1 /* -----------------------------------------------------------------------------
2 * $Id: Linker.c,v 1.18 2001/02/06 14:44:53 simonmar Exp $
4 * (c) The GHC Team, 2000
8 * ---------------------------------------------------------------------------*/
16 #include "StoragePriv.h"
18 #ifdef HAVE_SYS_TYPES_H
19 #include <sys/types.h>
22 #ifdef HAVE_SYS_STAT_H
30 #ifdef GHCI /* endif is right at end of file */
32 #if defined(linux_TARGET_OS) || defined(solaris2_TARGET_OS) || defined(freebsd_TARGET_OS)
34 #elif defined(cygwin32_TARGET_OS) || defined (mingw32_TARGET_OS)
35 #define OBJFORMAT_PEi386
38 /* A bucket in the symbol hash-table. Primarily, maps symbol names to
39 * absolute addresses. All symbols from a given module are linked
40 * together, so they can be freed at the same time. There's also a
41 * bucket link field for the hash table.
43 typedef struct _SymbolVal {
48 typedef enum { OBJECT_LOADED, OBJECT_RESOLVED } OStatus;
50 /* Indication of section kinds for loaded objects. Needed by
51 the GC for deciding whether or not a pointer on the stack
54 typedef enum { SECTIONKIND_CODE_OR_RODATA,
57 SECTIONKIND_NOINFOAVAIL }
60 typedef struct { void* start; void* end; SectionKind kind; }
63 /* Top-level structure for an object module. One of these is allocated
64 * for each object file in use.
66 typedef struct _ObjectCode {
70 char* formatName; /* eg "ELF32", "DLL", "COFF", etc. */
75 /* ptr to malloc'd lump of memory holding the obj file */
78 /* The section-kind entries for this object module. Dynamically expands. */
82 /* Allow a chain of these things */
83 struct _ObjectCode * next;
87 /* Hash table mapping symbol names to Symbol */
88 /*Str*/HashTable *symhash;
90 /* List of currently loaded objects */
93 #if defined(OBJFORMAT_ELF)
94 static int ocVerifyImage_ELF ( ObjectCode* oc );
95 static int ocGetNames_ELF ( ObjectCode* oc );
96 static int ocResolve_ELF ( ObjectCode* oc );
97 #elif defined(OBJFORMAT_PEi386)
98 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
99 static int ocGetNames_PEi386 ( ObjectCode* oc );
100 static int ocResolve_PEi386 ( ObjectCode* oc );
103 /* -----------------------------------------------------------------------------
104 * Built-in symbols from the RTS
107 #define RTS_SYMBOLS \
109 Sym(stg_gc_enter_1) \
116 Sym(stg_gc_unbx_r1) \
121 SymX(stg_update_PAP) \
122 SymX(stg_ap_2_upd_info) \
123 SymX(stg_ap_3_upd_info) \
124 SymX(stg_ap_4_upd_info) \
125 SymX(stg_ap_5_upd_info) \
126 SymX(stg_ap_6_upd_info) \
127 SymX(stg_ap_7_upd_info) \
128 SymX(stg_ap_8_upd_info) \
129 SymX(stg_sel_0_upd_info) \
130 SymX(stg_sel_1_upd_info) \
131 SymX(stg_sel_2_upd_info) \
132 SymX(stg_sel_3_upd_info) \
133 SymX(stg_sel_4_upd_info) \
134 SymX(stg_sel_5_upd_info) \
135 SymX(stg_sel_6_upd_info) \
136 SymX(stg_sel_7_upd_info) \
137 SymX(stg_sel_8_upd_info) \
138 SymX(stg_sel_9_upd_info) \
139 SymX(stg_sel_10_upd_info) \
140 SymX(stg_sel_11_upd_info) \
141 SymX(stg_sel_12_upd_info) \
142 SymX(stg_sel_13_upd_info) \
143 SymX(stg_sel_14_upd_info) \
144 SymX(stg_sel_15_upd_info) \
145 SymX(stg_upd_frame_info) \
146 SymX(stg_seq_frame_info) \
147 SymX(stg_CAF_BLACKHOLE_info) \
148 SymX(stg_IND_STATIC_info) \
149 SymX(stg_EMPTY_MVAR_info) \
150 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
151 SymX(stg_WEAK_info) \
152 SymX(stg_CHARLIKE_closure) \
153 SymX(stg_INTLIKE_closure) \
155 SymX(newBCOzh_fast) \
156 SymX(mkApUpd0zh_fast) \
157 SymX(putMVarzh_fast) \
158 SymX(newMVarzh_fast) \
159 SymX(takeMVarzh_fast) \
160 SymX(tryTakeMVarzh_fast) \
166 SymX(killThreadzh_fast) \
167 SymX(waitReadzh_fast) \
168 SymX(waitWritezh_fast) \
169 SymX(suspendThread) \
171 SymX(stackOverflow) \
172 SymX(int2Integerzh_fast) \
173 SymX(word2Integerzh_fast) \
174 SymX(mkForeignObjzh_fast) \
175 SymX(__encodeDouble) \
176 SymX(decodeDoublezh_fast) \
177 SymX(decodeFloatzh_fast) \
178 SymX(gcdIntegerzh_fast) \
179 SymX(newArrayzh_fast) \
180 SymX(unsafeThawArrayzh_fast) \
181 SymX(newByteArrayzh_fast) \
182 SymX(newMutVarzh_fast) \
183 SymX(quotRemIntegerzh_fast) \
184 SymX(quotIntegerzh_fast) \
185 SymX(remIntegerzh_fast) \
186 SymX(divExactIntegerzh_fast) \
187 SymX(divModIntegerzh_fast) \
188 SymX(timesIntegerzh_fast) \
189 SymX(minusIntegerzh_fast) \
190 SymX(plusIntegerzh_fast) \
191 SymX(andIntegerzh_fast) \
192 SymX(orIntegerzh_fast) \
193 SymX(xorIntegerzh_fast) \
194 SymX(complementIntegerzh_fast) \
195 SymX(mkWeakzh_fast) \
196 SymX(makeStableNamezh_fast) \
197 SymX(finalizzeWeakzh_fast) \
198 SymX(blockAsyncExceptionszh_fast) \
