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 #include "RtsTypeable.h"
32 #ifdef HAVE_SYS_TYPES_H
33 #include <sys/types.h>
39 #ifdef HAVE_SYS_STAT_H
43 #if defined(HAVE_DLFCN_H)
47 #if defined(cygwin32_HOST_OS)
52 #ifdef HAVE_SYS_TIME_H
56 #include <sys/fcntl.h>
57 #include <sys/termios.h>
58 #include <sys/utime.h>
59 #include <sys/utsname.h>
63 #if defined(ia64_HOST_ARCH) || defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
68 #if defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
76 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
77 # define OBJFORMAT_ELF
78 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
79 # define OBJFORMAT_PEi386
82 #elif defined(darwin_HOST_OS)
83 # define OBJFORMAT_MACHO
84 # include <mach-o/loader.h>
85 # include <mach-o/nlist.h>
86 # include <mach-o/reloc.h>
87 # include <mach-o/dyld.h>
88 #if defined(powerpc_HOST_ARCH)
89 # include <mach-o/ppc/reloc.h>
93 /* Hash table mapping symbol names to Symbol */
94 static /*Str*/HashTable *symhash;
100 /* Hash table mapping symbol names to StgStablePtr */
101 static /*Str*/HashTable *stablehash;
102 rootEntry *root_ptr_table = NULL;
103 static rootEntry *root_ptr_free = NULL;
105 static unsigned int RPT_size = 0;
107 /* List of currently loaded objects */
108 ObjectCode *objects = NULL; /* initially empty */
110 #if defined(OBJFORMAT_ELF)
111 static int ocVerifyImage_ELF ( ObjectCode* oc );
112 static int ocGetNames_ELF ( ObjectCode* oc );
113 static int ocResolve_ELF ( ObjectCode* oc );
114 #if defined(powerpc_HOST_ARCH)
115 static int ocAllocateJumpIslands_ELF ( ObjectCode* oc );
117 #elif defined(OBJFORMAT_PEi386)
118 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
119 static int ocGetNames_PEi386 ( ObjectCode* oc );
120 static int ocResolve_PEi386 ( ObjectCode* oc );
121 #elif defined(OBJFORMAT_MACHO)
122 static int ocVerifyImage_MachO ( ObjectCode* oc );
123 static int ocGetNames_MachO ( ObjectCode* oc );
124 static int ocResolve_MachO ( ObjectCode* oc );
126 static int machoGetMisalignment( FILE * );
127 #ifdef powerpc_HOST_ARCH
128 static int ocAllocateJumpIslands_MachO ( ObjectCode* oc );
129 static void machoInitSymbolsWithoutUnderscore( void );
133 #if defined(x86_64_HOST_ARCH)
134 static void*x86_64_high_symbol( char *lbl, void *addr );
137 /* -----------------------------------------------------------------------------
138 * Built-in symbols from the RTS
141 typedef struct _RtsSymbolVal {
148 #define Maybe_Stable_Names SymX(mkWeakzh_fast) \
149 SymX(makeStableNamezh_fast) \
150 SymX(finalizzeWeakzh_fast)
152 /* These are not available in GUM!!! -- HWL */
153 #define Maybe_Stable_Names
156 #if !defined (mingw32_HOST_OS)
157 #define RTS_POSIX_ONLY_SYMBOLS \
158 SymX(signal_handlers) \
159 SymX(stg_sig_install) \
163 #if defined (cygwin32_HOST_OS)
164 #define RTS_MINGW_ONLY_SYMBOLS /**/
165 /* Don't have the ability to read import libs / archives, so
166 * we have to stupidly list a lot of what libcygwin.a
169 #define RTS_CYGWIN_ONLY_SYMBOLS \
247 #elif !defined(mingw32_HOST_OS)
248 #define RTS_MINGW_ONLY_SYMBOLS /**/
249 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
250 #else /* defined(mingw32_HOST_OS) */
251 #define RTS_POSIX_ONLY_SYMBOLS /**/
252 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
254 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
256 #define RTS_MINGW_EXTRA_SYMS \
257 Sym(_imp____mb_cur_max) \
260 #define RTS_MINGW_EXTRA_SYMS
263 /* These are statically linked from the mingw libraries into the ghc
264 executable, so we have to employ this hack. */
265 #define RTS_MINGW_ONLY_SYMBOLS \
266 SymX(asyncReadzh_fast) \
267 SymX(asyncWritezh_fast) \
268 SymX(asyncDoProczh_fast) \
280 SymX(getservbyname) \
281 SymX(getservbyport) \
282 SymX(getprotobynumber) \
283 SymX(getprotobyname) \
284 SymX(gethostbyname) \
285 SymX(gethostbyaddr) \
332 SymX(rts_InstallConsoleEvent) \
333 SymX(rts_ConsoleHandlerDone) \
335 Sym(_imp___timezone) \
345 RTS_MINGW_EXTRA_SYMS \
349 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
350 #define RTS_DARWIN_ONLY_SYMBOLS \
351 Sym(asprintf$LDBLStub) \
355 Sym(fprintf$LDBLStub) \
356 Sym(fscanf$LDBLStub) \
357 Sym(fwprintf$LDBLStub) \
358 Sym(fwscanf$LDBLStub) \
359 Sym(printf$LDBLStub) \
360 Sym(scanf$LDBLStub) \
361 Sym(snprintf$LDBLStub) \
362 Sym(sprintf$LDBLStub) \
363 Sym(sscanf$LDBLStub) \
364 Sym(strtold$LDBLStub) \
365 Sym(swprintf$LDBLStub) \
366 Sym(swscanf$LDBLStub) \
367 Sym(syslog$LDBLStub) \
368 Sym(vasprintf$LDBLStub) \
370 Sym(verrc$LDBLStub) \
371 Sym(verrx$LDBLStub) \
372 Sym(vfprintf$LDBLStub) \
373 Sym(vfscanf$LDBLStub) \
374 Sym(vfwprintf$LDBLStub) \
375 Sym(vfwscanf$LDBLStub) \
376 Sym(vprintf$LDBLStub) \
377 Sym(vscanf$LDBLStub) \
378 Sym(vsnprintf$LDBLStub) \
379 Sym(vsprintf$LDBLStub) \
380 Sym(vsscanf$LDBLStub) \
381 Sym(vswprintf$LDBLStub) \
382 Sym(vswscanf$LDBLStub) \
383 Sym(vsyslog$LDBLStub) \
384 Sym(vwarn$LDBLStub) \
385 Sym(vwarnc$LDBLStub) \
386 Sym(vwarnx$LDBLStub) \
387 Sym(vwprintf$LDBLStub) \
388 Sym(vwscanf$LDBLStub) \
390 Sym(warnc$LDBLStub) \
391 Sym(warnx$LDBLStub) \
392 Sym(wcstold$LDBLStub) \
393 Sym(wprintf$LDBLStub) \
396 #define RTS_DARWIN_ONLY_SYMBOLS
400 # define MAIN_CAP_SYM SymX(MainCapability)
402 # define MAIN_CAP_SYM
405 #if !defined(mingw32_HOST_OS)
406 #define RTS_USER_SIGNALS_SYMBOLS \
407 SymX(setIOManagerPipe)
409 #define RTS_USER_SIGNALS_SYMBOLS /* nothing */
412 #ifdef TABLES_NEXT_TO_CODE
413 #define RTS_RET_SYMBOLS /* nothing */
415 #define RTS_RET_SYMBOLS \
416 SymX(stg_enter_ret) \
417 SymX(stg_gc_fun_ret) \
424 SymX(stg_ap_pv_ret) \
425 SymX(stg_ap_pp_ret) \
426 SymX(stg_ap_ppv_ret) \
427 SymX(stg_ap_ppp_ret) \
428 SymX(stg_ap_pppv_ret) \
429 SymX(stg_ap_pppp_ret) \
430 SymX(stg_ap_ppppp_ret) \
431 SymX(stg_ap_pppppp_ret)
434 #define RTS_SYMBOLS \
437 SymX(stg_enter_info) \
438 SymX(stg_gc_void_info) \
439 SymX(__stg_gc_enter_1) \
440 SymX(stg_gc_noregs) \
441 SymX(stg_gc_unpt_r1_info) \
442 SymX(stg_gc_unpt_r1) \
443 SymX(stg_gc_unbx_r1_info) \
444 SymX(stg_gc_unbx_r1) \
445 SymX(stg_gc_f1_info) \
447 SymX(stg_gc_d1_info) \
449 SymX(stg_gc_l1_info) \
452 SymX(stg_gc_fun_info) \
454 SymX(stg_gc_gen_info) \
455 SymX(stg_gc_gen_hp) \
457 SymX(stg_gen_yield) \
458 SymX(stg_yield_noregs) \
459 SymX(stg_yield_to_interpreter) \
460 SymX(stg_gen_block) \
461 SymX(stg_block_noregs) \
463 SymX(stg_block_takemvar) \
464 SymX(stg_block_putmvar) \
465 SymX(stg_seq_frame_info) \
467 SymX(MallocFailHook) \
469 SymX(OutOfHeapHook) \
470 SymX(StackOverflowHook) \
471 SymX(__encodeDouble) \
472 SymX(__encodeFloat) \
476 SymX(__gmpz_cmp_si) \
477 SymX(__gmpz_cmp_ui) \
478 SymX(__gmpz_get_si) \
479 SymX(__gmpz_get_ui) \
480 SymX(__int_encodeDouble) \
481 SymX(__int_encodeFloat) \
482 SymX(andIntegerzh_fast) \
483 SymX(atomicallyzh_fast) \
487 SymX(blockAsyncExceptionszh_fast) \
489 SymX(catchRetryzh_fast) \
490 SymX(catchSTMzh_fast) \
491 SymX(closure_flags) \
493 SymX(cmpIntegerzh_fast) \
494 SymX(cmpIntegerIntzh_fast) \
495 SymX(complementIntegerzh_fast) \
496 SymX(createAdjustor) \
497 SymX(decodeDoublezh_fast) \
498 SymX(decodeFloatzh_fast) \
501 SymX(deRefWeakzh_fast) \
502 SymX(deRefStablePtrzh_fast) \
503 SymX(dirty_MUT_VAR) \
504 SymX(divExactIntegerzh_fast) \
505 SymX(divModIntegerzh_fast) \
507 SymX(forkOnzh_fast) \
509 SymX(forkOS_createThread) \
510 SymX(freeHaskellFunctionPtr) \
511 SymX(freeStablePtr) \
512 SymX(getOrSetTypeableStore) \
513 SymX(gcdIntegerzh_fast) \
514 SymX(gcdIntegerIntzh_fast) \
515 SymX(gcdIntzh_fast) \
524 SymX(hs_perform_gc) \
525 SymX(hs_free_stable_ptr) \
526 SymX(hs_free_fun_ptr) \
528 SymX(int2Integerzh_fast) \
529 SymX(integer2Intzh_fast) \
530 SymX(integer2Wordzh_fast) \
531 SymX(isCurrentThreadBoundzh_fast) \
532 SymX(isDoubleDenormalized) \
533 SymX(isDoubleInfinite) \
535 SymX(isDoubleNegativeZero) \
536 SymX(isEmptyMVarzh_fast) \
537 SymX(isFloatDenormalized) \
538 SymX(isFloatInfinite) \
540 SymX(isFloatNegativeZero) \
541 SymX(killThreadzh_fast) \
543 SymX(insertStableSymbol) \
546 SymX(makeStablePtrzh_fast) \
547 SymX(minusIntegerzh_fast) \
548 SymX(mkApUpd0zh_fast) \
549 SymX(myThreadIdzh_fast) \
550 SymX(labelThreadzh_fast) \
551 SymX(newArrayzh_fast) \
552 SymX(newBCOzh_fast) \
553 SymX(newByteArrayzh_fast) \
554 SymX_redirect(newCAF, newDynCAF) \
555 SymX(newMVarzh_fast) \
556 SymX(newMutVarzh_fast) \
557 SymX(newTVarzh_fast) \
558 SymX(atomicModifyMutVarzh_fast) \
559 SymX(newPinnedByteArrayzh_fast) \
561 SymX(orIntegerzh_fast) \
563 SymX(performMajorGC) \
564 SymX(plusIntegerzh_fast) \
567 SymX(putMVarzh_fast) \
568 SymX(quotIntegerzh_fast) \
569 SymX(quotRemIntegerzh_fast) \
571 SymX(raiseIOzh_fast) \
572 SymX(readTVarzh_fast) \
573 SymX(remIntegerzh_fast) \
574 SymX(resetNonBlockingFd) \
579 SymX(rts_checkSchedStatus) \
582 SymX(rts_evalLazyIO) \
583 SymX(rts_evalStableIO) \
587 SymX(rts_getDouble) \
592 SymX(rts_getFunPtr) \
593 SymX(rts_getStablePtr) \
594 SymX(rts_getThreadId) \
596 SymX(rts_getWord32) \
609 SymX(rts_mkStablePtr) \
617 SymX(rtsSupportsBoundThreads) \
618 SymX(__hscore_get_saved_termios) \
619 SymX(__hscore_set_saved_termios) \
621 SymX(startupHaskell) \
622 SymX(shutdownHaskell) \
623 SymX(shutdownHaskellAndExit) \
624 SymX(stable_ptr_table) \
625 SymX(stackOverflow) \
626 SymX(stg_CAF_BLACKHOLE_info) \
627 SymX(awakenBlockedQueue) \
628 SymX(stg_CHARLIKE_closure) \
629 SymX(stg_EMPTY_MVAR_info) \
630 SymX(stg_IND_STATIC_info) \
631 SymX(stg_INTLIKE_closure) \
632 SymX(stg_MUT_ARR_PTRS_DIRTY_info) \
633 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
634 SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \
635 SymX(stg_WEAK_info) \
636 SymX(stg_ap_v_info) \
637 SymX(stg_ap_f_info) \
638 SymX(stg_ap_d_info) \
639 SymX(stg_ap_l_info) \
640 SymX(stg_ap_n_info) \
641 SymX(stg_ap_p_info) \
642 SymX(stg_ap_pv_info) \
643 SymX(stg_ap_pp_info) \
644 SymX(stg_ap_ppv_info) \
645 SymX(stg_ap_ppp_info) \
646 SymX(stg_ap_pppv_info) \
647 SymX(stg_ap_pppp_info) \
648 SymX(stg_ap_ppppp_info) \
649 SymX(stg_ap_pppppp_info) \
650 SymX(stg_ap_0_fast) \
651 SymX(stg_ap_v_fast) \
652 SymX(stg_ap_f_fast) \
653 SymX(stg_ap_d_fast) \
654 SymX(stg_ap_l_fast) \
655 SymX(stg_ap_n_fast) \
656 SymX(stg_ap_p_fast) \
657 SymX(stg_ap_pv_fast) \
658 SymX(stg_ap_pp_fast) \
659 SymX(stg_ap_ppv_fast) \
660 SymX(stg_ap_ppp_fast) \
661 SymX(stg_ap_pppv_fast) \
662 SymX(stg_ap_pppp_fast) \
663 SymX(stg_ap_ppppp_fast) \
664 SymX(stg_ap_pppppp_fast) \
665 SymX(stg_ap_1_upd_info) \
666 SymX(stg_ap_2_upd_info) \
667 SymX(stg_ap_3_upd_info) \
668 SymX(stg_ap_4_upd_info) \
669 SymX(stg_ap_5_upd_info) \
670 SymX(stg_ap_6_upd_info) \
671 SymX(stg_ap_7_upd_info) \
673 SymX(stg_sel_0_upd_info) \
674 SymX(stg_sel_10_upd_info) \
675 SymX(stg_sel_11_upd_info) \
676 SymX(stg_sel_12_upd_info) \
677 SymX(stg_sel_13_upd_info) \
678 SymX(stg_sel_14_upd_info) \
679 SymX(stg_sel_15_upd_info) \
680 SymX(stg_sel_1_upd_info) \
681 SymX(stg_sel_2_upd_info) \
682 SymX(stg_sel_3_upd_info) \
683 SymX(stg_sel_4_upd_info) \
684 SymX(stg_sel_5_upd_info) \
