cmd/compile, cmd/link, runtime: make defers low-cost through inline code and extra funcdata

Generate inline code at defer time to save the args of defer calls to unique
(autotmp) stack slots, and generate inline code at exit time to check which defer
calls were made and make the associated function/method/interface calls. We
remember that a particular defer statement was reached by storing in the deferBits
variable (always stored on the stack). At exit time, we check the bits of the
deferBits variable to determine which defer function calls to make (in reverse
order). These low-cost defers are only used for functions where no defers
appear in loops. In addition, we don't do these low-cost defers if there are too
many defer statements or too many exits in a function (to limit code increase).

When a function uses open-coded defers, we produce extra
FUNCDATA_OpenCodedDeferInfo information that specifies the number of defers, and
for each defer, the stack slots where the closure and associated args have been
stored. The funcdata also includes the location of the deferBits variable.
Therefore, for panics, we can use this funcdata to determine exactly which defers
are active, and call the appropriate functions/methods/closures with the correct
arguments for each active defer.

In order to unwind the stack correctly after a recover(), we need to add an extra
code segment to functions with open-coded defers that simply calls deferreturn()
and returns. This segment is not reachable by the normal function, but is returned
to by the runtime during recovery. We set the liveness information of this
deferreturn() to be the same as the liveness at the first function call during the
last defer exit code (so all return values and all stack slots needed by the defer
calls will be live).

I needed to increase the stackguard constant from 880 to 896, because of a small
amount of new code in deferreturn().

The -N flag disables open-coded defers. '-d defer' prints out the kind of defer
being used at each defer statement (heap-allocated, stack-allocated, or
open-coded).

Cost of defer statement  [ go test -run NONE -bench BenchmarkDefer$ runtime ]
  With normal (stack-allocated) defers only:         35.4  ns/op
  With open-coded defers:                             5.6  ns/op
  Cost of function call alone (remove defer keyword): 4.4  ns/op

Text size increase (including funcdata) for go cmd without/with open-coded defers:  0.09%

The average size increase (including funcdata) for only the functions that use
open-coded defers is 1.1%.

The cost of a panic followed by a recover got noticeably slower, since panic
processing now requires a scan of the stack for open-coded defer frames. This scan
is required, even if no frames are using open-coded defers:

Cost of panic and recover [ go test -run NONE -bench BenchmarkPanicRecover runtime ]
  Without open-coded defers:        62.0 ns/op
  With open-coded defers:           255  ns/op

A CGO Go-to-C-to-Go benchmark got noticeably faster because of open-coded defers:

CGO Go-to-C-to-Go benchmark [cd misc/cgo/test; go test -run NONE -bench BenchmarkCGoCallback ]
  Without open-coded defers:        443 ns/op
  With open-coded defers:           347 ns/op

Updates #14939 (defer performance)
Updates #34481 (design doc)

Change-Id: I51a389860b9676cfa1b84722f5fb84d3c4ee9e28
Reviewed-on: https://go-review.googlesource.com/c/go/+/190098Reviewed-by: Austin Clements <austin@google.com>

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

@@ -371,6 +371,7 @@ func (e *Escape) stmt(n *Node) {
 		e.stmts(n.Right.Ninit)
 		e.call(e.addrs(n.List), n.Right, nil)
 	case ORETURN:
+		e.curfn.Func.numReturns++
 		results := e.curfn.Type.Results().FieldSlice()
 		for i, v := range n.List.Slice() {
 			e.assign(asNode(results[i].Nname), v, "return", n)
@@ -378,6 +379,16 @@ func (e *Escape) stmt(n *Node) {
 	case OCALLFUNC, OCALLMETH, OCALLINTER, OCLOSE, OCOPY, ODELETE, OPANIC, OPRINT, OPRINTN, ORECOVER:
 		e.call(nil, n, nil)
 	case OGO, ODEFER:
+		if n.Op == ODEFER {
+			e.curfn.Func.SetHasDefer(true)
+			e.curfn.Func.numDefers++
+			if e.curfn.Func.numDefers > maxOpenDefers {
+				// Don't allow open defers if there are more than
+				// 8 defers in the function, since we use a single
+				// byte to record active defers.
+				e.curfn.Func.SetOpenCodedDeferDisallowed(true)
+			}
+		}
 		e.stmts(n.Left.Ninit)
 		e.call(nil, n.Left, n)
 
