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(".rdata",sectab_i->Name)||
1967 0==strcmp(".rodata",sectab_i->Name))
1968 kind = SECTIONKIND_CODE_OR_RODATA;
1969 if (0==strcmp(".data",sectab_i->Name) ||
1970 0==strcmp(".bss",sectab_i->Name))
1971 kind = SECTIONKIND_RWDATA;
1973 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
1974 sz = sectab_i->SizeOfRawData;
1975 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
1977 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
1978 end = start + sz - 1;
1980 if (kind == SECTIONKIND_OTHER
1981 /* Ignore sections called which contain stabs debugging
1983 && 0 != strcmp(".stab", sectab_i->Name)
1984 && 0 != strcmp(".stabstr", sectab_i->Name)
1986 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
1990 if (kind != SECTIONKIND_OTHER && end >= start) {
1991 addSection(oc, kind, start, end);
1992 addProddableBlock(oc, start, end - start + 1);
1996 /* Copy exported symbols into the ObjectCode. */
1998 oc->n_symbols = hdr->NumberOfSymbols;
1999 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2000 "ocGetNames_PEi386(oc->symbols)");
2001 /* Call me paranoid; I don't care. */
2002 for (i = 0; i < oc->n_symbols; i++)
2003 oc->symbols[i] = NULL;
2007 COFF_symbol* symtab_i;
2008 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2009 symtab_i = (COFF_symbol*)
2010 myindex ( sizeof_COFF_symbol, symtab, i );
2014 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2015 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2016 /* This symbol is global and defined, viz, exported */
2017 /* for MYIMAGE_SYMCLASS_EXTERNAL
2018 && !MYIMAGE_SYM_UNDEFINED,
2019 the address of the symbol is:
2020 address of relevant section + offset in section
2022 COFF_section* sectabent
2023 = (COFF_section*) myindex ( sizeof_COFF_section,
2025 symtab_i->SectionNumber-1 );
2026 addr = ((UChar*)(oc->image))
2027 + (sectabent->PointerToRawData
2031 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2032 && symtab_i->Value > 0) {
2033 /* This symbol isn't in any section at all, ie, global bss.
2034 Allocate zeroed space for it. */
2035 addr = stgCallocBytes(1, symtab_i->Value,
2036 "ocGetNames_PEi386(non-anonymous bss)");
2037 addSection(oc, SECTIONKIND_RWDATA, addr,
2038 ((UChar*)addr) + symtab_i->Value - 1);
2039 addProddableBlock(oc, addr, symtab_i->Value);
2040 /* debugBelch("BSS section at 0x%x\n", addr); */
2043 if (addr != NULL ) {
2044 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2045 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2046 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2047 ASSERT(i >= 0 && i < oc->n_symbols);
2048 /* cstring_from_COFF_symbol_name always succeeds. */
2049 oc->symbols[i] = sname;
2050 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2054 "IGNORING symbol %d\n"
2058 printName ( symtab_i->Name, strtab );
2067 (Int32)(symtab_i->SectionNumber),
2068 (UInt32)symtab_i->Type,
2069 (UInt32)symtab_i->StorageClass,
2070 (UInt32)symtab_i->NumberOfAuxSymbols
2075 i += symtab_i->NumberOfAuxSymbols;
2084 ocResolve_PEi386 ( ObjectCode* oc )
2087 COFF_section* sectab;
2088 COFF_symbol* symtab;
2098 /* ToDo: should be variable-sized? But is at least safe in the
2099 sense of buffer-overrun-proof. */
2101 /* debugBelch("resolving for %s\n", oc->fileName); */
2103 hdr = (COFF_header*)(oc->image);
2104 sectab = (COFF_section*) (
2105 ((UChar*)(oc->image))
2106 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2108 symtab = (COFF_symbol*) (
2109 ((UChar*)(oc->image))
2110 + hdr->PointerToSymbolTable
2112 strtab = ((UChar*)(oc->image))
2113 + hdr->PointerToSymbolTable
2114 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2116 for (i = 0; i < hdr->NumberOfSections; i++) {
2117 COFF_section* sectab_i
2119 myindex ( sizeof_COFF_section, sectab, i );
2122 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2125 /* Ignore sections called which contain stabs debugging
2127 if (0 == strcmp(".stab", sectab_i->Name)
2128 || 0 == strcmp(".stabstr", sectab_i->Name))
2131 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2132 /* If the relocation field (a short) has overflowed, the
2133 * real count can be found in the first reloc entry.
2135 * See Section 4.1 (last para) of the PE spec (rev6.0).