685 SymX(stg_sel_6_upd_info) \
686 SymX(stg_sel_7_upd_info) \
687 SymX(stg_sel_8_upd_info) \
688 SymX(stg_sel_9_upd_info) \
689 SymX(stg_upd_frame_info) \
690 SymX(suspendThread) \
691 SymX(takeMVarzh_fast) \
692 SymX(timesIntegerzh_fast) \
693 SymX(tryPutMVarzh_fast) \
694 SymX(tryTakeMVarzh_fast) \
695 SymX(unblockAsyncExceptionszh_fast) \
697 SymX(unsafeThawArrayzh_fast) \
698 SymX(waitReadzh_fast) \
699 SymX(waitWritezh_fast) \
700 SymX(word2Integerzh_fast) \
701 SymX(writeTVarzh_fast) \
702 SymX(xorIntegerzh_fast) \
704 SymX(stg_interp_constr_entry) \
705 SymX(stg_interp_constr1_entry) \
706 SymX(stg_interp_constr2_entry) \
707 SymX(stg_interp_constr3_entry) \
708 SymX(stg_interp_constr4_entry) \
709 SymX(stg_interp_constr5_entry) \
710 SymX(stg_interp_constr6_entry) \
711 SymX(stg_interp_constr7_entry) \
712 SymX(stg_interp_constr8_entry) \
715 SymX(getAllocations) \
718 RTS_USER_SIGNALS_SYMBOLS
720 #ifdef SUPPORT_LONG_LONGS
721 #define RTS_LONG_LONG_SYMS \
722 SymX(int64ToIntegerzh_fast) \
723 SymX(word64ToIntegerzh_fast)
725 #define RTS_LONG_LONG_SYMS /* nothing */
728 // 64-bit support functions in libgcc.a
729 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
730 #define RTS_LIBGCC_SYMBOLS \
740 #elif defined(ia64_HOST_ARCH)
741 #define RTS_LIBGCC_SYMBOLS \
749 #define RTS_LIBGCC_SYMBOLS
752 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
753 // Symbols that don't have a leading underscore
754 // on Mac OS X. They have to receive special treatment,
755 // see machoInitSymbolsWithoutUnderscore()
756 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
761 /* entirely bogus claims about types of these symbols */
762 #define Sym(vvv) extern void vvv(void);
763 #define SymX(vvv) /**/
764 #define SymX_redirect(vvv,xxx) /**/
768 RTS_POSIX_ONLY_SYMBOLS
769 RTS_MINGW_ONLY_SYMBOLS
770 RTS_CYGWIN_ONLY_SYMBOLS
771 RTS_DARWIN_ONLY_SYMBOLS
777 #ifdef LEADING_UNDERSCORE
778 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
780 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
783 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
785 #define SymX(vvv) Sym(vvv)
787 // SymX_redirect allows us to redirect references to one symbol to
788 // another symbol. See newCAF/newDynCAF for an example.
789 #define SymX_redirect(vvv,xxx) \
790 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
793 static RtsSymbolVal rtsSyms[] = {
797 RTS_POSIX_ONLY_SYMBOLS
798 RTS_MINGW_ONLY_SYMBOLS
799 RTS_CYGWIN_ONLY_SYMBOLS
801 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
802 // dyld stub code contains references to this,
803 // but it should never be called because we treat
804 // lazy pointers as nonlazy.
805 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
807 { 0, 0 } /* sentinel */
811 /* -----------------------------------------------------------------------------
812 * Utilities for handling root pointers.
813 * -------------------------------------------------------------------------- */
816 #define INIT_RPT_SIZE 64
819 initFreeList(rootEntry *table, nat n, rootEntry *free)
823 for (p = table + n - 1; p >= table; p--) {
827 root_ptr_free = table;
831 initRootPtrTable(void)
836 RPT_size = INIT_RPT_SIZE;
837 root_ptr_table = stgMallocBytes(RPT_size * sizeof(rootEntry),
840 initFreeList(root_ptr_table,INIT_RPT_SIZE,NULL);
845 enlargeRootPtrTable(void)
847 nat old_RPT_size = RPT_size;
849 // 2nd and subsequent times
852 stgReallocBytes(root_ptr_table,
853 RPT_size * sizeof(rootEntry),
854 "enlargeRootPtrTable");
856 initFreeList(root_ptr_table + old_RPT_size, old_RPT_size, NULL);
860 addRootObject(void *addr)
864 if (root_ptr_free == NULL) {
865 enlargeRootPtrTable();
868 rt = root_ptr_free - root_ptr_table;
869 root_ptr_free = (rootEntry*)(root_ptr_free->addr);
870 root_ptr_table[rt].addr = addr;
873 /* -----------------------------------------------------------------------------
874 * Treat root pointers as roots for the garbage collector.
875 * -------------------------------------------------------------------------- */
878 markRootPtrTable(evac_fn evac)
880 rootEntry *p, *end_root_ptr_table;
883 end_root_ptr_table = &root_ptr_table[RPT_size];
885 for (p = root_ptr_table; p < end_root_ptr_table; p++) {
888 if (q && (q < (P_)root_ptr_table || q >= (P_)end_root_ptr_table)) {
889 evac((StgClosure **)p->addr);
894 /* -----------------------------------------------------------------------------
895 * End of utilities for handling root pointers.
896 * -------------------------------------------------------------------------- */
899 /* -----------------------------------------------------------------------------
900 * Insert symbols into hash tables, checking for duplicates.
902 static void ghciInsertStrHashTable ( char* obj_name,
908 if (lookupHashTable(table, (StgWord)key) == NULL)
910 insertStrHashTable(table, (StgWord)key, data);
915 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
917 "whilst processing object file\n"
919 "This could be caused by:\n"
920 " * Loading two different object files which export the same symbol\n"
921 " * Specifying the same object file twice on the GHCi command line\n"
922 " * An incorrect `package.conf' entry, causing some object to be\n"
924 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
933 /* -----------------------------------------------------------------------------
934 * initialize the object linker
938 static int linker_init_done = 0 ;
940 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
941 static void *dl_prog_handle;
949 /* Make initLinker idempotent, so we can call it
950 before evey relevant operation; that means we
951 don't need to initialise the linker separately */
952 if (linker_init_done == 1) { return; } else {
953 linker_init_done = 1;
956 stablehash = allocStrHashTable();
957 symhash = allocStrHashTable();
959 /* populate the symbol table with stuff from the RTS */
960 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
961 ghciInsertStrHashTable("(GHCi built-in symbols)",
962 symhash, sym->lbl, sym->addr);
964 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
965 machoInitSymbolsWithoutUnderscore();
968 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
969 # if defined(RTLD_DEFAULT)
970 dl_prog_handle = RTLD_DEFAULT;
972 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
973 # endif /* RTLD_DEFAULT */
977 /* -----------------------------------------------------------------------------
978 * Loading DLL or .so dynamic libraries
979 * -----------------------------------------------------------------------------
981 * Add a DLL from which symbols may be found. In the ELF case, just
982 * do RTLD_GLOBAL-style add, so no further messing around needs to
983 * happen in order that symbols in the loaded .so are findable --
984 * lookupSymbol() will subsequently see them by dlsym on the program's
985 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
987 * In the PEi386 case, open the DLLs and put handles to them in a
988 * linked list. When looking for a symbol, try all handles in the
989 * list. This means that we need to load even DLLs that are guaranteed
990 * to be in the ghc.exe image already, just so we can get a handle
991 * to give to loadSymbol, so that we can find the symbols. For such
992 * libraries, the LoadLibrary call should be a no-op except for returning
997 #if defined(OBJFORMAT_PEi386)
998 /* A record for storing handles into DLLs. */
1003 struct _OpenedDLL* next;
1008 /* A list thereof. */
1009 static OpenedDLL* opened_dlls = NULL;
1013 addDLL( char *dll_name )
1015 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1016 /* ------------------- ELF DLL loader ------------------- */
1022 hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
1025 /* dlopen failed; return a ptr to the error msg. */
1027 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1034 # elif defined(OBJFORMAT_PEi386)
1035 /* ------------------- Win32 DLL loader ------------------- */
1043 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1045 /* See if we've already got it, and ignore if so. */
1046 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1047 if (0 == strcmp(o_dll->name, dll_name))
1051 /* The file name has no suffix (yet) so that we can try
1052 both foo.dll and foo.drv
1054 The documentation for LoadLibrary says:
1055 If no file name extension is specified in the lpFileName
1056 parameter, the default library extension .dll is
1057 appended. However, the file name string can include a trailing
1058 point character (.) to indicate that the module name has no
1061 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1062 sprintf(buf, "%s.DLL", dll_name);
1063 instance = LoadLibrary(buf);
1064 if (instance == NULL) {
1065 sprintf(buf, "%s.DRV", dll_name); // KAA: allow loading of drivers (like winspool.drv)
1066 instance = LoadLibrary(buf);
1067 if (instance == NULL) {
1070 /* LoadLibrary failed; return a ptr to the error msg. */
1071 return "addDLL: unknown error";
1076 /* Add this DLL to the list of DLLs in which to search for symbols. */
1077 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1078 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1079 strcpy(o_dll->name, dll_name);
1080 o_dll->instance = instance;
1081 o_dll->next = opened_dlls;
1082 opened_dlls = o_dll;
1086 barf("addDLL: not implemented on this platform");
1090 /* -----------------------------------------------------------------------------
1091 * insert a stable symbol in the hash table
1095 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1097 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1101 /* -----------------------------------------------------------------------------
1102 * insert a symbol in the hash table
1105 insertSymbol(char* obj_name, char* key, void* data)
1107 ghciInsertStrHashTable(obj_name, symhash, key, data);
1110 /* -----------------------------------------------------------------------------
1111 * lookup a symbol in the hash table
1114 lookupSymbol( char *lbl )
1118 ASSERT(symhash != NULL);
1119 val = lookupStrHashTable(symhash, lbl);
1122 # if defined(OBJFORMAT_ELF)
1123 # if defined(x86_64_HOST_ARCH)
1124 val = dlsym(dl_prog_handle, lbl);
1125 if (val >= (void *)0x80000000) {
1127 new_val = x86_64_high_symbol(lbl, val);
1128 IF_DEBUG(linker,debugBelch("lookupSymbol: relocating out of range symbol: %s = %p, now %p\n", lbl, val, new_val));
1134 return dlsym(dl_prog_handle, lbl);
1136 # elif defined(OBJFORMAT_MACHO)
1137 if(NSIsSymbolNameDefined(lbl)) {
1138 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1139 return NSAddressOfSymbol(symbol);
1143 # elif defined(OBJFORMAT_PEi386)
1146 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1147 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
1148 if (lbl[0] == '_') {
1149 /* HACK: if the name has an initial underscore, try stripping
1150 it off & look that up first. I've yet to verify whether there's
1151 a Rule that governs whether an initial '_' *should always* be
1152 stripped off when mapping from import lib name to the DLL name.
