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(signal_handlers) \
146 SymX(stg_sig_install) \
150 #if defined (cygwin32_HOST_OS)
151 #define RTS_MINGW_ONLY_SYMBOLS /**/
152 /* Don't have the ability to read import libs / archives, so
153 * we have to stupidly list a lot of what libcygwin.a
156 #define RTS_CYGWIN_ONLY_SYMBOLS \
234 #elif !defined(mingw32_HOST_OS)
235 #define RTS_MINGW_ONLY_SYMBOLS /**/
236 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
237 #else /* defined(mingw32_HOST_OS) */
238 #define RTS_POSIX_ONLY_SYMBOLS /**/
239 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
241 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
243 #define RTS_MINGW_EXTRA_SYMS \
244 Sym(_imp____mb_cur_max) \
247 #define RTS_MINGW_EXTRA_SYMS
250 /* These are statically linked from the mingw libraries into the ghc
251 executable, so we have to employ this hack. */
252 #define RTS_MINGW_ONLY_SYMBOLS \
253 SymX(asyncReadzh_fast) \
254 SymX(asyncWritezh_fast) \
255 SymX(asyncDoProczh_fast) \
267 SymX(getservbyname) \
268 SymX(getservbyport) \
269 SymX(getprotobynumber) \
270 SymX(getprotobyname) \
271 SymX(gethostbyname) \
272 SymX(gethostbyaddr) \
306 SymX(rts_InstallConsoleEvent) \
307 SymX(rts_ConsoleHandlerDone) \
309 Sym(_imp___timezone) \
318 RTS_MINGW_EXTRA_SYMS \
322 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
323 #define RTS_DARWIN_ONLY_SYMBOLS \
324 Sym(asprintf$LDBLStub) \
328 Sym(fprintf$LDBLStub) \
329 Sym(fscanf$LDBLStub) \
330 Sym(fwprintf$LDBLStub) \
331 Sym(fwscanf$LDBLStub) \
332 Sym(printf$LDBLStub) \
333 Sym(scanf$LDBLStub) \
334 Sym(snprintf$LDBLStub) \
335 Sym(sprintf$LDBLStub) \
336 Sym(sscanf$LDBLStub) \
337 Sym(strtold$LDBLStub) \
338 Sym(swprintf$LDBLStub) \
339 Sym(swscanf$LDBLStub) \
340 Sym(syslog$LDBLStub) \
341 Sym(vasprintf$LDBLStub) \
343 Sym(verrc$LDBLStub) \
344 Sym(verrx$LDBLStub) \
345 Sym(vfprintf$LDBLStub) \
346 Sym(vfscanf$LDBLStub) \
347 Sym(vfwprintf$LDBLStub) \
348 Sym(vfwscanf$LDBLStub) \
349 Sym(vprintf$LDBLStub) \
350 Sym(vscanf$LDBLStub) \
351 Sym(vsnprintf$LDBLStub) \
352 Sym(vsprintf$LDBLStub) \
353 Sym(vsscanf$LDBLStub) \
354 Sym(vswprintf$LDBLStub) \
355 Sym(vswscanf$LDBLStub) \
356 Sym(vsyslog$LDBLStub) \
357 Sym(vwarn$LDBLStub) \
358 Sym(vwarnc$LDBLStub) \
359 Sym(vwarnx$LDBLStub) \
360 Sym(vwprintf$LDBLStub) \
361 Sym(vwscanf$LDBLStub) \
363 Sym(warnc$LDBLStub) \
364 Sym(warnx$LDBLStub) \
365 Sym(wcstold$LDBLStub) \
366 Sym(wprintf$LDBLStub) \
369 #define RTS_DARWIN_ONLY_SYMBOLS
373 # define MAIN_CAP_SYM SymX(MainCapability)
375 # define MAIN_CAP_SYM
378 #if !defined(mingw32_HOST_OS)
379 #define RTS_USER_SIGNALS_SYMBOLS \
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) \
586 SymX(__hscore_get_saved_termios) \
587 SymX(__hscore_set_saved_termios) \
589 SymX(startupHaskell) \
590 SymX(shutdownHaskell) \
591 SymX(shutdownHaskellAndExit) \
592 SymX(stable_ptr_table) \
593 SymX(stackOverflow) \
594 SymX(stg_CAF_BLACKHOLE_info) \
595 SymX(awakenBlockedQueue) \
596 SymX(stg_CHARLIKE_closure) \
597 SymX(stg_EMPTY_MVAR_info) \
598 SymX(stg_IND_STATIC_info) \
599 SymX(stg_INTLIKE_closure) \
600 SymX(stg_MUT_ARR_PTRS_FROZEN_info) \
601 SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \
602 SymX(stg_WEAK_info) \
603 SymX(stg_ap_0_info) \
604 SymX(stg_ap_v_info) \
605 SymX(stg_ap_f_info) \
606 SymX(stg_ap_d_info) \
607 SymX(stg_ap_l_info) \
608 SymX(stg_ap_n_info) \
609 SymX(stg_ap_p_info) \
610 SymX(stg_ap_pv_info) \
611 SymX(stg_ap_pp_info) \
612 SymX(stg_ap_ppv_info) \
613 SymX(stg_ap_ppp_info) \
614 SymX(stg_ap_pppv_info) \
615 SymX(stg_ap_pppp_info) \
616 SymX(stg_ap_ppppp_info) \
617 SymX(stg_ap_pppppp_info) \
618 SymX(stg_ap_1_upd_info) \
619 SymX(stg_ap_2_upd_info) \
620 SymX(stg_ap_3_upd_info) \
621 SymX(stg_ap_4_upd_info) \
622 SymX(stg_ap_5_upd_info) \
623 SymX(stg_ap_6_upd_info) \
624 SymX(stg_ap_7_upd_info) \
626 SymX(stg_sel_0_upd_info) \
627 SymX(stg_sel_10_upd_info) \
628 SymX(stg_sel_11_upd_info) \
629 SymX(stg_sel_12_upd_info) \
630 SymX(stg_sel_13_upd_info) \
631 SymX(stg_sel_14_upd_info) \
632 SymX(stg_sel_15_upd_info) \
633 SymX(stg_sel_1_upd_info) \
634 SymX(stg_sel_2_upd_info) \
635 SymX(stg_sel_3_upd_info) \
636 SymX(stg_sel_4_upd_info) \
637 SymX(stg_sel_5_upd_info) \
638 SymX(stg_sel_6_upd_info) \
639 SymX(stg_sel_7_upd_info) \
640 SymX(stg_sel_8_upd_info) \
641 SymX(stg_sel_9_upd_info) \
642 SymX(stg_upd_frame_info) \
643 SymX(suspendThread) \
644 SymX(takeMVarzh_fast) \
645 SymX(timesIntegerzh_fast) \
646 SymX(tryPutMVarzh_fast) \
647 SymX(tryTakeMVarzh_fast) \
648 SymX(unblockAsyncExceptionszh_fast) \
650 SymX(unsafeThawArrayzh_fast) \
651 SymX(waitReadzh_fast) \
652 SymX(waitWritezh_fast) \
653 SymX(word2Integerzh_fast) \
654 SymX(writeTVarzh_fast) \
655 SymX(xorIntegerzh_fast) \
657 RTS_USER_SIGNALS_SYMBOLS
659 #ifdef SUPPORT_LONG_LONGS
660 #define RTS_LONG_LONG_SYMS \
661 SymX(int64ToIntegerzh_fast) \
662 SymX(word64ToIntegerzh_fast)
664 #define RTS_LONG_LONG_SYMS /* nothing */
667 // 64-bit support functions in libgcc.a
668 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
669 #define RTS_LIBGCC_SYMBOLS \
679 #elif defined(ia64_HOST_ARCH)
680 #define RTS_LIBGCC_SYMBOLS \
688 #define RTS_LIBGCC_SYMBOLS
691 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
692 // Symbols that don't have a leading underscore
693 // on Mac OS X. They have to receive special treatment,
694 // see machoInitSymbolsWithoutUnderscore()
695 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
700 /* entirely bogus claims about types of these symbols */
701 #define Sym(vvv) extern void vvv(void);
702 #define SymX(vvv) /**/
703 #define SymX_redirect(vvv,xxx) /**/
707 RTS_POSIX_ONLY_SYMBOLS
708 RTS_MINGW_ONLY_SYMBOLS
709 RTS_CYGWIN_ONLY_SYMBOLS
710 RTS_DARWIN_ONLY_SYMBOLS
716 #ifdef LEADING_UNDERSCORE
717 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
719 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
722 #define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
724 #define SymX(vvv) Sym(vvv)
726 // SymX_redirect allows us to redirect references to one symbol to
727 // another symbol. See newCAF/newDynCAF for an example.
728 #define SymX_redirect(vvv,xxx) \
729 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
732 static RtsSymbolVal rtsSyms[] = {
736 RTS_POSIX_ONLY_SYMBOLS
737 RTS_MINGW_ONLY_SYMBOLS
738 RTS_CYGWIN_ONLY_SYMBOLS
740 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
741 // dyld stub code contains references to this,
742 // but it should never be called because we treat
743 // lazy pointers as nonlazy.
744 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
746 { 0, 0 } /* sentinel */
749 /* -----------------------------------------------------------------------------
750 * Insert symbols into hash tables, checking for duplicates.
