1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
25 #include "LinkerInternals.h"
30 #ifdef HAVE_SYS_TYPES_H
31 #include <sys/types.h>
37 #ifdef HAVE_SYS_STAT_H
41 #if defined(HAVE_DLFCN_H)
45 #if defined(cygwin32_HOST_OS)
50 #ifdef HAVE_SYS_TIME_H
54 #include <sys/fcntl.h>
55 #include <sys/termios.h>
56 #include <sys/utime.h>
57 #include <sys/utsname.h>
61 #if defined(ia64_HOST_ARCH) || defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
66 #if defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
74 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
75 # define OBJFORMAT_ELF
76 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
77 # define OBJFORMAT_PEi386
80 #elif defined(darwin_HOST_OS)
81 # include <mach-o/ppc/reloc.h>
82 # define OBJFORMAT_MACHO
83 # include <mach-o/loader.h>
84 # include <mach-o/nlist.h>
85 # include <mach-o/reloc.h>
86 # include <mach-o/dyld.h>
89 /* Hash table mapping symbol names to Symbol */
90 static /*Str*/HashTable *symhash;
92 /* List of currently loaded objects */
93 ObjectCode *objects = NULL; /* initially empty */
95 #if defined(OBJFORMAT_ELF)
96 static int ocVerifyImage_ELF ( ObjectCode* oc );
97 static int ocGetNames_ELF ( ObjectCode* oc );
98 static int ocResolve_ELF ( ObjectCode* oc );
99 #if defined(powerpc_HOST_ARCH)
100 static int ocAllocateJumpIslands_ELF ( ObjectCode* oc );
102 #elif defined(OBJFORMAT_PEi386)
103 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
104 static int ocGetNames_PEi386 ( ObjectCode* oc );
105 static int ocResolve_PEi386 ( ObjectCode* oc );
106 #elif defined(OBJFORMAT_MACHO)
107 static int ocAllocateJumpIslands_MachO ( ObjectCode* oc );
108 static int ocVerifyImage_MachO ( ObjectCode* oc );
109 static int ocGetNames_MachO ( ObjectCode* oc );
110 static int ocResolve_MachO ( ObjectCode* oc );
112 static void machoInitSymbolsWithoutUnderscore( void );
115 /* -----------------------------------------------------------------------------
116 * Built-in symbols from the RTS
119 typedef struct _RtsSymbolVal {
126 #define Maybe_ForeignObj SymX(mkForeignObjzh_fast)
128 #define Maybe_Stable_Names SymX(mkWeakzh_fast) \
129 SymX(makeStableNamezh_fast) \
130 SymX(finalizzeWeakzh_fast)
132 /* These are not available in GUM!!! -- HWL */
133 #define Maybe_ForeignObj
134 #define Maybe_Stable_Names
137 #if !defined (mingw32_HOST_OS)
138 #define RTS_POSIX_ONLY_SYMBOLS \
139 SymX(stg_sig_install) \
143 #if defined (cygwin32_HOST_OS)
144 #define RTS_MINGW_ONLY_SYMBOLS /**/
145 /* Don't have the ability to read import libs / archives, so
146 * we have to stupidly list a lot of what libcygwin.a
149 #define RTS_CYGWIN_ONLY_SYMBOLS \
227 #elif !defined(mingw32_HOST_OS)
228 #define RTS_MINGW_ONLY_SYMBOLS /**/
229 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
230 #else /* defined(mingw32_HOST_OS) */
231 #define RTS_POSIX_ONLY_SYMBOLS /**/
232 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
234 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
236 #define RTS_MINGW_EXTRA_SYMS \
237 Sym(_imp____mb_cur_max) \
240 #define RTS_MINGW_EXTRA_SYMS
243 /* These are statically linked from the mingw libraries into the ghc
244 executable, so we have to employ this hack. */
245 #define RTS_MINGW_ONLY_SYMBOLS \
246 SymX(asyncReadzh_fast) \
247 SymX(asyncWritezh_fast) \
248 SymX(asyncDoProczh_fast) \
260 SymX(getservbyname) \
261 SymX(getservbyport) \
262 SymX(getprotobynumber) \
263 SymX(getprotobyname) \
264 SymX(gethostbyname) \
265 SymX(gethostbyaddr) \
299 SymX(stg_InstallConsoleEvent) \
301 Sym(_imp___timezone) \
310 RTS_MINGW_EXTRA_SYMS \
315 # define MAIN_CAP_SYM SymX(MainCapability)
317 # define MAIN_CAP_SYM
320 #ifdef TABLES_NEXT_TO_CODE
321 #define RTS_RET_SYMBOLS /* nothing */
323 #define RTS_RET_SYMBOLS \
324 SymX(stg_enter_ret) \
325 SymX(stg_gc_fun_ret) \
333 SymX(stg_ap_pv_ret) \
334 SymX(stg_ap_pp_ret) \
335 SymX(stg_ap_ppv_ret) \
336 SymX(stg_ap_ppp_ret) \
337 SymX(stg_ap_pppv_ret) \
338 SymX(stg_ap_pppp_ret) \
339 SymX(stg_ap_ppppp_ret) \
340 SymX(stg_ap_pppppp_ret)
343 #define RTS_SYMBOLS \
347 SymX(stg_enter_info) \
348 SymX(stg_gc_void_info) \
349 SymX(__stg_gc_enter_1) \
350 SymX(stg_gc_noregs) \
351 SymX(stg_gc_unpt_r1_info) \
352 SymX(stg_gc_unpt_r1) \
353 SymX(stg_gc_unbx_r1_info) \
354 SymX(stg_gc_unbx_r1) \
355 SymX(stg_gc_f1_info) \
357 SymX(stg_gc_d1_info) \
359 SymX(stg_gc_l1_info) \
362 SymX(stg_gc_fun_info) \
364 SymX(stg_gc_gen_info) \
365 SymX(stg_gc_gen_hp) \
367 SymX(stg_gen_yield) \
368 SymX(stg_yield_noregs) \
369 SymX(stg_yield_to_interpreter) \
370 SymX(stg_gen_block) \
371 SymX(stg_block_noregs) \
373 SymX(stg_block_takemvar) \
374 SymX(stg_block_putmvar) \
375 SymX(stg_seq_frame_info) \
377 SymX(MallocFailHook) \
379 SymX(OutOfHeapHook) \
380 SymX(StackOverflowHook) \
381 SymX(__encodeDouble) \
382 SymX(__encodeFloat) \
386 SymX(__gmpz_cmp_si) \
387 SymX(__gmpz_cmp_ui) \
388 SymX(__gmpz_get_si) \
389 SymX(__gmpz_get_ui) \
390 SymX(__int_encodeDouble) \
391 SymX(__int_encodeFloat) \
392 SymX(andIntegerzh_fast) \
393 SymX(atomicallyzh_fast) \
397 SymX(blockAsyncExceptionszh_fast) \
399 SymX(catchRetryzh_fast) \
400 SymX(catchSTMzh_fast) \
401 SymX(closure_flags) \
403 SymX(cmpIntegerzh_fast) \
404 SymX(cmpIntegerIntzh_fast) \
405 SymX(complementIntegerzh_fast) \
406 SymX(createAdjustor) \
407 SymX(decodeDoublezh_fast) \
408 SymX(decodeFloatzh_fast) \
411 SymX(deRefWeakzh_fast) \
412 SymX(deRefStablePtrzh_fast) \
413 SymX(divExactIntegerzh_fast) \
414 SymX(divModIntegerzh_fast) \
417 SymX(forkOS_createThread) \
418 SymX(freeHaskellFunctionPtr) \
419 SymX(freeStablePtr) \
420 SymX(gcdIntegerzh_fast) \
421 SymX(gcdIntegerIntzh_fast) \
422 SymX(gcdIntzh_fast) \
431 SymX(hs_perform_gc) \
432 SymX(hs_free_stable_ptr) \
433 SymX(hs_free_fun_ptr) \
435 SymX(int2Integerzh_fast) \
436 SymX(integer2Intzh_fast) \
437 SymX(integer2Wordzh_fast) \
438 SymX(isCurrentThreadBoundzh_fast) \
439 SymX(isDoubleDenormalized) \
440 SymX(isDoubleInfinite) \
442 SymX(isDoubleNegativeZero) \
443 SymX(isEmptyMVarzh_fast) \
444 SymX(isFloatDenormalized) \
445 SymX(isFloatInfinite) \
447 SymX(isFloatNegativeZero) \
448 SymX(killThreadzh_fast) \
451 SymX(makeStablePtrzh_fast) \
452 SymX(minusIntegerzh_fast) \
453 SymX(mkApUpd0zh_fast) \
454 SymX(myThreadIdzh_fast) \
455 SymX(labelThreadzh_fast) \
456 SymX(newArrayzh_fast) \
457 SymX(newBCOzh_fast) \
458 SymX(newByteArrayzh_fast) \
459 SymX_redirect(newCAF, newDynCAF) \
460 SymX(newMVarzh_fast) \
461 SymX(newMutVarzh_fast) \
462 SymX(newTVarzh_fast) \
463 SymX(atomicModifyMutVarzh_fast) \
464 SymX(newPinnedByteArrayzh_fast) \
465 SymX(orIntegerzh_fast) \
467 SymX(performMajorGC) \
468 SymX(plusIntegerzh_fast) \
471 SymX(putMVarzh_fast) \
472 SymX(quotIntegerzh_fast) \
473 SymX(quotRemIntegerzh_fast) \
475 SymX(raiseIOzh_fast) \
476 SymX(readTVarzh_fast) \
477 SymX(remIntegerzh_fast) \
478 SymX(resetNonBlockingFd) \
483 SymX(rts_checkSchedStatus) \
486 SymX(rts_evalLazyIO) \
487 SymX(rts_evalStableIO) \
491 SymX(rts_getDouble) \
496 SymX(rts_getFunPtr) \
497 SymX(rts_getStablePtr) \
498 SymX(rts_getThreadId) \
500 SymX(rts_getWord32) \
513 SymX(rts_mkStablePtr) \
521 SymX(rtsSupportsBoundThreads) \
523 SymX(__hscore_get_saved_termios) \
524 SymX(__hscore_set_saved_termios) \
526 SymX(startupHaskell) \
527 SymX(shutdownHaskell) \
528 SymX(shutdownHaskellAndExit) \
529 SymX(stable_ptr_table) \
530 SymX(stackOverflow) \
531 SymX(stg_CAF_BLACKHOLE_info) \
532 SymX(awakenBlockedQueue) \
533 SymX(stg_CHARLIKE_closure) \
534 SymX(stg_EMPTY_MVAR_info) \
535 SymX(stg_IND_STATIC_info) \
536 SymX(stg_INTLIKE_closure) \
537 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
538 SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \
539 SymX(stg_WEAK_info) \
540 SymX(stg_ap_0_info) \
541 SymX(stg_ap_v_info) \
542 SymX(stg_ap_f_info) \
543 SymX(stg_ap_d_info) \
544 SymX(stg_ap_l_info) \
545 SymX(stg_ap_n_info) \
546 SymX(stg_ap_p_info) \
547 SymX(stg_ap_pv_info) \
548 SymX(stg_ap_pp_info) \
549 SymX(stg_ap_ppv_info) \
550 SymX(stg_ap_ppp_info) \
551 SymX(stg_ap_pppv_info) \
552 SymX(stg_ap_pppp_info) \
553 SymX(stg_ap_ppppp_info) \
554 SymX(stg_ap_pppppp_info) \
555 SymX(stg_ap_1_upd_info) \
556 SymX(stg_ap_2_upd_info) \
557 SymX(stg_ap_3_upd_info) \
558 SymX(stg_ap_4_upd_info) \
559 SymX(stg_ap_5_upd_info) \
560 SymX(stg_ap_6_upd_info) \
561 SymX(stg_ap_7_upd_info) \
563 SymX(stg_sel_0_upd_info) \
564 SymX(stg_sel_10_upd_info) \
565 SymX(stg_sel_11_upd_info) \
566 SymX(stg_sel_12_upd_info) \
567 SymX(stg_sel_13_upd_info) \
568 SymX(stg_sel_14_upd_info) \
569 SymX(stg_sel_15_upd_info) \
570 SymX(stg_sel_1_upd_info) \
571 SymX(stg_sel_2_upd_info) \
572 SymX(stg_sel_3_upd_info) \
573 SymX(stg_sel_4_upd_info) \
574 SymX(stg_sel_5_upd_info) \
575 SymX(stg_sel_6_upd_info) \
576 SymX(stg_sel_7_upd_info) \
577 SymX(stg_sel_8_upd_info) \
578 SymX(stg_sel_9_upd_info) \
579 SymX(stg_upd_frame_info) \
580 SymX(suspendThread) \
581 SymX(takeMVarzh_fast) \
582 SymX(timesIntegerzh_fast) \
583 SymX(tryPutMVarzh_fast) \
584 SymX(tryTakeMVarzh_fast) \
585 SymX(unblockAsyncExceptionszh_fast) \
587 SymX(unsafeThawArrayzh_fast) \
588 SymX(waitReadzh_fast) \
589 SymX(waitWritezh_fast) \
590 SymX(word2Integerzh_fast) \
591 SymX(writeTVarzh_fast) \
592 SymX(xorIntegerzh_fast) \
595 #ifdef SUPPORT_LONG_LONGS
596 #define RTS_LONG_LONG_SYMS \
597 SymX(int64ToIntegerzh_fast) \
598 SymX(word64ToIntegerzh_fast)
600 #define RTS_LONG_LONG_SYMS /* nothing */
603 // 64-bit support functions in libgcc.a
604 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
605 #define RTS_LIBGCC_SYMBOLS \
615 #elif defined(ia64_HOST_ARCH)
616 #define RTS_LIBGCC_SYMBOLS \
624 #define RTS_LIBGCC_SYMBOLS
627 #ifdef darwin_HOST_OS
628 // Symbols that don't have a leading underscore
629 // on Mac OS X. They have to receive special treatment,
630 // see machoInitSymbolsWithoutUnderscore()
631 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
636 /* entirely bogus claims about types of these symbols */
637 #define Sym(vvv) extern void vvv(void);
638 #define SymX(vvv) /**/
639 #define SymX_redirect(vvv,xxx) /**/
643 RTS_POSIX_ONLY_SYMBOLS
644 RTS_MINGW_ONLY_SYMBOLS
645 RTS_CYGWIN_ONLY_SYMBOLS
651 #ifdef LEADING_UNDERSCORE
652 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
654 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
657 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
659 #define SymX(vvv) Sym(vvv)
661 // SymX_redirect allows us to redirect references to one symbol to
662 // another symbol. See newCAF/newDynCAF for an example.
663 #define SymX_redirect(vvv,xxx) \
664 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
667 static RtsSymbolVal rtsSyms[] = {
671 RTS_POSIX_ONLY_SYMBOLS
672 RTS_MINGW_ONLY_SYMBOLS
673 RTS_CYGWIN_ONLY_SYMBOLS
675 { 0, 0 } /* sentinel */
678 /* -----------------------------------------------------------------------------
679 * Insert symbols into hash tables, checking for duplicates.
681 static void ghciInsertStrHashTable ( char* obj_name,
687 if (lookupHashTable(table, (StgWord)key) == NULL)
689 insertStrHashTable(table, (StgWord)key, data);
694 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
696 "whilst processing object file\n"
698 "This could be caused by:\n"
699 " * Loading two different object files which export the same symbol\n"
700 " * Specifying the same object file twice on the GHCi command line\n"
701 " * An incorrect `package.conf' entry, causing some object to be\n"
703 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
712 /* -----------------------------------------------------------------------------
713 * initialize the object linker
717 static int linker_init_done = 0 ;
719 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
720 static void *dl_prog_handle;
723 /* dlopen(NULL,..) doesn't work so we grab libc explicitly */
724 #if defined(openbsd_HOST_OS)
725 static void *dl_libc_handle;
733 /* Make initLinker idempotent, so we can call it
734 before evey relevant operation; that means we
735 don't need to initialise the linker separately */
736 if (linker_init_done == 1) { return; } else {
737 linker_init_done = 1;
740 symhash = allocStrHashTable();
742 /* populate the symbol table with stuff from the RTS */
743 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
744 ghciInsertStrHashTable("(GHCi built-in symbols)",
745 symhash, sym->lbl, sym->addr);
747 # if defined(OBJFORMAT_MACHO)
748 machoInitSymbolsWithoutUnderscore();
751 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
752 # if defined(RTLD_DEFAULT)
753 dl_prog_handle = RTLD_DEFAULT;
755 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
756 # if defined(openbsd_HOST_OS)
757 dl_libc_handle = dlopen("libc.so", RTLD_LAZY);
759 # endif /* RTLD_DEFAULT */
763 /* -----------------------------------------------------------------------------
764 * Loading DLL or .so dynamic libraries
765 * -----------------------------------------------------------------------------
767 * Add a DLL from which symbols may be found. In the ELF case, just
768 * do RTLD_GLOBAL-style add, so no further messing around needs to
769 * happen in order that symbols in the loaded .so are findable --
770 * lookupSymbol() will subsequently see them by dlsym on the program's
771 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
773 * In the PEi386 case, open the DLLs and put handles to them in a
774 * linked list. When looking for a symbol, try all handles in the
775 * list. This means that we need to load even DLLs that are guaranteed
776 * to be in the ghc.exe image already, just so we can get a handle
777 * to give to loadSymbol, so that we can find the symbols. For such
778 * libraries, the LoadLibrary call should be a no-op except for returning
783 #if defined(OBJFORMAT_PEi386)
784 /* A record for storing handles into DLLs. */
789 struct _OpenedDLL* next;
794 /* A list thereof. */
795 static OpenedDLL* opened_dlls = NULL;
799 addDLL( char *dll_name )
801 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
802 /* ------------------- ELF DLL loader ------------------- */
807 // If we load libHSbase_cbits_dyn.[so|dylib],
808 // then we know that we need to activate another newCAF
809 // related hack in Storage.c because we can't redirect
810 // newCAF to newDynCAF with the system dynamic linker.
811 #ifdef OBJFORMAT_MACHO
812 const char *hsbase = "/libHSbase_cbits_dyn.dylib";
814 const char *hsbase = "/libHSbase_cbits_dyn.so";
816 int namelen = strlen(dll_name);
817 int baselen = strlen(hsbase);
818 if(namelen > baselen && !strcmp(dll_name + namelen - baselen, hsbase))
826 hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
829 /* dlopen failed; return a ptr to the error msg. */
831 if (errmsg == NULL) errmsg = "addDLL: unknown error";
838 # elif defined(OBJFORMAT_PEi386)
839 /* ------------------- Win32 DLL loader ------------------- */
847 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
849 /* See if we've already got it, and ignore if so. */
850 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
851 if (0 == strcmp(o_dll->name, dll_name))
855 /* The file name has no suffix (yet) so that we can try
856 both foo.dll and foo.drv
858 The documentation for LoadLibrary says:
859 If no file name extension is specified in the lpFileName
860 parameter, the default library extension .dll is
861 appended. However, the file name string can include a trailing
862 point character (.) to indicate that the module name has no
865 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
866 sprintf(buf, "%s.DLL", dll_name);
867 instance = LoadLibrary(buf);
868 if (instance == NULL) {
869 sprintf(buf, "%s.DRV", dll_name); // KAA: allow loading of drivers (like winspool.drv)
870 instance = LoadLibrary(buf);
871 if (instance == NULL) {
874 /* LoadLibrary failed; return a ptr to the error msg. */
875 return "addDLL: unknown error";
880 /* Add this DLL to the list of DLLs in which to search for symbols. */
881 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
882 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
883 strcpy(o_dll->name, dll_name);
884 o_dll->instance = instance;
885 o_dll->next = opened_dlls;
890 barf("addDLL: not implemented on this platform");
894 /* -----------------------------------------------------------------------------
895 * lookup a symbol in the hash table
898 lookupSymbol( char *lbl )
902 ASSERT(symhash != NULL);
903 val = lookupStrHashTable(symhash, lbl);
906 # if defined(OBJFORMAT_ELF)
907 # if defined(openbsd_HOST_OS)
908 val = dlsym(dl_prog_handle, lbl);
909 return (val != NULL) ? val : dlsym(dl_libc_handle,lbl);
910 # else /* not openbsd */
911 return dlsym(dl_prog_handle, lbl);
913 # elif defined(OBJFORMAT_MACHO)
914 if(NSIsSymbolNameDefined(lbl)) {
915 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
916 return NSAddressOfSymbol(symbol);
920 # elif defined(OBJFORMAT_PEi386)
923 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
924 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
926 /* HACK: if the name has an initial underscore, try stripping
927 it off & look that up first. I've yet to verify whether there's
928 a Rule that governs whether an initial '_' *should always* be
929 stripped off when mapping from import lib name to the DLL name.
931 sym = GetProcAddress(o_dll->instance, (lbl+1));
933 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
937 sym = GetProcAddress(o_dll->instance, lbl);
939 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
954 __attribute((unused))
956 lookupLocalSymbol( ObjectCode* oc, char *lbl )
960 val = lookupStrHashTable(oc->lochash, lbl);
970 /* -----------------------------------------------------------------------------
971 * Debugging aid: look in GHCi's object symbol tables for symbols
972 * within DELTA bytes of the specified address, and show their names.
975 void ghci_enquire ( char* addr );
977 void ghci_enquire ( char* addr )
982 const int DELTA = 64;
987 for (oc = objects; oc; oc = oc->next) {
988 for (i = 0; i < oc->n_symbols; i++) {
989 sym = oc->symbols[i];
990 if (sym == NULL) continue;
991 // debugBelch("enquire %p %p\n", sym, oc->lochash);
993 if (oc->lochash != NULL) {
994 a = lookupStrHashTable(oc->lochash, sym);
997 a = lookupStrHashTable(symhash, sym);
1000 // debugBelch("ghci_enquire: can't find %s\n", sym);
1002 else if (addr-DELTA <= a && a <= addr+DELTA) {
1003 debugBelch("%p + %3d == `%s'\n", addr, a - addr, sym);
1010 #ifdef ia64_HOST_ARCH
1011 static unsigned int PLTSize(void);
1014 /* -----------------------------------------------------------------------------
1015 * Load an obj (populate the global symbol table, but don't resolve yet)
1017 * Returns: 1 if ok, 0 on error.
