Commit 1ac637c7 authored by Russ Cox's avatar Russ Cox

cmd/compile: recognize Syscall-like functions for liveness analysis

Consider this code:

	func f(*int)

	func g() {
		p := new(int)
		f(p)
	}

where f is an assembly function.
In general liveness analysis assumes that during the call to f, p is dead
in this frame. If f has retained p, p will be found alive in f's frame and keep
the new(int) from being garbage collected. This is all correct and works.
We use the Go func declaration for f to give the assembly function
liveness information (the arguments are assumed live for the entire call).

Now consider this code:

	func h1() {
		p := new(int)
		syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p)))
	}

Here syscall.Syscall is taking the place of f, but because its arguments
are uintptr, the liveness analysis and the garbage collector ignore them.
Since p is no longer live in h once the call starts, if the garbage collector
scans the stack while the system call is blocked, it will find no reference
to the new(int) and reclaim it. If the kernel is going to write to *p once
the call finishes, reclaiming the memory is a mistake.

We can't change the arguments or the liveness information for
syscall.Syscall itself, both for compatibility and because sometimes the
arguments really are integers, and the garbage collector will get quite upset
if it finds an integer where it expects a pointer. The problem is that
these arguments are fundamentally untyped.

The solution we have taken in the syscall package's wrappers in past
releases is to insert a call to a dummy function named "use", to make
it look like the argument is live during the call to syscall.Syscall:

	func h2() {
		p := new(int)
		syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p)))
		use(unsafe.Pointer(p))
	}

Keeping p alive during the call means that if the garbage collector
scans the stack during the system call now, it will find the reference to p.

Unfortunately, this approach is not available to users outside syscall,
because 'use' is unexported, and people also have to realize they need
to use it and do so. There is much existing code using syscall.Syscall
without a 'use'-like function. That code will fail very occasionally in
mysterious ways (see #13372).

This CL fixes all that existing code by making the compiler do the right
thing automatically, without any code modifications. That is, it takes h1
above, which is incorrect code today, and makes it correct code.

Specifically, if the compiler sees a foreign func definition (one
without a body) that has uintptr arguments, it marks those arguments
as "unsafe uintptrs". If it later sees the function being called
with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x
as having escaped, and it makes sure to hold x in a live temporary
variable until the call returns, so that the garbage collector cannot
reclaim whatever heap memory x points to.

For now I am leaving the explicit calls to use in package syscall,
but they can be removed early in a future cycle (likely Go 1.7).

The rule has no effect on escape analysis, only on liveness analysis.

Fixes #13372.

Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500
Reviewed-on: https://go-review.googlesource.com/18584Reviewed-by: default avatarIan Lance Taylor <iant@golang.org>
parent 66330d8c
......@@ -510,6 +510,13 @@ the stack pointer may change during any function call:
even pointers to stack data must not be kept in local variables.
</p>
<p>
Assembly functions should always be given Go prototypes,
both to provide pointer information for the arguments and results
and to let <code>go</code> <code>vet</code> check that
the offsets being used to access them are correct.
</p>
<h2 id="architectures">Architecture-specific details</h2>
<p>
......
......@@ -34,6 +34,7 @@ var progtable = [x86.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the Intel opcode.
......
......@@ -33,6 +33,7 @@ var progtable = [arm.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the Intel opcode.
......
......@@ -34,6 +34,7 @@ var progtable = [arm64.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the Power opcode.
......
......@@ -1808,6 +1808,13 @@ recurse:
e.pdepth--
}
// This special tag is applied to uintptr variables
// that we believe may hold unsafe.Pointers for
// calls into assembly functions.
// It is logically a constant, but using a var
// lets us take the address below to get a *string.
var unsafeUintptrTag = "unsafe-uintptr"
func esctag(e *EscState, func_ *Node) {
func_.Esc = EscFuncTagged
......@@ -1822,6 +1829,29 @@ func esctag(e *EscState, func_ *Node) {
}
}
// Assume that uintptr arguments must be held live across the call.
// This is most important for syscall.Syscall.
// See golang.org/issue/13372.
// This really doesn't have much to do with escape analysis per se,
// but we are reusing the ability to annotate an individual function
// argument and pass those annotations along to importing code.
narg := 0
for t := getinargx(func_.Type).Type; t != nil; t = t.Down {
narg++
if t.Type.Etype == TUINTPTR {
if Debug['m'] != 0 {
var name string
if t.Sym != nil {
name = t.Sym.Name
} else {
name = fmt.Sprintf("arg#%d", narg)
}
Warnl(int(func_.Lineno), "%v assuming %v is unsafe uintptr", funcSym(func_), name)
}
t.Note = &unsafeUintptrTag
}
}
return
}
......
......@@ -605,6 +605,9 @@ func Tempname(nn *Node, t *Type) {
n.Esc = EscNever
n.Name.Curfn = Curfn
Curfn.Func.Dcl = list(Curfn.Func.Dcl, n)
if Debug['h'] != 0 {
println("H", n, n.Orig, funcSym(Curfn).Name)
}
dowidth(t)
n.Xoffset = 0
......@@ -868,6 +871,9 @@ func gen(n *Node) {
case OVARKILL:
gvarkill(n.Left)
case OVARLIVE:
gvarlive(n.Left)
}
ret:
......
......@@ -185,7 +185,7 @@ func fixautoused(p *obj.Prog) {
continue
}
if (p.As == obj.AVARDEF || p.As == obj.AVARKILL) && p.To.Node != nil && !((p.To.Node).(*Node)).Used {
if (p.As == obj.AVARDEF || p.As == obj.AVARKILL || p.As == obj.AVARLIVE) && p.To.Node != nil && !((p.To.Node).(*Node)).Used {
// Cannot remove VARDEF instruction, because - unlike TYPE handled above -
// VARDEFs are interspersed with other code, and a jump might be using the
// VARDEF as a target. Replace with a no-op instead. A later pass will remove
......
......@@ -243,6 +243,12 @@ func cleantempnopop(mark *NodeList, order *Order, out **NodeList) {
var kill *Node
for l := order.temp; l != mark; l = l.Next {
if l.N.Name.Keepalive {
l.N.Name.Keepalive = false
kill = Nod(OVARLIVE, l.N, nil)
typecheck(&kill, Etop)
*out = list(*out, kill)
}
kill = Nod(OVARKILL, l.N, nil)
typecheck(&kill, Etop)
*out = list(*out, kill)
......@@ -375,6 +381,28 @@ func ordercall(n *Node, order *Order) {
orderexpr(&n.Left, order, nil)
orderexpr(&n.Right, order, nil) // ODDDARG temp
ordercallargs(&n.List, order)
if n.Op == OCALLFUNC {
for l, t := n.List, getinargx(n.Left.Type).Type; l != nil && t != nil; l, t = l.Next, t.Down {
// Check for "unsafe-uintptr" tag provided by escape analysis.
// If present and the argument is really a pointer being converted
// to uintptr, arrange for the pointer to be kept alive until the call
// returns, by copying it into a temp and marking that temp
// still alive when we pop the temp stack.
if t.Note != nil && *t.Note == unsafeUintptrTag {
xp := &l.N
for (*xp).Op == OCONVNOP && !Isptr[(*xp).Type.Etype] {
xp = &(*xp).Left
}
x := *xp
if Isptr[x.Type.Etype] {
x = ordercopyexpr(x, x.Type, order, 0)
x.Name.Keepalive = true
*xp = x
}
}
}
}
}
// Ordermapassign appends n to order->out, introducing temporaries
......@@ -464,7 +492,7 @@ func orderstmt(n *Node, order *Order) {
default:
Fatalf("orderstmt %v", Oconv(int(n.Op), 0))
case OVARKILL:
case OVARKILL, OVARLIVE:
order.out = list(order.out, n)
case OAS:
......
......@@ -94,7 +94,11 @@ func gvardefx(n *Node, as int) {
switch n.Class {
case PAUTO, PPARAM, PPARAMOUT:
Thearch.Gins(as, nil, n)
if as == obj.AVARLIVE {
