From: Russ Cox Date: Fri, 4 Jan 2013 22:03:57 +0000 (-0500) Subject: cmd/ld: use native-endian symbol values in symbol table X-Git-Tag: go1.1rc2~1480 X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=4e2aa9bff0eb76f98bb3b238c7f56bea6330f70d;p=gostls13.git cmd/ld: use native-endian symbol values in symbol table The Plan 9 symbol table format defines big-endian symbol values for portability, but we want to be able to generate an ELF object file and let the host linker link it, as part of the solution to issue 4069. The symbol table itself, since it is loaded into memory at run time, must be filled in by the final host linker, using relocation directives to set the symbol values. On a little-endian machine, the linker will only fill in little-endian values during relocation, so we are forced to use little-endian symbol values. To preserve most of the original portability of the symbol table format, we make the table itself say whether it uses big- or little-endian values. If the table begins with the magic sequence fe ff ff ff 00 00 then the actual table begins after those six bytes and contains little-endian symbol values. Otherwise, the table is in the original format and contains big-endian symbol values. The magic sequence looks like an "end of table" entry (the fifth byte is zero), so legacy readers will see a little-endian table as an empty table. All the gc architectures are little-endian today, so the practical effect of this CL is to make all the generated tables little-endian, but if a big-endian system comes along, ld will not generate the magic sequence, and the various readers will fall back to the original big-endian interpretation. R=ken2 CC=golang-dev https://golang.org/cl/7066043 --- diff --git a/doc/go1.1.html b/doc/go1.1.html index eaaa90c580..55c76562e0 100644 --- a/doc/go1.1.html +++ b/doc/go1.1.html @@ -66,6 +66,15 @@ to adjust frame pointer offsets. The implementation now includes a built-in data race detector.

+

Symbol table

+ +

+In the gc toolchain, the symbol table format has been extended to allow +little-endian encoding of symbol values, and the extension is used in +binaries generated by the Go 1.1 version of the gc linker. +To the Go 1.0 toolchain and libraries, these new symbol tables appear empty. +

+

Changes to the standard library

debug/elf

diff --git a/src/cmd/ld/data.c b/src/cmd/ld/data.c index 89f73ec99f..88808d75f7 100644 --- a/src/cmd/ld/data.c +++ b/src/cmd/ld/data.c @@ -150,14 +150,14 @@ relocsym(Sym *s) int32 i, off, siz, fl; vlong o; uchar *cast; - + cursym = s; memset(&p, 0, sizeof p); for(r=s->r; rr+s->nr; r++) { off = r->off; siz = r->siz; - if(off < 0 || off+(siz&~Rbig) > s->np) { - diag("%s: invalid relocation %d+%d not in [%d,%d)", s->name, off, siz&~Rbig, 0, s->np); + if(off < 0 || off+siz > s->np) { + diag("%s: invalid relocation %d+%d not in [%d,%d)", s->name, off, siz, 0, s->np); continue; } if(r->sym != S && (r->sym->type & SMASK == 0 || r->sym->type & SMASK == SXREF)) { @@ -198,20 +198,6 @@ relocsym(Sym *s) default: cursym = s; diag("bad