2137 * Nov2003 update: the GNU linker still doesn't correctly
2138 * handle the generation of relocatable object files with
2139 * overflown relocations. Hence the output to warn of potential
2142 COFF_reloc* rel = (COFF_reloc*)
2143 myindex ( sizeof_COFF_reloc, reltab, 0 );
2144 noRelocs = rel->VirtualAddress;
2145 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2149 noRelocs = sectab_i->NumberOfRelocations;
2154 for (; j < noRelocs; j++) {
2156 COFF_reloc* reltab_j
2158 myindex ( sizeof_COFF_reloc, reltab, j );
2160 /* the location to patch */
2162 ((UChar*)(oc->image))
2163 + (sectab_i->PointerToRawData
2164 + reltab_j->VirtualAddress
2165 - sectab_i->VirtualAddress )
2167 /* the existing contents of pP */
2169 /* the symbol to connect to */
2170 sym = (COFF_symbol*)
2171 myindex ( sizeof_COFF_symbol,
2172 symtab, reltab_j->SymbolTableIndex );
2175 "reloc sec %2d num %3d: type 0x%-4x "
2176 "vaddr 0x%-8x name `",
2178 (UInt32)reltab_j->Type,
2179 reltab_j->VirtualAddress );
2180 printName ( sym->Name, strtab );
2181 debugBelch("'\n" ));
2183 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2184 COFF_section* section_sym
2185 = findPEi386SectionCalled ( oc, sym->Name );
2187 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2190 S = ((UInt32)(oc->image))
2191 + (section_sym->PointerToRawData
2194 copyName ( sym->Name, strtab, symbol, 1000-1 );
2195 (void*)S = lookupLocalSymbol( oc, symbol );
2196 if ((void*)S != NULL) goto foundit;
2197 (void*)S = lookupSymbol( symbol );
2198 if ((void*)S != NULL) goto foundit;
2199 zapTrailingAtSign ( symbol );
2200 (void*)S = lookupLocalSymbol( oc, symbol );
2201 if ((void*)S != NULL) goto foundit;
2202 (void*)S = lookupSymbol( symbol );
2203 if ((void*)S != NULL) goto foundit;
2204 /* Newline first because the interactive linker has printed "linking..." */
2205 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2209 checkProddableBlock(oc, pP);
2210 switch (reltab_j->Type) {
2211 case MYIMAGE_REL_I386_DIR32:
2214 case MYIMAGE_REL_I386_REL32:
2215 /* Tricky. We have to insert a displacement at
2216 pP which, when added to the PC for the _next_
2217 insn, gives the address of the target (S).
2218 Problem is to know the address of the next insn
2219 when we only know pP. We assume that this
2220 literal field is always the last in the insn,
2221 so that the address of the next insn is pP+4
2222 -- hence the constant 4.
2223 Also I don't know if A should be added, but so
2224 far it has always been zero.
2227 *pP = S - ((UInt32)pP) - 4;
2230 debugBelch("%s: unhandled PEi386 relocation type %d",
2231 oc->fileName, reltab_j->Type);
2238 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2242 #endif /* defined(OBJFORMAT_PEi386) */
2245 /* --------------------------------------------------------------------------
2247 * ------------------------------------------------------------------------*/
2249 #if defined(OBJFORMAT_ELF)
2254 #if defined(sparc_HOST_ARCH)
2255 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2256 #elif defined(i386_HOST_ARCH)
2257 # define ELF_TARGET_386 /* Used inside <elf.h> */
2258 #elif defined(x86_64_HOST_ARCH)
2259 # define ELF_TARGET_X64_64
2261 #elif defined (ia64_HOST_ARCH)
2262 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2264 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2265 # define ELF_NEED_GOT /* needs Global Offset Table */
2266 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2269 #if !defined(openbsd_HOST_OS)
2272 /* openbsd elf has things in different places, with diff names */
2273 #include <elf_abi.h>
2274 #include <machine/reloc.h>
2275 #define R_386_32 RELOC_32
2276 #define R_386_PC32 RELOC_PC32
2280 * Define a set of types which can be used for both ELF32 and ELF64
2284 #define ELFCLASS ELFCLASS64
2285 #define Elf_Addr Elf64_Addr
2286 #define Elf_Word Elf64_Word
2287 #define Elf_Sword Elf64_Sword
2288 #define Elf_Ehdr Elf64_Ehdr
2289 #define Elf_Phdr Elf64_Phdr
2290 #define Elf_Shdr Elf64_Shdr
2291 #define Elf_Sym Elf64_Sym
2292 #define Elf_Rel Elf64_Rel
2293 #define Elf_Rela Elf64_Rela
2294 #define ELF_ST_TYPE ELF64_ST_TYPE
2295 #define ELF_ST_BIND ELF64_ST_BIND
2296 #define ELF_R_TYPE ELF64_R_TYPE
2297 #define ELF_R_SYM ELF64_R_SYM
2299 #define ELFCLASS ELFCLASS32
2300 #define Elf_Addr Elf32_Addr
2301 #define Elf_Word Elf32_Word
2302 #define Elf_Sword Elf32_Sword
2303 #define Elf_Ehdr Elf32_Ehdr
2304 #define Elf_Phdr Elf32_Phdr
2305 #define Elf_Shdr Elf32_Shdr
2306 #define Elf_Sym Elf32_Sym
2307 #define Elf_Rel Elf32_Rel
2308 #define Elf_Rela Elf32_Rela
2310 #define ELF_ST_TYPE ELF32_ST_TYPE
2313 #define ELF_ST_BIND ELF32_ST_BIND
2316 #define ELF_R_TYPE ELF32_R_TYPE
2319 #define ELF_R_SYM ELF32_R_SYM
2325 * Functions to allocate entries in dynamic sections. Currently we simply
2326 * preallocate a large number, and we don't check if a entry for the given
2327 * target already exists (a linear search is too slow). Ideally these
2328 * entries would be associated with symbols.