@@ -872,8 +883,13 @@ func (e *Escape) augmentParamHole(k EscHole, where *Node) EscHole {
 	// non-transient location to avoid arguments from being
 	// transiently allocated.
 	if where.Op == ODEFER && e.loopDepth == 1 {
-		where.Esc = EscNever // force stack allocation of defer record (see ssa.go)
+		// force stack allocation of defer record, unless open-coded
+		// defers are used (see ssa.go)
+		where.Esc = EscNever
 		return e.later(k)
+	} else if where.Op == ODEFER {
+		// If any defer occurs in a loop, open-coded defers cannot be used
+		e.curfn.Func.SetOpenCodedDeferDisallowed(true)
 	}
 
 	return e.heapHole()

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

@@ -52,6 +52,7 @@ var (
 	Debug_typecheckinl int
 	Debug_gendwarfinl  int
 	Debug_softfloat    int
+	Debug_defer        int
 )
 
 // Debug arguments.
@@ -81,6 +82,7 @@ var debugtab = []struct {
 	{"typecheckinl", "eager typechecking of inline function bodies", &Debug_typecheckinl},
 	{"dwarfinl", "print information about DWARF inlined function creation", &Debug_gendwarfinl},
 	{"softfloat", "force compiler to emit soft-float code", &Debug_softfloat},
+	{"defer", "print information about defer compilation", &Debug_defer},
 }
 
 const debugHelpHeader = `usage: -d arg[,arg]* and arg is <key>[=<value>]

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

@@ -294,6 +294,9 @@ func addGCLocals() {
 			}
 			ggloblsym(x, int32(len(x.P)), attr)
 		}
+		if x := s.Func.OpenCodedDeferInfo; x != nil {
+			ggloblsym(x, int32(len(x.P)), obj.RODATA|obj.DUPOK)
+		}
 	}
 }
 

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

@@ -338,6 +338,7 @@ func deferstruct(stksize int64) *types.Type {
 		makefield("siz", types.Types[TUINT32]),
 		makefield("started", types.Types[TBOOL]),
 		makefield("heap", types.Types[TBOOL]),
+		makefield("openDefer", types.Types[TBOOL]),
 		makefield("sp", types.Types[TUINTPTR]),
 		makefield("pc", types.Types[TUINTPTR]),
 		// Note: the types here don't really matter. Defer structures
@@ -346,6 +347,9 @@ func deferstruct(stksize int64) *types.Type {
 		makefield("fn", types.Types[TUINTPTR]),
 		makefield("_panic", types.Types[TUINTPTR]),
 		makefield("link", types.Types[TUINTPTR]),
+		makefield("framepc", types.Types[TUINTPTR]),
+		makefield("varp", types.Types[TUINTPTR]),
+		makefield("fd", types.Types[TUINTPTR]),
 		makefield("args", argtype),
 	}
 

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

@@ -20,7 +20,7 @@ func TestSizeof(t *testing.T) {
 		_32bit uintptr     // size on 32bit platforms
 		_64bit uintptr     // size on 64bit platforms
 	}{
-		{Func{}, 116, 208},
+		{Func{}, 124, 224},
 		{Name{}, 32, 56},
 		{Param{}, 24, 48},
 		{Node{}, 76, 128},

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

@@ -491,7 +491,9 @@ type Func struct {
 
 	Pragma syntax.Pragma // go:xxx function annotations
 
-	flags bitset16
+	flags      bitset16
+	numDefers  int // number of defer calls in the function
+	numReturns int // number of explicit returns in the function
 
 	// nwbrCalls records the LSyms of functions called by this
 	// function for go:nowritebarrierrec analysis. Only filled in
@@ -527,34 +529,37 @@ const (
 	funcNeedctxt                  // function uses context register (has closure variables)
 	funcReflectMethod             // function calls reflect.Type.Method or MethodByName
 	funcIsHiddenClosure
-	funcHasDefer            // contains a defer statement
-	funcNilCheckDisabled    // disable nil checks when compiling this function
-	funcInlinabilityChecked // inliner has already determined whether the function is inlinable
-	funcExportInline        // include inline body in export data
-	funcInstrumentBody      // add race/msan instrumentation during SSA construction
+	funcHasDefer                 // contains a defer statement
+	funcNilCheckDisabled         // disable nil checks when compiling this function
+	funcInlinabilityChecked      // inliner has already determined whether the function is inlinable
+	funcExportInline             // include inline body in export data
+	funcInstrumentBody           // add race/msan instrumentation during SSA construction
+	funcOpenCodedDeferDisallowed // can't do open-coded defers
 )
 
