cmd/9l: support internal linking

This implements the ELF relocations and dynamic linking tables
necessary to support internal linking on ppc64.  It also marks ppc64le
ELF files as ABI v2; failing to do this doesn't seem to confuse the
loader, but it does confuse libbfd (and hence gdb, objdump, etc).

Change-Id: I559dddf89b39052e1b6288a4dd5e72693b5355e4
Reviewed-on: https://go-review.googlesource.com/2006Reviewed-by: Russ Cox <rsc@golang.org>

include/link.h

@@ -212,6 +212,7 @@ enum
 	SMACHO,	/* Mach-O __nl_symbol_ptr */
 	SMACHOGOT,
 	SWINDOWS,
+	SELFGOT,	/* also .toc in ppc64 ABI */
 	SNOPTRDATA,
 	SINITARR,
 	SDATA,
@@ -256,12 +257,20 @@ enum
 	R_PLT1,
 	R_PLT2,
 	R_USEFIELD,
+	R_POWER_TOC,		// ELF R_PPC64_TOC16*
 };
 
 // Reloc.variant
 enum
 {
 	RV_NONE,		// identity variant
+	RV_POWER_LO,		// x & 0xFFFF
+	RV_POWER_HI,		// x >> 16
+	RV_POWER_HA,		// (x + 0x8000) >> 16
+	RV_POWER_DS,		// x & 0xFFFC, check x&0x3 == 0
+
+	RV_CHECK_OVERFLOW = 1<<8,	// check overflow flag
+	RV_TYPE_MASK = (RV_CHECK_OVERFLOW - 1),
 };
 
 // Auto.type

src/cmd/9l/obj.c

@@ -80,7 +80,8 @@ archinit(void)
 			INITRND = 4096;
 		break;
 	case Hlinux:	/* ppc64 elf */
-		debug['d'] = 1;	// TODO(minux): dynamic linking is not supported yet.
+		if(strcmp(thestring, "ppc64") == 0)
+			debug['d'] = 1;	// TODO(austin): ELF ABI v1 not supported yet
 		elfinit();
 		HEADR = ELFRESERVE;
 		if(INITTEXT == -1)

src/cmd/ld/data.c

@@ -49,7 +49,11 @@ datcmp(LSym *s1, LSym *s2)
 {
 	if(s1->type != s2->type)
 		return (int)s1->type - (int)s2->type;
-	if(s1->size != s2->size) {
+	// For ppc64, we want to interleave the .got and .toc sections
+	// from input files.  Both are type SELFGOT, so in that case
+	// fall through to the name comparison (conveniently, .got
+	// sorts before .toc).
+	if(s1->type != SELFGOT && s1->size != s2->size) {
 		if(s1->size < s2->size)
 			return -1;
 		return +1;
@@ -920,7 +924,7 @@ dodata(void)
 	vlong datsize;
 	Section *sect;
 	Segment *segro;
-	LSym *s, *last, **l;
+	LSym *s, *last, **l, *toc;
 	LSym *gcdata, *gcbss;
 	ProgGen gen;
 
@@ -994,7 +998,7 @@ dodata(void)
 
 	/* writable ELF sections */
 	datsize = 0;
-	for(; s != nil && s->type < SNOPTRDATA; s = s->next) {
+	for(; s != nil && s->type < SELFGOT; s = s->next) {
 		sect = addsection(&segdata, s->name, 06);
 		sect->align = symalign(s);
 		datsize = rnd(datsize, sect->align);
@@ -1006,6 +1010,33 @@ dodata(void)
 		sect->len = datsize - sect->vaddr;
 	}
 
+	/* .got (and .toc on ppc64) */
+	if(s->type == SELFGOT) {
+		sect = addsection(&segdata, ".got", 06);
+		sect->align = maxalign(s, SELFGOT);
+		datsize = rnd(datsize, sect->align);
+		sect->vaddr = datsize;
+		for(; s != nil && s->type == SELFGOT; s = s->next) {
+			datsize = aligndatsize(datsize, s);
+			s->sect = sect;
+			s->type = SDATA;
+			s->value = datsize - sect->vaddr;
+
+			// Resolve .TOC. symbol for this object file (ppc64)
+			toc = linkrlookup(ctxt, ".TOC.", s->version);
+			if(toc != nil) {
+				toc->sect = sect;
+				toc->outer = s;
+				toc->sub = s->sub;
+				s->sub = toc;
+
+				toc->value = 0x8000;
+			}
+			growdatsize(&datsize, s);
+		}
+		sect->len = datsize - sect->vaddr;
+	}
+
 	/* pointer-free data */
 	sect = addsection(&segdata, ".noptrdata", 06);
 	sect->align = maxalign(s, SINITARR-1);

src/cmd/ld/elf.c

@@ -306,12 +306,18 @@ elfwritedynent(LSym *s, int tag, uint64 val)
 