1020 loadObj( char *path )
1027 void *map_addr = NULL;
1034 /* debugBelch("loadObj %s\n", path ); */
1036 /* Check that we haven't already loaded this object.
1037 Ignore requests to load multiple times */
1041 for (o = objects; o; o = o->next) {
1042 if (0 == strcmp(o->fileName, path)) {
1044 break; /* don't need to search further */
1048 IF_DEBUG(linker, debugBelch(
1049 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1050 "same object file twice:\n"
1052 "GHCi will ignore this, but be warned.\n"
1054 return 1; /* success */
1058 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1060 # if defined(OBJFORMAT_ELF)
1061 oc->formatName = "ELF";
1062 # elif defined(OBJFORMAT_PEi386)
1063 oc->formatName = "PEi386";
1064 # elif defined(OBJFORMAT_MACHO)
1065 oc->formatName = "Mach-O";
1068 barf("loadObj: not implemented on this platform");
1071 r = stat(path, &st);
1072 if (r == -1) { return 0; }
1074 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1075 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1076 strcpy(oc->fileName, path);
1078 oc->fileSize = st.st_size;
1080 oc->sections = NULL;
1081 oc->lochash = allocStrHashTable();
1082 oc->proddables = NULL;
1084 /* chain it onto the list of objects */
1089 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1091 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1093 #if defined(openbsd_HOST_OS)
1094 fd = open(path, O_RDONLY, S_IRUSR);
1096 fd = open(path, O_RDONLY);
1099 barf("loadObj: can't open `%s'", path);
1101 pagesize = getpagesize();
1103 #ifdef ia64_HOST_ARCH
1104 /* The PLT needs to be right before the object */
1105 n = ROUND_UP(PLTSize(), pagesize);
1106 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1107 if (oc->plt == MAP_FAILED)
1108 barf("loadObj: can't allocate PLT");
1111 map_addr = oc->plt + n;
1114 n = ROUND_UP(oc->fileSize, pagesize);
1116 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
1117 * small memory model on this architecture (see gcc docs,
1120 #ifdef x86_64_HOST_ARCH
1121 #define EXTRA_MAP_FLAGS MAP_32BIT
1123 #define EXTRA_MAP_FLAGS 0
1126 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1127 MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0);
1128 if (oc->image == MAP_FAILED)
1129 barf("loadObj: can't map `%s'", path);
1133 #else /* !USE_MMAP */
1135 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1137 /* load the image into memory */
1138 f = fopen(path, "rb");
1140 barf("loadObj: can't read `%s'", path);
1142 n = fread ( oc->image, 1, oc->fileSize, f );
1143 if (n != oc->fileSize)
1144 barf("loadObj: error whilst reading `%s'", path);
1148 #endif /* USE_MMAP */
1150 # if defined(OBJFORMAT_MACHO)
1151 r = ocAllocateJumpIslands_MachO ( oc );
1152 if (!r) { return r; }
1153 # elif defined(OBJFORMAT_ELF) && defined(powerpc_HOST_ARCH)
1154 r = ocAllocateJumpIslands_ELF ( oc );
1155 if (!r) { return r; }
1158 /* verify the in-memory image */
1159 # if defined(OBJFORMAT_ELF)
1160 r = ocVerifyImage_ELF ( oc );
1161 # elif defined(OBJFORMAT_PEi386)
1162 r = ocVerifyImage_PEi386 ( oc );
1163 # elif defined(OBJFORMAT_MACHO)
1164 r = ocVerifyImage_MachO ( oc );
1166 barf("loadObj: no verify method");
1168 if (!r) { return r; }
1170 /* build the symbol list for this image */
1171 # if defined(OBJFORMAT_ELF)
1172 r = ocGetNames_ELF ( oc );
1173 # elif defined(OBJFORMAT_PEi386)
1174 r = ocGetNames_PEi386 ( oc );
1175 # elif defined(OBJFORMAT_MACHO)
1176 r = ocGetNames_MachO ( oc );
1178 barf("loadObj: no getNames method");
1180 if (!r) { return r; }
1182 /* loaded, but not resolved yet */
1183 oc->status = OBJECT_LOADED;
1188 /* -----------------------------------------------------------------------------
1189 * resolve all the currently unlinked objects in memory
1191 * Returns: 1 if ok, 0 on error.
1201 for (oc = objects; oc; oc = oc->next) {
1202 if (oc->status != OBJECT_RESOLVED) {
1203 # if defined(OBJFORMAT_ELF)
1204 r = ocResolve_ELF ( oc );
1205 # elif defined(OBJFORMAT_PEi386)
1206 r = ocResolve_PEi386 ( oc );
1207 # elif defined(OBJFORMAT_MACHO)
1208 r = ocResolve_MachO ( oc );
1210 barf("resolveObjs: not implemented on this platform");
1212 if (!r) { return r; }
1213 oc->status = OBJECT_RESOLVED;
1219 /* -----------------------------------------------------------------------------
1220 * delete an object from the pool
1223 unloadObj( char *path )
1225 ObjectCode *oc, *prev;
1227 ASSERT(symhash != NULL);
1228 ASSERT(objects != NULL);
1233 for (oc = objects; oc; prev = oc, oc = oc->next) {
1234 if (!strcmp(oc->fileName,path)) {
1236 /* Remove all the mappings for the symbols within this
1241 for (i = 0; i < oc->n_symbols; i++) {
1242 if (oc->symbols[i] != NULL) {
1243 removeStrHashTable(symhash, oc->symbols[i], NULL);
1251 prev->next = oc->next;
1254 /* We're going to leave this in place, in case there are
1255 any pointers from the heap into it: */
1256 /* stgFree(oc->image); */
1257 stgFree(oc->fileName);
1258 stgFree(oc->symbols);
1259 stgFree(oc->sections);
1260 /* The local hash table should have been freed at the end
1261 of the ocResolve_ call on it. */
1262 ASSERT(oc->lochash == NULL);
1268 errorBelch("unloadObj: can't find `%s' to unload", path);
1272 /* -----------------------------------------------------------------------------
1273 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1274 * which may be prodded during relocation, and abort if we try and write
1275 * outside any of these.
1277 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1280 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1281 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1285 pb->next = oc->proddables;
1286 oc->proddables = pb;
1289 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1292 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1293 char* s = (char*)(pb->start);
1294 char* e = s + pb->size - 1;
1295 char* a = (char*)addr;
1296 /* Assumes that the biggest fixup involves a 4-byte write. This
1297 probably needs to be changed to 8 (ie, +7) on 64-bit
1299 if (a >= s && (a+3) <= e) return;
1301 barf("checkProddableBlock: invalid fixup in runtime linker");
1304 /* -----------------------------------------------------------------------------
1305 * Section management.
1307 static void addSection ( ObjectCode* oc, SectionKind kind,
1308 void* start, void* end )
1310 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1314 s->next = oc->sections;
1317 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1318 start, ((char*)end)-1, end - start + 1, kind );
1323 /* --------------------------------------------------------------------------
1324 * PowerPC specifics (jump islands)
1325 * ------------------------------------------------------------------------*/
1327 #if defined(powerpc_HOST_ARCH)
1330 ocAllocateJumpIslands
1332 Allocate additional space at the end of the object file image to make room
1335 PowerPC relative branch instructions have a 24 bit displacement field.
1336 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1337 If a particular imported symbol is outside this range, we have to redirect
1338 the jump to a short piece of new code that just loads the 32bit absolute
1339 address and jumps there.
1340 This function just allocates space for one 16 byte ppcJumpIsland for every
1341 undefined symbol in the object file. The code for the islands is filled in by
1342 makeJumpIsland below.
1345 static int ocAllocateJumpIslands( ObjectCode* oc, int count, int first )
1354 // round up to the nearest 4
1355 aligned = (oc->fileSize + 3) & ~3;
1358 #ifndef linux_HOST_OS /* mremap is a linux extension */
1359 #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined
1362 pagesize = getpagesize();
1363 n = ROUND_UP( oc->fileSize, pagesize );
1364 m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize );
1366 /* The effect of this mremap() call is only the ensure that we have
1367 * a sufficient number of virtually contiguous pages. As returned from
1368 * mremap, the pages past the end of the file are not backed. We give
1369 * them a backing by using MAP_FIXED to map in anonymous pages.
1371 if( (oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE )) == MAP_FAILED )
1373 errorBelch( "Unable to mremap for Jump Islands\n" );
1377 if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC,
1378 MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED )
1380 errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" );
1385 oc->image = stgReallocBytes( oc->image,
1386 aligned + sizeof (ppcJumpIsland) * count,
1387 "ocAllocateJumpIslands" );
1388 #endif /* USE_MMAP */
1390 oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned);
1391 memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count );
1394 oc->jump_islands = NULL;
1396 oc->island_start_symbol = first;
1397 oc->n_islands = count;
1402 static unsigned long makeJumpIsland( ObjectCode* oc,
1403 unsigned long symbolNumber,
1404 unsigned long target )
1406 ppcJumpIsland *island;
1408 if( symbolNumber < oc->island_start_symbol ||
1409 symbolNumber - oc->island_start_symbol > oc->n_islands)
1412 island = &oc->jump_islands[symbolNumber - oc->island_start_symbol];
1414 // lis r12, hi16(target)
1415 island->lis_r12 = 0x3d80;
1416 island->hi_addr = target >> 16;
1418 // ori r12, r12, lo16(target)
1419 island->ori_r12_r12 = 0x618c;
1420 island->lo_addr = target & 0xffff;
1423 island->mtctr_r12 = 0x7d8903a6;
1426 island->bctr = 0x4e800420;
1428 return (unsigned long) island;
1432 ocFlushInstructionCache
1434 Flush the data & instruction caches.
1435 Because the PPC has split data/instruction caches, we have to
1436 do that whenever we modify code at runtime.
1439 static void ocFlushInstructionCache( ObjectCode *oc )
1441 int n = (oc->fileSize + sizeof( ppcJumpIsland ) * oc->n_islands + 3) / 4;
1442 unsigned long *p = (unsigned long *) oc->image;
1446 __asm__ volatile ( "dcbf 0,%0\n\t"
1454 __asm__ volatile ( "sync\n\t"
1460 /* --------------------------------------------------------------------------
1461 * PEi386 specifics (Win32 targets)
1462 * ------------------------------------------------------------------------*/
1464 /* The information for this linker comes from
1465 Microsoft Portable Executable
1466 and Common Object File Format Specification
1467 revision 5.1 January 1998
1468 which SimonM says comes from the MS Developer Network CDs.
1470 It can be found there (on older CDs), but can also be found
1473 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1475 (this is Rev 6.0 from February 1999).
