cmd/internal/obj/mips, cmd/link: add support TLS relocation for mips64x

a new relocation R_ADDRMIPSTLS is added, which resolves to 16-bit offset
of a TLS address on mips64x.

Change-Id: Ic60d0e1ba49ff1c433cead242f5884677ab227a5
Reviewed-on: https://go-review.googlesource.com/19804
Reviewed-by: Minux Ma <minux@golang.org>

src/cmd/internal/obj/link.go

@@ -585,6 +585,9 @@ const (
 	// R_ADDRMIPSU (only used on mips64) resolves to the sign-adjusted "upper" 16
 	// bits (bit 16-31) of an external address, by encoding it into the instruction.
 	R_ADDRMIPSU
+	// R_ADDRMIPSTLS (only used on mips64) resolves to the low 16 bits of a TLS
+	// address (offset from thread pointer), by encoding it into the instruction.
+	R_ADDRMIPSTLS
 )
 
 type Auto struct {

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

@@ -243,6 +243,7 @@ const (
 	C_LACON /* $n(REG) where int16 < n <= int32 */
 	C_LECON
 	C_DACON /* $n(REG) where int32 < n */
+	C_STCON /* $tlsvar */
 	C_SBRA
 	C_LBRA
 	C_SAUTO
@@ -254,6 +255,7 @@ const (
 	C_LOREG
 	C_GOK
 	C_ADDR
+	C_TLS
 	C_TEXTSIZE
 
 	C_NCLASS /* must be the last */

src/cmd/internal/obj/mips/anames0.go

@@ -26,6 +26,7 @@ var cnames0 = []string{
 	"LACON",
 	"LECON",
 	"DACON",
+	"STCON",
 	"SBRA",
 	"LBRA",
 	"SAUTO",
@@ -37,6 +38,7 @@ var cnames0 = []string{
 	"LOREG",
 	"GOK",
 	"ADDR",
+	"TLS",
 	"TEXTSIZE",
 	"NCLASS",
 }

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

@@ -141,6 +141,11 @@ var optab = []Optab{
 	{AMOVV, C_REG, C_NONE, C_ADDR, 50, 12, 0},
 	{AMOVB, C_REG, C_NONE, C_ADDR, 50, 12, 0},
 	{AMOVBU, C_REG, C_NONE, C_ADDR, 50, 12, 0},
+	{AMOVW, C_REG, C_NONE, C_TLS, 53, 8, 0},
+	{AMOVWU, C_REG, C_NONE, C_TLS, 53, 8, 0},
+	{AMOVV, C_REG, C_NONE, C_TLS, 53, 8, 0},
+	{AMOVB, C_REG, C_NONE, C_TLS, 53, 8, 0},
+	{AMOVBU, C_REG, C_NONE, C_TLS, 53, 8, 0},
 
 	{AMOVW, C_LEXT, C_NONE, C_REG, 36, 12, REGSB},
 	{AMOVWU, C_LEXT, C_NONE, C_REG, 36, 12, REGSB},
@@ -162,6 +167,11 @@ var optab = []Optab{
 	{AMOVV, C_ADDR, C_NONE, C_REG, 51, 12, 0},
 	{AMOVB, C_ADDR, C_NONE, C_REG, 51, 12, 0},
 	{AMOVBU, C_ADDR, C_NONE, C_REG, 51, 12, 0},
+	{AMOVW, C_TLS, C_NONE, C_REG, 54, 8, 0},
+	{AMOVWU, C_TLS, C_NONE, C_REG, 54, 8, 0},
+	{AMOVV, C_TLS, C_NONE, C_REG, 54, 8, 0},
+	{AMOVB, C_TLS, C_NONE, C_REG, 54, 8, 0},
+	{AMOVBU, C_TLS, C_NONE, C_REG, 54, 8, 0},
 
 	{AMOVW, C_SECON, C_NONE, C_REG, 3, 4, REGSB},
 	{AMOVV, C_SECON, C_NONE, C_REG, 3, 4, REGSB},
@@ -175,6 +185,8 @@ var optab = []Optab{
 	{AMOVV, C_ADDCON, C_NONE, C_REG, 3, 4, REGZERO},
 	{AMOVW, C_ANDCON, C_NONE, C_REG, 3, 4, REGZERO},
 	{AMOVV, C_ANDCON, C_NONE, C_REG, 3, 4, REGZERO},
+	{AMOVW, C_STCON, C_NONE, C_REG, 55, 8, 0},
+	{AMOVV, C_STCON, C_NONE, C_REG, 55, 8, 0},
 
