cmd/link: process is_stmt data into dwarf line tables

To improve debugging, instructions should be annotated with
DWARF is_stmt.  The DWARF default before was is_stmt=1, and
to remove "jumpy" stepping the optimizer was tagging
instructions with a no-position position, which interferes
with the accuracy of profiling information.  This allows
that to be corrected, and also allows more "jumpy" positions
to be annotated with is_stmt=0 (these changes were not made
for 1.10 because of worries about further messing up
profiling).

The is_stmt values are placed in a pc-encoded table and
passed through a symbol derived from the name of the
function and processed in the linker alongside its
processing of each function's pc/line tables.

The only change in binary size is in the .debug_line tables
measured with "objdump -h --section=.debug_line go1.test"
For go1.test, these are 2614 bytes larger,
or 0.72% of the size of .debug_line,
or 0.025% of the file size.

This will increase in proportion to how much the is_stmt
flag is used (toggled).

Change-Id: Ic1f1aeccff44591ad0494d29e1a0202a3c506a7a
Reviewed-on: https://go-review.googlesource.com/93664
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>

4 files changed, 40 insertions, 5 deletions

diff --git a/src/cmd/internal/obj/link.go b/src/cmd/internal/obj/link.go
index a3a9c9ffe8..16e4e1410d 100644
--- a/src/cmd/internal/obj/link.go
+++ b/src/cmd/internal/obj/link.go
@@ -395,6 +395,7 @@ type FuncInfo struct {
 	dwarfLocSym    *LSym
 	dwarfRangesSym *LSym
 	dwarfAbsFnSym  *LSym
+	dwarfIsStmtSym *LSym
 
 	GCArgs   LSym
 	GCLocals LSym
diff --git a/src/cmd/internal/obj/objfile.go b/src/cmd/internal/obj/objfile.go
index b5f5790a50..91b48b5e08 100644
--- a/src/cmd/internal/obj/objfile.go
+++ b/src/cmd/internal/obj/objfile.go
@@ -497,7 +497,7 @@ func (c dwCtxt) Logf(format string, args ...interface{}) {
 	c.Link.Logf(format, args...)
 }
 
-func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfLocSym, dwarfRangesSym, dwarfAbsFnSym *LSym) {
+func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfLocSym, dwarfRangesSym, dwarfAbsFnSym, dwarfIsStmtSym *LSym) {
 	if s.Type != objabi.STEXT {
 		ctxt.Diag("dwarfSym of non-TEXT %v", s)
 	}
@@ -510,9 +510,10 @@ func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfLocSym, dwarfRangesSym,
 		if s.WasInlined() {
 			s.Func.dwarfAbsFnSym = ctxt.DwFixups.AbsFuncDwarfSym(s)
 		}
+		s.Func.dwarfIsStmtSym = ctxt.LookupDerived(s, dwarf.IsStmtPrefix+s.Name)
 
 	}
-	return s.Func.dwarfInfoSym, s.Func.dwarfLocSym, s.Func.dwarfRangesSym, s.Func.dwarfAbsFnSym
+	return s.Func.dwarfInfoSym, s.Func.dwarfLocSym, s.Func.dwarfRangesSym, s.Func.dwarfAbsFnSym, s.Func.dwarfIsStmtSym
 }
 
 func (s *LSym) Len() int64 {
@@ -536,13 +537,14 @@ func (ctxt *Link) fileSymbol(fn *LSym) *LSym {
 // TEXT symbol 's'. The various DWARF symbols must already have been
 // initialized in InitTextSym.
 func (ctxt *Link) populateDWARF(curfn interface{}, s *LSym, myimportpath string) {
-	info, loc, ranges, absfunc := ctxt.dwarfSym(s)
+	info, loc, ranges, absfunc, _ := ctxt.dwarfSym(s)
 	if info.Size != 0 {
 		ctxt.Diag("makeFuncDebugEntry double process %v", s)
 	}
 	var scopes []dwarf.Scope
 	var inlcalls dwarf.InlCalls
 	if ctxt.DebugInfo != nil {
+		stmtData(ctxt, s)
 		scopes, inlcalls = ctxt.DebugInfo(s, curfn)
 	}
 	var err error
diff --git a/src/cmd/internal/obj/pcln.go b/src/cmd/internal/obj/pcln.go
index d1d36cf685..1d5831d3cb 100644
--- a/src/cmd/internal/obj/pcln.go
+++ b/src/cmd/internal/obj/pcln.go
@@ -4,7 +4,10 @@
 
 package obj
 
-import "log"
+import (
+	"cmd/internal/src"
+	"log"
+)
 
 func addvarint(d *Pcdata, v uint32) {
 	for ; v >= 0x80; v >>= 7 {
@@ -230,6 +233,25 @@ func pctospadj(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg in
 	return oldval + p.Spadj
 }
 
+// pctostmt returns either,
+// if phase==0, then whether the current instruction is a step-target (Dwarf is_stmt)
+// else (phase == 1), zero.
+//
+func pctostmt(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
+	if phase == 1 {
+		return 0 // Ignored; also different from initial value of -1, if that ever matters.
+	}
+	s := p.Pos.IsStmt()
+	if s == src.PosIsStmt {
+		return 1
+	}
+	if s == src.PosNotStmt { // includes NoSrcPos case
+		return 0
+	}
+	// Line numbers in .s files will have no special setting, therefore default to is_stmt=1.
+	return 1
+}
+
 // pctopcdata computes the pcdata value in effect at p.
 // A PCDATA instruction sets the value in effect at future
 // non-PCDATA instructions.
@@ -248,6 +270,13 @@ func pctopcdata(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg i
 	return int32(p.To.Offset)
 }
 
+// stmtData writes out pc-linked is_stmt data for eventual use in the DWARF line numbering table.
+func stmtData(ctxt *Link, cursym *LSym) {
+	var pctostmtData Pcdata
+	funcpctab(ctxt, &pctostmtData, cursym, "pctostmt", pctostmt, nil)
+	cursym.Func.dwarfIsStmtSym.P = pctostmtData.P
+}
+
 func linkpcln(ctxt *Link, cursym *LSym) {
 	pcln := &cursym.Func.Pcln
 
diff --git a/src/cmd/internal/obj/plist.go b/src/cmd/internal/obj/plist.go
index e2609da35d..8e70404774 100644
--- a/src/cmd/internal/obj/plist.go
+++ b/src/cmd/internal/obj/plist.go
@@ -136,7 +136,7 @@ func (ctxt *Link) InitTextSym(s *LSym, flag int) {
 	ctxt.Text = append(ctxt.Text, s)
 
 	// Set up DWARF entries for s.
-	info, loc, ranges, _ := ctxt.dwarfSym(s)
+	info, loc, ranges, _, isstmt := ctxt.dwarfSym(s)
 	info.Type = objabi.SDWARFINFO
 	info.Set(AttrDuplicateOK, s.DuplicateOK())
 	if loc != nil {
@@ -147,6 +147,9 @@ func (ctxt *Link) InitTextSym(s *LSym, flag int) {
 	ranges.Type = objabi.SDWARFRANGE
 	ranges.Set(AttrDuplicateOK, s.DuplicateOK())
 	ctxt.Data = append(ctxt.Data, info, ranges)
+	isstmt.Type = objabi.SDWARFMISC
+	isstmt.Set(AttrDuplicateOK, s.DuplicateOK())
+	ctxt.Data = append(ctxt.Data, isstmt)
 
 	// Set up the function's gcargs and gclocals.
 	// They will be filled in later if needed.