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>.
23 #include "sm/Storage.h"
26 #include "LinkerInternals.h"
29 #include "StgPrimFloat.h" // for __int_encodeFloat etc.
32 #if !defined(mingw32_HOST_OS)
33 #include "posix/Signals.h"
36 // get protos for is*()
39 #ifdef HAVE_SYS_TYPES_H
40 #include <sys/types.h>
48 #ifdef HAVE_SYS_STAT_H
52 #if defined(HAVE_DLFCN_H)
56 #if defined(cygwin32_HOST_OS)
61 #ifdef HAVE_SYS_TIME_H
65 #include <sys/fcntl.h>
66 #include <sys/termios.h>
67 #include <sys/utime.h>
68 #include <sys/utsname.h>
72 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS) || defined(darwin_HOST_OS)
83 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
84 # define OBJFORMAT_ELF
85 # include <regex.h> // regex is already used by dlopen() so this is OK
86 // to use here without requiring an additional lib
87 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
88 # define OBJFORMAT_PEi386
91 #elif defined(darwin_HOST_OS)
92 # define OBJFORMAT_MACHO
94 # include <mach-o/loader.h>
95 # include <mach-o/nlist.h>
96 # include <mach-o/reloc.h>
97 #if !defined(HAVE_DLFCN_H)
98 # include <mach-o/dyld.h>
100 #if defined(powerpc_HOST_ARCH)
101 # include <mach-o/ppc/reloc.h>
103 #if defined(x86_64_HOST_ARCH)
104 # include <mach-o/x86_64/reloc.h>
108 #if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS)
112 /* Hash table mapping symbol names to Symbol */
113 static /*Str*/HashTable *symhash;
115 /* Hash table mapping symbol names to StgStablePtr */
116 static /*Str*/HashTable *stablehash;
118 /* List of currently loaded objects */
119 ObjectCode *objects = NULL; /* initially empty */
121 static HsInt loadOc( ObjectCode* oc );
122 static ObjectCode* mkOc( char *path, char *image, int imageSize,
123 char *archiveMemberName
125 #ifdef darwin_HOST_OS
131 #if defined(OBJFORMAT_ELF)
132 static int ocVerifyImage_ELF ( ObjectCode* oc );
133 static int ocGetNames_ELF ( ObjectCode* oc );
134 static int ocResolve_ELF ( ObjectCode* oc );
135 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
136 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
138 #elif defined(OBJFORMAT_PEi386)
139 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
140 static int ocGetNames_PEi386 ( ObjectCode* oc );
141 static int ocResolve_PEi386 ( ObjectCode* oc );
142 static void *lookupSymbolInDLLs ( unsigned char *lbl );
143 static void zapTrailingAtSign ( unsigned char *sym );
144 #elif defined(OBJFORMAT_MACHO)
145 static int ocVerifyImage_MachO ( ObjectCode* oc );
146 static int ocGetNames_MachO ( ObjectCode* oc );
147 static int ocResolve_MachO ( ObjectCode* oc );
150 static int machoGetMisalignment( FILE * );
152 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
153 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
155 #ifdef powerpc_HOST_ARCH
156 static void machoInitSymbolsWithoutUnderscore( void );
160 /* on x86_64 we have a problem with relocating symbol references in
161 * code that was compiled without -fPIC. By default, the small memory
162 * model is used, which assumes that symbol references can fit in a
163 * 32-bit slot. The system dynamic linker makes this work for
164 * references to shared libraries by either (a) allocating a jump
165 * table slot for code references, or (b) moving the symbol at load
166 * time (and copying its contents, if necessary) for data references.
168 * We unfortunately can't tell whether symbol references are to code
169 * or data. So for now we assume they are code (the vast majority
170 * are), and allocate jump-table slots. Unfortunately this will
171 * SILENTLY generate crashing code for data references. This hack is
172 * enabled by X86_64_ELF_NONPIC_HACK.
174 * One workaround is to use shared Haskell libraries. This is
175 * coming. Another workaround is to keep the static libraries but
176 * compile them with -fPIC, because that will generate PIC references
177 * to data which can be relocated. The PIC code is still too green to
178 * do this systematically, though.
