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(rts_InstallConsoleEvent) \
300 SymX(rts_ConsoleHandlerDone) \
302 Sym(_imp___timezone) \
311 RTS_MINGW_EXTRA_SYMS \
316 # define MAIN_CAP_SYM SymX(MainCapability)
318 # define MAIN_CAP_SYM
321 #if !defined(mingw32_HOST_OS)
322 #define RTS_USER_SIGNALS_SYMBOLS \
323 SymX(startSignalHandler) \
324 SymX(setIOManagerPipe)
326 #define RTS_USER_SIGNALS_SYMBOLS /* nothing */
329 #ifdef TABLES_NEXT_TO_CODE
330 #define RTS_RET_SYMBOLS /* nothing */
332 #define RTS_RET_SYMBOLS \
333 SymX(stg_enter_ret) \
334 SymX(stg_gc_fun_ret) \
342 SymX(stg_ap_pv_ret) \
343 SymX(stg_ap_pp_ret) \
344 SymX(stg_ap_ppv_ret) \
345 SymX(stg_ap_ppp_ret) \
346 SymX(stg_ap_pppv_ret) \
347 SymX(stg_ap_pppp_ret) \
348 SymX(stg_ap_ppppp_ret) \
349 SymX(stg_ap_pppppp_ret)
352 #define RTS_SYMBOLS \
356 SymX(stg_enter_info) \
357 SymX(stg_gc_void_info) \
358 SymX(__stg_gc_enter_1) \
359 SymX(stg_gc_noregs) \
360 SymX(stg_gc_unpt_r1_info) \
361 SymX(stg_gc_unpt_r1) \
362 SymX(stg_gc_unbx_r1_info) \
363 SymX(stg_gc_unbx_r1) \
364 SymX(stg_gc_f1_info) \
366 SymX(stg_gc_d1_info) \
368 SymX(stg_gc_l1_info) \
371 SymX(stg_gc_fun_info) \
373 SymX(stg_gc_gen_info) \
374 SymX(stg_gc_gen_hp) \
376 SymX(stg_gen_yield) \
377 SymX(stg_yield_noregs) \
378 SymX(stg_yield_to_interpreter) \
379 SymX(stg_gen_block) \
380 SymX(stg_block_noregs) \
382 SymX(stg_block_takemvar) \
383 SymX(stg_block_putmvar) \
384 SymX(stg_seq_frame_info) \
386 SymX(MallocFailHook) \
388 SymX(OutOfHeapHook) \
389 SymX(StackOverflowHook) \
390 SymX(__encodeDouble) \
391 SymX(__encodeFloat) \
395 SymX(__gmpz_cmp_si) \
396 SymX(__gmpz_cmp_ui) \
397 SymX(__gmpz_get_si) \
398 SymX(__gmpz_get_ui) \
399 SymX(__int_encodeDouble) \
400 SymX(__int_encodeFloat) \
401 SymX(andIntegerzh_fast) \
402 SymX(atomicallyzh_fast) \
406 SymX(blockAsyncExceptionszh_fast) \
408 SymX(catchRetryzh_fast) \
409 SymX(catchSTMzh_fast) \
410 SymX(closure_flags) \
412 SymX(cmpIntegerzh_fast) \
413 SymX(cmpIntegerIntzh_fast) \
414 SymX(complementIntegerzh_fast) \
415 SymX(createAdjustor) \
416 SymX(decodeDoublezh_fast) \
417 SymX(decodeFloatzh_fast) \
420 SymX(deRefWeakzh_fast) \
421 SymX(deRefStablePtrzh_fast) \
422 SymX(divExactIntegerzh_fast) \
423 SymX(divModIntegerzh_fast) \
426 SymX(forkOS_createThread) \
427 SymX(freeHaskellFunctionPtr) \
428 SymX(freeStablePtr) \
429 SymX(gcdIntegerzh_fast) \
430 SymX(gcdIntegerIntzh_fast) \
431 SymX(gcdIntzh_fast) \
440 SymX(hs_perform_gc) \
441 SymX(hs_free_stable_ptr) \
442 SymX(hs_free_fun_ptr) \
444 SymX(int2Integerzh_fast) \
445 SymX(integer2Intzh_fast) \
446 SymX(integer2Wordzh_fast) \
447 SymX(isCurrentThreadBoundzh_fast) \
448 SymX(isDoubleDenormalized) \
449 SymX(isDoubleInfinite) \
451 SymX(isDoubleNegativeZero) \
452 SymX(isEmptyMVarzh_fast) \
453 SymX(isFloatDenormalized) \
454 SymX(isFloatInfinite) \
456 SymX(isFloatNegativeZero) \
457 SymX(killThreadzh_fast) \
460 SymX(makeStablePtrzh_fast) \
461 SymX(minusIntegerzh_fast) \
462 SymX(mkApUpd0zh_fast) \
463 SymX(myThreadIdzh_fast) \
464 SymX(labelThreadzh_fast) \
465 SymX(newArrayzh_fast) \
466 SymX(newBCOzh_fast) \
467 SymX(newByteArrayzh_fast) \
468 SymX_redirect(newCAF, newDynCAF) \
469 SymX(newMVarzh_fast) \
470 SymX(newMutVarzh_fast) \
471 SymX(newTVarzh_fast) \
472 SymX(atomicModifyMutVarzh_fast) \
473 SymX(newPinnedByteArrayzh_fast) \
474 SymX(orIntegerzh_fast) \
476 SymX(performMajorGC) \
477 SymX(plusIntegerzh_fast) \
480 SymX(putMVarzh_fast) \
481 SymX(quotIntegerzh_fast) \
482 SymX(quotRemIntegerzh_fast) \
484 SymX(raiseIOzh_fast) \
485 SymX(readTVarzh_fast) \
486 SymX(remIntegerzh_fast) \
487 SymX(resetNonBlockingFd) \
492 SymX(rts_checkSchedStatus) \
495 SymX(rts_evalLazyIO) \
496 SymX(rts_evalStableIO) \
500 SymX(rts_getDouble) \
505 SymX(rts_getFunPtr) \
506 SymX(rts_getStablePtr) \
507 SymX(rts_getThreadId) \
509 SymX(rts_getWord32) \
522 SymX(rts_mkStablePtr) \
530 SymX(rtsSupportsBoundThreads) \
532 SymX(__hscore_get_saved_termios) \
533 SymX(__hscore_set_saved_termios) \
535 SymX(startupHaskell) \
536 SymX(shutdownHaskell) \
537 SymX(shutdownHaskellAndExit) \
538 SymX(stable_ptr_table) \
539 SymX(stackOverflow) \
540 SymX(stg_CAF_BLACKHOLE_info) \
541 SymX(awakenBlockedQueue) \
542 SymX(stg_CHARLIKE_closure) \
543 SymX(stg_EMPTY_MVAR_info) \
544 SymX(stg_IND_STATIC_info) \
545 SymX(stg_INTLIKE_closure) \
546 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
547 SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \
548 SymX(stg_WEAK_info) \
549 SymX(stg_ap_0_info) \
550 SymX(stg_ap_v_info) \
551 SymX(stg_ap_f_info) \
552 SymX(stg_ap_d_info) \
553 SymX(stg_ap_l_info) \
554 SymX(stg_ap_n_info) \
555 SymX(stg_ap_p_info) \
556 SymX(stg_ap_pv_info) \
557 SymX(stg_ap_pp_info) \
558 SymX(stg_ap_ppv_info) \
559 SymX(stg_ap_ppp_info) \
560 SymX(stg_ap_pppv_info) \
561 SymX(stg_ap_pppp_info) \
562 SymX(stg_ap_ppppp_info) \
563 SymX(stg_ap_pppppp_info) \
564 SymX(stg_ap_1_upd_info) \
565 SymX(stg_ap_2_upd_info) \
566 SymX(stg_ap_3_upd_info) \
567 SymX(stg_ap_4_upd_info) \
568 SymX(stg_ap_5_upd_info) \
569 SymX(stg_ap_6_upd_info) \
570 SymX(stg_ap_7_upd_info) \
572 SymX(stg_sel_0_upd_info) \
573 SymX(stg_sel_10_upd_info) \
574 SymX(stg_sel_11_upd_info) \
575 SymX(stg_sel_12_upd_info) \
576 SymX(stg_sel_13_upd_info) \
577 SymX(stg_sel_14_upd_info) \
578 SymX(stg_sel_15_upd_info) \
579 SymX(stg_sel_1_upd_info) \
580 SymX(stg_sel_2_upd_info) \
581 SymX(stg_sel_3_upd_info) \
582 SymX(stg_sel_4_upd_info) \
583 SymX(stg_sel_5_upd_info) \
584 SymX(stg_sel_6_upd_info) \
585 SymX(stg_sel_7_upd_info) \
586 SymX(stg_sel_8_upd_info) \
587 SymX(stg_sel_9_upd_info) \
588 SymX(stg_upd_frame_info) \
589 SymX(suspendThread) \
590 SymX(takeMVarzh_fast) \
591 SymX(timesIntegerzh_fast) \
592 SymX(tryPutMVarzh_fast) \
593 SymX(tryTakeMVarzh_fast) \
594 SymX(unblockAsyncExceptionszh_fast) \
596 SymX(unsafeThawArrayzh_fast) \
597 SymX(waitReadzh_fast) \
598 SymX(waitWritezh_fast) \
599 SymX(word2Integerzh_fast) \
600 SymX(writeTVarzh_fast) \
601 SymX(xorIntegerzh_fast) \
603 RTS_USER_SIGNALS_SYMBOLS
605 #ifdef SUPPORT_LONG_LONGS
606 #define RTS_LONG_LONG_SYMS \
607 SymX(int64ToIntegerzh_fast) \
608 SymX(word64ToIntegerzh_fast)
610 #define RTS_LONG_LONG_SYMS /* nothing */
613 // 64-bit support functions in libgcc.a
614 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
615 #define RTS_LIBGCC_SYMBOLS \
625 #elif defined(ia64_HOST_ARCH)
626 #define RTS_LIBGCC_SYMBOLS \
634 #define RTS_LIBGCC_SYMBOLS
637 #ifdef darwin_HOST_OS
638 // Symbols that don't have a leading underscore
639 // on Mac OS X. They have to receive special treatment,
640 // see machoInitSymbolsWithoutUnderscore()
641 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
646 /* entirely bogus claims about types of these symbols */
647 #define Sym(vvv) extern void vvv(void);
648 #define SymX(vvv) /**/
649 #define SymX_redirect(vvv,xxx) /**/
653 RTS_POSIX_ONLY_SYMBOLS
654 RTS_MINGW_ONLY_SYMBOLS
655 RTS_CYGWIN_ONLY_SYMBOLS
661 #ifdef LEADING_UNDERSCORE
662 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
664 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
667 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
669 #define SymX(vvv) Sym(vvv)
671 // SymX_redirect allows us to redirect references to one symbol to
672 // another symbol. See newCAF/newDynCAF for an example.
673 #define SymX_redirect(vvv,xxx) \
674 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
677 static RtsSymbolVal rtsSyms[] = {
681 RTS_POSIX_ONLY_SYMBOLS
682 RTS_MINGW_ONLY_SYMBOLS
683 RTS_CYGWIN_ONLY_SYMBOLS
685 { 0, 0 } /* sentinel */
688 /* -----------------------------------------------------------------------------
689 * Insert symbols into hash tables, checking for duplicates.
691 static void ghciInsertStrHashTable ( char* obj_name,
697 if (lookupHashTable(table, (StgWord)key) == NULL)
699 insertStrHashTable(table, (StgWord)key, data);
704 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
706 "whilst processing object file\n"
708 "This could be caused by:\n"
709 " * Loading two different object files which export the same symbol\n"
710 " * Specifying the same object file twice on the GHCi command line\n"
711 " * An incorrect `package.conf' entry, causing some object to be\n"
713 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
722 /* -----------------------------------------------------------------------------
723 * initialize the object linker
727 static int linker_init_done = 0 ;
729 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
730 static void *dl_prog_handle;
733 /* dlopen(NULL,..) doesn't work so we grab libc explicitly */
734 #if defined(openbsd_HOST_OS)
735 static void *dl_libc_handle;
743 /* Make initLinker idempotent, so we can call it
744 before evey relevant operation; that means we
745 don't need to initialise the linker separately */
746 if (linker_init_done == 1) { return; } else {
747 linker_init_done = 1;
750 symhash = allocStrHashTable();
752 /* populate the symbol table with stuff from the RTS */
753 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
754 ghciInsertStrHashTable("(GHCi built-in symbols)",
755 symhash, sym->lbl, sym->addr);
757 # if defined(OBJFORMAT_MACHO)
758 machoInitSymbolsWithoutUnderscore();
761 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
762 # if defined(RTLD_DEFAULT)
763 dl_prog_handle = RTLD_DEFAULT;
765 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
766 # if defined(openbsd_HOST_OS)
767 dl_libc_handle = dlopen("libc.so", RTLD_LAZY);
769 # endif /* RTLD_DEFAULT */
773 /* -----------------------------------------------------------------------------
774 * Loading DLL or .so dynamic libraries
775 * -----------------------------------------------------------------------------
777 * Add a DLL from which symbols may be found. In the ELF case, just
778 * do RTLD_GLOBAL-style add, so no further messing around needs to
779 * happen in order that symbols in the loaded .so are findable --
780 * lookupSymbol() will subsequently see them by dlsym on the program's
781 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
783 * In the PEi386 case, open the DLLs and put handles to them in a
784 * linked list. When looking for a symbol, try all handles in the
785 * list. This means that we need to load even DLLs that are guaranteed
786 * to be in the ghc.exe image already, just so we can get a handle
787 * to give to loadSymbol, so that we can find the symbols. For such
788 * libraries, the LoadLibrary call should be a no-op except for returning
793 #if defined(OBJFORMAT_PEi386)
794 /* A record for storing handles into DLLs. */
799 struct _OpenedDLL* next;
804 /* A list thereof. */
805 static OpenedDLL* opened_dlls = NULL;
809 addDLL( char *dll_name )
811 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
812 /* ------------------- ELF DLL loader ------------------- */
818 hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
821 /* dlopen failed; return a ptr to the error msg. */
823 if (errmsg == NULL) errmsg = "addDLL: unknown error";
830 # elif defined(OBJFORMAT_PEi386)
831 /* ------------------- Win32 DLL loader ------------------- */
839 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
841 /* See if we've already got it, and ignore if so. */
842 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
843 if (0 == strcmp(o_dll->name, dll_name))
847 /* The file name has no suffix (yet) so that we can try
848 both foo.dll and foo.drv
850 The documentation for LoadLibrary says:
851 If no file name extension is specified in the lpFileName
852 parameter, the default library extension .dll is
853 appended. However, the file name string can include a trailing
854 point character (.) to indicate that the module name has no
857 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
858 sprintf(buf, "%s.DLL", dll_name);
859 instance = LoadLibrary(buf);
860 if (instance == NULL) {
861 sprintf(buf, "%s.DRV", dll_name); // KAA: allow loading of drivers (like winspool.drv)
862 instance = LoadLibrary(buf);
863 if (instance == NULL) {
866 /* LoadLibrary failed; return a ptr to the error msg. */
867 return "addDLL: unknown error";
872 /* Add this DLL to the list of DLLs in which to search for symbols. */
873 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
874 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
875 strcpy(o_dll->name, dll_name);
876 o_dll->instance = instance;
877 o_dll->next = opened_dlls;
882 barf("addDLL: not implemented on this platform");
886 /* -----------------------------------------------------------------------------
887 * lookup a symbol in the hash table
890 lookupSymbol( char *lbl )
894 ASSERT(symhash != NULL);
895 val = lookupStrHashTable(symhash, lbl);
898 # if defined(OBJFORMAT_ELF)
899 # if defined(openbsd_HOST_OS)
900 val = dlsym(dl_prog_handle, lbl);
901 return (val != NULL) ? val : dlsym(dl_libc_handle,lbl);
902 # else /* not openbsd */
903 return dlsym(dl_prog_handle, lbl);
905 # elif defined(OBJFORMAT_MACHO)
906 if(NSIsSymbolNameDefined(lbl)) {
907 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
908 return NSAddressOfSymbol(symbol);
912 # elif defined(OBJFORMAT_PEi386)
915 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
916 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
918 /* HACK: if the name has an initial underscore, try stripping
919 it off & look that up first. I've yet to verify whether there's
920 a Rule that governs whether an initial '_' *should always* be
921 stripped off when mapping from import lib name to the DLL name.
923 sym = GetProcAddress(o_dll->instance, (lbl+1));
925 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
929 sym = GetProcAddress(o_dll->instance, lbl);
931 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
946 __attribute((unused))
948 lookupLocalSymbol( ObjectCode* oc, char *lbl )
952 val = lookupStrHashTable(oc->lochash, lbl);
962 /* -----------------------------------------------------------------------------
963 * Debugging aid: look in GHCi's object symbol tables for symbols
964 * within DELTA bytes of the specified address, and show their names.
967 void ghci_enquire ( char* addr );
969 void ghci_enquire ( char* addr )
974 const int DELTA = 64;
979 for (oc = objects; oc; oc = oc->next) {
980 for (i = 0; i < oc->n_symbols; i++) {
981 sym = oc->symbols[i];
982 if (sym == NULL) continue;
983 // debugBelch("enquire %p %p\n", sym, oc->lochash);
985 if (oc->lochash != NULL) {
986 a = lookupStrHashTable(oc->lochash, sym);
989 a = lookupStrHashTable(symhash, sym);
992 // debugBelch("ghci_enquire: can't find %s\n", sym);
994 else if (addr-DELTA <= a && a <= addr+DELTA) {
995 debugBelch("%p + %3d == `%s'\n", addr, a - addr, sym);
1002 #ifdef ia64_HOST_ARCH
1003 static unsigned int PLTSize(void);
1006 /* -----------------------------------------------------------------------------
1007 * Load an obj (populate the global symbol table, but don't resolve yet)
1009 * Returns: 1 if ok, 0 on error.
1012 loadObj( char *path )
1019 void *map_addr = NULL;
1026 /* debugBelch("loadObj %s\n", path ); */
1028 /* Check that we haven't already loaded this object.
1029 Ignore requests to load multiple times */
1033 for (o = objects; o; o = o->next) {
1034 if (0 == strcmp(o->fileName, path)) {
1036 break; /* don't need to search further */
1040 IF_DEBUG(linker, debugBelch(
1041 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1042 "same object file twice:\n"
1044 "GHCi will ignore this, but be warned.\n"
1046 return 1; /* success */
1050 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1052 # if defined(OBJFORMAT_ELF)
1053 oc->formatName = "ELF";
1054 # elif defined(OBJFORMAT_PEi386)
1055 oc->formatName = "PEi386";
1056 # elif defined(OBJFORMAT_MACHO)
1057 oc->formatName = "Mach-O";
1060 barf("loadObj: not implemented on this platform");
1063 r = stat(path, &st);
1064 if (r == -1) { return 0; }
1066 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1067 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1068 strcpy(oc->fileName, path);
1070 oc->fileSize = st.st_size;
1072 oc->sections = NULL;
1073 oc->lochash = allocStrHashTable();
1074 oc->proddables = NULL;
1076 /* chain it onto the list of objects */
1081 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1083 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1085 #if defined(openbsd_HOST_OS)
1086 fd = open(path, O_RDONLY, S_IRUSR);
1088 fd = open(path, O_RDONLY);
1091 barf("loadObj: can't open `%s'", path);
1093 pagesize = getpagesize();
1095 #ifdef ia64_HOST_ARCH
1096 /* The PLT needs to be right before the object */
1097 n = ROUND_UP(PLTSize(), pagesize);
1098 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1099 if (oc->plt == MAP_FAILED)
1100 barf("loadObj: can't allocate PLT");
1103 map_addr = oc->plt + n;
1106 n = ROUND_UP(oc->fileSize, pagesize);
1108 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
1109 * small memory model on this architecture (see gcc docs,
1112 #ifdef x86_64_HOST_ARCH
1113 #define EXTRA_MAP_FLAGS MAP_32BIT
1115 #define EXTRA_MAP_FLAGS 0
1118 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1119 MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0);
1120 if (oc->image == MAP_FAILED)
1121 barf("loadObj: can't map `%s'", path);
1125 #else /* !USE_MMAP */
1127 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1129 /* load the image into memory */
1130 f = fopen(path, "rb");
1132 barf("loadObj: can't read `%s'", path);
1134 n = fread ( oc->image, 1, oc->fileSize, f );
1135 if (n != oc->fileSize)
1136 barf("loadObj: error whilst reading `%s'", path);
1140 #endif /* USE_MMAP */
1142 # if defined(OBJFORMAT_MACHO)
1143 r = ocAllocateJumpIslands_MachO ( oc );
1144 if (!r) { return r; }
1145 # elif defined(OBJFORMAT_ELF) && defined(powerpc_HOST_ARCH)
1146 r = ocAllocateJumpIslands_ELF ( oc );
1147 if (!r) { return r; }
1150 /* verify the in-memory image */
1151 # if defined(OBJFORMAT_ELF)
1152 r = ocVerifyImage_ELF ( oc );
1153 # elif defined(OBJFORMAT_PEi386)
1154 r = ocVerifyImage_PEi386 ( oc );
1155 # elif defined(OBJFORMAT_MACHO)
1156 r = ocVerifyImage_MachO ( oc );
1158 barf("loadObj: no verify method");
1160 if (!r) { return r; }
1162 /* build the symbol list for this image */
1163 # if defined(OBJFORMAT_ELF)
1164 r = ocGetNames_ELF ( oc );
1165 # elif defined(OBJFORMAT_PEi386)
1166 r = ocGetNames_PEi386 ( oc );
1167 # elif defined(OBJFORMAT_MACHO)
1168 r = ocGetNames_MachO ( oc );
1170 barf("loadObj: no getNames method");
1172 if (!r) { return r; }
1174 /* loaded, but not resolved yet */
1175 oc->status = OBJECT_LOADED;
1180 /* -----------------------------------------------------------------------------
1181 * resolve all the currently unlinked objects in memory
1183 * Returns: 1 if ok, 0 on error.