199 SymX(unblockAsyncExceptionszh_fast) \
201 SymX(isDoubleInfinite) \
202 SymX(isDoubleDenormalized) \
203 SymX(isDoubleNegativeZero) \
204 SymX(__encodeFloat) \
206 SymX(isFloatInfinite) \
207 SymX(isFloatDenormalized) \
208 SymX(isFloatNegativeZero) \
209 SymX(__int_encodeFloat) \
210 SymX(__int_encodeDouble) \
211 SymX(__gmpz_cmp_si) \
212 SymX(__gmpz_cmp_ui) \
217 SymX(resetNonBlockingFd) \
219 SymX(stable_ptr_table) \
220 SymX(shutdownHaskellAndExit) \
221 Sym(stg_enterStackTop) \
222 Sym(stg_yield_to_interpreter) \
226 Sym(__init_PrelGHC) \
227 SymX(freeHaskellFunctionPtr) \
230 SymX(NoRunnableThreadsHook) \
231 SymX(StackOverflowHook) \
232 SymX(OutOfHeapHook) \
233 SymX(MallocFailHook) \
234 SymX(PatErrorHdrHook) \
237 SymX(PostTraceHook) \
238 SymX(stg_sig_install) \
240 SymX(createAdjustor) \
242 SymX(rts_mkStablePtr) \
245 SymX(rts_checkSchedStatus) \
248 #ifndef SUPPORT_LONG_LONGS
249 #define RTS_LONG_LONG_SYMS /* nothing */
251 #define RTS_LONG_LONG_SYMS \
264 SymX(stg_remWord64) \
265 SymX(stg_quotWord64) \
267 SymX(stg_quotInt64) \
268 SymX(stg_negateInt64) \
269 SymX(stg_plusInt64) \
270 SymX(stg_minusInt64) \
271 SymX(stg_timesInt64) \
277 SymX(stg_shiftRL64) \
278 SymX(stg_iShiftL64) \
279 SymX(stg_iShiftRL64) \
280 SymX(stg_iShiftRA64) \
281 SymX(stg_intToInt64) \
282 SymX(stg_int64ToInt) \
283 SymX(stg_int64ToWord64) \
284 SymX(stg_wordToWord64) \
285 SymX(stg_word64ToWord) \
286 SymX(stg_word64ToInt64) \
287 SymX(int64ToIntegerzh_fast) \
288 SymX(word64ToIntegerzh_fast)
289 #endif /* SUPPORT_LONG_LONGS */
291 /* entirely bogus claims about types of these symbols */
292 #define Sym(vvv) extern void (vvv);
293 #define SymX(vvv) /**/
298 #ifdef LEADING_UNDERSCORE
299 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
301 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
304 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
306 #define SymX(vvv) Sym(vvv)
308 static SymbolVal rtsSyms[] = {
311 { 0, 0 } /* sentinel */
314 /* -----------------------------------------------------------------------------
315 * initialize the object linker
317 static void *dl_prog_handle;
324 symhash = allocStrHashTable();
326 /* populate the symbol table with stuff from the RTS */
327 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
328 insertStrHashTable(symhash, sym->lbl, sym);
331 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
334 /* -----------------------------------------------------------------------------
335 * lookup a symbol in the hash table
338 lookupSymbol( char *lbl )
341 val = lookupStrHashTable(symhash, lbl);
344 return dlsym(dl_prog_handle, lbl);
350 /* -----------------------------------------------------------------------------
351 * Load an obj (populate the global symbol table, but don't resolve yet)
353 * Returns: 1 if ok, 0 on error.
356 loadObj( char *path )
364 /* assert that we haven't already loaded this object */
367 for (o = objects; o; o = o->next)
368 ASSERT(strcmp(o->fileName, path));
372 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
374 # if defined(OBJFORMAT_ELF)
375 oc->formatName = "ELF";
376 # elif defined(OBJFORMAT_PEi386)
377 oc->formatName = "PEi386";
380 barf("loadObj: not implemented on this platform");
384 if (r == -1) { return 0; }
386 /* sigh, stdup() isn't a POSIX function, so do it the long way */
387 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
388 strcpy(oc->fileName, path);
390 oc->fileSize = st.st_size;
391 oc->image = stgMallocBytes( st.st_size, "loadObj(image)" );
395 /* chain it onto the list of objects */
399 /* load the image into memory */
400 f = fopen(path, "rb");
402 barf("loadObj: can't read `%s'", path);
404 n = fread ( oc->image, 1, oc->fileSize, f );
405 if (n != oc->fileSize) {
407 barf("loadObj: error whilst reading `%s'", path);
410 /* verify the in-memory image */
411 # if defined(OBJFORMAT_ELF)
412 r = ocVerifyImage_ELF ( oc );
413 # elif defined(OBJFORMAT_PEi386)
414 r = ocVerifyImage_PEi386 ( oc );
416 barf("loadObj: no verify method");
418 if (!r) { return r; }
420 /* build the symbol list for this image */
421 # if defined(OBJFORMAT_ELF)
422 r = ocGetNames_ELF ( oc );
423 # elif defined(OBJFORMAT_PEi386)
424 r = ocGetNames_PEi386 ( oc );
426 barf("loadObj: no getNames method");
428 if (!r) { return r; }
430 /* loaded, but not resolved yet */
431 oc->status = OBJECT_LOADED;
436 /* -----------------------------------------------------------------------------
437 * resolve all the currently unlinked objects in memory
439 * Returns: 1 if ok, 0 on error.
