cmd/internal/obj, cmd/link: access global data via GOT when dynlinking on ppc64le

Change-Id: I79c60241df6c785f35371e70c777a7bd6e93571c
Reviewed-on: https://go-review.googlesource.com/15968
Reviewed-by: Russ Cox <rsc@golang.org>

6 files changed, 159 insertions, 0 deletions

diff --git a/src/cmd/internal/obj/link.go b/src/cmd/internal/obj/link.go
index a7e0c2f863..ee0b4e9fa4 100644
--- a/src/cmd/internal/obj/link.go
+++ b/src/cmd/internal/obj/link.go
@@ -500,6 +500,11 @@ const (
 	// bits of the address are not 0.
 	R_ADDRPOWER_DS
 
+	// R_ADDRPOWER_PCREL relocates a D-form, DS-form instruction sequence like
+	// R_ADDRPOWER_DS but inserts the offset of the GOT slot for the referenced symbol
+	// from the TOC rather than the symbol's address.
+	R_ADDRPOWER_GOT
+
 	// R_ADDRPOWER_PCREL relocates two D-form instructions like R_ADDRPOWER, but
 	// inserts the displacement from the place being relocated to the address of the
 	// the relocated symbol instead of just its address.
diff --git a/src/cmd/internal/obj/ppc64/a.out.go b/src/cmd/internal/obj/ppc64/a.out.go
index d28e1e895e..4f6cdcde06 100644
--- a/src/cmd/internal/obj/ppc64/a.out.go
+++ b/src/cmd/internal/obj/ppc64/a.out.go
@@ -221,6 +221,7 @@ const (
 	C_ANY
 	C_GOK
 	C_ADDR
+	C_GOTADDR
 	C_TLS_LE
 	C_TLS_IE
 	C_TEXTSIZE
diff --git a/src/cmd/internal/obj/ppc64/anames9.go b/src/cmd/internal/obj/ppc64/anames9.go
index 1b5d564dfe..d7140b1c8a 100644
--- a/src/cmd/internal/obj/ppc64/anames9.go
+++ b/src/cmd/internal/obj/ppc64/anames9.go
@@ -35,6 +35,7 @@ var cnames9 = []string{
 	"ANY",
 	"GOK",
 	"ADDR",
+	"GOTADDR",
 	"TLS_LE",
 	"TLS_IE",
 	"TEXTSIZE",
diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go
index 25b9aef830..25deeada42 100644
--- a/src/cmd/internal/obj/ppc64/asm9.go
+++ b/src/cmd/internal/obj/ppc64/asm9.go
@@ -248,6 +248,8 @@ var optab = []Optab{
 	{AMOVD, C_TLS_LE, C_NONE, C_NONE, C_REG, 79, 4, 0},
 	{AMOVD, C_TLS_IE, C_NONE, C_NONE, C_REG, 80, 8, 0},
 
+	{AMOVD, C_GOTADDR, C_NONE, C_NONE, C_REG, 81, 8, 0},
+
 	/* load constant */
 	{AMOVD, C_SECON, C_NONE, C_NONE, C_REG, 3, 4, REGSB},
 	{AMOVD, C_SACON, C_NONE, C_NONE, C_REG, 3, 4, REGSP},
@@ -598,6 +600,9 @@ func aclass(ctxt *obj.Link, a *obj.Addr) int {
 			}
 			return C_LEXT
 
+		case obj.NAME_GOTREF:
+			return C_GOTADDR
+
 		case obj.NAME_AUTO:
 			ctxt.Instoffset = int64(ctxt.Autosize) + a.Offset
 			if ctxt.Instoffset >= -BIG && ctxt.Instoffset < BIG {
@@ -2503,6 +2508,19 @@ func asmout(ctxt *obj.Link, p *obj.Prog, o *Optab, out []uint32) {
 		rel.Sym = p.From.Sym
 		rel.Type = obj.R_POWER_TLS_IE
 