1193 for (oc = objects; oc; oc = oc->next) {
1194 if (oc->status != OBJECT_RESOLVED) {
1195 # if defined(OBJFORMAT_ELF)
1196 r = ocResolve_ELF ( oc );
1197 # elif defined(OBJFORMAT_PEi386)
1198 r = ocResolve_PEi386 ( oc );
1199 # elif defined(OBJFORMAT_MACHO)
1200 r = ocResolve_MachO ( oc );
1202 barf("resolveObjs: not implemented on this platform");
1204 if (!r) { return r; }
1205 oc->status = OBJECT_RESOLVED;
1211 /* -----------------------------------------------------------------------------
1212 * delete an object from the pool
1215 unloadObj( char *path )
1217 ObjectCode *oc, *prev;
1219 ASSERT(symhash != NULL);
1220 ASSERT(objects != NULL);
1225 for (oc = objects; oc; prev = oc, oc = oc->next) {
1226 if (!strcmp(oc->fileName,path)) {
1228 /* Remove all the mappings for the symbols within this
1233 for (i = 0; i < oc->n_symbols; i++) {
1234 if (oc->symbols[i] != NULL) {
1235 removeStrHashTable(symhash, oc->symbols[i], NULL);
1243 prev->next = oc->next;
1246 /* We're going to leave this in place, in case there are
1247 any pointers from the heap into it: */
1248 /* stgFree(oc->image); */
1249 stgFree(oc->fileName);
1250 stgFree(oc->symbols);
1251 stgFree(oc->sections);
1252 /* The local hash table should have been freed at the end
1253 of the ocResolve_ call on it. */
1254 ASSERT(oc->lochash == NULL);
1260 errorBelch("unloadObj: can't find `%s' to unload", path);
1264 /* -----------------------------------------------------------------------------
1265 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1266 * which may be prodded during relocation, and abort if we try and write
1267 * outside any of these.
1269 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1272 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1273 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1277 pb->next = oc->proddables;
1278 oc->proddables = pb;
1281 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1284 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1285 char* s = (char*)(pb->start);
1286 char* e = s + pb->size - 1;
1287 char* a = (char*)addr;
1288 /* Assumes that the biggest fixup involves a 4-byte write. This
1289 probably needs to be changed to 8 (ie, +7) on 64-bit
1291 if (a >= s && (a+3) <= e) return;
1293 barf("checkProddableBlock: invalid fixup in runtime linker");
1296 /* -----------------------------------------------------------------------------
1297 * Section management.
1299 static void addSection ( ObjectCode* oc, SectionKind kind,
1300 void* start, void* end )
1302 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1306 s->next = oc->sections;
1309 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1310 start, ((char*)end)-1, end - start + 1, kind );
1315 /* --------------------------------------------------------------------------
1316 * PowerPC specifics (jump islands)
1317 * ------------------------------------------------------------------------*/
1319 #if defined(powerpc_HOST_ARCH)
1322 ocAllocateJumpIslands
1324 Allocate additional space at the end of the object file image to make room
1327 PowerPC relative branch instructions have a 24 bit displacement field.
1328 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1329 If a particular imported symbol is outside this range, we have to redirect
1330 the jump to a short piece of new code that just loads the 32bit absolute
1331 address and jumps there.
1332 This function just allocates space for one 16 byte ppcJumpIsland for every
1333 undefined symbol in the object file. The code for the islands is filled in by
1334 makeJumpIsland below.
1337 static int ocAllocateJumpIslands( ObjectCode* oc, int count, int first )
1346 // round up to the nearest 4
1347 aligned = (oc->fileSize + 3) & ~3;
1350 #ifndef linux_HOST_OS /* mremap is a linux extension */
1351 #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined
1354 pagesize = getpagesize();
1355 n = ROUND_UP( oc->fileSize, pagesize );
1356 m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize );
1358 /* The effect of this mremap() call is only the ensure that we have
1359 * a sufficient number of virtually contiguous pages. As returned from
1360 * mremap, the pages past the end of the file are not backed. We give
1361 * them a backing by using MAP_FIXED to map in anonymous pages.
1363 if( (oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE )) == MAP_FAILED )
1365 errorBelch( "Unable to mremap for Jump Islands\n" );
1369 if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC,
1370 MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED )
1372 errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" );
1377 oc->image = stgReallocBytes( oc->image,
1378 aligned + sizeof (ppcJumpIsland) * count,
1379 "ocAllocateJumpIslands" );
1380 #endif /* USE_MMAP */
1382 oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned);
1383 memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count );
1386 oc->jump_islands = NULL;
1388 oc->island_start_symbol = first;
1389 oc->n_islands = count;
1394 static unsigned long makeJumpIsland( ObjectCode* oc,
1395 unsigned long symbolNumber,
1396 unsigned long target )
1398 ppcJumpIsland *island;
1400 if( symbolNumber < oc->island_start_symbol ||
1401 symbolNumber - oc->island_start_symbol > oc->n_islands)
1404 island = &oc->jump_islands[symbolNumber - oc->island_start_symbol];
1406 // lis r12, hi16(target)
1407 island->lis_r12 = 0x3d80;
1408 island->hi_addr = target >> 16;
1410 // ori r12, r12, lo16(target)
1411 island->ori_r12_r12 = 0x618c;
1412 island->lo_addr = target & 0xffff;
1415 island->mtctr_r12 = 0x7d8903a6;
1418 island->bctr = 0x4e800420;
1420 return (unsigned long) island;
1424 ocFlushInstructionCache
1426 Flush the data & instruction caches.
1427 Because the PPC has split data/instruction caches, we have to
1428 do that whenever we modify code at runtime.
1431 static void ocFlushInstructionCache( ObjectCode *oc )
1433 int n = (oc->fileSize + sizeof( ppcJumpIsland ) * oc->n_islands + 3) / 4;
1434 unsigned long *p = (unsigned long *) oc->image;
1438 __asm__ volatile ( "dcbf 0,%0\n\t"
1446 __asm__ volatile ( "sync\n\t"
1452 /* --------------------------------------------------------------------------
1453 * PEi386 specifics (Win32 targets)
1454 * ------------------------------------------------------------------------*/
1456 /* The information for this linker comes from
1457 Microsoft Portable Executable
1458 and Common Object File Format Specification
1459 revision 5.1 January 1998
1460 which SimonM says comes from the MS Developer Network CDs.
1462 It can be found there (on older CDs), but can also be found
1465 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1467 (this is Rev 6.0 from February 1999).
1469 Things move, so if that fails, try searching for it via
1471 http://www.google.com/search?q=PE+COFF+specification
1473 The ultimate reference for the PE format is the Winnt.h
1474 header file that comes with the Platform SDKs; as always,
1475 implementations will drift wrt their documentation.
1477 A good background article on the PE format is Matt Pietrek's
1478 March 1994 article in Microsoft System Journal (MSJ)
1479 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1480 Win32 Portable Executable File Format." The info in there
1481 has recently been updated in a two part article in
1482 MSDN magazine, issues Feb and March 2002,
1483 "Inside Windows: An In-Depth Look into the Win32 Portable
1484 Executable File Format"
1486 John Levine's book "Linkers and Loaders" contains useful
1491 #if defined(OBJFORMAT_PEi386)
1495 typedef unsigned char UChar;
1496 typedef unsigned short UInt16;
1497 typedef unsigned int UInt32;
1504 UInt16 NumberOfSections;
1505 UInt32 TimeDateStamp;
1506 UInt32 PointerToSymbolTable;
1507 UInt32 NumberOfSymbols;
1508 UInt16 SizeOfOptionalHeader;
1509 UInt16 Characteristics;
1513 #define sizeof_COFF_header 20
1520 UInt32 VirtualAddress;
1521 UInt32 SizeOfRawData;
1522 UInt32 PointerToRawData;
1523 UInt32 PointerToRelocations;
1524 UInt32 PointerToLinenumbers;
1525 UInt16 NumberOfRelocations;
1526 UInt16 NumberOfLineNumbers;
1527 UInt32 Characteristics;
1531 #define sizeof_COFF_section 40
1538 UInt16 SectionNumber;
1541 UChar NumberOfAuxSymbols;
1545 #define sizeof_COFF_symbol 18
1550 UInt32 VirtualAddress;
1551 UInt32 SymbolTableIndex;
1556 #define sizeof_COFF_reloc 10
1559 /* From PE spec doc, section 3.3.2 */
1560 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1561 windows.h -- for the same purpose, but I want to know what I'm
1563 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1564 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1565 #define MYIMAGE_FILE_DLL 0x2000
1566 #define MYIMAGE_FILE_SYSTEM 0x1000
1567 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1568 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1569 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1571 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1572 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1573 #define MYIMAGE_SYM_CLASS_STATIC 3
1574 #define MYIMAGE_SYM_UNDEFINED 0
1576 /* From PE spec doc, section 4.1 */
1577 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1578 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1579 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1581 /* From PE spec doc, section 5.2.1 */
1582 #define MYIMAGE_REL_I386_DIR32 0x0006
1583 #define MYIMAGE_REL_I386_REL32 0x0014
1586 /* We use myindex to calculate array addresses, rather than
1587 simply doing the normal subscript thing. That's because
1588 some of the above structs have sizes which are not
1589 a whole number of words. GCC rounds their sizes up to a
1590 whole number of words, which means that the address calcs
1591 arising from using normal C indexing or pointer arithmetic
1592 are just plain wrong. Sigh.