752 static void ghciInsertStrHashTable ( char* obj_name,
758 if (lookupHashTable(table, (StgWord)key) == NULL)
760 insertStrHashTable(table, (StgWord)key, data);
765 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
767 "whilst processing object file\n"
769 "This could be caused by:\n"
770 " * Loading two different object files which export the same symbol\n"
771 " * Specifying the same object file twice on the GHCi command line\n"
772 " * An incorrect `package.conf' entry, causing some object to be\n"
774 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
783 /* -----------------------------------------------------------------------------
784 * initialize the object linker
788 static int linker_init_done = 0 ;
790 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
791 static void *dl_prog_handle;
794 /* dlopen(NULL,..) doesn't work so we grab libc explicitly */
795 #if defined(openbsd_HOST_OS)
796 static void *dl_libc_handle;
804 /* Make initLinker idempotent, so we can call it
805 before evey relevant operation; that means we
806 don't need to initialise the linker separately */
807 if (linker_init_done == 1) { return; } else {
808 linker_init_done = 1;
811 symhash = allocStrHashTable();
813 /* populate the symbol table with stuff from the RTS */
814 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
815 ghciInsertStrHashTable("(GHCi built-in symbols)",
816 symhash, sym->lbl, sym->addr);
818 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
819 machoInitSymbolsWithoutUnderscore();
822 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
823 # if defined(RTLD_DEFAULT)
824 dl_prog_handle = RTLD_DEFAULT;
826 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
827 # if defined(openbsd_HOST_OS)
828 dl_libc_handle = dlopen("libc.so", RTLD_LAZY);
830 # endif /* RTLD_DEFAULT */
834 /* -----------------------------------------------------------------------------
835 * Loading DLL or .so dynamic libraries
836 * -----------------------------------------------------------------------------
838 * Add a DLL from which symbols may be found. In the ELF case, just
839 * do RTLD_GLOBAL-style add, so no further messing around needs to
840 * happen in order that symbols in the loaded .so are findable --
841 * lookupSymbol() will subsequently see them by dlsym on the program's
842 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
844 * In the PEi386 case, open the DLLs and put handles to them in a
845 * linked list. When looking for a symbol, try all handles in the
846 * list. This means that we need to load even DLLs that are guaranteed
847 * to be in the ghc.exe image already, just so we can get a handle
848 * to give to loadSymbol, so that we can find the symbols. For such
849 * libraries, the LoadLibrary call should be a no-op except for returning
854 #if defined(OBJFORMAT_PEi386)
855 /* A record for storing handles into DLLs. */
860 struct _OpenedDLL* next;
865 /* A list thereof. */
866 static OpenedDLL* opened_dlls = NULL;
870 addDLL( char *dll_name )
872 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
873 /* ------------------- ELF DLL loader ------------------- */
879 hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
882 /* dlopen failed; return a ptr to the error msg. */
884 if (errmsg == NULL) errmsg = "addDLL: unknown error";
891 # elif defined(OBJFORMAT_PEi386)
892 /* ------------------- Win32 DLL loader ------------------- */
900 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
902 /* See if we've already got it, and ignore if so. */
903 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
904 if (0 == strcmp(o_dll->name, dll_name))
908 /* The file name has no suffix (yet) so that we can try
909 both foo.dll and foo.drv
911 The documentation for LoadLibrary says:
912 If no file name extension is specified in the lpFileName
913 parameter, the default library extension .dll is
914 appended. However, the file name string can include a trailing
915 point character (.) to indicate that the module name has no
918 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
919 sprintf(buf, "%s.DLL", dll_name);
920 instance = LoadLibrary(buf);
921 if (instance == NULL) {
922 sprintf(buf, "%s.DRV", dll_name); // KAA: allow loading of drivers (like winspool.drv)
923 instance = LoadLibrary(buf);
924 if (instance == NULL) {
927 /* LoadLibrary failed; return a ptr to the error msg. */
928 return "addDLL: unknown error";
933 /* Add this DLL to the list of DLLs in which to search for symbols. */
934 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
935 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
936 strcpy(o_dll->name, dll_name);
937 o_dll->instance = instance;
938 o_dll->next = opened_dlls;
943 barf("addDLL: not implemented on this platform");
947 /* -----------------------------------------------------------------------------
948 * lookup a symbol in the hash table
951 lookupSymbol( char *lbl )
955 ASSERT(symhash != NULL);
956 val = lookupStrHashTable(symhash, lbl);
959 # if defined(OBJFORMAT_ELF)
960 # if defined(openbsd_HOST_OS)
961 val = dlsym(dl_prog_handle, lbl);
962 return (val != NULL) ? val : dlsym(dl_libc_handle,lbl);
963 # elif defined(x86_64_HOST_ARCH)
964 val = dlsym(dl_prog_handle, lbl);
965 if (val >= (void *)0x80000000) {
967 new_val = x86_64_high_symbol(lbl, val);
968 IF_DEBUG(linker,debugBelch("lookupSymbol: relocating out of range symbol: %s = %p, now %p\n", lbl, val, new_val));
973 # else /* not openbsd */
974 return dlsym(dl_prog_handle, lbl);
976 # elif defined(OBJFORMAT_MACHO)
977 if(NSIsSymbolNameDefined(lbl)) {
978 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
979 return NSAddressOfSymbol(symbol);
983 # elif defined(OBJFORMAT_PEi386)
986 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
987 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
989 /* HACK: if the name has an initial underscore, try stripping
990 it off & look that up first. I've yet to verify whether there's
991 a Rule that governs whether an initial '_' *should always* be
992 stripped off when mapping from import lib name to the DLL name.
994 sym = GetProcAddress(o_dll->instance, (lbl+1));
996 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
1000 sym = GetProcAddress(o_dll->instance, lbl);
1002 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
1017 __attribute((unused))
1019 lookupLocalSymbol( ObjectCode* oc, char *lbl )
1023 val = lookupStrHashTable(oc->lochash, lbl);
1033 /* -----------------------------------------------------------------------------
1034 * Debugging aid: look in GHCi's object symbol tables for symbols
1035 * within DELTA bytes of the specified address, and show their names.
1038 void ghci_enquire ( char* addr );
1040 void ghci_enquire ( char* addr )
1045 const int DELTA = 64;
1050 for (oc = objects; oc; oc = oc->next) {
1051 for (i = 0; i < oc->n_symbols; i++) {
1052 sym = oc->symbols[i];
1053 if (sym == NULL) continue;
1054 // debugBelch("enquire %p %p\n", sym, oc->lochash);
1056 if (oc->lochash != NULL) {
1057 a = lookupStrHashTable(oc->lochash, sym);
1060 a = lookupStrHashTable(symhash, sym);
1063 // debugBelch("ghci_enquire: can't find %s\n", sym);
1065 else if (addr-DELTA <= a && a <= addr+DELTA) {
1066 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1073 #ifdef ia64_HOST_ARCH
1074 static unsigned int PLTSize(void);
1077 /* -----------------------------------------------------------------------------
1078 * Load an obj (populate the global symbol table, but don't resolve yet)
1080 * Returns: 1 if ok, 0 on error.
1083 loadObj( char *path )
1090 void *map_addr = NULL;
1097 /* debugBelch("loadObj %s\n", path ); */
1099 /* Check that we haven't already loaded this object.
1100 Ignore requests to load multiple times */
1104 for (o = objects; o; o = o->next) {
1105 if (0 == strcmp(o->fileName, path)) {
1107 break; /* don't need to search further */
1111 IF_DEBUG(linker, debugBelch(
1112 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1113 "same object file twice:\n"
1115 "GHCi will ignore this, but be warned.\n"
1117 return 1; /* success */
1121 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1123 # if defined(OBJFORMAT_ELF)
1124 oc->formatName = "ELF";
1125 # elif defined(OBJFORMAT_PEi386)
1126 oc->formatName = "PEi386";
1127 # elif defined(OBJFORMAT_MACHO)
1128 oc->formatName = "Mach-O";
1131 barf("loadObj: not implemented on this platform");
1134 r = stat(path, &st);
1135 if (r == -1) { return 0; }
1137 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1138 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1139 strcpy(oc->fileName, path);
1141 oc->fileSize = st.st_size;
1143 oc->sections = NULL;
1144 oc->lochash = allocStrHashTable();
1145 oc->proddables = NULL;
1147 /* chain it onto the list of objects */
1152 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1154 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1156 #if defined(openbsd_HOST_OS)
1157 fd = open(path, O_RDONLY, S_IRUSR);
1159 fd = open(path, O_RDONLY);
1162 barf("loadObj: can't open `%s'", path);
1164 pagesize = getpagesize();
1166 #ifdef ia64_HOST_ARCH
1167 /* The PLT needs to be right before the object */
1168 n = ROUND_UP(PLTSize(), pagesize);
1169 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1170 if (oc->plt == MAP_FAILED)
1171 barf("loadObj: can't allocate PLT");
1174 map_addr = oc->plt + n;
1177 n = ROUND_UP(oc->fileSize, pagesize);
1179 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
1180 * small memory model on this architecture (see gcc docs,
1183 #ifdef x86_64_HOST_ARCH
1184 #define EXTRA_MAP_FLAGS MAP_32BIT
1186 #define EXTRA_MAP_FLAGS 0
1189 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1190 MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0);
1191 if (oc->image == MAP_FAILED)
1192 barf("loadObj: can't map `%s'", path);
1196 #else /* !USE_MMAP */
1198 /* load the image into memory */
1199 f = fopen(path, "rb");
1201 barf("loadObj: can't read `%s'", path);
1203 #ifdef darwin_HOST_OS
1204 // In a Mach-O .o file, all sections can and will be misaligned
1205 // if the total size of the headers is not a multiple of the
1206 // desired alignment. This is fine for .o files that only serve
1207 // as input for the static linker, but it's not fine for us,
1208 // as SSE (used by gcc for floating point) and Altivec require
1209 // 16-byte alignment.
1210 // We calculate the correct alignment from the header before
1211 // reading the file, and then we misalign oc->image on purpose so
1212 // that the actual sections end up aligned again.
1213 misalignment = machoGetMisalignment(f);
1214 oc->misalignment = misalignment;
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 )
1431 int misalignment = 0;
1433 misalignment = oc->misalignment;
1438 // round up to the nearest 4
1439 aligned = (oc->fileSize + 3) & ~3;
1442 #ifndef linux_HOST_OS /* mremap is a linux extension */
1443 #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined
1446 pagesize = getpagesize();
1447 n = ROUND_UP( oc->fileSize, pagesize );
1448 m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize );
1450 /* If we have a half-page-size file and map one page of it then
1451 * the part of the page after the size of the file remains accessible.
1452 * If, however, we map in 2 pages, the 2nd page is not accessible
1453 * and will give a "Bus Error" on access. To get around this, we check
1454 * if we need any extra pages for the jump islands and map them in
1455 * anonymously. We must check that we actually require extra pages
1456 * otherwise the attempt to mmap 0 pages of anonymous memory will
1462 /* The effect of this mremap() call is only the ensure that we have
1463 * a sufficient number of virtually contiguous pages. As returned from
1464 * mremap, the pages past the end of the file are not backed. We give
1465 * them a backing by using MAP_FIXED to map in anonymous pages.
1467 oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE );
1469 if( oc->image == MAP_FAILED )
1471 errorBelch( "Unable to mremap for Jump Islands\n" );
1475 if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC,
1476 MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED )
1478 errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" );
1484 oc->image -= misalignment;
1485 oc->image = stgReallocBytes( oc->image,
1487 aligned + sizeof (ppcJumpIsland) * count,
1488 "ocAllocateJumpIslands" );
1489 oc->image += misalignment;
1490 #endif /* USE_MMAP */
1492 oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned);
1493 memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count );
1496 oc->jump_islands = NULL;
1498 oc->island_start_symbol = first;
1499 oc->n_islands = count;
1504 static unsigned long makeJumpIsland( ObjectCode* oc,
1505 unsigned long symbolNumber,
1506 unsigned long target )
1508 ppcJumpIsland *island;
1510 if( symbolNumber < oc->island_start_symbol ||
1511 symbolNumber - oc->island_start_symbol > oc->n_islands)
1514 island = &oc->jump_islands[symbolNumber - oc->island_start_symbol];
1516 // lis r12, hi16(target)
1517 island->lis_r12 = 0x3d80;
1518 island->hi_addr = target >> 16;
1520 // ori r12, r12, lo16(target)
1521 island->ori_r12_r12 = 0x618c;
1522 island->lo_addr = target & 0xffff;
1525 island->mtctr_r12 = 0x7d8903a6;
1528 island->bctr = 0x4e800420;
1530 return (unsigned long) island;
1534 ocFlushInstructionCache
1536 Flush the data & instruction caches.
1537 Because the PPC has split data/instruction caches, we have to
1538 do that whenever we modify code at runtime.
1541 static void ocFlushInstructionCache( ObjectCode *oc )
1543 int n = (oc->fileSize + sizeof( ppcJumpIsland ) * oc->n_islands + 3) / 4;
1544 unsigned long *p = (unsigned long *) oc->image;
1548 __asm__ volatile ( "dcbf 0,%0\n\t"
1556 __asm__ volatile ( "sync\n\t"
1562 /* --------------------------------------------------------------------------
1563 * PEi386 specifics (Win32 targets)
1564 * ------------------------------------------------------------------------*/
1566 /* The information for this linker comes from
1567 Microsoft Portable Executable
1568 and Common Object File Format Specification
1569 revision 5.1 January 1998
1570 which SimonM says comes from the MS Developer Network CDs.
1572 It can be found there (on older CDs), but can also be found
1575 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1577 (this is Rev 6.0 from February 1999).
1579 Things move, so if that fails, try searching for it via
1581 http://www.google.com/search?q=PE+COFF+specification
1583 The ultimate reference for the PE format is the Winnt.h
1584 header file that comes with the Platform SDKs; as always,
1585 implementations will drift wrt their documentation.
1587 A good background article on the PE format is Matt Pietrek's
1588 March 1994 article in Microsoft System Journal (MSJ)
1589 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1590 Win32 Portable Executable File Format." The info in there
1591 has recently been updated in a two part article in
1592 MSDN magazine, issues Feb and March 2002,
1593 "Inside Windows: An In-Depth Look into the Win32 Portable
1594 Executable File Format"
1596 John Levine's book "Linkers and Loaders" contains useful
1601 #if defined(OBJFORMAT_PEi386)
1605 typedef unsigned char UChar;
1606 typedef unsigned short UInt16;
1607 typedef unsigned int UInt32;
1614 UInt16 NumberOfSections;
1615 UInt32 TimeDateStamp;
1616 UInt32 PointerToSymbolTable;
1617 UInt32 NumberOfSymbols;
1618 UInt16 SizeOfOptionalHeader;
1619 UInt16 Characteristics;
1623 #define sizeof_COFF_header 20
1630 UInt32 VirtualAddress;
1631 UInt32 SizeOfRawData;
1632 UInt32 PointerToRawData;
1633 UInt32 PointerToRelocations;
1634 UInt32 PointerToLinenumbers;
1635 UInt16 NumberOfRelocations;
1636 UInt16 NumberOfLineNumbers;
1637 UInt32 Characteristics;
1641 #define sizeof_COFF_section 40
1648 UInt16 SectionNumber;
1651 UChar NumberOfAuxSymbols;
1655 #define sizeof_COFF_symbol 18
1660 UInt32 VirtualAddress;
1661 UInt32 SymbolTableIndex;
1666 #define sizeof_COFF_reloc 10
1669 /* From PE spec doc, section 3.3.2 */
1670 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1671 windows.h -- for the same purpose, but I want to know what I'm
1673 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1674 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1675 #define MYIMAGE_FILE_DLL 0x2000
1676 #define MYIMAGE_FILE_SYSTEM 0x1000
1677 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1678 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1679 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1681 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1682 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1683 #define MYIMAGE_SYM_CLASS_STATIC 3
1684 #define MYIMAGE_SYM_UNDEFINED 0
1686 /* From PE spec doc, section 4.1 */
1687 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1688 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1689 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1691 /* From PE spec doc, section 5.2.1 */
1692 #define MYIMAGE_REL_I386_DIR32 0x0006
1693 #define MYIMAGE_REL_I386_REL32 0x0014
1696 /* We use myindex to calculate array addresses, rather than
1697 simply doing the normal subscript thing. That's because
1698 some of the above structs have sizes which are not
1699 a whole number of words. GCC rounds their sizes up to a
1700 whole number of words, which means that the address calcs
1701 arising from using normal C indexing or pointer arithmetic
1702 are just plain wrong. Sigh.
1705 myindex ( int scale, void* base, int index )
1708 ((UChar*)base) + scale * index;
1713 printName ( UChar* name, UChar* strtab )
1715 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1716 UInt32 strtab_offset = * (UInt32*)(name+4);
1717 debugBelch("%s", strtab + strtab_offset );
1720 for (i = 0; i < 8; i++) {
1721 if (name[i] == 0) break;
1722 debugBelch("%c", name[i] );
1729 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1731 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1732 UInt32 strtab_offset = * (UInt32*)(name+4);
1733 strncpy ( dst, strtab+strtab_offset, dstSize );
1739 if (name[i] == 0) break;
1749 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
1752 /* If the string is longer than 8 bytes, look in the
1753 string table for it -- this will be correctly zero terminated.
1755 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1756 UInt32 strtab_offset = * (UInt32*)(name+4);
1757 return ((UChar*)strtab) + strtab_offset;
1759 /* Otherwise, if shorter than 8 bytes, return the original,
1760 which by defn is correctly terminated.
1762 if (name[7]==0) return name;
1763 /* The annoying case: 8 bytes. Copy into a temporary
1764 (which is never freed ...)
1766 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
1768 strncpy(newstr,name,8);
1774 /* Just compares the short names (first 8 chars) */
1775 static COFF_section *
1776 findPEi386SectionCalled ( ObjectCode* oc, char* name )
1780 = (COFF_header*)(oc->image);
1781 COFF_section* sectab
1783 ((UChar*)(oc->image))
1784 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1786 for (i = 0; i < hdr->NumberOfSections; i++) {
1789 COFF_section* section_i
1791 myindex ( sizeof_COFF_section, sectab, i );
1792 n1 = (UChar*) &(section_i->Name);
1794 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
1795 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
1796 n1[6]==n2[6] && n1[7]==n2[7])
1805 zapTrailingAtSign ( UChar* sym )
1807 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
1809 if (sym[0] == 0) return;
1811 while (sym[i] != 0) i++;
1814 while (j > 0 && my_isdigit(sym[j])) j--;
1815 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
1821 ocVerifyImage_PEi386 ( ObjectCode* oc )
1826 COFF_section* sectab;
1827 COFF_symbol* symtab;
1829 /* debugBelch("\nLOADING %s\n", oc->fileName); */
1830 hdr = (COFF_header*)(oc->image);
1831 sectab = (COFF_section*) (
1832 ((UChar*)(oc->image))
1833 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
1835 symtab = (COFF_symbol*) (
1836 ((UChar*)(oc->image))
1837 + hdr->PointerToSymbolTable
1839 strtab = ((UChar*)symtab)
1840 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
1842 if (hdr->Machine != 0x14c) {
1843 errorBelch("%s: Not x86 PEi386", oc->fileName);
1846 if (hdr->SizeOfOptionalHeader != 0) {
1847 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
1850 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
1851 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
1852 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
1853 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
1854 errorBelch("%s: Not a PEi386 object file", oc->fileName);
1857 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
1858 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
1859 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
1861 (int)(hdr->Characteristics));
1864 /* If the string table size is way crazy, this might indicate that
1865 there are more than 64k relocations, despite claims to the
1866 contrary. Hence this test. */
1867 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
1869 if ( (*(UInt32*)strtab) > 600000 ) {
1870 /* Note that 600k has no special significance other than being
1871 big enough to handle the almost-2MB-sized lumps that
1872 constitute HSwin32*.o. */
1873 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
1878 /* No further verification after this point; only debug printing. */
1880 IF_DEBUG(linker, i=1);
1881 if (i == 0) return 1;
1883 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
1884 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
1885 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
1888 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
1889 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
1890 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
1891 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
1892 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
1893 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
1894 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
1896 /* Print the section table. */
1898 for (i = 0; i < hdr->NumberOfSections; i++) {
1900 COFF_section* sectab_i
1902 myindex ( sizeof_COFF_section, sectab, i );
1909 printName ( sectab_i->Name, strtab );
1919 sectab_i->VirtualSize,
1920 sectab_i->VirtualAddress,
1921 sectab_i->SizeOfRawData,
1922 sectab_i->PointerToRawData,
1923 sectab_i->NumberOfRelocations,
1924 sectab_i->PointerToRelocations,
1925 sectab_i->PointerToRawData
1927 reltab = (COFF_reloc*) (
1928 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
1931 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
1932 /* If the relocation field (a short) has overflowed, the
1933 * real count can be found in the first reloc entry.
1935 * See Section 4.1 (last para) of the PE spec (rev6.0).
1937 COFF_reloc* rel = (COFF_reloc*)
1938 myindex ( sizeof_COFF_reloc, reltab, 0 );
1939 noRelocs = rel->VirtualAddress;
1942 noRelocs = sectab_i->NumberOfRelocations;
1946 for (; j < noRelocs; j++) {
1948 COFF_reloc* rel = (COFF_reloc*)
1949 myindex ( sizeof_COFF_reloc, reltab, j );
1951 " type 0x%-4x vaddr 0x%-8x name `",
1953 rel->VirtualAddress );
1954 sym = (COFF_symbol*)
1955 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
1956 /* Hmm..mysterious looking offset - what's it for? SOF */
1957 printName ( sym->Name, strtab -10 );
1964 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
1965 debugBelch("---START of string table---\n");
1966 for (i = 4; i < *(Int32*)strtab; i++) {
1968 debugBelch("\n"); else
1969 debugBelch("%c", strtab[i] );
1971 debugBelch("--- END of string table---\n");
1976 COFF_symbol* symtab_i;
1977 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
1978 symtab_i = (COFF_symbol*)
1979 myindex ( sizeof_COFF_symbol, symtab, i );
1985 printName ( symtab_i->Name, strtab );
1994 (Int32)(symtab_i->SectionNumber),
1995 (UInt32)symtab_i->Type,
1996 (UInt32)symtab_i->StorageClass,
1997 (UInt32)symtab_i->NumberOfAuxSymbols
1999 i += symtab_i->NumberOfAuxSymbols;
2009 ocGetNames_PEi386 ( ObjectCode* oc )
2012 COFF_section* sectab;
2013 COFF_symbol* symtab;
2020 hdr = (COFF_header*)(oc->image);
2021 sectab = (COFF_section*) (
2022 ((UChar*)(oc->image))
2023 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2025 symtab = (COFF_symbol*) (
2026 ((UChar*)(oc->image))
2027 + hdr->PointerToSymbolTable
2029 strtab = ((UChar*)(oc->image))
2030 + hdr->PointerToSymbolTable
2031 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2033 /* Allocate space for any (local, anonymous) .bss sections. */
2035 for (i = 0; i < hdr->NumberOfSections; i++) {
2038 COFF_section* sectab_i
2040 myindex ( sizeof_COFF_section, sectab, i );
2041 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2042 /* sof 10/05: the PE spec text isn't too clear regarding what
2043 * the SizeOfRawData field is supposed to hold for object
2044 * file sections containing just uninitialized data -- for executables,
2045 * it is supposed to be zero; unclear what it's supposed to be
2046 * for object files. However, VirtualSize is guaranteed to be
2047 * zero for object files, which definitely suggests that SizeOfRawData
2048 * will be non-zero (where else would the size of this .bss section be
2049 * stored?) Looking at the COFF_section info for incoming object files,
2050 * this certainly appears to be the case.
2052 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2053 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2054 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2055 * variable decls into to the .bss section. (The specific function in Q which
2056 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2058 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2059 /* This is a non-empty .bss section. Allocate zeroed space for
2060 it, and set its PointerToRawData field such that oc->image +
2061 PointerToRawData == addr_of_zeroed_space. */
2062 bss_sz = sectab_i->VirtualSize;
2063 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2064 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2065 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2066 addProddableBlock(oc, zspace, bss_sz);
2067 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2070 /* Copy section information into the ObjectCode. */
2072 for (i = 0; i < hdr->NumberOfSections; i++) {
2078 = SECTIONKIND_OTHER;
2079 COFF_section* sectab_i
2081 myindex ( sizeof_COFF_section, sectab, i );
2082 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2085 /* I'm sure this is the Right Way to do it. However, the
2086 alternative of testing the sectab_i->Name field seems to
2087 work ok with Cygwin.
2089 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2090 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2091 kind = SECTIONKIND_CODE_OR_RODATA;
2094 if (0==strcmp(".text",sectab_i->Name) ||
2095 0==strcmp(".rdata",sectab_i->Name)||
2096 0==strcmp(".rodata",sectab_i->Name))
2097 kind = SECTIONKIND_CODE_OR_RODATA;
2098 if (0==strcmp(".data",sectab_i->Name) ||
2099 0==strcmp(".bss",sectab_i->Name))
2100 kind = SECTIONKIND_RWDATA;
2102 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2103 sz = sectab_i->SizeOfRawData;
2104 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2106 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2107 end = start + sz - 1;
2109 if (kind == SECTIONKIND_OTHER
2110 /* Ignore sections called which contain stabs debugging
2112 && 0 != strcmp(".stab", sectab_i->Name)
2113 && 0 != strcmp(".stabstr", sectab_i->Name)
2114 /* ignore constructor section for now */
2115 && 0 != strcmp(".ctors", sectab_i->Name)
2117 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2121 if (kind != SECTIONKIND_OTHER && end >= start) {
2122 addSection(oc, kind, start, end);
2123 addProddableBlock(oc, start, end - start + 1);
2127 /* Copy exported symbols into the ObjectCode. */
2129 oc->n_symbols = hdr->NumberOfSymbols;
2130 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2131 "ocGetNames_PEi386(oc->symbols)");
2132 /* Call me paranoid; I don't care. */
2133 for (i = 0; i < oc->n_symbols; i++)
2134 oc->symbols[i] = NULL;
2138 COFF_symbol* symtab_i;
2139 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2140 symtab_i = (COFF_symbol*)
2141 myindex ( sizeof_COFF_symbol, symtab, i );
2145 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2146 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2147 /* This symbol is global and defined, viz, exported */
2148 /* for MYIMAGE_SYMCLASS_EXTERNAL
2149 && !MYIMAGE_SYM_UNDEFINED,
2150 the address of the symbol is:
2151 address of relevant section + offset in section
2153 COFF_section* sectabent
2154 = (COFF_section*) myindex ( sizeof_COFF_section,
2156 symtab_i->SectionNumber-1 );
2157 addr = ((UChar*)(oc->image))
2158 + (sectabent->PointerToRawData
2162 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2163 && symtab_i->Value > 0) {
2164 /* This symbol isn't in any section at all, ie, global bss.
2165 Allocate zeroed space for it. */
2166 addr = stgCallocBytes(1, symtab_i->Value,
2167 "ocGetNames_PEi386(non-anonymous bss)");
2168 addSection(oc, SECTIONKIND_RWDATA, addr,
2169 ((UChar*)addr) + symtab_i->Value - 1);
2170 addProddableBlock(oc, addr, symtab_i->Value);
2171 /* debugBelch("BSS section at 0x%x\n", addr); */
2174 if (addr != NULL ) {
2175 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2176 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2177 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2178 ASSERT(i >= 0 && i < oc->n_symbols);
2179 /* cstring_from_COFF_symbol_name always succeeds. */
2180 oc->symbols[i] = sname;
2181 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2185 "IGNORING symbol %d\n"
2189 printName ( symtab_i->Name, strtab );
2198 (Int32)(symtab_i->SectionNumber),
2199 (UInt32)symtab_i->Type,
2200 (UInt32)symtab_i->StorageClass,
2201 (UInt32)symtab_i->NumberOfAuxSymbols
2206 i += symtab_i->NumberOfAuxSymbols;
2215 ocResolve_PEi386 ( ObjectCode* oc )
2218 COFF_section* sectab;
2219 COFF_symbol* symtab;
2229 /* ToDo: should be variable-sized? But is at least safe in the
2230 sense of buffer-overrun-proof. */
2232 /* debugBelch("resolving for %s\n", oc->fileName); */
2234 hdr = (COFF_header*)(oc->image);
2235 sectab = (COFF_section*) (
2236 ((UChar*)(oc->image))
2237 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2239 symtab = (COFF_symbol*) (
2240 ((UChar*)(oc->image))
2241 + hdr->PointerToSymbolTable
2243 strtab = ((UChar*)(oc->image))
2244 + hdr->PointerToSymbolTable
2245 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2247 for (i = 0; i < hdr->NumberOfSections; i++) {
2248 COFF_section* sectab_i
2250 myindex ( sizeof_COFF_section, sectab, i );
2253 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2256 /* Ignore sections called which contain stabs debugging
2258 if (0 == strcmp(".stab", sectab_i->Name)
2259 || 0 == strcmp(".stabstr", sectab_i->Name)
2260 || 0 == strcmp(".ctors", sectab_i->Name))
2263 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2264 /* If the relocation field (a short) has overflowed, the
2265 * real count can be found in the first reloc entry.
2267 * See Section 4.1 (last para) of the PE spec (rev6.0).
2269 * Nov2003 update: the GNU linker still doesn't correctly
2270 * handle the generation of relocatable object files with
2271 * overflown relocations. Hence the output to warn of potential
2274 COFF_reloc* rel = (COFF_reloc*)
2275 myindex ( sizeof_COFF_reloc, reltab, 0 );
2276 noRelocs = rel->VirtualAddress;
2278 /* 10/05: we now assume (and check for) a GNU ld that is capable
2279 * of handling object files with (>2^16) of relocs.
2282 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2287 noRelocs = sectab_i->NumberOfRelocations;
2292 for (; j < noRelocs; j++) {
2294 COFF_reloc* reltab_j
2296 myindex ( sizeof_COFF_reloc, reltab, j );
2298 /* the location to patch */
2300 ((UChar*)(oc->image))
2301 + (sectab_i->PointerToRawData
2302 + reltab_j->VirtualAddress
2303 - sectab_i->VirtualAddress )
2305 /* the existing contents of pP */
2307 /* the symbol to connect to */
2308 sym = (COFF_symbol*)
2309 myindex ( sizeof_COFF_symbol,
2310 symtab, reltab_j->SymbolTableIndex );
2313 "reloc sec %2d num %3d: type 0x%-4x "
2314 "vaddr 0x%-8x name `",
2316 (UInt32)reltab_j->Type,
2317 reltab_j->VirtualAddress );
2318 printName ( sym->Name, strtab );
2319 debugBelch("'\n" ));
2321 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2322 COFF_section* section_sym
2323 = findPEi386SectionCalled ( oc, sym->Name );
2325 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2328 S = ((UInt32)(oc->image))
2329 + (section_sym->PointerToRawData
2332 copyName ( sym->Name, strtab, symbol, 1000-1 );
2333 (void*)S = lookupLocalSymbol( oc, symbol );
2334 if ((void*)S != NULL) goto foundit;
2335 (void*)S = lookupSymbol( symbol );
2336 if ((void*)S != NULL) goto foundit;
2337 zapTrailingAtSign ( symbol );
2338 (void*)S = lookupLocalSymbol( oc, symbol );
2339 if ((void*)S != NULL) goto foundit;
2340 (void*)S = lookupSymbol( symbol );
2341 if ((void*)S != NULL) goto foundit;
2342 /* Newline first because the interactive linker has printed "linking..." */
2343 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2347 checkProddableBlock(oc, pP);
2348 switch (reltab_j->Type) {
2349 case MYIMAGE_REL_I386_DIR32:
2352 case MYIMAGE_REL_I386_REL32:
2353 /* Tricky. We have to insert a displacement at
2354 pP which, when added to the PC for the _next_
2355 insn, gives the address of the target (S).
2356 Problem is to know the address of the next insn
2357 when we only know pP. We assume that this
2358 literal field is always the last in the insn,
2359 so that the address of the next insn is pP+4
2360 -- hence the constant 4.
2361 Also I don't know if A should be added, but so
2362 far it has always been zero.
2364 SOF 05/2005: 'A' (old contents of *pP) have been observed
2365 to contain values other than zero (the 'wx' object file
2366 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2367 So, add displacement to old value instead of asserting
2368 A to be zero. Fixes wxhaskell-related crashes, and no other
2369 ill effects have been observed.
2371 Update: the reason why we're seeing these more elaborate
2372 relocations is due to a switch in how the NCG compiles SRTs
2373 and offsets to them from info tables. SRTs live in .(ro)data,
2374 while info tables live in .text, causing GAS to emit REL32/DISP32
2375 relocations with non-zero values. Adding the displacement is
2376 the right thing to do.
2378 *pP = S - ((UInt32)pP) - 4 + A;
2381 debugBelch("%s: unhandled PEi386 relocation type %d",
2382 oc->fileName, reltab_j->Type);
2389 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2393 #endif /* defined(OBJFORMAT_PEi386) */
2396 /* --------------------------------------------------------------------------
2398 * ------------------------------------------------------------------------*/
2400 #if defined(OBJFORMAT_ELF)
2405 #if defined(sparc_HOST_ARCH)
2406 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2407 #elif defined(i386_HOST_ARCH)
2408 # define ELF_TARGET_386 /* Used inside <elf.h> */
2409 #elif defined(x86_64_HOST_ARCH)
2410 # define ELF_TARGET_X64_64
2412 #elif defined (ia64_HOST_ARCH)
2413 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2415 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2416 # define ELF_NEED_GOT /* needs Global Offset Table */
2417 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2420 #if !defined(openbsd_HOST_OS)
2423 /* openbsd elf has things in different places, with diff names */
2424 #include <elf_abi.h>
2425 #include <machine/reloc.h>
2426 #define R_386_32 RELOC_32
2427 #define R_386_PC32 RELOC_PC32
2431 * Define a set of types which can be used for both ELF32 and ELF64
2435 #define ELFCLASS ELFCLASS64
2436 #define Elf_Addr Elf64_Addr
2437 #define Elf_Word Elf64_Word
2438 #define Elf_Sword Elf64_Sword
2439 #define Elf_Ehdr Elf64_Ehdr
2440 #define Elf_Phdr Elf64_Phdr
2441 #define Elf_Shdr Elf64_Shdr
2442 #define Elf_Sym Elf64_Sym
2443 #define Elf_Rel Elf64_Rel
2444 #define Elf_Rela Elf64_Rela
2445 #define ELF_ST_TYPE ELF64_ST_TYPE
2446 #define ELF_ST_BIND ELF64_ST_BIND
2447 #define ELF_R_TYPE ELF64_R_TYPE
2448 #define ELF_R_SYM ELF64_R_SYM
2450 #define ELFCLASS ELFCLASS32
2451 #define Elf_Addr Elf32_Addr
2452 #define Elf_Word Elf32_Word
2453 #define Elf_Sword Elf32_Sword
2454 #define Elf_Ehdr Elf32_Ehdr
2455 #define Elf_Phdr Elf32_Phdr
2456 #define Elf_Shdr Elf32_Shdr
2457 #define Elf_Sym Elf32_Sym
2458 #define Elf_Rel Elf32_Rel
2459 #define Elf_Rela Elf32_Rela
2461 #define ELF_ST_TYPE ELF32_ST_TYPE
2464 #define ELF_ST_BIND ELF32_ST_BIND
2467 #define ELF_R_TYPE ELF32_R_TYPE
2470 #define ELF_R_SYM ELF32_R_SYM
2476 * Functions to allocate entries in dynamic sections. Currently we simply
2477 * preallocate a large number, and we don't check if a entry for the given
2478 * target already exists (a linear search is too slow). Ideally these
2479 * entries would be associated with symbols.
2482 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2483 #define GOT_SIZE 0x20000
2484 #define FUNCTION_TABLE_SIZE 0x10000
2485 #define PLT_SIZE 0x08000
2488 static Elf_Addr got[GOT_SIZE];
2489 static unsigned int gotIndex;
2490 static Elf_Addr gp_val = (Elf_Addr)got;
2493 allocateGOTEntry(Elf_Addr target)
2497 if (gotIndex >= GOT_SIZE)
2498 barf("Global offset table overflow");
2500 entry = &got[gotIndex++];
2502 return (Elf_Addr)entry;
2506 #ifdef ELF_FUNCTION_DESC
2512 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2513 static unsigned int functionTableIndex;
2516 allocateFunctionDesc(Elf_Addr target)
2518 FunctionDesc *entry;
2520 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2521 barf("Function table overflow");
2523 entry = &functionTable[functionTableIndex++];
2525 entry->gp = (Elf_Addr)gp_val;
2526 return (Elf_Addr)entry;
2530 copyFunctionDesc(Elf_Addr target)
2532 FunctionDesc *olddesc = (FunctionDesc *)target;
2533 FunctionDesc *newdesc;
2535 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2536 newdesc->gp = olddesc->gp;
2537 return (Elf_Addr)newdesc;
2542 #ifdef ia64_HOST_ARCH
2543 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2544 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2546 static unsigned char plt_code[] =
2548 /* taken from binutils bfd/elfxx-ia64.c */
2549 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2550 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2551 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2552 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2553 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2554 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2557 /* If we can't get to the function descriptor via gp, take a local copy of it */
2558 #define PLT_RELOC(code, target) { \
2559 Elf64_Sxword rel_value = target - gp_val; \
2560 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2561 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2563 ia64_reloc_gprel22((Elf_Addr)code, target); \
2568 unsigned char code[sizeof(plt_code)];
2572 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2574 PLTEntry *plt = (PLTEntry *)oc->plt;
2577 if (oc->pltIndex >= PLT_SIZE)
2578 barf("Procedure table overflow");
2580 entry = &plt[oc->pltIndex++];
2581 memcpy(entry->code, plt_code, sizeof(entry->code));
2582 PLT_RELOC(entry->code, target);
2583 return (Elf_Addr)entry;
2589 return (PLT_SIZE * sizeof(PLTEntry));
2594 #if x86_64_HOST_ARCH
2595 // On x86_64, 32-bit relocations are often used, which requires that
2596 // we can resolve a symbol to a 32-bit offset. However, shared
2597 // libraries are placed outside the 2Gb area, which leaves us with a
2598 // problem when we need to give a 32-bit offset to a symbol in a
2601 // For a function symbol, we can allocate a bounce sequence inside the
2602 // 2Gb area and resolve the symbol to this. The bounce sequence is
2603 // simply a long jump instruction to the real location of the symbol.
2605 // For data references, we're screwed.
2608 unsigned char jmp[8]; /* 6 byte instruction: jmpq *0x00000002(%rip) */
2612 #define X86_64_BB_SIZE 1024
2614 static x86_64_bounce *x86_64_bounce_buffer = NULL;
2615 static nat x86_64_bb_next_off;
2618 x86_64_high_symbol( char *lbl, void *addr )
2620 x86_64_bounce *bounce;
2622 if ( x86_64_bounce_buffer == NULL ||
2623 x86_64_bb_next_off >= X86_64_BB_SIZE ) {
2624 x86_64_bounce_buffer =
2625 mmap(NULL, X86_64_BB_SIZE * sizeof(x86_64_bounce),
2626 PROT_EXEC|PROT_READ|PROT_WRITE,
2627 MAP_PRIVATE|MAP_32BIT|MAP_ANONYMOUS, -1, 0);
2628 if (x86_64_bounce_buffer == MAP_FAILED) {
2629 barf("x86_64_high_symbol: mmap failed");
2631 x86_64_bb_next_off = 0;
2633 bounce = &x86_64_bounce_buffer[x86_64_bb_next_off];
2634 bounce->jmp[0] = 0xff;
2635 bounce->jmp[1] = 0x25;
2636 bounce->jmp[2] = 0x02;
2637 bounce->jmp[3] = 0x00;
2638 bounce->jmp[4] = 0x00;
2639 bounce->jmp[5] = 0x00;
2640 bounce->addr = addr;
2641 x86_64_bb_next_off++;
2643 IF_DEBUG(linker, debugBelch("x86_64: allocated bounce entry for %s->%p at %p\n",
2644 lbl, addr, bounce));
2646 insertStrHashTable(symhash, lbl, bounce);
2653 * Generic ELF functions
2657 findElfSection ( void* objImage, Elf_Word sh_type )
2659 char* ehdrC = (char*)objImage;
2660 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2661 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2662 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2666 for (i = 0; i < ehdr->e_shnum; i++) {
2667 if (shdr[i].sh_type == sh_type
2668 /* Ignore the section header's string table. */
2669 && i != ehdr->e_shstrndx
2670 /* Ignore string tables named .stabstr, as they contain
2672 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2674 ptr = ehdrC + shdr[i].sh_offset;
2681 #if defined(ia64_HOST_ARCH)
2683 findElfSegment ( void* objImage, Elf_Addr vaddr )
2685 char* ehdrC = (char*)objImage;
2686 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2687 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2688 Elf_Addr segaddr = 0;
2691 for (i = 0; i < ehdr->e_phnum; i++) {
2692 segaddr = phdr[i].p_vaddr;
2693 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2701 ocVerifyImage_ELF ( ObjectCode* oc )
2705 int i, j, nent, nstrtab, nsymtabs;
2709 char* ehdrC = (char*)(oc->image);
2710 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2712 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2713 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2714 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2715 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2716 errorBelch("%s: not an ELF object", oc->fileName);
2720 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2721 errorBelch("%s: unsupported ELF format", oc->fileName);
2725 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2726 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2728 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2729 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2731 errorBelch("%s: unknown endiannness", oc->fileName);
2735 if (ehdr->e_type != ET_REL) {
2736 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2739 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2741 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2742 switch (ehdr->e_machine) {
2743 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2744 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2746 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2748 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2750 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2752 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2753 errorBelch("%s: unknown architecture", oc->fileName);
2757 IF_DEBUG(linker,debugBelch(
2758 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
2759 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2761 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2763 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2765 if (ehdr->e_shstrndx == SHN_UNDEF) {
2766 errorBelch("%s: no section header string table", oc->fileName);
2769 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
2771 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2774 for (i = 0; i < ehdr->e_shnum; i++) {
2775 IF_DEBUG(linker,debugBelch("%2d: ", i ));
2776 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
2777 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
2778 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
2779 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
2780 ehdrC + shdr[i].sh_offset,
2781 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
2783 if (shdr[i].sh_type == SHT_REL) {
2784 IF_DEBUG(linker,debugBelch("Rel " ));
2785 } else if (shdr[i].sh_type == SHT_RELA) {
2786 IF_DEBUG(linker,debugBelch("RelA " ));
2788 IF_DEBUG(linker,debugBelch(" "));
2791 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
2795 IF_DEBUG(linker,debugBelch( "\nString tables" ));
2798 for (i = 0; i < ehdr->e_shnum; i++) {
2799 if (shdr[i].sh_type == SHT_STRTAB
2800 /* Ignore the section header's string table. */
2801 && i != ehdr->e_shstrndx
2802 /* Ignore string tables named .stabstr, as they contain
2804 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2806 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
2807 strtab = ehdrC + shdr[i].sh_offset;
2812 errorBelch("%s: no string tables, or too many", oc->fileName);
2817 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
2818 for (i = 0; i < ehdr->e_shnum; i++) {
2819 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2820 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
2822 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2823 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2824 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
2826 (long)shdr[i].sh_size % sizeof(Elf_Sym)
2828 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
2829 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
2832 for (j = 0; j < nent; j++) {
2833 IF_DEBUG(linker,debugBelch(" %2d ", j ));
2834 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
2835 (int)stab[j].st_shndx,
2836 (int)stab[j].st_size,
2837 (char*)stab[j].st_value ));
2839 IF_DEBUG(linker,debugBelch("type=" ));
2840 switch (ELF_ST_TYPE(stab[j].st_info)) {
2841 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
2842 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
2843 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
2844 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
2845 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
2846 default: IF_DEBUG(linker,debugBelch("? " )); break;
2848 IF_DEBUG(linker,debugBelch(" " ));
2850 IF_DEBUG(linker,debugBelch("bind=" ));
2851 switch (ELF_ST_BIND(stab[j].st_info)) {
2852 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
2853 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
2854 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
2855 default: IF_DEBUG(linker,debugBelch("? " )); break;
2857 IF_DEBUG(linker,debugBelch(" " ));
2859 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
2863 if (nsymtabs == 0) {
2864 errorBelch("%s: didn't find any symbol tables", oc->fileName);
2871 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
2875 if (hdr->sh_type == SHT_PROGBITS
2876 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
2877 /* .text-style section */
2878 return SECTIONKIND_CODE_OR_RODATA;
2881 if (hdr->sh_type == SHT_PROGBITS
2882 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2883 /* .data-style section */
2884 return SECTIONKIND_RWDATA;
2887 if (hdr->sh_type == SHT_PROGBITS
2888 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
2889 /* .rodata-style section */
2890 return SECTIONKIND_CODE_OR_RODATA;
2893 if (hdr->sh_type == SHT_NOBITS
2894 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
2895 /* .bss-style section */
2897 return SECTIONKIND_RWDATA;
2900 return SECTIONKIND_OTHER;
2905 ocGetNames_ELF ( ObjectCode* oc )
2910 char* ehdrC = (char*)(oc->image);
2911 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2912 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
2913 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2915 ASSERT(symhash != NULL);
2918 errorBelch("%s: no strtab", oc->fileName);
2923 for (i = 0; i < ehdr->e_shnum; i++) {
2924 /* Figure out what kind of section it is. Logic derived from
2925 Figure 1.14 ("Special Sections") of the ELF document
2926 ("Portable Formats Specification, Version 1.1"). */
2928 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
2930 if (is_bss && shdr[i].sh_size > 0) {
2931 /* This is a non-empty .bss section. Allocate zeroed space for
2932 it, and set its .sh_offset field such that
2933 ehdrC + .sh_offset == addr_of_zeroed_space. */
2934 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
2935 "ocGetNames_ELF(BSS)");
2936 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
2938 debugBelch("BSS section at 0x%x, size %d\n",
2939 zspace, shdr[i].sh_size);
2943 /* fill in the section info */
2944 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
2945 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
2946 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
2947 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
2950 if (shdr[i].sh_type != SHT_SYMTAB) continue;
2952 /* copy stuff into this module's object symbol table */
2953 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
2954 nent = shdr[i].sh_size / sizeof(Elf_Sym);
2956 oc->n_symbols = nent;
2957 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2958 "ocGetNames_ELF(oc->symbols)");
2960 for (j = 0; j < nent; j++) {
2962 char isLocal = FALSE; /* avoids uninit-var warning */
2964 char* nm = strtab + stab[j].st_name;
2965 int secno = stab[j].st_shndx;
2967 /* Figure out if we want to add it; if so, set ad to its
2968 address. Otherwise leave ad == NULL. */
2970 if (secno == SHN_COMMON) {
2972 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
2974 debugBelch("COMMON symbol, size %d name %s\n",
2975 stab[j].st_size, nm);
2977 /* Pointless to do addProddableBlock() for this area,
2978 since the linker should never poke around in it. */
2981 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
2982 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
2984 /* and not an undefined symbol */
2985 && stab[j].st_shndx != SHN_UNDEF
2986 /* and not in a "special section" */
2987 && stab[j].st_shndx < SHN_LORESERVE
2989 /* and it's a not a section or string table or anything silly */
2990 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
2991 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
2992 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
2995 /* Section 0 is the undefined section, hence > and not >=. */
2996 ASSERT(secno > 0 && secno < ehdr->e_shnum);
2998 if (shdr[secno].sh_type == SHT_NOBITS) {
2999 debugBelch(" BSS symbol, size %d off %d name %s\n",
3000 stab[j].st_size, stab[j].st_value, nm);
3003 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3004 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3007 #ifdef ELF_FUNCTION_DESC
3008 /* dlsym() and the initialisation table both give us function
3009 * descriptors, so to be consistent we store function descriptors
3010 * in the symbol table */
3011 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3012 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3014 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s",
3015 ad, oc->fileName, nm ));
3020 /* And the decision is ... */
3024 oc->symbols[j] = nm;
3027 /* Ignore entirely. */
3029 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3033 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3034 strtab + stab[j].st_name ));
3037 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3038 (int)ELF_ST_BIND(stab[j].st_info),
3039 (int)ELF_ST_TYPE(stab[j].st_info),
3040 (int)stab[j].st_shndx,
3041 strtab + stab[j].st_name
3044 oc->symbols[j] = NULL;
3053 /* Do ELF relocations which lack an explicit addend. All x86-linux
3054 relocations appear to be of this form. */
3056 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3057 Elf_Shdr* shdr, int shnum,
3058 Elf_Sym* stab, char* strtab )
3063 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3064 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3065 int target_shndx = shdr[shnum].sh_info;
3066 int symtab_shndx = shdr[shnum].sh_link;
3068 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3069 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3070 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3071 target_shndx, symtab_shndx ));
3073 /* Skip sections that we're not interested in. */
3076 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3077 if (kind == SECTIONKIND_OTHER) {
3078 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3083 for (j = 0; j < nent; j++) {
3084 Elf_Addr offset = rtab[j].r_offset;
3085 Elf_Addr info = rtab[j].r_info;
3087 Elf_Addr P = ((Elf_Addr)targ) + offset;
3088 Elf_Word* pP = (Elf_Word*)P;
3094 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3095 j, (void*)offset, (void*)info ));
3097 IF_DEBUG(linker,debugBelch( " ZERO" ));
3100 Elf_Sym sym = stab[ELF_R_SYM(info)];
3101 /* First see if it is a local symbol. */
3102 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3103 /* Yes, so we can get the address directly from the ELF symbol
3105 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3107 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3108 + stab[ELF_R_SYM(info)].st_value);
3111 /* No, so look up the name in our global table. */
3112 symbol = strtab + sym.st_name;
3113 S_tmp = lookupSymbol( symbol );
3114 S = (Elf_Addr)S_tmp;
3117 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3120 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3123 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3124 (void*)P, (void*)S, (void*)A ));
3125 checkProddableBlock ( oc, pP );
3129 switch (ELF_R_TYPE(info)) {
3130 # ifdef i386_HOST_ARCH
3131 case R_386_32: *pP = value; break;
3132 case R_386_PC32: *pP = value - P; break;
3135 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3136 oc->fileName, (lnat)ELF_R_TYPE(info));
3144 /* Do ELF relocations for which explicit addends are supplied.
3145 sparc-solaris relocations appear to be of this form. */
3147 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3148 Elf_Shdr* shdr, int shnum,
3149 Elf_Sym* stab, char* strtab )
3152 char *symbol = NULL;
3154 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3155 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3156 int target_shndx = shdr[shnum].sh_info;
3157 int symtab_shndx = shdr[shnum].sh_link;
3159 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3160 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3161 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3162 target_shndx, symtab_shndx ));
3164 for (j = 0; j < nent; j++) {
3165 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3166 /* This #ifdef only serves to avoid unused-var warnings. */
3167 Elf_Addr offset = rtab[j].r_offset;
3168 Elf_Addr P = targ + offset;
3170 Elf_Addr info = rtab[j].r_info;
3171 Elf_Addr A = rtab[j].r_addend;
3175 # if defined(sparc_HOST_ARCH)
3176 Elf_Word* pP = (Elf_Word*)P;
3178 # elif defined(ia64_HOST_ARCH)
3179 Elf64_Xword *pP = (Elf64_Xword *)P;
3181 # elif defined(powerpc_HOST_ARCH)
3185 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3186 j, (void*)offset, (void*)info,
3189 IF_DEBUG(linker,debugBelch( " ZERO" ));
3192 Elf_Sym sym = stab[ELF_R_SYM(info)];
3193 /* First see if it is a local symbol. */
3194 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3195 /* Yes, so we can get the address directly from the ELF symbol
3197 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3199 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3200 + stab[ELF_R_SYM(info)].st_value);
3201 #ifdef ELF_FUNCTION_DESC
3202 /* Make a function descriptor for this function */
3203 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3204 S = allocateFunctionDesc(S + A);
3209 /* No, so look up the name in our global table. */
3210 symbol = strtab + sym.st_name;
3211 S_tmp = lookupSymbol( symbol );
3212 S = (Elf_Addr)S_tmp;
3214 #ifdef ELF_FUNCTION_DESC
3215 /* If a function, already a function descriptor - we would
3216 have to copy it to add an offset. */
3217 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3218 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3222 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3225 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3228 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3229 (void*)P, (void*)S, (void*)A ));
3230 /* checkProddableBlock ( oc, (void*)P ); */
3234 switch (ELF_R_TYPE(info)) {
3235 # if defined(sparc_HOST_ARCH)
3236 case R_SPARC_WDISP30:
3237 w1 = *pP & 0xC0000000;
3238 w2 = (Elf_Word)((value - P) >> 2);
3239 ASSERT((w2 & 0xC0000000) == 0);
3244 w1 = *pP & 0xFFC00000;
3245 w2 = (Elf_Word)(value >> 10);
3246 ASSERT((w2 & 0xFFC00000) == 0);
3252 w2 = (Elf_Word)(value & 0x3FF);
3253 ASSERT((w2 & ~0x3FF) == 0);
3257 /* According to the Sun documentation:
3259 This relocation type resembles R_SPARC_32, except it refers to an
3260 unaligned word. That is, the word to be relocated must be treated
3261 as four separate bytes with arbitrary alignment, not as a word
3262 aligned according to the architecture requirements.
3264 (JRS: which means that freeloading on the R_SPARC_32 case
3265 is probably wrong, but hey ...)
3269 w2 = (Elf_Word)value;
3272 # elif defined(ia64_HOST_ARCH)
3273 case R_IA64_DIR64LSB:
3274 case R_IA64_FPTR64LSB:
3277 case R_IA64_PCREL64LSB:
3280 case R_IA64_SEGREL64LSB:
3281 addr = findElfSegment(ehdrC, value);
3284 case R_IA64_GPREL22:
3285 ia64_reloc_gprel22(P, value);
3287 case R_IA64_LTOFF22:
3288 case R_IA64_LTOFF22X:
3289 case R_IA64_LTOFF_FPTR22:
3290 addr = allocateGOTEntry(value);
3291 ia64_reloc_gprel22(P, addr);
3293 case R_IA64_PCREL21B:
3294 ia64_reloc_pcrel21(P, S, oc);
3297 /* This goes with R_IA64_LTOFF22X and points to the load to
3298 * convert into a move. We don't implement relaxation. */
3300 # elif defined(powerpc_HOST_ARCH)
3301 case R_PPC_ADDR16_LO:
3302 *(Elf32_Half*) P = value;
3305 case R_PPC_ADDR16_HI:
3306 *(Elf32_Half*) P = value >> 16;
3309 case R_PPC_ADDR16_HA:
3310 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3314 *(Elf32_Word *) P = value;
3318 *(Elf32_Word *) P = value - P;
3324 if( delta << 6 >> 6 != delta )
3326 value = makeJumpIsland( oc, ELF_R_SYM(info), value );
3329 if( value == 0 || delta << 6 >> 6 != delta )
3331 barf( "Unable to make ppcJumpIsland for #%d",
3337 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3338 | (delta & 0x3fffffc);
3342 #if x86_64_HOST_ARCH
3344 *(Elf64_Xword *)P = value;
3349 StgInt64 off = value - P;
3350 if (off >= 0x7fffffffL || off < -0x80000000L) {
3351 barf("R_X86_64_PC32 relocation out of range: %s = %p",
3354 *(Elf64_Word *)P = (Elf64_Word)off;
3359 if (value >= 0x7fffffffL) {
3360 barf("R_X86_64_32 relocation out of range: %s = %p\n",
3363 *(Elf64_Word *)P = (Elf64_Word)value;
3367 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3368 barf("R_X86_64_32S relocation out of range: %s = %p\n",
3371 *(Elf64_Sword *)P = (Elf64_Sword)value;
3376 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3377 oc->fileName, (lnat)ELF_R_TYPE(info));
3386 ocResolve_ELF ( ObjectCode* oc )
3390 Elf_Sym* stab = NULL;
3391 char* ehdrC = (char*)(oc->image);
3392 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3393 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3395 /* first find "the" symbol table */
3396 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3398 /* also go find the string table */
3399 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3401 if (stab == NULL || strtab == NULL) {
3402 errorBelch("%s: can't find string or symbol table", oc->fileName);
3406 /* Process the relocation sections. */
3407 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3408 if (shdr[shnum].sh_type == SHT_REL) {
3409 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3410 shnum, stab, strtab );
3414 if (shdr[shnum].sh_type == SHT_RELA) {
3415 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3416 shnum, stab, strtab );
3421 /* Free the local symbol table; we won't need it again. */
3422 freeHashTable(oc->lochash, NULL);
3425 #if defined(powerpc_HOST_ARCH)
3426 ocFlushInstructionCache( oc );
3434 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3435 * at the front. The following utility functions pack and unpack instructions, and
3436 * take care of the most common relocations.
3439 #ifdef ia64_HOST_ARCH
3442 ia64_extract_instruction(Elf64_Xword *target)
3445 int slot = (Elf_Addr)target & 3;
3446 target = (Elf_Addr)target & ~3;
3454 return ((w1 >> 5) & 0x1ffffffffff);
3456 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3460 barf("ia64_extract_instruction: invalid slot %p", target);
3465 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3467 int slot = (Elf_Addr)target & 3;
3468 target = (Elf_Addr)target & ~3;
3473 *target |= value << 5;
3476 *target |= value << 46;
3477 *(target+1) |= value >> 18;
3480 *(target+1) |= value << 23;
3486 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3488 Elf64_Xword instruction;
3489 Elf64_Sxword rel_value;
3491 rel_value = value - gp_val;
3492 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3493 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3495 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3496 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3497 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3498 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3499 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3500 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3504 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3506 Elf64_Xword instruction;
3507 Elf64_Sxword rel_value;
3510 entry = allocatePLTEntry(value, oc);
3512 rel_value = (entry >> 4) - (target >> 4);
3513 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3514 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3516 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3517 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3518 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3519 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3525 * PowerPC ELF specifics
3528 #ifdef powerpc_HOST_ARCH
3530 static int ocAllocateJumpIslands_ELF( ObjectCode *oc )
3536 ehdr = (Elf_Ehdr *) oc->image;
3537 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3539 for( i = 0; i < ehdr->e_shnum; i++ )
3540 if( shdr[i].sh_type == SHT_SYMTAB )
3543 if( i == ehdr->e_shnum )
3545 errorBelch( "This ELF file contains no symtab" );
3549 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3551 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3552 shdr[i].sh_entsize, sizeof( Elf_Sym ) );
3557 return ocAllocateJumpIslands( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3560 #endif /* powerpc */
3564 /* --------------------------------------------------------------------------
3566 * ------------------------------------------------------------------------*/
3568 #if defined(OBJFORMAT_MACHO)
3571 Support for MachO linking on Darwin/MacOS X
3572 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3574 I hereby formally apologize for the hackish nature of this code.
3575 Things that need to be done:
3576 *) implement ocVerifyImage_MachO
3577 *) add still more sanity checks.
3580 #ifdef powerpc_HOST_ARCH
3581 static int ocAllocateJumpIslands_MachO(ObjectCode* oc)
3583 struct mach_header *header = (struct mach_header *) oc->image;
3584 struct load_command *lc = (struct load_command *) (header + 1);
3587 for( i = 0; i < header->ncmds; i++ )
3589 if( lc->cmd == LC_SYMTAB )
3591 // Find out the first and last undefined external
3592 // symbol, so we don't have to allocate too many
3594 struct symtab_command *symLC = (struct symtab_command *) lc;
3595 unsigned min = symLC->nsyms, max = 0;
3596 struct nlist *nlist =
3597 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3599 for(i=0;i<symLC->nsyms;i++)
3601 if(nlist[i].n_type & N_STAB)
3603 else if(nlist[i].n_type & N_EXT)
3605 if((nlist[i].n_type & N_TYPE) == N_UNDF
3606 && (nlist[i].n_value == 0))
3616 return ocAllocateJumpIslands(oc, max - min + 1, min);
3621 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3623 return ocAllocateJumpIslands(oc,0,0);
3627 static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED)
3629 // FIXME: do some verifying here
3633 static int resolveImports(
3636 struct symtab_command *symLC,
3637 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3638 unsigned long *indirectSyms,
3639 struct nlist *nlist)
3643 for(i=0;i*4<sect->size;i++)
3645 // according to otool, reserved1 contains the first index into the indirect symbol table
3646 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3647 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3650 if((symbol->n_type & N_TYPE) == N_UNDF
3651 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3652 addr = (void*) (symbol->n_value);
3653 else if((addr = lookupLocalSymbol(oc,nm)) != NULL)
3656 addr = lookupSymbol(nm);
3659 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3663 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3664 ((void**)(image + sect->offset))[i] = addr;
3670 static unsigned long relocateAddress(
3673 struct section* sections,
3674 unsigned long address)
3677 for(i = 0; i < nSections; i++)
3679 if(sections[i].addr <= address
3680 && address < sections[i].addr + sections[i].size)
3682 return (unsigned long)oc->image
3683 + sections[i].offset + address - sections[i].addr;
3686 barf("Invalid Mach-O file:"
3687 "Address out of bounds while relocating object file");
3691 static int relocateSection(
3694 struct symtab_command *symLC, struct nlist *nlist,
3695 int nSections, struct section* sections, struct section *sect)
3697 struct relocation_info *relocs;
3700 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3702 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3704 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
3706 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
3710 relocs = (struct relocation_info*) (image + sect->reloff);
3714 if(relocs[i].r_address & R_SCATTERED)
3716 struct scattered_relocation_info *scat =
3717 (struct scattered_relocation_info*) &relocs[i];
3721 if(scat->r_length == 2)
3723 unsigned long word = 0;
3724 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
3725 checkProddableBlock(oc,wordPtr);
3727 // Note on relocation types:
3728 // i386 uses the GENERIC_RELOC_* types,
3729 // while ppc uses special PPC_RELOC_* types.
3730 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
3731 // in both cases, all others are different.
3732 // Therefore, we use GENERIC_RELOC_VANILLA
3733 // and GENERIC_RELOC_PAIR instead of the PPC variants,
3734 // and use #ifdefs for the other types.
3736 // Step 1: Figure out what the relocated value should be
3737 if(scat->r_type == GENERIC_RELOC_VANILLA)
3739 word = *wordPtr + (unsigned long) relocateAddress(
3746 #ifdef powerpc_HOST_ARCH
3747 else if(scat->r_type == PPC_RELOC_SECTDIFF
3748 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
3749 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
3750 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
3752 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
3755 struct scattered_relocation_info *pair =
3756 (struct scattered_relocation_info*) &relocs[i+1];
3758 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
3759 barf("Invalid Mach-O file: "
3760 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
3762 word = (unsigned long)
3763 (relocateAddress(oc, nSections, sections, scat->r_value)
3764 - relocateAddress(oc, nSections, sections, pair->r_value));
3767 #ifdef powerpc_HOST_ARCH
3768 else if(scat->r_type == PPC_RELOC_HI16
3769 || scat->r_type == PPC_RELOC_LO16
3770 || scat->r_type == PPC_RELOC_HA16
3771 || scat->r_type == PPC_RELOC_LO14)
3772 { // these are generated by label+offset things
3773 struct relocation_info *pair = &relocs[i+1];
3774 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
3775 barf("Invalid Mach-O file: "
3776 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
3778 if(scat->r_type == PPC_RELOC_LO16)
3780 word = ((unsigned short*) wordPtr)[1];
3781 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3783 else if(scat->r_type == PPC_RELOC_LO14)
3785 barf("Unsupported Relocation: PPC_RELOC_LO14");
3786 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
3787 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3789 else if(scat->r_type == PPC_RELOC_HI16)
3791 word = ((unsigned short*) wordPtr)[1] << 16;
3792 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3794 else if(scat->r_type == PPC_RELOC_HA16)
3796 word = ((unsigned short*) wordPtr)[1] << 16;
3797 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3801 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
3808 continue; // ignore the others
3810 #ifdef powerpc_HOST_ARCH
3811 if(scat->r_type == GENERIC_RELOC_VANILLA
3812 || scat->r_type == PPC_RELOC_SECTDIFF)
3814 if(scat->r_type == GENERIC_RELOC_VANILLA
3815 || scat->r_type == GENERIC_RELOC_SECTDIFF)
3820 #ifdef powerpc_HOST_ARCH
3821 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
3823 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3825 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
3827 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3829 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
3831 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3832 + ((word & (1<<15)) ? 1 : 0);
3838 continue; // FIXME: I hope it's OK to ignore all the others.
3842 struct relocation_info *reloc = &relocs[i];
3843 if(reloc->r_pcrel && !reloc->r_extern)
3846 if(reloc->r_length == 2)
3848 unsigned long word = 0;
3849 #ifdef powerpc_HOST_ARCH
3850 unsigned long jumpIsland = 0;
3851 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
3852 // to avoid warning and to catch
3856 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
3857 checkProddableBlock(oc,wordPtr);
3859 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3863 #ifdef powerpc_HOST_ARCH
3864 else if(reloc->r_type == PPC_RELOC_LO16)
3866 word = ((unsigned short*) wordPtr)[1];
3867 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
3869 else if(reloc->r_type == PPC_RELOC_HI16)
3871 word = ((unsigned short*) wordPtr)[1] << 16;
3872 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
3874 else if(reloc->r_type == PPC_RELOC_HA16)
3876 word = ((unsigned short*) wordPtr)[1] << 16;
3877 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
3879 else if(reloc->r_type == PPC_RELOC_BR24)
3882 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
3886 if(!reloc->r_extern)
3889 sections[reloc->r_symbolnum-1].offset
3890 - sections[reloc->r_symbolnum-1].addr
3897 struct nlist *symbol = &nlist[reloc->r_symbolnum];
3898 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3899 void *symbolAddress = lookupSymbol(nm);
3902 errorBelch("\nunknown symbol `%s'", nm);
3908 #ifdef powerpc_HOST_ARCH
3909 // In the .o file, this should be a relative jump to NULL
3910 // and we'll change it to a relative jump to the symbol
3911 ASSERT(-word == reloc->r_address);
3912 jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,(unsigned long) symbolAddress);
3915 offsetToJumpIsland = word + jumpIsland
3916 - (((long)image) + sect->offset - sect->addr);
3919 word += (unsigned long) symbolAddress
3920 - (((long)image) + sect->offset - sect->addr);
3924 word += (unsigned long) symbolAddress;
3928 if(reloc->r_type == GENERIC_RELOC_VANILLA)
3933 #ifdef powerpc_HOST_ARCH
3934 else if(reloc->r_type == PPC_RELOC_LO16)
3936 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
3939 else if(reloc->r_type == PPC_RELOC_HI16)
3941 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
3944 else if(reloc->r_type == PPC_RELOC_HA16)
3946 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
3947 + ((word & (1<<15)) ? 1 : 0);
3950 else if(reloc->r_type == PPC_RELOC_BR24)
3952 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3954 // The branch offset is too large.
3955 // Therefore, we try to use a jump island.
3958 barf("unconditional relative branch out of range: "
3959 "no jump island available");
3962 word = offsetToJumpIsland;
3963 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
3964 barf("unconditional relative branch out of range: "
3965 "jump island out of range");
3967 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
3972 barf("\nunknown relocation %d",reloc->r_type);
3979 static int ocGetNames_MachO(ObjectCode* oc)
3981 char *image = (char*) oc->image;
3982 struct mach_header *header = (struct mach_header*) image;
3983 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
3984 unsigned i,curSymbol = 0;
3985 struct segment_command *segLC = NULL;
3986 struct section *sections;
3987 struct symtab_command *symLC = NULL;
3988 struct nlist *nlist;
3989 unsigned long commonSize = 0;
3990 char *commonStorage = NULL;
3991 unsigned long commonCounter;
3993 for(i=0;i<header->ncmds;i++)
3995 if(lc->cmd == LC_SEGMENT)
3996 segLC = (struct segment_command*) lc;
3997 else if(lc->cmd == LC_SYMTAB)
3998 symLC = (struct symtab_command*) lc;
3999 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4002 sections = (struct section*) (segLC+1);
4003 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4006 for(i=0;i<segLC->nsects;i++)
4008 if(sections[i].size == 0)
4011 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4013 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4014 "ocGetNames_MachO(common symbols)");
4015 sections[i].offset = zeroFillArea - image;
4018 if(!strcmp(sections[i].sectname,"__text"))
4019 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4020 (void*) (image + sections[i].offset),
4021 (void*) (image + sections[i].offset + sections[i].size));
4022 else if(!strcmp(sections[i].sectname,"__const"))
4023 addSection(oc, SECTIONKIND_RWDATA,
4024 (void*) (image + sections[i].offset),
4025 (void*) (image + sections[i].offset + sections[i].size));
4026 else if(!strcmp(sections[i].sectname,"__data"))
4027 addSection(oc, SECTIONKIND_RWDATA,
4028 (void*) (image + sections[i].offset),
4029 (void*) (image + sections[i].offset + sections[i].size));
4030 else if(!strcmp(sections[i].sectname,"__bss")
4031 || !strcmp(sections[i].sectname,"__common"))
4032 addSection(oc, SECTIONKIND_RWDATA,
4033 (void*) (image + sections[i].offset),
4034 (void*) (image + sections[i].offset + sections[i].size));
4036 addProddableBlock(oc, (void*) (image + sections[i].offset),
4040 // count external symbols defined here
4044 for(i=0;i<symLC->nsyms;i++)
4046 if(nlist[i].n_type & N_STAB)
4048 else if(nlist[i].n_type & N_EXT)
4050 if((nlist[i].n_type & N_TYPE) == N_UNDF
4051 && (nlist[i].n_value != 0))
4053 commonSize += nlist[i].n_value;
4056 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4061 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4062 "ocGetNames_MachO(oc->symbols)");
4066 for(i=0;i<symLC->nsyms;i++)
4068 if(nlist[i].n_type & N_STAB)
4070 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4072 if(nlist[i].n_type & N_EXT)
4074 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4075 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4077 + sections[nlist[i].n_sect-1].offset
4078 - sections[nlist[i].n_sect-1].addr
4079 + nlist[i].n_value);
4080 oc->symbols[curSymbol++] = nm;
4084 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4085 ghciInsertStrHashTable(oc->fileName, oc->lochash, nm,
4087 + sections[nlist[i].n_sect-1].offset
4088 - sections[nlist[i].n_sect-1].addr
4089 + nlist[i].n_value);
4095 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4096 commonCounter = (unsigned long)commonStorage;
4099 for(i=0;i<symLC->nsyms;i++)
4101 if((nlist[i].n_type & N_TYPE) == N_UNDF
4102 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4104 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4105 unsigned long sz = nlist[i].n_value;
4107 nlist[i].n_value = commonCounter;
4109 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4110 (void*)commonCounter);
4111 oc->symbols[curSymbol++] = nm;
4113 commonCounter += sz;
4120 static int ocResolve_MachO(ObjectCode* oc)
4122 char *image = (char*) oc->image;
4123 struct mach_header *header = (struct mach_header*) image;
4124 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4126 struct segment_command *segLC = NULL;
4127 struct section *sections, *la_ptrs = NULL, *nl_ptrs = NULL;
4128 struct symtab_command *symLC = NULL;
4129 struct dysymtab_command *dsymLC = NULL;
4130 struct nlist *nlist;
4132 for(i=0;i<header->ncmds;i++)
4134 if(lc->cmd == LC_SEGMENT)
4135 segLC = (struct segment_command*) lc;
4136 else if(lc->cmd == LC_SYMTAB)
4137 symLC = (struct symtab_command*) lc;
4138 else if(lc->cmd == LC_DYSYMTAB)
4139 dsymLC = (struct dysymtab_command*) lc;
4140 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4143 sections = (struct section*) (segLC+1);
4144 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4147 for(i=0;i<segLC->nsects;i++)
4149 if(!strcmp(sections[i].sectname,"__la_symbol_ptr"))
4150 la_ptrs = §ions[i];
4151 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr"))
4152 nl_ptrs = §ions[i];
4153 else if(!strcmp(sections[i].sectname,"__la_sym_ptr2"))
4154 la_ptrs = §ions[i];
4155 else if(!strcmp(sections[i].sectname,"__la_sym_ptr3"))
4156 la_ptrs = §ions[i];
4161 unsigned long *indirectSyms
4162 = (unsigned long*) (image + dsymLC->indirectsymoff);
4165 if(!resolveImports(oc,image,symLC,la_ptrs,indirectSyms,nlist))
4168 if(!resolveImports(oc,image,symLC,nl_ptrs,indirectSyms,nlist))
4172 for(i=0;i<segLC->nsects;i++)
4174 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4178 /* Free the local symbol table; we won't need it again. */
4179 freeHashTable(oc->lochash, NULL);
4182 #if defined (powerpc_HOST_ARCH)
4183 ocFlushInstructionCache( oc );
4189 #ifdef powerpc_HOST_ARCH
4191 * The Mach-O object format uses leading underscores. But not everywhere.
4192 * There is a small number of runtime support functions defined in
4193 * libcc_dynamic.a whose name does not have a leading underscore.
4194 * As a consequence, we can't get their address from C code.
4195 * We have to use inline assembler just to take the address of a function.
4199 static void machoInitSymbolsWithoutUnderscore()
4201 extern void* symbolsWithoutUnderscore[];
4202 void **p = symbolsWithoutUnderscore;
4203 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4207 __asm__ volatile(".long " # x);
4209 RTS_MACHO_NOUNDERLINE_SYMBOLS
4211 __asm__ volatile(".text");
4215 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4217 RTS_MACHO_NOUNDERLINE_SYMBOLS
4224 * Figure out by how much to shift the entire Mach-O file in memory
4225 * when loading so that its single segment ends up 16-byte-aligned
4227 static int machoGetMisalignment( FILE * f )
4229 struct mach_header header;
4232 fread(&header, sizeof(header), 1, f);
4235 if(header.magic != MH_MAGIC)
4238 misalignment = (header.sizeofcmds + sizeof(header))
4241 return misalignment ? (16 - misalignment) : 0;