1477 Things move, so if that fails, try searching for it via
1479 http://www.google.com/search?q=PE+COFF+specification
1481 The ultimate reference for the PE format is the Winnt.h
1482 header file that comes with the Platform SDKs; as always,
1483 implementations will drift wrt their documentation.
1485 A good background article on the PE format is Matt Pietrek's
1486 March 1994 article in Microsoft System Journal (MSJ)
1487 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1488 Win32 Portable Executable File Format." The info in there
1489 has recently been updated in a two part article in
1490 MSDN magazine, issues Feb and March 2002,
1491 "Inside Windows: An In-Depth Look into the Win32 Portable
1492 Executable File Format"
1494 John Levine's book "Linkers and Loaders" contains useful
1499 #if defined(OBJFORMAT_PEi386)
1503 typedef unsigned char UChar;
1504 typedef unsigned short UInt16;
1505 typedef unsigned int UInt32;
1512 UInt16 NumberOfSections;
1513 UInt32 TimeDateStamp;
1514 UInt32 PointerToSymbolTable;
1515 UInt32 NumberOfSymbols;
1516 UInt16 SizeOfOptionalHeader;
1517 UInt16 Characteristics;
1521 #define sizeof_COFF_header 20
1528 UInt32 VirtualAddress;
1529 UInt32 SizeOfRawData;
1530 UInt32 PointerToRawData;
1531 UInt32 PointerToRelocations;
1532 UInt32 PointerToLinenumbers;
1533 UInt16 NumberOfRelocations;
1534 UInt16 NumberOfLineNumbers;
1535 UInt32 Characteristics;
1539 #define sizeof_COFF_section 40
1546 UInt16 SectionNumber;
1549 UChar NumberOfAuxSymbols;
1553 #define sizeof_COFF_symbol 18
1558 UInt32 VirtualAddress;
1559 UInt32 SymbolTableIndex;
1564 #define sizeof_COFF_reloc 10
1567 /* From PE spec doc, section 3.3.2 */
1568 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1569 windows.h -- for the same purpose, but I want to know what I'm
1571 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1572 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1573 #define MYIMAGE_FILE_DLL 0x2000
1574 #define MYIMAGE_FILE_SYSTEM 0x1000
1575 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1576 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1577 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1579 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1580 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1581 #define MYIMAGE_SYM_CLASS_STATIC 3
1582 #define MYIMAGE_SYM_UNDEFINED 0
1584 /* From PE spec doc, section 4.1 */
1585 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1586 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1587 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1589 /* From PE spec doc, section 5.2.1 */
1590 #define MYIMAGE_REL_I386_DIR32 0x0006
1591 #define MYIMAGE_REL_I386_REL32 0x0014
1594 /* We use myindex to calculate array addresses, rather than
1595 simply doing the normal subscript thing. That's because
1596 some of the above structs have sizes which are not
1597 a whole number of words. GCC rounds their sizes up to a
1598 whole number of words, which means that the address calcs
1599 arising from using normal C indexing or pointer arithmetic
1600 are just plain wrong. Sigh.
1603 myindex ( int scale, void* base, int index )
1606 ((UChar*)base) + scale * index;
1611 printName ( UChar* name, UChar* strtab )
1613 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1614 UInt32 strtab_offset = * (UInt32*)(name+4);
1615 debugBelch("%s", strtab + strtab_offset );
1618 for (i = 0; i < 8; i++) {
1619 if (name[i] == 0) break;
1620 debugBelch("%c", name[i] );
1627 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1629 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1630 UInt32 strtab_offset = * (UInt32*)(name+4);
1631 strncpy ( dst, strtab+strtab_offset, dstSize );
1637 if (name[i] == 0) break;
1647 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1650 /* If the string is longer than 8 bytes, look in the
1651 string table for it -- this will be correctly zero terminated.
1653 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1654 UInt32 strtab_offset = * (UInt32*)(name+4);
1655 return ((UChar*)strtab) + strtab_offset;
1657 /* Otherwise, if shorter than 8 bytes, return the original,
1658 which by defn is correctly terminated.
1660 if (name[7]==0) return name;
1661 /* The annoying case: 8 bytes. Copy into a temporary
1662 (which is never freed ...)
1664 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1666 strncpy(newstr,name,8);
1672 /* Just compares the short names (first 8 chars) */
1673 static COFF_section *
1674 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1678 = (COFF_header*)(oc->image);
1679 COFF_section* sectab
1681 ((UChar*)(oc->image))
1682 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1684 for (i = 0; i < hdr->NumberOfSections; i++) {
1687 COFF_section* section_i
1689 myindex ( sizeof_COFF_section, sectab, i );
1690 n1 = (UChar*) &(section_i->Name);
1692 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1693 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1694 n1[6]==n2[6] && n1[7]==n2[7])
1703 zapTrailingAtSign ( UChar* sym )
1705 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1707 if (sym[0] == 0) return;
1709 while (sym[i] != 0) i++;
1712 while (j > 0 && my_isdigit(sym[j])) j--;
1713 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1719 ocVerifyImage_PEi386 ( ObjectCode* oc )
1724 COFF_section* sectab;
1725 COFF_symbol* symtab;
1727 /* debugBelch("\nLOADING %s\n", oc->fileName); */
1728 hdr = (COFF_header*)(oc->image);
1729 sectab = (COFF_section*) (
1730 ((UChar*)(oc->image))
1731 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1733 symtab = (COFF_symbol*) (
1734 ((UChar*)(oc->image))
1735 + hdr->PointerToSymbolTable
1737 strtab = ((UChar*)symtab)
1738 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1740 if (hdr->Machine != 0x14c) {
1741 errorBelch("Not x86 PEi386");
1744 if (hdr->SizeOfOptionalHeader != 0) {
1745 errorBelch("PEi386 with nonempty optional header");
1748 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
1749 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
1750 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
1751 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
1752 errorBelch("Not a PEi386 object file");
1755 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
1756 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
1757 errorBelch("Invalid PEi386 word size or endiannness: %d",
1758 (int)(hdr->Characteristics));
1761 /* If the string table size is way crazy, this might indicate that
1762 there are more than 64k relocations, despite claims to the
1763 contrary. Hence this test. */
1764 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
1766 if ( (*(UInt32*)strtab) > 600000 ) {
1767 /* Note that 600k has no special significance other than being
1768 big enough to handle the almost-2MB-sized lumps that
1769 constitute HSwin32*.o. */
1770 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
1775 /* No further verification after this point; only debug printing. */
1777 IF_DEBUG(linker, i=1);
1778 if (i == 0) return 1;
1780 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
1781 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
1782 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
1785 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
1786 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
1787 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
1788 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
1789 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
1790 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
1791 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
1793 /* Print the section table. */
1795 for (i = 0; i < hdr->NumberOfSections; i++) {
1797 COFF_section* sectab_i
1799 myindex ( sizeof_COFF_section, sectab, i );
1806 printName ( sectab_i->Name, strtab );
1816 sectab_i->VirtualSize,
1817 sectab_i->VirtualAddress,
1818 sectab_i->SizeOfRawData,
1819 sectab_i->PointerToRawData,
1820 sectab_i->NumberOfRelocations,
1821 sectab_i->PointerToRelocations,
1822 sectab_i->PointerToRawData
1824 reltab = (COFF_reloc*) (
1825 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1828 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
1829 /* If the relocation field (a short) has overflowed, the
1830 * real count can be found in the first reloc entry.
1832 * See Section 4.1 (last para) of the PE spec (rev6.0).
1834 COFF_reloc* rel = (COFF_reloc*)
1835 myindex ( sizeof_COFF_reloc, reltab, 0 );
1836 noRelocs = rel->VirtualAddress;
1839 noRelocs = sectab_i->NumberOfRelocations;
1843 for (; j < noRelocs; j++) {
1845 COFF_reloc* rel = (COFF_reloc*)
1846 myindex ( sizeof_COFF_reloc, reltab, j );
1848 " type 0x%-4x vaddr 0x%-8x name `",
1850 rel->VirtualAddress );
1851 sym = (COFF_symbol*)
1852 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
1853 /* Hmm..mysterious looking offset - what's it for? SOF */
1854 printName ( sym->Name, strtab -10 );
1861 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
1862 debugBelch("---START of string table---\n");
1863 for (i = 4; i < *(Int32*)strtab; i++) {
1865 debugBelch("\n"); else
1866 debugBelch("%c", strtab[i] );
1868 debugBelch("--- END of string table---\n");
1873 COFF_symbol* symtab_i;
1874 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1875 symtab_i = (COFF_symbol*)
1876 myindex ( sizeof_COFF_symbol, symtab, i );
1882 printName ( symtab_i->Name, strtab );
1891 (Int32)(symtab_i->SectionNumber),
1892 (UInt32)symtab_i->Type,
1893 (UInt32)symtab_i->StorageClass,
1894 (UInt32)symtab_i->NumberOfAuxSymbols
1896 i += symtab_i->NumberOfAuxSymbols;
1906 ocGetNames_PEi386 ( ObjectCode* oc )
1909 COFF_section* sectab;
1910 COFF_symbol* symtab;
1917 hdr = (COFF_header*)(oc->image);
1918 sectab = (COFF_section*) (
1919 ((UChar*)(oc->image))
1920 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1922 symtab = (COFF_symbol*) (
1923 ((UChar*)(oc->image))
1924 + hdr->PointerToSymbolTable
1926 strtab = ((UChar*)(oc->image))
1927 + hdr->PointerToSymbolTable
1928 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1930 /* Allocate space for any (local, anonymous) .bss sections. */
1932 for (i = 0; i < hdr->NumberOfSections; i++) {
1934 COFF_section* sectab_i
1936 myindex ( sizeof_COFF_section, sectab, i );
1937 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
1938 if (sectab_i->VirtualSize == 0) continue;
1939 /* This is a non-empty .bss section. Allocate zeroed space for
1940 it, and set its PointerToRawData field such that oc->image +
1941 PointerToRawData == addr_of_zeroed_space. */
1942 zspace = stgCallocBytes(1, sectab_i->VirtualSize,
1943 "ocGetNames_PEi386(anonymous bss)");
1944 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
1945 addProddableBlock(oc, zspace, sectab_i->VirtualSize);
1946 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
1949 /* Copy section information into the ObjectCode. */
1951 for (i = 0; i < hdr->NumberOfSections; i++) {
1957 = SECTIONKIND_OTHER;
1958 COFF_section* sectab_i
1960 myindex ( sizeof_COFF_section, sectab, i );
1961 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
1964 /* I'm sure this is the Right Way to do it. However, the
1965 alternative of testing the sectab_i->Name field seems to
1966 work ok with Cygwin.
