cmd/internal/obj/mips: add support of LLV, SCV, NOOP instructions

LLV and SCV are 64-bit load-linked and store-conditional. They
were used in runtime as #define WORD. Change them to normal
instruction form.

NOOP is hardware no-op. It was written as WORD $0. Make a name
for it for better disassembly output.

Fixes #12561.
Fixes #18238.

Change-Id: I82c667ce756fa83ef37b034b641e8c4366335e83
Reviewed-on: https://go-review.googlesource.com/40297Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/cmd/asm/internal/asm/testdata/mips64.s

@@ -39,6 +39,9 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 	MOVV	16(R1), R2
 	MOVV	(R1), R2
 
+	LL	(R1), R2 // c0220000
+	LLV	(R1), R2 // d0220000
+
 //	LMOVB rreg ',' rreg
 //	{
 //		outcode(int($1), &$2, 0, &$4);
@@ -98,6 +101,9 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 	MOVV	R1, 16(R2)
 	MOVV	R1, (R2)
 
+	SC	R1, (R2) // e0410000
+	SCV	R1, (R2) // f0410000
+
 //	LMOVB rreg ',' addr
 //	{
 //		outcode(int($1), &$2, 0, &$4);
@@ -238,11 +244,11 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 label0:
 	JMP	1(PC)
 	BEQ	R1, 2(PC)
-	JMP	label0+0 // JMP 64
+	JMP	label0+0 // JMP 68
 	BEQ	R1, 2(PC)
 	JAL	1(PC) // CALL 1(PC)
 	BEQ	R1, 2(PC)
-	JAL	label0+0 // CALL 64
+	JAL	label0+0 // CALL 68
 
 //	LBRA addr
 //	{
@@ -266,7 +272,7 @@ label0:
 //	}
 label1:
 	BEQ	R1, 1(PC)
-	BEQ	R1, label1 // BEQ R1, 79
+	BEQ	R1, label1 // BEQ R1, 83
 
 //	LBRA rreg ',' sreg ',' rel
 //	{
@@ -274,7 +280,7 @@ label1:
 //	}
 label2:
 	BEQ	R1, R2, 1(PC)
-	BEQ	R1, R2, label2 // BEQ R1, R2, 81
+	BEQ	R1, R2, label2 // BEQ R1, R2, 85
 
 //
 // other integer conditional branch
@@ -285,7 +291,7 @@ label2:
 //	}
 label3:
 	BLTZ	R1, 1(PC)
-	BLTZ	R1, label3 // BLTZ R1, 83
+	BLTZ	R1, label3 // BLTZ R1, 87
 
 //
 // floating point conditional branch
@@ -293,7 +299,7 @@ label3:
 //	LBRA rel
 label4:
 	BFPT	1(PC)
-	BFPT	label4 // BFPT 85
+	BFPT	label4 // BFPT 89
 
 
 //
@@ -327,7 +333,9 @@ label4:
 //
 // WORD
 //
-	WORD	$1
+	WORD	$1	// 00000001
+	NOOP		// 00000000
+	SYNC		// 0000000f
 
 //
 // NOP

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

@@ -299,6 +299,7 @@ const (
 	ADIVW
 	AGOK
 	ALL
+	ALLV
 	ALUI
 	AMOVB
 	AMOVBU
@@ -323,12 +324,14 @@ const (
 	ANEGD
 	ANEGF
 	ANEGW
+	ANOOP // hardware nop
 	ANOR
 	AOR
 	AREM
 	AREMU
 	ARFE
 	ASC
+	ASCV
 	ASGT
 	ASGTU
 	ASLL

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

@@ -45,6 +45,7 @@ var Anames = []string{
 	"DIVW",
 	"GOK",
 	"LL",
+	"LLV",
 	"LUI",
 	"MOVB",
 	"MOVBU",
@@ -69,12 +70,14 @@ var Anames = []string{
 	"NEGD",
 	"NEGF",
 	"NEGW",
+	"NOOP",
 	"NOR",
 	"OR",
 	"REM",
 	"REMU",
 	"RFE",
 	"SC",
+	"SCV",
 	"SGT",
 	"SGTU",
 	"SLL",

src/cmd/internal/obj/mips/asm0.go

@@ -129,6 +129,7 @@ var optab = []Optab{
 	{AMOVWL, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0},
 	{AMOVVL, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64},
 	{ASC, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0},
+	{ASCV, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64},
 
 	{AMOVW, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64},
 	{AMOVWU, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64},
@@ -152,6 +153,7 @@ var optab = []Optab{
 	{AMOVWL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0},
 	{AMOVVL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64},
 	{ALL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0},
+	{ALLV, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64},
 
