cmd/link: insert trampolines for too-far jumps on ARM

ARM direct CALL/JMP instruction has 24 bit offset, which can only
encodes jumps within +/-32M. When the target is too far, the top
bits get truncated and the program jumps wild.

This CL detects too-far jumps and automatically insert trampolines,
currently only internal linking on ARM.

It is necessary to make the following changes to the linker:
- Resolve direct jump relocs when assigning addresses to functions.
  this allows trampoline insertion without moving all code that
  already laid down.
- Lay down packages in dependency order, so that when resolving a
  inter-package direct jump reloc, the target address is already
  known. Intra-package jumps are assumed never too far.
- a linker flag -debugtramp is added for debugging trampolines:
    "-debugtramp=1 -v" prints trampoline debug message
    "-debugtramp=2"    forces all inter-package jump to use
                       trampolines (currently ARM only)
    "-debugtramp=2 -v" does both
- Some data structures are changed for bookkeeping.

On ARM, pseudo DIV/DIVU/MOD/MODU instructions now clobber R8
(unfortunate). In the standard library there is no ARM assembly
code that uses these instructions, and the compiler no longer emits
them (CL 29390).

all.bash passes with -debugtramp=2, except a disassembly test (this
is unavoidable as we changed the instruction).

TBD: debug info of trampolines?

Fixes #17028.

Change-Id: Idcce347ea7e0af77c4079041a160b2f6e114b474
Reviewed-on: https://go-review.googlesource.com/29397
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

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

@@ -99,6 +99,7 @@ type Arch struct {
 	Archinit         func(*Link)
 	Archreloc        func(*Link, *Reloc, *Symbol, *int64) int
 	Archrelocvariant func(*Link, *Reloc, *Symbol, int64) int64
+	Trampoline       func(*Link, *Reloc, *Symbol)
 	Asmb             func(*Link)
 	Elfreloc1        func(*Link, *Reloc, int64) int
 	Elfsetupplt      func(*Link)
@@ -331,8 +332,7 @@ func errorexit() {
 	Exit(0)
 }
 
-func loadinternal(ctxt *Link, name string) {
-	found := 0
+func loadinternal(ctxt *Link, name string) *Library {
 	for i := 0; i < len(ctxt.Libdir); i++ {
 		if *FlagLinkshared {
 			shlibname := filepath.Join(ctxt.Libdir[i], name+".shlibname")
@@ -340,9 +340,7 @@ func loadinternal(ctxt *Link, name string) {
 				ctxt.Logf("searching for %s.a in %s\n", name, shlibname)
 			}
 			if _, err := os.Stat(shlibname); err == nil {
-				addlibpath(ctxt, "internal", "internal", "", name, shlibname)
-				found = 1
-				break
+				return addlibpath(ctxt, "internal", "internal", "", name, shlibname)
 			}
 		}
 		pname := filepath.Join(ctxt.Libdir[i], name+".a")
@@ -350,15 +348,12 @@ func loadinternal(ctxt *Link, name string) {
 			ctxt.Logf("searching for %s.a in %s\n", name, pname)
 		}
 		if _, err := os.Stat(pname); err == nil {
-			addlibpath(ctxt, "internal", "internal", pname, name, "")
-			found = 1
-			break
+			return addlibpath(ctxt, "internal", "internal", pname, name, "")
 		}
 	}
 
-	if found == 0 {
-		ctxt.Logf("warning: unable to find %s.a\n", name)
-	}
+	ctxt.Logf("warning: unable to find %s.a\n", name)
+	return nil
 }
 
 // findLibPathCmd uses cmd command to find gcc library libname.
@@ -617,6 +612,38 @@ func (ctxt *Link) loadlib() {
 	}
 
 	importcycles()
+
+	// put symbols into Textp
+	// do it in postorder so that packages are laid down in dependency order
+	// internal first, then everything else
+	ctxt.Library = postorder(ctxt.Library)
+	for _, doInternal := range [2]bool{true, false} {
+		for _, lib := range ctxt.Library {
+			if isRuntimeDepPkg(lib.Pkg) != doInternal {
+				continue
+			}
+			ctxt.Textp = append(ctxt.Textp, lib.textp...)
+			for _, s := range lib.dupTextSyms {
+				if !s.Attr.OnList() {
+					ctxt.Textp = append(ctxt.Textp, s)
+					s.Attr |= AttrOnList
+				}
+			}
+		}
+	}
+
+	if len(ctxt.Shlibs) > 0 {
+		// We might have overwritten some functions above (this tends to happen for the
+		// autogenerated type equality/hashing functions) and we don't want to generated
+		// pcln table entries for these any more so remove them from Textp.
+		textp := make([]*Symbol, 0, len(ctxt.Textp))
+		for _, s := range ctxt.Textp {
+			if s.Type != obj.SDYNIMPORT {
+				textp = append(textp, s)
+			}
+		}
+		ctxt.Textp = textp
+	}
 }
 
