Return Address RA", whatever that is.
+	// XMM registers. %xmmN => XN.
+	REG_X0:  17,
+	REG_X1:  18,
+	REG_X2:  19,
+	REG_X3:  20,
+	REG_X4:  21,
+	REG_X5:  22,
+	REG_X6:  23,
+	REG_X7:  24,
+	REG_X8:  25,
+	REG_X9:  26,
+	REG_X10: 27,
+	REG_X11: 28,
+	REG_X12: 29,
+	REG_X13: 30,
+	REG_X14: 31,
+	REG_X15: 32,
+	// ST registers. %stN => FN.
+	REG_F0: 33,
+	REG_F1: 34,
+	REG_F2: 35,
+	REG_F3: 36,
+	REG_F4: 37,
+	REG_F5: 38,
+	REG_F6: 39,
+	REG_F7: 40,
+	// MMX registers. %mmN => MN.
+	REG_M0: 41,
+	REG_M1: 42,
+	REG_M2: 43,
+	REG_M3: 44,
+	REG_M4: 45,
+	REG_M5: 46,
+	REG_M6: 47,
+	REG_M7: 48,
+	// 48 is flags, which doesn't have a name.
+	REG_ES: 50,
+	REG_CS: 51,
+	REG_SS: 52,
+	REG_DS: 53,
+	REG_FS: 54,
+	REG_GS: 55,
+	// 58 and 59 are {fs,gs}base, which don't have names.
+	REG_TR:   62,
+	REG_LDTR: 63,
+	// 64-66 are mxcsr, fcw, fsw, which don't have names.
+}
+
+// https://www.uclibc.org/docs/psABI-i386.pdf, table 2.14
+var X86DWARFRegisters = map[int16]int16{
+	REG_AX: 0,
+	REG_CX: 1,
+	REG_DX: 2,
+	REG_BX: 3,
+	REG_SP: 4,
+	REG_BP: 5,
+	REG_SI: 6,
+	REG_DI: 7,
+	// 8 is "Return Address RA", whatever that is.
+	// 9 is flags, which doesn't have a name.
+	// ST registers. %stN => FN.
+	REG_F0: 11,
+	REG_F1: 12,
+	REG_F2: 13,
+	REG_F3: 14,
+	REG_F4: 15,
+	REG_F5: 16,
+	REG_F6: 17,
+	REG_F7: 18,
+	// XMM registers. %xmmN => XN.
+	REG_X0: 21,
+	REG_X1: 22,
+	REG_X2: 23,
+	REG_X3: 24,
+	REG_X4: 25,
+	REG_X5: 26,
+	REG_X6: 27,
+	REG_X7: 28,
+	// MMX registers. %mmN => MN.
+	REG_M0: 29,
+	REG_M1: 30,
+	REG_M2: 31,
+	REG_M3: 32,
+	REG_M4: 33,
+	REG_M5: 34,
+	REG_M6: 35,
+	REG_M7: 36,
+	// 39 is mxcsr, which doesn't have a name.
+	REG_ES:   40,
+	REG_CS:   41,
+	REG_SS:   42,
+	REG_DS:   43,
+	REG_FS:   44,
+	REG_GS:   45,
+	REG_TR:   48,
+	REG_LDTR: 49,
+}

src/cmd/internal/obj/x86/obj6.go

@@ -1231,28 +1231,31 @@ var unaryDst = map[obj.As]bool{
 }
 
 var Linkamd64 = obj.LinkArch{
-	Arch:       sys.ArchAMD64,
-	Init:       instinit,
-	Preprocess: preprocess,
-	Assemble:   span6,
-	Progedit:   progedit,
-	UnaryDst:   unaryDst,
+	Arch:           sys.ArchAMD64,
+	Init:           instinit,
+	Preprocess:     preprocess,
+	Assemble:       span6,
+	Progedit:       progedit,
+	UnaryDst:       unaryDst,
+	DWARFRegisters: AMD64DWARFRegisters,
 }
 
 var Linkamd64p32 = obj.LinkArch{
-	Arch:       sys.ArchAMD64P32,
-	Init:       instinit,
-	Preprocess: preprocess,
-	Assemble:   span6,
-	Progedit:   progedit,
-	UnaryDst:   unaryDst,
+	Arch:           sys.ArchAMD64P32,
+	Init:           instinit,
+	Preprocess:     preprocess,
+	Assemble:       span6,
+	Progedit:       progedit,
+	UnaryDst:       unaryDst,
+	DWARFRegisters: AMD64DWARFRegisters,
 }
 
 var Link386 = obj.LinkArch{
-	Arch:       sys.Arch386,
-	Init:       instinit,
-	Preprocess: preprocess,
-	Assemble:   span6,
-	Progedit:   progedit,
-	UnaryDst:   unaryDst,
+	Arch:           sys.Arch386,
+	Init:           instinit,
+	Preprocess:     preprocess,
+	Assemble:       span6,
+	Progedit:       progedit,
+	UnaryDst:       unaryDst,
+	DWARFRegisters: AMD64DWARFRegisters,
 }

src/cmd/internal/objabi/symkind.go

@@ -57,4 +57,5 @@ const (
 	// Debugging data
 	SDWARFINFO
 	SDWARFRANGE
+	SDWARFLOC
 )

src/cmd/internal/objabi/symkind_string.go

@@ -4,9 +4,9 @@ package objabi
 
 import "fmt"
 