2331 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2332 #define GOT_SIZE 0x20000
2333 #define FUNCTION_TABLE_SIZE 0x10000
2334 #define PLT_SIZE 0x08000
2337 static Elf_Addr got[GOT_SIZE];
2338 static unsigned int gotIndex;
2339 static Elf_Addr gp_val = (Elf_Addr)got;
2342 allocateGOTEntry(Elf_Addr target)
2346 if (gotIndex >= GOT_SIZE)
2347 barf("Global offset table overflow");
2349 entry = &got[gotIndex++];
2351 return (Elf_Addr)entry;
2355 #ifdef ELF_FUNCTION_DESC
2361 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2362 static unsigned int functionTableIndex;
2365 allocateFunctionDesc(Elf_Addr target)
2367 FunctionDesc *entry;
2369 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2370 barf("Function table overflow");
2372 entry = &functionTable[functionTableIndex++];
2374 entry->gp = (Elf_Addr)gp_val;
2375 return (Elf_Addr)entry;
2379 copyFunctionDesc(Elf_Addr target)
2381 FunctionDesc *olddesc = (FunctionDesc *)target;
2382 FunctionDesc *newdesc;
2384 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2385 newdesc->gp = olddesc->gp;
2386 return (Elf_Addr)newdesc;
2391 #ifdef ia64_HOST_ARCH
2392 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2393 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2395 static unsigned char plt_code[] =
2397 /* taken from binutils bfd/elfxx-ia64.c */
2398 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2399 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2400 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2401 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2402 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2403 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2406 /* If we can't get to the function descriptor via gp, take a local copy of it */
2407 #define PLT_RELOC(code, target) { \
2408 Elf64_Sxword rel_value = target - gp_val; \
2409 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2410 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2412 ia64_reloc_gprel22((Elf_Addr)code, target); \
2417 unsigned char code[sizeof(plt_code)];
2421 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2423 PLTEntry *plt = (PLTEntry *)oc->plt;
2426 if (oc->pltIndex >= PLT_SIZE)
2427 barf("Procedure table overflow");
2429 entry = &plt[oc->pltIndex++];
2430 memcpy(entry->code, plt_code, sizeof(entry->code));
2431 PLT_RELOC(entry->code, target);
2432 return (Elf_Addr)entry;
2438 return (PLT_SIZE * sizeof(PLTEntry));
2444 * Generic ELF functions
2448 findElfSection ( void* objImage, Elf_Word sh_type )
2450 char* ehdrC = (char*)objImage;
2451 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2452 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2453 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2457 for (i = 0; i < ehdr->e_shnum; i++) {
2458 if (shdr[i].sh_type == sh_type
2459 /* Ignore the section header's string table. */
2460 && i != ehdr->e_shstrndx
2461 /* Ignore string tables named .stabstr, as they contain
2463 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2465 ptr = ehdrC + shdr[i].sh_offset;
2472 #if defined(ia64_HOST_ARCH)
2474 findElfSegment ( void* objImage, Elf_Addr vaddr )
2476 char* ehdrC = (char*)objImage;
2477 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2478 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2479 Elf_Addr segaddr = 0;
2482 for (i = 0; i < ehdr->e_phnum; i++) {
2483 segaddr = phdr[i].p_vaddr;
2484 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2492 ocVerifyImage_ELF ( ObjectCode* oc )
2496 int i, j, nent, nstrtab, nsymtabs;
2500 char* ehdrC = (char*)(oc->image);
2501 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2503 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2504 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2505 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2506 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2507 errorBelch("%s: not an ELF object", oc->fileName);
2511 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2512 errorBelch("%s: unsupported ELF format", oc->fileName);
2516 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2517 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2519 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2520 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2522 errorBelch("%s: unknown endiannness", oc->fileName);
2526 if (ehdr->e_type != ET_REL) {
2527 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2530 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2532 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2533 switch (ehdr->e_machine) {
2534 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2535 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2537 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2539 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2541 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2543 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2544 errorBelch("%s: unknown architecture", oc->fileName);
2548 IF_DEBUG(linker,debugBelch(
2549 "\nSection header table: start %d, n_entries %d, ent_size %d\n",
2550 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2552 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2554 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2556 if (ehdr->e_shstrndx == SHN_UNDEF) {
2557 errorBelch("%s: no section header string table", oc->fileName);
2560 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2562 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2565 for (i = 0; i < ehdr->e_shnum; i++) {
2566 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2567 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2568 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2569 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2570 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2571 ehdrC + shdr[i].sh_offset,
2572 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2574 if (shdr[i].sh_type == SHT_REL) {
2575 IF_DEBUG(linker,debugBelch("Rel " ));
2576 } else if (shdr[i].sh_type == SHT_RELA) {
2577 IF_DEBUG(linker,debugBelch("RelA " ));
2579 IF_DEBUG(linker,debugBelch(" "));
2582 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2586 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2589 for (i = 0; i < ehdr->e_shnum; i++) {
2590 if (shdr[i].sh_type == SHT_STRTAB
2591 /* Ignore the section header's string table. */
2592 && i != ehdr->e_shstrndx
2593 /* Ignore string tables named .stabstr, as they contain
2595 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2597 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2598 strtab = ehdrC + shdr[i].sh_offset;
2603 errorBelch("%s: no string tables, or too many", oc->fileName);
2608 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2609 for (i = 0; i < ehdr->e_shnum; i++) {
2610 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2611 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2613 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2614 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2615 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%d rem)\n",
2617 shdr[i].sh_size % sizeof(Elf_Sym)
2619 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2620 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2623 for (j = 0; j < nent; j++) {
2624 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2625 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
2626 (int)stab[j].st_shndx,
2627 (int)stab[j].st_size,
2628 (char*)stab[j].st_value ));
2630 IF_DEBUG(linker,debugBelch("type=" ));
2631 switch (ELF_ST_TYPE(stab[j].st_info)) {
2632 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
2633 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
2634 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
2635 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
2636 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
2637 default: IF_DEBUG(linker,debugBelch("? " )); break;
2639 IF_DEBUG(linker,debugBelch(" " ));
2641 IF_DEBUG(linker,debugBelch("bind=" ));
2642 switch (ELF_ST_BIND(stab[j].st_info)) {
2643 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
2644 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
2645 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
2646 default: IF_DEBUG(linker,debugBelch("? " )); break;
2648 IF_DEBUG(linker,debugBelch(" " ));
2650 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
2654 if (nsymtabs == 0) {
2655 errorBelch("%s: didn't find any symbol tables", oc->fileName);
2662 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
2666 if (hdr->sh_type == SHT_PROGBITS
2667 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
2668 /* .text-style section */
2669 return SECTIONKIND_CODE_OR_RODATA;
2672 if (hdr->sh_type == SHT_PROGBITS
2673 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2674 /* .data-style section */
2675 return SECTIONKIND_RWDATA;
2678 if (hdr->sh_type == SHT_PROGBITS
2679 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
2680 /* .rodata-style section */
2681 return SECTIONKIND_CODE_OR_RODATA;
2684 if (hdr->sh_type == SHT_NOBITS
2685 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2686 /* .bss-style section */
2688 return SECTIONKIND_RWDATA;
2691 return SECTIONKIND_OTHER;
2696 ocGetNames_ELF ( ObjectCode* oc )
2701 char* ehdrC = (char*)(oc->image);
2702 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2703 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
2704 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2706 ASSERT(symhash != NULL);
2709 errorBelch("%s: no strtab", oc->fileName);
2714 for (i = 0; i < ehdr->e_shnum; i++) {
2715 /* Figure out what kind of section it is. Logic derived from
2716 Figure 1.14 ("Special Sections") of the ELF document
2717 ("Portable Formats Specification, Version 1.1"). */
2719 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
2721 if (is_bss && shdr[i].sh_size > 0) {
2722 /* This is a non-empty .bss section. Allocate zeroed space for
2723 it, and set its .sh_offset field such that
2724 ehdrC + .sh_offset == addr_of_zeroed_space. */
2725 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
2726 "ocGetNames_ELF(BSS)");
2727 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
2729 debugBelch("BSS section at 0x%x, size %d\n",
2730 zspace, shdr[i].sh_size);
2734 /* fill in the section info */
2735 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
2736 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
2737 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
2738 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
2741 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2743 /* copy stuff into this module's object symbol table */
2744 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2745 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2747 oc->n_symbols = nent;
2748 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2749 "ocGetNames_ELF(oc->symbols)");
2751 for (j = 0; j < nent; j++) {
2753 char isLocal = FALSE; /* avoids uninit-var warning */
2755 char* nm = strtab + stab[j].st_name;
2756 int secno = stab[j].st_shndx;
2758 /* Figure out if we want to add it; if so, set ad to its
2759 address. Otherwise leave ad == NULL. */
2761 if (secno == SHN_COMMON) {
2763 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
2765 debugBelch("COMMON symbol, size %d name %s\n",
2766 stab[j].st_size, nm);
2768 /* Pointless to do addProddableBlock() for this area,
2769 since the linker should never poke around in it. */
2772 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
2773 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
2775 /* and not an undefined symbol */
2776 && stab[j].st_shndx != SHN_UNDEF
2777 /* and not in a "special section" */
2778 && stab[j].st_shndx < SHN_LORESERVE
2780 /* and it's a not a section or string table or anything silly */
2781 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
2782 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
2783 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
2786 /* Section 0 is the undefined section, hence > and not >=. */
2787 ASSERT(secno > 0 && secno < ehdr->e_shnum);
2789 if (shdr[secno].sh_type == SHT_NOBITS) {
2790 debugBelch(" BSS symbol, size %d off %d name %s\n",
2791 stab[j].st_size, stab[j].st_value, nm);
2794 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
2795 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
2798 #ifdef ELF_FUNCTION_DESC
2799 /* dlsym() and the initialisation table both give us function
2800 * descriptors, so to be consistent we store function descriptors
2801 * in the symbol table */
2802 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
2803 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
2805 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
2806 ad, oc->fileName, nm ));
2811 /* And the decision is ... */
2815 oc->symbols[j] = nm;
2818 /* Ignore entirely. */
2820 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
2824 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
2825 strtab + stab[j].st_name ));
2828 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
2829 (int)ELF_ST_BIND(stab[j].st_info),
2830 (int)ELF_ST_TYPE(stab[j].st_info),
2831 (int)stab[j].st_shndx,
2832 strtab + stab[j].st_name
2835 oc->symbols[j] = NULL;
2844 /* Do ELF relocations which lack an explicit addend. All x86-linux
2845 relocations appear to be of this form. */
2847 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
2848 Elf_Shdr* shdr, int shnum,
2849 Elf_Sym* stab, char* strtab )
2854 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
2855 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
2856 int target_shndx = shdr[shnum].sh_info;
2857 int symtab_shndx = shdr[shnum].sh_link;
2859 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2860 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
2861 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2862 target_shndx, symtab_shndx ));
2864 /* Skip sections that we're not interested in. */
2867 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
2868 if (kind == SECTIONKIND_OTHER) {
2869 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
2874 for (j = 0; j < nent; j++) {
2875 Elf_Addr offset = rtab[j].r_offset;
2876 Elf_Addr info = rtab[j].r_info;
2878 Elf_Addr P = ((Elf_Addr)targ) + offset;
2879 Elf_Word* pP = (Elf_Word*)P;
2885 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
2886 j, (void*)offset, (void*)info ));
2888 IF_DEBUG(linker,debugBelch( " ZERO" ));
2891 Elf_Sym sym = stab[ELF_R_SYM(info)];
2892 /* First see if it is a local symbol. */
2893 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2894 /* Yes, so we can get the address directly from the ELF symbol
2896 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2898 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2899 + stab[ELF_R_SYM(info)].st_value);
2902 /* No, so look up the name in our global table. */
2903 symbol = strtab + sym.st_name;
2904 S_tmp = lookupSymbol( symbol );
2905 S = (Elf_Addr)S_tmp;
2908 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2911 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
2914 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
2915 (void*)P, (void*)S, (void*)A ));
2916 checkProddableBlock ( oc, pP );
2920 switch (ELF_R_TYPE(info)) {
2921 # ifdef i386_HOST_ARCH
2922 case R_386_32: *pP = value; break;
2923 case R_386_PC32: *pP = value - P; break;
2926 errorBelch("%s: unhandled ELF relocation(Rel) type %d\n",
2927 oc->fileName, ELF_R_TYPE(info));
2935 /* Do ELF relocations for which explicit addends are supplied.
2936 sparc-solaris relocations appear to be of this form. */
2938 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
2939 Elf_Shdr* shdr, int shnum,
2940 Elf_Sym* stab, char* strtab )
2945 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
2946 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
2947 int target_shndx = shdr[shnum].sh_info;
2948 int symtab_shndx = shdr[shnum].sh_link;
2950 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
2951 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
2952 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
2953 target_shndx, symtab_shndx ));
2955 for (j = 0; j < nent; j++) {
2956 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
2957 /* This #ifdef only serves to avoid unused-var warnings. */
2958 Elf_Addr offset = rtab[j].r_offset;
2959 Elf_Addr P = targ + offset;
2961 Elf_Addr info = rtab[j].r_info;
2962 Elf_Addr A = rtab[j].r_addend;
2966 # if defined(sparc_HOST_ARCH)
2967 Elf_Word* pP = (Elf_Word*)P;
2969 # elif defined(ia64_HOST_ARCH)
2970 Elf64_Xword *pP = (Elf64_Xword *)P;
2972 # elif defined(powerpc_HOST_ARCH)
2976 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
2977 j, (void*)offset, (void*)info,
2980 IF_DEBUG(linker,debugBelch( " ZERO" ));
2983 Elf_Sym sym = stab[ELF_R_SYM(info)];
2984 /* First see if it is a local symbol. */
2985 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
2986 /* Yes, so we can get the address directly from the ELF symbol
2988 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
2990 (ehdrC + shdr[ sym.st_shndx ].sh_offset
2991 + stab[ELF_R_SYM(info)].st_value);
2992 #ifdef ELF_FUNCTION_DESC
2993 /* Make a function descriptor for this function */
2994 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
2995 S = allocateFunctionDesc(S + A);
3000 /* No, so look up the name in our global table. */
3001 symbol = strtab + sym.st_name;
3002 S_tmp = lookupSymbol( symbol );
3003 S = (Elf_Addr)S_tmp;
3005 #ifdef ELF_FUNCTION_DESC
3006 /* If a function, already a function descriptor - we would
3007 have to copy it to add an offset. */
3008 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3009 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3013 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3016 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3019 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3020 (void*)P, (void*)S, (void*)A ));
3021 /* checkProddableBlock ( oc, (void*)P ); */
3025 switch (ELF_R_TYPE(info)) {
3026 # if defined(sparc_HOST_ARCH)
3027 case R_SPARC_WDISP30:
3028 w1 = *pP & 0xC0000000;
3029 w2 = (Elf_Word)((value - P) >> 2);
3030 ASSERT((w2 & 0xC0000000) == 0);
3035 w1 = *pP & 0xFFC00000;
3036 w2 = (Elf_Word)(value >> 10);
3037 ASSERT((w2 & 0xFFC00000) == 0);
3043 w2 = (Elf_Word)(value & 0x3FF);
3044 ASSERT((w2 & ~0x3FF) == 0);
3048 /* According to the Sun documentation:
3050 This relocation type resembles R_SPARC_32, except it refers to an
3051 unaligned word. That is, the word to be relocated must be treated
3052 as four separate bytes with arbitrary alignment, not as a word
3053 aligned according to the architecture requirements.
3055 (JRS: which means that freeloading on the R_SPARC_32 case
3056 is probably wrong, but hey ...)
3060 w2 = (Elf_Word)value;
3063 # elif defined(ia64_HOST_ARCH)
3064 case R_IA64_DIR64LSB:
3065 case R_IA64_FPTR64LSB:
3068 case R_IA64_PCREL64LSB:
3071 case R_IA64_SEGREL64LSB:
3072 addr = findElfSegment(ehdrC, value);
3075 case R_IA64_GPREL22:
3076 ia64_reloc_gprel22(P, value);
3078 case R_IA64_LTOFF22:
3079 case R_IA64_LTOFF22X:
3080 case R_IA64_LTOFF_FPTR22:
3081 addr = allocateGOTEntry(value);
3082 ia64_reloc_gprel22(P, addr);
3084 case R_IA64_PCREL21B:
3085 ia64_reloc_pcrel21(P, S, oc);
3088 /* This goes with R_IA64_LTOFF22X and points to the load to
3089 * convert into a move. We don't implement relaxation. */
3091 # elif defined(powerpc_HOST_ARCH)
3092 case R_PPC_ADDR16_LO:
3093 *(Elf32_Half*) P = value;
3096 case R_PPC_ADDR16_HI:
3097 *(Elf32_Half*) P = value >> 16;
3100 case R_PPC_ADDR16_HA:
3101 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3105 *(Elf32_Word *) P = value;
3109 *(Elf32_Word *) P = value - P;
3115 if( delta << 6 >> 6 != delta )
3117 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3120 if( value == 0 || delta << 6 >> 6 != delta )
3122 barf( "Unable to make ppcJumpIsland for #%d",
3128 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3129 | (delta & 0x3fffffc);
3135 *(Elf64_Xword *)P = value;
3139 *(Elf64_Word *)P = (Elf64_Word) (value - P);
3143 *(Elf64_Word *)P = (Elf64_Word)value;
3147 *(Elf64_Sword *)P = (Elf64_Sword)value;
3152 errorBelch("%s: unhandled ELF relocation(RelA) type %d\n",
3153 oc->fileName, ELF_R_TYPE(info));
3162 ocResolve_ELF ( ObjectCode* oc )
3166 Elf_Sym* stab = NULL;
3167 char* ehdrC = (char*)(oc->image);
3168 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3169 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3171 /* first find "the" symbol table */
3172 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3174 /* also go find the string table */
3175 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3177 if (stab == NULL || strtab == NULL) {
3178 errorBelch("%s: can't find string or symbol table", oc->fileName);
3182 /* Process the relocation sections. */
3183 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3184 if (shdr[shnum].sh_type == SHT_REL) {
3185 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3186 shnum, stab, strtab );
3190 if (shdr[shnum].sh_type == SHT_RELA) {
3191 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3192 shnum, stab, strtab );
3197 /* Free the local symbol table; we won't need it again. */
3198 freeHashTable(oc->lochash, NULL);
3201 #if defined(powerpc_HOST_ARCH)
3202 ocFlushInstructionCache( oc );
3210 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3211 * at the front. The following utility functions pack and unpack instructions, and
3212 * take care of the most common relocations.
3215 #ifdef ia64_HOST_ARCH
3218 ia64_extract_instruction(Elf64_Xword *target)
3221 int slot = (Elf_Addr)target & 3;
3222 (Elf_Addr)target &= ~3;
3230 return ((w1 >> 5) & 0x1ffffffffff);
3232 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3236 barf("ia64_extract_instruction: invalid slot %p", target);
3241 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3243 int slot = (Elf_Addr)target & 3;
3244 (Elf_Addr)target &= ~3;
3249 *target |= value << 5;
3252 *target |= value << 46;
3253 *(target+1) |= value >> 18;
3256 *(target+1) |= value << 23;
3262 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3264 Elf64_Xword instruction;
3265 Elf64_Sxword rel_value;
3267 rel_value = value - gp_val;
3268 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3269 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3271 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3272 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3273 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3274 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3275 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3276 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3280 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3282 Elf64_Xword instruction;
3283 Elf64_Sxword rel_value;
3286 entry = allocatePLTEntry(value, oc);
3288 rel_value = (entry >> 4) - (target >> 4);
3289 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3290 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3292 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3293 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3294 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3295 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3301 * PowerPC ELF specifics
3304 #ifdef powerpc_HOST_ARCH
3306 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3312 ehdr = (Elf_Ehdr *) oc->image;
3313 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3315 for( i = 0; i < ehdr->e_shnum; i++ )
3316 if( shdr[i].sh_type == SHT_SYMTAB )
3319 if( i == ehdr->e_shnum )
3321 errorBelch( "This ELF file contains no symtab" );
3325 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3327 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3328 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3333 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3336 #endif /* powerpc */
3340 /* --------------------------------------------------------------------------
3342 * ------------------------------------------------------------------------*/
3344 #if defined(OBJFORMAT_MACHO)
3347 Support for MachO linking on Darwin/MacOS X on PowerPC chips
3348 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3350 I hereby formally apologize for the hackish nature of this code.
3351 Things that need to be done:
3352 *) implement ocVerifyImage_MachO
3353 *) add still more sanity checks.
3356 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3358 struct mach_header *header = (struct mach_header *) oc->image;
3359 struct load_command *lc = (struct load_command *) (header + 1);
3362 for( i = 0; i < header->ncmds; i++ )
3364 if( lc->cmd == LC_SYMTAB )
3366 // Find out the first and last undefined external
3367 // symbol, so we don't have to allocate too many
3369 struct symtab_command *symLC = (struct symtab_command *) lc;
3370 unsigned min = symLC->nsyms, max = 0;
3371 struct nlist *nlist =
3372 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3374 for(i=0;i<symLC->nsyms;i++)
3376 if(nlist[i].n_type & N_STAB)
3378 else if(nlist[i].n_type & N_EXT)
3380 if((nlist[i].n_type & N_TYPE) == N_UNDF
3381 && (nlist[i].n_value == 0))
3391 return ocAllocateJumpIslands(oc, max - min + 1, min);
3396 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3398 return ocAllocateJumpIslands(oc,0,0);
3401 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3403 // FIXME: do some verifying here
3407 static int resolveImports(
3410 struct symtab_command *symLC,
3411 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3412 unsigned long *indirectSyms,
3413 struct nlist *nlist)
3417 for(i=0;i*4<sect->size;i++)
3419 // according to otool, reserved1 contains the first index into the indirect symbol table
3420 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3421 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3424 if((symbol->n_type & N_TYPE) == N_UNDF
3425 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3426 addr = (void*) (symbol->n_value);
3427 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3430 addr = lookupSymbol(nm);
3433 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3437 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3438 ((void**)(image + sect->offset))[i] = addr;
3444 static unsigned long relocateAddress(
3447 struct section* sections,
3448 unsigned long address)
3451 for(i = 0; i < nSections; i++)
3453 if(sections[i].addr <= address
3454 && address < sections[i].addr + sections[i].size)
3456 return (unsigned long)oc->image
3457 + sections[i].offset + address - sections[i].addr;
3460 barf("Invalid Mach-O file:"
3461 "Address out of bounds while relocating object file");
3465 static int relocateSection(
3468 struct symtab_command *symLC, struct nlist *nlist,
3469 int nSections, struct section* sections, struct section *sect)
3471 struct relocation_info *relocs;