 void
 elfwritedynentsym(LSym *s, int tag, LSym *t)
+{
+	elfwritedynentsymplus(s, tag, t, 0);
+}
+
+void
+elfwritedynentsymplus(LSym *s, int tag, LSym *t, vlong add)
 {
 	if(elf64)
 		adduint64(ctxt, s, tag);
 	else
 		adduint32(ctxt, s, tag);
-	addaddr(ctxt, s, t);
+	addaddrplus(ctxt, s, t, add);
 }
 
 void
@@ -977,6 +983,8 @@ doelf(void)
 		addstring(shstrtab, ".interp");
 		addstring(shstrtab, ".hash");
 		addstring(shstrtab, ".got");
+		if(thechar == '9')
+			addstring(shstrtab, ".glink");
 		addstring(shstrtab, ".got.plt");
 		addstring(shstrtab, ".dynamic");
 		addstring(shstrtab, ".dynsym");
@@ -1020,7 +1028,14 @@ doelf(void)
 		/* global offset table */
 		s = linklookup(ctxt, ".got", 0);
 		s->reachable = 1;
-		s->type = SELFSECT; // writable
+		s->type = SELFGOT; // writable
+
+		/* ppc64 glink resolver */
+		if(thechar == '9') {
+			s = linklookup(ctxt, ".glink", 0);
+			s->reachable = 1;
+			s->type = SELFRXSECT;
+		}
 
 		/* hash */
 		s = linklookup(ctxt, ".hash", 0);
@@ -1033,6 +1048,11 @@ doelf(void)
 
 		s = linklookup(ctxt, ".plt", 0);
 		s->reachable = 1;
+		if(thechar == '9')
+			// In the ppc64 ABI, .plt is a data section
+			// written by the dynamic linker.
+			s->type = SELFSECT;
+		else
 			s->type = SELFRXSECT;
 		
 		elfsetupplt();
@@ -1080,8 +1100,14 @@ doelf(void)
 		if(rpath)
 			elfwritedynent(s, DT_RUNPATH, addstring(dynstr, rpath));
 
+		if(thechar == '9')
+			elfwritedynentsym(s, DT_PLTGOT, linklookup(ctxt, ".plt", 0));
+		else
 			elfwritedynentsym(s, DT_PLTGOT, linklookup(ctxt, ".got.plt", 0));
 
+		if(thechar == '9')
+			elfwritedynent(s, DT_PPC64_OPT, 0);
+
 		// Solaris dynamic linker can't handle an empty .rela.plt if
 		// DT_JMPREL is emitted so we have to defer generation of DT_PLTREL,
 		// DT_PLTRELSZ, and DT_JMPREL dynamic entries until after we know the
@@ -1309,6 +1335,7 @@ asmbelf(vlong symo)
 
 		switch(eh->machine) {
 		case EM_X86_64:
+		case EM_PPC64:
 			sh = elfshname(".rela.plt");
 			sh->type = SHT_RELA;
 			sh->flags = SHF_ALLOC;
@@ -1345,16 +1372,33 @@ asmbelf(vlong symo)
 			break;
 		}
 
+		if(eh->machine == EM_PPC64) {
+			sh = elfshname(".glink");
+			sh->type = SHT_PROGBITS;
+			sh->flags = SHF_ALLOC+SHF_EXECINSTR;
+			sh->addralign = 4;
+			shsym(sh, linklookup(ctxt, ".glink", 0));
+		}
+
 		sh = elfshname(".plt");
 		sh->type = SHT_PROGBITS;
 		sh->flags = SHF_ALLOC+SHF_EXECINSTR;
 		if(eh->machine == EM_X86_64)
 			sh->entsize = 16;
-		else
+		else if(eh->machine == EM_PPC64) {
+			// On ppc64, this is just a table of addresses
+			// filled by the dynamic linker
+			sh->type = SHT_NOBITS;
+			sh->flags = SHF_ALLOC+SHF_WRITE;
+			sh->entsize = 8;
+		} else
 			sh->entsize = 4;
-		sh->addralign = 4;
+		sh->addralign = sh->entsize;
 		shsym(sh, linklookup(ctxt, ".plt", 0));
 
+		// On ppc64, .got comes from the input files, so don't
+		// create it here, and .got.plt is not used.
+		if(eh->machine != EM_PPC64) {
 			sh = elfshname(".got");
 			sh->type = SHT_PROGBITS;
 			sh->flags = SHF_ALLOC+SHF_WRITE;
@@ -1368,6 +1412,7 @@ asmbelf(vlong symo)
 			sh->entsize = RegSize;
 			sh->addralign = RegSize;
 			shsym(sh, linklookup(ctxt, ".got.plt", 0));
+		}
 		
 		sh = elfshname(".hash");
 		sh->type = SHT_HASH;

src/cmd/ld/elf.h

@@ -317,6 +317,9 @@ typedef struct {
 #define	DT_VERNEEDNUM	0x6fffffff
 #define	DT_VERSYM	0x6ffffff0
 
