cmd/internal/obj, cmd/link, runtime: native-ish support for tls on arm64

Fixes #10560

Change-Id: Iedffd9c236c4fbb386c3afc52c5a1457f96ef122
Reviewed-on: https://go-review.googlesource.com/13991Reviewed-by: David Crawshaw <crawshaw@golang.org>

src/cmd/internal/obj/arm64/a.out.go

@@ -321,6 +321,7 @@ const (
 	C_LOREG
 
 	C_ADDR // TODO(aram): explain difference from C_VCONADDR
+	C_TLS  // TLS var, i.e. memory address containing offset for the var
 	C_ROFF // register offset (including register extended)
 
 	C_GOK

src/cmd/internal/obj/arm64/anames7.go

@@ -55,6 +55,7 @@ var cnames7 = []string{
 	"UOREG64K",
 	"LOREG",
 	"ADDR",
+	"TLS",
 	"ROFF",
 	"GOK",
 	"TEXTSIZE",

src/cmd/internal/obj/arm64/asm7.go

@@ -270,6 +270,7 @@ var optab = []Optab{
 	{AMOVH, C_ADDR, C_NONE, C_REG, 65, 12, 0, 0, 0},
 	{AMOVW, C_ADDR, C_NONE, C_REG, 65, 12, 0, 0, 0},
 	{AMOVD, C_ADDR, C_NONE, C_REG, 65, 12, 0, 0, 0},
+	{AMOVD, C_TLS, C_NONE, C_REG, 69, 4, 0, 0, 0},
 	{AMUL, C_REG, C_REG, C_REG, 15, 4, 0, 0, 0},
 	{AMUL, C_REG, C_NONE, C_REG, 15, 4, 0, 0, 0},
 	{AMADD, C_REG, C_REG, C_REG, 15, 4, 0, 0, 0},
@@ -968,6 +969,9 @@ 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
@@ -2753,6 +2757,18 @@ func asmout(ctxt *obj.Link, p *obj.Prog, o *Optab, out []uint32) {
 		rel.Add = p.From.Offset
 		rel.Type = obj.R_ADDRARM64
 
+	case 69: /* movd $tlsvar, reg -> movz reg, 0 + reloc */
+		o1 = opirr(ctxt, AMOVZ)
+		o1 |= uint32(p.To.Reg & 31)
+		rel := obj.Addrel(ctxt.Cursym)
+		rel.Off = int32(ctxt.Pc)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Type = obj.R_ARM64_TLS_LE
+		if p.From.Offset != 0 {
+			ctxt.Diag("invalid offset on MOVW $tlsvar")
+		}
+
 	// This is supposed to be something that stops execution.
 	// It's not supposed to be reached, ever, but if it is, we'd
 	// like to be able to tell how we got there.  Assemble as

src/cmd/internal/obj/link.go

@@ -432,6 +432,20 @@ const (
 	R_USEFIELD
 	R_POWER_TOC
 	R_GOTPCREL
+
+	// Platform dependent relocations. Architectures with fixed width instructions
+	// have the inherent issue that a 32-bit (or 64-bit!) displacement cannot be
+	// stuffed into a 32-bit instruction, so an address needs to be spread across
+	// several instructions, and in turn this requires a sequence of relocations, each
+	// updating a part of an instruction.  This leads to relocation codes that are
+	// inherently processor specific.
+
+	// Arm64.
+
+	// Set a MOV[NZ] immediate field to bits [15:0] of the offset from the thread
+	// local base to the thread local variable defined by the referenced (thread
+	// local) symbol. Error if the offset does not fit into 16 bits.
+	R_ARM64_TLS_LE
 )
 
 type Auto struct {

src/cmd/link/internal/arm64/asm.go

@@ -73,6 +73,9 @@ func elfreloc1(r *ld.Reloc, sectoff int64) int {
 		ld.Thearch.Vput(uint64(sectoff + 4))
 		ld.Thearch.Vput(ld.R_AARCH64_ADD_ABS_LO12_NC | uint64(elfsym)<<32)
 
+	case obj.R_ARM64_TLS_LE:
+		ld.Thearch.Vput(ld.R_AARCH64_TLSLE_MOVW_TPREL_G0 | uint64(elfsym)<<32)
+
 	case obj.R_CALLARM64:
 		if r.Siz != 4 {
 			return -1
@@ -225,7 +228,8 @@ func archreloc(r *ld.Reloc, s *ld.LSym, val *int64) int {
 
 			return 0
 
-		case obj.R_CALLARM64:
+		case obj.R_CALLARM64,
+			obj.R_ARM64_TLS_LE:
 			r.Done = 0
 			r.Xsym = r.Sym
 			r.Xadd = r.Add
@@ -269,6 +273,20 @@ func archreloc(r *ld.Reloc, s *ld.LSym, val *int64) int {
 		}
 		return 0
 