3474 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3476 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3480 relocs = (struct relocation_info*) (image + sect->reloff);
3484 if(relocs[i].r_address & R_SCATTERED)
3486 struct scattered_relocation_info *scat =
3487 (struct scattered_relocation_info*) &relocs[i];
3491 if(scat->r_length == 2)
3493 unsigned long word = 0;
3494 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3495 checkProddableBlock(oc,wordPtr);
3497 // Step 1: Figure out what the relocated value should be
3498 if(scat->r_type == GENERIC_RELOC_VANILLA)
3500 word = *wordPtr + (unsigned long) relocateAddress(
3507 else if(scat->r_type == PPC_RELOC_SECTDIFF
3508 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3509 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3510 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3512 struct scattered_relocation_info *pair =
3513 (struct scattered_relocation_info*) &relocs[i+1];
3515 if(!pair->r_scattered || pair->r_type != PPC_RELOC_PAIR)
3516 barf("Invalid Mach-O file: "
3517 "PPC_RELOC_*_SECTDIFF not followed by PPC_RELOC_PAIR");
3519 word = (unsigned long)
3520 (relocateAddress(oc, nSections, sections, scat->r_value)
3521 - relocateAddress(oc, nSections, sections, pair->r_value));
3524 else if(scat->r_type == PPC_RELOC_HI16
3525 || scat->r_type == PPC_RELOC_LO16
3526 || scat->r_type == PPC_RELOC_HA16
3527 || scat->r_type == PPC_RELOC_LO14)
3528 { // these are generated by label+offset things
3529 struct relocation_info *pair = &relocs[i+1];
3530 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3531 barf("Invalid Mach-O file: "
3532 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3534 if(scat->r_type == PPC_RELOC_LO16)
3536 word = ((unsigned short*) wordPtr)[1];
3537 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3539 else if(scat->r_type == PPC_RELOC_LO14)
3541 barf("Unsupported Relocation: PPC_RELOC_LO14");
3542 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3543 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3545 else if(scat->r_type == PPC_RELOC_HI16)
3547 word = ((unsigned short*) wordPtr)[1] << 16;
3548 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3550 else if(scat->r_type == PPC_RELOC_HA16)
3552 word = ((unsigned short*) wordPtr)[1] << 16;
3553 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3557 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
3563 continue; // ignore the others
3565 if(scat->r_type == GENERIC_RELOC_VANILLA
3566 || scat->r_type == PPC_RELOC_SECTDIFF)
3570 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
3572 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3574 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
3576 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3578 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
3580 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3581 + ((word & (1<<15)) ? 1 : 0);
3586 continue; // FIXME: I hope it's OK to ignore all the others.
3590 struct relocation_info *reloc = &relocs[i];
3591 if(reloc->r_pcrel && !reloc->r_extern)
3594 if(reloc->r_length == 2)
3596 unsigned long word = 0;
3597 unsigned long jumpIsland = 0;
3598 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
3599 // to avoid warning and to catch
3602 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
3603 checkProddableBlock(oc,wordPtr);
3605 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3609 else if(reloc->r_type == PPC_RELOC_LO16)
3611 word = ((unsigned short*) wordPtr)[1];
3612 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3614 else if(reloc->r_type == PPC_RELOC_HI16)
3616 word = ((unsigned short*) wordPtr)[1] << 16;
3617 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3619 else if(reloc->r_type == PPC_RELOC_HA16)
3621 word = ((unsigned short*) wordPtr)[1] << 16;
3622 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3624 else if(reloc->r_type == PPC_RELOC_BR24)
3627 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
3631 if(!reloc->r_extern)
3634 sections[reloc->r_symbolnum-1].offset
3635 - sections[reloc->r_symbolnum-1].addr
3642 struct nlist *symbol = &nlist[reloc->r_symbolnum];
3643 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3644 void *symbolAddress = lookupSymbol(nm);
3647 errorBelch("\nunknown symbol `%s'", nm);
3653 // In the .o file, this should be a relative jump to NULL
3654 // and we'll change it to a jump to a relative jump to the symbol
3655 ASSERT(-word == reloc->r_address);
3656 word = (unsigned long) symbolAddress;
3657 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,word);
3658 word -= ((long)image) + sect->offset + reloc->r_address;
3661 offsetToJumpIsland = jumpIsland
3662 - (((long)image) + sect->offset + reloc->r_address);
3667 word += (unsigned long) symbolAddress;
3671 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3676 else if(reloc->r_type == PPC_RELOC_LO16)
3678 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3681 else if(reloc->r_type == PPC_RELOC_HI16)
3683 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3686 else if(reloc->r_type == PPC_RELOC_HA16)
3688 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3689 + ((word & (1<<15)) ? 1 : 0);
3692 else if(reloc->r_type == PPC_RELOC_BR24)
3694 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3696 // The branch offset is too large.
3697 // Therefore, we try to use a jump island.