1154 sym = GetProcAddress(o_dll->instance, (lbl+1));
1156 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
1160 sym = GetProcAddress(o_dll->instance, lbl);
1162 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
1177 __attribute((unused))
1179 lookupLocalSymbol( ObjectCode* oc, char *lbl )
1183 val = lookupStrHashTable(oc->lochash, lbl);
1193 /* -----------------------------------------------------------------------------
1194 * Debugging aid: look in GHCi's object symbol tables for symbols
1195 * within DELTA bytes of the specified address, and show their names.
1198 void ghci_enquire ( char* addr );
1200 void ghci_enquire ( char* addr )
1205 const int DELTA = 64;
1210 for (oc = objects; oc; oc = oc->next) {
1211 for (i = 0; i < oc->n_symbols; i++) {
1212 sym = oc->symbols[i];
1213 if (sym == NULL) continue;
1214 // debugBelch("enquire %p %p\n", sym, oc->lochash);
1216 if (oc->lochash != NULL) {
1217 a = lookupStrHashTable(oc->lochash, sym);
1220 a = lookupStrHashTable(symhash, sym);
1223 // debugBelch("ghci_enquire: can't find %s\n", sym);
1225 else if (addr-DELTA <= a && a <= addr+DELTA) {
1226 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1233 #ifdef ia64_HOST_ARCH
1234 static unsigned int PLTSize(void);
1237 /* -----------------------------------------------------------------------------
1238 * Load an obj (populate the global symbol table, but don't resolve yet)
1240 * Returns: 1 if ok, 0 on error.
1243 loadObj( char *path )
1250 void *map_addr = NULL;
1257 /* debugBelch("loadObj %s\n", path ); */
1259 /* Check that we haven't already loaded this object.
1260 Ignore requests to load multiple times */
1264 for (o = objects; o; o = o->next) {
1265 if (0 == strcmp(o->fileName, path)) {
1267 break; /* don't need to search further */
1271 IF_DEBUG(linker, debugBelch(
1272 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1273 "same object file twice:\n"
1275 "GHCi will ignore this, but be warned.\n"
1277 return 1; /* success */
1281 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1283 # if defined(OBJFORMAT_ELF)
1284 oc->formatName = "ELF";
1285 # elif defined(OBJFORMAT_PEi386)
1286 oc->formatName = "PEi386";
1287 # elif defined(OBJFORMAT_MACHO)
1288 oc->formatName = "Mach-O";
1291 barf("loadObj: not implemented on this platform");
1294 r = stat(path, &st);
1295 if (r == -1) { return 0; }
1297 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1298 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1299 strcpy(oc->fileName, path);
1301 oc->fileSize = st.st_size;
1303 oc->sections = NULL;
1304 oc->lochash = allocStrHashTable();
1305 oc->proddables = NULL;
1307 /* chain it onto the list of objects */
1312 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1314 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1316 #if defined(openbsd_HOST_OS)
1317 fd = open(path, O_RDONLY, S_IRUSR);
1319 fd = open(path, O_RDONLY);
1322 barf("loadObj: can't open `%s'", path);
1324 pagesize = getpagesize();
1326 #ifdef ia64_HOST_ARCH
1327 /* The PLT needs to be right before the object */
1328 n = ROUND_UP(PLTSize(), pagesize);
1329 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1330 if (oc->plt == MAP_FAILED)
1331 barf("loadObj: can't allocate PLT");
1334 map_addr = oc->plt + n;
1337 n = ROUND_UP(oc->fileSize, pagesize);
1339 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
1340 * small memory model on this architecture (see gcc docs,
1343 #ifdef x86_64_HOST_ARCH
1344 #define EXTRA_MAP_FLAGS MAP_32BIT
1346 #define EXTRA_MAP_FLAGS 0
1349 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1350 MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0);
1351 if (oc->image == MAP_FAILED)
1352 barf("loadObj: can't map `%s'", path);
1356 #else /* !USE_MMAP */
1358 /* load the image into memory */
1359 f = fopen(path, "rb");
1361 barf("loadObj: can't read `%s'", path);
1363 #ifdef darwin_HOST_OS
1364 // In a Mach-O .o file, all sections can and will be misaligned
1365 // if the total size of the headers is not a multiple of the
1366 // desired alignment. This is fine for .o files that only serve
1367 // as input for the static linker, but it's not fine for us,
1368 // as SSE (used by gcc for floating point) and Altivec require
1369 // 16-byte alignment.
1370 // We calculate the correct alignment from the header before
1371 // reading the file, and then we misalign oc->image on purpose so
1372 // that the actual sections end up aligned again.
1373 misalignment = machoGetMisalignment(f);
1374 oc->misalignment = misalignment;
1379 oc->image = stgMallocBytes(oc->fileSize + misalignment, "loadObj(image)");
1380 oc->image += misalignment;
1382 n = fread ( oc->image, 1, oc->fileSize, f );
1383 if (n != oc->fileSize)
1384 barf("loadObj: error whilst reading `%s'", path);
1388 #endif /* USE_MMAP */
1390 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1391 r = ocAllocateJumpIslands_MachO ( oc );
1392 if (!r) { return r; }
1393 # elif defined(OBJFORMAT_ELF) && defined(powerpc_HOST_ARCH)
1394 r = ocAllocateJumpIslands_ELF ( oc );
1395 if (!r) { return r; }
1398 /* verify the in-memory image */
1399 # if defined(OBJFORMAT_ELF)
1400 r = ocVerifyImage_ELF ( oc );
1401 # elif defined(OBJFORMAT_PEi386)
1402 r = ocVerifyImage_PEi386 ( oc );
1403 # elif defined(OBJFORMAT_MACHO)
1404 r = ocVerifyImage_MachO ( oc );
1406 barf("loadObj: no verify method");
1408 if (!r) { return r; }
1410 /* build the symbol list for this image */
1411 # if defined(OBJFORMAT_ELF)
1412 r = ocGetNames_ELF ( oc );
1413 # elif defined(OBJFORMAT_PEi386)
1414 r = ocGetNames_PEi386 ( oc );
1415 # elif defined(OBJFORMAT_MACHO)
1416 r = ocGetNames_MachO ( oc );
1418 barf("loadObj: no getNames method");
1420 if (!r) { return r; }
1422 /* loaded, but not resolved yet */
1423 oc->status = OBJECT_LOADED;
1428 /* -----------------------------------------------------------------------------
1429 * resolve all the currently unlinked objects in memory
1431 * Returns: 1 if ok, 0 on error.
1441 for (oc = objects; oc; oc = oc->next) {
1442 if (oc->status != OBJECT_RESOLVED) {
1443 # if defined(OBJFORMAT_ELF)
1444 r = ocResolve_ELF ( oc );
1445 # elif defined(OBJFORMAT_PEi386)
1446 r = ocResolve_PEi386 ( oc );
1447 # elif defined(OBJFORMAT_MACHO)
1448 r = ocResolve_MachO ( oc );
1450 barf("resolveObjs: not implemented on this platform");
1452 if (!r) { return r; }
1453 oc->status = OBJECT_RESOLVED;
1459 /* -----------------------------------------------------------------------------
1460 * delete an object from the pool
1463 unloadObj( char *path )
1465 ObjectCode *oc, *prev;
1467 ASSERT(symhash != NULL);
1468 ASSERT(objects != NULL);
1473 for (oc = objects; oc; prev = oc, oc = oc->next) {
1474 if (!strcmp(oc->fileName,path)) {
1476 /* Remove all the mappings for the symbols within this
1481 for (i = 0; i < oc->n_symbols; i++) {
1482 if (oc->symbols[i] != NULL) {
1483 removeStrHashTable(symhash, oc->symbols[i], NULL);
1491 prev->next = oc->next;
1494 /* We're going to leave this in place, in case there are
1495 any pointers from the heap into it: */
1496 /* stgFree(oc->image); */
1497 stgFree(oc->fileName);
1498 stgFree(oc->symbols);
1499 stgFree(oc->sections);
1500 /* The local hash table should have been freed at the end
1501 of the ocResolve_ call on it. */
1502 ASSERT(oc->lochash == NULL);
1508 errorBelch("unloadObj: can't find `%s' to unload", path);
1512 /* -----------------------------------------------------------------------------
1513 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1514 * which may be prodded during relocation, and abort if we try and write
1515 * outside any of these.
1517 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1520 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1521 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1525 pb->next = oc->proddables;
1526 oc->proddables = pb;
1529 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1532 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1533 char* s = (char*)(pb->start);
1534 char* e = s + pb->size - 1;
1535 char* a = (char*)addr;
1536 /* Assumes that the biggest fixup involves a 4-byte write. This
1537 probably needs to be changed to 8 (ie, +7) on 64-bit
1539 if (a >= s && (a+3) <= e) return;
1541 barf("checkProddableBlock: invalid fixup in runtime linker");
1544 /* -----------------------------------------------------------------------------
1545 * Section management.
1547 static void addSection ( ObjectCode* oc, SectionKind kind,
1548 void* start, void* end )
1550 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1554 s->next = oc->sections;
1557 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1558 start, ((char*)end)-1, end - start + 1, kind );
1563 /* --------------------------------------------------------------------------
1564 * PowerPC specifics (jump islands)
1565 * ------------------------------------------------------------------------*/
1567 #if defined(powerpc_HOST_ARCH)
1570 ocAllocateJumpIslands
1572 Allocate additional space at the end of the object file image to make room
1575 PowerPC relative branch instructions have a 24 bit displacement field.
1576 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1577 If a particular imported symbol is outside this range, we have to redirect
1578 the jump to a short piece of new code that just loads the 32bit absolute
1579 address and jumps there.
1580 This function just allocates space for one 16 byte ppcJumpIsland for every
1581 undefined symbol in the object file. The code for the islands is filled in by
1582 makeJumpIsland below.
1585 static int ocAllocateJumpIslands( ObjectCode* oc, int count, int first )
1591 int misalignment = 0;
1593 misalignment = oc->misalignment;
1598 // round up to the nearest 4
1599 aligned = (oc->fileSize + 3) & ~3;
1602 #ifndef linux_HOST_OS /* mremap is a linux extension */
1603 #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined
1606 pagesize = getpagesize();
1607 n = ROUND_UP( oc->fileSize, pagesize );
1608 m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize );
1610 /* If we have a half-page-size file and map one page of it then
1611 * the part of the page after the size of the file remains accessible.
1612 * If, however, we map in 2 pages, the 2nd page is not accessible
1613 * and will give a "Bus Error" on access. To get around this, we check
1614 * if we need any extra pages for the jump islands and map them in
1615 * anonymously. We must check that we actually require extra pages
1616 * otherwise the attempt to mmap 0 pages of anonymous memory will
1622 /* The effect of this mremap() call is only the ensure that we have
1623 * a sufficient number of virtually contiguous pages. As returned from
1624 * mremap, the pages past the end of the file are not backed. We give
1625 * them a backing by using MAP_FIXED to map in anonymous pages.
1627 oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE );
1629 if( oc->image == MAP_FAILED )
1631 errorBelch( "Unable to mremap for Jump Islands\n" );
1635 if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC,
1636 MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED )
1638 errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" );
1644 oc->image -= misalignment;
1645 oc->image = stgReallocBytes( oc->image,
1647 aligned + sizeof (ppcJumpIsland) * count,
1648 "ocAllocateJumpIslands" );
1649 oc->image += misalignment;
1650 #endif /* USE_MMAP */
1652 oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned);
1653 memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count );
1656 oc->jump_islands = NULL;
1658 oc->island_start_symbol = first;
1659 oc->n_islands = count;
1664 static unsigned long makeJumpIsland( ObjectCode* oc,
1665 unsigned long symbolNumber,
1666 unsigned long target )
1668 ppcJumpIsland *island;
1670 if( symbolNumber < oc->island_start_symbol ||
1671 symbolNumber - oc->island_start_symbol > oc->n_islands)
1674 island = &oc->jump_islands[symbolNumber - oc->island_start_symbol];
1676 // lis r12, hi16(target)
1677 island->lis_r12 = 0x3d80;
1678 island->hi_addr = target >> 16;
1680 // ori r12, r12, lo16(target)
1681 island->ori_r12_r12 = 0x618c;
1682 island->lo_addr = target & 0xffff;
1685 island->mtctr_r12 = 0x7d8903a6;
1688 island->bctr = 0x4e800420;
1690 return (unsigned long) island;
1694 ocFlushInstructionCache
1696 Flush the data & instruction caches.
1697 Because the PPC has split data/instruction caches, we have to
1698 do that whenever we modify code at runtime.
1701 static void ocFlushInstructionCache( ObjectCode *oc )
1703 int n = (oc->fileSize + sizeof( ppcJumpIsland ) * oc->n_islands + 3) / 4;
1704 unsigned long *p = (unsigned long *) oc->image;
1708 __asm__ volatile ( "dcbf 0,%0\n\t"
1716 __asm__ volatile ( "sync\n\t"
1722 /* --------------------------------------------------------------------------
1723 * PEi386 specifics (Win32 targets)
1724 * ------------------------------------------------------------------------*/
1726 /* The information for this linker comes from
1727 Microsoft Portable Executable
1728 and Common Object File Format Specification
1729 revision 5.1 January 1998
1730 which SimonM says comes from the MS Developer Network CDs.
1732 It can be found there (on older CDs), but can also be found
1735 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1737 (this is Rev 6.0 from February 1999).
1739 Things move, so if that fails, try searching for it via
1741 http://www.google.com/search?q=PE+COFF+specification
1743 The ultimate reference for the PE format is the Winnt.h
1744 header file that comes with the Platform SDKs; as always,
1745 implementations will drift wrt their documentation.
1747 A good background article on the PE format is Matt Pietrek's
1748 March 1994 article in Microsoft System Journal (MSJ)
1749 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1750 Win32 Portable Executable File Format." The info in there
1751 has recently been updated in a two part article in
1752 MSDN magazine, issues Feb and March 2002,
1753 "Inside Windows: An In-Depth Look into the Win32 Portable
1754 Executable File Format"
1756 John Levine's book "Linkers and Loaders" contains useful
1761 #if defined(OBJFORMAT_PEi386)
1765 typedef unsigned char UChar;
1766 typedef unsigned short UInt16;
1767 typedef unsigned int UInt32;
1774 UInt16 NumberOfSections;
1775 UInt32 TimeDateStamp;
1776 UInt32 PointerToSymbolTable;
1777 UInt32 NumberOfSymbols;
1778 UInt16 SizeOfOptionalHeader;
1779 UInt16 Characteristics;
1783 #define sizeof_COFF_header 20
1790 UInt32 VirtualAddress;
1791 UInt32 SizeOfRawData;
1792 UInt32 PointerToRawData;
1793 UInt32 PointerToRelocations;
1794 UInt32 PointerToLinenumbers;
1795 UInt16 NumberOfRelocations;
1796 UInt16 NumberOfLineNumbers;
1797 UInt32 Characteristics;
1801 #define sizeof_COFF_section 40
1808 UInt16 SectionNumber;
1811 UChar NumberOfAuxSymbols;
1815 #define sizeof_COFF_symbol 18
1820 UInt32 VirtualAddress;
1821 UInt32 SymbolTableIndex;
1826 #define sizeof_COFF_reloc 10
1829 /* From PE spec doc, section 3.3.2 */
1830 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1831 windows.h -- for the same purpose, but I want to know what I'm
1833 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1834 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1835 #define MYIMAGE_FILE_DLL 0x2000
1836 #define MYIMAGE_FILE_SYSTEM 0x1000
1837 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1838 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1839 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1841 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1842 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1843 #define MYIMAGE_SYM_CLASS_STATIC 3
1844 #define MYIMAGE_SYM_UNDEFINED 0
1846 /* From PE spec doc, section 4.1 */
1847 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1848 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1849 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1851 /* From PE spec doc, section 5.2.1 */
1852 #define MYIMAGE_REL_I386_DIR32 0x0006
1853 #define MYIMAGE_REL_I386_REL32 0x0014
1856 /* We use myindex to calculate array addresses, rather than
1857 simply doing the normal subscript thing. That's because
1858 some of the above structs have sizes which are not
1859 a whole number of words. GCC rounds their sizes up to a
1860 whole number of words, which means that the address calcs
1861 arising from using normal C indexing or pointer arithmetic
1862 are just plain wrong. Sigh.
1865 myindex ( int scale, void* base, int index )
1868 ((UChar*)base) + scale * index;
1873 printName ( UChar* name, UChar* strtab )
1875 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1876 UInt32 strtab_offset = * (UInt32*)(name+4);
1877 debugBelch("%s", strtab + strtab_offset );
1880 for (i = 0; i < 8; i++) {
1881 if (name[i] == 0) break;
1882 debugBelch("%c", name[i] );
1889 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1891 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1892 UInt32 strtab_offset = * (UInt32*)(name+4);
1893 strncpy ( dst, strtab+strtab_offset, dstSize );
1899 if (name[i] == 0) break;
1909 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1912 /* If the string is longer than 8 bytes, look in the
1913 string table for it -- this will be correctly zero terminated.
1915 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1916 UInt32 strtab_offset = * (UInt32*)(name+4);
1917 return ((UChar*)strtab) + strtab_offset;
1919 /* Otherwise, if shorter than 8 bytes, return the original,
1920 which by defn is correctly terminated.
1922 if (name[7]==0) return name;
1923 /* The annoying case: 8 bytes. Copy into a temporary
1924 (which is never freed ...)
1926 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1928 strncpy(newstr,name,8);
1934 /* Just compares the short names (first 8 chars) */
1935 static COFF_section *
1936 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1940 = (COFF_header*)(oc->image);
1941 COFF_section* sectab
1943 ((UChar*)(oc->image))
1944 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1946 for (i = 0; i < hdr->NumberOfSections; i++) {
1949 COFF_section* section_i
1951 myindex ( sizeof_COFF_section, sectab, i );
1952 n1 = (UChar*) &(section_i->Name);
1954 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1955 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1956 n1[6]==n2[6] && n1[7]==n2[7])
1965 zapTrailingAtSign ( UChar* sym )
1967 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1969 if (sym[0] == 0) return;
1971 while (sym[i] != 0) i++;
1974 while (j > 0 && my_isdigit(sym[j])) j--;
1975 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1981 ocVerifyImage_PEi386 ( ObjectCode* oc )
1986 COFF_section* sectab;
1987 COFF_symbol* symtab;
1989 /* debugBelch("\nLOADING %s\n", oc->fileName); */
1990 hdr = (COFF_header*)(oc->image);
1991 sectab = (COFF_section*) (
1992 ((UChar*)(oc->image))
1993 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1995 symtab = (COFF_symbol*) (
1996 ((UChar*)(oc->image))
1997 + hdr->PointerToSymbolTable
1999 strtab = ((UChar*)symtab)
2000 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2002 if (hdr->Machine != 0x14c) {
2003 errorBelch("%s: Not x86 PEi386", oc->fileName);
2006 if (hdr->SizeOfOptionalHeader != 0) {
2007 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2010 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2011 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2012 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2013 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2014 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2017 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2018 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2019 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2021 (int)(hdr->Characteristics));
2024 /* If the string table size is way crazy, this might indicate that
2025 there are more than 64k relocations, despite claims to the
2026 contrary. Hence this test. */
2027 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2029 if ( (*(UInt32*)strtab) > 600000 ) {
2030 /* Note that 600k has no special significance other than being
2031 big enough to handle the almost-2MB-sized lumps that
2032 constitute HSwin32*.o. */
2033 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2038 /* No further verification after this point; only debug printing. */
2040 IF_DEBUG(linker, i=1);
2041 if (i == 0) return 1;
2043 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2044 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2045 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2048 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2049 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2050 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2051 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2052 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2053 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2054 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2056 /* Print the section table. */
2058 for (i = 0; i < hdr->NumberOfSections; i++) {
2060 COFF_section* sectab_i
2062 myindex ( sizeof_COFF_section, sectab, i );
2069 printName ( sectab_i->Name, strtab );
2079 sectab_i->VirtualSize,
2080 sectab_i->VirtualAddress,
2081 sectab_i->SizeOfRawData,
2082 sectab_i->PointerToRawData,
2083 sectab_i->NumberOfRelocations,
2084 sectab_i->PointerToRelocations,
2085 sectab_i->PointerToRawData
2087 reltab = (COFF_reloc*) (
2088 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2091 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2092 /* If the relocation field (a short) has overflowed, the
2093 * real count can be found in the first reloc entry.
2095 * See Section 4.1 (last para) of the PE spec (rev6.0).
2097 COFF_reloc* rel = (COFF_reloc*)
2098 myindex ( sizeof_COFF_reloc, reltab, 0 );
2099 noRelocs = rel->VirtualAddress;
2102 noRelocs = sectab_i->NumberOfRelocations;
2106 for (; j < noRelocs; j++) {
2108 COFF_reloc* rel = (COFF_reloc*)
2109 myindex ( sizeof_COFF_reloc, reltab, j );
2111 " type 0x%-4x vaddr 0x%-8x name `",
2113 rel->VirtualAddress );
2114 sym = (COFF_symbol*)
2115 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2116 /* Hmm..mysterious looking offset - what's it for? SOF */
2117 printName ( sym->Name, strtab -10 );
2124 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2125 debugBelch("---START of string table---\n");
2126 for (i = 4; i < *(Int32*)strtab; i++) {
2128 debugBelch("\n"); else
2129 debugBelch("%c", strtab[i] );
2131 debugBelch("--- END of string table---\n");
2136 COFF_symbol* symtab_i;
2137 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2138 symtab_i = (COFF_symbol*)
2139 myindex ( sizeof_COFF_symbol, symtab, i );
2145 printName ( symtab_i->Name, strtab );
2154 (Int32)(symtab_i->SectionNumber),
2155 (UInt32)symtab_i->Type,
2156 (UInt32)symtab_i->StorageClass,
2157 (UInt32)symtab_i->NumberOfAuxSymbols
2159 i += symtab_i->NumberOfAuxSymbols;
2169 ocGetNames_PEi386 ( ObjectCode* oc )
2172 COFF_section* sectab;
2173 COFF_symbol* symtab;
2180 hdr = (COFF_header*)(oc->image);
2181 sectab = (COFF_section*) (
2182 ((UChar*)(oc->image))
2183 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2185 symtab = (COFF_symbol*) (
2186 ((UChar*)(oc->image))
2187 + hdr->PointerToSymbolTable
2189 strtab = ((UChar*)(oc->image))
2190 + hdr->PointerToSymbolTable
2191 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2193 /* Allocate space for any (local, anonymous) .bss sections. */
2195 for (i = 0; i < hdr->NumberOfSections; i++) {
2198 COFF_section* sectab_i
2200 myindex ( sizeof_COFF_section, sectab, i );
2201 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2202 /* sof 10/05: the PE spec text isn't too clear regarding what
2203 * the SizeOfRawData field is supposed to hold for object
2204 * file sections containing just uninitialized data -- for executables,
2205 * it is supposed to be zero; unclear what it's supposed to be
2206 * for object files. However, VirtualSize is guaranteed to be
2207 * zero for object files, which definitely suggests that SizeOfRawData
2208 * will be non-zero (where else would the size of this .bss section be
2209 * stored?) Looking at the COFF_section info for incoming object files,
2210 * this certainly appears to be the case.