+	case obj.R_ARM64_TLS_LE:
+		r.Done = 0
+		if ld.HEADTYPE != obj.Hlinux {
+			ld.Diag("TLS reloc on unsupported OS %s", ld.Headstr(int(ld.HEADTYPE)))
+		}
+		// The TCB is two pointers. This is not documented anywhere, but is
+		// de facto part of the ABI.
+		v := r.Sym.Value + int64(2*ld.Thearch.Ptrsize)
+		if v < 0 || v >= 32678 {
+			ld.Diag("TLS offset out of range %d", v)
+		}
+		*val |= v << 5
+		return 0
+
 	case obj.R_CALLARM64:
 		t := (ld.Symaddr(r.Sym) + r.Add) - (s.Value + int64(r.Off))
 		if t >= 1<<27 || t < -1<<27 {

src/cmd/link/internal/ld/elf.go

@@ -350,11 +350,14 @@ const (
 	R_X86_64_GOTTPOFF = 22
 	R_X86_64_TPOFF32  = 23
 	R_X86_64_COUNT    = 24
+
 	R_AARCH64_ABS64               = 257
 	R_AARCH64_ABS32               = 258
 	R_AARCH64_CALL26              = 283
 	R_AARCH64_ADR_PREL_PG_HI21    = 275
 	R_AARCH64_ADD_ABS_LO12_NC     = 277
+	R_AARCH64_TLSLE_MOVW_TPREL_G0 = 547
+
 	R_ALPHA_NONE           = 0
 	R_ALPHA_REFLONG        = 1
 	R_ALPHA_REFQUAD        = 2
@@ -384,6 +387,7 @@ const (
 	R_ALPHA_JMP_SLOT       = 26
 	R_ALPHA_RELATIVE       = 27
 	R_ALPHA_COUNT          = 28
+
 	R_ARM_NONE          = 0
 	R_ARM_PC24          = 1
 	R_ARM_ABS32         = 2
@@ -424,6 +428,7 @@ const (
 	R_ARM_RPC24         = 254
 	R_ARM_RBASE         = 255
 	R_ARM_COUNT         = 38
+
 	R_386_NONE         = 0
 	R_386_32           = 1
 	R_386_PC32         = 2
@@ -456,6 +461,7 @@ const (
 	R_386_TLS_DTPOFF32 = 36
 	R_386_TLS_TPOFF32  = 37
 	R_386_COUNT        = 38
+
 	R_PPC_NONE            = 0
 	R_PPC_ADDR32          = 1
 	R_PPC_ADDR24          = 2
@@ -535,6 +541,7 @@ const (
 	R_PPC_EMB_BIT_FLD     = 115
 	R_PPC_EMB_RELSDA      = 116
 	R_PPC_EMB_COUNT       = R_PPC_EMB_RELSDA - R_PPC_EMB_NADDR32 + 1
+
 	R_PPC64_REL24       = R_PPC_REL24
 	R_PPC64_JMP_SLOT    = R_PPC_JMP_SLOT
 	R_PPC64_ADDR64      = 38
@@ -547,6 +554,7 @@ const (
 	R_PPC64_REL16_LO    = 250
 	R_PPC64_REL16_HI    = 251
 	R_PPC64_REL16_HA    = 252
+
 	R_SPARC_NONE     = 0
 	R_SPARC_8        = 1
 	R_SPARC_16       = 2
@@ -603,6 +611,7 @@ const (
 	R_SPARC_REGISTER = 53
 	R_SPARC_UA64     = 54
 	R_SPARC_UA16     = 55
+
 	ARM_MAGIC_TRAMP_NUMBER = 0x5c000003
 )
 

src/runtime/asm_arm64.s

@@ -38,7 +38,7 @@ TEXT runtime·rt0_go(SB),NOSPLIT,$0
 #ifdef TLSG_IS_VARIABLE
 	MOVD	$runtime·tls_g(SB), R2 	// arg 2: &tls_g
 #else
-	MOVD	0, R2		        // arg 2: not used when using platform's TLS
+	MOVD	$0, R2		        // arg 2: not used when using platform's TLS
 #endif
 	MOVD	$setg_gcc<>(SB), R1	// arg 1: setg
 	MOVD	g, R0			// arg 0: G

src/runtime/tls_arm64.s

@@ -18,13 +18,8 @@ TEXT runtime·load_g(SB),NOSPLIT,$0
 	// Darwin sometimes returns unaligned pointers
 	AND	$0xfffffffffffffff8, R0
 #endif
-#ifdef TLSG_IS_VARIABLE
 	MOVD	runtime·tls_g(SB), R27
 	ADD	R27, R0
-#else
-	// TODO(minux): use real TLS relocation, instead of hard-code for Linux
-	ADD	$0x10, R0
-#endif
 	MOVD	0(R0), g
 
 nocgo:
@@ -40,13 +35,8 @@ TEXT runtime·save_g(SB),NOSPLIT,$0
 	// Darwin sometimes returns unaligned pointers
 	AND	$0xfffffffffffffff8, R0
 #endif
-#ifdef TLSG_IS_VARIABLE
 	MOVD	runtime·tls_g(SB), R27
 	ADD	R27, R0
-#else
-	// TODO(minux): use real TLS relocation, instead of hard-code for Linux
-	ADD	$0x10, R0
-#endif
 	MOVD	g, 0(R0)
 
 nocgo:
@@ -54,4 +44,6 @@ nocgo:
 
 #ifdef TLSG_IS_VARIABLE
 GLOBL runtime·tls_g+0(SB), NOPTR, $8
+#else
+GLOBL runtime·tls_g+0(SB), TLSBSS, $8
 #endif