1 /* -----------------------------------------------------------------------------
2 * $Id: Linker.c,v 1.78 2002/01/23 11:29:12 sewardj Exp $
4 * (c) The GHC Team, 2000, 2001
8 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
16 #include "LinkerInternals.h"
18 #include "StoragePriv.h"
21 #ifdef HAVE_SYS_TYPES_H
22 #include <sys/types.h>
25 #ifdef HAVE_SYS_STAT_H
33 #if defined(linux_TARGET_OS) || defined(solaris2_TARGET_OS) || defined(freebsd_TARGET_OS)
34 # define OBJFORMAT_ELF
35 #elif defined(cygwin32_TARGET_OS) || defined (mingw32_TARGET_OS)
36 # define OBJFORMAT_PEi386
40 /* Hash table mapping symbol names to Symbol */
41 /*Str*/HashTable *symhash;
43 #if defined(OBJFORMAT_ELF)
44 static int ocVerifyImage_ELF ( ObjectCode* oc );
45 static int ocGetNames_ELF ( ObjectCode* oc );
46 static int ocResolve_ELF ( ObjectCode* oc );
47 #elif defined(OBJFORMAT_PEi386)
48 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
49 static int ocGetNames_PEi386 ( ObjectCode* oc );
50 static int ocResolve_PEi386 ( ObjectCode* oc );
53 /* -----------------------------------------------------------------------------
54 * Built-in symbols from the RTS
57 typedef struct _RtsSymbolVal {
64 #define Maybe_ForeignObj SymX(mkForeignObjzh_fast)
66 #define Maybe_Stable_Names SymX(mkWeakzh_fast) \
67 SymX(makeStableNamezh_fast) \
68 SymX(finalizzeWeakzh_fast)
70 /* These are not available in GUM!!! -- HWL */
71 #define Maybe_ForeignObj
72 #define Maybe_Stable_Names
75 #if !defined (mingw32_TARGET_OS)
77 #define RTS_POSIX_ONLY_SYMBOLS \
78 SymX(stg_sig_install) \
80 #define RTS_MINGW_ONLY_SYMBOLS /**/
84 #define RTS_POSIX_ONLY_SYMBOLS
86 /* These are statically linked from the mingw libraries into the ghc
87 executable, so we have to employ this hack. */
88 #define RTS_MINGW_ONLY_SYMBOLS \
100 SymX(getservbyname) \
101 SymX(getservbyport) \
102 SymX(getprotobynumber) \
103 SymX(getprotobyname) \
104 SymX(gethostbyname) \
105 SymX(gethostbyaddr) \
140 Sym(_imp___timezone) \
155 # define MAIN_CAP_SYM SymX(MainCapability)
157 # define MAIN_CAP_SYM
160 #define RTS_SYMBOLS \
164 Sym(__stginit_PrelGHC) \
168 Sym(stg_enterStackTop) \
171 SymX(__stg_gc_enter_1) \
173 SymX(stg_gc_noregs) \
175 SymX(stg_gc_unbx_r1) \
176 SymX(stg_gc_unpt_r1) \
177 SymX(stg_gc_ut_0_1) \
178 SymX(stg_gc_ut_1_0) \
180 SymX(stg_yield_to_interpreter) \
183 SymX(MallocFailHook) \
184 SymX(NoRunnableThreadsHook) \
186 SymX(OutOfHeapHook) \
187 SymX(PatErrorHdrHook) \
188 SymX(PostTraceHook) \
190 SymX(StackOverflowHook) \
191 SymX(__encodeDouble) \
192 SymX(__encodeFloat) \
195 SymX(__gmpz_cmp_si) \
196 SymX(__gmpz_cmp_ui) \
197 SymX(__gmpz_get_si) \
198 SymX(__gmpz_get_ui) \
199 SymX(__int_encodeDouble) \
200 SymX(__int_encodeFloat) \
201 SymX(andIntegerzh_fast) \
202 SymX(blockAsyncExceptionszh_fast) \
205 SymX(complementIntegerzh_fast) \
206 SymX(cmpIntegerzh_fast) \
207 SymX(cmpIntegerIntzh_fast) \
208 SymX(createAdjustor) \
209 SymX(decodeDoublezh_fast) \
210 SymX(decodeFloatzh_fast) \
213 SymX(deRefWeakzh_fast) \
214 SymX(deRefStablePtrzh_fast) \
215 SymX(divExactIntegerzh_fast) \
216 SymX(divModIntegerzh_fast) \
218 SymX(freeHaskellFunctionPtr) \
219 SymX(freeStablePtr) \
220 SymX(gcdIntegerzh_fast) \
221 SymX(gcdIntegerIntzh_fast) \
222 SymX(gcdIntzh_fast) \
225 SymX(int2Integerzh_fast) \
226 SymX(integer2Intzh_fast) \
227 SymX(integer2Wordzh_fast) \
228 SymX(isDoubleDenormalized) \
229 SymX(isDoubleInfinite) \
231 SymX(isDoubleNegativeZero) \
232 SymX(isEmptyMVarzh_fast) \
233 SymX(isFloatDenormalized) \
234 SymX(isFloatInfinite) \
236 SymX(isFloatNegativeZero) \
237 SymX(killThreadzh_fast) \
238 SymX(makeStablePtrzh_fast) \
239 SymX(minusIntegerzh_fast) \
240 SymX(mkApUpd0zh_fast) \
241 SymX(myThreadIdzh_fast) \
242 SymX(newArrayzh_fast) \
243 SymX(newBCOzh_fast) \
244 SymX(newByteArrayzh_fast) \
246 SymX(newMVarzh_fast) \
247 SymX(newMutVarzh_fast) \
248 SymX(newPinnedByteArrayzh_fast) \
249 SymX(orIntegerzh_fast) \
251 SymX(plusIntegerzh_fast) \
254 SymX(putMVarzh_fast) \
255 SymX(quotIntegerzh_fast) \
256 SymX(quotRemIntegerzh_fast) \
258 SymX(remIntegerzh_fast) \
259 SymX(resetNonBlockingFd) \
262 SymX(rts_checkSchedStatus) \
265 SymX(rts_evalLazyIO) \
270 SymX(rts_getDouble) \
275 SymX(rts_getStablePtr) \
276 SymX(rts_getThreadId) \
278 SymX(rts_getWord32) \
290 SymX(rts_mkStablePtr) \
299 SymX(shutdownHaskellAndExit) \
300 SymX(stable_ptr_table) \
301 SymX(stackOverflow) \
302 SymX(stg_CAF_BLACKHOLE_info) \
303 SymX(stg_CHARLIKE_closure) \
304 SymX(stg_EMPTY_MVAR_info) \
305 SymX(stg_IND_STATIC_info) \
306 SymX(stg_INTLIKE_closure) \
307 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
308 SymX(stg_WEAK_info) \
309 SymX(stg_ap_1_upd_info) \
310 SymX(stg_ap_2_upd_info) \
311 SymX(stg_ap_3_upd_info) \
312 SymX(stg_ap_4_upd_info) \
313 SymX(stg_ap_5_upd_info) \
314 SymX(stg_ap_6_upd_info) \
315 SymX(stg_ap_7_upd_info) \
316 SymX(stg_ap_8_upd_info) \
318 SymX(stg_sel_0_upd_info) \
319 SymX(stg_sel_10_upd_info) \
320 SymX(stg_sel_11_upd_info) \
321 SymX(stg_sel_12_upd_info) \
322 SymX(stg_sel_13_upd_info) \
323 SymX(stg_sel_14_upd_info) \
324 SymX(stg_sel_15_upd_info) \
325 SymX(stg_sel_1_upd_info) \
326 SymX(stg_sel_2_upd_info) \
327 SymX(stg_sel_3_upd_info) \
328 SymX(stg_sel_4_upd_info) \
329 SymX(stg_sel_5_upd_info) \
330 SymX(stg_sel_6_upd_info) \
331 SymX(stg_sel_7_upd_info) \
332 SymX(stg_sel_8_upd_info) \
333 SymX(stg_sel_9_upd_info) \
334 SymX(stg_seq_frame_info) \
335 SymX(stg_upd_frame_info) \
336 SymX(__stg_update_PAP) \
337 SymX(suspendThread) \
338 SymX(takeMVarzh_fast) \
339 SymX(timesIntegerzh_fast) \
340 SymX(tryPutMVarzh_fast) \
341 SymX(tryTakeMVarzh_fast) \
342 SymX(unblockAsyncExceptionszh_fast) \