2212 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2213 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2214 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2215 * variable decls into to the .bss section. (The specific function in Q which
2216 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2218 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2219 /* This is a non-empty .bss section. Allocate zeroed space for
2220 it, and set its PointerToRawData field such that oc->image +
2221 PointerToRawData == addr_of_zeroed_space. */
2222 bss_sz = sectab_i->VirtualSize;
2223 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2224 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2225 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2226 addProddableBlock(oc, zspace, bss_sz);
2227 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2230 /* Copy section information into the ObjectCode. */
2232 for (i = 0; i < hdr->NumberOfSections; i++) {
2238 = SECTIONKIND_OTHER;
2239 COFF_section* sectab_i
2241 myindex ( sizeof_COFF_section, sectab, i );
2242 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2245 /* I'm sure this is the Right Way to do it. However, the
2246 alternative of testing the sectab_i->Name field seems to
2247 work ok with Cygwin.
2249 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2250 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2251 kind = SECTIONKIND_CODE_OR_RODATA;
2254 if (0==strcmp(".text",sectab_i->Name) ||
2255 0==strcmp(".rdata",sectab_i->Name)||
2256 0==strcmp(".rodata",sectab_i->Name))
2257 kind = SECTIONKIND_CODE_OR_RODATA;
2258 if (0==strcmp(".data",sectab_i->Name) ||
2259 0==strcmp(".bss",sectab_i->Name))
2260 kind = SECTIONKIND_RWDATA;
2262 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2263 sz = sectab_i->SizeOfRawData;
2264 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2266 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2267 end = start + sz - 1;
2269 if (kind == SECTIONKIND_OTHER
2270 /* Ignore sections called which contain stabs debugging
2272 && 0 != strcmp(".stab", sectab_i->Name)
2273 && 0 != strcmp(".stabstr", sectab_i->Name)
2274 /* ignore constructor section for now */
2275 && 0 != strcmp(".ctors", sectab_i->Name)
2276 /* ignore section generated from .ident */
2277 && 0!= strcmp("/4", sectab_i->Name)
2279 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2283 if (kind != SECTIONKIND_OTHER && end >= start) {
2284 addSection(oc, kind, start, end);
2285 addProddableBlock(oc, start, end - start + 1);
2289 /* Copy exported symbols into the ObjectCode. */
2291 oc->n_symbols = hdr->NumberOfSymbols;
2292 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2293 "ocGetNames_PEi386(oc->symbols)");
2294 /* Call me paranoid; I don't care. */
2295 for (i = 0; i < oc->n_symbols; i++)
2296 oc->symbols[i] = NULL;
2300 COFF_symbol* symtab_i;
2301 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2302 symtab_i = (COFF_symbol*)
2303 myindex ( sizeof_COFF_symbol, symtab, i );
2307 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2308 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2309 /* This symbol is global and defined, viz, exported */
2310 /* for MYIMAGE_SYMCLASS_EXTERNAL
2311 && !MYIMAGE_SYM_UNDEFINED,
2312 the address of the symbol is:
2313 address of relevant section + offset in section
2315 COFF_section* sectabent
2316 = (COFF_section*) myindex ( sizeof_COFF_section,
2318 symtab_i->SectionNumber-1 );
2319 addr = ((UChar*)(oc->image))
2320 + (sectabent->PointerToRawData
2324 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2325 && symtab_i->Value > 0) {
2326 /* This symbol isn't in any section at all, ie, global bss.
2327 Allocate zeroed space for it. */
2328 addr = stgCallocBytes(1, symtab_i->Value,
2329 "ocGetNames_PEi386(non-anonymous bss)");
2330 addSection(oc, SECTIONKIND_RWDATA, addr,
2331 ((UChar*)addr) + symtab_i->Value - 1);
2332 addProddableBlock(oc, addr, symtab_i->Value);
2333 /* debugBelch("BSS section at 0x%x\n", addr); */
2336 if (addr != NULL ) {
2337 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2338 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2339 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2340 ASSERT(i >= 0 && i < oc->n_symbols);
2341 /* cstring_from_COFF_symbol_name always succeeds. */
2342 oc->symbols[i] = sname;
2343 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2347 "IGNORING symbol %d\n"
2351 printName ( symtab_i->Name, strtab );
2360 (Int32)(symtab_i->SectionNumber),
2361 (UInt32)symtab_i->Type,
2362 (UInt32)symtab_i->StorageClass,
2363 (UInt32)symtab_i->NumberOfAuxSymbols
2368 i += symtab_i->NumberOfAuxSymbols;
2377 ocResolve_PEi386 ( ObjectCode* oc )
2380 COFF_section* sectab;
2381 COFF_symbol* symtab;
2391 /* ToDo: should be variable-sized? But is at least safe in the
2392 sense of buffer-overrun-proof. */
2394 /* debugBelch("resolving for %s\n", oc->fileName); */
2396 hdr = (COFF_header*)(oc->image);
2397 sectab = (COFF_section*) (
2398 ((UChar*)(oc->image))
2399 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2401 symtab = (COFF_symbol*) (
2402 ((UChar*)(oc->image))
2403 + hdr->PointerToSymbolTable
2405 strtab = ((UChar*)(oc->image))
2406 + hdr->PointerToSymbolTable
2407 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2409 for (i = 0; i < hdr->NumberOfSections; i++) {
2410 COFF_section* sectab_i
2412 myindex ( sizeof_COFF_section, sectab, i );
2415 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2418 /* Ignore sections called which contain stabs debugging
2420 if (0 == strcmp(".stab", sectab_i->Name)
2421 || 0 == strcmp(".stabstr", sectab_i->Name)
2422 || 0 == strcmp(".ctors", sectab_i->Name))
2425 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2426 /* If the relocation field (a short) has overflowed, the
2427 * real count can be found in the first reloc entry.
2429 * See Section 4.1 (last para) of the PE spec (rev6.0).
2431 * Nov2003 update: the GNU linker still doesn't correctly
2432 * handle the generation of relocatable object files with
2433 * overflown relocations. Hence the output to warn of potential
2436 COFF_reloc* rel = (COFF_reloc*)
2437 myindex ( sizeof_COFF_reloc, reltab, 0 );
2438 noRelocs = rel->VirtualAddress;
2440 /* 10/05: we now assume (and check for) a GNU ld that is capable
2441 * of handling object files with (>2^16) of relocs.
2444 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2449 noRelocs = sectab_i->NumberOfRelocations;
2454 for (; j < noRelocs; j++) {
2456 COFF_reloc* reltab_j
2458 myindex ( sizeof_COFF_reloc, reltab, j );
2460 /* the location to patch */
2462 ((UChar*)(oc->image))
2463 + (sectab_i->PointerToRawData
2464 + reltab_j->VirtualAddress
2465 - sectab_i->VirtualAddress )
2467 /* the existing contents of pP */
2469 /* the symbol to connect to */
2470 sym = (COFF_symbol*)
2471 myindex ( sizeof_COFF_symbol,
2472 symtab, reltab_j->SymbolTableIndex );
2475 "reloc sec %2d num %3d: type 0x%-4x "
2476 "vaddr 0x%-8x name `",
2478 (UInt32)reltab_j->Type,
2479 reltab_j->VirtualAddress );
2480 printName ( sym->Name, strtab );
2481 debugBelch("'\n" ));
2483 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2484 COFF_section* section_sym
2485 = findPEi386SectionCalled ( oc, sym->Name );
2487 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2490 S = ((UInt32)(oc->image))
2491 + (section_sym->PointerToRawData
2494 copyName ( sym->Name, strtab, symbol, 1000-1 );
2495 (void*)S = lookupLocalSymbol( oc, symbol );
2496 if ((void*)S != NULL) goto foundit;
2497 (void*)S = lookupSymbol( symbol );
2498 if ((void*)S != NULL) goto foundit;
2499 zapTrailingAtSign ( symbol );
2500 (void*)S = lookupLocalSymbol( oc, symbol );
2501 if ((void*)S != NULL) goto foundit;
2502 (void*)S = lookupSymbol( symbol );
2503 if ((void*)S != NULL) goto foundit;
2504 /* Newline first because the interactive linker has printed "linking..." */
2505 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2509 checkProddableBlock(oc, pP);
2510 switch (reltab_j->Type) {
2511 case MYIMAGE_REL_I386_DIR32:
2514 case MYIMAGE_REL_I386_REL32:
2515 /* Tricky. We have to insert a displacement at
2516 pP which, when added to the PC for the _next_
2517 insn, gives the address of the target (S).
2518 Problem is to know the address of the next insn
2519 when we only know pP. We assume that this
2520 literal field is always the last in the insn,
2521 so that the address of the next insn is pP+4
2522 -- hence the constant 4.
2523 Also I don't know if A should be added, but so
2524 far it has always been zero.
2526 SOF 05/2005: 'A' (old contents of *pP) have been observed
2527 to contain values other than zero (the 'wx' object file
2528 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2529 So, add displacement to old value instead of asserting
2530 A to be zero. Fixes wxhaskell-related crashes, and no other
2531 ill effects have been observed.
2533 Update: the reason why we're seeing these more elaborate
2534 relocations is due to a switch in how the NCG compiles SRTs
2535 and offsets to them from info tables. SRTs live in .(ro)data,
2536 while info tables live in .text, causing GAS to emit REL32/DISP32
2537 relocations with non-zero values. Adding the displacement is
2538 the right thing to do.
2540 *pP = S - ((UInt32)pP) - 4 + A;
2543 debugBelch("%s: unhandled PEi386 relocation type %d",
2544 oc->fileName, reltab_j->Type);
2551 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2555 #endif /* defined(OBJFORMAT_PEi386) */
2558 /* --------------------------------------------------------------------------
2560 * ------------------------------------------------------------------------*/
2562 #if defined(OBJFORMAT_ELF)
2567 #if defined(sparc_HOST_ARCH)
2568 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2569 #elif defined(i386_HOST_ARCH)
2570 # define ELF_TARGET_386 /* Used inside <elf.h> */
2571 #elif defined(x86_64_HOST_ARCH)
2572 # define ELF_TARGET_X64_64
2574 #elif defined (ia64_HOST_ARCH)
2575 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2577 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2578 # define ELF_NEED_GOT /* needs Global Offset Table */
2579 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2582 #if !defined(openbsd_HOST_OS)
2585 /* openbsd elf has things in different places, with diff names */
2586 #include <elf_abi.h>
2587 #include <machine/reloc.h>
2588 #define R_386_32 RELOC_32
2589 #define R_386_PC32 RELOC_PC32
2593 * Define a set of types which can be used for both ELF32 and ELF64
2597 #define ELFCLASS ELFCLASS64
2598 #define Elf_Addr Elf64_Addr
2599 #define Elf_Word Elf64_Word
2600 #define Elf_Sword Elf64_Sword
2601 #define Elf_Ehdr Elf64_Ehdr
2602 #define Elf_Phdr Elf64_Phdr
2603 #define Elf_Shdr Elf64_Shdr
2604 #define Elf_Sym Elf64_Sym
2605 #define Elf_Rel Elf64_Rel
2606 #define Elf_Rela Elf64_Rela
2607 #define ELF_ST_TYPE ELF64_ST_TYPE
2608 #define ELF_ST_BIND ELF64_ST_BIND
2609 #define ELF_R_TYPE ELF64_R_TYPE
2610 #define ELF_R_SYM ELF64_R_SYM
2612 #define ELFCLASS ELFCLASS32
2613 #define Elf_Addr Elf32_Addr
2614 #define Elf_Word Elf32_Word
2615 #define Elf_Sword Elf32_Sword
2616 #define Elf_Ehdr Elf32_Ehdr
2617 #define Elf_Phdr Elf32_Phdr
2618 #define Elf_Shdr Elf32_Shdr
2619 #define Elf_Sym Elf32_Sym
2620 #define Elf_Rel Elf32_Rel
2621 #define Elf_Rela Elf32_Rela
2623 #define ELF_ST_TYPE ELF32_ST_TYPE
2626 #define ELF_ST_BIND ELF32_ST_BIND
2629 #define ELF_R_TYPE ELF32_R_TYPE
2632 #define ELF_R_SYM ELF32_R_SYM
2638 * Functions to allocate entries in dynamic sections. Currently we simply
2639 * preallocate a large number, and we don't check if a entry for the given
2640 * target already exists (a linear search is too slow). Ideally these
2641 * entries would be associated with symbols.