reloc size %#ux for %s", siz, r->sym->name); - case 4 + Rbig: - fl = o; - s->p[off] = fl>>24; - s->p[off+1] = fl>>16; - s->p[off+2] = fl>>8; - s->p[off+3] = fl; - break; - case 4 + Rlittle: - fl = o; - s->p[off] = fl; - s->p[off+1] = fl>>8; - s->p[off+2] = fl>>16; - s->p[off+3] = fl>>24; - break; case 4: fl = o; cast = (uchar*)&fl; @@ -223,7 +209,7 @@ relocsym(Sym *s) for(i=0; i<8; i++) s->p[off+i] = cast[inuxi8[i]]; break; - } + } } } @@ -231,7 +217,7 @@ void reloc(void) { Sym *s; - + if(debug['v']) Bprint(&bso, "%5.2f reloc\n", cputime()); Bflush(&bso); @@ -246,10 +232,10 @@ void dynrelocsym(Sym *s) { Reloc *r; - + if(HEADTYPE == Hwindows) { Sym *rel, *targ; - + rel = lookup(".rel", 0); if(s == rel) return; @@ -259,7 +245,7 @@ dynrelocsym(Sym *s) targ->plt = rel->size; r->sym = rel; r->add = targ->plt; - + // jmp *addr if(thechar == '8') { adduint8(rel, 0xff); @@ -291,7 +277,7 @@ void dynreloc(void) { Sym *s; - + // -d supresses dynamic loader format, so we may as well not // compute these sections or mark their symbols as reachable. if(debug['d'] && HEADTYPE != Hwindows) @@ -364,12 +350,12 @@ savedata(Sym *s, Prog *p, char *pn) break; } break; - + case D_SCONST: for(i=0; ip[off+i] = p->to.scon[i]; break; - + case D_CONST: if(p->to.sym) goto Addr; @@ -450,12 +436,12 @@ blk(Sym *start, int32 addr, int32 size) errorexit(); } } - + for(; addr < eaddr; addr++) cput(0); cflush(); } - + void codeblk(int32 addr, int32 size) { @@ -498,7 +484,7 @@ codeblk(int32 addr, int32 size) Bprint(&bso, "%.6llux\t%-20s | foreign text\n", (vlong)addr, sym->name); n = sym->size; q = sym->p; - + while(n >= 16) { Bprint(&bso, "%.6ux\t%-20.16I\n", addr, q); addr += 16; @@ -510,7 +496,7 @@ codeblk(int32 addr, int32 size) addr += n; continue; } - + Bprint(&bso, "%.6llux\t%-20s | %P\n", (vlong)sym->value, sym->name, p); for(p = p->link; p != P; p = p->link) { if(p->link != P) @@ -532,7 +518,7 @@ codeblk(int32 addr, int32 size) } Bflush(&bso); } - + void datblk(int32 addr, int32 size) { @@ -595,7 +581,7 @@ addstrdata(char *name, char *value) { Sym *s, *sp; char *p; - + p = smprint("%s.str", name); sp = lookup(p, 0); free(p); @@ -775,7 +761,7 @@ addpcrelplus(Sym *s, Sym *t, int32 add) { vlong i; Reloc *r; - + if(s->type == 0) s->type = SDATA; s->reachable = 1; @@ -972,7 +958,7 @@ dodata(void) * symbol, which is itself data. */ dynreloc(); - + /* some symbols may no longer belong in datap (Mach-O) */ for(l=&datap; (s=*l) != nil; ) { if(s->type <= STEXT || SXREF <= s->type) @@ -1099,7 +1085,7 @@ dodata(void) } sect->len = datsize - sect->vaddr; datsize = rnd(datsize, PtrSize); - + /* gcdata */ sect = addsection(&segtext, ".gcdata", 04); sect->vaddr = datsize; @@ -1169,7 +1155,7 @@ textaddress(void) addsection(&segtext, ".text", 05); // Assign PCs in text segment. - // Could parallelize, by assigning to text + // Could parallelize, by assigning to text // and then letting threads copy down, but probably not worth it. sect = segtext.sect; va = INITTEXT; @@ -1192,7 +1178,7 @@ textaddress(void) } va += sym->size; } - + // Align end of code