343 SymX(unsafeThawArrayzh_fast) \
344 SymX(waitReadzh_fast) \
345 SymX(waitWritezh_fast) \
346 SymX(word2Integerzh_fast) \
347 SymX(xorIntegerzh_fast) \
350 #ifndef SUPPORT_LONG_LONGS
351 #define RTS_LONG_LONG_SYMS /* nothing */
353 #define RTS_LONG_LONG_SYMS \
354 SymX(int64ToIntegerzh_fast) \
355 SymX(word64ToIntegerzh_fast)
356 #endif /* SUPPORT_LONG_LONGS */
358 /* entirely bogus claims about types of these symbols */
359 #define Sym(vvv) extern void (vvv);
360 #define SymX(vvv) /**/
363 RTS_POSIX_ONLY_SYMBOLS
364 RTS_MINGW_ONLY_SYMBOLS
368 #ifdef LEADING_UNDERSCORE
369 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
371 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
374 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
376 #define SymX(vvv) Sym(vvv)
378 static RtsSymbolVal rtsSyms[] = {
381 RTS_POSIX_ONLY_SYMBOLS
382 RTS_MINGW_ONLY_SYMBOLS
383 { 0, 0 } /* sentinel */
386 /* -----------------------------------------------------------------------------
387 * Insert symbols into hash tables, checking for duplicates.
389 static void ghciInsertStrHashTable ( char* obj_name,
395 if (lookupHashTable(table, (StgWord)key) == NULL)
397 insertStrHashTable(table, (StgWord)key, data);
402 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
404 "whilst processing object file\n"
406 "This could be caused by:\n"
407 " * Loading two different object files which export the same symbol\n"
408 " * Specifying the same object file twice on the GHCi command line\n"
409 " * An incorrect `package.conf' entry, causing some object to be\n"
411 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
420 /* -----------------------------------------------------------------------------
421 * initialize the object linker
423 #if defined(OBJFORMAT_ELF)
424 static void *dl_prog_handle;
432 symhash = allocStrHashTable();
434 /* populate the symbol table with stuff from the RTS */
435 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
436 ghciInsertStrHashTable("(GHCi built-in symbols)",
437 symhash, sym->lbl, sym->addr);
439 # if defined(OBJFORMAT_ELF)
440 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
444 /* -----------------------------------------------------------------------------
445 * Add a DLL from which symbols may be found. In the ELF case, just
446 * do RTLD_GLOBAL-style add, so no further messing around needs to
447 * happen in order that symbols in the loaded .so are findable --
448 * lookupSymbol() will subsequently see them by dlsym on the program's
449 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
451 * In the PEi386 case, open the DLLs and put handles to them in a
452 * linked list. When looking for a symbol, try all handles in the
456 #if defined(OBJFORMAT_PEi386)
457 /* A record for storing handles into DLLs. */
462 struct _OpenedDLL* next;
467 /* A list thereof. */
468 static OpenedDLL* opened_dlls = NULL;
474 addDLL ( __attribute((unused)) char* path, char* dll_name )
476 # if defined(OBJFORMAT_ELF)
481 if (path == NULL || strlen(path) == 0) {
482 buf = stgMallocBytes(strlen(dll_name) + 10, "addDll");
483 sprintf(buf, "lib%s.so", dll_name);
485 buf = stgMallocBytes(strlen(path) + 1 + strlen(dll_name) + 10, "addDll");
486 sprintf(buf, "%s/lib%s.so", path, dll_name);
488 hdl = dlopen(buf, RTLD_NOW | RTLD_GLOBAL );
491 /* dlopen failed; return a ptr to the error msg. */
493 if (errmsg == NULL) errmsg = "addDLL: unknown error";
500 # elif defined(OBJFORMAT_PEi386)
502 /* Add this DLL to the list of DLLs in which to search for symbols.
503 The path argument is ignored. */
508 /* fprintf(stderr, "\naddDLL; path=`%s', dll_name = `%s'\n", path, dll_name); */
510 /* See if we've already got it, and ignore if so. */
511 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
512 if (0 == strcmp(o_dll->name, dll_name))
516 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
517 sprintf(buf, "%s.DLL", dll_name);
518 instance = LoadLibrary(buf);
520 if (instance == NULL) {
521 /* LoadLibrary failed; return a ptr to the error msg. */
522 return "addDLL: unknown error";
525 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
526 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
527 strcpy(o_dll->name, dll_name);
528 o_dll->instance = instance;
529 o_dll->next = opened_dlls;
534 barf("addDLL: not implemented on this platform");
538 /* -----------------------------------------------------------------------------
539 * lookup a symbol in the hash table
542 lookupSymbol( char *lbl )
545 ASSERT(symhash != NULL);
546 val = lookupStrHashTable(symhash, lbl);
549 # if defined(OBJFORMAT_ELF)
550 return dlsym(dl_prog_handle, lbl);
551 # elif defined(OBJFORMAT_PEi386)
554 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
555 /* fprintf(stderr, "look in %s for %s\n", o_dll->name, lbl); */
557 /* HACK: if the name has an initial underscore, try stripping
558 it off & look that up first. I've yet to verify whether there's
559 a Rule that governs whether an initial '_' *should always* be
560 stripped off when mapping from import lib name to the DLL name.
562 sym = GetProcAddress(o_dll->instance, (lbl+1));
564 /*fprintf(stderr, "found %s in %s\n", lbl+1,o_dll->name); fflush(stderr);*/
568 sym = GetProcAddress(o_dll->instance, lbl);
570 /*fprintf(stderr, "found %s in %s\n", lbl,o_dll->name); fflush(stderr);*/
585 __attribute((unused))
587 lookupLocalSymbol( ObjectCode* oc, char *lbl )
590 val = lookupStrHashTable(oc->lochash, lbl);
600 /* -----------------------------------------------------------------------------
601 * Debugging aid: look in GHCi's object symbol tables for symbols
602 * within DELTA bytes of the specified address, and show their names.