2644 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2645 #define GOT_SIZE 0x20000
2646 #define FUNCTION_TABLE_SIZE 0x10000
2647 #define PLT_SIZE 0x08000
2650 static Elf_Addr got[GOT_SIZE];
2651 static unsigned int gotIndex;
2652 static Elf_Addr gp_val = (Elf_Addr)got;
2655 allocateGOTEntry(Elf_Addr target)
2659 if (gotIndex >= GOT_SIZE)
2660 barf("Global offset table overflow");
2662 entry = &got[gotIndex++];
2664 return (Elf_Addr)entry;
2668 #ifdef ELF_FUNCTION_DESC
2674 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2675 static unsigned int functionTableIndex;
2678 allocateFunctionDesc(Elf_Addr target)
2680 FunctionDesc *entry;
2682 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2683 barf("Function table overflow");
2685 entry = &functionTable[functionTableIndex++];
2687 entry->gp = (Elf_Addr)gp_val;
2688 return (Elf_Addr)entry;
2692 copyFunctionDesc(Elf_Addr target)
2694 FunctionDesc *olddesc = (FunctionDesc *)target;
2695 FunctionDesc *newdesc;
2697 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2698 newdesc->gp = olddesc->gp;
2699 return (Elf_Addr)newdesc;
2704 #ifdef ia64_HOST_ARCH
2705 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2706 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2708 static unsigned char plt_code[] =
2710 /* taken from binutils bfd/elfxx-ia64.c */
2711 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2712 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2713 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2714 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2715 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2716 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2719 /* If we can't get to the function descriptor via gp, take a local copy of it */
2720 #define PLT_RELOC(code, target) { \
2721 Elf64_Sxword rel_value = target - gp_val; \
2722 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2723 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2725 ia64_reloc_gprel22((Elf_Addr)code, target); \
2730 unsigned char code[sizeof(plt_code)];
2734 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2736 PLTEntry *plt = (PLTEntry *)oc->plt;
2739 if (oc->pltIndex >= PLT_SIZE)
2740 barf("Procedure table overflow");
2742 entry = &plt[oc->pltIndex++];
2743 memcpy(entry->code, plt_code, sizeof(entry->code));
2744 PLT_RELOC(entry->code, target);
2745 return (Elf_Addr)entry;
2751 return (PLT_SIZE * sizeof(PLTEntry));
2756 #if x86_64_HOST_ARCH
2757 // On x86_64, 32-bit relocations are often used, which requires that
2758 // we can resolve a symbol to a 32-bit offset. However, shared
2759 // libraries are placed outside the 2Gb area, which leaves us with a
2760 // problem when we need to give a 32-bit offset to a symbol in a
2763 // For a function symbol, we can allocate a bounce sequence inside the
2764 // 2Gb area and resolve the symbol to this. The bounce sequence is
2765 // simply a long jump instruction to the real location of the symbol.
2767 // For data references, we're screwed.
2770 unsigned char jmp[8]; /* 6 byte instruction: jmpq *0x00000002(%rip) */
2774 #define X86_64_BB_SIZE 1024
2776 static x86_64_bounce *x86_64_bounce_buffer = NULL;
2777 static nat x86_64_bb_next_off;
2780 x86_64_high_symbol( char *lbl, void *addr )
2782 x86_64_bounce *bounce;
2784 if ( x86_64_bounce_buffer == NULL ||
2785 x86_64_bb_next_off >= X86_64_BB_SIZE ) {
2786 x86_64_bounce_buffer =
2787 mmap(NULL, X86_64_BB_SIZE * sizeof(x86_64_bounce),
2788 PROT_EXEC|PROT_READ|PROT_WRITE,
2789 MAP_PRIVATE|MAP_32BIT|MAP_ANONYMOUS, -1, 0);
2790 if (x86_64_bounce_buffer == MAP_FAILED) {
2791 barf("x86_64_high_symbol: mmap failed");
2793 x86_64_bb_next_off = 0;
2795 bounce = &x86_64_bounce_buffer[x86_64_bb_next_off];
2796 bounce->jmp[0] = 0xff;
2797 bounce->jmp[1] = 0x25;
2798 bounce->jmp[2] = 0x02;
2799 bounce->jmp[3] = 0x00;
2800 bounce->jmp[4] = 0x00;
2801 bounce->jmp[5] = 0x00;
2802 bounce->addr = addr;
2803 x86_64_bb_next_off++;
2805 IF_DEBUG(linker, debugBelch("x86_64: allocated bounce entry for %s->%p at %p\n",
2806 lbl, addr, bounce));
2808 insertStrHashTable(symhash, lbl, bounce);
2815 * Generic ELF functions
2819 findElfSection ( void* objImage, Elf_Word sh_type )
2821 char* ehdrC = (char*)objImage;
2822 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2823 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2824 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2828 for (i = 0; i < ehdr->e_shnum; i++) {
2829 if (shdr[i].sh_type == sh_type
2830 /* Ignore the section header's string table. */
2831 && i != ehdr->e_shstrndx
2832 /* Ignore string tables named .stabstr, as they contain
2834 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2836 ptr = ehdrC + shdr[i].sh_offset;
2843 #if defined(ia64_HOST_ARCH)
2845 findElfSegment ( void* objImage, Elf_Addr vaddr )
2847 char* ehdrC = (char*)objImage;
2848 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2849 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2850 Elf_Addr segaddr = 0;
2853 for (i = 0; i < ehdr->e_phnum; i++) {
2854 segaddr = phdr[i].p_vaddr;
2855 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2863 ocVerifyImage_ELF ( ObjectCode* oc )
2867 int i, j, nent, nstrtab, nsymtabs;
2871 char* ehdrC = (char*)(oc->image);
2872 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2874 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2875 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2876 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2877 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2878 errorBelch("%s: not an ELF object", oc->fileName);
2882 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2883 errorBelch("%s: unsupported ELF format", oc->fileName);
2887 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2888 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2890 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2891 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2893 errorBelch("%s: unknown endiannness", oc->fileName);
2897 if (ehdr->e_type != ET_REL) {
2898 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2901 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2903 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2904 switch (ehdr->e_machine) {
2905 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2906 #ifdef EM_SPARC32PLUS
2907 case EM_SPARC32PLUS:
2909 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2911 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2913 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2915 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2917 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2918 errorBelch("%s: unknown architecture", oc->fileName);
2922 IF_DEBUG(linker,debugBelch(
2923 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
2924 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2926 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2928 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2930 if (ehdr->e_shstrndx == SHN_UNDEF) {
2931 errorBelch("%s: no section header string table", oc->fileName);
2934 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2936 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2939 for (i = 0; i < ehdr->e_shnum; i++) {
2940 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2941 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2942 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2943 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2944 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2945 ehdrC + shdr[i].sh_offset,
2946 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2948 if (shdr[i].sh_type == SHT_REL) {
2949 IF_DEBUG(linker,debugBelch("Rel " ));
2950 } else if (shdr[i].sh_type == SHT_RELA) {
2951 IF_DEBUG(linker,debugBelch("RelA " ));
2953 IF_DEBUG(linker,debugBelch(" "));
2956 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2960 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2963 for (i = 0; i < ehdr->e_shnum; i++) {
2964 if (shdr[i].sh_type == SHT_STRTAB
2965 /* Ignore the section header's string table. */
2966 && i != ehdr->e_shstrndx
2967 /* Ignore string tables named .stabstr, as they contain
2969 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2971 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2972 strtab = ehdrC + shdr[i].sh_offset;
2977 errorBelch("%s: no string tables, or too many", oc->fileName);
2982 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2983 for (i = 0; i < ehdr->e_shnum; i++) {
2984 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2985 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2987 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2988 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2989 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
2991 (long)shdr[i].sh_size % sizeof(Elf_Sym)
2993 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2994 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2997 for (j = 0; j < nent; j++) {
2998 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2999 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3000 (int)stab[j].st_shndx,
3001 (int)stab[j].st_size,
3002 (char*)stab[j].st_value ));
3004 IF_DEBUG(linker,debugBelch("type=" ));
3005 switch (ELF_ST_TYPE(stab[j].st_info)) {
3006 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3007 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3008 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3009 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3010 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3011 default: IF_DEBUG(linker,debugBelch("? " )); break;
3013 IF_DEBUG(linker,debugBelch(" " ));
3015 IF_DEBUG(linker,debugBelch("bind=" ));
3016 switch (ELF_ST_BIND(stab[j].st_info)) {
3017 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3018 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3019 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3020 default: IF_DEBUG(linker,debugBelch("? " )); break;
3022 IF_DEBUG(linker,debugBelch(" " ));
3024 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3028 if (nsymtabs == 0) {
3029 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3036 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3040 if (hdr->sh_type == SHT_PROGBITS
3041 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3042 /* .text-style section */
3043 return SECTIONKIND_CODE_OR_RODATA;
3046 if (hdr->sh_type == SHT_PROGBITS
3047 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3048 /* .data-style section */
3049 return SECTIONKIND_RWDATA;
3052 if (hdr->sh_type == SHT_PROGBITS
3053 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3054 /* .rodata-style section */
3055 return SECTIONKIND_CODE_OR_RODATA;
3058 if (hdr->sh_type == SHT_NOBITS
3059 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3060 /* .bss-style section */
3062 return SECTIONKIND_RWDATA;
3065 return SECTIONKIND_OTHER;
3070 ocGetNames_ELF ( ObjectCode* oc )
3075 char* ehdrC = (char*)(oc->image);
3076 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3077 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3078 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3080 ASSERT(symhash != NULL);
3083 errorBelch("%s: no strtab", oc->fileName);
3088 for (i = 0; i < ehdr->e_shnum; i++) {
3089 /* Figure out what kind of section it is. Logic derived from
3090 Figure 1.14 ("Special Sections") of the ELF document
3091 ("Portable Formats Specification, Version 1.1"). */
3093 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3095 if (is_bss && shdr[i].sh_size > 0) {
3096 /* This is a non-empty .bss section. Allocate zeroed space for
3097 it, and set its .sh_offset field such that
3098 ehdrC + .sh_offset == addr_of_zeroed_space. */
3099 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3100 "ocGetNames_ELF(BSS)");
3101 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3103 debugBelch("BSS section at 0x%x, size %d\n",
3104 zspace, shdr[i].sh_size);
3108 /* fill in the section info */
3109 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3110 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3111 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3112 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3115 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3117 /* copy stuff into this module's object symbol table */
3118 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3119 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3121 oc->n_symbols = nent;
3122 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3123 "ocGetNames_ELF(oc->symbols)");
3125 for (j = 0; j < nent; j++) {
3127 char isLocal = FALSE; /* avoids uninit-var warning */
3129 char* nm = strtab + stab[j].st_name;
3130 int secno = stab[j].st_shndx;
3132 /* Figure out if we want to add it; if so, set ad to its
3133 address. Otherwise leave ad == NULL. */
3135 if (secno == SHN_COMMON) {
3137 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3139 debugBelch("COMMON symbol, size %d name %s\n",
3140 stab[j].st_size, nm);
3142 /* Pointless to do addProddableBlock() for this area,
3143 since the linker should never poke around in it. */
3146 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3147 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3149 /* and not an undefined symbol */
3150 && stab[j].st_shndx != SHN_UNDEF
3151 /* and not in a "special section" */
3152 && stab[j].st_shndx < SHN_LORESERVE
3154 /* and it's a not a section or string table or anything silly */
3155 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3156 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3157 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3160 /* Section 0 is the undefined section, hence > and not >=. */
3161 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3163 if (shdr[secno].sh_type == SHT_NOBITS) {
3164 debugBelch(" BSS symbol, size %d off %d name %s\n",
3165 stab[j].st_size, stab[j].st_value, nm);
3168 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3169 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3172 #ifdef ELF_FUNCTION_DESC
3173 /* dlsym() and the initialisation table both give us function
3174 * descriptors, so to be consistent we store function descriptors
3175 * in the symbol table */
3176 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3177 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3179 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
3180 ad, oc->fileName, nm ));
3185 /* And the decision is ... */
3189 oc->symbols[j] = nm;
3192 /* Ignore entirely. */
3194 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3198 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3199 strtab + stab[j].st_name ));
3202 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3203 (int)ELF_ST_BIND(stab[j].st_info),
3204 (int)ELF_ST_TYPE(stab[j].st_info),
3205 (int)stab[j].st_shndx,
3206 strtab + stab[j].st_name
3209 oc->symbols[j] = NULL;
3218 /* Do ELF relocations which lack an explicit addend. All x86-linux
3219 relocations appear to be of this form. */
3221 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3222 Elf_Shdr* shdr, int shnum,
3223 Elf_Sym* stab, char* strtab )
3228 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3229 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3230 int target_shndx = shdr[shnum].sh_info;
3231 int symtab_shndx = shdr[shnum].sh_link;
3233 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3234 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3235 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3236 target_shndx, symtab_shndx ));
3238 /* Skip sections that we're not interested in. */
3241 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3242 if (kind == SECTIONKIND_OTHER) {
3243 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3248 for (j = 0; j < nent; j++) {
3249 Elf_Addr offset = rtab[j].r_offset;
3250 Elf_Addr info = rtab[j].r_info;
3252 Elf_Addr P = ((Elf_Addr)targ) + offset;
3253 Elf_Word* pP = (Elf_Word*)P;
3258 StgStablePtr stablePtr;
3261 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3262 j, (void*)offset, (void*)info ));
3264 IF_DEBUG(linker,debugBelch( " ZERO" ));
3267 Elf_Sym sym = stab[ELF_R_SYM(info)];
3268 /* First see if it is a local symbol. */
3269 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3270 /* Yes, so we can get the address directly from the ELF symbol
3272 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3274 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3275 + stab[ELF_R_SYM(info)].st_value);
3278 symbol = strtab + sym.st_name;
3279 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3280 if (NULL == stablePtr) {
3281 /* No, so look up the name in our global table. */
3282 S_tmp = lookupSymbol( symbol );
3283 S = (Elf_Addr)S_tmp;
3285 stableVal = deRefStablePtr( stablePtr );
3286 addRootObject((void*)P);
3288 S = (Elf_Addr)S_tmp;
3292 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3295 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3298 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3299 (void*)P, (void*)S, (void*)A ));
3300 checkProddableBlock ( oc, pP );
3304 switch (ELF_R_TYPE(info)) {
3305 # ifdef i386_HOST_ARCH
3306 case R_386_32: *pP = value; break;
3307 case R_386_PC32: *pP = value - P; break;
3310 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3311 oc->fileName, (lnat)ELF_R_TYPE(info));
3319 /* Do ELF relocations for which explicit addends are supplied.
