cmd/compile: better modeling of escape across loop levels

Brief background on "why heap allocate".  Things can be
forced to the heap for the following reasons:

1) address published, hence lifetime unknown.
2) size unknown/too large, cannot be stack allocated
3) multiplicity unknown/too large, cannot be stack allocated
4) reachable from heap (not necessarily published)

The bug here is a case of failing to enforce 4) when an
object Y was reachable from a heap allocation X forced
because of 3).  It was found in the case of a closure
allocated within a loop (X) and assigned to a variable
outside the loop (multiplicity unknown) where the closure
also captured a map (Y) declared outside the loop (reachable
from heap). Note the variable declared outside the loop (Y)
is not published, has known size, and known multiplicity
(one). The only reason for heap allocation is that it was
reached from a heap allocated item (X), but because that was
not forced by publication, it has to be tracked by loop
level, but escape-loop level was not tracked and thus a bug
results.

The fix is that when a heap allocation is newly discovered,
use its looplevel as the minimum loop level for downstream
escape flooding.

Every attempt to generalize this bug to X-in-loop-
references-Y-outside loop succeeded, so the fix was aimed
to be general.  Anywhere that loop level forces heap
allocation, the loop level is tracked.  This is not yet
tested for all possible X and Y, but it is correctness-
conservative and because it caused only one trivial
regression in the escape tests, it is probably also
performance-conservative.

The new test checks the following:
1) in the map case, that if fn escapes, so does the map.
2) in the map case, if fn does not escape, neither does the map.
3) in the &x case, that if fn escapes, so does &x.
4) in the &x case, if fn does not escape, neither does &x.

Fixes #13799.

Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb
Reviewed-on: https://go-review.googlesource.com/18234
Run-TryBot: David Chase <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Russ Cox <[email protected]>

Author: dr2chase

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

@@ -299,12 +299,13 @@ func (l Level) guaranteedDereference() int {
 }
 
 type NodeEscState struct {
-	Curfn        *Node
-	Escflowsrc   *NodeList // flow(this, src)
-	Escretval    *NodeList // on OCALLxxx, list of dummy return values
-	Escloopdepth int32     // -1: global, 0: return variables, 1:function top level, increased inside function for every loop or label to mark scopes
-	Esclevel     Level
-	Walkgen      uint32
+	Curfn             *Node
+	Escflowsrc        *NodeList // flow(this, src)
+	Escretval         *NodeList // on OCALLxxx, list of dummy return values
+	Escloopdepth      int32     // -1: global, 0: return variables, 1:function top level, increased inside function for every loop or label to mark scopes
+	Esclevel          Level
+	Walkgen           uint32
+	Maxextraloopdepth int32
 }
 
 func (e *EscState) nodeEscState(n *Node) *NodeEscState {
@@ -1579,7 +1580,13 @@ func funcOutputAndInput(dst, src *Node) bool {
 		src.Op == ONAME && src.Class == PPARAM && src.Name.Curfn == dst.Name.Curfn
 }
 
+const NOTALOOPDEPTH = -1
+
 func escwalk(e *EscState, level Level, dst *Node, src *Node) {
+	escwalkBody(e, level, dst, src, NOTALOOPDEPTH)
+}
+
+func escwalkBody(e *EscState, level Level, dst *Node, src *Node, extraloopdepth int32) {
 	if src.Op == OLITERAL {
 		return
 	}
@@ -1590,16 +1597,29 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 		// convergence.
 		level = level.min(srcE.Esclevel)
 		if level == srcE.Esclevel {
-			return
+			// Have we been here already with an extraloopdepth,
+			// or is the extraloopdepth provided no improvement on
+			// what's already been seen?
+			if srcE.Maxextraloopdepth >= extraloopdepth || srcE.Escloopdepth >= extraloopdepth {
+				return
+			}
+			srcE.Maxextraloopdepth = extraloopdepth
 		}
+	} else { // srcE.Walkgen < e.walkgen -- first time, reset this.
+		srcE.Maxextraloopdepth = NOTALOOPDEPTH
 	}
 