 	{AMOVW, C_UCON, C_NONE, C_REG, 24, 4, 0},
 	{AMOVV, C_UCON, C_NONE, C_REG, 24, 4, 0},
@@ -476,6 +488,9 @@ func aclass(ctxt *obj.Link, a *obj.Addr) int {
 			}
 			ctxt.Instoffset = a.Offset
 			if a.Sym != nil { // use relocation
+				if a.Sym.Type == obj.STLSBSS {
+					return C_TLS
+				}
 				return C_ADDR
 			}
 			return C_LEXT
@@ -539,6 +554,9 @@ func aclass(ctxt *obj.Link, a *obj.Addr) int {
 			}
 
 			ctxt.Instoffset = a.Offset
+			if s.Type == obj.STLSBSS {
+				return C_STCON // address of TLS variable
+			}
 			return C_LECON
 
 		case obj.NAME_AUTO:
@@ -1387,6 +1405,40 @@ func asmout(ctxt *obj.Link, p *obj.Prog, o *Optab, out []uint32) {
 		rel2.Sym = p.From.Sym
 		rel2.Add = p.From.Offset
 		rel2.Type = obj.R_ADDRMIPS
+
+	case 53: /* mov r, tlsvar ==> rdhwr + sw o(r3) */
+		// clobbers R3 !
+		// load thread pointer with RDHWR, R3 is used for fast kernel emulation on Linux
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(opirr(ctxt, p.As), uint32(0), uint32(REG_R3), uint32(p.From.Reg))
+		rel := obj.Addrel(ctxt.Cursym)
+		rel.Off = int32(ctxt.Pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.To.Sym
+		rel.Add = p.To.Offset
+		rel.Type = obj.R_ADDRMIPSTLS
+
+	case 54: /* mov tlsvar, r ==> rdhwr + lw o(r3) */
+		// clobbers R3 !
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(opirr(ctxt, -p.As), uint32(0), uint32(REG_R3), uint32(p.To.Reg))
+		rel := obj.Addrel(ctxt.Cursym)
+		rel.Off = int32(ctxt.Pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = obj.R_ADDRMIPSTLS
+
+	case 55: /* mov $tlsvar, r ==> rdhwr + add */
+		// clobbers R3 !
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(opirr(ctxt, AADDVU), uint32(0), uint32(REG_R3), uint32(p.To.Reg))
+		rel := obj.Addrel(ctxt.Cursym)
+		rel.Off = int32(ctxt.Pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = obj.R_ADDRMIPSTLS
 	}
 
 	out[0] = o1

src/cmd/link/internal/mips64/asm.go

@@ -81,6 +81,9 @@ func elfreloc1(r *ld.Reloc, sectoff int64) int {
 	case obj.R_ADDRMIPSU:
 		ld.Cput(ld.R_MIPS_HI16)
 
+	case obj.R_ADDRMIPSTLS:
+		ld.Cput(ld.R_MIPS_TLS_TPREL_LO16)
+
 	case obj.R_CALLMIPS,
 		obj.R_JMPMIPS:
 		ld.Cput(ld.R_MIPS_26)
@@ -123,7 +126,8 @@ func archreloc(r *ld.Reloc, s *ld.LSym, val *int64) int {
 
 			return 0
 
-		case obj.R_CALLMIPS,
+		case obj.R_ADDRMIPSTLS,
+			obj.R_CALLMIPS,
 			obj.R_JMPMIPS:
 			r.Done = 0
 			r.Xsym = r.Sym
@@ -152,6 +156,16 @@ func archreloc(r *ld.Reloc, s *ld.LSym, val *int64) int {
 		}
 		return 0
 
+	case obj.R_ADDRMIPSTLS:
+		// thread pointer is at 0x7000 offset from the start of TLS data area
+		t := ld.Symaddr(r.Sym) + r.Add - 0x7000
+		if t < -32768 || t >= 32678 {
+			ld.Diag("TLS offset out of range %d", t)
+		}
+		o1 := ld.SysArch.ByteOrder.Uint32(s.P[r.Off:])
+		*val = int64(o1&0xffff0000 | uint32(t)&0xffff)
+		return 0
+
 	case obj.R_CALLMIPS,
 		obj.R_JMPMIPS:
 		// Low 26 bits = (S + A) >> 2