447 for (oc = objects; oc; oc = oc->next) {
448 if (oc->status != OBJECT_RESOLVED) {
449 # if defined(OBJFORMAT_ELF)
450 r = ocResolve_ELF ( oc );
451 # elif defined(OBJFORMAT_PEi386)
452 r = ocResolve_PEi386 ( oc );
454 barf("link: not implemented on this platform");
456 if (!r) { return r; }
457 oc->status = OBJECT_RESOLVED;
463 /* -----------------------------------------------------------------------------
464 * delete an object from the pool
467 unloadObj( char *path )
469 ObjectCode *oc, *prev;
472 for (oc = objects; oc; prev = oc, oc = oc->next) {
473 if (!strcmp(oc->fileName,path)) {
475 /* Remove all the mappings for the symbols within this
480 for (s = oc->symbols; s < oc->symbols + oc->n_symbols; s++) {
481 if (s->lbl != NULL) {
482 removeStrHashTable(symhash, s->lbl, NULL);
490 prev->next = oc->next;
493 /* We're going to leave this in place, in case there are
494 any pointers from the heap into it: */
495 /* free(oc->image); */
504 belch("unloadObj: can't find `%s' to unload", path);
508 /* --------------------------------------------------------------------------
509 * PEi386 specifics (Win32 targets)
510 * ------------------------------------------------------------------------*/
512 /* The information for this linker comes from
513 Microsoft Portable Executable
514 and Common Object File Format Specification
515 revision 5.1 January 1998
516 which SimonM says comes from the MS Developer Network CDs.
520 #if defined(OBJFORMAT_PEi386)
524 typedef unsigned char UChar;
525 typedef unsigned short UInt16;
526 typedef unsigned int UInt32;
533 UInt16 NumberOfSections;
534 UInt32 TimeDateStamp;
535 UInt32 PointerToSymbolTable;
536 UInt32 NumberOfSymbols;
537 UInt16 SizeOfOptionalHeader;
538 UInt16 Characteristics;
542 #define sizeof_COFF_header 20
549 UInt32 VirtualAddress;
550 UInt32 SizeOfRawData;
551 UInt32 PointerToRawData;
552 UInt32 PointerToRelocations;
553 UInt32 PointerToLinenumbers;
554 UInt16 NumberOfRelocations;
555 UInt16 NumberOfLineNumbers;
556 UInt32 Characteristics;
560 #define sizeof_COFF_section 40
567 UInt16 SectionNumber;
570 UChar NumberOfAuxSymbols;
574 #define sizeof_COFF_symbol 18
579 UInt32 VirtualAddress;
580 UInt32 SymbolTableIndex;
585 #define sizeof_COFF_reloc 10
588 /* From PE spec doc, section 3.3.2 */
589 #define IMAGE_FILE_RELOCS_STRIPPED 0x0001
590 #define IMAGE_FILE_EXECUTABLE_IMAGE 0x0002
591 #define IMAGE_FILE_DLL 0x2000
592 #define IMAGE_FILE_SYSTEM 0x1000
593 #define IMAGE_FILE_BYTES_REVERSED_HI 0x8000
594 #define IMAGE_FILE_BYTES_REVERSED_LO 0x0080
595 #define IMAGE_FILE_32BIT_MACHINE 0x0100
597 /* From PE spec doc, section 5.4.2 and 5.4.4 */
598 #define IMAGE_SYM_CLASS_EXTERNAL 2
599 #define IMAGE_SYM_CLASS_STATIC 3
600 #define IMAGE_SYM_UNDEFINED 0
602 /* From PE spec doc, section 4.1 */
603 #define IMAGE_SCN_CNT_CODE 0x00000020
604 #define IMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
606 /* From PE spec doc, section 5.2.1 */
607 #define IMAGE_REL_I386_DIR32 0x0006
608 #define IMAGE_REL_I386_REL32 0x0014
611 /* We use myindex to calculate array addresses, rather than
612 simply doing the normal subscript thing. That's because
613 some of the above structs have sizes which are not
614 a whole number of words. GCC rounds their sizes up to a
615 whole number of words, which means that the address calcs
616 arising from using normal C indexing or pointer arithmetic
617 are just plain wrong. Sigh.
620 myindex ( int scale, int index, void* base )
623 ((UChar*)base) + scale * index;
628 printName ( UChar* name, UChar* strtab )
630 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
631 UInt32 strtab_offset = * (UInt32*)(name+4);
632 fprintf ( stderr, "%s", strtab + strtab_offset );
635 for (i = 0; i < 8; i++) {
636 if (name[i] == 0) break;
637 fprintf ( stderr, "%c", name[i] );
644 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
646 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
647 UInt32 strtab_offset = * (UInt32*)(name+4);
648 strncpy ( dst, strtab+strtab_offset, dstSize );
654 if (name[i] == 0) break;
664 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
667 /* If the string is longer than 8 bytes, look in the
668 string table for it -- this will be correctly zero terminated.
670 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
671 UInt32 strtab_offset = * (UInt32*)(name+4);
672 return ((UChar*)strtab) + strtab_offset;
674 /* Otherwise, if shorter than 8 bytes, return the original,
675 which by defn is correctly terminated.
677 if (name[7]==0) return name;
678 /* The annoying case: 8 bytes. Copy into a temporary
679 (which is never freed ...)
683 strncpy(newstr,name,8);
690 /* Just compares the short names (first 8 chars) */
691 static COFF_section *
692 findPEi386SectionCalled ( ObjectCode* oc, char* name )
696 = (COFF_header*)(oc->image);
699 ((UChar*)(oc->image))
700 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
702 for (i = 0; i < hdr->NumberOfSections; i++) {
705 COFF_section* section_i
707 myindex ( sizeof_COFF_section, i, sectab );
708 n1 = (UChar*) &(section_i->Name);
710 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
711 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
712 n1[6]==n2[6] && n1[7]==n2[7])
721 zapTrailingAtSign ( UChar* sym )
724 if (sym[0] == 0) return;
726 while (sym[i] != 0) i++;
729 while (j > 0 && isdigit(sym[j])) j--;
730 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
735 ocVerifyImage_PEi386 ( ObjectCode* oc )
739 COFF_section* sectab;
743 hdr = (COFF_header*)(oc->image);
744 sectab = (COFF_section*) (
745 ((UChar*)(oc->image))
746 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
748 symtab = (COFF_symbol*) (
749 ((UChar*)(oc->image))
750 + hdr->PointerToSymbolTable
752 strtab = ((UChar*)(oc->image))
753 + hdr->PointerToSymbolTable
754 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
756 if (hdr->Machine != 0x14c) {
757 oc->errMsg("Not x86 PEi386");
760 if (hdr->SizeOfOptionalHeader != 0) {
761 oc->errMsg("PEi386 with nonempty optional header");
764 if ( /* (hdr->Characteristics & IMAGE_FILE_RELOCS_STRIPPED) || */
765 (hdr->Characteristics & IMAGE_FILE_EXECUTABLE_IMAGE) ||
766 (hdr->Characteristics & IMAGE_FILE_DLL) ||
767 (hdr->Characteristics & IMAGE_FILE_SYSTEM) ) {
768 oc->errMsg("Not a PEi386 object file");
771 if ( (hdr->Characteristics & IMAGE_FILE_BYTES_REVERSED_HI) ||
772 !(hdr->Characteristics & IMAGE_FILE_32BIT_MACHINE) ) {
773 oc->errMsg("Invalid PEi386 word size or endiannness");
777 if (!verb) return TRUE;
778 /* No further verification after this point; only debug printing. */
781 "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
783 "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
785 "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
787 fprintf ( stderr, "\n" );
789 "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
791 "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
793 "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
795 "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
797 "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
799 "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
801 "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
803 fprintf ( stderr, "\n" );
804 fprintf ( stderr, "string table has size 0x%x\n", * (UInt32*)strtab );
805 fprintf ( stderr, "---START of string table---\n");
806 for (i = 4; i < *(UInt32*)strtab; i++) {
808 fprintf ( stderr, "\n"); else
809 fprintf( stderr, "%c", strtab[i] );
811 fprintf ( stderr, "--- END of string table---\n");
813 fprintf ( stderr, "\n" );
814 for (i = 0; i < hdr->NumberOfSections; i++) {
816 COFF_section* sectab_i
818 myindex ( sizeof_COFF_section, i, sectab );
825 printName ( sectab_i->Name, strtab );
834 sectab_i->VirtualSize,
835 sectab_i->VirtualAddress,
836 sectab_i->SizeOfRawData,
837 sectab_i->PointerToRawData,
838 sectab_i->NumberOfRelocations,
839 sectab_i->PointerToRelocations
841 reltab = (COFF_reloc*) (
842 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
844 for (j = 0; j < sectab_i->NumberOfRelocations; j++) {
846 COFF_reloc* rel = (COFF_reloc*)
847 myindex ( sizeof_COFF_reloc, j, reltab );
849 " type 0x%-4x vaddr 0x%-8x name `",
851 rel->VirtualAddress );
853 myindex ( sizeof_COFF_symbol, rel->SymbolTableIndex, symtab );
854 printName ( sym->Name, strtab );
855 fprintf ( stderr, "'\n" );
857 fprintf ( stderr, "\n" );
861 fprintf ( stderr, "\n" );
864 COFF_symbol* symtab_i;
865 if (i >= hdr->NumberOfSymbols) break;
866 symtab_i = (COFF_symbol*)
867 myindex ( sizeof_COFF_symbol, i, symtab );
873 printName ( symtab_i->Name, strtab );
882 (Int32)(symtab_i->SectionNumber) - 1,
883 (UInt32)symtab_i->Type,
884 (UInt32)symtab_i->StorageClass,
885 (UInt32)symtab_i->NumberOfAuxSymbols
887 i += symtab_i->NumberOfAuxSymbols;
891 fprintf ( stderr, "\n" );
898 ocGetNames_PEi386 ( ObjectCode* oc )
901 COFF_section* sectab;