 	srcE.Walkgen = e.walkgen
 	srcE.Esclevel = level
+	modSrcLoopdepth := srcE.Escloopdepth
+
+	if extraloopdepth > modSrcLoopdepth {
+		modSrcLoopdepth = extraloopdepth
+	}
 
 	if Debug['m'] > 1 {
-		fmt.Printf("escwalk: level:%d depth:%d %.*s op=%v %v(%v) scope:%v[%d]\n",
-			level, e.pdepth, e.pdepth, "\t\t\t\t\t\t\t\t\t\t", Oconv(int(src.Op), 0), Nconv(src, obj.FmtShort), Jconv(src, obj.FmtShort), e.curfnSym(src), srcE.Escloopdepth)
+		fmt.Printf("escwalk: level:%d depth:%d %.*s op=%v %v(%v) scope:%v[%d] extraloopdepth=%v\n",
+			level, e.pdepth, e.pdepth, "\t\t\t\t\t\t\t\t\t\t", Oconv(int(src.Op), 0), Nconv(src, obj.FmtShort), Jconv(src, obj.FmtShort), e.curfnSym(src), srcE.Escloopdepth, extraloopdepth)
 	}
 
 	e.pdepth++
@@ -1638,7 +1658,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 		}
 	}
 
-	leaks = level.int() <= 0 && level.guaranteedDereference() <= 0 && dstE.Escloopdepth < srcE.Escloopdepth
+	leaks = level.int() <= 0 && level.guaranteedDereference() <= 0 && dstE.Escloopdepth < modSrcLoopdepth
 
 	switch src.Op {
 	case ONAME:
@@ -1650,7 +1670,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 						Warnl(int(src.Lineno), "leaking param content: %v", Nconv(src, obj.FmtShort))
 					} else {
 						Warnl(int(src.Lineno), "leaking param content: %v level=%v dst.eld=%v src.eld=%v dst=%v",
-							Nconv(src, obj.FmtShort), level, dstE.Escloopdepth, srcE.Escloopdepth, Nconv(dst, obj.FmtShort))
+							Nconv(src, obj.FmtShort), level, dstE.Escloopdepth, modSrcLoopdepth, Nconv(dst, obj.FmtShort))
 					}
 				}
 			} else {
@@ -1660,7 +1680,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 						Warnl(int(src.Lineno), "leaking param: %v", Nconv(src, obj.FmtShort))
 					} else {
 						Warnl(int(src.Lineno), "leaking param: %v level=%v dst.eld=%v src.eld=%v dst=%v",
-							Nconv(src, obj.FmtShort), level, dstE.Escloopdepth, srcE.Escloopdepth, Nconv(dst, obj.FmtShort))
+							Nconv(src, obj.FmtShort), level, dstE.Escloopdepth, modSrcLoopdepth, Nconv(dst, obj.FmtShort))
 					}
 				}
 			}
@@ -1686,15 +1706,17 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 				}
 				if Debug['m'] > 1 {
 					Warnl(int(src.Lineno), "%v escapes to heap, level=%v, dst.eld=%v, src.eld=%v",
-						Nconv(p, obj.FmtShort), level, dstE.Escloopdepth, srcE.Escloopdepth)
+						Nconv(p, obj.FmtShort), level, dstE.Escloopdepth, modSrcLoopdepth)
 				} else {
 					Warnl(int(src.Lineno), "%v escapes to heap", Nconv(p, obj.FmtShort))
 				}
 			}
+			escwalkBody(e, level.dec(), dst, src.Left, modSrcLoopdepth)
+			extraloopdepth = modSrcLoopdepth // passes to recursive case, seems likely a no-op
+		} else {
+			escwalk(e, level.dec(), dst, src.Left)
 		}
 
-		escwalk(e, level.dec(), dst, src.Left)
-
 	case OAPPEND:
 		escwalk(e, level, dst, src.List.N)
 