-func (f *Func) Dupok() bool               { return f.flags&funcDupok != 0 }
-func (f *Func) Wrapper() bool             { return f.flags&funcWrapper != 0 }
-func (f *Func) Needctxt() bool            { return f.flags&funcNeedctxt != 0 }
-func (f *Func) ReflectMethod() bool       { return f.flags&funcReflectMethod != 0 }
-func (f *Func) IsHiddenClosure() bool     { return f.flags&funcIsHiddenClosure != 0 }
-func (f *Func) HasDefer() bool            { return f.flags&funcHasDefer != 0 }
-func (f *Func) NilCheckDisabled() bool    { return f.flags&funcNilCheckDisabled != 0 }
-func (f *Func) InlinabilityChecked() bool { return f.flags&funcInlinabilityChecked != 0 }
-func (f *Func) ExportInline() bool        { return f.flags&funcExportInline != 0 }
-func (f *Func) InstrumentBody() bool      { return f.flags&funcInstrumentBody != 0 }
-
-func (f *Func) SetDupok(b bool)               { f.flags.set(funcDupok, b) }
-func (f *Func) SetWrapper(b bool)             { f.flags.set(funcWrapper, b) }
-func (f *Func) SetNeedctxt(b bool)            { f.flags.set(funcNeedctxt, b) }
-func (f *Func) SetReflectMethod(b bool)       { f.flags.set(funcReflectMethod, b) }
-func (f *Func) SetIsHiddenClosure(b bool)     { f.flags.set(funcIsHiddenClosure, b) }
-func (f *Func) SetHasDefer(b bool)            { f.flags.set(funcHasDefer, b) }
-func (f *Func) SetNilCheckDisabled(b bool)    { f.flags.set(funcNilCheckDisabled, b) }
-func (f *Func) SetInlinabilityChecked(b bool) { f.flags.set(funcInlinabilityChecked, b) }
-func (f *Func) SetExportInline(b bool)        { f.flags.set(funcExportInline, b) }
-func (f *Func) SetInstrumentBody(b bool)      { f.flags.set(funcInstrumentBody, b) }
+func (f *Func) Dupok() bool                    { return f.flags&funcDupok != 0 }
+func (f *Func) Wrapper() bool                  { return f.flags&funcWrapper != 0 }
+func (f *Func) Needctxt() bool                 { return f.flags&funcNeedctxt != 0 }
+func (f *Func) ReflectMethod() bool            { return f.flags&funcReflectMethod != 0 }
+func (f *Func) IsHiddenClosure() bool          { return f.flags&funcIsHiddenClosure != 0 }
+func (f *Func) HasDefer() bool                 { return f.flags&funcHasDefer != 0 }
+func (f *Func) NilCheckDisabled() bool         { return f.flags&funcNilCheckDisabled != 0 }
+func (f *Func) InlinabilityChecked() bool      { return f.flags&funcInlinabilityChecked != 0 }
+func (f *Func) ExportInline() bool             { return f.flags&funcExportInline != 0 }
+func (f *Func) InstrumentBody() bool           { return f.flags&funcInstrumentBody != 0 }
+func (f *Func) OpenCodedDeferDisallowed() bool { return f.flags&funcOpenCodedDeferDisallowed != 0 }
+
+func (f *Func) SetDupok(b bool)                    { f.flags.set(funcDupok, b) }
+func (f *Func) SetWrapper(b bool)                  { f.flags.set(funcWrapper, b) }
+func (f *Func) SetNeedctxt(b bool)                 { f.flags.set(funcNeedctxt, b) }
+func (f *Func) SetReflectMethod(b bool)            { f.flags.set(funcReflectMethod, b) }
+func (f *Func) SetIsHiddenClosure(b bool)          { f.flags.set(funcIsHiddenClosure, b) }
+func (f *Func) SetHasDefer(b bool)                 { f.flags.set(funcHasDefer, b) }
+func (f *Func) SetNilCheckDisabled(b bool)         { f.flags.set(funcNilCheckDisabled, b) }
+func (f *Func) SetInlinabilityChecked(b bool)      { f.flags.set(funcInlinabilityChecked, b) }
+func (f *Func) SetExportInline(b bool)             { f.flags.set(funcExportInline, b) }
+func (f *Func) SetInstrumentBody(b bool)           { f.flags.set(funcInstrumentBody, b) }
+func (f *Func) SetOpenCodedDeferDisallowed(b bool) { f.flags.set(funcOpenCodedDeferDisallowed, b) }
 