so that rodata starts aligned. // 128 bytes is likely overkill but definitely cheap. va = rnd(va, 128); @@ -1267,7 +1253,7 @@ address(void) for(sub = sym->sub; sub != nil; sub = sub->sub) sub->value += sym->value; } - + xdefine("text", STEXT, text->vaddr); xdefine("etext", STEXT, text->vaddr + text->len); xdefine("rodata", SRODATA, rodata->vaddr); diff --git a/src/cmd/ld/lib.h b/src/cmd/ld/lib.h index cf334a4bf7..0a582b9b10 100644 --- a/src/cmd/ld/lib.h +++ b/src/cmd/ld/lib.h @@ -31,7 +31,7 @@ enum { Sxxx, - + /* order here is order in output file */ STEXT, SMACHOPLT, @@ -246,12 +246,6 @@ struct Endian extern Endian be, le; -// relocation size bits -enum { - Rbig = 128, - Rlittle = 64, -}; - /* set by call to mywhatsys() */ extern char* goroot; extern char* goarch; diff --git a/src/cmd/ld/symtab.c b/src/cmd/ld/symtab.c index a27b181edc..92627f5986 100644 --- a/src/cmd/ld/symtab.c +++ b/src/cmd/ld/symtab.c @@ -199,7 +199,7 @@ static void slputb(int32 v) { uchar *p; - + symgrow(symt, symt->size+4); p = symt->p + symt->size; *p++ = v>>24; @@ -209,6 +209,22 @@ slputb(int32 v) symt->size += 4; } +static void +slputl(int32 v) +{ + uchar *p; + + symgrow(symt, symt->size+4); + p = symt->p + symt->size; + *p++ = v; + *p++ = v>>8; + *p++ = v>>16; + *p = v>>24; + symt->size += 4; +} + +static void (*slput)(int32); + void wputl(ushort w) { @@ -269,15 +285,24 @@ putsymb(Sym *s, char *name, int t, vlong v, vlong size, int ver, Sym *typ) // l = 8; if(s != nil) { rel = addrel(symt); - rel->siz = l + Rbig; + rel->siz = l; rel->sym = s; rel->type = D_ADDR; rel->off = symt->size; v = 0; - } - if(l == 8) - slputb(v>>32); - slputb(v); + } + + if(l == 8) { + if(slput == slputl) { + slputl(v); + slputl(v>>32); + } else { + slputb(v>>32); + slputb(v); + } + } else + slput(v); + if(ver) t += 'a' - 'A'; scput(t+0x80); /* 0x80 is variable length */ @@ -306,8 +331,8 @@ putsymb(Sym *s, char *name, int t, vlong v, vlong size, int ver, Sym *typ) rel->off = symt->size; } if(l == 8) - slputb(0); - slputb(0); + slput(0); + slput(0); if(debug['n']) { if(t == 'z' || t == 'Z') { @@ -356,7 +381,7 @@ symtab(void) xdefine("end", SBSS, 0); xdefine("epclntab", SRODATA, 0); xdefine("esymtab", SRODATA, 0); - + // pseudo-symbols to mark locations of type, string, and go string data. s = lookup("type.*", 0); s->type = STYPE; @@ -372,7 +397,7 @@ symtab(void) symt->type = SSYMTAB; symt->size = 0; symt->reachable = 1; - + // assign specific types so that they sort together. // within a type they sort by size, so the .* symbols // just defined above will be first. @@ -396,5 +421,25 @@ symtab(void) if(debug['s']) return; + + switch(thechar) { + default: + diag("unknown architecture %c", thechar); + errorexit(); + case '5': + case '6': + case '8': + // magic entry to denote little-endian symbol table + slputl(0xfffffffe); + scput(0); + scput(0); + slput = slputl; + break; + case 'v': + // big-endian (in case one comes along) + slput = slputb; + break; + } + genasmsym(putsymb); } diff --git a/src/libmach/sym.c b/src/libmach/sym.c index 1512d7a4fe..bb758addb5 