605 void ghci_enquire ( char* addr );
607 void ghci_enquire ( char* addr )
612 const int DELTA = 64;
614 for (oc = objects; oc; oc = oc->next) {
615 for (i = 0; i < oc->n_symbols; i++) {
616 sym = oc->symbols[i];
617 if (sym == NULL) continue;
618 /* fprintf(stderr, "enquire %p %p\n", sym, oc->lochash); */
620 if (oc->lochash != NULL)
621 a = lookupStrHashTable(oc->lochash, sym);
623 a = lookupStrHashTable(symhash, sym);
625 /* fprintf(stderr, "ghci_enquire: can't find %s\n", sym); */
627 else if (addr-DELTA <= a && a <= addr+DELTA) {
628 fprintf(stderr, "%p + %3d == `%s'\n", addr, a - addr, sym);
636 /* -----------------------------------------------------------------------------
637 * Load an obj (populate the global symbol table, but don't resolve yet)
639 * Returns: 1 if ok, 0 on error.
642 loadObj( char *path )
649 /* fprintf(stderr, "loadObj %s\n", path ); */
651 /* Check that we haven't already loaded this object. Don't give up
652 at this stage; ocGetNames_* will barf later. */
656 for (o = objects; o; o = o->next) {
657 if (0 == strcmp(o->fileName, path))
663 "GHCi runtime linker: warning: looks like you're trying to load the\n"
664 "same object file twice:\n"
666 "GHCi will continue, but a duplicate-symbol error may shortly follow.\n"
672 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
674 # if defined(OBJFORMAT_ELF)
675 oc->formatName = "ELF";
676 # elif defined(OBJFORMAT_PEi386)
677 oc->formatName = "PEi386";
680 barf("loadObj: not implemented on this platform");
684 if (r == -1) { return 0; }
686 /* sigh, strdup() isn't a POSIX function, so do it the long way */
687 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
688 strcpy(oc->fileName, path);
690 oc->fileSize = st.st_size;
691 oc->image = stgMallocBytes( st.st_size, "loadObj(image)" );
694 oc->lochash = allocStrHashTable();
695 oc->proddables = NULL;
697 /* chain it onto the list of objects */
701 /* load the image into memory */
702 f = fopen(path, "rb");
704 barf("loadObj: can't read `%s'", path);
706 n = fread ( oc->image, 1, oc->fileSize, f );
707 if (n != oc->fileSize) {
709 barf("loadObj: error whilst reading `%s'", path);
712 /* verify the in-memory image */
713 # if defined(OBJFORMAT_ELF)
714 r = ocVerifyImage_ELF ( oc );
715 # elif defined(OBJFORMAT_PEi386)
716 r = ocVerifyImage_PEi386 ( oc );
718 barf("loadObj: no verify method");
720 if (!r) { return r; }
722 /* build the symbol list for this image */
723 # if defined(OBJFORMAT_ELF)
724 r = ocGetNames_ELF ( oc );
725 # elif defined(OBJFORMAT_PEi386)
726 r = ocGetNames_PEi386 ( oc );
728 barf("loadObj: no getNames method");
730 if (!r) { return r; }
732 /* loaded, but not resolved yet */
733 oc->status = OBJECT_LOADED;
738 /* -----------------------------------------------------------------------------
739 * resolve all the currently unlinked objects in memory
741 * Returns: 1 if ok, 0 on error.
749 for (oc = objects; oc; oc = oc->next) {
750 if (oc->status != OBJECT_RESOLVED) {
751 # if defined(OBJFORMAT_ELF)
752 r = ocResolve_ELF ( oc );
753 # elif defined(OBJFORMAT_PEi386)
754 r = ocResolve_PEi386 ( oc );
756 barf("resolveObjs: not implemented on this platform");
758 if (!r) { return r; }
759 oc->status = OBJECT_RESOLVED;
765 /* -----------------------------------------------------------------------------
766 * delete an object from the pool
769 unloadObj( char *path )
771 ObjectCode *oc, *prev;
773 ASSERT(symhash != NULL);
774 ASSERT(objects != NULL);
777 for (oc = objects; oc; prev = oc, oc = oc->next) {
778 if (!strcmp(oc->fileName,path)) {
780 /* Remove all the mappings for the symbols within this
785 for (i = 0; i < oc->n_symbols; i++) {
786 if (oc->symbols[i] != NULL) {
787 removeStrHashTable(symhash, oc->symbols[i], NULL);
795 prev->next = oc->next;
798 /* We're going to leave this in place, in case there are
799 any pointers from the heap into it: */
800 /* free(oc->image); */
804 /* The local hash table should have been freed at the end
805 of the ocResolve_ call on it. */
806 ASSERT(oc->lochash == NULL);
812 belch("unloadObj: can't find `%s' to unload", path);
816 /* -----------------------------------------------------------------------------
817 * Sanity checking. For each ObjectCode, maintain a list of address ranges
818 * which may be prodded during relocation, and abort if we try and write
819 * outside any of these.
821 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
824 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
825 /* fprintf(stderr, "aPB %p %p %d\n", oc, start, size); */
829 pb->next = oc->proddables;
833 static void checkProddableBlock ( ObjectCode* oc, void* addr )
836 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
837 char* s = (char*)(pb->start);
838 char* e = s + pb->size - 1;
839 char* a = (char*)addr;
840 /* Assumes that the biggest fixup involves a 4-byte write. This
841 probably needs to be changed to 8 (ie, +7) on 64-bit
843 if (a >= s && (a+3) <= e) return;
845 barf("checkProddableBlock: invalid fixup in runtime linker");
848 /* -----------------------------------------------------------------------------
849 * Section management.
851 static void addSection ( ObjectCode* oc, SectionKind kind,
852 void* start, void* end )
854 Section* s = stgMallocBytes(sizeof(Section), "addSection");
858 s->next = oc->sections;
864 /* --------------------------------------------------------------------------
865 * PEi386 specifics (Win32 targets)
866 * ------------------------------------------------------------------------*/
868 /* The information for this linker comes from
869 Microsoft Portable Executable
870 and Common Object File Format Specification
871 revision 5.1 January 1998
872 which SimonM says comes from the MS Developer Network CDs.
876 #if defined(OBJFORMAT_PEi386)
880 typedef unsigned char UChar;
881 typedef unsigned short UInt16;
882 typedef unsigned int UInt32;
889 UInt16 NumberOfSections;
890 UInt32 TimeDateStamp;
891 UInt32 PointerToSymbolTable;
892 UInt32 NumberOfSymbols;
893 UInt16 SizeOfOptionalHeader;
894 UInt16 Characteristics;
898 #define sizeof_COFF_header 20
905 UInt32 VirtualAddress;
906 UInt32 SizeOfRawData;
907 UInt32 PointerToRawData;
908 UInt32 PointerToRelocations;
909 UInt32 PointerToLinenumbers;
910 UInt16 NumberOfRelocations;
911 UInt16 NumberOfLineNumbers;
912 UInt32 Characteristics;
916 #define sizeof_COFF_section 40
923 UInt16 SectionNumber;
926 UChar NumberOfAuxSymbols;
930 #define sizeof_COFF_symbol 18
935 UInt32 VirtualAddress;
936 UInt32 SymbolTableIndex;
941 #define sizeof_COFF_reloc 10
944 /* From PE spec doc, section 3.3.2 */
945 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
946 windows.h -- for the same purpose, but I want to know what I'm
948 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
949 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
950 #define MYIMAGE_FILE_DLL 0x2000
951 #define MYIMAGE_FILE_SYSTEM 0x1000
952 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
953 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
954 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
956 /* From PE spec doc, section 5.4.2 and 5.4.4 */
957 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
958 #define MYIMAGE_SYM_CLASS_STATIC 3
959 #define MYIMAGE_SYM_UNDEFINED 0
961 /* From PE spec doc, section 4.1 */
962 #define MYIMAGE_SCN_CNT_CODE 0x00000020
963 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
965 /* From PE spec doc, section 5.2.1 */
966 #define MYIMAGE_REL_I386_DIR32 0x0006
967 #define MYIMAGE_REL_I386_REL32 0x0014
970 /* We use myindex to calculate array addresses, rather than
971 simply doing the normal subscript thing. That's because
972 some of the above structs have sizes which are not
973 a whole number of words. GCC rounds their sizes up to a
974 whole number of words, which means that the address calcs
975 arising from using normal C indexing or pointer arithmetic
976 are just plain wrong. Sigh.