1968 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
1969 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
1970 kind = SECTIONKIND_CODE_OR_RODATA;
1973 if (0==strcmp(".text",sectab_i->Name) ||
1974 0==strcmp(".rodata",sectab_i->Name))
1975 kind = SECTIONKIND_CODE_OR_RODATA;
1976 if (0==strcmp(".data",sectab_i->Name) ||
1977 0==strcmp(".bss",sectab_i->Name))
1978 kind = SECTIONKIND_RWDATA;
1980 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
1981 sz = sectab_i->SizeOfRawData;
1982 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
1984 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
1985 end = start + sz - 1;
1987 if (kind == SECTIONKIND_OTHER
1988 /* Ignore sections called which contain stabs debugging
1990 && 0 != strcmp(".stab", sectab_i->Name)
1991 && 0 != strcmp(".stabstr", sectab_i->Name)
1993 errorBelch("Unknown PEi386 section name `%s'", sectab_i->Name);
1997 if (kind != SECTIONKIND_OTHER && end >= start) {
1998 addSection(oc, kind, start, end);
1999 addProddableBlock(oc, start, end - start + 1);
2003 /* Copy exported symbols into the ObjectCode. */
2005 oc->n_symbols = hdr->NumberOfSymbols;
2006 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2007 "ocGetNames_PEi386(oc->symbols)");
2008 /* Call me paranoid; I don't care. */
2009 for (i = 0; i < oc->n_symbols; i++)
2010 oc->symbols[i] = NULL;
2014 COFF_symbol* symtab_i;
2015 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2016 symtab_i = (COFF_symbol*)
2017 myindex ( sizeof_COFF_symbol, symtab, i );
2021 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2022 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2023 /* This symbol is global and defined, viz, exported */
2024 /* for MYIMAGE_SYMCLASS_EXTERNAL
2025 && !MYIMAGE_SYM_UNDEFINED,
2026 the address of the symbol is:
2027 address of relevant section + offset in section
2029 COFF_section* sectabent
2030 = (COFF_section*) myindex ( sizeof_COFF_section,
2032 symtab_i->SectionNumber-1 );
2033 addr = ((UChar*)(oc->image))
2034 + (sectabent->PointerToRawData
2038 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2039 && symtab_i->Value > 0) {
2040 /* This symbol isn't in any section at all, ie, global bss.
2041 Allocate zeroed space for it. */
2042 addr = stgCallocBytes(1, symtab_i->Value,
2043 "ocGetNames_PEi386(non-anonymous bss)");
2044 addSection(oc, SECTIONKIND_RWDATA, addr,
2045 ((UChar*)addr) + symtab_i->Value - 1);
2046 addProddableBlock(oc, addr, symtab_i->Value);
2047 /* debugBelch("BSS section at 0x%x\n", addr); */
2050 if (addr != NULL ) {
2051 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2052 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2053 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2054 ASSERT(i >= 0 && i < oc->n_symbols);
2055 /* cstring_from_COFF_symbol_name always succeeds. */
2056 oc->symbols[i] = sname;
2057 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2061 "IGNORING symbol %d\n"
2065 printName ( symtab_i->Name, strtab );
2074 (Int32)(symtab_i->SectionNumber),
2075 (UInt32)symtab_i->Type,
2076 (UInt32)symtab_i->StorageClass,
2077 (UInt32)symtab_i->NumberOfAuxSymbols
2082 i += symtab_i->NumberOfAuxSymbols;
2091 ocResolve_PEi386 ( ObjectCode* oc )
2094 COFF_section* sectab;
2095 COFF_symbol* symtab;
2105 /* ToDo: should be variable-sized? But is at least safe in the
2106 sense of buffer-overrun-proof. */
2108 /* debugBelch("resolving for %s\n", oc->fileName); */
2110 hdr = (COFF_header*)(oc->image);
2111 sectab = (COFF_section*) (
2112 ((UChar*)(oc->image))
2113 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2115 symtab = (COFF_symbol*) (
2116 ((UChar*)(oc->image))
2117 + hdr->PointerToSymbolTable
2119 strtab = ((UChar*)(oc->image))
2120 + hdr->PointerToSymbolTable
2121 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2123 for (i = 0; i < hdr->NumberOfSections; i++) {
2124 COFF_section* sectab_i
2126 myindex ( sizeof_COFF_section, sectab, i );
2129 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2132 /* Ignore sections called which contain stabs debugging
2134 if (0 == strcmp(".stab", sectab_i->Name)
2135 || 0 == strcmp(".stabstr", sectab_i->Name))
2138 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2139 /* If the relocation field (a short) has overflowed, the
2140 * real count can be found in the first reloc entry.
2142 * See Section 4.1 (last para) of the PE spec (rev6.0).
2144 * Nov2003 update: the GNU linker still doesn't correctly
2145 * handle the generation of relocatable object files with
2146 * overflown relocations. Hence the output to warn of potential
2149 COFF_reloc* rel = (COFF_reloc*)
2150 myindex ( sizeof_COFF_reloc, reltab, 0 );
2151 noRelocs = rel->VirtualAddress;
2152 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2156 noRelocs = sectab_i->NumberOfRelocations;
2161 for (; j < noRelocs; j++) {
2163 COFF_reloc* reltab_j
2165 myindex ( sizeof_COFF_reloc, reltab, j );
2167 /* the location to patch */
2169 ((UChar*)(oc->image))
2170 + (sectab_i->PointerToRawData
2171 + reltab_j->VirtualAddress
2172 - sectab_i->VirtualAddress )
2174 /* the existing contents of pP */
2176 /* the symbol to connect to */
2177 sym = (COFF_symbol*)
2178 myindex ( sizeof_COFF_symbol,
2179 symtab, reltab_j->SymbolTableIndex );
2182 "reloc sec %2d num %3d: type 0x%-4x "
2183 "vaddr 0x%-8x name `",
2185 (UInt32)reltab_j->Type,
2186 reltab_j->VirtualAddress );
2187 printName ( sym->Name, strtab );
2188 debugBelch("'\n" ));
2190 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2191 COFF_section* section_sym
2192 = findPEi386SectionCalled ( oc, sym->Name );
2194 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2197 S = ((UInt32)(oc->image))
2198 + (section_sym->PointerToRawData
2201 copyName ( sym->Name, strtab, symbol, 1000-1 );
2202 (void*)S = lookupLocalSymbol( oc, symbol );
2203 if ((void*)S != NULL) goto foundit;
2204 (void*)S = lookupSymbol( symbol );
2205 if ((void*)S != NULL) goto foundit;
2206 zapTrailingAtSign ( symbol );
2207 (void*)S = lookupLocalSymbol( oc, symbol );
2208 if ((void*)S != NULL) goto foundit;
2209 (void*)S = lookupSymbol( symbol );
2210 if ((void*)S != NULL) goto foundit;
2211 /* Newline first because the interactive linker has printed "linking..." */
2212 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2216 checkProddableBlock(oc, pP);
2217 switch (reltab_j->Type) {
2218 case MYIMAGE_REL_I386_DIR32:
2221 case MYIMAGE_REL_I386_REL32:
2222 /* Tricky. We have to insert a displacement at
2223 pP which, when added to the PC for the _next_
2224 insn, gives the address of the target (S).
2225 Problem is to know the address of the next insn
2226 when we only know pP. We assume that this
2227 literal field is always the last in the insn,
2228 so that the address of the next insn is pP+4
2229 -- hence the constant 4.
2230 Also I don't know if A should be added, but so
2231 far it has always been zero.
2234 *pP = S - ((UInt32)pP) - 4;
2237 debugBelch("%s: unhandled PEi386 relocation type %d",
2238 oc->fileName, reltab_j->Type);
2245 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2249 #endif /* defined(OBJFORMAT_PEi386) */
2252 /* --------------------------------------------------------------------------
2254 * ------------------------------------------------------------------------*/
2256 #if defined(OBJFORMAT_ELF)
2261 #if defined(sparc_HOST_ARCH)
2262 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2263 #elif defined(i386_HOST_ARCH)
2264 # define ELF_TARGET_386 /* Used inside <elf.h> */
2265 #elif defined(x86_64_HOST_ARCH)
2266 # define ELF_TARGET_X64_64
2268 #elif defined (ia64_HOST_ARCH)
2269 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2271 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2272 # define ELF_NEED_GOT /* needs Global Offset Table */
2273 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2276 #if !defined(openbsd_HOST_OS)
2279 /* openbsd elf has things in different places, with diff names */
2280 #include <elf_abi.h>
2281 #include <machine/reloc.h>
2282 #define R_386_32 RELOC_32
2283 #define R_386_PC32 RELOC_PC32
2287 * Define a set of types which can be used for both ELF32 and ELF64
2291 #define ELFCLASS ELFCLASS64
2292 #define Elf_Addr Elf64_Addr
2293 #define Elf_Word Elf64_Word
2294 #define Elf_Sword Elf64_Sword
2295 #define Elf_Ehdr Elf64_Ehdr
2296 #define Elf_Phdr Elf64_Phdr
2297 #define Elf_Shdr Elf64_Shdr
2298 #define Elf_Sym Elf64_Sym
2299 #define Elf_Rel Elf64_Rel
2300 #define Elf_Rela Elf64_Rela
2301 #define ELF_ST_TYPE ELF64_ST_TYPE
2302 #define ELF_ST_BIND ELF64_ST_BIND
2303 #define ELF_R_TYPE ELF64_R_TYPE
2304 #define ELF_R_SYM ELF64_R_SYM
2306 #define ELFCLASS ELFCLASS32
2307 #define Elf_Addr Elf32_Addr
2308 #define Elf_Word Elf32_Word
2309 #define Elf_Sword Elf32_Sword
2310 #define Elf_Ehdr Elf32_Ehdr
2311 #define Elf_Phdr Elf32_Phdr
2312 #define Elf_Shdr Elf32_Shdr
2313 #define Elf_Sym Elf32_Sym
2314 #define Elf_Rel Elf32_Rel
2315 #define Elf_Rela Elf32_Rela
2317 #define ELF_ST_TYPE ELF32_ST_TYPE
2320 #define ELF_ST_BIND ELF32_ST_BIND
2323 #define ELF_R_TYPE ELF32_R_TYPE
2326 #define ELF_R_SYM ELF32_R_SYM
2332 * Functions to allocate entries in dynamic sections. Currently we simply
2333 * preallocate a large number, and we don't check if a entry for the given
2334 * target already exists (a linear search is too slow). Ideally these
2335 * entries would be associated with symbols.
