cmd/internal/obj/ppc64: implement full operand support for l*arx instructions

The current implementation of l*arx instructions does not accept non-zero
offsets in RA nor the EH field. This change adds full functionality to those
instructions.

Updates #23845

Change-Id: If113f70d11de5f35f8389520b049390dbc40e863
Reviewed-on: https://go-review.googlesource.com/99635
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>

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

@@ -1199,6 +1199,26 @@ label1:
 	CALL	foo(SB)
 	RET	foo(SB)
 
+// load-and-reserve
+//	L*AR (RB)(RA*1),EH,RT produces
+//	l*arx RT,RA,RB,EH
+//
+//	Extended forms also accepted. Assumes RA=0, EH=0:
+//	L*AR (RB),RT
+//	L*AR (RB),EH,RT
+	LBAR (R4)(R3*1), $1, R5
+	LBAR (R4), $0, R5
+	LBAR (R3), R5
+	LHAR (R4)(R3*1), $1, R5
+	LHAR (R4), $0, R5
+	LHAR (R3), R5
+	LWAR (R4)(R3*1), $1, R5
+	LWAR (R4), $0, R5
+	LWAR (R3), R5
+	LDAR (R4)(R3*1), $1, R5
+	LDAR (R4), $0, R5
+	LDAR (R3), R5
+
 // END
 //
 //	LEND	comma // asm doesn't support the trailing comma.

src/cmd/asm/internal/asm/testdata/ppc64enc.s

@@ -84,4 +84,18 @@ TEXT asmtest(SB),DUPOK|NOSPLIT,$0
 	XOR $1234567, R5                // 641f001263ffd6877fe52a78
 	XOR $1234567, R5, R3            // 641f001263ffd6877fe32a78
 
+	// load-and-reserve
+	LBAR (R4)(R3*1),$1,R5           // 7ca32069
+	LBAR (R4),$0,R5                 // 7ca02068
+	LBAR (R3),R5                    // 7ca01868
+	LHAR (R4)(R3*1),$1,R5           // 7ca320e9
+	LHAR (R4),$0,R5                 // 7ca020e8
+	LHAR (R3),R5                    // 7ca018e8
+	LWAR (R4)(R3*1),$1,R5           // 7ca32029
+	LWAR (R4),$0,R5                 // 7ca02028
+	LWAR (R3),R5                    // 7ca01828
+	LDAR (R4)(R3*1),$1,R5           // 7ca320a9
+	LDAR (R4),$0,R5                 // 7ca020a8
+	LDAR (R3),R5                    // 7ca018a8
+
 	RET

src/cmd/internal/obj/ppc64/a.out.go

@@ -519,6 +519,7 @@ const (
 	AISEL
 	AMOVMW
 	ALBAR
+	ALHAR
 	ALSW
 	ALWAR
 	ALWSYNC

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

@@ -126,6 +126,7 @@ var Anames = []string{
 	"ISEL",
 	"MOVMW",
 	"LBAR",
+	"LHAR",
 	"LSW",
 	"LWAR",
 	"LWSYNC",

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

@@ -581,6 +581,8 @@ var optab = []Optab{
 	{AECIWX, C_ZOREG, C_REG, C_NONE, C_REG, 45, 4, 0},
 	{AECOWX, C_REG, C_NONE, C_NONE, C_ZOREG, 44, 4, 0},
 	{AECIWX, C_ZOREG, C_NONE, C_NONE, C_REG, 45, 4, 0},
+	{ALDAR, C_ZOREG, C_NONE, C_NONE, C_REG, 45, 4, 0},
+	{ALDAR, C_ZOREG, C_NONE, C_ANDCON, C_REG, 45, 4, 0},
 	{AEIEIO, C_NONE, C_NONE, C_NONE, C_NONE, 46, 4, 0},
 	{ATLBIE, C_REG, C_NONE, C_NONE, C_NONE, 49, 4, 0},
 	{ATLBIE, C_SCON, C_NONE, C_NONE, C_REG, 49, 4, 0},
@@ -1791,10 +1793,10 @@ func buildop(ctxt *obj.Link) {
 			opset(AFMOVS, r0)
 			opset(AFMOVSU, r0)
 
-		case AECIWX:
+		case ALDAR:
 			opset(ALBAR, r0)
+			opset(ALHAR, r0)
 			opset(ALWAR, r0)
-			opset(ALDAR, r0)
 
 		case ASYSCALL: /* just the op; flow of control */
 			opset(ARFI, r0)
@@ -1861,6 +1863,7 @@ func buildop(ctxt *obj.Link) {
 			AVMSUMUDM,
 			AADDEX,
 			ACMPEQB,
+			AECIWX,
 			obj.ANOP,
 			obj.ATEXT,
 			obj.AUNDEF,
@@ -1993,6 +1996,11 @@ func AOP_Z23I(op uint32, d uint32, a uint32, b uint32, c uint32) uint32 {
 	return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (c&3)<<7
 }
 
+/* X-form, 3-register operands + EH field */
+func AOP_RRRI(op uint32, d uint32, a uint32, b uint32, c uint32) uint32 {
+	return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (c & 1)
+}
+
 func LOP_RRR(op uint32, a uint32, s uint32, b uint32) uint32 {
 	return op | (s&31)<<21 | (a&31)<<16 | (b&31)<<11
 }
@@ -2994,8 +3002,24 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		o1 = AOP_RRR(c.opstorex(p.As), uint32(p.From.Reg), uint32(p.To.Index), uint32(p.To.Reg))
 
 	case 45: /* indexed load */
-		o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg))
-
+		switch p.As {
+		/* The assembler accepts a 4-operand l*arx instruction. The fourth operand is an Exclusive Access Hint (EH) */
+		/* The EH field can be used as a lock acquire/release hint as follows: */
+		/* 0 = Atomic Update (fetch-and-operate or similar algorithm) */
+		/* 1 = Exclusive Access (lock acquire and release) */
+		case ALBAR, ALHAR, ALWAR, ALDAR:
+			if p.From3Type() != obj.TYPE_NONE {
+				eh := int(c.regoff(p.GetFrom3()))
+				if eh > 1 {
+					c.ctxt.Diag("illegal EH field\n%v", p)
+				}
+				o1 = AOP_RRRI(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg), uint32(eh))
+			} else {
+				o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg))
+			}
+		default:
+			o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg))
+		}
 	case 46: /* plain op */
 		o1 = c.oprrr(p.As)
 
@@ -4765,10 +4789,12 @@ func (c *ctxt9) oploadx(a obj.As) uint32 {
 		return OPVCC(31, 310, 0, 0) /* eciwx */
 	case ALBAR:
 		return OPVCC(31, 52, 0, 0) /* lbarx */
+	case ALHAR:
+		return OPVCC(31, 116, 0, 0) /* lharx */
 	case ALWAR:
 		return OPVCC(31, 20, 0, 0) /* lwarx */
 	case ALDAR:
-		return OPVCC(31, 84, 0, 0)
+		return OPVCC(31, 84, 0, 0) /* ldarx */
 	case ALSW:
 		return OPVCC(31, 533, 0, 0) /* lswx */
 	case AMOVD: