1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
25 #include "LinkerInternals.h"
30 #ifdef HAVE_SYS_TYPES_H
31 #include <sys/types.h>
37 #ifdef HAVE_SYS_STAT_H
41 #if defined(HAVE_DLFCN_H)
45 #if defined(cygwin32_HOST_OS)
50 #ifdef HAVE_SYS_TIME_H
54 #include <sys/fcntl.h>
55 #include <sys/termios.h>
56 #include <sys/utime.h>
57 #include <sys/utsname.h>
61 #if defined(ia64_HOST_ARCH) || defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
66 #if defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS)
74 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
75 # define OBJFORMAT_ELF
76 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
77 # define OBJFORMAT_PEi386
80 #elif defined(darwin_HOST_OS)
81 # define OBJFORMAT_MACHO
82 # include <mach-o/loader.h>
83 # include <mach-o/nlist.h>
84 # include <mach-o/reloc.h>
85 # include <mach-o/dyld.h>
86 #if defined(powerpc_HOST_ARCH)
87 # include <mach-o/ppc/reloc.h>
91 /* Hash table mapping symbol names to Symbol */
92 static /*Str*/HashTable *symhash;
94 /* List of currently loaded objects */
95 ObjectCode *objects = NULL; /* initially empty */
97 #if defined(OBJFORMAT_ELF)
98 static int ocVerifyImage_ELF ( ObjectCode* oc );
99 static int ocGetNames_ELF ( ObjectCode* oc );
100 static int ocResolve_ELF ( ObjectCode* oc );
101 #if defined(powerpc_HOST_ARCH)
102 static int ocAllocateJumpIslands_ELF ( ObjectCode* oc );
104 #elif defined(OBJFORMAT_PEi386)
105 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
106 static int ocGetNames_PEi386 ( ObjectCode* oc );
107 static int ocResolve_PEi386 ( ObjectCode* oc );
108 #elif defined(OBJFORMAT_MACHO)
109 static int ocVerifyImage_MachO ( ObjectCode* oc );
110 static int ocGetNames_MachO ( ObjectCode* oc );
111 static int ocResolve_MachO ( ObjectCode* oc );
113 static int machoGetMisalignment( FILE * );
114 #ifdef powerpc_HOST_ARCH
115 static int ocAllocateJumpIslands_MachO ( ObjectCode* oc );
116 static void machoInitSymbolsWithoutUnderscore( void );
120 #if defined(x86_64_HOST_ARCH)
121 static void*x86_64_high_symbol( char *lbl, void *addr );
124 /* -----------------------------------------------------------------------------
125 * Built-in symbols from the RTS
128 typedef struct _RtsSymbolVal {
135 #define Maybe_Stable_Names SymX(mkWeakzh_fast) \
136 SymX(makeStableNamezh_fast) \
137 SymX(finalizzeWeakzh_fast)
139 /* These are not available in GUM!!! -- HWL */
140 #define Maybe_Stable_Names
143 #if !defined (mingw32_HOST_OS)
144 #define RTS_POSIX_ONLY_SYMBOLS \
145 SymX(stg_sig_install) \
149 #if defined (cygwin32_HOST_OS)
150 #define RTS_MINGW_ONLY_SYMBOLS /**/
151 /* Don't have the ability to read import libs / archives, so
152 * we have to stupidly list a lot of what libcygwin.a
155 #define RTS_CYGWIN_ONLY_SYMBOLS \
233 #elif !defined(mingw32_HOST_OS)
234 #define RTS_MINGW_ONLY_SYMBOLS /**/
235 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
236 #else /* defined(mingw32_HOST_OS) */
237 #define RTS_POSIX_ONLY_SYMBOLS /**/
238 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
240 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
242 #define RTS_MINGW_EXTRA_SYMS \
243 Sym(_imp____mb_cur_max) \
246 #define RTS_MINGW_EXTRA_SYMS
249 /* These are statically linked from the mingw libraries into the ghc
250 executable, so we have to employ this hack. */
251 #define RTS_MINGW_ONLY_SYMBOLS \
252 SymX(asyncReadzh_fast) \
253 SymX(asyncWritezh_fast) \
254 SymX(asyncDoProczh_fast) \
266 SymX(getservbyname) \
267 SymX(getservbyport) \
268 SymX(getprotobynumber) \
269 SymX(getprotobyname) \
270 SymX(gethostbyname) \
271 SymX(gethostbyaddr) \
305 SymX(rts_InstallConsoleEvent) \
306 SymX(rts_ConsoleHandlerDone) \
308 Sym(_imp___timezone) \
317 RTS_MINGW_EXTRA_SYMS \
321 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
322 #define RTS_DARWIN_ONLY_SYMBOLS \
323 Sym(asprintf$LDBLStub) \
327 Sym(fprintf$LDBLStub) \
328 Sym(fscanf$LDBLStub) \
329 Sym(fwprintf$LDBLStub) \
330 Sym(fwscanf$LDBLStub) \
331 Sym(printf$LDBLStub) \
332 Sym(scanf$LDBLStub) \
333 Sym(snprintf$LDBLStub) \
334 Sym(sprintf$LDBLStub) \
335 Sym(sscanf$LDBLStub) \
336 Sym(strtold$LDBLStub) \
337 Sym(swprintf$LDBLStub) \
338 Sym(swscanf$LDBLStub) \
339 Sym(syslog$LDBLStub) \
340 Sym(vasprintf$LDBLStub) \
342 Sym(verrc$LDBLStub) \
343 Sym(verrx$LDBLStub) \
344 Sym(vfprintf$LDBLStub) \
345 Sym(vfscanf$LDBLStub) \
346 Sym(vfwprintf$LDBLStub) \
347 Sym(vfwscanf$LDBLStub) \
348 Sym(vprintf$LDBLStub) \
349 Sym(vscanf$LDBLStub) \
350 Sym(vsnprintf$LDBLStub) \
351 Sym(vsprintf$LDBLStub) \
352 Sym(vsscanf$LDBLStub) \
353 Sym(vswprintf$LDBLStub) \
354 Sym(vswscanf$LDBLStub) \
355 Sym(vsyslog$LDBLStub) \
356 Sym(vwarn$LDBLStub) \
357 Sym(vwarnc$LDBLStub) \
358 Sym(vwarnx$LDBLStub) \
359 Sym(vwprintf$LDBLStub) \
360 Sym(vwscanf$LDBLStub) \
362 Sym(warnc$LDBLStub) \
363 Sym(warnx$LDBLStub) \
364 Sym(wcstold$LDBLStub) \
365 Sym(wprintf$LDBLStub) \
368 #define RTS_DARWIN_ONLY_SYMBOLS
372 # define MAIN_CAP_SYM SymX(MainCapability)
374 # define MAIN_CAP_SYM
377 #if !defined(mingw32_HOST_OS)
378 #define RTS_USER_SIGNALS_SYMBOLS \
379 SymX(startSignalHandler) \
380 SymX(setIOManagerPipe)
382 #define RTS_USER_SIGNALS_SYMBOLS /* nothing */
385 #ifdef TABLES_NEXT_TO_CODE
386 #define RTS_RET_SYMBOLS /* nothing */
388 #define RTS_RET_SYMBOLS \
389 SymX(stg_enter_ret) \
390 SymX(stg_gc_fun_ret) \
398 SymX(stg_ap_pv_ret) \
399 SymX(stg_ap_pp_ret) \
400 SymX(stg_ap_ppv_ret) \
401 SymX(stg_ap_ppp_ret) \
402 SymX(stg_ap_pppv_ret) \
403 SymX(stg_ap_pppp_ret) \
404 SymX(stg_ap_ppppp_ret) \
405 SymX(stg_ap_pppppp_ret)
408 #define RTS_SYMBOLS \
411 SymX(stg_enter_info) \
412 SymX(stg_gc_void_info) \
413 SymX(__stg_gc_enter_1) \
414 SymX(stg_gc_noregs) \
415 SymX(stg_gc_unpt_r1_info) \
416 SymX(stg_gc_unpt_r1) \
417 SymX(stg_gc_unbx_r1_info) \
418 SymX(stg_gc_unbx_r1) \
419 SymX(stg_gc_f1_info) \
421 SymX(stg_gc_d1_info) \
423 SymX(stg_gc_l1_info) \
426 SymX(stg_gc_fun_info) \
428 SymX(stg_gc_gen_info) \
429 SymX(stg_gc_gen_hp) \
431 SymX(stg_gen_yield) \
432 SymX(stg_yield_noregs) \
433 SymX(stg_yield_to_interpreter) \
434 SymX(stg_gen_block) \
435 SymX(stg_block_noregs) \
437 SymX(stg_block_takemvar) \
438 SymX(stg_block_putmvar) \
439 SymX(stg_seq_frame_info) \
441 SymX(MallocFailHook) \
443 SymX(OutOfHeapHook) \
444 SymX(StackOverflowHook) \
445 SymX(__encodeDouble) \
446 SymX(__encodeFloat) \
450 SymX(__gmpz_cmp_si) \
451 SymX(__gmpz_cmp_ui) \
452 SymX(__gmpz_get_si) \
453 SymX(__gmpz_get_ui) \
454 SymX(__int_encodeDouble) \
455 SymX(__int_encodeFloat) \
456 SymX(andIntegerzh_fast) \
457 SymX(atomicallyzh_fast) \
461 SymX(blockAsyncExceptionszh_fast) \
463 SymX(catchRetryzh_fast) \
464 SymX(catchSTMzh_fast) \
465 SymX(closure_flags) \
467 SymX(cmpIntegerzh_fast) \
468 SymX(cmpIntegerIntzh_fast) \
469 SymX(complementIntegerzh_fast) \
470 SymX(createAdjustor) \
471 SymX(decodeDoublezh_fast) \
472 SymX(decodeFloatzh_fast) \
475 SymX(deRefWeakzh_fast) \
476 SymX(deRefStablePtrzh_fast) \
477 SymX(divExactIntegerzh_fast) \
478 SymX(divModIntegerzh_fast) \
481 SymX(forkOS_createThread) \
482 SymX(freeHaskellFunctionPtr) \
483 SymX(freeStablePtr) \
484 SymX(gcdIntegerzh_fast) \
485 SymX(gcdIntegerIntzh_fast) \
486 SymX(gcdIntzh_fast) \
495 SymX(hs_perform_gc) \
496 SymX(hs_free_stable_ptr) \
497 SymX(hs_free_fun_ptr) \
499 SymX(int2Integerzh_fast) \
500 SymX(integer2Intzh_fast) \
501 SymX(integer2Wordzh_fast) \
502 SymX(isCurrentThreadBoundzh_fast) \
503 SymX(isDoubleDenormalized) \
504 SymX(isDoubleInfinite) \
506 SymX(isDoubleNegativeZero) \
507 SymX(isEmptyMVarzh_fast) \
508 SymX(isFloatDenormalized) \
509 SymX(isFloatInfinite) \
511 SymX(isFloatNegativeZero) \
512 SymX(killThreadzh_fast) \
515 SymX(makeStablePtrzh_fast) \
516 SymX(minusIntegerzh_fast) \
517 SymX(mkApUpd0zh_fast) \
518 SymX(myThreadIdzh_fast) \
519 SymX(labelThreadzh_fast) \
520 SymX(newArrayzh_fast) \
521 SymX(newBCOzh_fast) \
522 SymX(newByteArrayzh_fast) \
523 SymX_redirect(newCAF, newDynCAF) \
524 SymX(newMVarzh_fast) \
525 SymX(newMutVarzh_fast) \
526 SymX(newTVarzh_fast) \
527 SymX(atomicModifyMutVarzh_fast) \
528 SymX(newPinnedByteArrayzh_fast) \
529 SymX(orIntegerzh_fast) \
531 SymX(performMajorGC) \
532 SymX(plusIntegerzh_fast) \
535 SymX(putMVarzh_fast) \
536 SymX(quotIntegerzh_fast) \
537 SymX(quotRemIntegerzh_fast) \
539 SymX(raiseIOzh_fast) \
540 SymX(readTVarzh_fast) \
541 SymX(remIntegerzh_fast) \
542 SymX(resetNonBlockingFd) \
547 SymX(rts_checkSchedStatus) \
550 SymX(rts_evalLazyIO) \
551 SymX(rts_evalStableIO) \
555 SymX(rts_getDouble) \
560 SymX(rts_getFunPtr) \
561 SymX(rts_getStablePtr) \
562 SymX(rts_getThreadId) \
564 SymX(rts_getWord32) \
577 SymX(rts_mkStablePtr) \
585 SymX(rtsSupportsBoundThreads) \
587 SymX(__hscore_get_saved_termios) \
588 SymX(__hscore_set_saved_termios) \
590 SymX(startupHaskell) \
591 SymX(shutdownHaskell) \
592 SymX(shutdownHaskellAndExit) \
593 SymX(stable_ptr_table) \
594 SymX(stackOverflow) \
595 SymX(stg_CAF_BLACKHOLE_info) \
596 SymX(awakenBlockedQueue) \
597 SymX(stg_CHARLIKE_closure) \
598 SymX(stg_EMPTY_MVAR_info) \
599 SymX(stg_IND_STATIC_info) \
600 SymX(stg_INTLIKE_closure) \
601 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
602 SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \
603 SymX(stg_WEAK_info) \
604 SymX(stg_ap_0_info) \
605 SymX(stg_ap_v_info) \
606 SymX(stg_ap_f_info) \
607 SymX(stg_ap_d_info) \
608 SymX(stg_ap_l_info) \
609 SymX(stg_ap_n_info) \
610 SymX(stg_ap_p_info) \
611 SymX(stg_ap_pv_info) \
612 SymX(stg_ap_pp_info) \
613 SymX(stg_ap_ppv_info) \
614 SymX(stg_ap_ppp_info) \
615 SymX(stg_ap_pppv_info) \
616 SymX(stg_ap_pppp_info) \
617 SymX(stg_ap_ppppp_info) \
618 SymX(stg_ap_pppppp_info) \
619 SymX(stg_ap_1_upd_info) \
620 SymX(stg_ap_2_upd_info) \
621 SymX(stg_ap_3_upd_info) \
622 SymX(stg_ap_4_upd_info) \
623 SymX(stg_ap_5_upd_info) \
624 SymX(stg_ap_6_upd_info) \
625 SymX(stg_ap_7_upd_info) \
627 SymX(stg_sel_0_upd_info) \
628 SymX(stg_sel_10_upd_info) \
629 SymX(stg_sel_11_upd_info) \
630 SymX(stg_sel_12_upd_info) \
631 SymX(stg_sel_13_upd_info) \
632 SymX(stg_sel_14_upd_info) \
633 SymX(stg_sel_15_upd_info) \
634 SymX(stg_sel_1_upd_info) \
635 SymX(stg_sel_2_upd_info) \
636 SymX(stg_sel_3_upd_info) \
637 SymX(stg_sel_4_upd_info) \
638 SymX(stg_sel_5_upd_info) \
639 SymX(stg_sel_6_upd_info) \
640 SymX(stg_sel_7_upd_info) \
641 SymX(stg_sel_8_upd_info) \
642 SymX(stg_sel_9_upd_info) \
643 SymX(stg_upd_frame_info) \
644 SymX(suspendThread) \
645 SymX(takeMVarzh_fast) \
646 SymX(timesIntegerzh_fast) \
647 SymX(tryPutMVarzh_fast) \
648 SymX(tryTakeMVarzh_fast) \
649 SymX(unblockAsyncExceptionszh_fast) \
651 SymX(unsafeThawArrayzh_fast) \
652 SymX(waitReadzh_fast) \
653 SymX(waitWritezh_fast) \
654 SymX(word2Integerzh_fast) \
655 SymX(writeTVarzh_fast) \
656 SymX(xorIntegerzh_fast) \
658 RTS_USER_SIGNALS_SYMBOLS
660 #ifdef SUPPORT_LONG_LONGS
661 #define RTS_LONG_LONG_SYMS \
662 SymX(int64ToIntegerzh_fast) \
663 SymX(word64ToIntegerzh_fast)
665 #define RTS_LONG_LONG_SYMS /* nothing */
668 // 64-bit support functions in libgcc.a
669 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
670 #define RTS_LIBGCC_SYMBOLS \
680 #elif defined(ia64_HOST_ARCH)
681 #define RTS_LIBGCC_SYMBOLS \
689 #define RTS_LIBGCC_SYMBOLS
692 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
693 // Symbols that don't have a leading underscore
694 // on Mac OS X. They have to receive special treatment,
695 // see machoInitSymbolsWithoutUnderscore()
696 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
701 /* entirely bogus claims about types of these symbols */
702 #define Sym(vvv) extern void vvv(void);
703 #define SymX(vvv) /**/
704 #define SymX_redirect(vvv,xxx) /**/
708 RTS_POSIX_ONLY_SYMBOLS
709 RTS_MINGW_ONLY_SYMBOLS
710 RTS_CYGWIN_ONLY_SYMBOLS
711 RTS_DARWIN_ONLY_SYMBOLS
717 #ifdef LEADING_UNDERSCORE
718 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
720 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
723 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
725 #define SymX(vvv) Sym(vvv)
727 // SymX_redirect allows us to redirect references to one symbol to
728 // another symbol. See newCAF/newDynCAF for an example.
729 #define SymX_redirect(vvv,xxx) \
730 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
733 static RtsSymbolVal rtsSyms[] = {
737 RTS_POSIX_ONLY_SYMBOLS
738 RTS_MINGW_ONLY_SYMBOLS
739 RTS_CYGWIN_ONLY_SYMBOLS
741 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
742 // dyld stub code contains references to this,
743 // but it should never be called because we treat
744 // lazy pointers as nonlazy.
745 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
747 { 0, 0 } /* sentinel */
750 /* -----------------------------------------------------------------------------
751 * Insert symbols into hash tables, checking for duplicates.
753 static void ghciInsertStrHashTable ( char* obj_name,
759 if (lookupHashTable(table, (StgWord)key) == NULL)
761 insertStrHashTable(table, (StgWord)key, data);
766 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
768 "whilst processing object file\n"
770 "This could be caused by:\n"
771 " * Loading two different object files which export the same symbol\n"
772 " * Specifying the same object file twice on the GHCi command line\n"
773 " * An incorrect `package.conf' entry, causing some object to be\n"
775 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
784 /* -----------------------------------------------------------------------------
785 * initialize the object linker
789 static int linker_init_done = 0 ;
791 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
792 static void *dl_prog_handle;
795 /* dlopen(NULL,..) doesn't work so we grab libc explicitly */
796 #if defined(openbsd_HOST_OS)
797 static void *dl_libc_handle;
805 /* Make initLinker idempotent, so we can call it
806 before evey relevant operation; that means we
807 don't need to initialise the linker separately */
808 if (linker_init_done == 1) { return; } else {
809 linker_init_done = 1;
812 symhash = allocStrHashTable();
814 /* populate the symbol table with stuff from the RTS */
815 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
816 ghciInsertStrHashTable("(GHCi built-in symbols)",
817 symhash, sym->lbl, sym->addr);
819 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
820 machoInitSymbolsWithoutUnderscore();
823 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
824 # if defined(RTLD_DEFAULT)
825 dl_prog_handle = RTLD_DEFAULT;
827 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
828 # if defined(openbsd_HOST_OS)
829 dl_libc_handle = dlopen("libc.so", RTLD_LAZY);
831 # endif /* RTLD_DEFAULT */
835 /* -----------------------------------------------------------------------------
836 * Loading DLL or .so dynamic libraries
837 * -----------------------------------------------------------------------------
839 * Add a DLL from which symbols may be found. In the ELF case, just
840 * do RTLD_GLOBAL-style add, so no further messing around needs to
841 * happen in order that symbols in the loaded .so are findable --
842 * lookupSymbol() will subsequently see them by dlsym on the program's
843 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
845 * In the PEi386 case, open the DLLs and put handles to them in a
846 * linked list. When looking for a symbol, try all handles in the
847 * list. This means that we need to load even DLLs that are guaranteed
848 * to be in the ghc.exe image already, just so we can get a handle
849 * to give to loadSymbol, so that we can find the symbols. For such
850 * libraries, the LoadLibrary call should be a no-op except for returning
855 #if defined(OBJFORMAT_PEi386)
856 /* A record for storing handles into DLLs. */
861 struct _OpenedDLL* next;
866 /* A list thereof. */
867 static OpenedDLL* opened_dlls = NULL;
871 addDLL( char *dll_name )
873 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
874 /* ------------------- ELF DLL loader ------------------- */
880 hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
883 /* dlopen failed; return a ptr to the error msg. */
885 if (errmsg == NULL) errmsg = "addDLL: unknown error";
892 # elif defined(OBJFORMAT_PEi386)
893 /* ------------------- Win32 DLL loader ------------------- */
901 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
903 /* See if we've already got it, and ignore if so. */
904 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
905 if (0 == strcmp(o_dll->name, dll_name))
909 /* The file name has no suffix (yet) so that we can try
910 both foo.dll and foo.drv
912 The documentation for LoadLibrary says:
913 If no file name extension is specified in the lpFileName
914 parameter, the default library extension .dll is
915 appended. However, the file name string can include a trailing
916 point character (.) to indicate that the module name has no
919 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
920 sprintf(buf, "%s.DLL", dll_name);
921 instance = LoadLibrary(buf);
922 if (instance == NULL) {
923 sprintf(buf, "%s.DRV", dll_name); // KAA: allow loading of drivers (like winspool.drv)
924 instance = LoadLibrary(buf);
925 if (instance == NULL) {
928 /* LoadLibrary failed; return a ptr to the error msg. */
929 return "addDLL: unknown error";
934 /* Add this DLL to the list of DLLs in which to search for symbols. */
935 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
936 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
937 strcpy(o_dll->name, dll_name);
938 o_dll->instance = instance;
939 o_dll->next = opened_dlls;
944 barf("addDLL: not implemented on this platform");
948 /* -----------------------------------------------------------------------------
949 * lookup a symbol in the hash table
952 lookupSymbol( char *lbl )
956 ASSERT(symhash != NULL);
957 val = lookupStrHashTable(symhash, lbl);
960 # if defined(OBJFORMAT_ELF)
961 # if defined(openbsd_HOST_OS)
962 val = dlsym(dl_prog_handle, lbl);
963 return (val != NULL) ? val : dlsym(dl_libc_handle,lbl);
964 # elif defined(x86_64_HOST_ARCH)
965 val = dlsym(dl_prog_handle, lbl);
966 if (val >= (void *)0x80000000) {
968 new_val = x86_64_high_symbol(lbl, val);
969 IF_DEBUG(linker,debugBelch("lookupSymbol: relocating out of range symbol: %s = %p, now %p\n", lbl, val, new_val));
974 # else /* not openbsd */
975 return dlsym(dl_prog_handle, lbl);
977 # elif defined(OBJFORMAT_MACHO)
978 if(NSIsSymbolNameDefined(lbl)) {
979 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
980 return NSAddressOfSymbol(symbol);
984 # elif defined(OBJFORMAT_PEi386)
987 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
988 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
990 /* HACK: if the name has an initial underscore, try stripping
991 it off & look that up first. I've yet to verify whether there's
992 a Rule that governs whether an initial '_' *should always* be
993 stripped off when mapping from import lib name to the DLL name.
995 sym = GetProcAddress(o_dll->instance, (lbl+1));
997 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
1001 sym = GetProcAddress(o_dll->instance, lbl);
1003 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
1018 __attribute((unused))
1020 lookupLocalSymbol( ObjectCode* oc, char *lbl )
1024 val = lookupStrHashTable(oc->lochash, lbl);
1034 /* -----------------------------------------------------------------------------
1035 * Debugging aid: look in GHCi's object symbol tables for symbols
1036 * within DELTA bytes of the specified address, and show their names.
1039 void ghci_enquire ( char* addr );
1041 void ghci_enquire ( char* addr )
1046 const int DELTA = 64;
1051 for (oc = objects; oc; oc = oc->next) {
1052 for (i = 0; i < oc->n_symbols; i++) {
1053 sym = oc->symbols[i];
1054 if (sym == NULL) continue;
1055 // debugBelch("enquire %p %p\n", sym, oc->lochash);
1057 if (oc->lochash != NULL) {
1058 a = lookupStrHashTable(oc->lochash, sym);
1061 a = lookupStrHashTable(symhash, sym);
1064 // debugBelch("ghci_enquire: can't find %s\n", sym);
1066 else if (addr-DELTA <= a && a <= addr+DELTA) {
1067 debugBelch("%p + %3d == `%s'\n", addr, a - addr, sym);
1074 #ifdef ia64_HOST_ARCH
1075 static unsigned int PLTSize(void);
1078 /* -----------------------------------------------------------------------------
1079 * Load an obj (populate the global symbol table, but don't resolve yet)
1081 * Returns: 1 if ok, 0 on error.
1084 loadObj( char *path )
1091 void *map_addr = NULL;
1098 /* debugBelch("loadObj %s\n", path ); */
1100 /* Check that we haven't already loaded this object.
1101 Ignore requests to load multiple times */
1105 for (o = objects; o; o = o->next) {
1106 if (0 == strcmp(o->fileName, path)) {
1108 break; /* don't need to search further */
1112 IF_DEBUG(linker, debugBelch(
1113 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1114 "same object file twice:\n"
1116 "GHCi will ignore this, but be warned.\n"
1118 return 1; /* success */
1122 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1124 # if defined(OBJFORMAT_ELF)
1125 oc->formatName = "ELF";
1126 # elif defined(OBJFORMAT_PEi386)
1127 oc->formatName = "PEi386";
1128 # elif defined(OBJFORMAT_MACHO)
1129 oc->formatName = "Mach-O";
1132 barf("loadObj: not implemented on this platform");
1135 r = stat(path, &st);
1136 if (r == -1) { return 0; }
1138 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1139 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1140 strcpy(oc->fileName, path);
1142 oc->fileSize = st.st_size;
1144 oc->sections = NULL;
1145 oc->lochash = allocStrHashTable();
1146 oc->proddables = NULL;
1148 /* chain it onto the list of objects */
1153 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1155 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1157 #if defined(openbsd_HOST_OS)
1158 fd = open(path, O_RDONLY, S_IRUSR);
1160 fd = open(path, O_RDONLY);
1163 barf("loadObj: can't open `%s'", path);
1165 pagesize = getpagesize();
1167 #ifdef ia64_HOST_ARCH
1168 /* The PLT needs to be right before the object */
1169 n = ROUND_UP(PLTSize(), pagesize);
1170 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1171 if (oc->plt == MAP_FAILED)
1172 barf("loadObj: can't allocate PLT");
1175 map_addr = oc->plt + n;
1178 n = ROUND_UP(oc->fileSize, pagesize);
1180 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
1181 * small memory model on this architecture (see gcc docs,
1184 #ifdef x86_64_HOST_ARCH
1185 #define EXTRA_MAP_FLAGS MAP_32BIT
1187 #define EXTRA_MAP_FLAGS 0
1190 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1191 MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0);
1192 if (oc->image == MAP_FAILED)
1193 barf("loadObj: can't map `%s'", path);
1197 #else /* !USE_MMAP */
1199 /* load the image into memory */
1200 f = fopen(path, "rb");
1202 barf("loadObj: can't read `%s'", path);
1204 #ifdef darwin_HOST_OS
1205 // In a Mach-O .o file, all sections can and will be misaligned
1206 // if the total size of the headers is not a multiple of the
1207 // desired alignment. This is fine for .o files that only serve
1208 // as input for the static linker, but it's not fine for us,
1209 // as SSE (used by gcc for floating point) and Altivec require
1210 // 16-byte alignment.
1211 // We calculate the correct alignment from the header before
1212 // reading the file, and then we misalign oc->image on purpose so
1213 // that the actual sections end up aligned again.
1214 misalignment = machoGetMisalignment(f);
1219 oc->image = stgMallocBytes(oc->fileSize + misalignment, "loadObj(image)");
1220 oc->image += misalignment;
1222 n = fread ( oc->image, 1, oc->fileSize, f );
1223 if (n != oc->fileSize)
1224 barf("loadObj: error whilst reading `%s'", path);
1228 #endif /* USE_MMAP */
1230 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1231 r = ocAllocateJumpIslands_MachO ( oc );
1232 if (!r) { return r; }
1233 # elif defined(OBJFORMAT_ELF) && defined(powerpc_HOST_ARCH)
1234 r = ocAllocateJumpIslands_ELF ( oc );
1235 if (!r) { return r; }
1238 /* verify the in-memory image */
1239 # if defined(OBJFORMAT_ELF)
1240 r = ocVerifyImage_ELF ( oc );
1241 # elif defined(OBJFORMAT_PEi386)
1242 r = ocVerifyImage_PEi386 ( oc );
1243 # elif defined(OBJFORMAT_MACHO)
1244 r = ocVerifyImage_MachO ( oc );
1246 barf("loadObj: no verify method");
1248 if (!r) { return r; }
1250 /* build the symbol list for this image */
1251 # if defined(OBJFORMAT_ELF)
1252 r = ocGetNames_ELF ( oc );
1253 # elif defined(OBJFORMAT_PEi386)
1254 r = ocGetNames_PEi386 ( oc );
1255 # elif defined(OBJFORMAT_MACHO)
1256 r = ocGetNames_MachO ( oc );
1258 barf("loadObj: no getNames method");
1260 if (!r) { return r; }
1262 /* loaded, but not resolved yet */
1263 oc->status = OBJECT_LOADED;
1268 /* -----------------------------------------------------------------------------
1269 * resolve all the currently unlinked objects in memory
1271 * Returns: 1 if ok, 0 on error.
1281 for (oc = objects; oc; oc = oc->next) {
1282 if (oc->status != OBJECT_RESOLVED) {
1283 # if defined(OBJFORMAT_ELF)
1284 r = ocResolve_ELF ( oc );
1285 # elif defined(OBJFORMAT_PEi386)
1286 r = ocResolve_PEi386 ( oc );
1287 # elif defined(OBJFORMAT_MACHO)
1288 r = ocResolve_MachO ( oc );
1290 barf("resolveObjs: not implemented on this platform");
1292 if (!r) { return r; }
1293 oc->status = OBJECT_RESOLVED;
1299 /* -----------------------------------------------------------------------------
1300 * delete an object from the pool
1303 unloadObj( char *path )
1305 ObjectCode *oc, *prev;
1307 ASSERT(symhash != NULL);
1308 ASSERT(objects != NULL);
1313 for (oc = objects; oc; prev = oc, oc = oc->next) {
1314 if (!strcmp(oc->fileName,path)) {
1316 /* Remove all the mappings for the symbols within this
1321 for (i = 0; i < oc->n_symbols; i++) {
1322 if (oc->symbols[i] != NULL) {
1323 removeStrHashTable(symhash, oc->symbols[i], NULL);
1331 prev->next = oc->next;
1334 /* We're going to leave this in place, in case there are
1335 any pointers from the heap into it: */
1336 /* stgFree(oc->image); */
1337 stgFree(oc->fileName);
1338 stgFree(oc->symbols);
1339 stgFree(oc->sections);
1340 /* The local hash table should have been freed at the end
1341 of the ocResolve_ call on it. */
1342 ASSERT(oc->lochash == NULL);
1348 errorBelch("unloadObj: can't find `%s' to unload", path);
1352 /* -----------------------------------------------------------------------------
1353 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1354 * which may be prodded during relocation, and abort if we try and write
1355 * outside any of these.
1357 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1360 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1361 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1365 pb->next = oc->proddables;
1366 oc->proddables = pb;
1369 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1372 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1373 char* s = (char*)(pb->start);
1374 char* e = s + pb->size - 1;
1375 char* a = (char*)addr;
1376 /* Assumes that the biggest fixup involves a 4-byte write. This
1377 probably needs to be changed to 8 (ie, +7) on 64-bit
1379 if (a >= s && (a+3) <= e) return;
1381 barf("checkProddableBlock: invalid fixup in runtime linker");
1384 /* -----------------------------------------------------------------------------
1385 * Section management.
1387 static void addSection ( ObjectCode* oc, SectionKind kind,
1388 void* start, void* end )
1390 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1394 s->next = oc->sections;
1397 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1398 start, ((char*)end)-1, end - start + 1, kind );
1403 /* --------------------------------------------------------------------------
1404 * PowerPC specifics (jump islands)
1405 * ------------------------------------------------------------------------*/
1407 #if defined(powerpc_HOST_ARCH)
1410 ocAllocateJumpIslands
1412 Allocate additional space at the end of the object file image to make room
1415 PowerPC relative branch instructions have a 24 bit displacement field.
1416 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1417 If a particular imported symbol is outside this range, we have to redirect
1418 the jump to a short piece of new code that just loads the 32bit absolute
1419 address and jumps there.
1420 This function just allocates space for one 16 byte ppcJumpIsland for every
1421 undefined symbol in the object file. The code for the islands is filled in by
1422 makeJumpIsland below.
1425 static int ocAllocateJumpIslands( ObjectCode* oc, int count, int first )
1434 // round up to the nearest 4
1435 aligned = (oc->fileSize + 3) & ~3;
1438 #ifndef linux_HOST_OS /* mremap is a linux extension */
1439 #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined
1442 pagesize = getpagesize();
1443 n = ROUND_UP( oc->fileSize, pagesize );
1444 m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize );
1446 /* The effect of this mremap() call is only the ensure that we have
1447 * a sufficient number of virtually contiguous pages. As returned from
1448 * mremap, the pages past the end of the file are not backed. We give
1449 * them a backing by using MAP_FIXED to map in anonymous pages.
1451 if( (oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE )) == MAP_FAILED )
1453 errorBelch( "Unable to mremap for Jump Islands\n" );
1457 if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC,
1458 MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED )
1460 errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" );
1465 oc->image = stgReallocBytes( oc->image,
1466 aligned + sizeof (ppcJumpIsland) * count,
1467 "ocAllocateJumpIslands" );
1468 #endif /* USE_MMAP */
1470 oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned);
1471 memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count );
1474 oc->jump_islands = NULL;
1476 oc->island_start_symbol = first;
1477 oc->n_islands = count;
1482 static unsigned long makeJumpIsland( ObjectCode* oc,
1483 unsigned long symbolNumber,
1484 unsigned long target )
1486 ppcJumpIsland *island;
1488 if( symbolNumber < oc->island_start_symbol ||
1489 symbolNumber - oc->island_start_symbol > oc->n_islands)
1492 island = &oc->jump_islands[symbolNumber - oc->island_start_symbol];
1494 // lis r12, hi16(target)
1495 island->lis_r12 = 0x3d80;
1496 island->hi_addr = target >> 16;
1498 // ori r12, r12, lo16(target)
1499 island->ori_r12_r12 = 0x618c;
1500 island->lo_addr = target & 0xffff;
1503 island->mtctr_r12 = 0x7d8903a6;
1506 island->bctr = 0x4e800420;
1508 return (unsigned long) island;
1512 ocFlushInstructionCache
1514 Flush the data & instruction caches.
1515 Because the PPC has split data/instruction caches, we have to
1516 do that whenever we modify code at runtime.
1519 static void ocFlushInstructionCache( ObjectCode *oc )
1521 int n = (oc->fileSize + sizeof( ppcJumpIsland ) * oc->n_islands + 3) / 4;
1522 unsigned long *p = (unsigned long *) oc->image;
1526 __asm__ volatile ( "dcbf 0,%0\n\t"
1534 __asm__ volatile ( "sync\n\t"
1540 /* --------------------------------------------------------------------------
1541 * PEi386 specifics (Win32 targets)
1542 * ------------------------------------------------------------------------*/
1544 /* The information for this linker comes from
1545 Microsoft Portable Executable
1546 and Common Object File Format Specification
1547 revision 5.1 January 1998
1548 which SimonM says comes from the MS Developer Network CDs.
1550 It can be found there (on older CDs), but can also be found
1553 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1555 (this is Rev 6.0 from February 1999).
1557 Things move, so if that fails, try searching for it via
1559 http://www.google.com/search?q=PE+COFF+specification
1561 The ultimate reference for the PE format is the Winnt.h
1562 header file that comes with the Platform SDKs; as always,
1563 implementations will drift wrt their documentation.
1565 A good background article on the PE format is Matt Pietrek's
1566 March 1994 article in Microsoft System Journal (MSJ)
1567 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1568 Win32 Portable Executable File Format." The info in there
1569 has recently been updated in a two part article in
1570 MSDN magazine, issues Feb and March 2002,
1571 "Inside Windows: An In-Depth Look into the Win32 Portable
1572 Executable File Format"
1574 John Levine's book "Linkers and Loaders" contains useful
1579 #if defined(OBJFORMAT_PEi386)
1583 typedef unsigned char UChar;
1584 typedef unsigned short UInt16;
1585 typedef unsigned int UInt32;
1592 UInt16 NumberOfSections;
1593 UInt32 TimeDateStamp;
1594 UInt32 PointerToSymbolTable;
1595 UInt32 NumberOfSymbols;
1596 UInt16 SizeOfOptionalHeader;
1597 UInt16 Characteristics;
1601 #define sizeof_COFF_header 20
1608 UInt32 VirtualAddress;
1609 UInt32 SizeOfRawData;
1610 UInt32 PointerToRawData;
1611 UInt32 PointerToRelocations;
1612 UInt32 PointerToLinenumbers;
1613 UInt16 NumberOfRelocations;
1614 UInt16 NumberOfLineNumbers;
1615 UInt32 Characteristics;
1619 #define sizeof_COFF_section 40
1626 UInt16 SectionNumber;
1629 UChar NumberOfAuxSymbols;
1633 #define sizeof_COFF_symbol 18
1638 UInt32 VirtualAddress;
1639 UInt32 SymbolTableIndex;
1644 #define sizeof_COFF_reloc 10
1647 /* From PE spec doc, section 3.3.2 */
1648 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1649 windows.h -- for the same purpose, but I want to know what I'm
1651 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1652 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1653 #define MYIMAGE_FILE_DLL 0x2000
1654 #define MYIMAGE_FILE_SYSTEM 0x1000
1655 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1656 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1657 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1659 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1660 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1661 #define MYIMAGE_SYM_CLASS_STATIC 3
1662 #define MYIMAGE_SYM_UNDEFINED 0
1664 /* From PE spec doc, section 4.1 */
1665 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1666 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1667 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1669 /* From PE spec doc, section 5.2.1 */
1670 #define MYIMAGE_REL_I386_DIR32 0x0006
1671 #define MYIMAGE_REL_I386_REL32 0x0014
1674 /* We use myindex to calculate array addresses, rather than
1675 simply doing the normal subscript thing. That's because
1676 some of the above structs have sizes which are not
1677 a whole number of words. GCC rounds their sizes up to a
1678 whole number of words, which means that the address calcs
1679 arising from using normal C indexing or pointer arithmetic
1680 are just plain wrong. Sigh.
1683 myindex ( int scale, void* base, int index )
1686 ((UChar*)base) + scale * index;
1691 printName ( UChar* name, UChar* strtab )
1693 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1694 UInt32 strtab_offset = * (UInt32*)(name+4);
1695 debugBelch("%s", strtab + strtab_offset );
1698 for (i = 0; i < 8; i++) {
1699 if (name[i] == 0) break;
1700 debugBelch("%c", name[i] );
1707 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1709 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1710 UInt32 strtab_offset = * (UInt32*)(name+4);
1711 strncpy ( dst, strtab+strtab_offset, dstSize );
1717 if (name[i] == 0) break;
1727 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1730 /* If the string is longer than 8 bytes, look in the
1731 string table for it -- this will be correctly zero terminated.
1733 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1734 UInt32 strtab_offset = * (UInt32*)(name+4);
1735 return ((UChar*)strtab) + strtab_offset;
1737 /* Otherwise, if shorter than 8 bytes, return the original,
1738 which by defn is correctly terminated.
1740 if (name[7]==0) return name;
1741 /* The annoying case: 8 bytes. Copy into a temporary
1742 (which is never freed ...)
1744 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1746 strncpy(newstr,name,8);
1752 /* Just compares the short names (first 8 chars) */
1753 static COFF_section *
1754 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1758 = (COFF_header*)(oc->image);
1759 COFF_section* sectab
1761 ((UChar*)(oc->image))
1762 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1764 for (i = 0; i < hdr->NumberOfSections; i++) {
1767 COFF_section* section_i
1769 myindex ( sizeof_COFF_section, sectab, i );
1770 n1 = (UChar*) &(section_i->Name);
1772 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1773 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1774 n1[6]==n2[6] && n1[7]==n2[7])
1783 zapTrailingAtSign ( UChar* sym )
1785 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1787 if (sym[0] == 0) return;
1789 while (sym[i] != 0) i++;
1792 while (j > 0 && my_isdigit(sym[j])) j--;
1793 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1799 ocVerifyImage_PEi386 ( ObjectCode* oc )
1804 COFF_section* sectab;
1805 COFF_symbol* symtab;
1807 /* debugBelch("\nLOADING %s\n", oc->fileName); */
1808 hdr = (COFF_header*)(oc->image);
1809 sectab = (COFF_section*) (
1810 ((UChar*)(oc->image))
1811 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1813 symtab = (COFF_symbol*) (
1814 ((UChar*)(oc->image))
1815 + hdr->PointerToSymbolTable
1817 strtab = ((UChar*)symtab)
1818 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1820 if (hdr->Machine != 0x14c) {
1821 errorBelch("%s: Not x86 PEi386", oc->fileName);
1824 if (hdr->SizeOfOptionalHeader != 0) {
1825 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
1828 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
1829 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
1830 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
1831 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
1832 errorBelch("%s: Not a PEi386 object file", oc->fileName);
1835 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
1836 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
1837 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
1839 (int)(hdr->Characteristics));
1842 /* If the string table size is way crazy, this might indicate that
1843 there are more than 64k relocations, despite claims to the
1844 contrary. Hence this test. */
1845 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
1847 if ( (*(UInt32*)strtab) > 600000 ) {
1848 /* Note that 600k has no special significance other than being
1849 big enough to handle the almost-2MB-sized lumps that
1850 constitute HSwin32*.o. */
1851 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
1856 /* No further verification after this point; only debug printing. */
1858 IF_DEBUG(linker, i=1);
1859 if (i == 0) return 1;
1861 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
1862 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
1863 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
1866 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
1867 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
1868 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
1869 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
1870 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
1871 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
1872 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
1874 /* Print the section table. */
1876 for (i = 0; i < hdr->NumberOfSections; i++) {
1878 COFF_section* sectab_i
1880 myindex ( sizeof_COFF_section, sectab, i );
1887 printName ( sectab_i->Name, strtab );
1897 sectab_i->VirtualSize,
1898 sectab_i->VirtualAddress,
1899 sectab_i->SizeOfRawData,
1900 sectab_i->PointerToRawData,
1901 sectab_i->NumberOfRelocations,
1902 sectab_i->PointerToRelocations,
1903 sectab_i->PointerToRawData
1905 reltab = (COFF_reloc*) (
1906 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1909 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
1910 /* If the relocation field (a short) has overflowed, the
1911 * real count can be found in the first reloc entry.
1913 * See Section 4.1 (last para) of the PE spec (rev6.0).
1915 COFF_reloc* rel = (COFF_reloc*)
1916 myindex ( sizeof_COFF_reloc, reltab, 0 );
1917 noRelocs = rel->VirtualAddress;
1920 noRelocs = sectab_i->NumberOfRelocations;
1924 for (; j < noRelocs; j++) {
1926 COFF_reloc* rel = (COFF_reloc*)
1927 myindex ( sizeof_COFF_reloc, reltab, j );
1929 " type 0x%-4x vaddr 0x%-8x name `",
1931 rel->VirtualAddress );
1932 sym = (COFF_symbol*)
1933 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
1934 /* Hmm..mysterious looking offset - what's it for? SOF */
1935 printName ( sym->Name, strtab -10 );
1942 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
1943 debugBelch("---START of string table---\n");
1944 for (i = 4; i < *(Int32*)strtab; i++) {
1946 debugBelch("\n"); else
1947 debugBelch("%c", strtab[i] );
1949 debugBelch("--- END of string table---\n");
1954 COFF_symbol* symtab_i;
1955 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1956 symtab_i = (COFF_symbol*)
1957 myindex ( sizeof_COFF_symbol, symtab, i );
1963 printName ( symtab_i->Name, strtab );
1972 (Int32)(symtab_i->SectionNumber),
1973 (UInt32)symtab_i->Type,
1974 (UInt32)symtab_i->StorageClass,
1975 (UInt32)symtab_i->NumberOfAuxSymbols
1977 i += symtab_i->NumberOfAuxSymbols;
1987 ocGetNames_PEi386 ( ObjectCode* oc )
1990 COFF_section* sectab;
1991 COFF_symbol* symtab;
1998 hdr = (COFF_header*)(oc->image);
1999 sectab = (COFF_section*) (
2000 ((UChar*)(oc->image))
2001 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2003 symtab = (COFF_symbol*) (
2004 ((UChar*)(oc->image))
2005 + hdr->PointerToSymbolTable
2007 strtab = ((UChar*)(oc->image))
2008 + hdr->PointerToSymbolTable
2009 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2011 /* Allocate space for any (local, anonymous) .bss sections. */
2013 for (i = 0; i < hdr->NumberOfSections; i++) {
2015 COFF_section* sectab_i
2017 myindex ( sizeof_COFF_section, sectab, i );
2018 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2019 if (sectab_i->VirtualSize == 0) continue;
2020 /* This is a non-empty .bss section. Allocate zeroed space for
2021 it, and set its PointerToRawData field such that oc->image +
2022 PointerToRawData == addr_of_zeroed_space. */
2023 zspace = stgCallocBytes(1, sectab_i->VirtualSize,
2024 "ocGetNames_PEi386(anonymous bss)");
2025 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2026 addProddableBlock(oc, zspace, sectab_i->VirtualSize);
2027 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2030 /* Copy section information into the ObjectCode. */
2032 for (i = 0; i < hdr->NumberOfSections; i++) {
2038 = SECTIONKIND_OTHER;
2039 COFF_section* sectab_i
2041 myindex ( sizeof_COFF_section, sectab, i );
2042 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2045 /* I'm sure this is the Right Way to do it. However, the
2046 alternative of testing the sectab_i->Name field seems to
2047 work ok with Cygwin.
2049 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2050 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2051 kind = SECTIONKIND_CODE_OR_RODATA;
2054 if (0==strcmp(".text",sectab_i->Name) ||
2055 0==strcmp(".rdata",sectab_i->Name)||
2056 0==strcmp(".rodata",sectab_i->Name))
2057 kind = SECTIONKIND_CODE_OR_RODATA;
2058 if (0==strcmp(".data",sectab_i->Name) ||
2059 0==strcmp(".bss",sectab_i->Name))
2060 kind = SECTIONKIND_RWDATA;
2062 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2063 sz = sectab_i->SizeOfRawData;
2064 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2066 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2067 end = start + sz - 1;
2069 if (kind == SECTIONKIND_OTHER
2070 /* Ignore sections called which contain stabs debugging
2072 && 0 != strcmp(".stab", sectab_i->Name)
2073 && 0 != strcmp(".stabstr", sectab_i->Name)
2074 /* ignore constructor section for now */
2075 && 0 != strcmp(".ctors", sectab_i->Name)
2077 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2081 if (kind != SECTIONKIND_OTHER && end >= start) {
2082 addSection(oc, kind, start, end);
2083 addProddableBlock(oc, start, end - start + 1);
2087 /* Copy exported symbols into the ObjectCode. */
2089 oc->n_symbols = hdr->NumberOfSymbols;
2090 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2091 "ocGetNames_PEi386(oc->symbols)");
2092 /* Call me paranoid; I don't care. */
2093 for (i = 0; i < oc->n_symbols; i++)
2094 oc->symbols[i] = NULL;
2098 COFF_symbol* symtab_i;
2099 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2100 symtab_i = (COFF_symbol*)
2101 myindex ( sizeof_COFF_symbol, symtab, i );
2105 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2106 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2107 /* This symbol is global and defined, viz, exported */
2108 /* for MYIMAGE_SYMCLASS_EXTERNAL
2109 && !MYIMAGE_SYM_UNDEFINED,
2110 the address of the symbol is:
2111 address of relevant section + offset in section
2113 COFF_section* sectabent
2114 = (COFF_section*) myindex ( sizeof_COFF_section,
2116 symtab_i->SectionNumber-1 );
2117 addr = ((UChar*)(oc->image))
2118 + (sectabent->PointerToRawData
2122 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2123 && symtab_i->Value > 0) {
2124 /* This symbol isn't in any section at all, ie, global bss.
2125 Allocate zeroed space for it. */
2126 addr = stgCallocBytes(1, symtab_i->Value,
2127 "ocGetNames_PEi386(non-anonymous bss)");
2128 addSection(oc, SECTIONKIND_RWDATA, addr,
2129 ((UChar*)addr) + symtab_i->Value - 1);
2130 addProddableBlock(oc, addr, symtab_i->Value);
2131 /* debugBelch("BSS section at 0x%x\n", addr); */
2134 if (addr != NULL ) {
2135 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2136 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2137 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2138 ASSERT(i >= 0 && i < oc->n_symbols);
2139 /* cstring_from_COFF_symbol_name always succeeds. */
2140 oc->symbols[i] = sname;
2141 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2145 "IGNORING symbol %d\n"
2149 printName ( symtab_i->Name, strtab );
2158 (Int32)(symtab_i->SectionNumber),
2159 (UInt32)symtab_i->Type,
2160 (UInt32)symtab_i->StorageClass,
2161 (UInt32)symtab_i->NumberOfAuxSymbols
2166 i += symtab_i->NumberOfAuxSymbols;
2175 ocResolve_PEi386 ( ObjectCode* oc )
2178 COFF_section* sectab;
2179 COFF_symbol* symtab;
2189 /* ToDo: should be variable-sized? But is at least safe in the
2190 sense of buffer-overrun-proof. */
2192 /* debugBelch("resolving for %s\n", oc->fileName); */
2194 hdr = (COFF_header*)(oc->image);
2195 sectab = (COFF_section*) (
2196 ((UChar*)(oc->image))
2197 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2199 symtab = (COFF_symbol*) (
2200 ((UChar*)(oc->image))
2201 + hdr->PointerToSymbolTable
2203 strtab = ((UChar*)(oc->image))
2204 + hdr->PointerToSymbolTable
2205 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2207 for (i = 0; i < hdr->NumberOfSections; i++) {
2208 COFF_section* sectab_i
2210 myindex ( sizeof_COFF_section, sectab, i );
2213 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2216 /* Ignore sections called which contain stabs debugging
2218 if (0 == strcmp(".stab", sectab_i->Name)
2219 || 0 == strcmp(".stabstr", sectab_i->Name)
2220 || 0 == strcmp(".ctors", sectab_i->Name))
2223 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2224 /* If the relocation field (a short) has overflowed, the
2225 * real count can be found in the first reloc entry.
2227 * See Section 4.1 (last para) of the PE spec (rev6.0).
2229 * Nov2003 update: the GNU linker still doesn't correctly
2230 * handle the generation of relocatable object files with
2231 * overflown relocations. Hence the output to warn of potential
2234 COFF_reloc* rel = (COFF_reloc*)
2235 myindex ( sizeof_COFF_reloc, reltab, 0 );
2236 noRelocs = rel->VirtualAddress;
2237 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2241 noRelocs = sectab_i->NumberOfRelocations;
2246 for (; j < noRelocs; j++) {
2248 COFF_reloc* reltab_j
2250 myindex ( sizeof_COFF_reloc, reltab, j );
2252 /* the location to patch */
2254 ((UChar*)(oc->image))
2255 + (sectab_i->PointerToRawData
2256 + reltab_j->VirtualAddress
2257 - sectab_i->VirtualAddress )
2259 /* the existing contents of pP */
2261 /* the symbol to connect to */
2262 sym = (COFF_symbol*)
2263 myindex ( sizeof_COFF_symbol,
2264 symtab, reltab_j->SymbolTableIndex );
2267 "reloc sec %2d num %3d: type 0x%-4x "
2268 "vaddr 0x%-8x name `",
2270 (UInt32)reltab_j->Type,
2271 reltab_j->VirtualAddress );
2272 printName ( sym->Name, strtab );
2273 debugBelch("'\n" ));
2275 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2276 COFF_section* section_sym
2277 = findPEi386SectionCalled ( oc, sym->Name );
2279 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2282 S = ((UInt32)(oc->image))
2283 + (section_sym->PointerToRawData
2286 copyName ( sym->Name, strtab, symbol, 1000-1 );
2287 (void*)S = lookupLocalSymbol( oc, symbol );
2288 if ((void*)S != NULL) goto foundit;
2289 (void*)S = lookupSymbol( symbol );
2290 if ((void*)S != NULL) goto foundit;
2291 zapTrailingAtSign ( symbol );
2292 (void*)S = lookupLocalSymbol( oc, symbol );
2293 if ((void*)S != NULL) goto foundit;
2294 (void*)S = lookupSymbol( symbol );
2295 if ((void*)S != NULL) goto foundit;
2296 /* Newline first because the interactive linker has printed "linking..." */
2297 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2301 checkProddableBlock(oc, pP);
2302 switch (reltab_j->Type) {
2303 case MYIMAGE_REL_I386_DIR32:
2306 case MYIMAGE_REL_I386_REL32:
2307 /* Tricky. We have to insert a displacement at
2308 pP which, when added to the PC for the _next_
2309 insn, gives the address of the target (S).
2310 Problem is to know the address of the next insn
2311 when we only know pP. We assume that this
2312 literal field is always the last in the insn,
2313 so that the address of the next insn is pP+4
2314 -- hence the constant 4.
2315 Also I don't know if A should be added, but so
2316 far it has always been zero.
2318 SOF 05/2005: 'A' (old contents of *pP) have been observed
2319 to contain values other than zero (the 'wx' object file
2320 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2321 So, add displacement to old value instead of asserting
2322 A to be zero. Fixes wxhaskell-related crashes, and no other
2323 ill effects have been observed.
2325 Update: the reason why we're seeing these more elaborate
2326 relocations is due to a switch in how the NCG compiles SRTs
2327 and offsets to them from info tables. SRTs live in .(ro)data,
2328 while info tables live in .text, causing GAS to emit REL32/DISP32
2329 relocations with non-zero values. Adding the displacement is
2330 the right thing to do.
2332 *pP = S - ((UInt32)pP) - 4 + A;
2335 debugBelch("%s: unhandled PEi386 relocation type %d",
2336 oc->fileName, reltab_j->Type);
2343 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2347 #endif /* defined(OBJFORMAT_PEi386) */
2350 /* --------------------------------------------------------------------------
2352 * ------------------------------------------------------------------------*/
2354 #if defined(OBJFORMAT_ELF)
2359 #if defined(sparc_HOST_ARCH)
2360 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2361 #elif defined(i386_HOST_ARCH)
2362 # define ELF_TARGET_386 /* Used inside <elf.h> */
2363 #elif defined(x86_64_HOST_ARCH)
2364 # define ELF_TARGET_X64_64
2366 #elif defined (ia64_HOST_ARCH)
2367 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2369 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2370 # define ELF_NEED_GOT /* needs Global Offset Table */
2371 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2374 #if !defined(openbsd_HOST_OS)
2377 /* openbsd elf has things in different places, with diff names */
2378 #include <elf_abi.h>
2379 #include <machine/reloc.h>
2380 #define R_386_32 RELOC_32
2381 #define R_386_PC32 RELOC_PC32
2385 * Define a set of types which can be used for both ELF32 and ELF64
2389 #define ELFCLASS ELFCLASS64
2390 #define Elf_Addr Elf64_Addr
2391 #define Elf_Word Elf64_Word
2392 #define Elf_Sword Elf64_Sword
2393 #define Elf_Ehdr Elf64_Ehdr
2394 #define Elf_Phdr Elf64_Phdr
2395 #define Elf_Shdr Elf64_Shdr
2396 #define Elf_Sym Elf64_Sym
2397 #define Elf_Rel Elf64_Rel
2398 #define Elf_Rela Elf64_Rela
2399 #define ELF_ST_TYPE ELF64_ST_TYPE
2400 #define ELF_ST_BIND ELF64_ST_BIND
2401 #define ELF_R_TYPE ELF64_R_TYPE
2402 #define ELF_R_SYM ELF64_R_SYM
2404 #define ELFCLASS ELFCLASS32
2405 #define Elf_Addr Elf32_Addr
2406 #define Elf_Word Elf32_Word
2407 #define Elf_Sword Elf32_Sword
2408 #define Elf_Ehdr Elf32_Ehdr
2409 #define Elf_Phdr Elf32_Phdr
2410 #define Elf_Shdr Elf32_Shdr
2411 #define Elf_Sym Elf32_Sym
2412 #define Elf_Rel Elf32_Rel
2413 #define Elf_Rela Elf32_Rela
2415 #define ELF_ST_TYPE ELF32_ST_TYPE
2418 #define ELF_ST_BIND ELF32_ST_BIND
2421 #define ELF_R_TYPE ELF32_R_TYPE
2424 #define ELF_R_SYM ELF32_R_SYM
2430 * Functions to allocate entries in dynamic sections. Currently we simply
2431 * preallocate a large number, and we don't check if a entry for the given
2432 * target already exists (a linear search is too slow). Ideally these
2433 * entries would be associated with symbols.
2436 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2437 #define GOT_SIZE 0x20000
2438 #define FUNCTION_TABLE_SIZE 0x10000
2439 #define PLT_SIZE 0x08000
2442 static Elf_Addr got[GOT_SIZE];
2443 static unsigned int gotIndex;
2444 static Elf_Addr gp_val = (Elf_Addr)got;
2447 allocateGOTEntry(Elf_Addr target)
2451 if (gotIndex >= GOT_SIZE)
2452 barf("Global offset table overflow");
2454 entry = &got[gotIndex++];
2456 return (Elf_Addr)entry;
2460 #ifdef ELF_FUNCTION_DESC
2466 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2467 static unsigned int functionTableIndex;
2470 allocateFunctionDesc(Elf_Addr target)
2472 FunctionDesc *entry;
2474 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2475 barf("Function table overflow");
2477 entry = &functionTable[functionTableIndex++];
2479 entry->gp = (Elf_Addr)gp_val;
2480 return (Elf_Addr)entry;
2484 copyFunctionDesc(Elf_Addr target)
2486 FunctionDesc *olddesc = (FunctionDesc *)target;
2487 FunctionDesc *newdesc;
2489 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2490 newdesc->gp = olddesc->gp;
2491 return (Elf_Addr)newdesc;
2496 #ifdef ia64_HOST_ARCH
2497 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2498 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2500 static unsigned char plt_code[] =
2502 /* taken from binutils bfd/elfxx-ia64.c */
2503 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2504 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2505 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2506 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2507 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2508 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2511 /* If we can't get to the function descriptor via gp, take a local copy of it */
2512 #define PLT_RELOC(code, target) { \
2513 Elf64_Sxword rel_value = target - gp_val; \
2514 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2515 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2517 ia64_reloc_gprel22((Elf_Addr)code, target); \
2522 unsigned char code[sizeof(plt_code)];
2526 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2528 PLTEntry *plt = (PLTEntry *)oc->plt;
2531 if (oc->pltIndex >= PLT_SIZE)
2532 barf("Procedure table overflow");
2534 entry = &plt[oc->pltIndex++];
2535 memcpy(entry->code, plt_code, sizeof(entry->code));
2536 PLT_RELOC(entry->code, target);
2537 return (Elf_Addr)entry;
2543 return (PLT_SIZE * sizeof(PLTEntry));
2548 #if x86_64_HOST_ARCH
2549 // On x86_64, 32-bit relocations are often used, which requires that
2550 // we can resolve a symbol to a 32-bit offset. However, shared
2551 // libraries are placed outside the 2Gb area, which leaves us with a
2552 // problem when we need to give a 32-bit offset to a symbol in a
2555 // For a function symbol, we can allocate a bounce sequence inside the
2556 // 2Gb area and resolve the symbol to this. The bounce sequence is
2557 // simply a long jump instruction to the real location of the symbol.
2559 // For data references, we're screwed.
2562 unsigned char jmp[8]; /* 6 byte instruction: jmpq *0x00000002(%rip) */
2566 #define X86_64_BB_SIZE 1024
2568 static x86_64_bounce *x86_64_bounce_buffer = NULL;
2569 static nat x86_64_bb_next_off;
2572 x86_64_high_symbol( char *lbl, void *addr )
2574 x86_64_bounce *bounce;
2576 if ( x86_64_bounce_buffer == NULL ||
2577 x86_64_bb_next_off >= X86_64_BB_SIZE ) {
2578 x86_64_bounce_buffer =
2579 mmap(NULL, X86_64_BB_SIZE * sizeof(x86_64_bounce),
2580 PROT_EXEC|PROT_READ|PROT_WRITE,
2581 MAP_PRIVATE|MAP_32BIT|MAP_ANONYMOUS, -1, 0);
2582 if (x86_64_bounce_buffer == MAP_FAILED) {
2583 barf("x86_64_high_symbol: mmap failed");
2585 x86_64_bb_next_off = 0;
2587 bounce = &x86_64_bounce_buffer[x86_64_bb_next_off];
2588 bounce->jmp[0] = 0xff;
2589 bounce->jmp[1] = 0x25;
2590 bounce->jmp[2] = 0x02;
2591 bounce->jmp[3] = 0x00;
2592 bounce->jmp[4] = 0x00;
2593 bounce->jmp[5] = 0x00;
2594 bounce->addr = addr;
2595 x86_64_bb_next_off++;
2597 IF_DEBUG(linker, debugBelch("x86_64: allocated bounce entry for %s->%p at %p\n",
2598 lbl, addr, bounce));
2600 insertStrHashTable(symhash, lbl, bounce);
2607 * Generic ELF functions
2611 findElfSection ( void* objImage, Elf_Word sh_type )
2613 char* ehdrC = (char*)objImage;
2614 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2615 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2616 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2620 for (i = 0; i < ehdr->e_shnum; i++) {
2621 if (shdr[i].sh_type == sh_type
2622 /* Ignore the section header's string table. */
2623 && i != ehdr->e_shstrndx
2624 /* Ignore string tables named .stabstr, as they contain
2626 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2628 ptr = ehdrC + shdr[i].sh_offset;
2635 #if defined(ia64_HOST_ARCH)
2637 findElfSegment ( void* objImage, Elf_Addr vaddr )
2639 char* ehdrC = (char*)objImage;
2640 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2641 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2642 Elf_Addr segaddr = 0;
2645 for (i = 0; i < ehdr->e_phnum; i++) {
2646 segaddr = phdr[i].p_vaddr;
2647 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2655 ocVerifyImage_ELF ( ObjectCode* oc )
2659 int i, j, nent, nstrtab, nsymtabs;
2663 char* ehdrC = (char*)(oc->image);
2664 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2666 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2667 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2668 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2669 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2670 errorBelch("%s: not an ELF object", oc->fileName);
2674 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2675 errorBelch("%s: unsupported ELF format", oc->fileName);
2679 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2680 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2682 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2683 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2685 errorBelch("%s: unknown endiannness", oc->fileName);
2689 if (ehdr->e_type != ET_REL) {
2690 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2693 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2695 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2696 switch (ehdr->e_machine) {
2697 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2698 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2700 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2702 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2704 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2706 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2707 errorBelch("%s: unknown architecture", oc->fileName);
2711 IF_DEBUG(linker,debugBelch(
2712 "\nSection header table: start %d, n_entries %d, ent_size %d\n",
2713 ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2715 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2717 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2719 if (ehdr->e_shstrndx == SHN_UNDEF) {
2720 errorBelch("%s: no section header string table", oc->fileName);
2723 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2725 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2728 for (i = 0; i < ehdr->e_shnum; i++) {
2729 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2730 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2731 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2732 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2733 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2734 ehdrC + shdr[i].sh_offset,
2735 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2737 if (shdr[i].sh_type == SHT_REL) {
2738 IF_DEBUG(linker,debugBelch("Rel " ));
2739 } else if (shdr[i].sh_type == SHT_RELA) {
2740 IF_DEBUG(linker,debugBelch("RelA " ));
2742 IF_DEBUG(linker,debugBelch(" "));
2745 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2749 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2752 for (i = 0; i < ehdr->e_shnum; i++) {
2753 if (shdr[i].sh_type == SHT_STRTAB
2754 /* Ignore the section header's string table. */
2755 && i != ehdr->e_shstrndx
2756 /* Ignore string tables named .stabstr, as they contain
2758 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2760 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2761 strtab = ehdrC + shdr[i].sh_offset;
2766 errorBelch("%s: no string tables, or too many", oc->fileName);
2771 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2772 for (i = 0; i < ehdr->e_shnum; i++) {
2773 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2774 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2776 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2777 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2778 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%d rem)\n",
2780 shdr[i].sh_size % sizeof(Elf_Sym)
2782 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2783 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2786 for (j = 0; j < nent; j++) {
2787 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2788 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
2789 (int)stab[j].st_shndx,
2790 (int)stab[j].st_size,
2791 (char*)stab[j].st_value ));
2793 IF_DEBUG(linker,debugBelch("type=" ));
2794 switch (ELF_ST_TYPE(stab[j].st_info)) {
2795 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
2796 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
2797 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
2798 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
2799 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
2800 default: IF_DEBUG(linker,debugBelch("? " )); break;
2802 IF_DEBUG(linker,debugBelch(" " ));
2804 IF_DEBUG(linker,debugBelch("bind=" ));
2805 switch (ELF_ST_BIND(stab[j].st_info)) {
2806 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
2807 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
2808 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
2809 default: IF_DEBUG(linker,debugBelch("? " )); break;
2811 IF_DEBUG(linker,debugBelch(" " ));
2813 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
2817 if (nsymtabs == 0) {
2818 errorBelch("%s: didn't find any symbol tables", oc->fileName);
2825 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
2829 if (hdr->sh_type == SHT_PROGBITS
2830 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
2831 /* .text-style section */
2832 return SECTIONKIND_CODE_OR_RODATA;
2835 if (hdr->sh_type == SHT_PROGBITS
2836 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2837 /* .data-style section */
2838 return SECTIONKIND_RWDATA;
2841 if (hdr->sh_type == SHT_PROGBITS
2842 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
2843 /* .rodata-style section */
2844 return SECTIONKIND_CODE_OR_RODATA;
2847 if (hdr->sh_type == SHT_NOBITS
2848 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2849 /* .bss-style section */
2851 return SECTIONKIND_RWDATA;
2854 return SECTIONKIND_OTHER;
2859 ocGetNames_ELF ( ObjectCode* oc )
2864 char* ehdrC = (char*)(oc->image);
2865 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2866 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
2867 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2869 ASSERT(symhash != NULL);
2872 errorBelch("%s: no strtab", oc->fileName);
2877 for (i = 0; i < ehdr->e_shnum; i++) {
2878 /* Figure out what kind of section it is. Logic derived from
2879 Figure 1.14 ("Special Sections") of the ELF document
2880 ("Portable Formats Specification, Version 1.1"). */
2882 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
2884 if (is_bss && shdr[i].sh_size > 0) {
2885 /* This is a non-empty .bss section. Allocate zeroed space for
2886 it, and set its .sh_offset field such that
2887 ehdrC + .sh_offset == addr_of_zeroed_space. */
2888 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
2889 "ocGetNames_ELF(BSS)");
2890 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
2892 debugBelch("BSS section at 0x%x, size %d\n",
2893 zspace, shdr[i].sh_size);
2897 /* fill in the section info */
2898 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
2899 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
2900 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
2901 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
2904 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2906 /* copy stuff into this module's object symbol table */
2907 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2908 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2910 oc->n_symbols = nent;
2911 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2912 "ocGetNames_ELF(oc->symbols)");
2914 for (j = 0; j < nent; j++) {
2916 char isLocal = FALSE; /* avoids uninit-var warning */
2918 char* nm = strtab + stab[j].st_name;
2919 int secno = stab[j].st_shndx;
2921 /* Figure out if we want to add it; if so, set ad to its
2922 address. Otherwise leave ad == NULL. */
2924 if (secno == SHN_COMMON) {
2926 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
2928 debugBelch("COMMON symbol, size %d name %s\n",
2929 stab[j].st_size, nm);
2931 /* Pointless to do addProddableBlock() for this area,
2932 since the linker should never poke around in it. */
2935 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
2936 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
2938 /* and not an undefined symbol */
2939 && stab[j].st_shndx != SHN_UNDEF
2940 /* and not in a "special section" */
2941 && stab[j].st_shndx < SHN_LORESERVE
2943 /* and it's a not a section or string table or anything silly */
2944 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
2945 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
2946 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
2949 /* Section 0 is the undefined section, hence > and not >=. */
2950 ASSERT(secno > 0 && secno < ehdr->e_shnum);
2952 if (shdr[secno].sh_type == SHT_NOBITS) {
2953 debugBelch(" BSS symbol, size %d off %d name %s\n",
2954 stab[j].st_size, stab[j].st_value, nm);
2957 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
2958 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
2961 #ifdef ELF_FUNCTION_DESC
2962 /* dlsym() and the initialisation table both give us function
2963 * descriptors, so to be consistent we store function descriptors
2964 * in the symbol table */
2965 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
2966 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
2968 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
2969 ad, oc->fileName, nm ));
2974 /* And the decision is ... */
2978 oc->symbols[j] = nm;
2981 /* Ignore entirely. */
2983 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
2987 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
2988 strtab + stab[j].st_name ));
2991 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
2992 (int)ELF_ST_BIND(stab[j].st_info),
2993 (int)ELF_ST_TYPE(stab[j].st_info),
2994 (int)stab[j].st_shndx,
2995 strtab + stab[j].st_name
2998 oc->symbols[j] = NULL;
3007 /* Do ELF relocations which lack an explicit addend. All x86-linux
3008 relocations appear to be of this form. */
3010 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3011 Elf_Shdr* shdr, int shnum,
3012 Elf_Sym* stab, char* strtab )
3017 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3018 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3019 int target_shndx = shdr[shnum].sh_info;
3020 int symtab_shndx = shdr[shnum].sh_link;
3022 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3023 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3024 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3025 target_shndx, symtab_shndx ));
3027 /* Skip sections that we're not interested in. */
3030 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3031 if (kind == SECTIONKIND_OTHER) {
3032 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3037 for (j = 0; j < nent; j++) {
3038 Elf_Addr offset = rtab[j].r_offset;
3039 Elf_Addr info = rtab[j].r_info;
3041 Elf_Addr P = ((Elf_Addr)targ) + offset;
3042 Elf_Word* pP = (Elf_Word*)P;
3048 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3049 j, (void*)offset, (void*)info ));
3051 IF_DEBUG(linker,debugBelch( " ZERO" ));
3054 Elf_Sym sym = stab[ELF_R_SYM(info)];
3055 /* First see if it is a local symbol. */
3056 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3057 /* Yes, so we can get the address directly from the ELF symbol
3059 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3061 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3062 + stab[ELF_R_SYM(info)].st_value);
3065 /* No, so look up the name in our global table. */
3066 symbol = strtab + sym.st_name;
3067 S_tmp = lookupSymbol( symbol );
3068 S = (Elf_Addr)S_tmp;
3071 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3074 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3077 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3078 (void*)P, (void*)S, (void*)A ));
3079 checkProddableBlock ( oc, pP );
3083 switch (ELF_R_TYPE(info)) {
3084 # ifdef i386_HOST_ARCH
3085 case R_386_32: *pP = value; break;
3086 case R_386_PC32: *pP = value - P; break;
3089 errorBelch("%s: unhandled ELF relocation(Rel) type %d\n",
3090 oc->fileName, ELF_R_TYPE(info));
3098 /* Do ELF relocations for which explicit addends are supplied.
