cmd/asm: detect invalid DS form offsets for ppc64x

While debugging a recent regression it was discovered that
the assembler for ppc64x was not always generating the correct
instruction for DS form loads and stores.  When an instruction
is DS form then the offset must be a multiple of 4, and if it
isn't then bits outside the offset field were being incorrectly
set resulting in unexpected and incorrect instructions.

This change adds a check to determine when the opcode is DS form
and then verifies that the offset is a multiple of 4 before
generating the instruction, otherwise logs an error.

This also changes a few asm files that were using unaligned offsets
for DS form loads and stores.  In the runtime package these were
instructions intended to cause a crash so using aligned or unaligned
offsets doesn't change that behavior.

Change-Id: Ie3a7e1e65dcc9933b54de7a46a054da8459cb56f
Reviewed-on: https://go-review.googlesource.com/40476
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>

src/cmd/asm/internal/asm/testdata/ppc64.s

@@ -24,7 +24,7 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	MOVW	foo<>+3(SB), R2
+	MOVW	foo<>+4(SB), R2
 	MOVW	16(R1), R2
 
 //	LMOVW regaddr ',' rreg
@@ -61,7 +61,7 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	FMOVD	foo<>+3(SB), F2
+	FMOVD	foo<>+4(SB), F2
 	FMOVD	16(R1), F2
 
 //	LFMOV regaddr ',' freg
@@ -86,7 +86,7 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	FMOVD	F2, foo<>+3(SB)
+	FMOVD	F2, foo<>+4(SB)
 	FMOVD	F2, 16(R1)
 
 //	LFMOV freg ',' regaddr
@@ -132,7 +132,7 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	FMOVD	F1, foo<>+3(SB)
+	FMOVD	F1, foo<>+4(SB)
 	FMOVD	F1, 16(R2)
 
 //	LMOVW freg ',' regaddr

src/cmd/internal/obj/ppc64/asm9.go

@@ -1985,29 +1985,48 @@ const (
 	DS_FORM
 )
 
-// opform returns the form (D_FORM or DS_FORM) of an instruction. Used to decide on
-// which relocation to use with a load or store and only supports the needed
-// instructions.
+// This function determines when a non-indexed load or store is D or
+// DS form for use in finding the size of the offset field in the instruction.
+// The size is needed when setting the offset value in the instruction
+// and when generating relocation for that field.
+// DS form instructions include: ld, ldu, lwa, std, stdu.  All other
+// loads and stores with an offset field are D form.  This function should
+// only be called with the same opcodes as are handled by opstore and opload.
 func (c *ctxt9) opform(insn uint32) int {
 	switch insn {
 	default:
 		c.ctxt.Diag("bad insn in loadform: %x", insn)
 	case OPVCC(58, 0, 0, 0), // ld
+		OPVCC(58, 0, 0, 1),        // ldu
 		OPVCC(58, 0, 0, 0) | 1<<1, // lwa
-		OPVCC(62, 0, 0, 0):        // std
+		OPVCC(62, 0, 0, 0),        // std
+		OPVCC(62, 0, 0, 1):        //stdu
 		return DS_FORM
 	case OP_ADDI, // add
 		OPVCC(32, 0, 0, 0), // lwz
-		OPVCC(42, 0, 0, 0), // lha
-		OPVCC(40, 0, 0, 0), // lhz
+		OPVCC(33, 0, 0, 0), // lwzu
 		OPVCC(34, 0, 0, 0), // lbz
-		OPVCC(50, 0, 0, 0), // lfd
+		OPVCC(35, 0, 0, 0), // lbzu
+		OPVCC(40, 0, 0, 0), // lhz
+		OPVCC(41, 0, 0, 0), // lhzu
+		OPVCC(42, 0, 0, 0), // lha
+		OPVCC(43, 0, 0, 0), // lhau
+		OPVCC(46, 0, 0, 0), // lmw
 		OPVCC(48, 0, 0, 0), // lfs
+		OPVCC(49, 0, 0, 0), // lfsu
+		OPVCC(50, 0, 0, 0), // lfd
+		OPVCC(51, 0, 0, 0), // lfdu
 		OPVCC(36, 0, 0, 0), // stw
-		OPVCC(44, 0, 0, 0), // sth
+		OPVCC(37, 0, 0, 0), // stwu
 		OPVCC(38, 0, 0, 0), // stb
+		OPVCC(39, 0, 0, 0), // stbu
+		OPVCC(44, 0, 0, 0), // sth
+		OPVCC(45, 0, 0, 0), // sthu
+		OPVCC(47, 0, 0, 0), // stmw
+		OPVCC(52, 0, 0, 0), // stfs
+		OPVCC(53, 0, 0, 0), // stfsu
 		OPVCC(54, 0, 0, 0), // stfd
-		OPVCC(52, 0, 0, 0): // stfs
+		OPVCC(55, 0, 0, 0): // stfdu
 		return D_FORM
 	}
 	return 0
@@ -2268,7 +2287,12 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			if int32(int16(v)) != v {
 				log.Fatalf("mishandled instruction %v", p)
 			}
-			o1 = AOP_IRR(c.opstore(p.As), uint32(p.From.Reg), uint32(r), uint32(v))
+			// Offsets in DS form stores must be a multiple of 4
+			inst := c.opstore(p.As)
+			if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+				log.Fatalf("invalid offset for DS form load/store %v", p)
+			}
+			o1 = AOP_IRR(inst, uint32(p.From.Reg), uint32(r), uint32(v))
 		}
 