+#define DT_PPC64_GLINK	(DT_LOPROC + 0)
+#define DT_PPC64_OPT	(DT_LOPROC + 3)
+
 /* Values for DT_FLAGS */
 /*	Indicates that the object being loaded may make reference to
 	the $ORIGIN substitution string */
@@ -700,6 +703,18 @@ typedef struct {
 					/* Count of defined relocation types. */
 #define	R_PPC_EMB_COUNT		(R_PPC_EMB_RELSDA - R_PPC_EMB_NADDR32 + 1)
 
+#define R_PPC64_REL24		R_PPC_REL24
+#define R_PPC64_JMP_SLOT	R_PPC_JMP_SLOT
+#define R_PPC64_ADDR64		38
+#define R_PPC64_TOC16		47
+#define R_PPC64_TOC16_LO	48
+#define R_PPC64_TOC16_HI	49
+#define R_PPC64_TOC16_HA	50
+#define R_PPC64_TOC16_DS	63
+#define R_PPC64_TOC16_LO_DS	64
+#define R_PPC64_REL16_LO	250
+#define R_PPC64_REL16_HI	251
+#define R_PPC64_REL16_HA	252
 
 #define R_SPARC_NONE		0
 #define R_SPARC_8		1
@@ -970,6 +985,7 @@ uint32	elfwritephdrs(void);
 uint32	elfwriteshdrs(void);
 void	elfwritedynent(LSym*, int, uint64);
 void	elfwritedynentsym(LSym*, int, LSym*);
+void	elfwritedynentsymplus(LSym*, int, LSym*, vlong);
 void	elfwritedynentsymsize(LSym*, int, LSym*);
 uint32	elfhash(uchar*);
 uint64	startelf(void);

src/cmd/ld/ldelf.c

@@ -554,6 +554,9 @@ ldelf(Biobuf *f, char *pkg, int64 len, char *pn)
 			s->type = STEXT;
 			break;
 		}
+		if(strcmp(sect->name, ".got") == 0 ||
+		   strcmp(sect->name, ".toc") == 0)
+			s->type = SELFGOT;
 		if(sect->type == ElfSectProgbits) {
 			s->p = sect->base;
 			s->np = sect->size;
@@ -812,6 +815,10 @@ readsym(ElfObj *obj, int i, ElfSym *sym, int needSym)
 	s = nil;
 	if(strcmp(sym->name, "_GLOBAL_OFFSET_TABLE_") == 0)
 		sym->name = ".got";
+	if(strcmp(sym->name, ".TOC.") == 0)
+		// Magic symbol on ppc64.  Will be set to this object
+		// file's .got+0x8000.
+		sym->bind = ElfSymBindLocal;
 	switch(sym->type) {
 	case ElfSymTypeSection:
 		s = obj->sect[sym->shndx].sym;
@@ -842,6 +849,15 @@ readsym(ElfObj *obj, int i, ElfSym *sym, int needSym)
 				// symbols, ignore these
 				break;
 			}
+			if(strcmp(sym->name, ".TOC.") == 0) {
+				// We need to be able to look this up,
+				// so put it in the hash table.
+				if(needSym) {
+					s = linklookup(ctxt, sym->name, ctxt->version);
+					s->type |= SHIDDEN;
+				}
+				break;
+			}
 			if(needSym) {
 				// local names and hidden visiblity global names are unique
 				// and should only reference by its index, not name, so we
@@ -892,6 +908,17 @@ reltype(char *pn, int elftype, uchar *siz)
 	switch(R(thechar, elftype)) {
 	default:
 		diag("%s: unknown relocation type %d; compiled without -fpic?", pn, elftype);
+	case R('9', R_PPC64_TOC16):
+	case R('9', R_PPC64_TOC16_LO):
+	case R('9', R_PPC64_TOC16_HI):
+	case R('9', R_PPC64_TOC16_HA):
+	case R('9', R_PPC64_TOC16_DS):
+	case R('9', R_PPC64_TOC16_LO_DS):
+	case R('9', R_PPC64_REL16_LO):
+	case R('9', R_PPC64_REL16_HI):
+	case R('9', R_PPC64_REL16_HA):
+		*siz = 2;
+		break;
 	case R('5', R_ARM_ABS32):
 	case R('5', R_ARM_GOT32):
 	case R('5', R_ARM_PLT32):
@@ -913,9 +940,11 @@ reltype(char *pn, int elftype, uchar *siz)
 	case R('8', R_386_PLT32):
 	case R('8', R_386_GOTOFF):
 	case R('8', R_386_GOTPC):
+	case R('9', R_PPC64_REL24):
 		*siz = 4;
 		break;
 	case R('6', R_X86_64_64):
+	case R('9', R_PPC64_ADDR64):
 		*siz = 8;
 		break;
 	}