 	{AMOVW, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64},
 	{AMOVWU, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64},
@@ -963,6 +965,7 @@ func buildop(ctxt *obj.Link) {
 
 		case ASYSCALL:
 			opset(ASYNC, r0)
+			opset(ANOOP, r0)
 			opset(ATLBP, r0)
 			opset(ATLBR, r0)
 			opset(ATLBWI, r0)
@@ -994,7 +997,9 @@ func buildop(ctxt *obj.Link) {
 			AJMP,
 			AMOVWU,
 			ALL,
+			ALLV,
 			ASC,
+			ASCV,
 			AWORD,
 			obj.ANOP,
 			obj.ATEXT,
@@ -1741,6 +1746,8 @@ func (c *ctxt0) oprrr(a obj.As) uint32 {
 
 	case ASYNC:
 		return OP(1, 7)
+	case ANOOP:
+		return 0
 
 	case ACMOVN:
 		return OP(1, 3)
@@ -1913,8 +1920,12 @@ func (c *ctxt0) opirr(a obj.As) uint32 {
 		return OP(6, 6)
 	case -ALL:
 		return SP(6, 0)
+	case -ALLV:
+		return SP(6, 4)
 	case ASC:
 		return SP(7, 0)
+	case ASCV:
+		return SP(7, 4)
 	}
 
 	if a < 0 {

src/cmd/internal/obj/mips/obj0.go

@@ -786,16 +786,8 @@ func (c *ctxt0) stacksplit(p *obj.Prog, framesize int32) *obj.Prog {
 
 func (c *ctxt0) addnop(p *obj.Prog) {
 	q := c.newprog()
-	// we want to use the canonical NOP (SLL $0,R0,R0) here,
-	// however, as the assembler will always replace $0
-	// as R0, we have to resort to manually encode the SLL
-	// instruction as WORD $0.
-	q.As = AWORD
+	q.As = ANOOP
 	q.Pos = p.Pos
-	q.From.Type = obj.TYPE_CONST
-	q.From.Name = obj.NAME_NONE
-	q.From.Offset = 0
-
 	q.Link = p.Link
 	p.Link = q
 }

src/runtime/internal/atomic/asm_mips64x.s

@@ -6,12 +6,6 @@
 
 #include "textflag.h"
 
-#define LL(base, rt)	WORD	$((060<<26)|((base)<<21)|((rt)<<16))
-#define LLV(base, rt)	WORD	$((064<<26)|((base)<<21)|((rt)<<16))
-#define SC(base, rt)	WORD	$((070<<26)|((base)<<21)|((rt)<<16))
-#define SCV(base, rt)	WORD	$((074<<26)|((base)<<21)|((rt)<<16))
-#define SYNC	WORD $0xf
-
 // bool cas(uint32 *ptr, uint32 old, uint32 new)
 // Atomically:
 //	if(*val == old){
@@ -26,9 +20,9 @@ TEXT ·Cas(SB), NOSPLIT, $0-17
 	SYNC
 cas_again:
 	MOVV	R5, R3
-	LL(1, 4)	// R4 = *R1
+	LL	(R1), R4
 	BNE	R2, R4, cas_fail
-	SC(1, 3)	// *R1 = R3
+	SC	R3, (R1)
 	BEQ	R3, cas_again
 	MOVV	$1, R1
 	MOVB	R1, ret+16(FP)
@@ -53,9 +47,9 @@ TEXT ·Cas64(SB), NOSPLIT, $0-25
 	SYNC
 cas64_again:
 	MOVV	R5, R3
-	LLV(1, 4)	// R4 = *R1
+	LLV	(R1), R4
 	BNE	R2, R4, cas64_fail
-	SCV(1, 3)	// *R1 = R3
+	SCV	R3, (R1)
 	BEQ	R3, cas64_again
 	MOVV	$1, R1
 	MOVB	R1, ret+24(FP)
@@ -104,10 +98,10 @@ TEXT ·Xadd(SB), NOSPLIT, $0-20
 	MOVV	ptr+0(FP), R2
 	MOVW	delta+8(FP), R3
 	SYNC
-	LL(2, 1)	// R1 = *R2
+	LL	(R2), R1
 	ADDU	R1, R3, R4
 	MOVV	R4, R1
-	SC(2, 4)	// *R2 = R4
+	SC	R4, (R2)
 	BEQ	R4, -4(PC)
 	MOVW	R1, ret+16(FP)
 	SYNC
@@ -117,10 +111,10 @@ TEXT ·Xadd64(SB), NOSPLIT, $0-24
 	MOVV	ptr+0(FP), R2
 	MOVV	delta+8(FP), R3
 	SYNC
-	LLV(2, 1)	// R1 = *R2
+	LLV	(R2), R1
 	ADDVU	R1, R3, R4
 	MOVV	R4, R1
-	SCV(2, 4)	// *R2 = R4
+	SCV	R4, (R2)
 	BEQ	R4, -4(PC)
 	MOVV	R1, ret+16(FP)
 	SYNC
@@ -132,8 +126,8 @@ TEXT ·Xchg(SB), NOSPLIT, $0-20
 
 	SYNC
 	MOVV	R5, R3
-	LL(2, 1)	// R1 = *R2
-	SC(2, 3)	// *R2 = R3
+	LL	(R2), R1
+	SC	R3, (R2)
 	BEQ	R3, -3(PC)
 	MOVW	R1, ret+16(FP)
 	SYNC
@@ -145,8 +139,8 @@ TEXT ·Xchg64(SB), NOSPLIT, $0-24
 