2338 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2339 #define GOT_SIZE 0x20000
2340 #define FUNCTION_TABLE_SIZE 0x10000
2341 #define PLT_SIZE 0x08000
2344 static Elf_Addr got[GOT_SIZE];
2345 static unsigned int gotIndex;
2346 static Elf_Addr gp_val = (Elf_Addr)got;
2349 allocateGOTEntry(Elf_Addr target)
2353 if (gotIndex >= GOT_SIZE)
2354 barf("Global offset table overflow");
2356 entry = &got[gotIndex++];
2358 return (Elf_Addr)entry;
2362 #ifdef ELF_FUNCTION_DESC
2368 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2369 static unsigned int functionTableIndex;
2372 allocateFunctionDesc(Elf_Addr target)
2374 FunctionDesc *entry;
2376 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2377 barf("Function table overflow");
2379 entry = &functionTable[functionTableIndex++];
2381 entry->gp = (Elf_Addr)gp_val;
2382 return (Elf_Addr)entry;
2386 copyFunctionDesc(Elf_Addr target)
2388 FunctionDesc *olddesc = (FunctionDesc *)target;
2389 FunctionDesc *newdesc;
2391 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2392 newdesc->gp = olddesc->gp;
2393 return (Elf_Addr)newdesc;
2398 #ifdef ia64_HOST_ARCH
2399 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2400 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2402 static unsigned char plt_code[] =
2404 /* taken from binutils bfd/elfxx-ia64.c */
2405 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2406 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2407 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2408 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2409 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2410 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2413 /* If we can't get to the function descriptor via gp, take a local copy of it */
2414 #define PLT_RELOC(code, target) { \
2415 Elf64_Sxword rel_value = target - gp_val; \
2416 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2417 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2419 ia64_reloc_gprel22((Elf_Addr)code, target); \
2424 unsigned char code[sizeof(plt_code)];
2428 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2430 PLTEntry *plt = (PLTEntry *)oc->plt;
2433 if (oc->pltIndex >= PLT_SIZE)
2434 barf("Procedure table overflow");
2436 entry = &plt[oc->pltIndex++];
2437 memcpy(entry->code, plt_code, sizeof(entry->code));
2438 PLT_RELOC(entry->code, target);
2439 return (Elf_Addr)entry;
2445 return (PLT_SIZE * sizeof(PLTEntry));
2451 * Generic ELF functions
2455 findElfSection ( void* objImage, Elf_Word sh_type )
2457 char* ehdrC = (char*)objImage;
2458 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2459 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2460 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2464 for (i = 0; i < ehdr->e_shnum; i++) {
2465 if (shdr[i].sh_type == sh_type
2466 /* Ignore the section header's string table. */
2467 && i != ehdr->e_shstrndx
2468 /* Ignore string tables named .stabstr, as they contain
2470 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2472 ptr = ehdrC + shdr[i].sh_offset;
2479 #if defined(ia64_HOST_ARCH)
2481 findElfSegment ( void* objImage, Elf_Addr vaddr )
2483 char* ehdrC = (char*)objImage;
2484 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2485 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2486 Elf_Addr segaddr = 0;
2489 for (i = 0; i < ehdr->e_phnum; i++) {
2490 segaddr = phdr[i].p_vaddr;
2491 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2499 ocVerifyImage_ELF ( ObjectCode* oc )
2503 int i, j, nent, nstrtab, nsymtabs;
2507 char* ehdrC = (char*)(oc->image);
2508 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2510 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2511 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2512 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2513 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2514 errorBelch("%s: not an ELF object", oc->fileName);
2518 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2519 errorBelch("%s: unsupported ELF format", oc->fileName);
2523 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2524 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2526 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2527 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2529 errorBelch("%s: unknown endiannness", oc->fileName);
2533 if (ehdr->e_type != ET_REL) {
2534 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2537 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2539 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2540 switch (ehdr->e_machine) {
2541 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2542 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2544 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2546 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2548 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2550 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2551 errorBelch("%s: unknown architecture", oc->fileName);
2555 IF_DEBUG(linker,debugBelch(
2556 "\nSection header table: start %d, n_entries %d, ent_size %d\n",
2557 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2559 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2561 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2563 if (ehdr->e_shstrndx == SHN_UNDEF) {
2564 errorBelch("%s: no section header string table", oc->fileName);
2567 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2569 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2572 for (i = 0; i < ehdr->e_shnum; i++) {
2573 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2574 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2575 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2576 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2577 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2578 ehdrC + shdr[i].sh_offset,
2579 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2581 if (shdr[i].sh_type == SHT_REL) {
2582 IF_DEBUG(linker,debugBelch("Rel " ));
2583 } else if (shdr[i].sh_type == SHT_RELA) {
2584 IF_DEBUG(linker,debugBelch("RelA " ));
2586 IF_DEBUG(linker,debugBelch(" "));
2589 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2593 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2596 for (i = 0; i < ehdr->e_shnum; i++) {
2597 if (shdr[i].sh_type == SHT_STRTAB
2598 /* Ignore the section header's string table. */
2599 && i != ehdr->e_shstrndx
2600 /* Ignore string tables named .stabstr, as they contain
2602 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2604 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2605 strtab = ehdrC + shdr[i].sh_offset;
2610 errorBelch("%s: no string tables, or too many", oc->fileName);
2615 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2616 for (i = 0; i < ehdr->e_shnum; i++) {
2617 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2618 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2620 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2621 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2622 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%d rem)\n",
2624 shdr[i].sh_size % sizeof(Elf_Sym)
2626 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2627 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2630 for (j = 0; j < nent; j++) {
2631 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2632 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
2633 (int)stab[j].st_shndx,
2634 (int)stab[j].st_size,
2635 (char*)stab[j].st_value ));
2637 IF_DEBUG(linker,debugBelch("type=" ));
2638 switch (ELF_ST_TYPE(stab[j].st_info)) {
2639 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
2640 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
2641 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
2642 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
2643 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
2644 default: IF_DEBUG(linker,debugBelch("? " )); break;
2646 IF_DEBUG(linker,debugBelch(" " ));
2648 IF_DEBUG(linker,debugBelch("bind=" ));
2649 switch (ELF_ST_BIND(stab[j].st_info)) {
2650 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
2651 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
2652 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
2653 default: IF_DEBUG(linker,debugBelch("? " )); break;
2655 IF_DEBUG(linker,debugBelch(" " ));
2657 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
2661 if (nsymtabs == 0) {
2662 errorBelch("%s: didn't find any symbol tables", oc->fileName);
2669 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
2673 if (hdr->sh_type == SHT_PROGBITS
2674 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
2675 /* .text-style section */
2676 return SECTIONKIND_CODE_OR_RODATA;
2679 if (hdr->sh_type == SHT_PROGBITS
2680 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2681 /* .data-style section */
2682 return SECTIONKIND_RWDATA;
2685 if (hdr->sh_type == SHT_PROGBITS
2686 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
2687 /* .rodata-style section */
2688 return SECTIONKIND_CODE_OR_RODATA;
2691 if (hdr->sh_type == SHT_NOBITS
2692 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2693 /* .bss-style section */
2695 return SECTIONKIND_RWDATA;
2698 return SECTIONKIND_OTHER;
2703 ocGetNames_ELF ( ObjectCode* oc )
2708 char* ehdrC = (char*)(oc->image);
2709 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2710 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
2711 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2713 ASSERT(symhash != NULL);
2716 errorBelch("%s: no strtab", oc->fileName);
2721 for (i = 0; i < ehdr->e_shnum; i++) {
2722 /* Figure out what kind of section it is. Logic derived from
2723 Figure 1.14 ("Special Sections") of the ELF document
2724 ("Portable Formats Specification, Version 1.1"). */
2726 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
2728 if (is_bss && shdr[i].sh_size > 0) {
2729 /* This is a non-empty .bss section. Allocate zeroed space for
2730 it, and set its .sh_offset field such that
2731 ehdrC + .sh_offset == addr_of_zeroed_space. */
2732 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
2733 "ocGetNames_ELF(BSS)");
2734 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
2736 debugBelch("BSS section at 0x%x, size %d\n",
2737 zspace, shdr[i].sh_size);
2741 /* fill in the section info */
2742 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
2743 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
2744 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
2745 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
2748 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2750 /* copy stuff into this module's object symbol table */
2751 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2752 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2754 oc->n_symbols = nent;
2755 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2756 "ocGetNames_ELF(oc->symbols)");
2758 for (j = 0; j < nent; j++) {
2760 char isLocal = FALSE; /* avoids uninit-var warning */
2762 char* nm = strtab + stab[j].st_name;
2763 int secno = stab[j].st_shndx;
2765 /* Figure out if we want to add it; if so, set ad to its
2766 address. Otherwise leave ad == NULL. */
2768 if (secno == SHN_COMMON) {
2770 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
2772 debugBelch("COMMON symbol, size %d name %s\n",
2773 stab[j].st_size, nm);
2775 /* Pointless to do addProddableBlock() for this area,
2776 since the linker should never poke around in it. */
2779 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
2780 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
2782 /* and not an undefined symbol */
2783 && stab[j].st_shndx != SHN_UNDEF
2784 /* and not in a "special section" */
2785 && stab[j].st_shndx < SHN_LORESERVE
2787 /* and it's a not a section or string table or anything silly */
2788 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
2789 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
2790 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
2793 /* Section 0 is the undefined section, hence > and not >=. */
2794 ASSERT(secno > 0 && secno < ehdr->e_shnum);
2796 if (shdr[secno].sh_type == SHT_NOBITS) {
2797 debugBelch(" BSS symbol, size %d off %d name %s\n",
2798 stab[j].st_size, stab[j].st_value, nm);
2801 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
2802 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
2805 #ifdef ELF_FUNCTION_DESC
2806 /* dlsym() and the initialisation table both give us function
2807 * descriptors, so to be consistent we store function descriptors
2808 * in the symbol table */
2809 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
2810 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
2812 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
2813 ad, oc->fileName, nm ));
2818 /* And the decision is ... */
2822 oc->symbols[j] = nm;
2825 /* Ignore entirely. */
2827 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
2831 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
2832 strtab + stab[j].st_name ));
2835 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
2836 (int)ELF_ST_BIND(stab[j].st_info),
2837 (int)ELF_ST_TYPE(stab[j].st_info),
2838 (int)stab[j].st_shndx,
2839 strtab + stab[j].st_name
2842 oc->symbols[j] = NULL;
2851 /* Do ELF relocations which lack an explicit addend. All x86-linux
2852 relocations appear to be of this form. */
2854 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
2855 Elf_Shdr* shdr, int shnum,
2856 Elf_Sym* stab, char* strtab )
2861 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
2862 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
2863 int target_shndx = shdr[shnum].sh_info;
2864 int symtab_shndx = shdr[shnum].sh_link;
2866 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2867 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
2868 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2869 target_shndx, symtab_shndx ));
2871 /* Skip sections that we're not interested in. */
2874 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
2875 if (kind == SECTIONKIND_OTHER) {
2876 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
2881 for (j = 0; j < nent; j++) {
2882 Elf_Addr offset = rtab[j].r_offset;
2883 Elf_Addr info = rtab[j].r_info;
2885 Elf_Addr P = ((Elf_Addr)targ) + offset;
2886 Elf_Word* pP = (Elf_Word*)P;
2892 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
2893 j, (void*)offset, (void*)info ));
2895 IF_DEBUG(linker,debugBelch( " ZERO" ));
2898 Elf_Sym sym = stab[ELF_R_SYM(info)];
2899 /* First see if it is a local symbol. */
2900 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2901 /* Yes, so we can get the address directly from the ELF symbol
2903 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2905 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2906 + stab[ELF_R_SYM(info)].st_value);
2909 /* No, so look up the name in our global table. */
2910 symbol = strtab + sym.st_name;
2911 S_tmp = lookupSymbol( symbol );
2912 S = (Elf_Addr)S_tmp;
2915 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2918 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
2921 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
2922 (void*)P, (void*)S, (void*)A ));
2923 checkProddableBlock ( oc, pP );
2927 switch (ELF_R_TYPE(info)) {
2928 # ifdef i386_HOST_ARCH
2929 case R_386_32: *pP = value; break;
2930 case R_386_PC32: *pP = value - P; break;
2933 errorBelch("%s: unhandled ELF relocation(Rel) type %d\n",
2934 oc->fileName, ELF_R_TYPE(info));
2942 /* Do ELF relocations for which explicit addends are supplied.