@@ -1704,6 +1726,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 			if Debug['m'] != 0 {
 				Warnl(int(src.Lineno), "%v escapes to heap", Nconv(src, obj.FmtShort))
 			}
+			extraloopdepth = modSrcLoopdepth
 		}
 		// similar to a slice arraylit and its args.
 		level = level.dec()
@@ -1737,6 +1760,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 			if Debug['m'] != 0 {
 				Warnl(int(src.Lineno), "%v escapes to heap", Nconv(src, obj.FmtShort))
 			}
+			extraloopdepth = modSrcLoopdepth
 		}
 
 	case ODOT,
@@ -1778,7 +1802,7 @@ func escwalk(e *EscState, level Level, dst *Node, src *Node) {
 recurse:
 	level = level.copy()
 	for ll := srcE.Escflowsrc; ll != nil; ll = ll.Next {
-		escwalk(e, level, dst, ll.N)
+		escwalkBody(e, level, dst, ll.N, extraloopdepth)
 	}
 
 	e.pdepth--

test/escape2.go

@@ -1204,7 +1204,7 @@ func foo126() {
 		// loopdepth 1
 		var i int // ERROR "moved to heap: i$"
 		func() {  // ERROR "foo126 func literal does not escape$"
-			px = &i // ERROR "&i escapes to heap$"
+			px = &i // ERROR "&i escapes to heap$"  "leaking closure reference i"
 		}()
 	}
 	_ = px

test/escape2n.go

@@ -1204,7 +1204,7 @@ func foo126() {
 		// loopdepth 1
 		var i int // ERROR "moved to heap: i$"
 		func() {  // ERROR "foo126 func literal does not escape$"
-			px = &i // ERROR "&i escapes to heap$"
+			px = &i // ERROR "&i escapes to heap$" "leaking closure reference i"
 		}()
 	}
 	_ = px