 func (f *Func) setWBPos(pos src.XPos) {
 	if Debug_wb != 0 {

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

@@ -213,7 +213,6 @@ func walkstmt(n *Node) *Node {
 		yyerror("case statement out of place")
 
 	case ODEFER:
-		Curfn.Func.SetHasDefer(true)
 		fallthrough
 	case OGO:
 		switch n.Left.Op {

src/cmd/compile/internal/ssa/deadstore.go

@@ -170,6 +170,11 @@ func elimDeadAutosGeneric(f *Func) {
 			return
 		case OpVarLive:
 			// Don't delete the auto if it needs to be kept alive.
+
+			// We depend on this check to keep the autotmp stack slots
+			// for open-coded defers from being removed (since they
+			// may not be used by the inline code, but will be used by
+			// panic processing).
 			n, ok := v.Aux.(GCNode)
 			if !ok || n.StorageClass() != ClassAuto {
 				return

src/cmd/compile/internal/ssa/func.go

@@ -32,8 +32,16 @@ type Func struct {
 	Type   *types.Type // type signature of the function.
 	Blocks []*Block    // unordered set of all basic blocks (note: not indexable by ID)
 	Entry  *Block      // the entry basic block
-	bid    idAlloc     // block ID allocator
-	vid    idAlloc     // value ID allocator
+
+	// If we are using open-coded defers, this is the first call to a deferred
+	// function in the final defer exit sequence that we generated. This call
+	// should be after all defer statements, and will have all args, etc. of
+	// all defer calls as live. The liveness info of this call will be used
+	// for the deferreturn/ret segment generated for functions with open-coded
+	// defers.
+	LastDeferExit *Value
+	bid           idAlloc // block ID allocator
+	vid           idAlloc // value ID allocator
 
 	// Given an environment variable used for debug hash match,
 	// what file (if any) receives the yes/no logging?

src/cmd/internal/obj/link.go

@@ -405,10 +405,11 @@ type FuncInfo struct {
 	dwarfAbsFnSym      *LSym
 	dwarfDebugLinesSym *LSym
 
-	GCArgs       *LSym
-	GCLocals     *LSym
-	GCRegs       *LSym
-	StackObjects *LSym
+	GCArgs             *LSym
+	GCLocals           *LSym
+	GCRegs             *LSym
+	StackObjects       *LSym
+	OpenCodedDeferInfo *LSym
 }
 
 type InlMark struct {

src/cmd/internal/objabi/funcdata.go

@@ -15,11 +15,12 @@ const (
 	PCDATA_StackMapIndex = 1
 	PCDATA_InlTreeIndex  = 2
 
-	FUNCDATA_ArgsPointerMaps   = 0
-	FUNCDATA_LocalsPointerMaps = 1
-	FUNCDATA_RegPointerMaps    = 2
-	FUNCDATA_StackObjects      = 3
-	FUNCDATA_InlTree           = 4
+	FUNCDATA_ArgsPointerMaps    = 0
+	FUNCDATA_LocalsPointerMaps  = 1
+	FUNCDATA_RegPointerMaps     = 2
+	FUNCDATA_StackObjects       = 3
+	FUNCDATA_InlTree            = 4
+	FUNCDATA_OpenCodedDeferInfo = 5
 
 	// ArgsSizeUnknown is set in Func.argsize to mark all functions
 	// whose argument size is unknown (C vararg functions, and

src/cmd/internal/objabi/funcid.go

@@ -85,6 +85,12 @@ func GetFuncID(name, file string) FuncID {
 		return FuncID_panicwrap
 	case "runtime.handleAsyncEvent":
 		return FuncID_handleAsyncEvent
+	case "runtime.deferreturn":
+		// Don't show in the call stack (used when invoking defer functions)
+		return FuncID_wrapper
+	case "runtime.runOpenDeferFrame":
+		// Don't show in the call stack (used when invoking defer functions)
+		return FuncID_wrapper
 	}
 	if file == "<autogenerated>" {
 		return FuncID_wrapper

src/cmd/internal/objabi/stack.go

@@ -18,7 +18,7 @@ const (
 )
 
 // Initialize StackGuard and StackLimit according to target system.
-var StackGuard = 880*stackGuardMultiplier() + StackSystem
+var StackGuard = 896*stackGuardMultiplier() + StackSystem
 var StackLimit = StackGuard - StackSystem - StackSmall
 