2943 sparc-solaris relocations appear to be of this form. */
2945 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
2946 Elf_Shdr* shdr, int shnum,
2947 Elf_Sym* stab, char* strtab )
2952 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
2953 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
2954 int target_shndx = shdr[shnum].sh_info;
2955 int symtab_shndx = shdr[shnum].sh_link;
2957 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2958 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
2959 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2960 target_shndx, symtab_shndx ));
2962 for (j = 0; j < nent; j++) {
2963 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
2964 /* This #ifdef only serves to avoid unused-var warnings. */
2965 Elf_Addr offset = rtab[j].r_offset;
2966 Elf_Addr P = targ + offset;
2968 Elf_Addr info = rtab[j].r_info;
2969 Elf_Addr A = rtab[j].r_addend;
2973 # if defined(sparc_HOST_ARCH)
2974 Elf_Word* pP = (Elf_Word*)P;
2976 # elif defined(ia64_HOST_ARCH)
2977 Elf64_Xword *pP = (Elf64_Xword *)P;
2979 # elif defined(powerpc_HOST_ARCH)
2983 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
2984 j, (void*)offset, (void*)info,
2987 IF_DEBUG(linker,debugBelch( " ZERO" ));
2990 Elf_Sym sym = stab[ELF_R_SYM(info)];
2991 /* First see if it is a local symbol. */
2992 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2993 /* Yes, so we can get the address directly from the ELF symbol
2995 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2997 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2998 + stab[ELF_R_SYM(info)].st_value);
2999 #ifdef ELF_FUNCTION_DESC
3000 /* Make a function descriptor for this function */
3001 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3002 S = allocateFunctionDesc(S + A);
3007 /* No, so look up the name in our global table. */
3008 symbol = strtab + sym.st_name;
3009 S_tmp = lookupSymbol( symbol );
3010 S = (Elf_Addr)S_tmp;
3012 #ifdef ELF_FUNCTION_DESC
3013 /* If a function, already a function descriptor - we would
3014 have to copy it to add an offset. */
3015 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3016 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3020 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3023 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3026 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3027 (void*)P, (void*)S, (void*)A ));
3028 /* checkProddableBlock ( oc, (void*)P ); */
3032 switch (ELF_R_TYPE(info)) {
3033 # if defined(sparc_HOST_ARCH)
3034 case R_SPARC_WDISP30:
3035 w1 = *pP & 0xC0000000;
3036 w2 = (Elf_Word)((value - P) >> 2);
3037 ASSERT((w2 & 0xC0000000) == 0);
3042 w1 = *pP & 0xFFC00000;
3043 w2 = (Elf_Word)(value >> 10);
3044 ASSERT((w2 & 0xFFC00000) == 0);
3050 w2 = (Elf_Word)(value & 0x3FF);
3051 ASSERT((w2 & ~0x3FF) == 0);
3055 /* According to the Sun documentation:
3057 This relocation type resembles R_SPARC_32, except it refers to an
3058 unaligned word. That is, the word to be relocated must be treated
3059 as four separate bytes with arbitrary alignment, not as a word
3060 aligned according to the architecture requirements.
3062 (JRS: which means that freeloading on the R_SPARC_32 case
3063 is probably wrong, but hey ...)
3067 w2 = (Elf_Word)value;
3070 # elif defined(ia64_HOST_ARCH)
3071 case R_IA64_DIR64LSB:
3072 case R_IA64_FPTR64LSB:
3075 case R_IA64_PCREL64LSB:
3078 case R_IA64_SEGREL64LSB:
3079 addr = findElfSegment(ehdrC, value);
3082 case R_IA64_GPREL22:
3083 ia64_reloc_gprel22(P, value);
3085 case R_IA64_LTOFF22:
3086 case R_IA64_LTOFF22X:
3087 case R_IA64_LTOFF_FPTR22:
3088 addr = allocateGOTEntry(value);
3089 ia64_reloc_gprel22(P, addr);
3091 case R_IA64_PCREL21B:
3092 ia64_reloc_pcrel21(P, S, oc);
3095 /* This goes with R_IA64_LTOFF22X and points to the load to
3096 * convert into a move. We don't implement relaxation. */
3098 # elif defined(powerpc_HOST_ARCH)
3099 case R_PPC_ADDR16_LO:
3100 *(Elf32_Half*) P = value;
3103 case R_PPC_ADDR16_HI:
3104 *(Elf32_Half*) P = value >> 16;
3107 case R_PPC_ADDR16_HA:
3108 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3112 *(Elf32_Word *) P = value;
3116 *(Elf32_Word *) P = value - P;
3122 if( delta << 6 >> 6 != delta )
3124 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3127 if( value == 0 || delta << 6 >> 6 != delta )
3129 barf( "Unable to make ppcJumpIsland for #%d",
3135 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3136 | (delta & 0x3fffffc);
3141 *(Elf64_Xword *)P = value;
3145 *(Elf64_Word *)P = (Elf64_Word) (value - P);
3149 *(Elf64_Word *)P = (Elf64_Word)value;
3153 *(Elf64_Sword *)P = (Elf64_Sword)value;
3157 errorBelch("%s: unhandled ELF relocation(RelA) type %d\n",
3158 oc->fileName, ELF_R_TYPE(info));
3167 ocResolve_ELF ( ObjectCode* oc )
3171 Elf_Sym* stab = NULL;
3172 char* ehdrC = (char*)(oc->image);
3173 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3174 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3176 /* first find "the" symbol table */
3177 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3179 /* also go find the string table */
3180 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3182 if (stab == NULL || strtab == NULL) {
3183 errorBelch("%s: can't find string or symbol table", oc->fileName);
3187 /* Process the relocation sections. */
3188 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3189 if (shdr[shnum].sh_type == SHT_REL) {
3190 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3191 shnum, stab, strtab );
3195 if (shdr[shnum].sh_type == SHT_RELA) {
3196 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3197 shnum, stab, strtab );
3202 /* Free the local symbol table; we won't need it again. */
3203 freeHashTable(oc->lochash, NULL);
3206 #if defined(powerpc_HOST_ARCH)
3207 ocFlushInstructionCache( oc );
3215 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3216 * at the front. The following utility functions pack and unpack instructions, and
3217 * take care of the most common relocations.
3220 #ifdef ia64_HOST_ARCH
3223 ia64_extract_instruction(Elf64_Xword *target)
3226 int slot = (Elf_Addr)target & 3;
3227 (Elf_Addr)target &= ~3;
3235 return ((w1 >> 5) & 0x1ffffffffff);
3237 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3241 barf("ia64_extract_instruction: invalid slot %p", target);
3246 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3248 int slot = (Elf_Addr)target & 3;
3249 (Elf_Addr)target &= ~3;
3254 *target |= value << 5;
3257 *target |= value << 46;
3258 *(target+1) |= value >> 18;
3261 *(target+1) |= value << 23;
3267 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3269 Elf64_Xword instruction;
3270 Elf64_Sxword rel_value;
3272 rel_value = value - gp_val;
3273 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3274 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3276 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3277 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3278 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3279 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3280 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3281 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3285 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3287 Elf64_Xword instruction;
3288 Elf64_Sxword rel_value;
3291 entry = allocatePLTEntry(value, oc);
3293 rel_value = (entry >> 4) - (target >> 4);
3294 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3295 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3297 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3298 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3299 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3300 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3306 * PowerPC ELF specifics
3309 #ifdef powerpc_HOST_ARCH
3311 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3317 ehdr = (Elf_Ehdr *) oc->image;
3318 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3320 for( i = 0; i < ehdr->e_shnum; i++ )
3321 if( shdr[i].sh_type == SHT_SYMTAB )
3324 if( i == ehdr->e_shnum )
3326 errorBelch( "This ELF file contains no symtab" );
3330 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3332 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3333 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3338 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3341 #endif /* powerpc */
3345 /* --------------------------------------------------------------------------
3347 * ------------------------------------------------------------------------*/
3349 #if defined(OBJFORMAT_MACHO)
3352 Support for MachO linking on Darwin/MacOS X on PowerPC chips
3353 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3355 I hereby formally apologize for the hackish nature of this code.
3356 Things that need to be done:
3357 *) implement ocVerifyImage_MachO
3358 *) add still more sanity checks.