3700 barf("unconditional relative branch out of range: "
3701 "no jump island available");
3704 word = offsetToJumpIsland;
3705 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3706 barf("unconditional relative branch out of range: "
3707 "jump island out of range");
3709 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
3713 barf("\nunknown relocation %d",reloc->r_type);
3720 static int ocGetNames_MachO(ObjectCode* oc)
3722 char *image = (char*) oc->image;
3723 struct mach_header *header = (struct mach_header*) image;
3724 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3725 unsigned i,curSymbol = 0;
3726 struct segment_command *segLC = NULL;
3727 struct section *sections;
3728 struct symtab_command *symLC = NULL;
3729 struct nlist *nlist;
3730 unsigned long commonSize = 0;
3731 char *commonStorage = NULL;
3732 unsigned long commonCounter;
3734 for(i=0;i<header->ncmds;i++)
3736 if(lc->cmd == LC_SEGMENT)
3737 segLC = (struct segment_command*) lc;
3738 else if(lc->cmd == LC_SYMTAB)
3739 symLC = (struct symtab_command*) lc;
3740 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3743 sections = (struct section*) (segLC+1);
3744 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3747 for(i=0;i<segLC->nsects;i++)
3749 if(sections[i].size == 0)
3752 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
3754 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
3755 "ocGetNames_MachO(common symbols)");
3756 sections[i].offset = zeroFillArea - image;
3759 if(!strcmp(sections[i].sectname,"__text"))
3760 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
3761 (void*) (image + sections[i].offset),
3762 (void*) (image + sections[i].offset + sections[i].size));
3763 else if(!strcmp(sections[i].sectname,"__const"))
3764 addSection(oc, SECTIONKIND_RWDATA,
3765 (void*) (image + sections[i].offset),
3766 (void*) (image + sections[i].offset + sections[i].size));
3767 else if(!strcmp(sections[i].sectname,"__data"))
3768 addSection(oc, SECTIONKIND_RWDATA,
3769 (void*) (image + sections[i].offset),
3770 (void*) (image + sections[i].offset + sections[i].size));
3771 else if(!strcmp(sections[i].sectname,"__bss")
3772 || !strcmp(sections[i].sectname,"__common"))
3773 addSection(oc, SECTIONKIND_RWDATA,
3774 (void*) (image + sections[i].offset),
3775 (void*) (image + sections[i].offset + sections[i].size));
3777 addProddableBlock(oc, (void*) (image + sections[i].offset),
3781 // count external symbols defined here
3785 for(i=0;i<symLC->nsyms;i++)
3787 if(nlist[i].n_type & N_STAB)
3789 else if(nlist[i].n_type & N_EXT)
3791 if((nlist[i].n_type & N_TYPE) == N_UNDF
3792 && (nlist[i].n_value != 0))
3794 commonSize += nlist[i].n_value;
3797 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3802 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3803 "ocGetNames_MachO(oc->symbols)");
3807 for(i=0;i<symLC->nsyms;i++)
3809 if(nlist[i].n_type & N_STAB)
3811 else if((nlist[i].n_type & N_TYPE) == N_SECT)
3813 if(nlist[i].n_type & N_EXT)
3815 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3816 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3818 + sections[nlist[i].n_sect-1].offset
3819 - sections[nlist[i].n_sect-1].addr
3820 + nlist[i].n_value);
3821 oc->symbols[curSymbol++] = nm;
3825 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3826 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
3828 + sections[nlist[i].n_sect-1].offset
3829 - sections[nlist[i].n_sect-1].addr
3830 + nlist[i].n_value);
3836 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
3837 commonCounter = (unsigned long)commonStorage;
3840 for(i=0;i<symLC->nsyms;i++)
3842 if((nlist[i].n_type & N_TYPE) == N_UNDF
3843 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
3845 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
3846 unsigned long sz = nlist[i].n_value;
3848 nlist[i].n_value = commonCounter;
3850 ghciInsertStrHashTable(oc->fileName, symhash, nm,
3851 (void*)commonCounter);
3852 oc->symbols[curSymbol++] = nm;
3854 commonCounter += sz;
3861 static int ocResolve_MachO(ObjectCode* oc)
3863 char *image = (char*) oc->image;
3864 struct mach_header *header = (struct mach_header*) image;
3865 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3867 struct segment_command *segLC = NULL;
3868 struct section *sections, *la_ptrs = NULL, *nl_ptrs = NULL;
3869 struct symtab_command *symLC = NULL;
3870 struct dysymtab_command *dsymLC = NULL;
3871 struct nlist *nlist;
3873 for(i=0;i<header->ncmds;i++)
3875 if(lc->cmd == LC_SEGMENT)
3876 segLC = (struct segment_command*) lc;
3877 else if(lc->cmd == LC_SYMTAB)
3878 symLC = (struct symtab_command*) lc;
3879 else if(lc->cmd == LC_DYSYMTAB)
3880 dsymLC = (struct dysymtab_command*) lc;
3881 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3884 sections = (struct section*) (segLC+1);
3885 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3888 for(i=0;i<segLC->nsects;i++)
3890 if(!strcmp(sections[i].sectname,"__la_symbol_ptr"))
3891 la_ptrs = §ions[i];
3892 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr"))
3893 nl_ptrs = §ions[i];
3898 unsigned long *indirectSyms
3899 = (unsigned long*) (image + dsymLC->indirectsymoff);
3902 if(!resolveImports(oc,image,symLC,la_ptrs,indirectSyms,nlist))
3905 if(!resolveImports(oc,image,symLC,nl_ptrs,indirectSyms,nlist))
3909 for(i=0;i<segLC->nsects;i++)
3911 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
3915 /* Free the local symbol table; we won't need it again. */
3916 freeHashTable(oc->lochash, NULL);
3919 #if defined (powerpc_HOST_ARCH)
3920 ocFlushInstructionCache( oc );
3927 * The Mach-O object format uses leading underscores. But not everywhere.
3928 * There is a small number of runtime support functions defined in
3929 * libcc_dynamic.a whose name does not have a leading underscore.
3930 * As a consequence, we can't get their address from C code.
3931 * We have to use inline assembler just to take the address of a function.
3935 static void machoInitSymbolsWithoutUnderscore()
3937 extern void* symbolsWithoutUnderscore[];
3938 void **p = symbolsWithoutUnderscore;
3939 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
3943 __asm__ volatile(".long " # x);
3945 RTS_MACHO_NOUNDERLINE_SYMBOLS
3947 __asm__ volatile(".text");
3951 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
3953 RTS_MACHO_NOUNDERLINE_SYMBOLS