3099 sparc-solaris relocations appear to be of this form. */
3101 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3102 Elf_Shdr* shdr, int shnum,
3103 Elf_Sym* stab, char* strtab )
3108 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3109 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3110 int target_shndx = shdr[shnum].sh_info;
3111 int symtab_shndx = shdr[shnum].sh_link;
3113 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3114 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3115 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3116 target_shndx, symtab_shndx ));
3118 for (j = 0; j < nent; j++) {
3119 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3120 /* This #ifdef only serves to avoid unused-var warnings. */
3121 Elf_Addr offset = rtab[j].r_offset;
3122 Elf_Addr P = targ + offset;
3124 Elf_Addr info = rtab[j].r_info;
3125 Elf_Addr A = rtab[j].r_addend;
3129 # if defined(sparc_HOST_ARCH)
3130 Elf_Word* pP = (Elf_Word*)P;
3132 # elif defined(ia64_HOST_ARCH)
3133 Elf64_Xword *pP = (Elf64_Xword *)P;
3135 # elif defined(powerpc_HOST_ARCH)
3139 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3140 j, (void*)offset, (void*)info,
3143 IF_DEBUG(linker,debugBelch( " ZERO" ));
3146 Elf_Sym sym = stab[ELF_R_SYM(info)];
3147 /* First see if it is a local symbol. */
3148 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3149 /* Yes, so we can get the address directly from the ELF symbol
3151 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3153 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3154 + stab[ELF_R_SYM(info)].st_value);
3155 #ifdef ELF_FUNCTION_DESC
3156 /* Make a function descriptor for this function */
3157 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3158 S = allocateFunctionDesc(S + A);
3163 /* No, so look up the name in our global table. */
3164 symbol = strtab + sym.st_name;
3165 S_tmp = lookupSymbol( symbol );
3166 S = (Elf_Addr)S_tmp;
3168 #ifdef ELF_FUNCTION_DESC
3169 /* If a function, already a function descriptor - we would
3170 have to copy it to add an offset. */
3171 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3172 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3176 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3179 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3182 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3183 (void*)P, (void*)S, (void*)A ));
3184 /* checkProddableBlock ( oc, (void*)P ); */
3188 switch (ELF_R_TYPE(info)) {
3189 # if defined(sparc_HOST_ARCH)
3190 case R_SPARC_WDISP30:
3191 w1 = *pP & 0xC0000000;
3192 w2 = (Elf_Word)((value - P) >> 2);
3193 ASSERT((w2 & 0xC0000000) == 0);
3198 w1 = *pP & 0xFFC00000;
3199 w2 = (Elf_Word)(value >> 10);
3200 ASSERT((w2 & 0xFFC00000) == 0);
3206 w2 = (Elf_Word)(value & 0x3FF);
3207 ASSERT((w2 & ~0x3FF) == 0);
3211 /* According to the Sun documentation:
3213 This relocation type resembles R_SPARC_32, except it refers to an
3214 unaligned word. That is, the word to be relocated must be treated
3215 as four separate bytes with arbitrary alignment, not as a word
3216 aligned according to the architecture requirements.
3218 (JRS: which means that freeloading on the R_SPARC_32 case
3219 is probably wrong, but hey ...)
3223 w2 = (Elf_Word)value;
3226 # elif defined(ia64_HOST_ARCH)
3227 case R_IA64_DIR64LSB:
3228 case R_IA64_FPTR64LSB:
3231 case R_IA64_PCREL64LSB:
3234 case R_IA64_SEGREL64LSB:
3235 addr = findElfSegment(ehdrC, value);
3238 case R_IA64_GPREL22:
3239 ia64_reloc_gprel22(P, value);
3241 case R_IA64_LTOFF22:
3242 case R_IA64_LTOFF22X:
3243 case R_IA64_LTOFF_FPTR22:
3244 addr = allocateGOTEntry(value);
3245 ia64_reloc_gprel22(P, addr);
3247 case R_IA64_PCREL21B:
3248 ia64_reloc_pcrel21(P, S, oc);
3251 /* This goes with R_IA64_LTOFF22X and points to the load to
3252 * convert into a move. We don't implement relaxation. */
3254 # elif defined(powerpc_HOST_ARCH)
3255 case R_PPC_ADDR16_LO:
3256 *(Elf32_Half*) P = value;
3259 case R_PPC_ADDR16_HI:
3260 *(Elf32_Half*) P = value >> 16;
3263 case R_PPC_ADDR16_HA:
3264 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3268 *(Elf32_Word *) P = value;
3272 *(Elf32_Word *) P = value - P;
3278 if( delta << 6 >> 6 != delta )
3280 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3283 if( value == 0 || delta << 6 >> 6 != delta )
3285 barf( "Unable to make ppcJumpIsland for #%d",
3291 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3292 | (delta & 0x3fffffc);
3296 #if x86_64_HOST_ARCH
3298 *(Elf64_Xword *)P = value;
3303 StgInt64 off = value - P;
3304 if (off >= 0x7fffffffL || off < -0x80000000L) {
3305 barf("R_X86_64_PC32 relocation out of range: %s = %p",
3308 *(Elf64_Word *)P = (Elf64_Word)off;
3313 if (value >= 0x7fffffffL) {
3314 barf("R_X86_64_32 relocation out of range: %s = %p\n",
3317 *(Elf64_Word *)P = (Elf64_Word)value;
3321 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3322 barf("R_X86_64_32S relocation out of range: %s = %p\n",
3325 *(Elf64_Sword *)P = (Elf64_Sword)value;
3330 errorBelch("%s: unhandled ELF relocation(RelA) type %d\n",
3331 oc->fileName, ELF_R_TYPE(info));
3340 ocResolve_ELF ( ObjectCode* oc )
3344 Elf_Sym* stab = NULL;
3345 char* ehdrC = (char*)(oc->image);
3346 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3347 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3349 /* first find "the" symbol table */
3350 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3352 /* also go find the string table */
3353 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3355 if (stab == NULL || strtab == NULL) {
3356 errorBelch("%s: can't find string or symbol table", oc->fileName);
3360 /* Process the relocation sections. */
3361 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3362 if (shdr[shnum].sh_type == SHT_REL) {
3363 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3364 shnum, stab, strtab );
3368 if (shdr[shnum].sh_type == SHT_RELA) {
3369 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3370 shnum, stab, strtab );
3375 /* Free the local symbol table; we won't need it again. */
3376 freeHashTable(oc->lochash, NULL);
3379 #if defined(powerpc_HOST_ARCH)
3380 ocFlushInstructionCache( oc );
3388 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3389 * at the front. The following utility functions pack and unpack instructions, and
3390 * take care of the most common relocations.
3393 #ifdef ia64_HOST_ARCH
3396 ia64_extract_instruction(Elf64_Xword *target)
3399 int slot = (Elf_Addr)target & 3;
3400 target = (Elf_Addr)target & ~3;
3408 return ((w1 >> 5) & 0x1ffffffffff);
3410 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3414 barf("ia64_extract_instruction: invalid slot %p", target);
3419 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3421 int slot = (Elf_Addr)target & 3;
3422 target = (Elf_Addr)target & ~3;
3427 *target |= value << 5;
3430 *target |= value << 46;
3431 *(target+1) |= value >> 18;
3434 *(target+1) |= value << 23;
3440 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3442 Elf64_Xword instruction;
3443 Elf64_Sxword rel_value;
3445 rel_value = value - gp_val;
3446 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3447 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3449 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3450 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3451 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3452 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3453 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3454 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3458 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3460 Elf64_Xword instruction;
3461 Elf64_Sxword rel_value;
3464 entry = allocatePLTEntry(value, oc);
3466 rel_value = (entry >> 4) - (target >> 4);
3467 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3468 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3470 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3471 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3472 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3473 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3479 * PowerPC ELF specifics
3482 #ifdef powerpc_HOST_ARCH
3484 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3490 ehdr = (Elf_Ehdr *) oc->image;
3491 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3493 for( i = 0; i < ehdr->e_shnum; i++ )
3494 if( shdr[i].sh_type == SHT_SYMTAB )
3497 if( i == ehdr->e_shnum )
3499 errorBelch( "This ELF file contains no symtab" );
3503 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3505 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3506 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3511 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3514 #endif /* powerpc */
3518 /* --------------------------------------------------------------------------
3520 * ------------------------------------------------------------------------*/
3522 #if defined(OBJFORMAT_MACHO)
3525 Support for MachO linking on Darwin/MacOS X
3526 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3528 I hereby formally apologize for the hackish nature of this code.
3529 Things that need to be done:
3530 *) implement ocVerifyImage_MachO
3531 *) add still more sanity checks.
3534 #ifdef powerpc_HOST_ARCH
3535 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3537 struct mach_header *header = (struct mach_header *) oc->image;
3538 struct load_command *lc = (struct load_command *) (header + 1);
3541 for( i = 0; i < header->ncmds; i++ )
3543 if( lc->cmd == LC_SYMTAB )
3545 // Find out the first and last undefined external
3546 // symbol, so we don't have to allocate too many
3548 struct symtab_command *symLC = (struct symtab_command *) lc;
3549 unsigned min = symLC->nsyms, max = 0;
3550 struct nlist *nlist =
3551 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3553 for(i=0;i<symLC->nsyms;i++)
3555 if(nlist[i].n_type & N_STAB)
3557 else if(nlist[i].n_type & N_EXT)
3559 if((nlist[i].n_type & N_TYPE) == N_UNDF
3560 && (nlist[i].n_value == 0))
3570 return ocAllocateJumpIslands(oc, max - min + 1, min);
3575 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3577 return ocAllocateJumpIslands(oc,0,0);
3581 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3583 // FIXME: do some verifying here
3587 static int resolveImports(
3590 struct symtab_command *symLC,
3591 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3592 unsigned long *indirectSyms,
3593 struct nlist *nlist)
3597 for(i=0;i*4<sect->size;i++)
3599 // according to otool, reserved1 contains the first index into the indirect symbol table
3600 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3601 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3604 if((symbol->n_type & N_TYPE) == N_UNDF
3605 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3606 addr = (void*) (symbol->n_value);
3607 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3610 addr = lookupSymbol(nm);
3613 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3617 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3618 ((void**)(image + sect->offset))[i] = addr;
3624 static unsigned long relocateAddress(
3627 struct section* sections,
3628 unsigned long address)
3631 for(i = 0; i < nSections; i++)
3633 if(sections[i].addr <= address
3634 && address < sections[i].addr + sections[i].size)
3636 return (unsigned long)oc->image
3637 + sections[i].offset + address - sections[i].addr;
3640 barf("Invalid Mach-O file:"
3641 "Address out of bounds while relocating object file");
3645 static int relocateSection(
3648 struct symtab_command *symLC, struct nlist *nlist,
3649 int nSections, struct section* sections, struct section *sect)
3651 struct relocation_info *relocs;
3654 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3656 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3658 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
3660 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
3664 relocs = (struct relocation_info*) (image + sect->reloff);
3668 if(relocs[i].r_address & R_SCATTERED)
3670 struct scattered_relocation_info *scat =
3671 (struct scattered_relocation_info*) &relocs[i];
3675 if(scat->r_length == 2)
3677 unsigned long word = 0;
3678 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3679 checkProddableBlock(oc,wordPtr);
3681 // Note on relocation types:
3682 // i386 uses the GENERIC_RELOC_* types,
3683 // while ppc uses special PPC_RELOC_* types.
3684 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
3685 // in both cases, all others are different.