-const _SymKind_name = "SxxxSTEXTSRODATASNOPTRDATASDATASBSSSNOPTRBSSSTLSBSSSDWARFINFOSDWARFRANGE"
+const _SymKind_name = "SxxxSTEXTSRODATASNOPTRDATASDATASBSSSNOPTRBSSSTLSBSSSDWARFINFOSDWARFRANGESDWARFLOC"
 
-var _SymKind_index = [...]uint8{0, 4, 9, 16, 26, 31, 35, 44, 51, 61, 72}
+var _SymKind_index = [...]uint8{0, 4, 9, 16, 26, 31, 35, 44, 51, 61, 72, 81}
 
 func (i SymKind) String() string {
 	if i >= SymKind(len(_SymKind_index)-1) {

src/cmd/link/internal/ld/data.go

@@ -1862,6 +1862,8 @@ func (ctxt *Link) dodata() {
 			sect = addsection(&Segdwarf, ".debug_info", 04)
 		case SDWARFRANGE:
 			sect = addsection(&Segdwarf, ".debug_ranges", 04)
+		case SDWARFLOC:
+			sect = addsection(&Segdwarf, ".debug_loc", 04)
 		default:
 			Errorf(dwarfp[i], "unknown DWARF section %v", curType)
 		}

src/cmd/link/internal/ld/dwarf.go

@@ -1579,10 +1579,35 @@ func dwarfgeneratedebugsyms(ctxt *Link) {
 	syms = writearanges(ctxt, syms)
 	syms = writegdbscript(ctxt, syms)
 	syms = append(syms, infosyms...)
+	syms = collectlocs(ctxt, syms, funcs)
 	syms = writeranges(ctxt, syms)
 	dwarfp = syms
 }
 
+func collectlocs(ctxt *Link, syms []*Symbol, funcs []*Symbol) []*Symbol {
+	empty := true
+	for _, fn := range funcs {
+		for _, reloc := range fn.R {
+			if reloc.Type == objabi.R_DWARFREF && strings.HasPrefix(reloc.Sym.Name, dwarf.LocPrefix) {
+				reloc.Sym.Attr |= AttrReachable | AttrNotInSymbolTable
+				syms = append(syms, reloc.Sym)
+				empty = false
+				// One location list entry per function, but many relocations to it. Don't duplicate.
+				break
+			}
+		}
+	}
+	// Don't emit .debug_loc if it's empty -- it makes the ARM linker mad.
+	if !empty {
+		locsym := ctxt.Syms.Lookup(".debug_loc", 0)
+		locsym.R = locsym.R[:0]
+		locsym.Type = SDWARFLOC
+		locsym.Attr |= AttrReachable
+		syms = append(syms, locsym)
+	}
+	return syms
+}
+
 /*
  *  Elf.
  */
@@ -1595,6 +1620,7 @@ func dwarfaddshstrings(ctxt *Link, shstrtab *Symbol) {
 	Addstring(shstrtab, ".debug_aranges")
 	Addstring(shstrtab, ".debug_frame")
 	Addstring(shstrtab, ".debug_info")
+	Addstring(shstrtab, ".debug_loc")
 	Addstring(shstrtab, ".debug_line")
 	Addstring(shstrtab, ".debug_pubnames")
 	Addstring(shstrtab, ".debug_pubtypes")
@@ -1602,6 +1628,7 @@ func dwarfaddshstrings(ctxt *Link, shstrtab *Symbol) {
 	Addstring(shstrtab, ".debug_ranges")
 	if Linkmode == LinkExternal {
 		Addstring(shstrtab, elfRelType+".debug_info")
+		Addstring(shstrtab, elfRelType+".debug_loc")
 		Addstring(shstrtab, elfRelType+".debug_aranges")
 		Addstring(shstrtab, elfRelType+".debug_line")
 		Addstring(shstrtab, elfRelType+".debug_frame")
@@ -1628,6 +1655,10 @@ func dwarfaddelfsectionsyms(ctxt *Link) {
 	putelfsectionsym(sym, sym.Sect.Elfsect.shnum)
 	sym = ctxt.Syms.Lookup(".debug_frame", 0)
 	putelfsectionsym(sym, sym.Sect.Elfsect.shnum)
+	sym = ctxt.Syms.Lookup(".debug_loc", 0)
+	if sym.Sect != nil {
+		putelfsectionsym(sym, sym.Sect.Elfsect.shnum)
+	}
 	sym = ctxt.Syms.Lookup(".debug_ranges", 0)
 	if sym.Sect != nil {
 		putelfsectionsym(sym, sym.Sect.Elfsect.shnum)