3320 sparc-solaris relocations appear to be of this form. */
3322 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3323 Elf_Shdr* shdr, int shnum,
3324 Elf_Sym* stab, char* strtab )
3327 char *symbol = NULL;
3329 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3330 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3331 int target_shndx = shdr[shnum].sh_info;
3332 int symtab_shndx = shdr[shnum].sh_link;
3334 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3335 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3336 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3337 target_shndx, symtab_shndx ));
3339 for (j = 0; j < nent; j++) {
3340 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3341 /* This #ifdef only serves to avoid unused-var warnings. */
3342 Elf_Addr offset = rtab[j].r_offset;
3343 Elf_Addr P = targ + offset;
3345 Elf_Addr info = rtab[j].r_info;
3346 Elf_Addr A = rtab[j].r_addend;
3350 # if defined(sparc_HOST_ARCH)
3351 Elf_Word* pP = (Elf_Word*)P;
3353 # elif defined(ia64_HOST_ARCH)
3354 Elf64_Xword *pP = (Elf64_Xword *)P;
3356 # elif defined(powerpc_HOST_ARCH)
3360 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3361 j, (void*)offset, (void*)info,
3364 IF_DEBUG(linker,debugBelch( " ZERO" ));
3367 Elf_Sym sym = stab[ELF_R_SYM(info)];
3368 /* First see if it is a local symbol. */
3369 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3370 /* Yes, so we can get the address directly from the ELF symbol
3372 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3374 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3375 + stab[ELF_R_SYM(info)].st_value);
3376 #ifdef ELF_FUNCTION_DESC
3377 /* Make a function descriptor for this function */
3378 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3379 S = allocateFunctionDesc(S + A);
3384 /* No, so look up the name in our global table. */
3385 symbol = strtab + sym.st_name;
3386 S_tmp = lookupSymbol( symbol );
3387 S = (Elf_Addr)S_tmp;
3389 #ifdef ELF_FUNCTION_DESC
3390 /* If a function, already a function descriptor - we would
3391 have to copy it to add an offset. */
3392 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3393 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3397 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3400 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3403 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3404 (void*)P, (void*)S, (void*)A ));
3405 /* checkProddableBlock ( oc, (void*)P ); */
3409 switch (ELF_R_TYPE(info)) {
3410 # if defined(sparc_HOST_ARCH)
3411 case R_SPARC_WDISP30:
3412 w1 = *pP & 0xC0000000;
3413 w2 = (Elf_Word)((value - P) >> 2);
3414 ASSERT((w2 & 0xC0000000) == 0);
3419 w1 = *pP & 0xFFC00000;
3420 w2 = (Elf_Word)(value >> 10);
3421 ASSERT((w2 & 0xFFC00000) == 0);
3427 w2 = (Elf_Word)(value & 0x3FF);
3428 ASSERT((w2 & ~0x3FF) == 0);
3432 /* According to the Sun documentation:
3434 This relocation type resembles R_SPARC_32, except it refers to an
3435 unaligned word. That is, the word to be relocated must be treated
3436 as four separate bytes with arbitrary alignment, not as a word
3437 aligned according to the architecture requirements.
3439 (JRS: which means that freeloading on the R_SPARC_32 case
3440 is probably wrong, but hey ...)
3444 w2 = (Elf_Word)value;
3447 # elif defined(ia64_HOST_ARCH)
3448 case R_IA64_DIR64LSB:
3449 case R_IA64_FPTR64LSB:
3452 case R_IA64_PCREL64LSB:
3455 case R_IA64_SEGREL64LSB:
3456 addr = findElfSegment(ehdrC, value);
3459 case R_IA64_GPREL22:
3460 ia64_reloc_gprel22(P, value);
3462 case R_IA64_LTOFF22:
3463 case R_IA64_LTOFF22X:
3464 case R_IA64_LTOFF_FPTR22:
3465 addr = allocateGOTEntry(value);
3466 ia64_reloc_gprel22(P, addr);
3468 case R_IA64_PCREL21B:
3469 ia64_reloc_pcrel21(P, S, oc);
3472 /* This goes with R_IA64_LTOFF22X and points to the load to
3473 * convert into a move. We don't implement relaxation. */
3475 # elif defined(powerpc_HOST_ARCH)
3476 case R_PPC_ADDR16_LO:
3477 *(Elf32_Half*) P = value;
3480 case R_PPC_ADDR16_HI:
3481 *(Elf32_Half*) P = value >> 16;
3484 case R_PPC_ADDR16_HA:
3485 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3489 *(Elf32_Word *) P = value;
3493 *(Elf32_Word *) P = value - P;
3499 if( delta << 6 >> 6 != delta )
3501 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3504 if( value == 0 || delta << 6 >> 6 != delta )
3506 barf( "Unable to make ppcJumpIsland for #%d",
3512 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3513 | (delta & 0x3fffffc);
3517 #if x86_64_HOST_ARCH
3519 *(Elf64_Xword *)P = value;
3524 StgInt64 off = value - P;
3525 if (off >= 0x7fffffffL || off < -0x80000000L) {
3526 barf("R_X86_64_PC32 relocation out of range: %s = %p",
3529 *(Elf64_Word *)P = (Elf64_Word)off;
3534 if (value >= 0x7fffffffL) {
3535 barf("R_X86_64_32 relocation out of range: %s = %p\n",
3538 *(Elf64_Word *)P = (Elf64_Word)value;
3542 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3543 barf("R_X86_64_32S relocation out of range: %s = %p\n",
3546 *(Elf64_Sword *)P = (Elf64_Sword)value;
3551 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3552 oc->fileName, (lnat)ELF_R_TYPE(info));
3561 ocResolve_ELF ( ObjectCode* oc )
3565 Elf_Sym* stab = NULL;
3566 char* ehdrC = (char*)(oc->image);
3567 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3568 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3570 /* first find "the" symbol table */
3571 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3573 /* also go find the string table */
3574 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3576 if (stab == NULL || strtab == NULL) {
3577 errorBelch("%s: can't find string or symbol table", oc->fileName);
3581 /* Process the relocation sections. */
3582 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3583 if (shdr[shnum].sh_type == SHT_REL) {
3584 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3585 shnum, stab, strtab );
3589 if (shdr[shnum].sh_type == SHT_RELA) {
3590 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3591 shnum, stab, strtab );
3596 /* Free the local symbol table; we won't need it again. */
3597 freeHashTable(oc->lochash, NULL);
3600 #if defined(powerpc_HOST_ARCH)
3601 ocFlushInstructionCache( oc );
3609 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3610 * at the front. The following utility functions pack and unpack instructions, and
3611 * take care of the most common relocations.
3614 #ifdef ia64_HOST_ARCH
3617 ia64_extract_instruction(Elf64_Xword *target)
3620 int slot = (Elf_Addr)target & 3;
3621 target = (Elf_Addr)target & ~3;
3629 return ((w1 >> 5) & 0x1ffffffffff);
3631 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3635 barf("ia64_extract_instruction: invalid slot %p", target);
3640 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3642 int slot = (Elf_Addr)target & 3;
3643 target = (Elf_Addr)target & ~3;
3648 *target |= value << 5;
3651 *target |= value << 46;
3652 *(target+1) |= value >> 18;
3655 *(target+1) |= value << 23;
3661 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3663 Elf64_Xword instruction;
3664 Elf64_Sxword rel_value;
3666 rel_value = value - gp_val;
3667 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3668 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3670 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3671 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3672 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3673 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3674 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3675 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3679 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3681 Elf64_Xword instruction;
3682 Elf64_Sxword rel_value;
3685 entry = allocatePLTEntry(value, oc);
3687 rel_value = (entry >> 4) - (target >> 4);
3688 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3689 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3691 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3692 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3693 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3694 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3700 * PowerPC ELF specifics
3703 #ifdef powerpc_HOST_ARCH
3705 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3711 ehdr = (Elf_Ehdr *) oc->image;
3712 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3714 for( i = 0; i < ehdr->e_shnum; i++ )
3715 if( shdr[i].sh_type == SHT_SYMTAB )
3718 if( i == ehdr->e_shnum )
3720 errorBelch( "This ELF file contains no symtab" );
3724 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3726 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3727 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3732 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3735 #endif /* powerpc */
3739 /* --------------------------------------------------------------------------
3741 * ------------------------------------------------------------------------*/
3743 #if defined(OBJFORMAT_MACHO)
3746 Support for MachO linking on Darwin/MacOS X
3747 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3749 I hereby formally apologize for the hackish nature of this code.
3750 Things that need to be done:
3751 *) implement ocVerifyImage_MachO
3752 *) add still more sanity checks.
3755 #ifdef powerpc_HOST_ARCH
3756 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3758 struct mach_header *header = (struct mach_header *) oc->image;
3759 struct load_command *lc = (struct load_command *) (header + 1);
3762 for( i = 0; i < header->ncmds; i++ )
3764 if( lc->cmd == LC_SYMTAB )
3766 // Find out the first and last undefined external
3767 // symbol, so we don't have to allocate too many
3769 struct symtab_command *symLC = (struct symtab_command *) lc;
3770 unsigned min = symLC->nsyms, max = 0;
3771 struct nlist *nlist =
3772 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3774 for(i=0;i<symLC->nsyms;i++)
3776 if(nlist[i].n_type & N_STAB)
3778 else if(nlist[i].n_type & N_EXT)
3780 if((nlist[i].n_type & N_TYPE) == N_UNDF
3781 && (nlist[i].n_value == 0))
3791 return ocAllocateJumpIslands(oc, max - min + 1, min);
3796 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3798 return ocAllocateJumpIslands(oc,0,0);
3802 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3804 // FIXME: do some verifying here
3808 static int resolveImports(
3811 struct symtab_command *symLC,
3812 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3813 unsigned long *indirectSyms,
3814 struct nlist *nlist)
3817 size_t itemSize = 4;
3820 int isJumpTable = 0;
3821 if(!strcmp(sect->sectname,"__jump_table"))
3825 ASSERT(sect->reserved2 == itemSize);
3829 for(i=0; i*itemSize < sect->size;i++)
3831 // according to otool, reserved1 contains the first index into the indirect symbol table
3832 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3833 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3836 if((symbol->n_type & N_TYPE) == N_UNDF
3837 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3838 addr = (void*) (symbol->n_value);
3839 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3842 addr = lookupSymbol(nm);
3845 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3853 checkProddableBlock(oc,image + sect->offset + i*itemSize);
3854 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
3855 *(unsigned*)(image + sect->offset + i*itemSize + 1)
3856 = (char*)addr - (image + sect->offset + i*itemSize + 5);
3861 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3862 ((void**)(image + sect->offset))[i] = addr;
3869 static unsigned long relocateAddress(
3872 struct section* sections,
3873 unsigned long address)
3876 for(i = 0; i < nSections; i++)
3878 if(sections[i].addr <= address
3879 && address < sections[i].addr + sections[i].size)
3881 return (unsigned long)oc->image
3882 + sections[i].offset + address - sections[i].addr;
3885 barf("Invalid Mach-O file:"
3886 "Address out of bounds while relocating object file");
3890 static int relocateSection(
3893 struct symtab_command *symLC, struct nlist *nlist,
3894 int nSections, struct section* sections, struct section *sect)
3896 struct relocation_info *relocs;
3899 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3901 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3903 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
3905 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
3909 relocs = (struct relocation_info*) (image + sect->reloff);
3913 if(relocs[i].r_address & R_SCATTERED)
3915 struct scattered_relocation_info *scat =
3916 (struct scattered_relocation_info*) &relocs[i];
3920 if(scat->r_length == 2)
3922 unsigned long word = 0;
3923 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3924 checkProddableBlock(oc,wordPtr);
3926 // Note on relocation types:
3927 // i386 uses the GENERIC_RELOC_* types,
3928 // while ppc uses special PPC_RELOC_* types.