+	case 81:
+		v := vregoff(ctxt, &p.To)
+		if v != 0 {
+			ctxt.Diag("invalid offset against GOT slot %v", p)
+		}
+
+		o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), REG_R2, 0)
+		o2 = AOP_IRR(uint32(opload(ctxt, AMOVD)), uint32(p.To.Reg), uint32(p.To.Reg), 0)
+		rel := obj.Addrel(ctxt.Cursym)
+		rel.Off = int32(ctxt.Pc)
+		rel.Siz = 8
+		rel.Sym = p.From.Sym
+		rel.Type = obj.R_ADDRPOWER_GOT
 	}
 
 	out[0] = o1
diff --git a/src/cmd/internal/obj/ppc64/obj9.go b/src/cmd/internal/obj/ppc64/obj9.go
index 389d6ac88d..a876b99a76 100644
--- a/src/cmd/internal/obj/ppc64/obj9.go
+++ b/src/cmd/internal/obj/ppc64/obj9.go
@@ -63,6 +63,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 			s.Size = 4
 			p.From.Type = obj.TYPE_MEM
 			p.From.Sym = s
+			p.From.Sym.Local = true
 			p.From.Name = obj.NAME_EXTERN
 			p.From.Offset = 0
 		}
@@ -75,6 +76,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 			s.Size = 8
 			p.From.Type = obj.TYPE_MEM
 			p.From.Sym = s
+			p.From.Sym.Local = true
 			p.From.Name = obj.NAME_EXTERN
 			p.From.Offset = 0
 		}
@@ -87,6 +89,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 			s.Size = 8
 			p.From.Type = obj.TYPE_MEM
 			p.From.Sym = s
+			p.From.Sym.Local = true
 			p.From.Name = obj.NAME_EXTERN
 			p.From.Offset = 0
 		}
@@ -112,6 +115,130 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 			p.As = AADD
 		}
 	}
+	if ctxt.Flag_dynlink {
+		rewriteToUseGot(ctxt, p)
+	}
+}
+
+// Rewrite p, if necessary, to access global data via the global offset table.
+func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
+	if p.As == obj.ADUFFCOPY || p.As == obj.ADUFFZERO {
+		//     ADUFFxxx $offset
+		// becomes
+		//     MOVD runtime.duffxxx@GOT, R12
+		//     ADD $offset, R12
+		//     MOVD R12, CTR
+		//     BL (CTR)
+		var sym *obj.LSym
+		if p.As == obj.ADUFFZERO {
+			sym = obj.Linklookup(ctxt, "runtime.duffzero", 0)
+		} else {
+			sym = obj.Linklookup(ctxt, "runtime.duffcopy", 0)
+		}
+		offset := p.To.Offset
+		p.As = AMOVD
+		p.From.Type = obj.TYPE_MEM
+		p.From.Name = obj.NAME_GOTREF
+		p.From.Sym = sym
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REG_R12
+		p.To.Name = obj.NAME_NONE
+		p.To.Offset = 0
+		p.To.Sym = nil
+		p1 := obj.Appendp(ctxt, p)
+		p1.As = AADD
+		p1.From.Type = obj.TYPE_CONST
+		p1.From.Offset = offset
+		p1.To.Type = obj.TYPE_REG
+		p1.To.Reg = REG_R12
+		p2 := obj.Appendp(ctxt, p1)
+		p2.As = AMOVD
+		p2.From.Type = obj.TYPE_REG
+		p2.From.Reg = REG_R12
+		p2.To.Type = obj.TYPE_REG
+		p2.To.Reg = REG_CTR
+		p3 := obj.Appendp(ctxt, p2)
+		p3.As = obj.ACALL
+		p3.From.Type = obj.TYPE_REG
+		p3.From.Reg = REG_R12
+		p3.To.Type = obj.TYPE_REG
+		p3.To.Reg = REG_CTR
+	}
+
+	// We only care about global data: NAME_EXTERN means a global
+	// symbol in the Go sense, and p.Sym.Local is true for a few
+	// internally defined symbols.
+	if p.From.Type == obj.TYPE_ADDR && p.From.Name == obj.NAME_EXTERN && !p.From.Sym.Local {
+		// MOVD $sym, Rx becomes MOVD sym@GOT, Rx
+		// MOVD $sym+<off>, Rx becomes MOVD sym@GOT, Rx; ADD <off>, Rx
+		if p.As != AMOVD {
+			ctxt.Diag("do not know how to handle TYPE_ADDR in %v with -dynlink", p)
+		}
+		if p.To.Type != obj.TYPE_REG {
+			ctxt.Diag("do not know how to handle LEAQ-type insn to non-register in %v with -dynlink", p)
+		}
+		p.From.Type = obj.TYPE_MEM
+		p.From.Name = obj.NAME_GOTREF
+		if p.From.Offset != 0 {
+			q := obj.Appendp(ctxt, p)
+			q.As = AADD
+			q.From.Type = obj.TYPE_CONST
+			q.From.Offset = p.From.Offset
+			q.To = p.To
+			p.From.Offset = 0
+		}
+	}
+	if p.From3 != nil && p.From3.Name == obj.NAME_EXTERN {
+		ctxt.Diag("don't know how to handle %v with -dynlink", p)
+	}
+	var source *obj.Addr
+	// MOVx sym, Ry becomes MOVD sym@GOT, REGTMP; MOVx (REGTMP), Ry
+	// MOVx Ry, sym becomes MOVD sym@GOT, REGTMP; MOVx Ry, (REGTMP)
+	// An addition may be inserted between the two MOVs if there is an offset.
+	if p.From.Name == obj.NAME_EXTERN && !p.From.Sym.Local {
+		if p.To.Name == obj.NAME_EXTERN && !p.To.Sym.Local {
+			ctxt.Diag("cannot handle NAME_EXTERN on both sides in %v with -dynlink", p)
+		}
+		source = &p.From
+	} else if p.To.Name == obj.NAME_EXTERN && !p.To.Sym.Local {
+		source = &p.To
+	} else {
+		return
+	}
+	if p.As == obj.ATEXT || p.As == obj.AFUNCDATA || p.As == obj.ACALL || p.As == obj.ARET || p.As == obj.AJMP {
+		return
+	}
+	if source.Sym.Type == obj.STLSBSS {
+		return
+	}
+	if source.Type != obj.TYPE_MEM {
+		ctxt.Diag("don't know how to handle %v with -dynlink", p)
+	}
+	p1 := obj.Appendp(ctxt, p)
+	p2 := obj.Appendp(ctxt, p1)
+
+	p1.As = AMOVD
+	p1.From.Type = obj.TYPE_MEM
+	p1.From.Sym = source.Sym
+	p1.From.Name = obj.NAME_GOTREF
+	p1.To.Type = obj.TYPE_REG
+	p1.To.Reg = REGTMP
+
+	p2.As = p.As
+	p2.From = p.From
+	p2.To = p.To
+	if p.From.Name == obj.NAME_EXTERN {
+		p2.From.Reg = REGTMP
+		p2.From.Name = obj.NAME_NONE
+		p2.From.Sym = nil
+	} else if p.To.Name == obj.NAME_EXTERN {
+		p2.To.Reg = REGTMP
+		p2.To.Name = obj.NAME_NONE
+		p2.To.Sym = nil
+	} else {
+		return
+	}
+	obj.Nopout(p)
 }
 