src/cmd/link/internal/ld/elf.go

@@ -1808,7 +1808,7 @@ func elfrelocsect(ctxt *Link, sect *Section, syms []*Symbol) {
 				continue
 			}
 			if r.Xsym == nil {
-				Errorf(sym, "missing xsym in relocation")
+				Errorf(sym, "missing xsym in relocation %#v %#v", r.Sym.Name, sym)
 				continue
 			}
 			if r.Xsym.ElfsymForReloc() == 0 {
@@ -2596,12 +2596,9 @@ elfobj:
 			elfshreloc(sect)
 		}
 		for _, s := range dwarfp {
-			if len(s.R) > 0 || s.Type == SDWARFINFO {
+			if len(s.R) > 0 || s.Type == SDWARFINFO || s.Type == SDWARFLOC {
 				elfshreloc(s.Sect)
 			}
-			if s.Type == SDWARFINFO {
-				break
-			}
 		}
 		// add a .note.GNU-stack section to mark the stack as non-executable
 		sh := elfshname(".note.GNU-stack")

src/cmd/link/internal/ld/symkind.go

@@ -105,6 +105,7 @@ const (
 	SDWARFSECT
 	SDWARFINFO
 	SDWARFRANGE
+	SDWARFLOC
 	SSUB       = SymKind(1 << 8)
 	SMASK      = SymKind(SSUB - 1)
 	SHIDDEN    = SymKind(1 << 9)
@@ -124,6 +125,7 @@ var abiSymKindToSymKind = [...]SymKind{
 	STLSBSS,
 	SDWARFINFO,
 	SDWARFRANGE,
+	SDWARFLOC,
 }
 
 // readOnly are the symbol kinds that form read-only sections. In some

src/cmd/link/internal/ld/symkind_string.go

@@ -4,9 +4,9 @@ package ld
 
 import "fmt"
 
-const _SymKind_name = "SxxxSTEXTSELFRXSECTSTYPESSTRINGSGOSTRINGSGOFUNCSGCBITSSRODATASFUNCTABSELFROSECTSMACHOPLTSTYPERELROSSTRINGRELROSGOSTRINGRELROSGOFUNCRELROSGCBITSRELROSRODATARELROSFUNCTABRELROSTYPELINKSITABLINKSSYMTABSPCLNTABSELFSECTSMACHOSMACHOGOTSWINDOWSSELFGOTSNOPTRDATASINITARRSDATASBSSSNOPTRBSSSTLSBSSSXREFSMACHOSYMSTRSMACHOSYMTABSMACHOINDIRECTPLTSMACHOINDIRECTGOTSFILESFILEPATHSCONSTSDYNIMPORTSHOSTOBJSDWARFSECTSDWARFINFOSDWARFRANGE"
+const _SymKind_name = "SxxxSTEXTSELFRXSECTSTYPESSTRINGSGOSTRINGSGOFUNCSGCBITSSRODATASFUNCTABSELFROSECTSMACHOPLTSTYPERELROSSTRINGRELROSGOSTRINGRELROSGOFUNCRELROSGCBITSRELROSRODATARELROSFUNCTABRELROSTYPELINKSITABLINKSSYMTABSPCLNTABSELFSECTSMACHOSMACHOGOTSWINDOWSSELFGOTSNOPTRDATASINITARRSDATASBSSSNOPTRBSSSTLSBSSSXREFSMACHOSYMSTRSMACHOSYMTABSMACHOINDIRECTPLTSMACHOINDIRECTGOTSFILESFILEPATHSCONSTSDYNIMPORTSHOSTOBJSDWARFSECTSDWARFINFOSDWARFRANGESDWARFLOC"
 
-var _SymKind_index = [...]uint16{0, 4, 9, 19, 24, 31, 40, 47, 54, 61, 69, 79, 88, 98, 110, 124, 136, 148, 160, 173, 182, 191, 198, 206, 214, 220, 229, 237, 244, 254, 262, 267, 271, 280, 287, 292, 304, 316, 333, 350, 355, 364, 370, 380, 388, 398, 408, 419}
+var _SymKind_index = [...]uint16{0, 4, 9, 19, 24, 31, 40, 47, 54, 61, 69, 79, 88, 98, 110, 124, 136, 148, 160, 173, 182, 191, 198, 206, 214, 220, 229, 237, 244, 254, 262, 267, 271, 280, 287, 292, 304, 316, 333, 350, 355, 364, 370, 380, 388, 398, 408, 419, 428}
 
 func (i SymKind) String() string {
 	if i < 0 || i >= SymKind(len(_SymKind_index)-1) {