Use llvm-symbolizer's JSON output to provide function start lines

When analyzing a perf.data profile converted automatically via perf_to_profile via pprof -raw perf.data, no function start lines (s=0) are present in any of the locations. With 813a5fb, this can be easily solved by using the same JSON frame data from llvm-symbolizer to provide StartLine for Function.start_line. This solves #823.
google · Sep 1, 2024 · a088559 · a088559
1 parent da1f7e9
commit a088559
Show file tree

Hide file tree

Showing 6 changed files with 22 additions and 17 deletions.
diff --git a/driver/driver.go b/driver/driver.go
@@ -186,10 +186,11 @@ type ObjFile interface {
 
 // A Frame describes a single line in a source file.
 type Frame struct {
-	Func   string // name of function
-	File   string // source file name
-	Line   int    // line in file
-	Column int    // column in file
+	Func      string // name of function
+	File      string // source file name
+	Line      int    // line in file
+	Column    int    // column in file
+	StartLine int    // start line of function (if available)
 }
 
 // A Sym describes a single symbol in an object file.

diff --git a/internal/binutils/addr2liner_llvm.go b/internal/binutils/addr2liner_llvm.go
@@ -151,18 +151,19 @@ func (d *llvmSymbolizer) readCodeFrames() ([]plugin.Frame, error) {
 		Address    string `json:"Address"`
 		ModuleName string `json:"ModuleName"`
 		Symbol     []struct {
-			Line         int    `json:"Line"`
-			Column       int    `json:"Column"`
-			FunctionName string `json:"FunctionName"`
-			FileName     string `json:"FileName"`
+			Line          int    `json:"Line"`
+			Column        int    `json:"Column"`
+			FunctionName  string `json:"FunctionName"`
+			FileName      string `json:"FileName"`
+			StartLine     int    `json:"StartLine"`
 		} `json:"Symbol"`
 	}
 	if err := json.Unmarshal([]byte(line), &frame); err != nil {
 		return nil, err
 	}
 	var stack []plugin.Frame
 	for _, s := range frame.Symbol {
-		stack = append(stack, plugin.Frame{Func: s.FunctionName, File: s.FileName, Line: s.Line, Column: s.Column})
+		stack = append(stack, plugin.Frame{Func: s.FunctionName, File: s.FileName, Line: s.Line, Column: s.Column, StartLine: s.StartLine})
 	}
 	return stack, nil
 }

diff --git a/internal/binutils/binutils_test.go b/internal/binutils/binutils_test.go
@@ -373,7 +373,7 @@ func TestObjFile(t *testing.T) {
 				t.Fatalf("SourceLine: unexpected error %v", err)
 			}
 			wantFrames := []plugin.Frame{
-				{Func: "main", File: "/tmp/hello.c", Line: 3},
+				{Func: "main", File: "/tmp/hello.c", Line: 3, StartLine: 3},
 			}
 			if !reflect.DeepEqual(gotFrames, wantFrames) {
 				t.Fatalf("SourceLine for main: got %v; want %v\n", gotFrames, wantFrames)
@@ -461,8 +461,8 @@ func TestLLVMSymbolizer(t *testing.T) {
 		frames []plugin.Frame
 	}{
 		{0x10, false, []plugin.Frame{
-			{Func: "Inlined_0x10", File: "foo.h", Line: 0, Column: 0},
-			{Func: "Func_0x10", File: "foo.c", Line: 2, Column: 1},
+			{Func: "Inlined_0x10", File: "foo.h", Line: 0, Column: 0, StartLine: 0},
+			{Func: "Func_0x10", File: "foo.c", Line: 2, Column: 1, StartLine: 2},
 		}},
 		{0x20, true, []plugin.Frame{
 			{Func: "foo_0x20", File: "0x20 8"},
@@ -480,6 +480,7 @@ func TestLLVMSymbolizer(t *testing.T) {
 			defer symbolizer.rw.close()
 
 			frames, err := symbolizer.addrInfo(c.addr)
+			t.Logf("EITA expect %v; got %v\n", c.frames, frames)
 			if err != nil {
 				t.Fatalf("LLVM: unexpected error %v", err)
 			}

diff --git a/internal/binutils/testdata/fake-llvm-symbolizer b/internal/binutils/testdata/fake-llvm-symbolizer
@@ -29,7 +29,7 @@ while read line; do
   addr=$3
   case ${kind} in
   CODE)
-    echo "{\"Address\":\"${addr}\",\"ModuleName\":\"${fname}\",\"Symbol\":[{\"Column\":0,\"FileName\":\"${fname}.h\",\"FunctionName\":\"Inlined_${addr}\",\"Line\":0},{\"Column\":1,\"FileName\":\"${fname}.c\",\"FunctionName\":\"Func_${addr}\",\"Line\":2}]}"
+    echo "{\"Address\":\"${addr}\",\"ModuleName\":\"${fname}\",\"Symbol\":[{\"Column\":0,\"FileName\":\"${fname}.h\",\"FunctionName\":\"Inlined_${addr}\",\"Line\":0,\"StartLine\":0},{\"Column\":1,\"FileName\":\"${fname}.c\",\"FunctionName\":\"Func_${addr}\",\"Line\":2,\"StartLine\":2}]}"
     ;;
   DATA)
     echo "{\"Address\":\"${addr}\",\"ModuleName\":\"${fname}\",\"Data\":{\"Name\":\"${fname}_${addr}\",\"Size\":\"0x8\",\"Start\":\"${addr}\"}}"

diff --git a/internal/plugin/plugin.go b/internal/plugin/plugin.go
@@ -159,10 +159,11 @@ type ObjFile interface {
 
 // A Frame describes a location in a single line in a source file.
 type Frame struct {
-	Func   string // name of function
-	File   string // source file name
-	Line   int    // line in file
-	Column int    // column in line (if available)
+	Func          string // name of function
+	File          string // source file name
+	Line          int    // line in file
+	Column        int    // column in line (if available)
+	StartLine     int    // start line of function (if available)
 }
 
 // A Sym describes a single symbol in an object file.

diff --git a/internal/symbolizer/symbolizer.go b/internal/symbolizer/symbolizer.go
@@ -228,6 +228,7 @@ func symbolizeOneMapping(m *profile.Mapping, locs []*profile.Location, obj plugi
 				Name:       frame.Func,
 				SystemName: frame.Func,
 				Filename:   frame.File,
+				StartLine:  int64(frame.StartLine),
 			})
 			l.Line[i] = profile.Line{
 				Function: f,