909 hdr = (COFF_header*)(oc->image);
910 sectab = (COFF_section*) (
911 ((UChar*)(oc->image))
912 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
914 symtab = (COFF_symbol*) (
915 ((UChar*)(oc->image))
916 + hdr->PointerToSymbolTable
918 strtab = ((UChar*)(oc->image))
919 + hdr->PointerToSymbolTable
920 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
922 /* Copy exported symbols into the ObjectCode. */
925 COFF_symbol* symtab_i;
926 if (i >= hdr->NumberOfSymbols) break;
927 symtab_i = (COFF_symbol*)
928 myindex ( sizeof_COFF_symbol, i, symtab );
930 if (symtab_i->StorageClass == IMAGE_SYM_CLASS_EXTERNAL &&
931 symtab_i->SectionNumber != IMAGE_SYM_UNDEFINED) {
933 /* This symbol is global and defined, viz, exported */
934 COFF_section* sectabent;
936 sname = cstring_from_COFF_symbol_name (
937 symtab_i->Name, strtab
940 oc->errMsg("Out of memory when copying PEi386 symbol");
944 /* for IMAGE_SYMCLASS_EXTERNAL
945 && !IMAGE_SYM_UNDEFINED,
946 the address of the symbol is:
947 address of relevant section + offset in section
949 sectabent = (COFF_section*)
950 myindex ( sizeof_COFF_section,
951 symtab_i->SectionNumber-1,
953 addr = ((UChar*)(oc->image))
954 + (sectabent->PointerToRawData
956 /* fprintf ( stderr, "addSymbol %p `%s'\n", addr,sname); */
957 if (!addSymbol(oc,sname,addr)) return FALSE;
959 i += symtab_i->NumberOfAuxSymbols;
963 oc->sections = stgMallocBytes( NumberOfSections * sizeof(Section),
964 "ocGetNamesPEi386" );
966 /* Copy section information into the ObjectCode. */
967 for (i = 0; i < hdr->NumberOfSections; i++) {
973 COFF_section* sectab_i
975 myindex ( sizeof_COFF_section, i, sectab );
976 /* fprintf ( stderr, "section name = %s\n", sectab_i->Name ); */
979 /* I'm sure this is the Right Way to do it. However, the
980 alternative of testing the sectab_i->Name field seems to
983 if (sectab_i->Characteristics & IMAGE_SCN_CNT_CODE ||
984 sectab_i->Characteristics & IMAGE_SCN_CNT_INITIALIZED_DATA)
985 kind = SECTIONKIND_CODE_OR_RODATA;
988 if (0==strcmp(".text",sectab_i->Name))
989 kind = SECTIONKIND_CODE_OR_RODATA;
990 if (0==strcmp(".data",sectab_i->Name) ||
991 0==strcmp(".bss",sectab_i->Name))
992 kind = SECTIONKIND_RWDATA;
994 start = ((UChar*)(oc->image))
995 + sectab_i->PointerToRawData;
997 + sectab_i->SizeOfRawData - 1;
999 if (kind != SECTIONKIND_OTHER) {
1000 addSection ( oc, start, end, kind );
1002 fprintf ( stderr, "unknown section name = `%s'\n",
1004 oc->errMsg("Unknown PEi386 section name");
1014 ocResolve_PEi386 ( ObjectCode* oc, int verb )
1017 COFF_section* sectab;
1018 COFF_symbol* symtab;
1026 char symbol[1000]; // ToDo
1028 hdr = (COFF_header*)(oc->image);
1029 sectab = (COFF_section*) (
1030 ((UChar*)(oc->image))
1031 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1033 symtab = (COFF_symbol*) (
1034 ((UChar*)(oc->image))
1035 + hdr->PointerToSymbolTable
1037 strtab = ((UChar*)(oc->image))
1038 + hdr->PointerToSymbolTable
1039 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1041 for (i = 0; i < hdr->NumberOfSections; i++) {
1042 COFF_section* sectab_i
1044 myindex ( sizeof_COFF_section, i, sectab );
1047 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1049 for (j = 0; j < sectab_i->NumberOfRelocations; j++) {
1051 COFF_reloc* reltab_j
1053 myindex ( sizeof_COFF_reloc, j, reltab );
1055 /* the location to patch */
1057 ((UChar*)(oc->image))
1058 + (sectab_i->PointerToRawData
1059 + reltab_j->VirtualAddress)
1061 /* the existing contents of pP */
1063 /* the symbol to connect to */
1064 sym = (COFF_symbol*)
1065 myindex ( sizeof_COFF_symbol,
1066 reltab_j->SymbolTableIndex, symtab );
1069 "reloc sec %2d num %3d: type 0x%-4x "
1070 "vaddr 0x%-8x name `",
1072 (UInt32)reltab_j->Type,
1073 reltab_j->VirtualAddress );
1074 printName ( sym->Name, strtab );
1075 fprintf ( stderr, "'\n" );
1078 if (sym->StorageClass == IMAGE_SYM_CLASS_STATIC) {
1079 COFF_section* section_sym
1080 = findPEi386SectionCalled ( oc, sym->Name );
1082 fprintf ( stderr, "bad section = `%s'\n", sym->Name );
1083 oc->errMsg("Can't find abovementioned PEi386 section");
1086 S = ((UInt32)(oc->image))
1087 + (section_sym->PointerToRawData
1090 copyName ( sym->Name, strtab, symbol, 1000 );
1091 zapTrailingAtSign ( symbol );
1092 S = (UInt32) ocLookupSym ( oc, symbol );
1094 S = (UInt32)(oc->clientLookup ( symbol ));
1096 belch("%s: unresolvable reference to `%s'", oc->fileName, symbol);
1101 switch (reltab_j->Type) {
1102 case IMAGE_REL_I386_DIR32:
1105 case IMAGE_REL_I386_REL32:
1106 /* Tricky. We have to insert a displacement at
1107 pP which, when added to the PC for the _next_
1108 insn, gives the address of the target (S).