979 myindex ( int scale, void* base, int index )
982 ((UChar*)base) + scale * index;
987 printName ( UChar* name, UChar* strtab )
989 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
990 UInt32 strtab_offset = * (UInt32*)(name+4);
991 fprintf ( stderr, "%s", strtab + strtab_offset );
994 for (i = 0; i < 8; i++) {
995 if (name[i] == 0) break;
996 fprintf ( stderr, "%c", name[i] );
1003 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1005 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1006 UInt32 strtab_offset = * (UInt32*)(name+4);
1007 strncpy ( dst, strtab+strtab_offset, dstSize );
1013 if (name[i] == 0) break;
1023 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1026 /* If the string is longer than 8 bytes, look in the
1027 string table for it -- this will be correctly zero terminated.
1029 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1030 UInt32 strtab_offset = * (UInt32*)(name+4);
1031 return ((UChar*)strtab) + strtab_offset;
1033 /* Otherwise, if shorter than 8 bytes, return the original,
1034 which by defn is correctly terminated.
1036 if (name[7]==0) return name;
1037 /* The annoying case: 8 bytes. Copy into a temporary
1038 (which is never freed ...)
1040 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1042 strncpy(newstr,name,8);
1048 /* Just compares the short names (first 8 chars) */
1049 static COFF_section *
1050 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1054 = (COFF_header*)(oc->image);
1055 COFF_section* sectab
1057 ((UChar*)(oc->image))
1058 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1060 for (i = 0; i < hdr->NumberOfSections; i++) {
1063 COFF_section* section_i
1065 myindex ( sizeof_COFF_section, sectab, i );
1066 n1 = (UChar*) &(section_i->Name);
1068 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1069 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1070 n1[6]==n2[6] && n1[7]==n2[7])
1079 zapTrailingAtSign ( UChar* sym )
1081 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1083 if (sym[0] == 0) return;
1085 while (sym[i] != 0) i++;
1088 while (j > 0 && my_isdigit(sym[j])) j--;
1089 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1095 ocVerifyImage_PEi386 ( ObjectCode* oc )
1099 COFF_section* sectab;
1100 COFF_symbol* symtab;
1102 /* fprintf(stderr, "\nLOADING %s\n", oc->fileName); */
1103 hdr = (COFF_header*)(oc->image);
1104 sectab = (COFF_section*) (
1105 ((UChar*)(oc->image))
1106 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1108 symtab = (COFF_symbol*) (
1109 ((UChar*)(oc->image))
1110 + hdr->PointerToSymbolTable
1112 strtab = ((UChar*)symtab)
1113 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1115 if (hdr->Machine != 0x14c) {
1116 belch("Not x86 PEi386");
1119 if (hdr->SizeOfOptionalHeader != 0) {
1120 belch("PEi386 with nonempty optional header");
1123 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
1124 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
1125 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
1126 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
1127 belch("Not a PEi386 object file");
1130 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
1131 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
1132 belch("Invalid PEi386 word size or endiannness: %d",
1133 (int)(hdr->Characteristics));
1136 /* If the string table size is way crazy, this might indicate that
1137 there are more than 64k relocations, despite claims to the
1138 contrary. Hence this test. */
1139 /* fprintf(stderr, "strtab size %d\n", * (UInt32*)strtab); */
1140 if (* (UInt32*)strtab > 600000) {
1141 /* Note that 600k has no special significance other than being
1142 big enough to handle the almost-2MB-sized lumps that
1143 constitute HSwin32*.o. */
1144 belch("PEi386 object has suspiciously large string table; > 64k relocs?");
1148 /* No further verification after this point; only debug printing. */
1150 IF_DEBUG(linker, i=1);
1151 if (i == 0) return 1;
1154 "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
1156 "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
1158 "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
1160 fprintf ( stderr, "\n" );
1162 "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
1164 "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
1166 "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
1168 "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
1170 "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
1172 "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
1174 "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
1176 /* Print the section table. */
1177 fprintf ( stderr, "\n" );
1178 for (i = 0; i < hdr->NumberOfSections; i++) {
1180 COFF_section* sectab_i
1182 myindex ( sizeof_COFF_section, sectab, i );
1189 printName ( sectab_i->Name, strtab );
1199 sectab_i->VirtualSize,
1200 sectab_i->VirtualAddress,
1201 sectab_i->SizeOfRawData,
1202 sectab_i->PointerToRawData,
1203 sectab_i->NumberOfRelocations,
1204 sectab_i->PointerToRelocations,
1205 sectab_i->PointerToRawData
1207 reltab = (COFF_reloc*) (
1208 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1211 for (j = 0; j < sectab_i->NumberOfRelocations; j++) {
1213 COFF_reloc* rel = (COFF_reloc*)
1214 myindex ( sizeof_COFF_reloc, reltab, j );
1216 " type 0x%-4x vaddr 0x%-8x name `",
1218 rel->VirtualAddress );
1219 sym = (COFF_symbol*)
1220 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
1221 printName ( sym->Name, strtab -10 );
1222 fprintf ( stderr, "'\n" );
1225 fprintf ( stderr, "\n" );
1227 fprintf ( stderr, "\n" );
1228 fprintf ( stderr, "string table has size 0x%x\n", * (UInt32*)strtab );
1229 fprintf ( stderr, "---START of string table---\n");
1230 for (i = 4; i < *(Int32*)strtab; i++) {
1232 fprintf ( stderr, "\n"); else
1233 fprintf( stderr, "%c", strtab[i] );
1235 fprintf ( stderr, "--- END of string table---\n");
1237 fprintf ( stderr, "\n" );
1240 COFF_symbol* symtab_i;
1241 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1242 symtab_i = (COFF_symbol*)
1243 myindex ( sizeof_COFF_symbol, symtab, i );
1249 printName ( symtab_i->Name, strtab );
1258 (Int32)(symtab_i->SectionNumber),
1259 (UInt32)symtab_i->Type,
1260 (UInt32)symtab_i->StorageClass,
1261 (UInt32)symtab_i->NumberOfAuxSymbols
1263 i += symtab_i->NumberOfAuxSymbols;
1267 fprintf ( stderr, "\n" );
1273 ocGetNames_PEi386 ( ObjectCode* oc )
1276 COFF_section* sectab;
1277 COFF_symbol* symtab;
1284 hdr = (COFF_header*)(oc->image);
1285 sectab = (COFF_section*) (
1286 ((UChar*)(oc->image))
1287 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1289 symtab = (COFF_symbol*) (
1290 ((UChar*)(oc->image))
1291 + hdr->PointerToSymbolTable
1293 strtab = ((UChar*)(oc->image))
1294 + hdr->PointerToSymbolTable
1295 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1297 /* Allocate space for any (local, anonymous) .bss sections. */
1299 for (i = 0; i < hdr->NumberOfSections; i++) {
1301 COFF_section* sectab_i
1303 myindex ( sizeof_COFF_section, sectab, i );
1304 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
1305 if (sectab_i->VirtualSize == 0) continue;
1306 /* This is a non-empty .bss section. Allocate zeroed space for
1307 it, and set its PointerToRawData field such that oc->image +
1308 PointerToRawData == addr_of_zeroed_space. */
1309 zspace = stgCallocBytes(1, sectab_i->VirtualSize,
1310 "ocGetNames_PEi386(anonymous bss)");
1311 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
1312 addProddableBlock(oc, zspace, sectab_i->VirtualSize);
1313 /* fprintf(stderr, "BSS anon section at 0x%x\n", zspace); */
1316 /* Copy section information into the ObjectCode. */
1318 for (i = 0; i < hdr->NumberOfSections; i++) {
1324 = SECTIONKIND_OTHER;
1325 COFF_section* sectab_i
1327 myindex ( sizeof_COFF_section, sectab, i );
1328 IF_DEBUG(linker, belch("section name = %s\n", sectab_i->Name ));
1331 /* I'm sure this is the Right Way to do it. However, the
1332 alternative of testing the sectab_i->Name field seems to
1333 work ok with Cygwin.
1335 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
1336 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
1337 kind = SECTIONKIND_CODE_OR_RODATA;
1340 if (0==strcmp(".text",sectab_i->Name) ||
1341 0==strcmp(".rodata",sectab_i->Name))
1342 kind = SECTIONKIND_CODE_OR_RODATA;
1343 if (0==strcmp(".data",sectab_i->Name) ||
1344 0==strcmp(".bss",sectab_i->Name))
1345 kind = SECTIONKIND_RWDATA;
1347 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
1348 sz = sectab_i->SizeOfRawData;
1349 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
1351 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
1352 end = start + sz - 1;
1354 if (kind == SECTIONKIND_OTHER
1355 /* Ignore sections called which contain stabs debugging
1357 && 0 != strcmp(".stab", sectab_i->Name)
1358 && 0 != strcmp(".stabstr", sectab_i->Name)
1360 belch("Unknown PEi386 section name `%s'", sectab_i->Name);
1364 if (kind != SECTIONKIND_OTHER && end >= start) {
1365 addSection(oc, kind, start, end);
1366 addProddableBlock(oc, start, end - start + 1);
1370 /* Copy exported symbols into the ObjectCode. */
1372 oc->n_symbols = hdr->NumberOfSymbols;
1373 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
1374 "ocGetNames_PEi386(oc->symbols)");
1375 /* Call me paranoid; I don't care. */
1376 for (i = 0; i < oc->n_symbols; i++)
1377 oc->symbols[i] = NULL;
1381 COFF_symbol* symtab_i;
1382 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1383 symtab_i = (COFF_symbol*)
1384 myindex ( sizeof_COFF_symbol, symtab, i );
1388 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
1389 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
1390 /* This symbol is global and defined, viz, exported */
1391 /* for MYIMAGE_SYMCLASS_EXTERNAL
1392 && !MYIMAGE_SYM_UNDEFINED,
1393 the address of the symbol is:
1394 address of relevant section + offset in section
1396 COFF_section* sectabent
1397 = (COFF_section*) myindex ( sizeof_COFF_section,
1399 symtab_i->SectionNumber-1 );
1400 addr = ((UChar*)(oc->image))
1401 + (sectabent->PointerToRawData
1405 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
1406 && symtab_i->Value > 0) {
1407 /* This symbol isn't in any section at all, ie, global bss.
1408 Allocate zeroed space for it. */
1409 addr = stgCallocBytes(1, symtab_i->Value,
1410 "ocGetNames_PEi386(non-anonymous bss)");
1411 addSection(oc, SECTIONKIND_RWDATA, addr,
1412 ((UChar*)addr) + symtab_i->Value - 1);
1413 addProddableBlock(oc, addr, symtab_i->Value);
1414 /* fprintf(stderr, "BSS section at 0x%x\n", addr); */
1418 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
1419 /* fprintf(stderr,"addSymbol %p `%s'\n", addr,sname); */
1420 IF_DEBUG(linker, belch("addSymbol %p `%s'\n", addr,sname);)
1421 ASSERT(i >= 0 && i < oc->n_symbols);
1422 /* cstring_from_COFF_symbol_name always succeeds. */
1423 oc->symbols[i] = sname;
1424 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
1428 "IGNORING symbol %d\n"
1432 printName ( symtab_i->Name, strtab );
1441 (Int32)(symtab_i->SectionNumber),
1442 (UInt32)symtab_i->Type,
1443 (UInt32)symtab_i->StorageClass,
1444 (UInt32)symtab_i->NumberOfAuxSymbols
1449 i += symtab_i->NumberOfAuxSymbols;
1458 ocResolve_PEi386 ( ObjectCode* oc )
1461 COFF_section* sectab;
1462 COFF_symbol* symtab;
1471 /* ToDo: should be variable-sized? But is at least safe in the
1472 sense of buffer-overrun-proof. */
1474 /* fprintf(stderr, "resolving for %s\n", oc->fileName); */
1476 hdr = (COFF_header*)(oc->image);
1477 sectab = (COFF_section*) (
1478 ((UChar*)(oc->image))
1479 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1481 symtab = (COFF_symbol*) (
1482 ((UChar*)(oc->image))
1483 + hdr->PointerToSymbolTable
1485 strtab = ((UChar*)(oc->image))
1486 + hdr->PointerToSymbolTable
1487 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1489 for (i = 0; i < hdr->NumberOfSections; i++) {
1490 COFF_section* sectab_i
1492 myindex ( sizeof_COFF_section, sectab, i );
1495 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1498 /* Ignore sections called which contain stabs debugging
1500 if (0 == strcmp(".stab", sectab_i->Name)
1501 || 0 == strcmp(".stabstr", sectab_i->Name))
1504 for (j = 0; j < sectab_i->NumberOfRelocations; j++) {
1506 COFF_reloc* reltab_j
1508 myindex ( sizeof_COFF_reloc, reltab, j );
1510 /* the location to patch */
1512 ((UChar*)(oc->image))
1513 + (sectab_i->PointerToRawData
1514 + reltab_j->VirtualAddress
1515 - sectab_i->VirtualAddress )
1517 /* the existing contents of pP */
1519 /* the symbol to connect to */
1520 sym = (COFF_symbol*)
1521 myindex ( sizeof_COFF_symbol,
1522 symtab, reltab_j->SymbolTableIndex );
1525 "reloc sec %2d num %3d: type 0x%-4x "
1526 "vaddr 0x%-8x name `",
1528 (UInt32)reltab_j->Type,
1529 reltab_j->VirtualAddress );
1530 printName ( sym->Name, strtab );
1531 fprintf ( stderr, "'\n" ));
1533 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
1534 COFF_section* section_sym
1535 = findPEi386SectionCalled ( oc, sym->Name );
1537 belch("%s: can't find section `%s'", oc->fileName, sym->Name);
1540 S = ((UInt32)(oc->image))
1541 + (section_sym->PointerToRawData
1544 copyName ( sym->Name, strtab, symbol, 1000-1 );
1545 (void*)S = lookupLocalSymbol( oc, symbol );
1546 if ((void*)S != NULL) goto foundit;
1547 (void*)S = lookupSymbol( symbol );
1548 if ((void*)S != NULL) goto foundit;
1549 zapTrailingAtSign ( symbol );
1550 (void*)S = lookupLocalSymbol( oc, symbol );
1551 if ((void*)S != NULL) goto foundit;
1552 (void*)S = lookupSymbol( symbol );
1553 if ((void*)S != NULL) goto foundit;
1554 belch("%s: unknown symbol `%s'", oc->fileName, symbol);
1558 checkProddableBlock(oc, pP);
1559 switch (reltab_j->Type) {
1560 case MYIMAGE_REL_I386_DIR32:
1563 case MYIMAGE_REL_I386_REL32:
1564 /* Tricky. We have to insert a displacement at
1565 pP which, when added to the PC for the _next_
1566 insn, gives the address of the target (S).