1595 myindex ( int scale, void* base, int index )
1598 ((UChar*)base) + scale * index;
1603 printName ( UChar* name, UChar* strtab )
1605 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1606 UInt32 strtab_offset = * (UInt32*)(name+4);
1607 debugBelch("%s", strtab + strtab_offset );
1610 for (i = 0; i < 8; i++) {
1611 if (name[i] == 0) break;
1612 debugBelch("%c", name[i] );
1619 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1621 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1622 UInt32 strtab_offset = * (UInt32*)(name+4);
1623 strncpy ( dst, strtab+strtab_offset, dstSize );
1629 if (name[i] == 0) break;
1639 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1642 /* If the string is longer than 8 bytes, look in the
1643 string table for it -- this will be correctly zero terminated.
1645 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1646 UInt32 strtab_offset = * (UInt32*)(name+4);
1647 return ((UChar*)strtab) + strtab_offset;
1649 /* Otherwise, if shorter than 8 bytes, return the original,
1650 which by defn is correctly terminated.
1652 if (name[7]==0) return name;
1653 /* The annoying case: 8 bytes. Copy into a temporary
1654 (which is never freed ...)
1656 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1658 strncpy(newstr,name,8);
1664 /* Just compares the short names (first 8 chars) */
1665 static COFF_section *
1666 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1670 = (COFF_header*)(oc->image);
1671 COFF_section* sectab
1673 ((UChar*)(oc->image))
1674 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1676 for (i = 0; i < hdr->NumberOfSections; i++) {
1679 COFF_section* section_i
1681 myindex ( sizeof_COFF_section, sectab, i );
1682 n1 = (UChar*) &(section_i->Name);
1684 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1685 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1686 n1[6]==n2[6] && n1[7]==n2[7])
1695 zapTrailingAtSign ( UChar* sym )
1697 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1699 if (sym[0] == 0) return;
1701 while (sym[i] != 0) i++;
1704 while (j > 0 && my_isdigit(sym[j])) j--;
1705 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1711 ocVerifyImage_PEi386 ( ObjectCode* oc )
1716 COFF_section* sectab;
1717 COFF_symbol* symtab;
1719 /* debugBelch("\nLOADING %s\n", oc->fileName); */
1720 hdr = (COFF_header*)(oc->image);
1721 sectab = (COFF_section*) (
1722 ((UChar*)(oc->image))
1723 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1725 symtab = (COFF_symbol*) (
1726 ((UChar*)(oc->image))
1727 + hdr->PointerToSymbolTable
1729 strtab = ((UChar*)symtab)
1730 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1732 if (hdr->Machine != 0x14c) {
1733 errorBelch("Not x86 PEi386");
1736 if (hdr->SizeOfOptionalHeader != 0) {
1737 errorBelch("PEi386 with nonempty optional header");
1740 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
1741 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
1742 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
1743 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
1744 errorBelch("Not a PEi386 object file");
1747 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
1748 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
1749 errorBelch("Invalid PEi386 word size or endiannness: %d",
1750 (int)(hdr->Characteristics));
1753 /* If the string table size is way crazy, this might indicate that
1754 there are more than 64k relocations, despite claims to the
1755 contrary. Hence this test. */
1756 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
1758 if ( (*(UInt32*)strtab) > 600000 ) {
1759 /* Note that 600k has no special significance other than being
1760 big enough to handle the almost-2MB-sized lumps that
1761 constitute HSwin32*.o. */
1762 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
1767 /* No further verification after this point; only debug printing. */
1769 IF_DEBUG(linker, i=1);
1770 if (i == 0) return 1;
1772 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
1773 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
1774 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
1777 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
1778 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
1779 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
1780 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
1781 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
1782 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
1783 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
1785 /* Print the section table. */
1787 for (i = 0; i < hdr->NumberOfSections; i++) {
1789 COFF_section* sectab_i
1791 myindex ( sizeof_COFF_section, sectab, i );
1798 printName ( sectab_i->Name, strtab );
1808 sectab_i->VirtualSize,
1809 sectab_i->VirtualAddress,
1810 sectab_i->SizeOfRawData,
1811 sectab_i->PointerToRawData,
1812 sectab_i->NumberOfRelocations,
1813 sectab_i->PointerToRelocations,
1814 sectab_i->PointerToRawData
1816 reltab = (COFF_reloc*) (
1817 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1820 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
1821 /* If the relocation field (a short) has overflowed, the
1822 * real count can be found in the first reloc entry.
1824 * See Section 4.1 (last para) of the PE spec (rev6.0).
1826 COFF_reloc* rel = (COFF_reloc*)
1827 myindex ( sizeof_COFF_reloc, reltab, 0 );
1828 noRelocs = rel->VirtualAddress;
1831 noRelocs = sectab_i->NumberOfRelocations;
1835 for (; j < noRelocs; j++) {
1837 COFF_reloc* rel = (COFF_reloc*)
1838 myindex ( sizeof_COFF_reloc, reltab, j );
1840 " type 0x%-4x vaddr 0x%-8x name `",
1842 rel->VirtualAddress );
1843 sym = (COFF_symbol*)
1844 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
1845 /* Hmm..mysterious looking offset - what's it for? SOF */
1846 printName ( sym->Name, strtab -10 );
1853 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
1854 debugBelch("---START of string table---\n");
1855 for (i = 4; i < *(Int32*)strtab; i++) {
1857 debugBelch("\n"); else
1858 debugBelch("%c", strtab[i] );
1860 debugBelch("--- END of string table---\n");
1865 COFF_symbol* symtab_i;
1866 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1867 symtab_i = (COFF_symbol*)
1868 myindex ( sizeof_COFF_symbol, symtab, i );
1874 printName ( symtab_i->Name, strtab );
1883 (Int32)(symtab_i->SectionNumber),
1884 (UInt32)symtab_i->Type,
1885 (UInt32)symtab_i->StorageClass,
1886 (UInt32)symtab_i->NumberOfAuxSymbols
1888 i += symtab_i->NumberOfAuxSymbols;
1898 ocGetNames_PEi386 ( ObjectCode* oc )
1901 COFF_section* sectab;
1902 COFF_symbol* symtab;
1909 hdr = (COFF_header*)(oc->image);
1910 sectab = (COFF_section*) (
1911 ((UChar*)(oc->image))
1912 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1914 symtab = (COFF_symbol*) (
1915 ((UChar*)(oc->image))
1916 + hdr->PointerToSymbolTable
1918 strtab = ((UChar*)(oc->image))
1919 + hdr->PointerToSymbolTable
1920 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1922 /* Allocate space for any (local, anonymous) .bss sections. */
1924 for (i = 0; i < hdr->NumberOfSections; i++) {
1926 COFF_section* sectab_i
1928 myindex ( sizeof_COFF_section, sectab, i );
1929 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
1930 if (sectab_i->VirtualSize == 0) continue;
1931 /* This is a non-empty .bss section. Allocate zeroed space for
1932 it, and set its PointerToRawData field such that oc->image +
1933 PointerToRawData == addr_of_zeroed_space. */
1934 zspace = stgCallocBytes(1, sectab_i->VirtualSize,
1935 "ocGetNames_PEi386(anonymous bss)");
1936 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
1937 addProddableBlock(oc, zspace, sectab_i->VirtualSize);
1938 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
1941 /* Copy section information into the ObjectCode. */
1943 for (i = 0; i < hdr->NumberOfSections; i++) {
1949 = SECTIONKIND_OTHER;
1950 COFF_section* sectab_i
1952 myindex ( sizeof_COFF_section, sectab, i );
1953 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
1956 /* I'm sure this is the Right Way to do it. However, the
1957 alternative of testing the sectab_i->Name field seems to
1958 work ok with Cygwin.
1960 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
1961 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
1962 kind = SECTIONKIND_CODE_OR_RODATA;
1965 if (0==strcmp(".text",sectab_i->Name) ||
1966 0==strcmp(".rodata",sectab_i->Name))
1967 kind = SECTIONKIND_CODE_OR_RODATA;
1968 if (0==strcmp(".data",sectab_i->Name) ||
1969 0==strcmp(".bss",sectab_i->Name))
1970 kind = SECTIONKIND_RWDATA;
1972 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
1973 sz = sectab_i->SizeOfRawData;
1974 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
1976 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
1977 end = start + sz - 1;
1979 if (kind == SECTIONKIND_OTHER
1980 /* Ignore sections called which contain stabs debugging
1982 && 0 != strcmp(".stab", sectab_i->Name)
1983 && 0 != strcmp(".stabstr", sectab_i->Name)
1985 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
1989 if (kind != SECTIONKIND_OTHER && end >= start) {
1990 addSection(oc, kind, start, end);
1991 addProddableBlock(oc, start, end - start + 1);
1995 /* Copy exported symbols into the ObjectCode. */
1997 oc->n_symbols = hdr->NumberOfSymbols;
1998 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
1999 "ocGetNames_PEi386(oc->symbols)");
2000 /* Call me paranoid; I don't care. */
2001 for (i = 0; i < oc->n_symbols; i++)
2002 oc->symbols[i] = NULL;
2006 COFF_symbol* symtab_i;
2007 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2008 symtab_i = (COFF_symbol*)
2009 myindex ( sizeof_COFF_symbol, symtab, i );
2013 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2014 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2015 /* This symbol is global and defined, viz, exported */
2016 /* for MYIMAGE_SYMCLASS_EXTERNAL
2017 && !MYIMAGE_SYM_UNDEFINED,
2018 the address of the symbol is:
2019 address of relevant section + offset in section
2021 COFF_section* sectabent
2022 = (COFF_section*) myindex ( sizeof_COFF_section,
2024 symtab_i->SectionNumber-1 );
2025 addr = ((UChar*)(oc->image))
2026 + (sectabent->PointerToRawData
2030 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2031 && symtab_i->Value > 0) {
2032 /* This symbol isn't in any section at all, ie, global bss.