1109 Problem is to know the address of the next insn
1110 when we only know pP. We assume that this
1111 literal field is always the last in the insn,
1112 so that the address of the next insn is pP+4
1113 -- hence the constant 4.
1114 Also I don't know if A should be added, but so
1115 far it has always been zero.
1118 *pP = S - ((UInt32)pP) - 4;
1122 "unhandled PEi386 relocation type %d\n",
1124 oc->errMsg("unhandled PEi386 relocation type");
1134 #endif /* defined(OBJFORMAT_PEi386) */
1137 /* --------------------------------------------------------------------------
1139 * ------------------------------------------------------------------------*/
1141 #if defined(OBJFORMAT_ELF)
1149 findElfSection ( void* objImage, Elf32_Word sh_type )
1152 char* ehdrC = (char*)objImage;
1153 Elf32_Ehdr* ehdr = ( Elf32_Ehdr*)ehdrC;
1154 Elf32_Shdr* shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1156 for (i = 0; i < ehdr->e_shnum; i++) {
1157 if (shdr[i].sh_type == sh_type &&
1158 i != ehdr->e_shstrndx) {
1159 ptr = ehdrC + shdr[i].sh_offset;
1168 ocVerifyImage_ELF ( ObjectCode* oc )
1172 int i, j, nent, nstrtab, nsymtabs;
1176 char* ehdrC = (char*)(oc->image);
1177 Elf32_Ehdr* ehdr = ( Elf32_Ehdr*)ehdrC;
1179 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1180 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1181 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1182 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
1183 belch("ocVerifyImage_ELF: not an ELF header");
1186 IF_DEBUG(linker,belch( "Is an ELF header" ));
1188 if (ehdr->e_ident[EI_CLASS] != ELFCLASS32) {
1189 belch("ocVerifyImage_ELF: not 32 bit ELF" );
1193 IF_DEBUG(linker,belch( "Is 32 bit ELF" ));
1195 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
1196 IF_DEBUG(linker,belch( "Is little-endian" ));
1198 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
1199 IF_DEBUG(linker,belch( "Is big-endian" ));
1201 belch("ocVerifyImage_ELF: unknown endiannness");
1205 if (ehdr->e_type != ET_REL) {
1206 belch("ocVerifyImage_ELF: not a relocatable object (.o) file");
1209 IF_DEBUG(linker, belch( "Is a relocatable object (.o) file" ));
1211 IF_DEBUG(linker,belch( "Architecture is " ));
1212 switch (ehdr->e_machine) {
1213 case EM_386: IF_DEBUG(linker,belch( "x86" )); break;
1214 case EM_SPARC: IF_DEBUG(linker,belch( "sparc" )); break;
1215 default: IF_DEBUG(linker,belch( "unknown" ));
1216 belch("ocVerifyImage_ELF: unknown architecture");
1220 IF_DEBUG(linker,belch(
1221 "\nSection header table: start %d, n_entries %d, ent_size %d",
1222 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
1224 ASSERT (ehdr->e_shentsize == sizeof(Elf32_Shdr));
1226 shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1228 if (ehdr->e_shstrndx == SHN_UNDEF) {
1229 belch("ocVerifyImage_ELF: no section header string table");
1232 IF_DEBUG(linker,belch( "Section header string table is section %d",
1234 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
1237 for (i = 0; i < ehdr->e_shnum; i++) {
1238 IF_DEBUG(linker,fprintf(stderr, "%2d: ", i ));
1239 IF_DEBUG(linker,fprintf(stderr, "type=%2d ", shdr[i].sh_type ));
1240 IF_DEBUG(linker,fprintf(stderr, "size=%4d ", shdr[i].sh_size ));
1241 IF_DEBUG(linker,fprintf(stderr, "offs=%4d ", shdr[i].sh_offset ));
1242 IF_DEBUG(linker,fprintf(stderr, " (%p .. %p) ",
1243 ehdrC + shdr[i].sh_offset,
1244 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
1246 if (shdr[i].sh_type == SHT_REL) {
1247 IF_DEBUG(linker,fprintf(stderr, "Rel " ));
1248 } else if (shdr[i].sh_type == SHT_RELA) {
1249 IF_DEBUG(linker,fprintf(stderr, "RelA " ));
1251 IF_DEBUG(linker,fprintf(stderr," "));
1254 IF_DEBUG(linker,fprintf(stderr, "sname=%s\n", sh_strtab + shdr[i].sh_name ));
1258 IF_DEBUG(linker,belch( "\nString tables" ));
1261 for (i = 0; i < ehdr->e_shnum; i++) {
1262 if (shdr[i].sh_type == SHT_STRTAB &&
1263 i != ehdr->e_shstrndx) {
1264 IF_DEBUG(linker,belch(" section %d is a normal string table", i ));
1265 strtab = ehdrC + shdr[i].sh_offset;