3361 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3363 struct mach_header *header = (struct mach_header *) oc->image;
3364 struct load_command *lc = (struct load_command *) (header + 1);
3367 for( i = 0; i < header->ncmds; i++ )
3369 if( lc->cmd == LC_SYMTAB )
3371 // Find out the first and last undefined external
3372 // symbol, so we don't have to allocate too many
3374 struct symtab_command *symLC = (struct symtab_command *) lc;
3375 unsigned min = symLC->nsyms, max = 0;
3376 struct nlist *nlist =
3377 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3379 for(i=0;i<symLC->nsyms;i++)
3381 if(nlist[i].n_type & N_STAB)
3383 else if(nlist[i].n_type & N_EXT)
3385 if((nlist[i].n_type & N_TYPE) == N_UNDF
3386 && (nlist[i].n_value == 0))
3396 return ocAllocateJumpIslands(oc, max - min + 1, min);
3401 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3403 return ocAllocateJumpIslands(oc,0,0);
3406 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3408 // FIXME: do some verifying here
3412 static int resolveImports(
3415 struct symtab_command *symLC,
3416 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3417 unsigned long *indirectSyms,
3418 struct nlist *nlist)
3422 for(i=0;i*4<sect->size;i++)
3424 // according to otool, reserved1 contains the first index into the indirect symbol table
3425 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3426 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3429 if((symbol->n_type & N_TYPE) == N_UNDF
3430 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3431 addr = (void*) (symbol->n_value);
3432 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3435 addr = lookupSymbol(nm);
3438 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3442 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3443 ((void**)(image + sect->offset))[i] = addr;
3449 static unsigned long relocateAddress(
3452 struct section* sections,
3453 unsigned long address)
3456 for(i = 0; i < nSections; i++)
3458 if(sections[i].addr <= address
3459 && address < sections[i].addr + sections[i].size)
3461 return (unsigned long)oc->image
3462 + sections[i].offset + address - sections[i].addr;
3465 barf("Invalid Mach-O file:"
3466 "Address out of bounds while relocating object file");
3470 static int relocateSection(
3473 struct symtab_command *symLC, struct nlist *nlist,
3474 int nSections, struct section* sections, struct section *sect)
3476 struct relocation_info *relocs;
3479 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3481 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3485 relocs = (struct relocation_info*) (image + sect->reloff);
3489 if(relocs[i].r_address & R_SCATTERED)
3491 struct scattered_relocation_info *scat =
3492 (struct scattered_relocation_info*) &relocs[i];
3496 if(scat->r_length == 2)
3498 unsigned long word = 0;
3499 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3500 checkProddableBlock(oc,wordPtr);
3502 // Step 1: Figure out what the relocated value should be
3503 if(scat->r_type == GENERIC_RELOC_VANILLA)
3505 word = *wordPtr + (unsigned long) relocateAddress(
3512 else if(scat->r_type == PPC_RELOC_SECTDIFF
3513 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3514 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3515 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3517 struct scattered_relocation_info *pair =
3518 (struct scattered_relocation_info*) &relocs[i+1];
3520 if(!pair->r_scattered || pair->r_type != PPC_RELOC_PAIR)
3521 barf("Invalid Mach-O file: "
3522 "PPC_RELOC_*_SECTDIFF not followed by PPC_RELOC_PAIR");
3524 word = (unsigned long)
3525 (relocateAddress(oc, nSections, sections, scat->r_value)
3526 - relocateAddress(oc, nSections, sections, pair->r_value));
3529 else if(scat->r_type == PPC_RELOC_HI16
3530 || scat->r_type == PPC_RELOC_LO16
3531 || scat->r_type == PPC_RELOC_HA16
3532 || scat->r_type == PPC_RELOC_LO14)
3533 { // these are generated by label+offset things
3534 struct relocation_info *pair = &relocs[i+1];
3535 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3536 barf("Invalid Mach-O file: "
3537 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3539 if(scat->r_type == PPC_RELOC_LO16)
3541 word = ((unsigned short*) wordPtr)[1];
3542 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3544 else if(scat->r_type == PPC_RELOC_LO14)
3546 barf("Unsupported Relocation: PPC_RELOC_LO14");
3547 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3548 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3550 else if(scat->r_type == PPC_RELOC_HI16)
3552 word = ((unsigned short*) wordPtr)[1] << 16;
3553 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3555 else if(scat->r_type == PPC_RELOC_HA16)
3557 word = ((unsigned short*) wordPtr)[1] << 16;
3558 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3562 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
3568 continue; // ignore the others
3570 if(scat->r_type == GENERIC_RELOC_VANILLA
3571 || scat->r_type == PPC_RELOC_SECTDIFF)
3575 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
3577 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3579 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
3581 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3583 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
3585 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3586 + ((word & (1<<15)) ? 1 : 0);
3591 continue; // FIXME: I hope it's OK to ignore all the others.
3595 struct relocation_info *reloc = &relocs[i];
3596 if(reloc->r_pcrel && !reloc->r_extern)
3599 if(reloc->r_length == 2)
3601 unsigned long word = 0;
3602 unsigned long jumpIsland = 0;
3603 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
3604 // to avoid warning and to catch
3607 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
3608 checkProddableBlock(oc,wordPtr);
3610 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3614 else if(reloc->r_type == PPC_RELOC_LO16)
3616 word = ((unsigned short*) wordPtr)[1];
3617 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3619 else if(reloc->r_type == PPC_RELOC_HI16)
3621 word = ((unsigned short*) wordPtr)[1] << 16;
3622 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3624 else if(reloc->r_type == PPC_RELOC_HA16)
3626 word = ((unsigned short*) wordPtr)[1] << 16;
3627 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3629 else if(reloc->r_type == PPC_RELOC_BR24)
3632 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
3636 if(!reloc->r_extern)
3639 sections[reloc->r_symbolnum-1].offset
3640 - sections[reloc->r_symbolnum-1].addr
3647 struct nlist *symbol = &nlist[reloc->r_symbolnum];
3648 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3649 void *symbolAddress = lookupSymbol(nm);
3652 errorBelch("\nunknown symbol `%s'", nm);
3658 // In the .o file, this should be a relative jump to NULL
3659 // and we'll change it to a jump to a relative jump to the symbol
3660 ASSERT(-word == reloc->r_address);
3661 word = (unsigned long) symbolAddress;
3662 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,word);
3663 word -= ((long)image) + sect->offset + reloc->r_address;
3666 offsetToJumpIsland = jumpIsland
3667 - (((long)image) + sect->offset + reloc->r_address);
3672 word += (unsigned long) symbolAddress;
3676 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3681 else if(reloc->r_type == PPC_RELOC_LO16)
3683 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3686 else if(reloc->r_type == PPC_RELOC_HI16)
3688 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3691 else if(reloc->r_type == PPC_RELOC_HA16)
3693 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3694 + ((word & (1<<15)) ? 1 : 0);
3697 else if(reloc->r_type == PPC_RELOC_BR24)
3699 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3701 // The branch offset is too large.
3702 // Therefore, we try to use a jump island.
3705 barf("unconditional relative branch out of range: "
3706 "no jump island available");
3709 word = offsetToJumpIsland;
3710 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3711 barf("unconditional relative branch out of range: "
3712 "jump island out of range");
3714 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
3718 barf("\nunknown relocation %d",reloc->r_type);
3725 static int ocGetNames_MachO(ObjectCode* oc)
3727 char *image = (char*) oc->image;
3728 struct mach_header *header = (struct mach_header*) image;
3729 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3730 unsigned i,curSymbol = 0;
3731 struct segment_command *segLC = NULL;
3732 struct section *sections;
3733 struct symtab_command *symLC = NULL;
3734 struct nlist *nlist;
3735 unsigned long commonSize = 0;
3736 char *commonStorage = NULL;
3737 unsigned long commonCounter;
3739 for(i=0;i<header->ncmds;i++)
3741 if(lc->cmd == LC_SEGMENT)
3742 segLC = (struct segment_command*) lc;
3743 else if(lc->cmd == LC_SYMTAB)
3744 symLC = (struct symtab_command*) lc;
3745 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3748 sections = (struct section*) (segLC+1);
3749 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3752 for(i=0;i<segLC->nsects;i++)
3754 if(sections[i].size == 0)
3757 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
3759 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
3760 "ocGetNames_MachO(common symbols)");
3761 sections[i].offset = zeroFillArea - image;
3764 if(!strcmp(sections[i].sectname,"__text"))
3765 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
3766 (void*) (image + sections[i].offset),
3767 (void*) (image + sections[i].offset + sections[i].size));
3768 else if(!strcmp(sections[i].sectname,"__const"))
3769 addSection(oc, SECTIONKIND_RWDATA,
3770 (void*) (image + sections[i].offset),
3771 (void*) (image + sections[i].offset + sections[i].size));
3772 else if(!strcmp(sections[i].sectname,"__data"))
3773 addSection(oc, SECTIONKIND_RWDATA,
3774 (void*) (image + sections[i].offset),
3775 (void*) (image + sections[i].offset + sections[i].size));
3776 else if(!strcmp(sections[i].sectname,"__bss")
3777 || !strcmp(sections[i].sectname,"__common"))
3778 addSection(oc, SECTIONKIND_RWDATA,
3779 (void*) (image + sections[i].offset),
3780 (void*) (image + sections[i].offset + sections[i].size));
3782 addProddableBlock(oc, (void*) (image + sections[i].offset),
3786 // count external symbols defined here
3790 for(i=0;i<symLC->nsyms;i++)
3792 if(nlist[i].n_type & N_STAB)
3794 else if(nlist[i].n_type & N_EXT)
3796 if((nlist[i].n_type & N_TYPE) == N_UNDF
3797 && (nlist[i].n_value != 0))
3799 commonSize += nlist[i].n_value;
3802 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3807 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3808 "ocGetNames_MachO(oc->symbols)");
3812 for(i=0;i<symLC->nsyms;i++)
3814 if(nlist[i].n_type & N_STAB)
3816 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3818 if(nlist[i].n_type & N_EXT)
3820 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3821 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3823 + sections[nlist[i].n_sect-1].offset
3824 - sections[nlist[i].n_sect-1].addr
3825 + nlist[i].n_value);
3826 oc->symbols[curSymbol++] = nm;
3830 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3831 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
3833 + sections[nlist[i].n_sect-1].offset
3834 - sections[nlist[i].n_sect-1].addr
3835 + nlist[i].n_value);
3841 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
3842 commonCounter = (unsigned long)commonStorage;
3845 for(i=0;i<symLC->nsyms;i++)
3847 if((nlist[i].n_type & N_TYPE) == N_UNDF
3848 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
3850 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3851 unsigned long sz = nlist[i].n_value;
3853 nlist[i].n_value = commonCounter;
3855 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3856 (void*)commonCounter);
3857 oc->symbols[curSymbol++] = nm;
3859 commonCounter += sz;
3866 static int ocResolve_MachO(ObjectCode* oc)
3868 char *image = (char*) oc->image;
3869 struct mach_header *header = (struct mach_header*) image;
3870 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3872 struct segment_command *segLC = NULL;
3873 struct section *sections, *la_ptrs = NULL, *nl_ptrs = NULL;
3874 struct symtab_command *symLC = NULL;
3875 struct dysymtab_command *dsymLC = NULL;
3876 struct nlist *nlist;
3878 for(i=0;i<header->ncmds;i++)
3880 if(lc->cmd == LC_SEGMENT)
3881 segLC = (struct segment_command*) lc;
3882 else if(lc->cmd == LC_SYMTAB)
3883 symLC = (struct symtab_command*) lc;
3884 else if(lc->cmd == LC_DYSYMTAB)
3885 dsymLC = (struct dysymtab_command*) lc;
3886 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3889 sections = (struct section*) (segLC+1);
3890 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3893 for(i=0;i<segLC->nsects;i++)
3895 if(!strcmp(sections[i].sectname,"__la_symbol_ptr"))
3896 la_ptrs = §ions[i];
3897 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr"))
3898 nl_ptrs = §ions[i];
3903 unsigned long *indirectSyms
3904 = (unsigned long*) (image + dsymLC->indirectsymoff);
3907 if(!resolveImports(oc,image,symLC,la_ptrs,indirectSyms,nlist))
3910 if(!resolveImports(oc,image,symLC,nl_ptrs,indirectSyms,nlist))
3914 for(i=0;i<segLC->nsects;i++)
3916 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
3920 /* Free the local symbol table; we won't need it again. */
3921 freeHashTable(oc->lochash, NULL);
3924 #if defined (powerpc_HOST_ARCH)
3925 ocFlushInstructionCache( oc );
3932 * The Mach-O object format uses leading underscores. But not everywhere.
3933 * There is a small number of runtime support functions defined in
3934 * libcc_dynamic.a whose name does not have a leading underscore.
3935 * As a consequence, we can't get their address from C code.
3936 * We have to use inline assembler just to take the address of a function.
3940 static void machoInitSymbolsWithoutUnderscore()
3942 extern void* symbolsWithoutUnderscore[];
3943 void **p = symbolsWithoutUnderscore;
3944 __asm__ volatile(".data\n_symbolsWithoutUnderscore:");
3948 __asm__ volatile(".long " # x);
3950 RTS_MACHO_NOUNDERLINE_SYMBOLS
3952 __asm__ volatile(".text");
3956 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
3958 RTS_MACHO_NOUNDERLINE_SYMBOLS