2033 Allocate zeroed space for it. */
2034 addr = stgCallocBytes(1, symtab_i->Value,
2035 "ocGetNames_PEi386(non-anonymous bss)");
2036 addSection(oc, SECTIONKIND_RWDATA, addr,
2037 ((UChar*)addr) + symtab_i->Value - 1);
2038 addProddableBlock(oc, addr, symtab_i->Value);
2039 /* debugBelch("BSS section at 0x%x\n", addr); */
2042 if (addr != NULL ) {
2043 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2044 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2045 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2046 ASSERT(i >= 0 && i < oc->n_symbols);
2047 /* cstring_from_COFF_symbol_name always succeeds. */
2048 oc->symbols[i] = sname;
2049 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2053 "IGNORING symbol %d\n"
2057 printName ( symtab_i->Name, strtab );
2066 (Int32)(symtab_i->SectionNumber),
2067 (UInt32)symtab_i->Type,
2068 (UInt32)symtab_i->StorageClass,
2069 (UInt32)symtab_i->NumberOfAuxSymbols
2074 i += symtab_i->NumberOfAuxSymbols;
2083 ocResolve_PEi386 ( ObjectCode* oc )
2086 COFF_section* sectab;
2087 COFF_symbol* symtab;
2097 /* ToDo: should be variable-sized? But is at least safe in the
2098 sense of buffer-overrun-proof. */
2100 /* debugBelch("resolving for %s\n", oc->fileName); */
2102 hdr = (COFF_header*)(oc->image);
2103 sectab = (COFF_section*) (
2104 ((UChar*)(oc->image))
2105 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2107 symtab = (COFF_symbol*) (
2108 ((UChar*)(oc->image))
2109 + hdr->PointerToSymbolTable
2111 strtab = ((UChar*)(oc->image))
2112 + hdr->PointerToSymbolTable
2113 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2115 for (i = 0; i < hdr->NumberOfSections; i++) {
2116 COFF_section* sectab_i
2118 myindex ( sizeof_COFF_section, sectab, i );
2121 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2124 /* Ignore sections called which contain stabs debugging
2126 if (0 == strcmp(".stab", sectab_i->Name)
2127 || 0 == strcmp(".stabstr", sectab_i->Name))
2130 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2131 /* If the relocation field (a short) has overflowed, the
2132 * real count can be found in the first reloc entry.
2134 * See Section 4.1 (last para) of the PE spec (rev6.0).
2136 * Nov2003 update: the GNU linker still doesn't correctly
2137 * handle the generation of relocatable object files with
2138 * overflown relocations. Hence the output to warn of potential
2141 COFF_reloc* rel = (COFF_reloc*)
2142 myindex ( sizeof_COFF_reloc, reltab, 0 );
2143 noRelocs = rel->VirtualAddress;
2144 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2148 noRelocs = sectab_i->NumberOfRelocations;
2153 for (; j < noRelocs; j++) {
2155 COFF_reloc* reltab_j
2157 myindex ( sizeof_COFF_reloc, reltab, j );
2159 /* the location to patch */
2161 ((UChar*)(oc->image))
2162 + (sectab_i->PointerToRawData
2163 + reltab_j->VirtualAddress
2164 - sectab_i->VirtualAddress )
2166 /* the existing contents of pP */
2168 /* the symbol to connect to */
2169 sym = (COFF_symbol*)
2170 myindex ( sizeof_COFF_symbol,
2171 symtab, reltab_j->SymbolTableIndex );
2174 "reloc sec %2d num %3d: type 0x%-4x "
2175 "vaddr 0x%-8x name `",
2177 (UInt32)reltab_j->Type,
2178 reltab_j->VirtualAddress );
2179 printName ( sym->Name, strtab );
2180 debugBelch("'\n" ));
2182 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2183 COFF_section* section_sym
2184 = findPEi386SectionCalled ( oc, sym->Name );
2186 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2189 S = ((UInt32)(oc->image))
2190 + (section_sym->PointerToRawData
2193 copyName ( sym->Name, strtab, symbol, 1000-1 );
2194 (void*)S = lookupLocalSymbol( oc, symbol );
2195 if ((void*)S != NULL) goto foundit;
2196 (void*)S = lookupSymbol( symbol );
2197 if ((void*)S != NULL) goto foundit;
2198 zapTrailingAtSign ( symbol );
2199 (void*)S = lookupLocalSymbol( oc, symbol );
2200 if ((void*)S != NULL) goto foundit;
2201 (void*)S = lookupSymbol( symbol );
2202 if ((void*)S != NULL) goto foundit;
2203 /* Newline first because the interactive linker has printed "linking..." */
2204 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2208 checkProddableBlock(oc, pP);
2209 switch (reltab_j->Type) {
2210 case MYIMAGE_REL_I386_DIR32:
2213 case MYIMAGE_REL_I386_REL32:
2214 /* Tricky. We have to insert a displacement at
2215 pP which, when added to the PC for the _next_
2216 insn, gives the address of the target (S).
2217 Problem is to know the address of the next insn
2218 when we only know pP. We assume that this
2219 literal field is always the last in the insn,
2220 so that the address of the next insn is pP+4
2221 -- hence the constant 4.
2222 Also I don't know if A should be added, but so
2223 far it has always been zero.
2226 *pP = S - ((UInt32)pP) - 4;
2229 debugBelch("%s: unhandled PEi386 relocation type %d",
2230 oc->fileName, reltab_j->Type);
2237 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2241 #endif /* defined(OBJFORMAT_PEi386) */
2244 /* --------------------------------------------------------------------------
2246 * ------------------------------------------------------------------------*/
2248 #if defined(OBJFORMAT_ELF)
2253 #if defined(sparc_HOST_ARCH)
2254 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2255 #elif defined(i386_HOST_ARCH)
2256 # define ELF_TARGET_386 /* Used inside <elf.h> */
2257 #elif defined(x86_64_HOST_ARCH)
2258 # define ELF_TARGET_X64_64
2260 #elif defined (ia64_HOST_ARCH)
2261 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2263 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2264 # define ELF_NEED_GOT /* needs Global Offset Table */
2265 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2268 #if !defined(openbsd_HOST_OS)
2271 /* openbsd elf has things in different places, with diff names */
2272 #include <elf_abi.h>
2273 #include <machine/reloc.h>
2274 #define R_386_32 RELOC_32
2275 #define R_386_PC32 RELOC_PC32
2279 * Define a set of types which can be used for both ELF32 and ELF64
2283 #define ELFCLASS ELFCLASS64
2284 #define Elf_Addr Elf64_Addr
2285 #define Elf_Word Elf64_Word
2286 #define Elf_Sword Elf64_Sword
2287 #define Elf_Ehdr Elf64_Ehdr
2288 #define Elf_Phdr Elf64_Phdr
2289 #define Elf_Shdr Elf64_Shdr
2290 #define Elf_Sym Elf64_Sym
2291 #define Elf_Rel Elf64_Rel
2292 #define Elf_Rela Elf64_Rela
2293 #define ELF_ST_TYPE ELF64_ST_TYPE
2294 #define ELF_ST_BIND ELF64_ST_BIND
2295 #define ELF_R_TYPE ELF64_R_TYPE
2296 #define ELF_R_SYM ELF64_R_SYM
2298 #define ELFCLASS ELFCLASS32
2299 #define Elf_Addr Elf32_Addr
2300 #define Elf_Word Elf32_Word
2301 #define Elf_Sword Elf32_Sword
2302 #define Elf_Ehdr Elf32_Ehdr
2303 #define Elf_Phdr Elf32_Phdr
2304 #define Elf_Shdr Elf32_Shdr
2305 #define Elf_Sym Elf32_Sym
2306 #define Elf_Rel Elf32_Rel
2307 #define Elf_Rela Elf32_Rela
2309 #define ELF_ST_TYPE ELF32_ST_TYPE
2312 #define ELF_ST_BIND ELF32_ST_BIND
2315 #define ELF_R_TYPE ELF32_R_TYPE
2318 #define ELF_R_SYM ELF32_R_SYM
2324 * Functions to allocate entries in dynamic sections. Currently we simply
2325 * preallocate a large number, and we don't check if a entry for the given
2326 * target already exists (a linear search is too slow). Ideally these
2327 * entries would be associated with symbols.