1270 belch("ocVerifyImage_ELF: no string tables, or too many");
1275 IF_DEBUG(linker,belch( "\nSymbol tables" ));
1276 for (i = 0; i < ehdr->e_shnum; i++) {
1277 if (shdr[i].sh_type != SHT_SYMTAB) continue;
1278 IF_DEBUG(linker,belch( "section %d is a symbol table", i ));
1280 stab = (Elf32_Sym*) (ehdrC + shdr[i].sh_offset);
1281 nent = shdr[i].sh_size / sizeof(Elf32_Sym);
1282 IF_DEBUG(linker,belch( " number of entries is apparently %d (%d rem)",
1284 shdr[i].sh_size % sizeof(Elf32_Sym)
1286 if (0 != shdr[i].sh_size % sizeof(Elf32_Sym)) {
1287 belch("ocVerifyImage_ELF: non-integral number of symbol table entries");
1290 for (j = 0; j < nent; j++) {
1291 IF_DEBUG(linker,fprintf(stderr, " %2d ", j ));
1292 IF_DEBUG(linker,fprintf(stderr, " sec=%-5d size=%-3d val=%5p ",
1293 (int)stab[j].st_shndx,
1294 (int)stab[j].st_size,
1295 (char*)stab[j].st_value ));
1297 IF_DEBUG(linker,fprintf(stderr, "type=" ));
1298 switch (ELF32_ST_TYPE(stab[j].st_info)) {
1299 case STT_NOTYPE: IF_DEBUG(linker,fprintf(stderr, "notype " )); break;
1300 case STT_OBJECT: IF_DEBUG(linker,fprintf(stderr, "object " )); break;
1301 case STT_FUNC : IF_DEBUG(linker,fprintf(stderr, "func " )); break;
1302 case STT_SECTION: IF_DEBUG(linker,fprintf(stderr, "section" )); break;
1303 case STT_FILE: IF_DEBUG(linker,fprintf(stderr, "file " )); break;
1304 default: IF_DEBUG(linker,fprintf(stderr, "? " )); break;
1306 IF_DEBUG(linker,fprintf(stderr, " " ));
1308 IF_DEBUG(linker,fprintf(stderr, "bind=" ));
1309 switch (ELF32_ST_BIND(stab[j].st_info)) {
1310 case STB_LOCAL : IF_DEBUG(linker,fprintf(stderr, "local " )); break;
1311 case STB_GLOBAL: IF_DEBUG(linker,fprintf(stderr, "global" )); break;
1312 case STB_WEAK : IF_DEBUG(linker,fprintf(stderr, "weak " )); break;
1313 default: IF_DEBUG(linker,fprintf(stderr, "? " )); break;
1315 IF_DEBUG(linker,fprintf(stderr, " " ));
1317 IF_DEBUG(linker,fprintf(stderr, "name=%s\n", strtab + stab[j].st_name ));
1321 if (nsymtabs == 0) {
1322 belch("ocVerifyImage_ELF: didn't find any symbol tables");
1331 ocGetNames_ELF ( ObjectCode* oc )
1336 char* ehdrC = (char*)(oc->image);
1337 Elf32_Ehdr* ehdr = (Elf32_Ehdr*)ehdrC;
1338 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
1339 Elf32_Shdr* shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1340 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
1343 belch("ocGetNames_ELF: no strtab");
1348 oc->sections = stgMallocBytes( ehdr->e_shnum * sizeof(Section),
1350 oc->n_sections = ehdr->e_shnum;
1352 for (i = 0; i < ehdr->e_shnum; i++) {
1354 /* make a section entry for relevant sections */
1355 SectionKind kind = SECTIONKIND_OTHER;
1356 if (!strcmp(".data",sh_strtab+shdr[i].sh_name) ||
1357 !strcmp(".data1",sh_strtab+shdr[i].sh_name))
1358 kind = SECTIONKIND_RWDATA;
1359 if (!strcmp(".text",sh_strtab+shdr[i].sh_name) ||
1360 !strcmp(".rodata",sh_strtab+shdr[i].sh_name) ||
1361 !strcmp(".rodata1",sh_strtab+shdr[i].sh_name))
1362 kind = SECTIONKIND_CODE_OR_RODATA;
1364 /* fill in the section info */
1365 oc->sections[i].start = ehdrC + shdr[i].sh_offset;
1366 oc->sections[i].end = ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1;
1367 oc->sections[i].kind = kind;
1369 if (shdr[i].sh_type != SHT_SYMTAB) continue;
1371 /* copy stuff into this module's object symbol table */
1372 stab = (Elf32_Sym*) (ehdrC + shdr[i].sh_offset);
1373 nent = shdr[i].sh_size / sizeof(Elf32_Sym);
1374 oc->symbols = malloc(nent * sizeof(SymbolVal));
1375 oc->n_symbols = nent;
1376 for (j = 0; j < nent; j++) {
1377 if ( ( ELF32_ST_BIND(stab[j].st_info)==STB_GLOBAL /* ||
1378 ELF32_ST_BIND(stab[j].st_info)==STB_LOCAL */
1380 /* and not an undefined symbol */
1381 && stab[j].st_shndx != SHN_UNDEF
1383 /* and it's a not a section or string table or anything silly */
1384 ( ELF32_ST_TYPE(stab[j].st_info)==STT_FUNC ||
1385 ELF32_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
1386 ELF32_ST_TYPE(stab[j].st_info)==STT_NOTYPE )
1388 char* nm = strtab + stab[j].st_name;