100644 --- a/src/libmach/sym.c +++ b/src/libmach/sym.c @@ -1,11 +1,11 @@ // Inferno libmach/sym.c // http://code.google.com/p/inferno-os/source/browse/utils/libmach/sym.c // -// Copyright © 1994-1999 Lucent Technologies Inc. -// Power PC support Copyright © 1995-2004 C H Forsyth (forsyth@terzarima.net). -// Portions Copyright © 1997-1999 Vita Nuova Limited. -// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com). -// Revisions Copyright © 2000-2004 Lucent Technologies Inc. and others. +// Copyright © 1994-1999 Lucent Technologies Inc. +// Power PC support Copyright © 1995-2004 C H Forsyth (forsyth@terzarima.net). +// Portions Copyright © 1997-1999 Vita Nuova Limited. +// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com). +// Revisions Copyright © 2000-2004 Lucent Technologies Inc. and others. // Portions Copyright © 2009 The Go Authors. All rights reserved. // // Permission is hereby granted, free of charge, to any person obtaining a copy @@ -113,12 +113,18 @@ syminit(int fd, Fhdr *fp) vlong vl; Biobuf b; int svalsz; + uvlong (*swav)(uvlong); + uint32 (*swal)(uint32); + uchar buf[6]; if(fp->symsz == 0) return 0; if(fp->type == FNONE) return 0; + swav = beswav; + swal = beswal; + cleansyms(); textseg(fp->txtaddr, fp); /* minimum symbol record size = 4+1+2 bytes */ @@ -129,6 +135,15 @@ syminit(int fd, Fhdr *fp) } Binit(&b, fd, OREAD); Bseek(&b, fp->symoff, 0); + memset(buf, 0, sizeof buf); + Bread(&b, buf, sizeof buf); + if(memcmp(buf, "\xfe\xff\xff\xff\x00\x00", 6) == 0) { + swav = leswav; + swal = leswal; + } else { + Bseek(&b, fp->symoff, 0); + } + nsym = 0; size = 0; for(p = symbols; size < fp->symsz; p++, nsym++) { @@ -136,13 +151,13 @@ syminit(int fd, Fhdr *fp) svalsz = 8; if(Bread(&b, &vl, 8) != 8) return symerrmsg(8, "symbol"); - p->value = beswav(vl); + p->value = swav(vl); } else{ svalsz = 4; if(Bread(&b, &l, 4) != 4) return symerrmsg(4, "symbol"); - p->value = (u32int)beswal(l); + p->value = (u32int)swal(l); } if(Bread(&b, &p->type, sizeof(p->type)) != sizeof(p->type)) return symerrmsg(sizeof(p->value), "symbol"); @@ -155,12 +170,12 @@ syminit(int fd, Fhdr *fp) if(svalsz == 8){ if(Bread(&b, &vl, 8) != 8) return symerrmsg(8, "symbol"); - p->gotype = beswav(vl); + p->gotype = swav(vl); } else{ if(Bread(&b, &l, 4) != 4) return symerrmsg(4, "symbol"); - p->gotype = (u32int)beswal(l); + p->gotype = (u32int)swal(l); } size += svalsz; diff --git a/src/pkg/debug/gosym/pclntab_test.go b/src/pkg/debug/gosym/pclntab_test.go index ade704335d..5f2242eba0 100644 --- a/src/pkg/debug/gosym/pclntab_test.go +++ b/src/pkg/debug/gosym/pclntab_test.go @@ -129,7 +129,7 @@ func TestLineFromAline(t *testing.T) { if !ok { t.Errorf("file %s starts on line %d", path, line) } else if line != ll+1 { - t.Errorf("expected next line of file %s to be %d, got %d", path, ll+1, line) + t.Fatalf("expected next line of file %s to be %d, got %d", path, ll+1, line) } lastline[path] = line } diff --git a/src/pkg/debug/gosym/symtab.go b/src/pkg/debug/gosym/symtab.go index 52d7d55a33..cc01e0b9d6 100644 --- a/src/pkg/debug/gosym/symtab.go +++ b/src/pkg/debug/gosym/symtab.go @@ -13,6 +13,7 @@ package gosym // and the Go format is the runtime source, specifically ../../runtime/symtab.c. import ( + "bytes" "encoding/binary" "fmt" "strconv" @@ -104,11 +105,18 @@ type sym struct { name []byte } +var littleEndianSymtab = []byte{0xFE, 0xFF, 0xFF, 0xFF, 0x00, 0x00} + func walksymtab(data []byte, fn func(sym) error) error { + var order binary.ByteOrder = binary.BigEndian + if bytes.HasPrefix(data, littleEndianSymtab) { + data = data[6:] + order = binary.LittleEndian + } var s sym p := data for len(p) >= 6 { - s.value = binary.BigEndian.Uint32(p[0:4]) + s.value = order.Uint32(p[0:4]) typ := p[4] if typ&0x80 == 0 { return &DecodingError{len(data) - len(p) + 4, "bad symbol type", typ} @@ -139,7 +147,7 @@ func walksymtab(data []byte, fn func(sym) error) error { } s.name = p[0:i] i += nnul - s.gotype = binary.BigEndian.Uint32(p[i : i+4]) + s.gotype = order.Uint32(p[i : i+4]) p = p[i+4:] fn(s) } diff --git a/src/pkg/runtime/symtab.c b/src/pkg/runtime/symtab.c index 74b0071476..5df9fd2d3d 100644 --- a/src/pkg/runtime/symtab.c +++ b/src/pkg/runtime/symtab.c @@ -40,13 +40,26 @@ walksymtab(void (*fn)(Sym*)) { byte *p, *ep, *q; Sym s; + int32 bigend; p = symtab; ep = esymtab; + + // Default is big-endian value encoding. + // If table begins fe ff ff ff 00 00, little-endian. + bigend = 1; + if(symtab[0] == 0xfe && symtab[1] == 0xff && symtab[2] == 0xff && symtab[3] == 0xff && symtab[4] == 0x00 && symtab[5] == 0x00) { + p += 6; + bigend = 0; + } while(p < ep) { if(p + 7 > ep) break; - s.value = ((uint32)p[0]<<24) | ((uint32)p[1]<<16) | ((uint32)p[2]<<8) | ((uint32)p[3]); + + if(bigend) + s.value = ((uint32)p[0]<<24) | ((uint32)p[1]<<16) | ((uint32)p[2]<<8) | ((uint32)p[3]); + else + s.value = ((uint32)p[3]<<24) | ((uint32)p[2]<<16) | ((uint32)p[1]<<8) | ((uint32)p[0]); if(!(p[4]&0x80)) break; @@ -304,7 +317,7 @@ splitpcln(void) line += *p++; else line -= *p++ - 64; - + // pc, line now match. // Because the state machine begins at pc==entry and line==0, // it can happen - just at the beginning! - that the update may @@ -326,7 +339,7 @@ splitpcln(void) while(f < ef && pc >= (f+1)->entry); f->pcln.array = p; // pc0 and ln0 are the starting values for - // the loop over f->pcln, so pc must be + // the loop over f->pcln, so pc must be // adjusted by the same pcquant update // that we're going to do as we continue our loop. f->pc0 = pc + pcquant; @@ -352,11 +365,11 @@ runtime·funcline(Func *f, uintptr targetpc) uintptr pc; int32 line; int32 pcquant; - + enum { debug = 0 }; - + switch(thechar) { case '5': pcquant = 4; @@ -383,7 +396,7 @@ runtime·funcline(Func *f, uintptr targetpc) if(debug && !runtime·panicking) runtime·printf("pc<%p targetpc=%p line=%d\n", pc, targetpc, line); - + // If the pc has advanced too far or we're out of data, // stop and the last known line number. if(pc > targetpc || p >= ep) @@ -519,7 +532,7 @@ static bool hasprefix(String s, int8 *p) { int32 i; - + for(i=0; i 1 || contains(f->name, ".") && !hasprefix(f->name, "runtime.");