2330 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2331 #define GOT_SIZE 0x20000
2332 #define FUNCTION_TABLE_SIZE 0x10000
2333 #define PLT_SIZE 0x08000
2336 static Elf_Addr got[GOT_SIZE];
2337 static unsigned int gotIndex;
2338 static Elf_Addr gp_val = (Elf_Addr)got;
2341 allocateGOTEntry(Elf_Addr target)
2345 if (gotIndex >= GOT_SIZE)
2346 barf("Global offset table overflow");
2348 entry = &got[gotIndex++];
2350 return (Elf_Addr)entry;
2354 #ifdef ELF_FUNCTION_DESC
2360 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2361 static unsigned int functionTableIndex;
2364 allocateFunctionDesc(Elf_Addr target)
2366 FunctionDesc *entry;
2368 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2369 barf("Function table overflow");
2371 entry = &functionTable[functionTableIndex++];
2373 entry->gp = (Elf_Addr)gp_val;
2374 return (Elf_Addr)entry;
2378 copyFunctionDesc(Elf_Addr target)
2380 FunctionDesc *olddesc = (FunctionDesc *)target;
2381 FunctionDesc *newdesc;
2383 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2384 newdesc->gp = olddesc->gp;
2385 return (Elf_Addr)newdesc;
2390 #ifdef ia64_HOST_ARCH
2391 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2392 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2394 static unsigned char plt_code[] =
2396 /* taken from binutils bfd/elfxx-ia64.c */
2397 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2398 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2399 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2400 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2401 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2402 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2405 /* If we can't get to the function descriptor via gp, take a local copy of it */
2406 #define PLT_RELOC(code, target) { \
2407 Elf64_Sxword rel_value = target - gp_val; \
2408 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2409 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2411 ia64_reloc_gprel22((Elf_Addr)code, target); \
2416 unsigned char code[sizeof(plt_code)];
2420 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2422 PLTEntry *plt = (PLTEntry *)oc->plt;
2425 if (oc->pltIndex >= PLT_SIZE)
2426 barf("Procedure table overflow");
2428 entry = &plt[oc->pltIndex++];
2429 memcpy(entry->code, plt_code, sizeof(entry->code));
2430 PLT_RELOC(entry->code, target);
2431 return (Elf_Addr)entry;
2437 return (PLT_SIZE * sizeof(PLTEntry));
2443 * Generic ELF functions
2447 findElfSection ( void* objImage, Elf_Word sh_type )
2449 char* ehdrC = (char*)objImage;
2450 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2451 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2452 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2456 for (i = 0; i < ehdr->e_shnum; i++) {
2457 if (shdr[i].sh_type == sh_type
2458 /* Ignore the section header's string table. */
2459 && i != ehdr->e_shstrndx
2460 /* Ignore string tables named .stabstr, as they contain
2462 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2464 ptr = ehdrC + shdr[i].sh_offset;
2471 #if defined(ia64_HOST_ARCH)
2473 findElfSegment ( void* objImage, Elf_Addr vaddr )
2475 char* ehdrC = (char*)objImage;
2476 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2477 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2478 Elf_Addr segaddr = 0;
2481 for (i = 0; i < ehdr->e_phnum; i++) {
2482 segaddr = phdr[i].p_vaddr;
2483 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2491 ocVerifyImage_ELF ( ObjectCode* oc )
2495 int i, j, nent, nstrtab, nsymtabs;
2499 char* ehdrC = (char*)(oc->image);
2500 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2502 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2503 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2504 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2505 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2506 errorBelch("%s: not an ELF object", oc->fileName);
2510 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2511 errorBelch("%s: unsupported ELF format", oc->fileName);
2515 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2516 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2518 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2519 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2521 errorBelch("%s: unknown endiannness", oc->fileName);
2525 if (ehdr->e_type != ET_REL) {
2526 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2529 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2531 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2532 switch (ehdr->e_machine) {
2533 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2534 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2536 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2538 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2540 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2542 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2543 errorBelch("%s: unknown architecture", oc->fileName);
2547 IF_DEBUG(linker,debugBelch(
2548 "\nSection header table: start %d, n_entries %d, ent_size %d\n",
2549 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2551 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2553 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2555 if (ehdr->e_shstrndx == SHN_UNDEF) {
2556 errorBelch("%s: no section header string table", oc->fileName);
2559 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2561 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2564 for (i = 0; i < ehdr->e_shnum; i++) {
2565 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2566 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2567 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2568 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2569 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2570 ehdrC + shdr[i].sh_offset,
2571 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2573 if (shdr[i].sh_type == SHT_REL) {
2574 IF_DEBUG(linker,debugBelch("Rel " ));
2575 } else if (shdr[i].sh_type == SHT_RELA) {
2576 IF_DEBUG(linker,debugBelch("RelA " ));
2578 IF_DEBUG(linker,debugBelch(" "));
2581 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2585 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2588 for (i = 0; i < ehdr->e_shnum; i++) {
2589 if (shdr[i].sh_type == SHT_STRTAB
2590 /* Ignore the section header's string table. */
2591 && i != ehdr->e_shstrndx
2592 /* Ignore string tables named .stabstr, as they contain
2594 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2596 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2597 strtab = ehdrC + shdr[i].sh_offset;
2602 errorBelch("%s: no string tables, or too many", oc->fileName);
2607 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2608 for (i = 0; i < ehdr->e_shnum; i++) {
2609 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2610 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2612 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2613 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2614 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%d rem)\n",
2616 shdr[i].sh_size % sizeof(Elf_Sym)
2618 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2619 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2622 for (j = 0; j < nent; j++) {
2623 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2624 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
2625 (int)stab[j].st_shndx,
2626 (int)stab[j].st_size,
2627 (char*)stab[j].st_value ));
2629 IF_DEBUG(linker,debugBelch("type=" ));
2630 switch (ELF_ST_TYPE(stab[j].st_info)) {
2631 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
2632 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
2633 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
2634 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
2635 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
2636 default: IF_DEBUG(linker,debugBelch("? " )); break;
2638 IF_DEBUG(linker,debugBelch(" " ));
2640 IF_DEBUG(linker,debugBelch("bind=" ));
2641 switch (ELF_ST_BIND(stab[j].st_info)) {
2642 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
2643 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
2644 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
2645 default: IF_DEBUG(linker,debugBelch("? " )); break;
2647 IF_DEBUG(linker,debugBelch(" " ));
2649 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
2653 if (nsymtabs == 0) {
2654 errorBelch("%s: didn't find any symbol tables", oc->fileName);
2661 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
2665 if (hdr->sh_type == SHT_PROGBITS
2666 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
2667 /* .text-style section */
2668 return SECTIONKIND_CODE_OR_RODATA;
2671 if (hdr->sh_type == SHT_PROGBITS
2672 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2673 /* .data-style section */
2674 return SECTIONKIND_RWDATA;
2677 if (hdr->sh_type == SHT_PROGBITS
2678 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
2679 /* .rodata-style section */
2680 return SECTIONKIND_CODE_OR_RODATA;
2683 if (hdr->sh_type == SHT_NOBITS
2684 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2685 /* .bss-style section */
2687 return SECTIONKIND_RWDATA;
2690 return SECTIONKIND_OTHER;
2695 ocGetNames_ELF ( ObjectCode* oc )
2700 char* ehdrC = (char*)(oc->image);
2701 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2702 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
2703 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2705 ASSERT(symhash != NULL);
2708 errorBelch("%s: no strtab", oc->fileName);
2713 for (i = 0; i < ehdr->e_shnum; i++) {
2714 /* Figure out what kind of section it is. Logic derived from
2715 Figure 1.14 ("Special Sections") of the ELF document
2716 ("Portable Formats Specification, Version 1.1"). */
2718 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
2720 if (is_bss && shdr[i].sh_size > 0) {
2721 /* This is a non-empty .bss section. Allocate zeroed space for
2722 it, and set its .sh_offset field such that
2723 ehdrC + .sh_offset == addr_of_zeroed_space. */
2724 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
2725 "ocGetNames_ELF(BSS)");
2726 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
2728 debugBelch("BSS section at 0x%x, size %d\n",
2729 zspace, shdr[i].sh_size);
2733 /* fill in the section info */
2734 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
2735 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
2736 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
2737 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
2740 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2742 /* copy stuff into this module's object symbol table */
2743 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2744 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2746 oc->n_symbols = nent;
2747 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2748 "ocGetNames_ELF(oc->symbols)");
2750 for (j = 0; j < nent; j++) {
2752 char isLocal = FALSE; /* avoids uninit-var warning */
2754 char* nm = strtab + stab[j].st_name;
2755 int secno = stab[j].st_shndx;
2757 /* Figure out if we want to add it; if so, set ad to its
2758 address. Otherwise leave ad == NULL. */
2760 if (secno == SHN_COMMON) {
2762 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
2764 debugBelch("COMMON symbol, size %d name %s\n",
2765 stab[j].st_size, nm);
2767 /* Pointless to do addProddableBlock() for this area,
2768 since the linker should never poke around in it. */
2771 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
2772 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
2774 /* and not an undefined symbol */
2775 && stab[j].st_shndx != SHN_UNDEF
2776 /* and not in a "special section" */
2777 && stab[j].st_shndx < SHN_LORESERVE
2779 /* and it's a not a section or string table or anything silly */
2780 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
2781 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
2782 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
2785 /* Section 0 is the undefined section, hence > and not >=. */
2786 ASSERT(secno > 0 && secno < ehdr->e_shnum);
2788 if (shdr[secno].sh_type == SHT_NOBITS) {
2789 debugBelch(" BSS symbol, size %d off %d name %s\n",
2790 stab[j].st_size, stab[j].st_value, nm);
2793 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
2794 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
2797 #ifdef ELF_FUNCTION_DESC
2798 /* dlsym() and the initialisation table both give us function
2799 * descriptors, so to be consistent we store function descriptors
2800 * in the symbol table */
2801 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
2802 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
2804 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
2805 ad, oc->fileName, nm ));
2810 /* And the decision is ... */
2814 oc->symbols[j] = nm;
2817 /* Ignore entirely. */
2819 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
2823 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
2824 strtab + stab[j].st_name ));
2827 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
2828 (int)ELF_ST_BIND(stab[j].st_info),
2829 (int)ELF_ST_TYPE(stab[j].st_info),
2830 (int)stab[j].st_shndx,
2831 strtab + stab[j].st_name
2834 oc->symbols[j] = NULL;
2843 /* Do ELF relocations which lack an explicit addend. All x86-linux
2844 relocations appear to be of this form. */
2846 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
2847 Elf_Shdr* shdr, int shnum,
2848 Elf_Sym* stab, char* strtab )
2853 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
2854 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
2855 int target_shndx = shdr[shnum].sh_info;
2856 int symtab_shndx = shdr[shnum].sh_link;
2858 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2859 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
2860 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2861 target_shndx, symtab_shndx ));
2863 /* Skip sections that we're not interested in. */
2866 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
2867 if (kind == SECTIONKIND_OTHER) {
2868 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
2873 for (j = 0; j < nent; j++) {
2874 Elf_Addr offset = rtab[j].r_offset;
2875 Elf_Addr info = rtab[j].r_info;
2877 Elf_Addr P = ((Elf_Addr)targ) + offset;
2878 Elf_Word* pP = (Elf_Word*)P;
2884 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
2885 j, (void*)offset, (void*)info ));
2887 IF_DEBUG(linker,debugBelch( " ZERO" ));
2890 Elf_Sym sym = stab[ELF_R_SYM(info)];
2891 /* First see if it is a local symbol. */
2892 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2893 /* Yes, so we can get the address directly from the ELF symbol
2895 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2897 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2898 + stab[ELF_R_SYM(info)].st_value);
2901 /* No, so look up the name in our global table. */
2902 symbol = strtab + sym.st_name;
2903 S_tmp = lookupSymbol( symbol );
2904 S = (Elf_Addr)S_tmp;
2907 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2910 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
2913 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
2914 (void*)P, (void*)S, (void*)A ));
2915 checkProddableBlock ( oc, pP );
2919 switch (ELF_R_TYPE(info)) {
2920 # ifdef i386_HOST_ARCH
2921 case R_386_32: *pP = value; break;
2922 case R_386_PC32: *pP = value - P; break;
2925 errorBelch("%s: unhandled ELF relocation(Rel) type %d\n",
2926 oc->fileName, ELF_R_TYPE(info));
2934 /* Do ELF relocations for which explicit addends are supplied.