3686 // Therefore, we use GENERIC_RELOC_VANILLA
3687 // and GENERIC_RELOC_PAIR instead of the PPC variants,
3688 // and use #ifdefs for the other types.
3690 // Step 1: Figure out what the relocated value should be
3691 if(scat->r_type == GENERIC_RELOC_VANILLA)
3693 word = *wordPtr + (unsigned long) relocateAddress(
3700 #ifdef powerpc_HOST_ARCH
3701 else if(scat->r_type == PPC_RELOC_SECTDIFF
3702 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3703 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3704 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3706 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
3709 struct scattered_relocation_info *pair =
3710 (struct scattered_relocation_info*) &relocs[i+1];
3712 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
3713 barf("Invalid Mach-O file: "
3714 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
3716 word = (unsigned long)
3717 (relocateAddress(oc, nSections, sections, scat->r_value)
3718 - relocateAddress(oc, nSections, sections, pair->r_value));
3721 #ifdef powerpc_HOST_ARCH
3722 else if(scat->r_type == PPC_RELOC_HI16
3723 || scat->r_type == PPC_RELOC_LO16
3724 || scat->r_type == PPC_RELOC_HA16
3725 || scat->r_type == PPC_RELOC_LO14)
3726 { // these are generated by label+offset things
3727 struct relocation_info *pair = &relocs[i+1];
3728 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3729 barf("Invalid Mach-O file: "
3730 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3732 if(scat->r_type == PPC_RELOC_LO16)
3734 word = ((unsigned short*) wordPtr)[1];
3735 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3737 else if(scat->r_type == PPC_RELOC_LO14)
3739 barf("Unsupported Relocation: PPC_RELOC_LO14");
3740 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3741 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3743 else if(scat->r_type == PPC_RELOC_HI16)
3745 word = ((unsigned short*) wordPtr)[1] << 16;
3746 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3748 else if(scat->r_type == PPC_RELOC_HA16)
3750 word = ((unsigned short*) wordPtr)[1] << 16;
3751 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3755 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
3762 continue; // ignore the others
3764 #ifdef powerpc_HOST_ARCH
3765 if(scat->r_type == GENERIC_RELOC_VANILLA
3766 || scat->r_type == PPC_RELOC_SECTDIFF)
3768 if(scat->r_type == GENERIC_RELOC_VANILLA
3769 || scat->r_type == GENERIC_RELOC_SECTDIFF)
3774 #ifdef powerpc_HOST_ARCH
3775 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
3777 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3779 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
3781 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3783 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
3785 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3786 + ((word & (1<<15)) ? 1 : 0);
3792 continue; // FIXME: I hope it's OK to ignore all the others.
3796 struct relocation_info *reloc = &relocs[i];
3797 if(reloc->r_pcrel && !reloc->r_extern)
3800 if(reloc->r_length == 2)
3802 unsigned long word = 0;
3803 #ifdef powerpc_HOST_ARCH
3804 unsigned long jumpIsland = 0;
3805 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
3806 // to avoid warning and to catch
3810 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
3811 checkProddableBlock(oc,wordPtr);
3813 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3817 #ifdef powerpc_HOST_ARCH
3818 else if(reloc->r_type == PPC_RELOC_LO16)
3820 word = ((unsigned short*) wordPtr)[1];
3821 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3823 else if(reloc->r_type == PPC_RELOC_HI16)
3825 word = ((unsigned short*) wordPtr)[1] << 16;
3826 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3828 else if(reloc->r_type == PPC_RELOC_HA16)
3830 word = ((unsigned short*) wordPtr)[1] << 16;
3831 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3833 else if(reloc->r_type == PPC_RELOC_BR24)
3836 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
3840 if(!reloc->r_extern)
3843 sections[reloc->r_symbolnum-1].offset
3844 - sections[reloc->r_symbolnum-1].addr
3851 struct nlist *symbol = &nlist[reloc->r_symbolnum];
3852 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3853 void *symbolAddress = lookupSymbol(nm);
3856 errorBelch("\nunknown symbol `%s'", nm);
3862 #ifdef powerpc_HOST_ARCH
3863 // In the .o file, this should be a relative jump to NULL
3864 // and we'll change it to a relative jump to the symbol
3865 ASSERT(-word == reloc->r_address);
3866 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,(unsigned long) symbolAddress);
3869 offsetToJumpIsland = word + jumpIsland
3870 - (((long)image) + sect->offset - sect->addr);
3873 word += (unsigned long) symbolAddress
3874 - (((long)image) + sect->offset - sect->addr);
3878 word += (unsigned long) symbolAddress;
3882 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3887 #ifdef powerpc_HOST_ARCH
3888 else if(reloc->r_type == PPC_RELOC_LO16)
3890 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3893 else if(reloc->r_type == PPC_RELOC_HI16)
3895 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3898 else if(reloc->r_type == PPC_RELOC_HA16)
3900 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3901 + ((word & (1<<15)) ? 1 : 0);
3904 else if(reloc->r_type == PPC_RELOC_BR24)
3906 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3908 // The branch offset is too large.
3909 // Therefore, we try to use a jump island.
3912 barf("unconditional relative branch out of range: "
3913 "no jump island available");
3916 word = offsetToJumpIsland;
3917 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3918 barf("unconditional relative branch out of range: "
3919 "jump island out of range");
3921 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
3926 barf("\nunknown relocation %d",reloc->r_type);
3933 static int ocGetNames_MachO(ObjectCode* oc)
3935 char *image = (char*) oc->image;
3936 struct mach_header *header = (struct mach_header*) image;
3937 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3938 unsigned i,curSymbol = 0;
3939 struct segment_command *segLC = NULL;
3940 struct section *sections;
3941 struct symtab_command *symLC = NULL;
3942 struct nlist *nlist;
3943 unsigned long commonSize = 0;
3944 char *commonStorage = NULL;
3945 unsigned long commonCounter;
3947 for(i=0;i<header->ncmds;i++)
3949 if(lc->cmd == LC_SEGMENT)
3950 segLC = (struct segment_command*) lc;
3951 else if(lc->cmd == LC_SYMTAB)
3952 symLC = (struct symtab_command*) lc;
3953 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
3956 sections = (struct section*) (segLC+1);
3957 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
3960 for(i=0;i<segLC->nsects;i++)
3962 if(sections[i].size == 0)
3965 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
3967 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
3968 "ocGetNames_MachO(common symbols)");
3969 sections[i].offset = zeroFillArea - image;
3972 if(!strcmp(sections[i].sectname,"__text"))
3973 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
3974 (void*) (image + sections[i].offset),
3975 (void*) (image + sections[i].offset + sections[i].size));
3976 else if(!strcmp(sections[i].sectname,"__const"))
3977 addSection(oc, SECTIONKIND_RWDATA,
3978 (void*) (image + sections[i].offset),
3979 (void*) (image + sections[i].offset + sections[i].size));
3980 else if(!strcmp(sections[i].sectname,"__data"))
3981 addSection(oc, SECTIONKIND_RWDATA,
3982 (void*) (image + sections[i].offset),
3983 (void*) (image + sections[i].offset + sections[i].size));
3984 else if(!strcmp(sections[i].sectname,"__bss")
3985 || !strcmp(sections[i].sectname,"__common"))
3986 addSection(oc, SECTIONKIND_RWDATA,
3987 (void*) (image + sections[i].offset),
3988 (void*) (image + sections[i].offset + sections[i].size));
3990 addProddableBlock(oc, (void*) (image + sections[i].offset),
3994 // count external symbols defined here
3998 for(i=0;i<symLC->nsyms;i++)
4000 if(nlist[i].n_type & N_STAB)
4002 else if(nlist[i].n_type & N_EXT)
4004 if((nlist[i].n_type & N_TYPE) == N_UNDF
4005 && (nlist[i].n_value != 0))
4007 commonSize += nlist[i].n_value;
4010 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4015 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4016 "ocGetNames_MachO(oc->symbols)");
4020 for(i=0;i<symLC->nsyms;i++)
4022 if(nlist[i].n_type & N_STAB)
4024 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4026 if(nlist[i].n_type & N_EXT)
4028 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4029 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4031 + sections[nlist[i].n_sect-1].offset
4032 - sections[nlist[i].n_sect-1].addr
4033 + nlist[i].n_value);
4034 oc->symbols[curSymbol++] = nm;
4038 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4039 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
4041 + sections[nlist[i].n_sect-1].offset
4042 - sections[nlist[i].n_sect-1].addr
4043 + nlist[i].n_value);
4049 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4050 commonCounter = (unsigned long)commonStorage;
4053 for(i=0;i<symLC->nsyms;i++)
4055 if((nlist[i].n_type & N_TYPE) == N_UNDF
4056 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4058 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4059 unsigned long sz = nlist[i].n_value;
4061 nlist[i].n_value = commonCounter;
4063 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4064 (void*)commonCounter);
4065 oc->symbols[curSymbol++] = nm;
4067 commonCounter += sz;
4074 static int ocResolve_MachO(ObjectCode* oc)
4076 char *image = (char*) oc->image;
4077 struct mach_header *header = (struct mach_header*) image;
4078 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4080 struct segment_command *segLC = NULL;
4081 struct section *sections, *la_ptrs = NULL, *nl_ptrs = NULL;
4082 struct symtab_command *symLC = NULL;
4083 struct dysymtab_command *dsymLC = NULL;
4084 struct nlist *nlist;
4086 for(i=0;i<header->ncmds;i++)
4088 if(lc->cmd == LC_SEGMENT)
4089 segLC = (struct segment_command*) lc;
4090 else if(lc->cmd == LC_SYMTAB)
4091 symLC = (struct symtab_command*) lc;
4092 else if(lc->cmd == LC_DYSYMTAB)
4093 dsymLC = (struct dysymtab_command*) lc;
4094 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4097 sections = (struct section*) (segLC+1);
4098 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4101 for(i=0;i<segLC->nsects;i++)
4103 if(!strcmp(sections[i].sectname,"__la_symbol_ptr"))
4104 la_ptrs = §ions[i];
4105 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr"))
4106 nl_ptrs = §ions[i];
4107 else if(!strcmp(sections[i].sectname,"__la_sym_ptr2"))
4108 la_ptrs = §ions[i];
4109 else if(!strcmp(sections[i].sectname,"__la_sym_ptr3"))
4110 la_ptrs = §ions[i];
4115 unsigned long *indirectSyms
4116 = (unsigned long*) (image + dsymLC->indirectsymoff);
4119 if(!resolveImports(oc,image,symLC,la_ptrs,indirectSyms,nlist))
4122 if(!resolveImports(oc,image,symLC,nl_ptrs,indirectSyms,nlist))
4126 for(i=0;i<segLC->nsects;i++)
4128 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4132 /* Free the local symbol table; we won't need it again. */
4133 freeHashTable(oc->lochash, NULL);
4136 #if defined (powerpc_HOST_ARCH)
4137 ocFlushInstructionCache( oc );
4143 #ifdef powerpc_HOST_ARCH
4145 * The Mach-O object format uses leading underscores. But not everywhere.
4146 * There is a small number of runtime support functions defined in
4147 * libcc_dynamic.a whose name does not have a leading underscore.
4148 * As a consequence, we can't get their address from C code.
4149 * We have to use inline assembler just to take the address of a function.
4153 static void machoInitSymbolsWithoutUnderscore()
4155 extern void* symbolsWithoutUnderscore[];
4156 void **p = symbolsWithoutUnderscore;
4157 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4161 __asm__ volatile(".long " # x);
4163 RTS_MACHO_NOUNDERLINE_SYMBOLS
4165 __asm__ volatile(".text");
4169 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4171 RTS_MACHO_NOUNDERLINE_SYMBOLS
4178 * Figure out by how much to shift the entire Mach-O file in memory
4179 * when loading so that its single segment ends up 16-byte-aligned
4181 static int machoGetMisalignment( FILE * f )
4183 struct mach_header header;
4186 fread(&header, sizeof(header), 1, f);
4189 if(header.magic != MH_MAGIC)
4192 misalignment = (header.sizeofcmds + sizeof(header))
4195 return misalignment ? (16 - misalignment) : 0;