 // stackGuardMultiplier returns a multiplier to apply to the default

src/cmd/link/internal/ld/pcln.go

@@ -11,6 +11,7 @@ import (
 	"cmd/internal/sys"
 	"cmd/link/internal/sym"
 	"encoding/binary"
+	"fmt"
 	"log"
 	"os"
 	"path/filepath"
@@ -255,13 +256,23 @@ func (ctxt *Link) pclntab() {
 			}
 			if r.Type.IsDirectJump() && r.Sym != nil && r.Sym.Name == "runtime.deferreturn" {
 				if ctxt.Arch.Family == sys.Wasm {
-					deferreturn = lastWasmAddr
+					deferreturn = lastWasmAddr - 1
 				} else {
 					// Note: the relocation target is in the call instruction, but
 					// is not necessarily the whole instruction (for instance, on
 					// x86 the relocation applies to bytes [1:5] of the 5 byte call
 					// instruction).
 					deferreturn = uint32(r.Off)
+					switch ctxt.Arch.Family {
+					case sys.AMD64, sys.I386, sys.MIPS, sys.MIPS64, sys.RISCV64:
+						deferreturn--
+					case sys.PPC64, sys.ARM, sys.ARM64:
+						// no change
+					case sys.S390X:
+						deferreturn -= 2
+					default:
+						panic(fmt.Sprint("Unhandled architecture:", ctxt.Arch.Family))
+					}
 				}
 				break // only need one
 			}

src/cmd/link/internal/ld/symtab.go

@@ -498,7 +498,8 @@ func (ctxt *Link) symtab() {
 		case strings.HasPrefix(s.Name, "gcargs."),
 			strings.HasPrefix(s.Name, "gclocals."),
 			strings.HasPrefix(s.Name, "gclocals·"),
-			strings.HasPrefix(s.Name, "inltree."):
+			strings.HasPrefix(s.Name, "inltree."),
+			strings.HasSuffix(s.Name, ".opendefer"):
 			s.Type = sym.SGOFUNC
 			s.Attr |= sym.AttrNotInSymbolTable
 			s.Outer = symgofunc

src/runtime/callers_test.go

@@ -5,6 +5,7 @@
 package runtime_test
 
 import (
+	"internal/race"
 	"reflect"
 	"runtime"
 	"strings"
@@ -12,19 +13,19 @@ import (
 )
 
 func f1(pan bool) []uintptr {
-	return f2(pan) // line 15
+	return f2(pan) // line 16
 }
 
 func f2(pan bool) []uintptr {
-	return f3(pan) // line 19
+	return f3(pan) // line 20
 }
 
 func f3(pan bool) []uintptr {
 	if pan {
-		panic("f3") // line 24
+		panic("f3") // line 25
 	}
 	ret := make([]uintptr, 20)
-	return ret[:runtime.Callers(0, ret)] // line 27
+	return ret[:runtime.Callers(0, ret)] // line 28
 }
 
 func testCallers(t *testing.T, pcs []uintptr, pan bool) {
@@ -48,16 +49,16 @@ func testCallers(t *testing.T, pcs []uintptr, pan bool) {
 
 	var f3Line int
 	if pan {
-		f3Line = 24
+		f3Line = 25
 	} else {
-		f3Line = 27
+		f3Line = 28
 	}
 	want := []struct {
 		name string
 		line int
 	}{
-		{"f1", 15},
-		{"f2", 19},
+		{"f1", 16},
+		{"f2", 20},
 		{"f3", f3Line},
 	}
 	for _, w := range want {
@@ -188,3 +189,36 @@ func TestCallersDivZeroPanic(t *testing.T) {
 		t.Fatal("did not see divide-by-sizer panic")
 	}
 }
+
+// This test will have a slightly different callstack if non-open-coded defers are
+// called (e.g. if race checks enabled), because of a difference in the way the
+// defer function is invoked.
+func TestCallersDeferNilFuncPanic(t *testing.T) {
+	if race.Enabled {
+		t.Skip("skipping TestCallersDeferNilFuncPanic under race detector")
+	}
+	// Make sure we don't have any extra frames on the stack (due to
+	// open-coded defer processing)
+	state := 1
+	want := []string{"runtime.Callers", "runtime_test.TestCallersDeferNilFuncPanic.func1",
+		"runtime.gopanic", "runtime.panicmem", "runtime.sigpanic",
+		"runtime_test.TestCallersDeferNilFuncPanic"}
+
+	defer func() {
+		if r := recover(); r == nil {
+			t.Fatal("did not panic")
+		}
+		pcs := make([]uintptr, 20)
+		pcs = pcs[:runtime.Callers(0, pcs)]
+		testCallersEqual(t, pcs, want)
+		if state == 1 {
+			t.Fatal("nil defer func panicked at defer time rather than function exit time")
+		}
+
+	}()
+	var f func()
+	defer f()
+	// Use the value of 'state' to make sure nil defer func f causes panic at
+	// function exit, rather than at the defer statement.
+	state = 2
+}