3929 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
3930 // in both cases, all others are different.
3931 // Therefore, we use GENERIC_RELOC_VANILLA
3932 // and GENERIC_RELOC_PAIR instead of the PPC variants,
3933 // and use #ifdefs for the other types.
3935 // Step 1: Figure out what the relocated value should be
3936 if(scat->r_type == GENERIC_RELOC_VANILLA)
3938 word = *wordPtr + (unsigned long) relocateAddress(
3945 #ifdef powerpc_HOST_ARCH
3946 else if(scat->r_type == PPC_RELOC_SECTDIFF
3947 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3948 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3949 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3951 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
3954 struct scattered_relocation_info *pair =
3955 (struct scattered_relocation_info*) &relocs[i+1];
3957 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
3958 barf("Invalid Mach-O file: "
3959 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
3961 word = (unsigned long)
3962 (relocateAddress(oc, nSections, sections, scat->r_value)
3963 - relocateAddress(oc, nSections, sections, pair->r_value));
3966 #ifdef powerpc_HOST_ARCH
3967 else if(scat->r_type == PPC_RELOC_HI16
3968 || scat->r_type == PPC_RELOC_LO16
3969 || scat->r_type == PPC_RELOC_HA16
3970 || scat->r_type == PPC_RELOC_LO14)
3971 { // these are generated by label+offset things
3972 struct relocation_info *pair = &relocs[i+1];
3973 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3974 barf("Invalid Mach-O file: "
3975 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3977 if(scat->r_type == PPC_RELOC_LO16)
3979 word = ((unsigned short*) wordPtr)[1];
3980 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3982 else if(scat->r_type == PPC_RELOC_LO14)
3984 barf("Unsupported Relocation: PPC_RELOC_LO14");
3985 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3986 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3988 else if(scat->r_type == PPC_RELOC_HI16)
3990 word = ((unsigned short*) wordPtr)[1] << 16;
3991 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3993 else if(scat->r_type == PPC_RELOC_HA16)
3995 word = ((unsigned short*) wordPtr)[1] << 16;
3996 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4000 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4007 continue; // ignore the others
4009 #ifdef powerpc_HOST_ARCH
4010 if(scat->r_type == GENERIC_RELOC_VANILLA
4011 || scat->r_type == PPC_RELOC_SECTDIFF)
4013 if(scat->r_type == GENERIC_RELOC_VANILLA
4014 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4019 #ifdef powerpc_HOST_ARCH
4020 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4022 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4024 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4026 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4028 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4030 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4031 + ((word & (1<<15)) ? 1 : 0);
4037 continue; // FIXME: I hope it's OK to ignore all the others.
4041 struct relocation_info *reloc = &relocs[i];
4042 if(reloc->r_pcrel && !reloc->r_extern)
4045 if(reloc->r_length == 2)
4047 unsigned long word = 0;
4048 #ifdef powerpc_HOST_ARCH
4049 unsigned long jumpIsland = 0;
4050 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4051 // to avoid warning and to catch
4055 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4056 checkProddableBlock(oc,wordPtr);
4058 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4062 #ifdef powerpc_HOST_ARCH
4063 else if(reloc->r_type == PPC_RELOC_LO16)
4065 word = ((unsigned short*) wordPtr)[1];
4066 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4068 else if(reloc->r_type == PPC_RELOC_HI16)
4070 word = ((unsigned short*) wordPtr)[1] << 16;
4071 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4073 else if(reloc->r_type == PPC_RELOC_HA16)
4075 word = ((unsigned short*) wordPtr)[1] << 16;
4076 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4078 else if(reloc->r_type == PPC_RELOC_BR24)
4081 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4085 if(!reloc->r_extern)
4088 sections[reloc->r_symbolnum-1].offset
4089 - sections[reloc->r_symbolnum-1].addr
4096 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4097 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4098 void *symbolAddress = lookupSymbol(nm);
4101 errorBelch("\nunknown symbol `%s'", nm);
4107 #ifdef powerpc_HOST_ARCH
4108 // In the .o file, this should be a relative jump to NULL
4109 // and we'll change it to a relative jump to the symbol
4110 ASSERT(-word == reloc->r_address);
4111 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,(unsigned long) symbolAddress);
4114 offsetToJumpIsland = word + jumpIsland
4115 - (((long)image) + sect->offset - sect->addr);
4118 word += (unsigned long) symbolAddress
4119 - (((long)image) + sect->offset - sect->addr);
4123 word += (unsigned long) symbolAddress;
4127 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4132 #ifdef powerpc_HOST_ARCH
4133 else if(reloc->r_type == PPC_RELOC_LO16)
4135 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4138 else if(reloc->r_type == PPC_RELOC_HI16)
4140 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4143 else if(reloc->r_type == PPC_RELOC_HA16)
4145 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4146 + ((word & (1<<15)) ? 1 : 0);
4149 else if(reloc->r_type == PPC_RELOC_BR24)
4151 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4153 // The branch offset is too large.
4154 // Therefore, we try to use a jump island.
4157 barf("unconditional relative branch out of range: "
4158 "no jump island available");
4161 word = offsetToJumpIsland;
4162 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4163 barf("unconditional relative branch out of range: "
4164 "jump island out of range");
4166 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4171 barf("\nunknown relocation %d",reloc->r_type);
4178 static int ocGetNames_MachO(ObjectCode* oc)
4180 char *image = (char*) oc->image;
4181 struct mach_header *header = (struct mach_header*) image;
4182 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4183 unsigned i,curSymbol = 0;
4184 struct segment_command *segLC = NULL;
4185 struct section *sections;
4186 struct symtab_command *symLC = NULL;
4187 struct nlist *nlist;
4188 unsigned long commonSize = 0;
4189 char *commonStorage = NULL;
4190 unsigned long commonCounter;
4192 for(i=0;i<header->ncmds;i++)
4194 if(lc->cmd == LC_SEGMENT)
4195 segLC = (struct segment_command*) lc;
4196 else if(lc->cmd == LC_SYMTAB)
4197 symLC = (struct symtab_command*) lc;
4198 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4201 sections = (struct section*) (segLC+1);
4202 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4205 for(i=0;i<segLC->nsects;i++)
4207 if(sections[i].size == 0)
4210 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4212 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4213 "ocGetNames_MachO(common symbols)");
4214 sections[i].offset = zeroFillArea - image;
4217 if(!strcmp(sections[i].sectname,"__text"))
4218 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4219 (void*) (image + sections[i].offset),
4220 (void*) (image + sections[i].offset + sections[i].size));
4221 else if(!strcmp(sections[i].sectname,"__const"))
4222 addSection(oc, SECTIONKIND_RWDATA,
4223 (void*) (image + sections[i].offset),
4224 (void*) (image + sections[i].offset + sections[i].size));
4225 else if(!strcmp(sections[i].sectname,"__data"))
4226 addSection(oc, SECTIONKIND_RWDATA,
4227 (void*) (image + sections[i].offset),
4228 (void*) (image + sections[i].offset + sections[i].size));
4229 else if(!strcmp(sections[i].sectname,"__bss")
4230 || !strcmp(sections[i].sectname,"__common"))
4231 addSection(oc, SECTIONKIND_RWDATA,
4232 (void*) (image + sections[i].offset),
4233 (void*) (image + sections[i].offset + sections[i].size));
4235 addProddableBlock(oc, (void*) (image + sections[i].offset),
4239 // count external symbols defined here
4243 for(i=0;i<symLC->nsyms;i++)
4245 if(nlist[i].n_type & N_STAB)
4247 else if(nlist[i].n_type & N_EXT)
4249 if((nlist[i].n_type & N_TYPE) == N_UNDF
4250 && (nlist[i].n_value != 0))
4252 commonSize += nlist[i].n_value;
4255 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4260 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4261 "ocGetNames_MachO(oc->symbols)");
4265 for(i=0;i<symLC->nsyms;i++)
4267 if(nlist[i].n_type & N_STAB)
4269 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4271 if(nlist[i].n_type & N_EXT)
4273 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4274 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4276 + sections[nlist[i].n_sect-1].offset
4277 - sections[nlist[i].n_sect-1].addr
4278 + nlist[i].n_value);
4279 oc->symbols[curSymbol++] = nm;
4283 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4284 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
4286 + sections[nlist[i].n_sect-1].offset
4287 - sections[nlist[i].n_sect-1].addr
4288 + nlist[i].n_value);
4294 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4295 commonCounter = (unsigned long)commonStorage;
4298 for(i=0;i<symLC->nsyms;i++)
4300 if((nlist[i].n_type & N_TYPE) == N_UNDF
4301 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4303 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4304 unsigned long sz = nlist[i].n_value;
4306 nlist[i].n_value = commonCounter;
4308 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4309 (void*)commonCounter);
4310 oc->symbols[curSymbol++] = nm;
4312 commonCounter += sz;
4319 static int ocResolve_MachO(ObjectCode* oc)
4321 char *image = (char*) oc->image;
4322 struct mach_header *header = (struct mach_header*) image;
4323 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4325 struct segment_command *segLC = NULL;
4326 struct section *sections;
4327 struct symtab_command *symLC = NULL;
4328 struct dysymtab_command *dsymLC = NULL;
4329 struct nlist *nlist;
4331 for(i=0;i<header->ncmds;i++)
4333 if(lc->cmd == LC_SEGMENT)
4334 segLC = (struct segment_command*) lc;
4335 else if(lc->cmd == LC_SYMTAB)
4336 symLC = (struct symtab_command*) lc;
4337 else if(lc->cmd == LC_DYSYMTAB)
4338 dsymLC = (struct dysymtab_command*) lc;
4339 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4342 sections = (struct section*) (segLC+1);
4343 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4348 unsigned long *indirectSyms
4349 = (unsigned long*) (image + dsymLC->indirectsymoff);
4351 for(i=0;i<segLC->nsects;i++)
4353 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4354 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4355 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4357 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4360 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4361 || !strcmp(sections[i].sectname,"__pointers"))
4363 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4366 else if(!strcmp(sections[i].sectname,"__jump_table"))
4368 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4374 for(i=0;i<segLC->nsects;i++)
4376 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4380 /* Free the local symbol table; we won't need it again. */
4381 freeHashTable(oc->lochash, NULL);
4384 #if defined (powerpc_HOST_ARCH)
4385 ocFlushInstructionCache( oc );
4391 #ifdef powerpc_HOST_ARCH
4393 * The Mach-O object format uses leading underscores. But not everywhere.
4394 * There is a small number of runtime support functions defined in
4395 * libcc_dynamic.a whose name does not have a leading underscore.
4396 * As a consequence, we can't get their address from C code.
4397 * We have to use inline assembler just to take the address of a function.
4401 static void machoInitSymbolsWithoutUnderscore()
4403 extern void* symbolsWithoutUnderscore[];
4404 void **p = symbolsWithoutUnderscore;
4405 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4409 __asm__ volatile(".long " # x);
4411 RTS_MACHO_NOUNDERLINE_SYMBOLS
4413 __asm__ volatile(".text");
4417 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4419 RTS_MACHO_NOUNDERLINE_SYMBOLS
4426 * Figure out by how much to shift the entire Mach-O file in memory
4427 * when loading so that its single segment ends up 16-byte-aligned
4429 static int machoGetMisalignment( FILE * f )
4431 struct mach_header header;
4434 fread(&header, sizeof(header), 1, f);
4437 if(header.magic != MH_MAGIC)
4440 misalignment = (header.sizeofcmds + sizeof(header))
4443 return misalignment ? (16 - misalignment) : 0;