1567 Problem is to know the address of the next insn
1568 when we only know pP. We assume that this
1569 literal field is always the last in the insn,
1570 so that the address of the next insn is pP+4
1571 -- hence the constant 4.
1572 Also I don't know if A should be added, but so
1573 far it has always been zero.
1576 *pP = S - ((UInt32)pP) - 4;
1579 belch("%s: unhandled PEi386 relocation type %d",
1580 oc->fileName, reltab_j->Type);
1587 IF_DEBUG(linker, belch("completed %s", oc->fileName));
1591 #endif /* defined(OBJFORMAT_PEi386) */
1594 /* --------------------------------------------------------------------------
1596 * ------------------------------------------------------------------------*/
1598 #if defined(OBJFORMAT_ELF)
1603 #if defined(sparc_TARGET_ARCH)
1604 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
1605 #elif defined(i386_TARGET_ARCH)
1606 # define ELF_TARGET_386 /* Used inside <elf.h> */
1608 /* There is a similar case for IA64 in the Solaris2 headers if this
1609 * ever becomes relevant.
1616 findElfSection ( void* objImage, Elf32_Word sh_type )
1619 char* ehdrC = (char*)objImage;
1620 Elf32_Ehdr* ehdr = (Elf32_Ehdr*)ehdrC;
1621 Elf32_Shdr* shdr = (Elf32_Shdr*)(ehdrC + ehdr->e_shoff);
1622 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
1624 for (i = 0; i < ehdr->e_shnum; i++) {
1625 if (shdr[i].sh_type == sh_type
1626 /* Ignore the section header's string table. */
1627 && i != ehdr->e_shstrndx
1628 /* Ignore string tables named .stabstr, as they contain
1630 && 0 != strncmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
1632 ptr = ehdrC + shdr[i].sh_offset;
1641 ocVerifyImage_ELF ( ObjectCode* oc )
1645 int i, j, nent, nstrtab, nsymtabs;
1649 char* ehdrC = (char*)(oc->image);
1650 Elf32_Ehdr* ehdr = ( Elf32_Ehdr*)ehdrC;
1652 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1653 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1654 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1655 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
1656 belch("%s: not an ELF header", oc->fileName);
1659 IF_DEBUG(linker,belch( "Is an ELF header" ));
1661 if (ehdr->e_ident[EI_CLASS] != ELFCLASS32) {
1662 belch("%s: not 32 bit ELF", oc->fileName);
1666 IF_DEBUG(linker,belch( "Is 32 bit ELF" ));
1668 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
1669 IF_DEBUG(linker,belch( "Is little-endian" ));
1671 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
1672 IF_DEBUG(linker,belch( "Is big-endian" ));
1674 belch("%s: unknown endiannness", oc->fileName);
1678 if (ehdr->e_type != ET_REL) {
1679 belch("%s: not a relocatable object (.o) file", oc->fileName);
1682 IF_DEBUG(linker, belch( "Is a relocatable object (.o) file" ));
1684 IF_DEBUG(linker,belch( "Architecture is " ));
1685 switch (ehdr->e_machine) {
1686 case EM_386: IF_DEBUG(linker,belch( "x86" )); break;
1687 case EM_SPARC: IF_DEBUG(linker,belch( "sparc" )); break;
1688 default: IF_DEBUG(linker,belch( "unknown" ));
1689 belch("%s: unknown architecture", oc->fileName);
1693 IF_DEBUG(linker,belch(
1694 "\nSection header table: start %d, n_entries %d, ent_size %d",
1695 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
1697 ASSERT (ehdr->e_shentsize == sizeof(Elf32_Shdr));
1699 shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1701 if (ehdr->e_shstrndx == SHN_UNDEF) {
1702 belch("%s: no section header string table", oc->fileName);
1705 IF_DEBUG(linker,belch( "Section header string table is section %d",
1707 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
1710 for (i = 0; i < ehdr->e_shnum; i++) {
1711 IF_DEBUG(linker,fprintf(stderr, "%2d: ", i ));
1712 IF_DEBUG(linker,fprintf(stderr, "type=%2d ", (int)shdr[i].sh_type ));
1713 IF_DEBUG(linker,fprintf(stderr, "size=%4d ", (int)shdr[i].sh_size ));
1714 IF_DEBUG(linker,fprintf(stderr, "offs=%4d ", (int)shdr[i].sh_offset ));
1715 IF_DEBUG(linker,fprintf(stderr, " (%p .. %p) ",
1716 ehdrC + shdr[i].sh_offset,
1717 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
1719 if (shdr[i].sh_type == SHT_REL) {
1720 IF_DEBUG(linker,fprintf(stderr, "Rel " ));
1721 } else if (shdr[i].sh_type == SHT_RELA) {
1722 IF_DEBUG(linker,fprintf(stderr, "RelA " ));
1724 IF_DEBUG(linker,fprintf(stderr," "));
1727 IF_DEBUG(linker,fprintf(stderr, "sname=%s\n", sh_strtab + shdr[i].sh_name ));
1731 IF_DEBUG(linker,belch( "\nString tables" ));
1734 for (i = 0; i < ehdr->e_shnum; i++) {
1735 if (shdr[i].sh_type == SHT_STRTAB
1736 /* Ignore the section header's string table. */
1737 && i != ehdr->e_shstrndx
1738 /* Ignore string tables named .stabstr, as they contain
1740 && 0 != strncmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
1742 IF_DEBUG(linker,belch(" section %d is a normal string table", i ));
1743 strtab = ehdrC + shdr[i].sh_offset;
1748 belch("%s: no string tables, or too many", oc->fileName);
1753 IF_DEBUG(linker,belch( "\nSymbol tables" ));
1754 for (i = 0; i < ehdr->e_shnum; i++) {
1755 if (shdr[i].sh_type != SHT_SYMTAB) continue;
1756 IF_DEBUG(linker,belch( "section %d is a symbol table", i ));
1758 stab = (Elf32_Sym*) (ehdrC + shdr[i].sh_offset);
1759 nent = shdr[i].sh_size / sizeof(Elf32_Sym);
1760 IF_DEBUG(linker,belch( " number of entries is apparently %d (%d rem)",
1762 shdr[i].sh_size % sizeof(Elf32_Sym)
1764 if (0 != shdr[i].sh_size % sizeof(Elf32_Sym)) {
1765 belch("%s: non-integral number of symbol table entries", oc->fileName);
1768 for (j = 0; j < nent; j++) {
1769 IF_DEBUG(linker,fprintf(stderr, " %2d ", j ));
1770 IF_DEBUG(linker,fprintf(stderr, " sec=%-5d size=%-3d val=%5p ",
1771 (int)stab[j].st_shndx,
1772 (int)stab[j].st_size,
1773 (char*)stab[j].st_value ));
1775 IF_DEBUG(linker,fprintf(stderr, "type=" ));
1776 switch (ELF32_ST_TYPE(stab[j].st_info)) {
1777 case STT_NOTYPE: IF_DEBUG(linker,fprintf(stderr, "notype " )); break;
1778 case STT_OBJECT: IF_DEBUG(linker,fprintf(stderr, "object " )); break;
1779 case STT_FUNC : IF_DEBUG(linker,fprintf(stderr, "func " )); break;
1780 case STT_SECTION: IF_DEBUG(linker,fprintf(stderr, "section" )); break;
1781 case STT_FILE: IF_DEBUG(linker,fprintf(stderr, "file " )); break;
1782 default: IF_DEBUG(linker,fprintf(stderr, "? " )); break;
1784 IF_DEBUG(linker,fprintf(stderr, " " ));
1786 IF_DEBUG(linker,fprintf(stderr, "bind=" ));
1787 switch (ELF32_ST_BIND(stab[j].st_info)) {
1788 case STB_LOCAL : IF_DEBUG(linker,fprintf(stderr, "local " )); break;
1789 case STB_GLOBAL: IF_DEBUG(linker,fprintf(stderr, "global" )); break;
1790 case STB_WEAK : IF_DEBUG(linker,fprintf(stderr, "weak " )); break;
1791 default: IF_DEBUG(linker,fprintf(stderr, "? " )); break;
1793 IF_DEBUG(linker,fprintf(stderr, " " ));