 	case 8: /* mov soreg, r ==> lbz/lhz/lwz o(r) */
@@ -2294,7 +2318,12 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			if int32(int16(v)) != v {
 				log.Fatalf("mishandled instruction %v", p)
 			}
-			o1 = AOP_IRR(c.opload(p.As), uint32(p.To.Reg), uint32(r), uint32(v))
+			// Offsets in DS form loads must be a multiple of 4
+			inst := c.opload(p.As)
+			if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+				log.Fatalf("invalid offset for DS form load/store %v", p)
+			}
+			o1 = AOP_IRR(inst, uint32(p.To.Reg), uint32(r), uint32(v))
 		}
 
 	case 9: /* movb soreg, r ==> lbz o(r),r2; extsb r2,r2 */
@@ -2789,8 +2818,13 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		if r == 0 {
 			r = int(o.param)
 		}
+		// Offsets in DS form stores must be a multiple of 4
+		inst := c.opstore(p.As)
+		if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+			log.Fatalf("invalid offset for DS form load/store %v", p)
+		}
 		o1 = AOP_IRR(OP_ADDIS, REGTMP, uint32(r), uint32(high16adjusted(v)))
-		o2 = AOP_IRR(c.opstore(p.As), uint32(p.From.Reg), REGTMP, uint32(v))
+		o2 = AOP_IRR(inst, uint32(p.From.Reg), REGTMP, uint32(v))
 
 	case 36: /* mov bz/h/hz lext/lauto/lreg,r ==> lbz/lha/lhz etc */
 		v := c.regoff(&p.From)
@@ -3120,19 +3154,34 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 	/* relocation operations */
 	case 74:
 		v := c.vregoff(&p.To)
-		o1, o2 = c.symbolAccess(p.To.Sym, v, p.From.Reg, c.opstore(p.As))
+		// Offsets in DS form stores must be a multiple of 4
+		inst := c.opstore(p.As)
+		if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+			log.Fatalf("invalid offset for DS form load/store %v", p)
+		}
+		o1, o2 = c.symbolAccess(p.To.Sym, v, p.From.Reg, inst)
 
 	//if(dlm) reloc(&p->to, p->pc, 1);
 
 	case 75:
 		v := c.vregoff(&p.From)
-		o1, o2 = c.symbolAccess(p.From.Sym, v, p.To.Reg, c.opload(p.As))
+		// Offsets in DS form loads must be a multiple of 4
+		inst := c.opload(p.As)
+		if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+			log.Fatalf("invalid offset for DS form load/store %v", p)
+		}
+		o1, o2 = c.symbolAccess(p.From.Sym, v, p.To.Reg, inst)
 
 	//if(dlm) reloc(&p->from, p->pc, 1);
 
 	case 76:
 		v := c.vregoff(&p.From)
-		o1, o2 = c.symbolAccess(p.From.Sym, v, p.To.Reg, c.opload(p.As))
+		// Offsets in DS form loads must be a multiple of 4
+		inst := c.opload(p.As)
+		if c.opform(inst) == DS_FORM && v&0x3 != 0 {
+			log.Fatalf("invalid offset for DS form load/store %v", p)
+		}
+		o1, o2 = c.symbolAccess(p.From.Sym, v, p.To.Reg, inst)
 		o3 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.To.Reg), 0)
 
 		//if(dlm) reloc(&p->from, p->pc, 1);

src/runtime/asm_ppc64x.s

@@ -85,14 +85,14 @@ nocgo:
 	// start this M
 	BL	runtime·mstart(SB)
 
-	MOVD	R0, 1(R0)
+	MOVD	R0, 0(R0)
 	RET
 
 DATA	runtime·mainPC+0(SB)/8,$runtime·main(SB)
 GLOBL	runtime·mainPC(SB),RODATA,$8
 
 TEXT runtime·breakpoint(SB),NOSPLIT|NOFRAME,$0-0
-	MOVD	R0, 2(R0) // TODO: TD
+	MOVD	R0, 0(R0) // TODO: TD
 	RET
 
 TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0

src/runtime/sys_linux_ppc64x.s

@@ -189,7 +189,7 @@ TEXT runtime·rtsigprocmask(SB),NOSPLIT|NOFRAME,$0-28
 	MOVW	size+24(FP), R6
 	SYSCALL	$SYS_rt_sigprocmask
 	BVC	2(PC)
-	MOVD	R0, 0xf1(R0)	// crash
+	MOVD	R0, 0xf0(R0)	// crash
 	RET
 
 TEXT runtime·rt_sigaction(SB),NOSPLIT|NOFRAME,$0-36
@@ -273,7 +273,7 @@ TEXT runtime·munmap(SB),NOSPLIT|NOFRAME,$0
 	MOVD	n+8(FP), R4
 	SYSCALL	$SYS_munmap
 	BVC	2(PC)
-	MOVD	R0, 0xf3(R0)
+	MOVD	R0, 0xf0(R0)
 	RET
 
 TEXT runtime·madvise(SB),NOSPLIT|NOFRAME,$0
@@ -366,7 +366,7 @@ TEXT runtime·sigaltstack(SB),NOSPLIT|NOFRAME,$0
 	MOVD	old+8(FP), R4
 	SYSCALL	$SYS_sigaltstack
 	BVC	2(PC)
-	MOVD	R0, 0xf1(R0)  // crash
+	MOVD	R0, 0xf0(R0)  // crash
 	RET
 
 TEXT runtime·osyield(SB),NOSPLIT|NOFRAME,$0