src/runtime/defer_test.go

@@ -15,11 +15,11 @@ import (
 // unconditional panic (hence no return from the function)
 func TestUnconditionalPanic(t *testing.T) {
 	defer func() {
-		if recover() == nil {
+		if recover() != "testUnconditional" {
 			t.Fatal("expected unconditional panic")
 		}
 	}()
-	panic("panic should be recovered")
+	panic("testUnconditional")
 }
 
 var glob int = 3
@@ -30,7 +30,7 @@ func TestOpenAndNonOpenDefers(t *testing.T) {
 	for {
 		// Non-open defer because in a loop
 		defer func(n int) {
-			if recover() == nil {
+			if recover() != "testNonOpenDefer" {
 				t.Fatal("expected testNonOpen panic")
 			}
 		}(3)
@@ -45,7 +45,7 @@ func TestOpenAndNonOpenDefers(t *testing.T) {
 //go:noinline
 func testOpen(t *testing.T, arg int) {
 	defer func(n int) {
-		if recover() == nil {
+		if recover() != "testOpenDefer" {
 			t.Fatal("expected testOpen panic")
 		}
 	}(4)
@@ -61,7 +61,7 @@ func TestNonOpenAndOpenDefers(t *testing.T) {
 	for {
 		// Non-open defer because in a loop
 		defer func(n int) {
-			if recover() == nil {
+			if recover() != "testNonOpenDefer" {
 				t.Fatal("expected testNonOpen panic")
 			}
 		}(3)
@@ -80,7 +80,7 @@ func TestConditionalDefers(t *testing.T) {
 	list = make([]int, 0, 10)
 
 	defer func() {
-		if recover() == nil {
+		if recover() != "testConditional" {
 			t.Fatal("expected panic")
 		}
 		want := []int{4, 2, 1}
@@ -106,7 +106,7 @@ func testConditionalDefers(n int) {
 			defer doappend(4)
 		}
 	}
-	panic("test")
+	panic("testConditional")
 }
 
 // Test that there is no compile-time or run-time error if an open-coded defer
@@ -174,3 +174,52 @@ func TestRecoverMatching(t *testing.T) {
 	}()
 	panic("panic1")
 }
+
+type nonSSAable [128]byte
+
+type bigStruct struct {
+	x, y, z, w, p, q int64
+}
+
+func mknonSSAable() nonSSAable {
+	globint1++
+	return nonSSAable{0, 0, 0, 0, 5}
+}
+
+var globint1, globint2 int
+
+//go:noinline
+func sideeffect(n int64) int64 {
+	globint2++
+	return n
+}
+
+// Test that nonSSAable arguments to defer are handled correctly and only evaluated once.
+func TestNonSSAableArgs(t *testing.T) {
+	globint1 = 0
+	globint2 = 0
+	var save1 byte
+	var save2 int64
+
+	defer func() {
+		if globint1 != 1 {
+			t.Fatal(fmt.Sprintf("globint1:  wanted: 1, got %v", globint1))
+		}
+		if save1 != 5 {
+			t.Fatal(fmt.Sprintf("save1:  wanted: 5, got %v", save1))
+		}
+		if globint2 != 1 {
+			t.Fatal(fmt.Sprintf("globint2:  wanted: 1, got %v", globint2))
+		}
+		if save2 != 2 {
+			t.Fatal(fmt.Sprintf("save2:  wanted: 2, got %v", save2))
+		}
+	}()
+
+	defer func(n nonSSAable) {
+		save1 = n[4]
+	}(mknonSSAable())
+	defer func(b bigStruct) {
+		save2 = b.y
+	}(bigStruct{1, 2, 3, 4, 5, sideeffect(6)})
+}

src/runtime/funcdata.h

@@ -17,6 +17,7 @@
 #define FUNCDATA_RegPointerMaps 2
 #define FUNCDATA_StackObjects 3
 #define FUNCDATA_InlTree 4
+#define FUNCDATA_OpenCodedDeferInfo 5 /* info for func with open-coded defers */
 
 // Pseudo-assembly statements.
 

src/runtime/runtime2.go

@@ -701,7 +701,7 @@ type _func struct {
 	nameoff int32   // function name
 
 	args        int32  // in/out args size
-	deferreturn uint32 // offset of a deferreturn block from entry, if any.
+	deferreturn uint32 // offset of start of a deferreturn call instruction from entry, if any.
 