1795 IF_DEBUG(linker,fprintf(stderr, "name=%s\n", strtab + stab[j].st_name ));
1799 if (nsymtabs == 0) {
1800 belch("%s: didn't find any symbol tables", oc->fileName);
1809 ocGetNames_ELF ( ObjectCode* oc )
1814 char* ehdrC = (char*)(oc->image);
1815 Elf32_Ehdr* ehdr = (Elf32_Ehdr*)ehdrC;
1816 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
1817 Elf32_Shdr* shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
1819 ASSERT(symhash != NULL);
1822 belch("%s: no strtab", oc->fileName);
1827 for (i = 0; i < ehdr->e_shnum; i++) {
1828 /* Figure out what kind of section it is. Logic derived from
1829 Figure 1.14 ("Special Sections") of the ELF document
1830 ("Portable Formats Specification, Version 1.1"). */
1831 Elf32_Shdr hdr = shdr[i];
1832 SectionKind kind = SECTIONKIND_OTHER;
1835 if (hdr.sh_type == SHT_PROGBITS
1836 && (hdr.sh_flags & SHF_ALLOC) && (hdr.sh_flags & SHF_EXECINSTR)) {
1837 /* .text-style section */
1838 kind = SECTIONKIND_CODE_OR_RODATA;
1841 if (hdr.sh_type == SHT_PROGBITS
1842 && (hdr.sh_flags & SHF_ALLOC) && (hdr.sh_flags & SHF_WRITE)) {
1843 /* .data-style section */
1844 kind = SECTIONKIND_RWDATA;
1847 if (hdr.sh_type == SHT_PROGBITS
1848 && (hdr.sh_flags & SHF_ALLOC) && !(hdr.sh_flags & SHF_WRITE)) {
1849 /* .rodata-style section */
1850 kind = SECTIONKIND_CODE_OR_RODATA;
1853 if (hdr.sh_type == SHT_NOBITS
1854 && (hdr.sh_flags & SHF_ALLOC) && (hdr.sh_flags & SHF_WRITE)) {
1855 /* .bss-style section */
1856 kind = SECTIONKIND_RWDATA;
1860 if (is_bss && shdr[i].sh_size > 0) {
1861 /* This is a non-empty .bss section. Allocate zeroed space for
1862 it, and set its .sh_offset field such that
1863 ehdrC + .sh_offset == addr_of_zeroed_space. */
1864 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
1865 "ocGetNames_ELF(BSS)");
1866 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
1868 fprintf(stderr, "BSS section at 0x%x, size %d\n",
1869 zspace, shdr[i].sh_size);
1873 /* fill in the section info */
1874 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
1875 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
1876 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0)
1877 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
1879 if (shdr[i].sh_type != SHT_SYMTAB) continue;
1881 /* copy stuff into this module's object symbol table */
1882 stab = (Elf32_Sym*) (ehdrC + shdr[i].sh_offset);
1883 nent = shdr[i].sh_size / sizeof(Elf32_Sym);
1885 oc->n_symbols = nent;
1886 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
1887 "ocGetNames_ELF(oc->symbols)");
1889 for (j = 0; j < nent; j++) {
1891 char isLocal = FALSE; /* avoids uninit-var warning */
1893 char* nm = strtab + stab[j].st_name;
1894 int secno = stab[j].st_shndx;
1896 /* Figure out if we want to add it; if so, set ad to its
1897 address. Otherwise leave ad == NULL. */
1899 if (secno == SHN_COMMON) {
1901 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
1903 fprintf(stderr, "COMMON symbol, size %d name %s\n",
1904 stab[j].st_size, nm);
1906 /* Pointless to do addProddableBlock() for this area,
1907 since the linker should never poke around in it. */
1910 if ( ( ELF32_ST_BIND(stab[j].st_info)==STB_GLOBAL
1911 || ELF32_ST_BIND(stab[j].st_info)==STB_LOCAL
1913 /* and not an undefined symbol */
1914 && stab[j].st_shndx != SHN_UNDEF
1915 /* and not in a "special section" */
1916 && stab[j].st_shndx < SHN_LORESERVE
1918 /* and it's a not a section or string table or anything silly */
1919 ( ELF32_ST_TYPE(stab[j].st_info)==STT_FUNC ||
1920 ELF32_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
1921 ELF32_ST_TYPE(stab[j].st_info)==STT_NOTYPE
1924 /* Section 0 is the undefined section, hence > and not >=. */
1925 ASSERT(secno > 0 && secno < ehdr->e_shnum);
1927 if (shdr[secno].sh_type == SHT_NOBITS) {
1928 fprintf(stderr, " BSS symbol, size %d off %d name %s\n",
1929 stab[j].st_size, stab[j].st_value, nm);
1932 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
1933 if (ELF32_ST_BIND(stab[j].st_info)==STB_LOCAL) {
1936 IF_DEBUG(linker,belch( "addOTabName(GLOB): %10p %s %s",
1937 ad, oc->fileName, nm ));
1942 /* And the decision is ... */
1946 oc->symbols[j] = nm;
1949 /* Ignore entirely. */
1951 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
1955 IF_DEBUG(linker,belch( "skipping `%s'",
1956 strtab + stab[j].st_name ));
1959 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
1960 (int)ELF32_ST_BIND(stab[j].st_info),
1961 (int)ELF32_ST_TYPE(stab[j].st_info),
1962 (int)stab[j].st_shndx,
1963 strtab + stab[j].st_name
1966 oc->symbols[j] = NULL;
1976 /* Do ELF relocations which lack an explicit addend. All x86-linux
1977 relocations appear to be of this form. */
1979 do_Elf32_Rel_relocations ( ObjectCode* oc, char* ehdrC,
1980 Elf32_Shdr* shdr, int shnum,
1981 Elf32_Sym* stab, char* strtab )
1986 Elf32_Rel* rtab = (Elf32_Rel*) (ehdrC + shdr[shnum].sh_offset);
1987 int nent = shdr[shnum].sh_size / sizeof(Elf32_Rel);
1988 int target_shndx = shdr[shnum].sh_info;
1989 int symtab_shndx = shdr[shnum].sh_link;
1990 stab = (Elf32_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
1991 targ = (Elf32_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
1992 IF_DEBUG(linker,belch( "relocations for section %d using symtab %d",
1993 target_shndx, symtab_shndx ));
1994 for (j = 0; j < nent; j++) {
1995 Elf32_Addr offset = rtab[j].r_offset;
1996 Elf32_Word info = rtab[j].r_info;
1998 Elf32_Addr P = ((Elf32_Addr)targ) + offset;
1999 Elf32_Word* pP = (Elf32_Word*)P;
2003 IF_DEBUG(linker,belch( "Rel entry %3d is raw(%6p %6p)",
2004 j, (void*)offset, (void*)info ));
2006 IF_DEBUG(linker,belch( " ZERO" ));
2009 Elf32_Sym sym = stab[ELF32_R_SYM(info)];
2010 /* First see if it is a local symbol. */
2011 if (ELF32_ST_BIND(sym.st_info) == STB_LOCAL) {
2012 /* Yes, so we can get the address directly from the ELF symbol
2014 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2016 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2017 + stab[ELF32_R_SYM(info)].st_value);
2020 /* No, so look up the name in our global table. */
2021 symbol = strtab + sym.st_name;
2022 (void*)S = lookupSymbol( symbol );
2025 belch("%s: unknown symbol `%s'", oc->fileName, symbol);
2028 IF_DEBUG(linker,belch( "`%s' resolves to %p", symbol, (void*)S ));
2030 IF_DEBUG(linker,belch( "Reloc: P = %p S = %p A = %p",
2031 (void*)P, (void*)S, (void*)A ));
2032 checkProddableBlock ( oc, pP );
2033 switch (ELF32_R_TYPE(info)) {
2034 # ifdef i386_TARGET_ARCH
2035 case R_386_32: *pP = S + A; break;
2036 case R_386_PC32: *pP = S + A - P; break;
2039 belch("%s: unhandled ELF relocation(Rel) type %d\n",
2040 oc->fileName, ELF32_R_TYPE(info));
2049 /* Do ELF relocations for which explicit addends are supplied.