2935 sparc-solaris relocations appear to be of this form. */
2937 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
2938 Elf_Shdr* shdr, int shnum,
2939 Elf_Sym* stab, char* strtab )
2944 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
2945 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
2946 int target_shndx = shdr[shnum].sh_info;
2947 int symtab_shndx = shdr[shnum].sh_link;
2949 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2950 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
2951 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2952 target_shndx, symtab_shndx ));
2954 for (j = 0; j < nent; j++) {
2955 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
2956 /* This #ifdef only serves to avoid unused-var warnings. */
2957 Elf_Addr offset = rtab[j].r_offset;
2958 Elf_Addr P = targ + offset;
2960 Elf_Addr info = rtab[j].r_info;
2961 Elf_Addr A = rtab[j].r_addend;
2965 # if defined(sparc_HOST_ARCH)
2966 Elf_Word* pP = (Elf_Word*)P;
2968 # elif defined(ia64_HOST_ARCH)
2969 Elf64_Xword *pP = (Elf64_Xword *)P;
2971 # elif defined(powerpc_HOST_ARCH)
2975 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
2976 j, (void*)offset, (void*)info,
2979 IF_DEBUG(linker,debugBelch( " ZERO" ));
2982 Elf_Sym sym = stab[ELF_R_SYM(info)];
2983 /* First see if it is a local symbol. */
2984 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2985 /* Yes, so we can get the address directly from the ELF symbol
2987 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2989 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2990 + stab[ELF_R_SYM(info)].st_value);
2991 #ifdef ELF_FUNCTION_DESC
2992 /* Make a function descriptor for this function */
2993 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
2994 S = allocateFunctionDesc(S + A);
2999 /* No, so look up the name in our global table. */
3000 symbol = strtab + sym.st_name;
3001 S_tmp = lookupSymbol( symbol );
3002 S = (Elf_Addr)S_tmp;
3004 #ifdef ELF_FUNCTION_DESC
3005 /* If a function, already a function descriptor - we would
3006 have to copy it to add an offset. */
3007 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3008 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3012 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3015 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3018 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3019 (void*)P, (void*)S, (void*)A ));
3020 /* checkProddableBlock ( oc, (void*)P ); */
3024 switch (ELF_R_TYPE(info)) {
3025 # if defined(sparc_HOST_ARCH)
3026 case R_SPARC_WDISP30:
3027 w1 = *pP & 0xC0000000;
3028 w2 = (Elf_Word)((value - P) >> 2);
3029 ASSERT((w2 & 0xC0000000) == 0);
3034 w1 = *pP & 0xFFC00000;
3035 w2 = (Elf_Word)(value >> 10);
3036 ASSERT((w2 & 0xFFC00000) == 0);
3042 w2 = (Elf_Word)(value & 0x3FF);
3043 ASSERT((w2 & ~0x3FF) == 0);
3047 /* According to the Sun documentation:
3049 This relocation type resembles R_SPARC_32, except it refers to an
3050 unaligned word. That is, the word to be relocated must be treated
3051 as four separate bytes with arbitrary alignment, not as a word
3052 aligned according to the architecture requirements.
3054 (JRS: which means that freeloading on the R_SPARC_32 case
3055 is probably wrong, but hey ...)
3059 w2 = (Elf_Word)value;
3062 # elif defined(ia64_HOST_ARCH)
3063 case R_IA64_DIR64LSB:
3064 case R_IA64_FPTR64LSB:
3067 case R_IA64_PCREL64LSB:
3070 case R_IA64_SEGREL64LSB:
3071 addr = findElfSegment(ehdrC, value);
3074 case R_IA64_GPREL22:
3075 ia64_reloc_gprel22(P, value);
3077 case R_IA64_LTOFF22:
3078 case R_IA64_LTOFF22X:
3079 case R_IA64_LTOFF_FPTR22:
3080 addr = allocateGOTEntry(value);
3081 ia64_reloc_gprel22(P, addr);
3083 case R_IA64_PCREL21B:
3084 ia64_reloc_pcrel21(P, S, oc);
3087 /* This goes with R_IA64_LTOFF22X and points to the load to
3088 * convert into a move. We don't implement relaxation. */
3090 # elif defined(powerpc_HOST_ARCH)
3091 case R_PPC_ADDR16_LO:
3092 *(Elf32_Half*) P = value;
3095 case R_PPC_ADDR16_HI:
3096 *(Elf32_Half*) P = value >> 16;
3099 case R_PPC_ADDR16_HA:
3100 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3104 *(Elf32_Word *) P = value;
3108 *(Elf32_Word *) P = value - P;
3114 if( delta << 6 >> 6 != delta )
3116 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3119 if( value == 0 || delta << 6 >> 6 != delta )
3121 barf( "Unable to make ppcJumpIsland for #%d",
3127 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3128 | (delta & 0x3fffffc);
3134 *(Elf64_Xword *)P = value;
3138 *(Elf64_Word *)P = (Elf64_Word) (value - P);
3142 *(Elf64_Word *)P = (Elf64_Word)value;
3146 *(Elf64_Sword *)P = (Elf64_Sword)value;
3151 errorBelch("%s: unhandled ELF relocation(RelA) type %d\n",
3152 oc->fileName, ELF_R_TYPE(info));
3161 ocResolve_ELF ( ObjectCode* oc )
3165 Elf_Sym* stab = NULL;
3166 char* ehdrC = (char*)(oc->image);
3167 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3168 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3170 /* first find "the" symbol table */
3171 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3173 /* also go find the string table */
3174 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3176 if (stab == NULL || strtab == NULL) {
3177 errorBelch("%s: can't find string or symbol table", oc->fileName);
3181 /* Process the relocation sections. */
3182 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3183 if (shdr[shnum].sh_type == SHT_REL) {
3184 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3185 shnum, stab, strtab );
3189 if (shdr[shnum].sh_type == SHT_RELA) {
3190 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3191 shnum, stab, strtab );
3196 /* Free the local symbol table; we won't need it again. */
3197 freeHashTable(oc->lochash, NULL);
3200 #if defined(powerpc_HOST_ARCH)
3201 ocFlushInstructionCache( oc );
3209 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3210 * at the front. The following utility functions pack and unpack instructions, and
3211 * take care of the most common relocations.
3214 #ifdef ia64_HOST_ARCH
3217 ia64_extract_instruction(Elf64_Xword *target)
3220 int slot = (Elf_Addr)target & 3;
3221 (Elf_Addr)target &= ~3;
3229 return ((w1 >> 5) & 0x1ffffffffff);
3231 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3235 barf("ia64_extract_instruction: invalid slot %p", target);
3240 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3242 int slot = (Elf_Addr)target & 3;
3243 (Elf_Addr)target &= ~3;
3248 *target |= value << 5;
3251 *target |= value << 46;
3252 *(target+1) |= value >> 18;
3255 *(target+1) |= value << 23;
3261 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3263 Elf64_Xword instruction;
3264 Elf64_Sxword rel_value;
3266 rel_value = value - gp_val;
3267 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3268 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3270 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3271 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3272 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3273 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3274 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3275 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3279 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3281 Elf64_Xword instruction;
3282 Elf64_Sxword rel_value;
3285 entry = allocatePLTEntry(value, oc);
3287 rel_value = (entry >> 4) - (target >> 4);
3288 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3289 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3291 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3292 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3293 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3294 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3300 * PowerPC ELF specifics
3303 #ifdef powerpc_HOST_ARCH
3305 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3311 ehdr = (Elf_Ehdr *) oc->image;
3312 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3314 for( i = 0; i < ehdr->e_shnum; i++ )
3315 if( shdr[i].sh_type == SHT_SYMTAB )
3318 if( i == ehdr->e_shnum )
3320 errorBelch( "This ELF file contains no symtab" );
3324 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3326 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3327 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3332 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3335 #endif /* powerpc */
3339 /* --------------------------------------------------------------------------
3341 * ------------------------------------------------------------------------*/
3343 #if defined(OBJFORMAT_MACHO)
3346 Support for MachO linking on Darwin/MacOS X on PowerPC chips
3347 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3349 I hereby formally apologize for the hackish nature of this code.
3350 Things that need to be done:
3351 *) implement ocVerifyImage_MachO
3352 *) add still more sanity checks.
3355 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3357 struct mach_header *header = (struct mach_header *) oc->image;
3358 struct load_command *lc = (struct load_command *) (header + 1);
3361 for( i = 0; i < header->ncmds; i++ )
3363 if( lc->cmd == LC_SYMTAB )
3365 // Find out the first and last undefined external
3366 // symbol, so we don't have to allocate too many
3368 struct symtab_command *symLC = (struct symtab_command *) lc;
3369 unsigned min = symLC->nsyms, max = 0;
3370 struct nlist *nlist =
3371 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3373 for(i=0;i<symLC->nsyms;i++)
3375 if(nlist[i].n_type & N_STAB)
3377 else if(nlist[i].n_type & N_EXT)
3379 if((nlist[i].n_type & N_TYPE) == N_UNDF
3380 && (nlist[i].n_value == 0))
3390 return ocAllocateJumpIslands(oc, max - min + 1, min);
3395 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3397 return ocAllocateJumpIslands(oc,0,0);
3400 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3402 // FIXME: do some verifying here
3406 static int resolveImports(
3409 struct symtab_command *symLC,
3410 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3411 unsigned long *indirectSyms,
3412 struct nlist *nlist)
3416 for(i=0;i*4<sect->size;i++)
3418 // according to otool, reserved1 contains the first index into the indirect symbol table
3419 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3420 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3423 if((symbol->n_type & N_TYPE) == N_UNDF
3424 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3425 addr = (void*) (symbol->n_value);
3426 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3429 addr = lookupSymbol(nm);
3432 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3436 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3437 ((void**)(image + sect->offset))[i] = addr;
3443 static unsigned long relocateAddress(
3446 struct section* sections,
3447 unsigned long address)
3450 for(i = 0; i < nSections; i++)
3452 if(sections[i].addr <= address
3453 && address < sections[i].addr + sections[i].size)
3455 return (unsigned long)oc->image
3456 + sections[i].offset + address - sections[i].addr;
3459 barf("Invalid Mach-O file:"
3460 "Address out of bounds while relocating object file");
3464 static int relocateSection(
3467 struct symtab_command *symLC, struct nlist *nlist,
3468 int nSections, struct section* sections, struct section *sect)
3470 struct relocation_info *relocs;
3473 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3475 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3479 relocs = (struct relocation_info*) (image + sect->reloff);
3483 if(relocs[i].r_address & R_SCATTERED)
3485 struct scattered_relocation_info *scat =
3486 (struct scattered_relocation_info*) &relocs[i];
3490 if(scat->r_length == 2)
3492 unsigned long word = 0;
3493 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3494 checkProddableBlock(oc,wordPtr);
3496 // Step 1: Figure out what the relocated value should be
3497 if(scat->r_type == GENERIC_RELOC_VANILLA)
3499 word = *wordPtr + (unsigned long) relocateAddress(
3506 else if(scat->r_type == PPC_RELOC_SECTDIFF
3507 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3508 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3509 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3511 struct scattered_relocation_info *pair =
3512 (struct scattered_relocation_info*) &relocs[i+1];
3514 if(!pair->r_scattered || pair->r_type != PPC_RELOC_PAIR)
3515 barf("Invalid Mach-O file: "
3516 "PPC_RELOC_*_SECTDIFF not followed by PPC_RELOC_PAIR");
3518 word = (unsigned long)
3519 (relocateAddress(oc, nSections, sections, scat->r_value)
3520 - relocateAddress(oc, nSections, sections, pair->r_value));
3523 else if(scat->r_type == PPC_RELOC_HI16
3524 || scat->r_type == PPC_RELOC_LO16
3525 || scat->r_type == PPC_RELOC_HA16
3526 || scat->r_type == PPC_RELOC_LO14)
3527 { // these are generated by label+offset things
3528 struct relocation_info *pair = &relocs[i+1];
3529 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3530 barf("Invalid Mach-O file: "
3531 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3533 if(scat->r_type == PPC_RELOC_LO16)
3535 word = ((unsigned short*) wordPtr)[1];
3536 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3538 else if(scat->r_type == PPC_RELOC_LO14)
3540 barf("Unsupported Relocation: PPC_RELOC_LO14");
3541 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3542 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3544 else if(scat->r_type == PPC_RELOC_HI16)
3546 word = ((unsigned short*) wordPtr)[1] << 16;
3547 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3549 else if(scat->r_type == PPC_RELOC_HA16)
3551 word = ((unsigned short*) wordPtr)[1] << 16;
3552 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3556 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
3562 continue; // ignore the others
3564 if(scat->r_type == GENERIC_RELOC_VANILLA
3565 || scat->r_type == PPC_RELOC_SECTDIFF)
3569 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
3571 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3573 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
3575 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3577 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
3579 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3580 + ((word & (1<<15)) ? 1 : 0);
3585 continue; // FIXME: I hope it's OK to ignore all the others.