1390 + shdr[ stab[j].st_shndx ].sh_offset
1394 IF_DEBUG(linker,belch( "addOTabName: %10p %s %s",
1395 ad, oc->fileName, nm ));
1396 oc->symbols[j].lbl = nm;
1397 oc->symbols[j].addr = ad;
1398 insertStrHashTable(symhash, nm, &(oc->symbols[j]));
1401 IF_DEBUG(linker,belch( "skipping `%s'", strtab +
1403 oc->symbols[j].lbl = NULL;
1404 oc->symbols[j].addr = NULL;
1414 ocResolve_ELF ( ObjectCode* oc )
1416 char *strtab, *symbol;
1418 Elf32_Sym* stab = NULL;
1419 char* ehdrC = (char*)(oc->image);
1420 Elf32_Ehdr* ehdr = (Elf32_Ehdr*) ehdrC;
1421 Elf32_Shdr* shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1424 /* first find "the" symbol table */
1425 stab = (Elf32_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
1427 /* also go find the string table */
1428 strtab = findElfSection ( ehdrC, SHT_STRTAB );
1430 if (stab == NULL || strtab == NULL) {
1431 belch("ocResolve_ELF: can't find string or symbol table");
1435 for (i = 0; i < ehdr->e_shnum; i++) {
1436 if (shdr[i].sh_type == SHT_REL ) {
1437 Elf32_Rel* rtab = (Elf32_Rel*) (ehdrC + shdr[i].sh_offset);
1438 int nent = shdr[i].sh_size / sizeof(Elf32_Rel);
1439 int target_shndx = shdr[i].sh_info;
1440 int symtab_shndx = shdr[i].sh_link;
1441 stab = (Elf32_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
1442 targ = (Elf32_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
1443 IF_DEBUG(linker,belch( "relocations for section %d using symtab %d",
1444 target_shndx, symtab_shndx ));
1445 for (j = 0; j < nent; j++) {
1446 Elf32_Addr offset = rtab[j].r_offset;
1447 Elf32_Word info = rtab[j].r_info;
1449 Elf32_Addr P = ((Elf32_Addr)targ) + offset;
1450 Elf32_Word* pP = (Elf32_Word*)P;
1454 IF_DEBUG(linker,belch( "Rel entry %3d is raw(%6p %6p) ",
1455 j, (void*)offset, (void*)info ));
1457 IF_DEBUG(linker,belch( " ZERO" ));
1460 /* First see if it is a nameless local symbol. */
1461 if (stab[ ELF32_R_SYM(info)].st_name == 0) {
1462 symbol = "(noname)";
1463 S = (Elf32_Addr)(ehdrC
1464 + shdr[stab[ELF32_R_SYM(info)].st_shndx ].sh_offset
1465 + stab[ELF32_R_SYM(info)].st_value
1468 /* No? Should be in the symbol table then. */
1469 symbol = strtab+stab[ ELF32_R_SYM(info)].st_name;
1470 (void *)S = lookupSymbol( symbol );
1473 barf("ocResolve_ELF: %s: unknown symbol `%s'",
1474 oc->fileName, symbol);
1476 IF_DEBUG(linker,belch( "`%s' resolves to %p", symbol, (void*)S ));
1478 IF_DEBUG(linker,fprintf ( stderr, "Reloc: P = %p S = %p A = %p\n", (void*)P, (void*)S, (void*)A ));
1479 switch (ELF32_R_TYPE(info)) {
1480 case R_386_32: *pP = S + A; break;
1481 case R_386_PC32: *pP = S + A - P; break;
1482 default: fprintf(stderr,
1483 "unhandled ELF relocation type %d",
1484 ELF32_R_TYPE(info));
1485 belch("ocResolve_ELF: unhandled ELF relocation type");
1492 if (shdr[i].sh_type == SHT_RELA) {
1493 belch("ocResolve_ELF: RelA style reloc table -- not yet done");
1504 /* -----------------------------------------------------------------------------
1505 * Look up an address to discover whether it is in text or data space.
1507 * Used by the garbage collector when walking the stack.
1508 * -------------------------------------------------------------------------- */
1510 static __inline__ SectionKind
1511 lookupSection ( void* addr )
1516 for ( oc = objects; oc; oc = oc->next ) {
1517 for (i = 0; i < oc->n_sections; i++) {
1518 if (oc->sections[i].start <= addr
1519 && addr <= oc->sections[i].end)
1520 return oc->sections[i].kind;
1523 return SECTIONKIND_OTHER;
1527 is_dynamically_loaded_code_or_rodata_ptr ( void* p )
1529 SectionKind sk = lookupSection(p);
1530 ASSERT (sk != SECTIONKIND_NOINFOAVAIL);
1531 return (sk == SECTIONKIND_CODE_OR_RODATA);
1536 is_dynamically_loaded_rwdata_ptr ( void* p )
1538 SectionKind sk = lookupSection(p);
1539 ASSERT (sk != SECTIONKIND_NOINFOAVAIL);
1540 return (sk == SECTIONKIND_RWDATA);
1545 is_not_dynamically_loaded_ptr ( void* p )
1547 SectionKind sk = lookupSection(p);
1548 ASSERT (sk != SECTIONKIND_NOINFOAVAIL);
1549 return (sk == SECTIONKIND_OTHER);