 	SYNC
 	MOVV	R5, R3
-	LLV(2, 1)	// R1 = *R2
-	SCV(2, 3)	// *R2 = R3
+	LLV	(R2), R1
+	SCV	R3, (R2)
 	BEQ	R3, -3(PC)
 	MOVV	R1, ret+16(FP)
 	SYNC
@@ -193,9 +187,9 @@ TEXT ·Or8(SB), NOSPLIT, $0-9
 	SLLV	R4, R2
 
 	SYNC
-	LL(3, 4)	// R4 = *R3
+	LL	(R3), R4
 	OR	R2, R4
-	SC(3, 4)	// *R3 = R4
+	SC	R4, (R3)
 	BEQ	R4, -4(PC)
 	SYNC
 	RET
@@ -223,9 +217,9 @@ TEXT ·And8(SB), NOSPLIT, $0-9
 	OR	R5, R2
 
 	SYNC
-	LL(3, 4)	// R4 = *R3
+	LL	(R3), R4
 	AND	R2, R4
-	SC(3, 4)	// *R3 = R4
+	SC	R4, (R3)
 	BEQ	R4, -4(PC)
 	SYNC
 	RET

src/sync/atomic/asm_mips64x.s

@@ -6,12 +6,6 @@
 
 #include "textflag.h"
 
-#define LL(base, rt)	WORD	$((060<<26)|((base)<<21)|((rt)<<16))
-#define LLV(base, rt)	WORD	$((064<<26)|((base)<<21)|((rt)<<16))
-#define SC(base, rt)	WORD	$((070<<26)|((base)<<21)|((rt)<<16))
-#define SCV(base, rt)	WORD	$((074<<26)|((base)<<21)|((rt)<<16))
-#define SYNC	WORD $0xf
-
 TEXT ·SwapInt32(SB),NOSPLIT,$0-20
 	JMP	·SwapUint32(SB)
 
@@ -20,8 +14,8 @@ TEXT ·SwapUint32(SB),NOSPLIT,$0-20
 	MOVW	new+8(FP), R5
 	SYNC
 	MOVV	R5, R3
-	LL(2, 1)	// R1 = *R2
-	SC(2, 3)	// *R2 = R3
+	LL	(R2), R1
+	SC	R3, (R2)
 	BEQ	R3, -3(PC)
 	MOVW	R1, old+16(FP)
 	SYNC
@@ -35,8 +29,8 @@ TEXT ·SwapUint64(SB),NOSPLIT,$0-24
 	MOVV	new+8(FP), R5
 	SYNC
 	MOVV	R5, R3
-	LLV(2, 1)	// R1 = *R2
-	SCV(2, 3)	// *R2 = R3
+	LLV	(R2), R1
+	SCV	R3, (R2)
 	BEQ	R3, -3(PC)
 	MOVV	R1, old+16(FP)
 	SYNC
@@ -55,9 +49,9 @@ TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-17
 	SYNC
 cas_again:
 	MOVV	R5, R3
-	LL(1, 4)	// R4 = *R1
+	LL	(R1), R4
 	BNE	R2, R4, cas_fail
-	SC(1, 3)	// *R1 = R3
+	SC	R3, (R1)
 	BEQ	R3, cas_again
 	MOVV	$1, R1
 	MOVB	R1, swapped+16(FP)
@@ -80,9 +74,9 @@ TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-25
 	SYNC
 cas64_again:
 	MOVV	R5, R3
-	LLV(1, 4)	// R4 = *R1
+	LLV	(R1), R4
 	BNE	R2, R4, cas64_fail
-	SCV(1, 3)	// *R1 = R3
+	SCV	R3, (R1)
 	BEQ	R3, cas64_again
 	MOVV	$1, R1
 	MOVB	R1, swapped+24(FP)
@@ -99,10 +93,10 @@ TEXT ·AddUint32(SB),NOSPLIT,$0-20
 	MOVV	addr+0(FP), R2
 	MOVW	delta+8(FP), R3
 	SYNC
-	LL(2, 1)	// R1 = *R2
+	LL	(R2), R1
 	ADDU	R1, R3, R4
 	MOVV	R4, R1
-	SC(2, 4)	// *R2 = R4
+	SC	R4, (R2)
 	BEQ	R4, -4(PC)
 	MOVW	R1, new+16(FP)
 	SYNC
@@ -118,10 +112,10 @@ TEXT ·AddUint64(SB),NOSPLIT,$0-24
 	MOVV	addr+0(FP), R2
 	MOVV	delta+8(FP), R3
 	SYNC
-	LLV(2, 1)	// R1 = *R2
+	LLV	(R2), R1
 	ADDVU	R1, R3, R4
 	MOVV	R4, R1
-	SCV(2, 4)	// *R2 = R4
+	SCV	R4, (R2)
 	BEQ	R4, -4(PC)
 	MOVV	R1, new+16(FP)
 	SYNC