runtime: write trace stack tab directly to trace buffer

Currently, the stack frame of (*traceStackTable).dump is 68KiB. We're
about to move (*traceStackTable).dump to the system stack, where we
often don't have this much room.

5140 bytes of this is an on-stack temporary buffer for constructing
potentially large trace events before copying these out to the actual
trace buffer.

Reduce the stack frame size by writing these events directly to the
trace buffer rather than temporary space. This introduces a couple
complications:

- The trace event starts with a varint encoding the event payload's
  length in bytes. These events are large and somewhat complicated, so
  it's hard to know the size ahead of time. That's not a problem with
  the temporary buffer because we can just construct the event and see
  how long it is. In order to support writing directly to the trace
  buffer, we reserve enough bytes for a maximum size varint and add
  support for populating a reserved space after the fact.

- Emitting a stack event calls traceFrameForPC, which can itself emit
  string events. If these were emitted in the middle of the stack
  event, it would corrupt the stream. We already allocate a []Frame to
  convert the PC slice to frames, and then convert each Frame into a
  traceFrame with trace string IDs, so we address this by combining
  these two steps into one so that all trace string events are emitted
  before we start constructing the stack event.

For #53979.

Change-Id: Ie60704be95199559c426b551f8e119b14e06ddac
Reviewed-on: https://go-review.googlesource.com/c/go/+/422954
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>

src/runtime/trace.go

@@ -935,6 +935,22 @@ func (buf *traceBuf) varint(v uint64) {
 	buf.pos = pos
 }
 
+// varintAt writes varint v at byte position pos in buf. This always
+// consumes traceBytesPerNumber bytes. This is intended for when the
+// caller needs to reserve space for a varint but can't populate it
+// until later.
+func (buf *traceBuf) varintAt(pos int, v uint64) {
+	for i := 0; i < traceBytesPerNumber; i++ {
+		if i < traceBytesPerNumber-1 {
+			buf.arr[pos] = 0x80 | byte(v)
+		} else {
+			buf.arr[pos] = byte(v)
+		}
+		v >>= 7
+		pos++
+	}
+}
+
 // byte appends v to buf.
 func (buf *traceBuf) byte(v byte) {
 	buf.arr[buf.pos] = v
@@ -1024,15 +1040,18 @@ func (tab *traceStackTable) newStack(n int) *traceStack {
 	return (*traceStack)(tab.mem.alloc(unsafe.Sizeof(traceStack{}) + uintptr(n)*goarch.PtrSize))
 }
 
-// allFrames returns all of the Frames corresponding to pcs.
+// traceFrames returns the frames corresponding to pcs. It may
-func allFrames(pcs []uintptr) []Frame {
+// allocate and may emit trace events.
-	frames := make([]Frame, 0, len(pcs))
+func traceFrames(bufp traceBufPtr, pcs []uintptr) ([]traceFrame, traceBufPtr) {
+	frames := make([]traceFrame, 0, len(pcs))
 	ci := CallersFrames(pcs)
 	for {
+		var frame traceFrame
 		f, more := ci.Next()
-		frames = append(frames, f)
+		frame, bufp = traceFrameForPC(bufp, 0, f)
+		frames = append(frames, frame)
 		if !more {
-			return frames
+			return frames, bufp
 		}
 	}
 }
@@ -1040,32 +1059,41 @@ func allFrames(pcs []uintptr) []Frame {
 // dump writes all previously cached stacks to trace buffers,
 // releases all memory and resets state.
 func (tab *traceStackTable) dump() {
-	var tmp [(2 + 4*traceStackSize) * traceBytesPerNumber]byte
 	bufp := traceFlush(0, 0)
 	for _, stk := range tab.tab {
 		stk := stk.ptr()
 		for ; stk != nil; stk = stk.link.ptr() {
-			tmpbuf := tmp[:0]
+			var frames []traceFrame
-			tmpbuf = traceAppend(tmpbuf, uint64(stk.id))
+			frames, bufp = traceFrames(bufp, stk.stack())
-			frames := allFrames(stk.stack())
+
-			tmpbuf = traceAppend(tmpbuf, uint64(len(frames)))
+			// Estimate the size of this record. This
-			for _, f := range frames {
+			// bound is pretty loose, but avoids counting
-				var frame traceFrame
+			// lots of varint sizes.
-				frame, bufp = traceFrameForPC(bufp, 0, f)
+			maxSize := 1 + traceBytesPerNumber + (2+4*len(frames))*traceBytesPerNumber
-				tmpbuf = traceAppend(tmpbuf, uint64(f.PC))
+			// Make sure we have enough buffer space.
-				tmpbuf = traceAppend(tmpbuf, uint64(frame.funcID))
+			if buf := bufp.ptr(); len(buf.arr)-buf.pos < maxSize {
-				tmpbuf = traceAppend(tmpbuf, uint64(frame.fileID))
-				tmpbuf = traceAppend(tmpbuf, uint64(frame.line))
-			}
-			// Now copy to the buffer.
-			size := 1 + traceBytesPerNumber + len(tmpbuf)
-			if buf := bufp.ptr(); len(buf.arr)-buf.pos < size {
 				bufp = traceFlush(bufp, 0)
 			}
+
+			// Emit header, with space reserved for length.
 			buf := bufp.ptr()
 			buf.byte(traceEvStack | 3<<traceArgCountShift)
-			buf.varint(uint64(len(tmpbuf)))
+			lenPos := buf.pos
-			buf.pos += copy(buf.arr[buf.pos:], tmpbuf)
+			buf.pos += traceBytesPerNumber
+
+			// Emit body.
+			recPos := buf.pos
+			buf.varint(uint64(stk.id))
+			buf.varint(uint64(len(frames)))
+			for _, frame := range frames {
+				buf.varint(uint64(frame.PC))
+				buf.varint(frame.funcID)
+				buf.varint(frame.fileID)
+				buf.varint(frame.line)
+			}
+
+			// Fill in size header.
+			buf.varintAt(lenPos, uint64(buf.pos-recPos))
 		}
 	}
 
@@ -1079,6 +1107,7 @@ func (tab *traceStackTable) dump() {
 }
 
 type traceFrame struct {
+	PC     uintptr
 	funcID uint64
 	fileID uint64
 	line   uint64
@@ -1089,6 +1118,7 @@ type traceFrame struct {
 func traceFrameForPC(buf traceBufPtr, pid int32, f Frame) (traceFrame, traceBufPtr) {
 	bufp := &buf
 	var frame traceFrame
+	frame.PC = f.PC
 
 	fn := f.Function
 	const maxLen = 1 << 10