 func preprocess(ctxt *obj.Link, cursym *obj.LSym) {
diff --git a/src/cmd/link/internal/ppc64/asm.go b/src/cmd/link/internal/ppc64/asm.go
index 706160d0fb..c16cd3cf7a 100644
--- a/src/cmd/link/internal/ppc64/asm.go
+++ b/src/cmd/link/internal/ppc64/asm.go
@@ -334,6 +334,12 @@ func elfreloc1(r *ld.Reloc, sectoff int64) int {
 		ld.Thearch.Vput(uint64(sectoff + 4))
 		ld.Thearch.Vput(ld.R_PPC64_ADDR16_LO_DS | uint64(elfsym)<<32)
 
+	case obj.R_ADDRPOWER_GOT:
+		ld.Thearch.Vput(ld.R_PPC64_GOT16_HA | uint64(elfsym)<<32)
+		ld.Thearch.Vput(uint64(r.Xadd))
+		ld.Thearch.Vput(uint64(sectoff + 4))
+		ld.Thearch.Vput(ld.R_PPC64_GOT16_LO_DS | uint64(elfsym)<<32)
+
 	case obj.R_ADDRPOWER_PCREL:
 		ld.Thearch.Vput(ld.R_PPC64_REL16_HA | uint64(elfsym)<<32)
 		ld.Thearch.Vput(uint64(r.Xadd))
@@ -464,6 +470,7 @@ func archreloc(r *ld.Reloc, s *ld.LSym, val *int64) int {
 			obj.R_ADDRPOWER_DS,
 			obj.R_ADDRPOWER_TOCREL,
 			obj.R_ADDRPOWER_TOCREL_DS,
+			obj.R_ADDRPOWER_GOT,
 			obj.R_ADDRPOWER_PCREL:
 			r.Done = 0