cmd/internal/obj, runtime: fixes for defer in 386 shared libraries

Any defer in a shared object crashed when GOARCH=386. This turns out to be two
bugs:

 1) Calls to morestack were not processed to be PIC safe (must have been
    possible to trigger this another way too)
 2) jmpdefer needs to rewind the return address of the deferred function past
    the instructions that load the GOT pointer into BX, not just past the call

Bug 2) requires re-introducing the a way for .s files to know when they are
being compiled for dynamic linking but I've tried to do that in as minimal
a way as possible.

Fixes #15916

Change-Id: Ia0d09b69ec272a176934176b8eaef5f3bfcacf04
Reviewed-on: https://go-review.googlesource.com/23623
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

misc/cgo/testshared/src/depBase/dep.go

@@ -17,5 +17,6 @@ func (d *Dep) Method() int {
 }
 
 func F() int {
+	defer func() {}()
 	return V
 }

src/cmd/go/build.go

@@ -2328,7 +2328,15 @@ func (gcToolchain) asm(b *builder, p *Package, obj, ofile, sfile string) error {
 	// Add -I pkg/GOOS_GOARCH so #include "textflag.h" works in .s files.
 	inc := filepath.Join(goroot, "pkg", "include")
 	sfile = mkAbs(p.Dir, sfile)
-	args := []interface{}{buildToolExec, tool("asm"), "-o", ofile, "-trimpath", b.work, "-I", obj, "-I", inc, "-D", "GOOS_" + goos, "-D", "GOARCH_" + goarch, buildAsmflags, sfile}
+	args := []interface{}{buildToolExec, tool("asm"), "-o", ofile, "-trimpath", b.work, "-I", obj, "-I", inc, "-D", "GOOS_" + goos, "-D", "GOARCH_" + goarch, buildAsmflags}
+	if p.ImportPath == "runtime" && goarch == "386" {
+		for _, arg := range buildAsmflags {
+			if arg == "-dynlink" {
+				args = append(args, "-D=GOBUILDMODE_shared=1")
+			}
+		}
+	}
+	args = append(args, sfile)
 	if err := b.run(p.Dir, p.ImportPath, nil, args...); err != nil {
 		return err
 	}

src/cmd/internal/obj/x86/obj6.go

@@ -1092,6 +1092,7 @@ func stacksplit(ctxt *obj.Link, p *obj.Prog, framesize int32, textarg int32) *ob
 	call.Mode = ctxt.Cursym.Text.Mode
 	call.As = obj.ACALL
 	call.To.Type = obj.TYPE_BRANCH
+	call.To.Name = obj.NAME_EXTERN
 	morestack := "runtime.morestack"
 	switch {
 	case ctxt.Cursym.Cfunc:
@@ -1100,8 +1101,17 @@ func stacksplit(ctxt *obj.Link, p *obj.Prog, framesize int32, textarg int32) *ob
 		morestack = "runtime.morestack_noctxt"
 	}
 	call.To.Sym = obj.Linklookup(ctxt, morestack, 0)
+	// When compiling 386 code for dynamic linking, the call needs to be adjusted
+	// to follow PIC rules. This in turn can insert more instructions, so we need
+	// to keep track of the start of the call (where the jump will be to) and the
+	// end (which following instructions are appended to).
+	callend := call
+	progedit(ctxt, callend)
+	for ; callend.Link != nil; callend = callend.Link {
+		progedit(ctxt, callend.Link)
+	}
 
-	jmp := obj.Appendp(ctxt, call)
+	jmp := obj.Appendp(ctxt, callend)
 	jmp.As = obj.AJMP
 	jmp.To.Type = obj.TYPE_BRANCH
 	jmp.Pcond = ctxt.Cursym.Text.Link

src/runtime/asm_386.s

@@ -539,13 +539,20 @@ TEXT ·publicationBarrier(SB),NOSPLIT,$0-0
 // void jmpdefer(fn, sp);
 // called from deferreturn.
 // 1. pop the caller
-// 2. sub 5 bytes from the callers return
+// 2. sub 5 bytes (the length of CALL & a 32 bit displacement) from the callers
+//    return (when building for shared libraries, subtract 16 bytes -- 5 bytes
+//    for CALL & displacement to call __x86.get_pc_thunk.cx, 6 bytes for the
+//    LEAL to load the offset into BX, and finally 5 for the call & displacement)
 // 3. jmp to the argument
 TEXT runtime·jmpdefer(SB), NOSPLIT, $0-8
 	MOVL	fv+0(FP), DX	// fn
 	MOVL	argp+4(FP), BX	// caller sp
 	LEAL	-4(BX), SP	// caller sp after CALL
+#ifdef GOBUILDMODE_shared
+	SUBL	$16, (SP)	// return to CALL again
+#else
 	SUBL	$5, (SP)	// return to CALL again
+#endif
 	MOVL	0(DX), BX
 	JMP	BX	// but first run the deferred function