 	pcsp      int32
 	pcfile    int32
@@ -774,7 +774,7 @@ func extendRandom(r []byte, n int) {
 }
 
 // A _defer holds an entry on the list of deferred calls.
-// If you add a field here, add code to clear it in freedefer.
+// If you add a field here, add code to clear it in freedefer and deferProcStack
 // This struct must match the code in cmd/compile/internal/gc/reflect.go:deferstruct
 // and cmd/compile/internal/gc/ssa.go:(*state).call.
 // Some defers will be allocated on the stack and some on the heap.
@@ -785,11 +785,27 @@ type _defer struct {
 	siz     int32 // includes both arguments and results
 	started bool
 	heap    bool
-	sp      uintptr // sp at time of defer
-	pc      uintptr
-	fn      *funcval
-	_panic  *_panic // panic that is running defer
-	link    *_defer
+	// openDefer indicates that this _defer is for a frame with open-coded
+	// defers. We have only one defer record for the entire frame (which may
+	// currently have 0, 1, or more defers active).
+	openDefer bool
+	sp        uintptr // sp at time of defer
+	pc        uintptr // pc at time of defer
+	fn        *funcval
+	_panic    *_panic // panic that is running defer
+	link      *_defer
+
+	// If openDefer is true, the fields below record values about the stack
+	// frame and associated function that has the open-coded defer(s). sp
+	// above will be the sp for the frame, and pc will be address of the
+	// deferreturn call in the function.
+	fd   unsafe.Pointer // funcdata for the function associated with the frame
+	varp uintptr        // value of varp for the stack frame
+	// framepc is the current pc associated with the stack frame. Together,
+	// with sp above (which is the sp associated with the stack frame),
+	// framepc/sp can be used as pc/sp pair to continue a stack trace via
+	// gentraceback().
+	framepc uintptr
 }
 
 // A _panic holds information about an active panic.

src/runtime/runtime_test.go

@@ -89,7 +89,7 @@ func BenchmarkDefer(b *testing.B) {
 }
 
 func defer1() {
-	defer func(x, y, z int) {
+	func(x, y, z int) {
 		if recover() != nil || x != 1 || y != 2 || z != 3 {
 			panic("bad recover")
 		}

src/runtime/stack.go

@@ -91,7 +91,7 @@ const (
 
 	// The stack guard is a pointer this many bytes above the
 	// bottom of the stack.
-	_StackGuard = 880*sys.StackGuardMultiplier + _StackSystem
+	_StackGuard = 896*sys.StackGuardMultiplier + _StackSystem
 
 	// After a stack split check the SP is allowed to be this
 	// many bytes below the stack guard. This saves an instruction
@@ -736,6 +736,8 @@ func adjustdefers(gp *g, adjinfo *adjustinfo) {
 		adjustpointer(adjinfo, unsafe.Pointer(&d.sp))
 		adjustpointer(adjinfo, unsafe.Pointer(&d._panic))
 		adjustpointer(adjinfo, unsafe.Pointer(&d.link))
+		adjustpointer(adjinfo, unsafe.Pointer(&d.varp))
+		adjustpointer(adjinfo, unsafe.Pointer(&d.fd))
 	}
 
 	// Adjust defer argument blocks the same way we adjust active stack frames.

src/runtime/symtab.go

@@ -216,11 +216,12 @@ const (
 	_PCDATA_StackMapIndex = 1
 	_PCDATA_InlTreeIndex  = 2
 
-	_FUNCDATA_ArgsPointerMaps   = 0
-	_FUNCDATA_LocalsPointerMaps = 1
-	_FUNCDATA_RegPointerMaps    = 2
-	_FUNCDATA_StackObjects      = 3
-	_FUNCDATA_InlTree           = 4
+	_FUNCDATA_ArgsPointerMaps    = 0
+	_FUNCDATA_LocalsPointerMaps  = 1
+	_FUNCDATA_RegPointerMaps     = 2
+	_FUNCDATA_StackObjects       = 3
+	_FUNCDATA_InlTree            = 4
+	_FUNCDATA_OpenCodedDeferInfo = 5
 
 	_ArgsSizeUnknown = -0x80000000
 )

test/defererrcheck.go

+// errorcheck -0 -l -d=defer
+
+// Copyright 2019 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.
+
+// check that open-coded defers are used in expected situations
+
+package main
+
+import "fmt"
+
+var glob = 3
+
+func f1() {
+
+	for i := 0; i < 10; i++ {
+		fmt.Println("loop")
+	}
+	defer func() { // ERROR "open-coded defer in function f1"
+		fmt.Println("defer")
+	}()
+}
+
+func f2() {
+	for {
+		defer func() { // ERROR "heap-allocated defer in function f2"
+			fmt.Println("defer1")
+		}()
+		if glob > 2 {
+			break
+		}
+	}
+	defer func() { // ERROR "stack-allocated defer in function f2"
+		fmt.Println("defer2")
+	}()
+}
+
+func f3() {
+	defer func() { // ERROR "stack-allocated defer in function f3"
+		fmt.Println("defer2")
+	}()
+	for {
+		defer func() { // ERROR "heap-allocated defer in function f3"
+			fmt.Println("defer1")
+		}()
+		if glob > 2 {
+			break
+		}
+	}
+}
+
+func f4() {
+	defer func() { // ERROR "open-coded defer in function f4"
+		fmt.Println("defer")
+	}()
+label:
+	fmt.Println("goto loop")
+	if glob > 2 {
+		goto label
+	}
+}
+
+func f5() {
+label:
+	fmt.Println("goto loop")
+	defer func() { // ERROR "heap-allocated defer in function f5"
+		fmt.Println("defer")
+	}()
+	if glob > 2 {
+		goto label
+	}
+}
+
+func f6() {
+label:
+	fmt.Println("goto loop")
+	if glob > 2 {
+		goto label
+	}
+	// The current analysis doesn't end a backward goto loop, so this defer is
+	// considered to be inside a loop
+	defer func() { // ERROR "heap-allocated defer in function f6"
+		fmt.Println("defer")
+	}()
+}

test/live.go

@@ -367,16 +367,19 @@ func f24() {
 	m2[[2]string{"x", "y"}] = nil
 }
 
-// defer should not cause spurious ambiguously live variables
-
+// Non-open-coded defers should not cause autotmps.  (Open-coded defers do create extra autotmps).
 func f25(b bool) {
-	defer g25()
+	for i := 0; i < 2; i++ {
+		// Put in loop to make sure defer is not open-coded
+		defer g25()
+	}
 	if b {
 		return
 	}
 	var x string
 	x = g14()
 	printstring(x)
+	return
 }
 
 func g25()
@@ -417,7 +420,8 @@ func f27defer(b bool) {
 		defer call27(func() { x++ }) // ERROR "stack object .autotmp_[0-9]+ struct \{"
 	}
 	defer call27(func() { x++ }) // ERROR "stack object .autotmp_[0-9]+ struct \{"
-	printnl()
+	printnl()                    // ERROR "live at call to printnl: .autotmp_[0-9]+ .autotmp_[0-9]+"
+	return                       // ERROR "live at call to call27: .autotmp_[0-9]+"
 }
 
 // and newproc (go) escapes to the heap
@@ -687,12 +691,12 @@ type R struct{ *T } // ERRORAUTO "live at entry to \(\*R\)\.Foo: \.this ptr" "li
 // In particular, at printint r must be live.
 func f41(p, q *int) (r *int) { // ERROR "live at entry to f41: p q$"
 	r = p
-	defer func() { // ERROR "live at call to deferprocStack: q r$" "live at call to deferreturn: r$"
+	defer func() {
 		recover()
 	}()
-	printint(0) // ERROR "live at call to printint: q r$"
+	printint(0) // ERROR "live at call to printint: q r .autotmp_[0-9]+$"
 	r = q
-	return // ERROR "live at call to deferreturn: r$"
+	return // ERROR "live at call to f41.func1: r .autotmp_[0-9]+$"
 }
 
 func f42() {

test/nosplit.go

@@ -309,17 +309,17 @@ TestCases:
 				name := m[1]
 				size, _ := strconv.Atoi(m[2])
 
-				// The limit was originally 128 but is now 752 (880-128).
+				// The limit was originally 128 but is now 768 (896-128).
 				// Instead of rewriting the test cases above, adjust
 				// the first stack frame to use up the extra bytes.
 				if i == 0 {
-					size += (880 - 128) - 128
+					size += (896 - 128) - 128
 					// Noopt builds have a larger stackguard.
 					// See ../src/cmd/dist/buildruntime.go:stackGuardMultiplier
 					// This increase is included in objabi.StackGuard
 					for _, s := range strings.Split(os.Getenv("GO_GCFLAGS"), " ") {
 						if s == "-N" {
-							size += 880
+							size += 896
 						}
 					}
 				}