3589 struct relocation_info *reloc = &relocs[i];
3590 if(reloc->r_pcrel && !reloc->r_extern)
3593 if(reloc->r_length == 2)
3595 unsigned long word = 0;
3596 unsigned long jumpIsland = 0;
3597 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
3598 // to avoid warning and to catch
3601 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
3602 checkProddableBlock(oc,wordPtr);
3604 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3608 else if(reloc->r_type == PPC_RELOC_LO16)
3610 word = ((unsigned short*) wordPtr)[1];
3611 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3613 else if(reloc->r_type == PPC_RELOC_HI16)
3615 word = ((unsigned short*) wordPtr)[1] << 16;
3616 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3618 else if(reloc->r_type == PPC_RELOC_HA16)
3620 word = ((unsigned short*) wordPtr)[1] << 16;
3621 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3623 else if(reloc->r_type == PPC_RELOC_BR24)
3626 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
3630 if(!reloc->r_extern)
3633 sections[reloc->r_symbolnum-1].offset
3634 - sections[reloc->r_symbolnum-1].addr
3641 struct nlist *symbol = &nlist[reloc->r_symbolnum];
3642 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3643 void *symbolAddress = lookupSymbol(nm);
3646 errorBelch("\nunknown symbol `%s'", nm);
3652 // In the .o file, this should be a relative jump to NULL
3653 // and we'll change it to a jump to a relative jump to the symbol
3654 ASSERT(-word == reloc->r_address);
3655 word = (unsigned long) symbolAddress;
3656 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,word);
3657 word -= ((long)image) + sect->offset + reloc->r_address;
3660 offsetToJumpIsland = jumpIsland
3661 - (((long)image) + sect->offset + reloc->r_address);
3666 word += (unsigned long) symbolAddress;
3670 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3675 else if(reloc->r_type == PPC_RELOC_LO16)
3677 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3680 else if(reloc->r_type == PPC_RELOC_HI16)
3682 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3685 else if(reloc->r_type == PPC_RELOC_HA16)
3687 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3688 + ((word & (1<<15)) ? 1 : 0);
3691 else if(reloc->r_type == PPC_RELOC_BR24)
3693 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3695 // The branch offset is too large.
3696 // Therefore, we try to use a jump island.
3699 barf("unconditional relative branch out of range: "
3700 "no jump island available");
3703 word = offsetToJumpIsland;
3704 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3705 barf("unconditional relative branch out of range: "
3706 "jump island out of range");
3708 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
3712 barf("\nunknown relocation %d",reloc->r_type);
3719 static int ocGetNames_MachO(ObjectCode* oc)
3721 char *image = (char*) oc->image;
3722 struct mach_header *header = (struct mach_header*) image;
3723 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3724 unsigned i,curSymbol = 0;
3725 struct segment_command *segLC = NULL;
3726 struct section *sections;
3727 struct symtab_command *symLC = NULL;
3728 struct nlist *nlist;
3729 unsigned long commonSize = 0;
3730 char *commonStorage = NULL;
3731 unsigned long commonCounter;
3733 for(i=0;i<header->ncmds;i++)
3735 if(lc->cmd == LC_SEGMENT)
3736 segLC = (struct segment_command*) lc;
3737 else if(lc->cmd == LC_SYMTAB)
3738 symLC = (struct symtab_command*) lc;
3739 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3742 sections = (struct section*) (segLC+1);
3743 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3746 for(i=0;i<segLC->nsects;i++)
3748 if(sections[i].size == 0)
3751 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
3753 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
3754 "ocGetNames_MachO(common symbols)");
3755 sections[i].offset = zeroFillArea - image;
3758 if(!strcmp(sections[i].sectname,"__text"))
3759 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
3760 (void*) (image + sections[i].offset),
3761 (void*) (image + sections[i].offset + sections[i].size));
3762 else if(!strcmp(sections[i].sectname,"__const"))
3763 addSection(oc, SECTIONKIND_RWDATA,
3764 (void*) (image + sections[i].offset),
3765 (void*) (image + sections[i].offset + sections[i].size));
3766 else if(!strcmp(sections[i].sectname,"__data"))
3767 addSection(oc, SECTIONKIND_RWDATA,
3768 (void*) (image + sections[i].offset),
3769 (void*) (image + sections[i].offset + sections[i].size));
3770 else if(!strcmp(sections[i].sectname,"__bss")
3771 || !strcmp(sections[i].sectname,"__common"))
3772 addSection(oc, SECTIONKIND_RWDATA,
3773 (void*) (image + sections[i].offset),
3774 (void*) (image + sections[i].offset + sections[i].size));
3776 addProddableBlock(oc, (void*) (image + sections[i].offset),
3780 // count external symbols defined here
3784 for(i=0;i<symLC->nsyms;i++)
3786 if(nlist[i].n_type & N_STAB)
3788 else if(nlist[i].n_type & N_EXT)
3790 if((nlist[i].n_type & N_TYPE) == N_UNDF
3791 && (nlist[i].n_value != 0))
3793 commonSize += nlist[i].n_value;
3796 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3801 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3802 "ocGetNames_MachO(oc->symbols)");
3806 for(i=0;i<symLC->nsyms;i++)
3808 if(nlist[i].n_type & N_STAB)
3810 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3812 if(nlist[i].n_type & N_EXT)
3814 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3815 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3817 + sections[nlist[i].n_sect-1].offset
3818 - sections[nlist[i].n_sect-1].addr
3819 + nlist[i].n_value);
3820 oc->symbols[curSymbol++] = nm;
3824 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3825 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
3827 + sections[nlist[i].n_sect-1].offset
3828 - sections[nlist[i].n_sect-1].addr
3829 + nlist[i].n_value);
3835 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
3836 commonCounter = (unsigned long)commonStorage;
3839 for(i=0;i<symLC->nsyms;i++)
3841 if((nlist[i].n_type & N_TYPE) == N_UNDF
3842 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
3844 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3845 unsigned long sz = nlist[i].n_value;
3847 nlist[i].n_value = commonCounter;
3849 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3850 (void*)commonCounter);
3851 oc->symbols[curSymbol++] = nm;
3853 commonCounter += sz;
3860 static int ocResolve_MachO(ObjectCode* oc)
3862 char *image = (char*) oc->image;
3863 struct mach_header *header = (struct mach_header*) image;
3864 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3866 struct segment_command *segLC = NULL;
3867 struct section *sections, *la_ptrs = NULL, *nl_ptrs = NULL;
3868 struct symtab_command *symLC = NULL;
3869 struct dysymtab_command *dsymLC = NULL;
3870 struct nlist *nlist;
3872 for(i=0;i<header->ncmds;i++)
3874 if(lc->cmd == LC_SEGMENT)
3875 segLC = (struct segment_command*) lc;
3876 else if(lc->cmd == LC_SYMTAB)
3877 symLC = (struct symtab_command*) lc;
3878 else if(lc->cmd == LC_DYSYMTAB)
3879 dsymLC = (struct dysymtab_command*) lc;
3880 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3883 sections = (struct section*) (segLC+1);
3884 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3887 for(i=0;i<segLC->nsects;i++)
3889 if(!strcmp(sections[i].sectname,"__la_symbol_ptr"))
3890 la_ptrs = §ions[i];
3891 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr"))
3892 nl_ptrs = §ions[i];
3897 unsigned long *indirectSyms
3898 = (unsigned long*) (image + dsymLC->indirectsymoff);
3901 if(!resolveImports(oc,image,symLC,la_ptrs,indirectSyms,nlist))
3904 if(!resolveImports(oc,image,symLC,nl_ptrs,indirectSyms,nlist))
3908 for(i=0;i<segLC->nsects;i++)
3910 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
3914 /* Free the local symbol table; we won't need it again. */
3915 freeHashTable(oc->lochash, NULL);
3918 #if defined (powerpc_HOST_ARCH)
3919 ocFlushInstructionCache( oc );
3926 * The Mach-O object format uses leading underscores. But not everywhere.
3927 * There is a small number of runtime support functions defined in
3928 * libcc_dynamic.a whose name does not have a leading underscore.
3929 * As a consequence, we can't get their address from C code.
3930 * We have to use inline assembler just to take the address of a function.
3934 static void machoInitSymbolsWithoutUnderscore()
3936 extern void* symbolsWithoutUnderscore[];
3937 void **p = symbolsWithoutUnderscore;
3938 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
3942 __asm__ volatile(".long " # x);
3944 RTS_MACHO_NOUNDERLINE_SYMBOLS
3946 __asm__ volatile(".text");
3950 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
3952 RTS_MACHO_NOUNDERLINE_SYMBOLS