2050 sparc-solaris relocations appear to be of this form. */
2052 do_Elf32_Rela_relocations ( ObjectCode* oc, char* ehdrC,
2053 Elf32_Shdr* shdr, int shnum,
2054 Elf32_Sym* stab, char* strtab )
2059 Elf32_Rela* rtab = (Elf32_Rela*) (ehdrC + shdr[shnum].sh_offset);
2060 int nent = shdr[shnum].sh_size / sizeof(Elf32_Rela);
2061 int target_shndx = shdr[shnum].sh_info;
2062 int symtab_shndx = shdr[shnum].sh_link;
2063 stab = (Elf32_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2064 targ = (Elf32_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
2065 IF_DEBUG(linker,belch( "relocations for section %d using symtab %d",
2066 target_shndx, symtab_shndx ));
2067 for (j = 0; j < nent; j++) {
2068 Elf32_Addr offset = rtab[j].r_offset;
2069 Elf32_Word info = rtab[j].r_info;
2070 Elf32_Sword addend = rtab[j].r_addend;
2071 Elf32_Addr P = ((Elf32_Addr)targ) + offset;
2072 Elf32_Addr A = addend; /* Do not delete this; it is used on sparc. */
2074 # if defined(sparc_TARGET_ARCH)
2075 /* This #ifdef only serves to avoid unused-var warnings. */
2076 Elf32_Word* pP = (Elf32_Word*)P;
2080 IF_DEBUG(linker,belch( "Rel entry %3d is raw(%6p %6p %6p) ",
2081 j, (void*)offset, (void*)info,
2084 IF_DEBUG(linker,belch( " ZERO" ));
2087 Elf32_Sym sym = stab[ELF32_R_SYM(info)];
2088 /* First see if it is a local symbol. */
2089 if (ELF32_ST_BIND(sym.st_info) == STB_LOCAL) {
2090 /* Yes, so we can get the address directly from the ELF symbol
2092 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2094 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2095 + stab[ELF32_R_SYM(info)].st_value);
2098 /* No, so look up the name in our global table. */
2099 symbol = strtab + sym.st_name;
2100 (void*)S = lookupSymbol( symbol );
2103 belch("%s: unknown symbol `%s'", oc->fileName, symbol);
2107 fprintf ( stderr, "S %p A %p S+A %p S+A-P %p\n",S,A,S+A,S+A-P);
2110 IF_DEBUG(linker,belch( "`%s' resolves to %p", symbol, (void*)S ));
2112 IF_DEBUG(linker,fprintf ( stderr, "Reloc: P = %p S = %p A = %p\n",
2113 (void*)P, (void*)S, (void*)A ));
2114 checkProddableBlock ( oc, (void*)P );
2115 switch (ELF32_R_TYPE(info)) {
2116 # if defined(sparc_TARGET_ARCH)
2117 case R_SPARC_WDISP30:
2118 w1 = *pP & 0xC0000000;
2119 w2 = (Elf32_Word)((S + A - P) >> 2);
2120 ASSERT((w2 & 0xC0000000) == 0);
2125 w1 = *pP & 0xFFC00000;
2126 w2 = (Elf32_Word)((S + A) >> 10);
2127 ASSERT((w2 & 0xFFC00000) == 0);
2133 w2 = (Elf32_Word)((S + A) & 0x3FF);
2134 ASSERT((w2 & ~0x3FF) == 0);
2138 /* According to the Sun documentation:
2140 This relocation type resembles R_SPARC_32, except it refers to an
2141 unaligned word. That is, the word to be relocated must be treated
2142 as four separate bytes with arbitrary alignment, not as a word
2143 aligned according to the architecture requirements.
2145 (JRS: which means that freeloading on the R_SPARC_32 case
2146 is probably wrong, but hey ...)
2150 w2 = (Elf32_Word)(S + A);
2155 belch("%s: unhandled ELF relocation(RelA) type %d\n",
2156 oc->fileName, ELF32_R_TYPE(info));
2166 ocResolve_ELF ( ObjectCode* oc )
2170 Elf32_Sym* stab = NULL;
2171 char* ehdrC = (char*)(oc->image);
2172 Elf32_Ehdr* ehdr = (Elf32_Ehdr*) ehdrC;
2173 Elf32_Shdr* shdr = (Elf32_Shdr*) (ehdrC + ehdr->e_shoff);
2174 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2176 /* first find "the" symbol table */
2177 stab = (Elf32_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
2179 /* also go find the string table */
2180 strtab = findElfSection ( ehdrC, SHT_STRTAB );
2182 if (stab == NULL || strtab == NULL) {
2183 belch("%s: can't find string or symbol table", oc->fileName);
2187 /* Process the relocation sections. */
2188 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
2190 /* Skip sections called ".rel.stab". These appear to contain
2191 relocation entries that, when done, make the stabs debugging
2192 info point at the right places. We ain't interested in all
2194 if (0 == strncmp(".rel.stab", sh_strtab + shdr[shnum].sh_name, 9))
2197 if (shdr[shnum].sh_type == SHT_REL ) {
2198 ok = do_Elf32_Rel_relocations ( oc, ehdrC, shdr,
2199 shnum, stab, strtab );
2203 if (shdr[shnum].sh_type == SHT_RELA) {
2204 ok = do_Elf32_Rela_relocations ( oc, ehdrC, shdr,
2205 shnum, stab, strtab );
2211 /* Free the local symbol table; we won't need it again. */
2212 freeHashTable(oc->lochash, NULL);