Thearch.Gins(as, n, nil)
} else {
Thearch.Gins(as, nil, n)
}
}
}
......@@ -106,13 +110,17 @@ func gvarkill(n *Node) {
gvardefx(n, obj.AVARKILL)
}
func gvarlive(n *Node) {
gvardefx(n, obj.AVARLIVE)
}
func removevardef(firstp *obj.Prog) {
for p := firstp; p != nil; p = p.Link {
for p.Link != nil && (p.Link.As == obj.AVARDEF || p.Link.As == obj.AVARKILL) {
for p.Link != nil && (p.Link.As == obj.AVARDEF || p.Link.As == obj.AVARKILL || p.Link.As == obj.AVARLIVE) {
p.Link = p.Link.Link
}
if p.To.Type == obj.TYPE_BRANCH {
for p.To.Val.(*obj.Prog) != nil && (p.To.Val.(*obj.Prog).As == obj.AVARDEF || p.To.Val.(*obj.Prog).As == obj.AVARKILL) {
for p.To.Val.(*obj.Prog) != nil && (p.To.Val.(*obj.Prog).As == obj.AVARDEF || p.To.Val.(*obj.Prog).As == obj.AVARKILL || p.To.Val.(*obj.Prog).As == obj.AVARLIVE) {
p.To.Val = p.To.Val.(*obj.Prog).Link
}
}
......
......@@ -806,7 +806,7 @@ func checkauto(fn *Node, p *obj.Prog, n *Node) {
return
}
fmt.Printf("checkauto %v: %v (%p; class=%d) not found in %v\n", Curfn, n, n, n.Class, p)
fmt.Printf("checkauto %v: %v (%p; class=%d) not found in %p %v\n", funcSym(Curfn), n, n, n.Class, p, p)
for l := fn.Func.Dcl; l != nil; l = l.Next {
fmt.Printf("\t%v (%p; class=%d)\n", l.N, l.N, l.N.Class)
}
......
......@@ -143,7 +143,7 @@ func instrumentnode(np **Node, init **NodeList, wr int, skip int) {
goto ret
// can't matter
case OCFUNC, OVARKILL:
case OCFUNC, OVARKILL, OVARLIVE:
goto ret
case OBLOCK:
......
......@@ -1073,6 +1073,9 @@ func regopt(firstp *obj.Prog) {
for f := firstf; f != nil; f = f.Link {
p := f.Prog
// AVARLIVE must be considered a use, do not skip it.
// Otherwise the variable will be optimized away,
// and the whole point of AVARLIVE is to keep it on the stack.
if p.As == obj.AVARDEF || p.As == obj.AVARKILL {
continue
}
......
......@@ -128,6 +128,7 @@ type Name struct {
Captured bool // is the variable captured by a closure
Byval bool // is the variable captured by value or by reference
Needzero bool // if it contains pointers, needs to be zeroed on function entry
Keepalive bool // mark value live across unknown assembly call
}
type Param struct {
......@@ -342,6 +343,7 @@ const (
OCFUNC // reference to c function pointer (not go func value)
OCHECKNIL // emit code to ensure pointer/interface not nil
OVARKILL // variable is dead
OVARLIVE // variable is alive
// thearch-specific registers
OREGISTER // a register, such as AX.
......
......@@ -2023,7 +2023,8 @@ OpSwitch:
OEMPTY,
OGOTO,
OXFALL,
OVARKILL:
OVARKILL,
OVARLIVE:
ok |= Etop
break OpSwitch
......
......@@ -216,7 +216,8 @@ func walkstmt(np **Node) {
ODCLCONST,
ODCLTYPE,
OCHECKNIL,
OVARKILL:
OVARKILL,
OVARLIVE:
break
case OBLOCK:
......
......@@ -34,6 +34,7 @@ var progtable = [mips.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the MIPS opcode.
......
......@@ -34,6 +34,7 @@ var progtable = [ppc64.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the Power opcode.
......
......@@ -40,6 +40,7 @@ var progtable = [x86.ALAST]obj.ProgInfo{
obj.ACHECKNIL: {Flags: gc.LeftRead},
obj.AVARDEF: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARKILL: {Flags: gc.Pseudo | gc.RightWrite},
obj.AVARLIVE: {Flags: gc.Pseudo | gc.LeftRead},
// NOP is an internal no-op that also stands
// for USED and SET annotations, not the Intel opcode.
......
......@@ -282,6 +282,7 @@ const (
AUSEFIELD
AVARDEF
AVARKILL
AVARLIVE
A_ARCHSPECIFIC
)
......
......@@ -608,7 +608,7 @@ func RegisterOpcode(lo int, Anames []string) {
}
func Aconv(a int) string {
if a < A_ARCHSPECIFIC {
if 0 <= a && a < len(Anames) {
return Anames[a]
}
for i := range aSpace {
......@@ -639,6 +639,7 @@ var Anames = []string{
"UNDEF",
"USEFIELD",
"VARDEF",
"VARLIVE",
"VARKILL",
}
......
......@@ -95,5 +95,8 @@ func (tv *Timeval) Nano() int64 {
// use is a no-op, but the compiler cannot see that it is.
// Calling use(p) ensures that p is kept live until that point.
// This was needed until Go 1.6 to call syscall.Syscall correctly.
// As of Go 1.6 the compiler handles that case automatically.
// The uses and definition of use can be removed early in the Go 1.7 cycle.
//go:noescape
func use(p unsafe.Pointer)
// errorcheck -0 -m -live
// +build !windows
// 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 escape analysis and liveness inferred for syscall.Syscall-like functions.
package p
import (
"syscall"
"unsafe"
)
func f(uintptr) // ERROR "f assuming arg#1 is unsafe uintptr"
func g() {
var t int
f(uintptr(unsafe.Pointer(&t))) // ERROR "live at call to f: autotmp" "g &t does not escape"
}
func h() {
var v int
syscall.Syscall(0, 1, uintptr(unsafe.Pointer(&v)), 2) // ERROR "live at call to Syscall: autotmp" "h &v does not escape"
}
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment