cmd/compile, cmd/link: separate stable and internal ABIs

This implements compiler and linker support for separating the
function calling ABI into two ABIs: a stable and an internal ABI. At
the moment, the two ABIs are identical, but we'll be able to evolve
the internal ABI without breaking existing assembly code that depends
on the stable ABI for calling to and from Go.

The Go compiler generates internal ABI symbols for all Go functions.
It uses the symabis information produced by the assembler to create
ABI wrappers whenever it encounters a body-less Go function that's
defined in assembly or a Go function that's referenced from assembly.

Since the two ABIs are currently identical, for the moment this is
implemented using "ABI alias" symbols, which are just forwarding
references to the native ABI symbol for a function. This way there's
no actual code involved in the ABI wrapper, which is good because
we're not deriving any benefit from it right now. Once the ABIs
diverge, we can eliminate ABI aliases.

The linker represents these different ABIs internally as different
versions of the same symbol. This way, the linker keeps us honest,
since every symbol definition and reference also specifies its
version. The linker is responsible for resolving ABI aliases.

Fixes #27539.

Change-Id: I197c52ec9f8fc435db8f7a4259029b20f6d65e95
Reviewed-on: https://go-review.googlesource.com/c/147160
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>

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

@@ -11,7 +11,18 @@ import (
 	"strconv"
 )
 
+// sysfunc looks up Go function name in package runtime. This function
+// must follow the internal calling convention.
 func sysfunc(name string) *obj.LSym {
+	s := Runtimepkg.Lookup(name)
+	s.SetFunc(true)
+	return s.Linksym()
+}
+
+// sysvar looks up a variable (or assembly function) name in package
+// runtime. If this is a function, it may have a special calling
+// convention.
+func sysvar(name string) *obj.LSym {
 	return Runtimepkg.Lookup(name).Linksym()
 }
 

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

@@ -187,7 +187,13 @@ func (pp *Progs) settext(fn *Node) {
 	ptxt.From.Sym = fn.Func.lsym
 }
 
-func (f *Func) initLSym() {
+// initLSym defines f's obj.LSym and initializes it based on the
+// properties of f. This includes setting the symbol flags and ABI and
+// creating and initializing related DWARF symbols.
+//
+// initLSym must be called exactly once per function and must be
+// called for both functions with bodies and functions without bodies.
+func (f *Func) initLSym(hasBody bool) {
 	if f.lsym != nil {
 		Fatalf("Func.initLSym called twice")
 	}
@@ -197,6 +203,61 @@ func (f *Func) initLSym() {
 		if f.Pragma&Systemstack != 0 {
 			f.lsym.Set(obj.AttrCFunc, true)
 		}
+
+		var aliasABI obj.ABI
+		needABIAlias := false
+		if abi, ok := symabiDefs[f.lsym.Name]; ok && abi == obj.ABI0 {
+			// Symbol is defined as ABI0. Create an
+			// Internal -> ABI0 wrapper.
+			f.lsym.SetABI(obj.ABI0)
+			needABIAlias, aliasABI = true, obj.ABIInternal
+		} else {
+			// No ABI override. Check that the symbol is
+			// using the expected ABI.
+			want := obj.ABIInternal
+			if f.lsym.ABI() != want {
+				Fatalf("function symbol %s has the wrong ABI %v, expected %v", f.lsym, f.lsym.ABI(), want)
+			}
+		}
+
+		if abi, ok := symabiRefs[f.lsym.Name]; ok && abi == obj.ABI0 {
+			// Symbol is referenced as ABI0. Create an
+			// ABI0 -> Internal wrapper if necessary.
+			if f.lsym.ABI() != obj.ABI0 {
+				needABIAlias, aliasABI = true, obj.ABI0
+			}
+		}
+
+		if !needABIAlias && allABIs {
+			// The compiler was asked to produce ABI
+			// wrappers for everything.
+			switch f.lsym.ABI() {
+			case obj.ABI0:
+				needABIAlias, aliasABI = true, obj.ABIInternal
+			case obj.ABIInternal:
+				needABIAlias, aliasABI = true, obj.ABI0
+			}
+		}
+
+		if needABIAlias {
+			// These LSyms have the same name as the
+			// native function, so we create them directly
+			// rather than looking them up. The uniqueness
+			// of f.lsym ensures uniqueness of asym.
+			asym := &obj.LSym{
+				Name: f.lsym.Name,
+				Type: objabi.SABIALIAS,
+				R:    []obj.Reloc{{Sym: f.lsym}}, // 0 size, so "informational"
+			}
+			asym.SetABI(aliasABI)
+			asym.Set(obj.AttrDuplicateOK, true)
+			Ctxt.ABIAliases = append(Ctxt.ABIAliases, asym)
+		}
+	}
+
+	if !hasBody {
+		// For body-less functions, we only create the LSym.
+		return
 	}
 
 	var flag int

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

@@ -15,6 +15,7 @@ import (
 
 	"cmd/compile/internal/syntax"
 	"cmd/compile/internal/types"
+	"cmd/internal/obj"
 	"cmd/internal/objabi"
 	"cmd/internal/src"
 )
@@ -250,6 +251,18 @@ func (p *noder) node() {
 		}
 	}
 
+	// The linker expects an ABI0 wrapper for all cgo-exported
+	// functions.
+	for _, prag := range p.pragcgobuf {
+		switch prag[0] {
+		case "cgo_export_static", "cgo_export_dynamic":
+			if symabiRefs == nil {
+				symabiRefs = make(map[string]obj.ABI)
+			}
+			symabiRefs[prag[1]] = obj.ABI0
+		}
+	}
+
 	pragcgobuf = append(pragcgobuf, p.pragcgobuf...)
 	lineno = src.NoXPos
 	clearImports()

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

@@ -198,6 +198,8 @@ func funccompile(fn *Node) {
 	dowidth(fn.Type)
 
 	if fn.Nbody.Len() == 0 {
+		// Initialize ABI wrappers if necessary.
+		fn.Func.initLSym(false)
 		emitptrargsmap(fn)
 		return
 	}
@@ -231,7 +233,7 @@ func compile(fn *Node) {
 	Curfn = nil
 
 	// Set up the function's LSym early to avoid data races with the assemblers.
-	fn.Func.initLSym()
+	fn.Func.initLSym(true)
 
 	// Make sure type syms are declared for all types that might
 	// be types of stack objects. We need to do this here

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

@@ -801,7 +801,7 @@ var (
 func dcommontype(lsym *obj.LSym, t *types.Type) int {
 	sizeofAlg := 2 * Widthptr
 	if algarray == nil {
-		algarray = sysfunc("algarray")
+		algarray = sysvar("algarray")
 	}
 	dowidth(t)
 	alg := algtype(t)
@@ -1618,7 +1618,7 @@ func dalgsym(t *types.Type) *obj.LSym {
 
 		if memhashvarlen == nil {
 			memhashvarlen = sysfunc("memhash_varlen")
-			memequalvarlen = sysfunc("memequal_varlen")
+			memequalvarlen = sysvar("memequal_varlen") // asm func
 		}
 
 		// make hash closure

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

@@ -68,9 +68,9 @@ func initssaconfig() {
 	assertI2I2 = sysfunc("assertI2I2")
 	deferproc = sysfunc("deferproc")
 	Deferreturn = sysfunc("deferreturn")
-	Duffcopy = sysfunc("duffcopy")
-	Duffzero = sysfunc("duffzero")
-	gcWriteBarrier = sysfunc("gcWriteBarrier")
+	Duffcopy = sysvar("duffcopy")             // asm func with special ABI
+	Duffzero = sysvar("duffzero")             // asm func with special ABI
+	gcWriteBarrier = sysvar("gcWriteBarrier") // asm func with special ABI
 	goschedguarded = sysfunc("goschedguarded")
 	growslice = sysfunc("growslice")
 	msanread = sysfunc("msanread")
@@ -86,25 +86,25 @@ func initssaconfig() {
 	racereadrange = sysfunc("racereadrange")
 	racewrite = sysfunc("racewrite")
 	racewriterange = sysfunc("racewriterange")
-	supportPopcnt = sysfunc("support_popcnt")
-	supportSSE41 = sysfunc("support_sse41")
-	arm64SupportAtomics = sysfunc("arm64_support_atomics")
+	supportPopcnt = sysvar("support_popcnt")              // bool
+	supportSSE41 = sysvar("support_sse41")                // bool
+	arm64SupportAtomics = sysvar("arm64_support_atomics") // bool
 	typedmemclr = sysfunc("typedmemclr")
 	typedmemmove = sysfunc("typedmemmove")
-	Udiv = sysfunc("udiv")
-	writeBarrier = sysfunc("writeBarrier")
-
-	// GO386=387 runtime functions
-	ControlWord64trunc = sysfunc("controlWord64trunc")
-	ControlWord32 = sysfunc("controlWord32")
-
-	// Wasm
-	WasmMove = sysfunc("wasmMove")
-	WasmZero = sysfunc("wasmZero")
-	WasmDiv = sysfunc("wasmDiv")
-	WasmTruncS = sysfunc("wasmTruncS")
-	WasmTruncU = sysfunc("wasmTruncU")
-	SigPanic = sysfunc("sigpanic")
+	Udiv = sysvar("udiv")                 // asm func with special ABI
+	writeBarrier = sysvar("writeBarrier") // struct { bool; ... }
+
+	// GO386=387 runtime definitions
+	ControlWord64trunc = sysvar("controlWord64trunc") // uint16
+	ControlWord32 = sysvar("controlWord32")           // uint16
+
+	// Wasm (all asm funcs with special ABIs)
+	WasmMove = sysvar("wasmMove")
+	WasmZero = sysvar("wasmZero")
+	WasmDiv = sysvar("wasmDiv")
+	WasmTruncS = sysvar("wasmTruncS")
+	WasmTruncU = sysvar("wasmTruncU")
+	SigPanic = sysvar("sigpanic")
 }
 
 // buildssa builds an SSA function for fn.

src/cmd/compile/internal/types/sym.go

@@ -77,6 +77,12 @@ func (sym *Sym) Linksym() *obj.LSym {
 	if sym == nil {
 		return nil
 	}
+	if sym.Func() {
+		// This is a function symbol. Mark it as "internal ABI".
+		return Ctxt.LookupInit(sym.LinksymName(), func(s *obj.LSym) {
+			s.SetABI(obj.ABIInternal)
+		})
+	}
 	return Ctxt.Lookup(sym.LinksymName())
 }
 

src/cmd/internal/obj/arm/asm5.go

@@ -1530,6 +1530,7 @@ func buildop(ctxt *obj.Link) {
 	}
 
 	deferreturn = ctxt.Lookup("runtime.deferreturn")
+	deferreturn.SetABI(obj.ABIInternal)
 
 	symdiv = ctxt.Lookup("runtime._div")
 	symdivu = ctxt.Lookup("runtime._divu")

src/cmd/internal/obj/wasm/wasmobj.go

@@ -126,7 +126,9 @@ func instinit(ctxt *obj.Link) {
 	morestackNoCtxt = ctxt.Lookup("runtime.morestack_noctxt")
 	gcWriteBarrier = ctxt.Lookup("runtime.gcWriteBarrier")
 	sigpanic = ctxt.Lookup("runtime.sigpanic")
+	sigpanic.SetABI(obj.ABIInternal)
 	deferreturn = ctxt.Lookup("runtime.deferreturn")
+	deferreturn.SetABI(obj.ABIInternal)
 	jmpdefer = ctxt.Lookup(`"".jmpdefer`)
 }
 

src/cmd/internal/obj/x86/asm6.go

@@ -2065,6 +2065,7 @@ func instinit(ctxt *obj.Link) {
 		plan9privates = ctxt.Lookup("_privates")
 	case objabi.Hnacl:
 		deferreturn = ctxt.Lookup("runtime.deferreturn")
+		deferreturn.SetABI(obj.ABIInternal)
 	}
 
 	for i := range avxOptab {

src/cmd/internal/objabi/symkind.go

@@ -60,6 +60,13 @@ const (
 	SDWARFRANGE
 	SDWARFLOC
 	SDWARFMISC
+	// ABI alias. An ABI alias symbol is an empty symbol with a
+	// single relocation with 0 size that references the native
+	// function implementation symbol.
+	//
+	// TODO(austin): Remove this and all uses once the compiler
+	// generates real ABI wrappers rather than symbol aliases.
+	SABIALIAS
 	// Update cmd/link/internal/sym/AbiSymKindToSymKind for new SymKind values.
 
 )

src/cmd/internal/objabi/symkind_string.go

@@ -4,9 +4,9 @@ package objabi
 
 import "strconv"
 
-const _SymKind_name = "SxxxSTEXTSRODATASNOPTRDATASDATASBSSSNOPTRBSSSTLSBSSSDWARFINFOSDWARFRANGESDWARFLOCSDWARFMISC"
+const _SymKind_name = "SxxxSTEXTSRODATASNOPTRDATASDATASBSSSNOPTRBSSSTLSBSSSDWARFINFOSDWARFRANGESDWARFLOCSDWARFMISCSABIALIAS"
 
-var _SymKind_index = [...]uint8{0, 4, 9, 16, 26, 31, 35, 44, 51, 61, 72, 81, 91}
+var _SymKind_index = [...]uint8{0, 4, 9, 16, 26, 31, 35, 44, 51, 61, 72, 81, 91, 100}
 
 func (i SymKind) String() string {
 	if i >= SymKind(len(_SymKind_index)-1) {

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

@@ -60,8 +60,8 @@ func deadcode(ctxt *Link) {
 	d.init()
 	d.flood()
 
-	callSym := ctxt.Syms.ROLookup("reflect.Value.Call", 0)
-	methSym := ctxt.Syms.ROLookup("reflect.Value.Method", 0)
+	callSym := ctxt.Syms.ROLookup("reflect.Value.Call", sym.SymVerABIInternal)
+	methSym := ctxt.Syms.ROLookup("reflect.Value.Method", sym.SymVerABIInternal)
 	reflectSeen := false
 
 	if ctxt.DynlinkingGo() {
@@ -257,7 +257,10 @@ func (d *deadcodepass) init() {
 	}
 
 	for _, name := range names {
+		// Mark symbol as an data/ABI0 symbol.
 		d.mark(d.ctxt.Syms.ROLookup(name, 0), nil)
+		// Also mark any Go functions (internal ABI).
+		d.mark(d.ctxt.Syms.ROLookup(name, sym.SymVerABIInternal), nil)
 	}
 }
 
@@ -308,6 +311,11 @@ func (d *deadcodepass) flood() {
 				// reachable.
 				continue
 			}
+			if r.Sym.Type == sym.SABIALIAS {
+				// Patch this relocation through the
+				// ABI alias before marking.
+				r.Sym = resolveABIAlias(r.Sym)
+			}
 			if r.Type != objabi.R_METHODOFF {
 				d.mark(r.Sym, s)
 				continue

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

@@ -25,6 +25,17 @@ func expandpkg(t0 string, pkg string) string {
 	return strings.Replace(t0, `"".`, pkg+".", -1)
 }
 
+func resolveABIAlias(s *sym.Symbol) *sym.Symbol {
+	if s.Type != sym.SABIALIAS {
+		return s
+	}
+	target := s.R[0].Sym
+	if target.Type == sym.SABIALIAS {
+		panic(fmt.Sprintf("ABI alias %s references another ABI alias %s", s, target))
+	}
+	return target
+}
+
 // TODO:
 //	generate debugging section in binary.
 //	once the dust settles, try to move some code to
@@ -191,6 +202,11 @@ func loadcgo(ctxt *Link, file string, pkg string, p string) {
 			}
 			local = expandpkg(local, pkg)
 
+			// The compiler arranges for an ABI0 wrapper
+			// to be available for all cgo-exported
+			// functions. Link.loadlib will resolve any
+			// ABI aliases we find here (since we may not
+			// yet know it's an alias).
 			s := ctxt.Syms.Lookup(local, 0)
 
 			switch ctxt.BuildMode {

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

@@ -169,7 +169,7 @@ func (ctxt *Link) DynlinkingGo() bool {
 
 // CanUsePlugins returns whether a plugins can be used
 func (ctxt *Link) CanUsePlugins() bool {
-	return ctxt.Syms.ROLookup("plugin.Open", 0) != nil
+	return ctxt.Syms.ROLookup("plugin.Open", sym.SymVerABIInternal) != nil
 }
 
 // UseRelro returns whether to make use of "read only relocations" aka
@@ -635,6 +635,19 @@ func (ctxt *Link) loadlib() {
 		}
 		ctxt.Textp = textp
 	}
+
+	// Resolve ABI aliases in the list of cgo-exported functions.
+	// This is necessary because we load the ABI0 symbol for all
+	// cgo exports.
+	for i, s := range dynexp {
+		if s.Type != sym.SABIALIAS {
+			continue
+		}
+		t := resolveABIAlias(s)
+		t.Attr |= s.Attr
+		t.SetExtname(s.Extname())
+		dynexp[i] = t
+	}
 }
 
 // mangleTypeSym shortens the names of symbols that represent Go types
@@ -651,7 +664,7 @@ func (ctxt *Link) loadlib() {
 // those programs loaded dynamically in multiple parts need these
 // symbols to have entries in the symbol table.
 func (ctxt *Link) mangleTypeSym() {
-	if ctxt.BuildMode != BuildModeShared && !ctxt.linkShared && ctxt.BuildMode != BuildModePlugin && ctxt.Syms.ROLookup("plugin.Open", 0) == nil {
+	if ctxt.BuildMode != BuildModeShared && !ctxt.linkShared && ctxt.BuildMode != BuildModePlugin && !ctxt.CanUsePlugins() {
 		return
 	}
 
@@ -1801,6 +1814,21 @@ func ldshlibsyms(ctxt *Link, shlib string) {
 				gcdataLocations[elfsym.Value+2*uint64(ctxt.Arch.PtrSize)+8+1*uint64(ctxt.Arch.PtrSize)] = lsym
 			}
 		}
+		// For function symbols, we don't know what ABI is
+		// available, so alias it under both ABIs.
+		//
+		// TODO(austin): This is almost certainly wrong once
+		// the ABIs are actually different. We might have to
+		// mangle Go function names in the .so to include the
+		// ABI.
+		if elf.ST_TYPE(elfsym.Info) == elf.STT_FUNC {
+			alias := ctxt.Syms.Lookup(elfsym.Name, sym.SymVerABIInternal)
+			if alias.Type != 0 {
+				continue
+			}
+			alias.Type = sym.SABIALIAS
+			alias.R = []sym.Reloc{{Sym: lsym}}
+		}
 	}
 	gcdataAddresses := make(map[*sym.Symbol]uint64)
 	if ctxt.Arch.Family == sys.ARM64 {

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

@@ -506,7 +506,7 @@ func (ctxt *Link) symtab() {
 		abihashgostr.AddAddr(ctxt.Arch, hashsym)
 		abihashgostr.AddUint(ctxt.Arch, uint64(hashsym.Size))
 	}
-	if ctxt.BuildMode == BuildModePlugin || ctxt.Syms.ROLookup("plugin.Open", 0) != nil {
+	if ctxt.BuildMode == BuildModePlugin || ctxt.CanUsePlugins() {
 		for _, l := range ctxt.Library {
 			s := ctxt.Syms.Lookup("go.link.pkghashbytes."+l.Pkg, 0)
 			s.Attr |= sym.AttrReachable

src/cmd/link/internal/ppc64/asm.go

@@ -133,7 +133,7 @@ func genplt(ctxt *ld.Link) {
 }
 
 func genaddmoduledata(ctxt *ld.Link) {
-	addmoduledata := ctxt.Syms.ROLookup("runtime.addmoduledata", 0)
+	addmoduledata := ctxt.Syms.ROLookup("runtime.addmoduledata", sym.SymVerABI0)
 	if addmoduledata.Type == sym.STEXT && ctxt.BuildMode != ld.BuildModePlugin {
 		return
 	}

src/cmd/link/internal/sym/symbols.go

@@ -43,6 +43,8 @@ func NewSymbols() *Symbols {
 	hash := make([]map[string]*Symbol, SymVerStatic)
 	// Preallocate about 2mb for hash of non static symbols
 	hash[0] = make(map[string]*Symbol, 100000)
+	// And another 1mb for internal ABI text symbols.
+	hash[SymVerABIInternal] = make(map[string]*Symbol, 50000)
 	return &Symbols{
 		hash:   hash,
 		Allsym: make([]*Symbol, 0, 100000),

src/cmd/link/internal/sym/symkind.go

@@ -109,6 +109,9 @@ const (
 	SDWARFRANGE
 	SDWARFLOC
 	SDWARFMISC // Not really a section; informs/affects other DWARF section generation
+
+	// ABI aliases (these never appear in the output)
+	SABIALIAS
 )
 
 // AbiSymKindToSymKind maps values read from object files (which are
@@ -126,6 +129,7 @@ var AbiSymKindToSymKind = [...]SymKind{
 	SDWARFRANGE,
 	SDWARFLOC,
 	SDWARFMISC,
+	SABIALIAS,
 }
 
 // ReadOnly are the symbol kinds that form read-only sections. In some

src/cmd/link/internal/sym/symkind_string.go

@@ -4,9 +4,9 @@ package sym
 
 import "strconv"
 
-const _SymKind_name = "SxxxSTEXTSELFRXSECTSTYPESSTRINGSGOSTRINGSGOFUNCSGCBITSSRODATASFUNCTABSELFROSECTSMACHOPLTSTYPERELROSSTRINGRELROSGOSTRINGRELROSGOFUNCRELROSGCBITSRELROSRODATARELROSFUNCTABRELROSTYPELINKSITABLINKSSYMTABSPCLNTABSELFSECTSMACHOSMACHOGOTSWINDOWSSELFGOTSNOPTRDATASINITARRSDATASBSSSNOPTRBSSSTLSBSSSXCOFFTOCSXREFSMACHOSYMSTRSMACHOSYMTABSMACHOINDIRECTPLTSMACHOINDIRECTGOTSFILEPATHSCONSTSDYNIMPORTSHOSTOBJSDWARFSECTSDWARFINFOSDWARFRANGESDWARFLOCSDWARFMISC"
+const _SymKind_name = "SxxxSTEXTSELFRXSECTSTYPESSTRINGSGOSTRINGSGOFUNCSGCBITSSRODATASFUNCTABSELFROSECTSMACHOPLTSTYPERELROSSTRINGRELROSGOSTRINGRELROSGOFUNCRELROSGCBITSRELROSRODATARELROSFUNCTABRELROSTYPELINKSITABLINKSSYMTABSPCLNTABSELFSECTSMACHOSMACHOGOTSWINDOWSSELFGOTSNOPTRDATASINITARRSDATASBSSSNOPTRBSSSTLSBSSSXCOFFTOCSXREFSMACHOSYMSTRSMACHOSYMTABSMACHOINDIRECTPLTSMACHOINDIRECTGOTSFILEPATHSCONSTSDYNIMPORTSHOSTOBJSDWARFSECTSDWARFINFOSDWARFRANGESDWARFLOCSDWARFMISCSABIALIAS"
 
-var _SymKind_index = [...]uint16{0, 4, 9, 19, 24, 31, 40, 47, 54, 61, 69, 79, 88, 98, 110, 124, 136, 148, 160, 173, 182, 191, 198, 206, 214, 220, 229, 237, 244, 254, 262, 267, 271, 280, 287, 296, 301, 313, 325, 342, 359, 368, 374, 384, 392, 402, 412, 423, 432, 442}
+var _SymKind_index = [...]uint16{0, 4, 9, 19, 24, 31, 40, 47, 54, 61, 69, 79, 88, 98, 110, 124, 136, 148, 160, 173, 182, 191, 198, 206, 214, 220, 229, 237, 244, 254, 262, 267, 271, 280, 287, 296, 301, 313, 325, 342, 359, 368, 374, 384, 392, 402, 412, 423, 432, 442, 451}
 
 func (i SymKind) String() string {
 	if i >= SymKind(len(_SymKind_index)-1) {