test/fixedbugs/issue13799.go

@@ -0,0 +1,190 @@
+// errorcheck -0 -m -l
+
+// Copyright 2015 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Test, using compiler diagnostic flags, that the escape analysis is working.
+// Compiles but does not run.  Inlining is disabled.
+// Registerization is disabled too (-N), which should
+// have no effect on escape analysis.
+
+package main
+
+import "fmt"
+
+func main() {
+	// Just run test over and over again. This main func is just for
+	// convenience; if test were the main func, we could also trigger
+	// the panic just by running the program over and over again
+	// (sometimes it takes 1 time, sometimes it takes ~4,000+).
+	for iter := 0; ; iter++ {
+		if iter%50 == 0 {
+			fmt.Println(iter) // ERROR "iter escapes to heap$" "main ... argument does not escape$"
+		}
+		test1(iter)
+		test2(iter)
+		test3(iter)
+		test4(iter)
+		test5(iter)
+		test6(iter)
+	}
+}
+
+func test1(iter int) {
+
+	const maxI = 500
+	m := make(map[int][]int) // ERROR "make\(map\[int\]\[\]int\) escapes to heap$"
+
+	// The panic seems to be triggered when m is modified inside a
+	// closure that is both recursively called and reassigned to in a
+	// loop.
+
+	// Cause of bug -- escape of closure failed to escape (shared) data structures
+	// of map.  Assign to fn declared outside of loop triggers escape of closure.
+	// Heap -> stack pointer eventually causes badness when stack reallocation
+	// occurs.
+
+	var fn func()               // ERROR "moved to heap: fn$"
+	for i := 0; i < maxI; i++ { // ERROR "moved to heap: i$"
+		// var fn func() // this makes it work, because fn stays off heap
+		j := 0        // ERROR "moved to heap: j$"
+		fn = func() { // ERROR "func literal escapes to heap$"
+			m[i] = append(m[i], 0) // ERROR "&i escapes to heap$"
+			if j < 25 {            // ERROR "&j escapes to heap$"
+				j++
+				fn() // ERROR "&fn escapes to heap$"
+			}
+		}
+		fn()
+	}
+
+	if len(m) != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, len(m) = %d", iter, maxI, len(m))) // ERROR "iter escapes to heap$" "len\(m\) escapes to heap$" "maxI escapes to heap$" "test1 ... argument does not escape$"
+	}
+}
+
+func test2(iter int) {
+
+	const maxI = 500
+	m := make(map[int][]int) // ERROR "test2 make\(map\[int\]\[\]int\) does not escape$"
+
+	// var fn func()
+	for i := 0; i < maxI; i++ {
+		var fn func() // this makes it work, because fn stays off heap
+		j := 0
+		fn = func() { // ERROR "test2 func literal does not escape$"
+			m[i] = append(m[i], 0)
+			if j < 25 {
+				j++
+				fn()
+			}
+		}
+		fn()
+	}
+
+	if len(m) != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, len(m) = %d", iter, maxI, len(m))) // ERROR "iter escapes to heap$" "len\(m\) escapes to heap$" "maxI escapes to heap$" "test2 ... argument does not escape$"
+	}
+}
+
+func test3(iter int) {
+
+	const maxI = 500
+	var x int // ERROR "moved to heap: x$"
+	m := &x   // ERROR "&x escapes to heap$"
+
+	var fn func() // ERROR "moved to heap: fn$"
+	for i := 0; i < maxI; i++ {
+		// var fn func() // this makes it work, because fn stays off heap
+		j := 0        // ERROR "moved to heap: j$"
+		fn = func() { // ERROR "func literal escapes to heap$"
+			if j < 100 { // ERROR "&j escapes to heap$"
+				j++
+				fn() // ERROR "&fn escapes to heap$"
+			} else {
+				*m = *m + 1
+			}
+		}
+		fn()
+	}
+
+	if *m != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, *m = %d", iter, maxI, *m)) // ERROR "\*m escapes to heap$" "iter escapes to heap$" "maxI escapes to heap$" "test3 ... argument does not escape$"
+	}
+}
+
+func test4(iter int) {
+
+	const maxI = 500
+	var x int
+	m := &x // ERROR "test4 &x does not escape$"
+
+	// var fn func()
+	for i := 0; i < maxI; i++ {
+		var fn func() // this makes it work, because fn stays off heap
+		j := 0
+		fn = func() { // ERROR "test4 func literal does not escape$"
+			if j < 100 {
+				j++
+				fn()
+			} else {
+				*m = *m + 1
+			}
+		}
+		fn()
+	}
+
+	if *m != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, *m = %d", iter, maxI, *m)) // ERROR "\*m escapes to heap$" "iter escapes to heap$" "maxI escapes to heap$" "test4 ... argument does not escape$"
+	}
+}
+
+type str struct {
+	m *int
+}
+
+func recur1(j int, s *str) { // ERROR "recur1 s does not escape"
+	if j < 100 {
+		j++
+		recur1(j, s)
+	} else {
+		*s.m++
+	}
+}
+
+func test5(iter int) {
+
+	const maxI = 500
+	var x int // ERROR "moved to heap: x$"
+	m := &x   // ERROR "&x escapes to heap$"
+
+	var fn *str
+	for i := 0; i < maxI; i++ {
+		// var fn *str // this makes it work, because fn stays off heap
+		fn = &str{m} // ERROR "&str literal escapes to heap"
+		recur1(0, fn)
+	}
+
+	if *m != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, *m = %d", iter, maxI, *m)) // ERROR "\*m escapes to heap$" "iter escapes to heap$" "maxI escapes to heap$" "test5 ... argument does not escape$"
+	}
+}
+
+func test6(iter int) {
+
+	const maxI = 500
+	var x int
+	m := &x // ERROR "&x does not escape$"
+
+	// var fn *str
+	for i := 0; i < maxI; i++ {
+		var fn *str  // this makes it work, because fn stays off heap
+		fn = &str{m} // ERROR "&str literal does not escape"
+		recur1(0, fn)
+	}
+
+	if *m != maxI {
+		panic(fmt.Sprintf("iter %d: maxI = %d, *m = %d", iter, maxI, *m)) // ERROR "\*m escapes to heap$" "iter escapes to heap$" "maxI escapes to heap$" "test6 ... argument does not escape$"
+	}
+}