 /*
@@ -671,7 +698,7 @@ func objfile(ctxt *Link, lib *Library) {
 		l := f.Seek(0, 2)
 
 		f.Seek(0, 0)
-		ldobj(ctxt, f, pkg, l, lib.File, lib.File, FileObj)
+		ldobj(ctxt, f, lib, l, lib.File, lib.File, FileObj)
 		f.Close()
 
 		return
@@ -733,7 +760,7 @@ func objfile(ctxt *Link, lib *Library) {
 
 		pname = fmt.Sprintf("%s(%s)", lib.File, arhdr.name)
 		l = atolwhex(arhdr.size)
-		ldobj(ctxt, f, pkg, l, pname, lib.File, ArchiveObj)
+		ldobj(ctxt, f, lib, l, pname, lib.File, ArchiveObj)
 	}
 
 out:
@@ -1219,9 +1246,10 @@ func hostlinkArchArgs() []string {
 // ldobj loads an input object. If it is a host object (an object
 // compiled by a non-Go compiler) it returns the Hostobj pointer. If
 // it is a Go object, it returns nil.
-func ldobj(ctxt *Link, f *bio.Reader, pkg string, length int64, pn string, file string, whence int) *Hostobj {
-	eof := f.Offset() + length
+func ldobj(ctxt *Link, f *bio.Reader, lib *Library, length int64, pn string, file string, whence int) *Hostobj {
+	pkg := pathtoprefix(lib.Pkg)
 
+	eof := f.Offset() + length
 	start := f.Offset()
 	c1 := bgetc(f)
 	c2 := bgetc(f)
@@ -1308,7 +1336,7 @@ func ldobj(ctxt *Link, f *bio.Reader, pkg string, length int64, pn string, file
 	ldpkg(ctxt, f, pkg, import1-import0-2, pn, whence) // -2 for !\n
 	f.Seek(import1, 0)
 
-	LoadObjFile(ctxt, f, pkg, eof-f.Offset(), pn)
+	LoadObjFile(ctxt, f, lib, eof-f.Offset(), pn)
 	return nil
 }
 
@@ -1476,17 +1504,6 @@ func ldshlibsyms(ctxt *Link, shlib string) {
 		}
 	}
 
-	// We might have overwritten some functions above (this tends to happen for the
-	// autogenerated type equality/hashing functions) and we don't want to generated
-	// pcln table entries for these any more so remove them from Textp.
-	textp := make([]*Symbol, 0, len(ctxt.Textp))
-	for _, s := range ctxt.Textp {
-		if s.Type != obj.SDYNIMPORT {
-			textp = append(textp, s)
-		}
-	}
-	ctxt.Textp = textp
-
 	ctxt.Shlibs = append(ctxt.Shlibs, Shlib{Path: libpath, Hash: hash, Deps: deps, File: f, gcdataAddresses: gcdataAddresses})
 }
 
@@ -2084,3 +2101,34 @@ func bgetc(r *bio.Reader) int {
 	}
 	return int(c)
 }
+
+type markKind uint8 // for postorder traversal
+const (
+	unvisited markKind = iota
+	visiting
+	visited
+)
+
+func postorder(libs []*Library) []*Library {
+	order := make([]*Library, 0, len(libs)) // hold the result
+	mark := make(map[*Library]markKind, len(libs))
+	for _, lib := range libs {
+		dfs(lib, mark, &order)
+	}
+	return order
+}
+
+func dfs(lib *Library, mark map[*Library]markKind, order *[]*Library) {
+	if mark[lib] == visited {
+		return
+	}
+	if mark[lib] == visiting {
+		panic("found import cycle while visiting " + lib.Pkg)
+	}
+	mark[lib] = visiting
+	for _, i := range lib.imports {
+		dfs(i, mark, order)
+	}
+	mark[lib] = visited
+	*order = append(*order, lib)
+}