181 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
183 * Naming Scheme for Symbol Macros
185 * SymI_*: symbol is internal to the RTS. It resides in an object
186 * file/library that is statically.
187 * SymE_*: symbol is external to the RTS library. It might be linked
190 * Sym*_HasProto : the symbol prototype is imported in an include file
191 * or defined explicitly
192 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
193 * default proto extern void sym(void);
195 #define X86_64_ELF_NONPIC_HACK 1
197 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
198 * small memory model on this architecture (see gcc docs,
201 * MAP_32BIT not available on OpenBSD/amd64
203 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
204 #define TRY_MAP_32BIT MAP_32BIT
206 #define TRY_MAP_32BIT 0
210 * Due to the small memory model (see above), on x86_64 we have to map
211 * all our non-PIC object files into the low 2Gb of the address space
212 * (why 2Gb and not 4Gb? Because all addresses must be reachable
213 * using a 32-bit signed PC-relative offset). On Linux we can do this
214 * using the MAP_32BIT flag to mmap(), however on other OSs
215 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
216 * can't do this. So on these systems, we have to pick a base address
217 * in the low 2Gb of the address space and try to allocate memory from
220 * We pick a default address based on the OS, but also make this
221 * configurable via an RTS flag (+RTS -xm)
223 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
225 #if defined(MAP_32BIT)
226 // Try to use MAP_32BIT
227 #define MMAP_32BIT_BASE_DEFAULT 0
230 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
233 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
236 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
237 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
238 #define MAP_ANONYMOUS MAP_ANON
241 /* -----------------------------------------------------------------------------
242 * Built-in symbols from the RTS
245 typedef struct _RtsSymbolVal {
250 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
251 SymI_HasProto(stg_mkWeakForeignEnvzh) \
252 SymI_HasProto(stg_makeStableNamezh) \
253 SymI_HasProto(stg_finalizzeWeakzh)
255 #if !defined (mingw32_HOST_OS)
256 #define RTS_POSIX_ONLY_SYMBOLS \
257 SymI_HasProto(__hscore_get_saved_termios) \
258 SymI_HasProto(__hscore_set_saved_termios) \
259 SymI_HasProto(shutdownHaskellAndSignal) \
260 SymI_HasProto(lockFile) \
261 SymI_HasProto(unlockFile) \
262 SymI_HasProto(signal_handlers) \
263 SymI_HasProto(stg_sig_install) \
264 SymI_NeedsProto(nocldstop)
267 #if defined (cygwin32_HOST_OS)
268 #define RTS_MINGW_ONLY_SYMBOLS /**/
269 /* Don't have the ability to read import libs / archives, so
270 * we have to stupidly list a lot of what libcygwin.a
273 #define RTS_CYGWIN_ONLY_SYMBOLS \
274 SymI_HasProto(regfree) \
275 SymI_HasProto(regexec) \
276 SymI_HasProto(regerror) \
277 SymI_HasProto(regcomp) \
278 SymI_HasProto(__errno) \
279 SymI_HasProto(access) \
280 SymI_HasProto(chmod) \
281 SymI_HasProto(chdir) \
282 SymI_HasProto(close) \
283 SymI_HasProto(creat) \
285 SymI_HasProto(dup2) \
286 SymI_HasProto(fstat) \
287 SymI_HasProto(fcntl) \
288 SymI_HasProto(getcwd) \
289 SymI_HasProto(getenv) \
290 SymI_HasProto(lseek) \
291 SymI_HasProto(open) \
292 SymI_HasProto(fpathconf) \
293 SymI_HasProto(pathconf) \
294 SymI_HasProto(stat) \
296 SymI_HasProto(tanh) \
297 SymI_HasProto(cosh) \
298 SymI_HasProto(sinh) \
299 SymI_HasProto(atan) \
300 SymI_HasProto(acos) \
301 SymI_HasProto(asin) \
307 SymI_HasProto(sqrt) \
308 SymI_HasProto(localtime_r) \
309 SymI_HasProto(gmtime_r) \
310 SymI_HasProto(mktime) \
311 SymI_NeedsProto(_imp___tzname) \
312 SymI_HasProto(gettimeofday) \
313 SymI_HasProto(timezone) \
314 SymI_HasProto(tcgetattr) \
315 SymI_HasProto(tcsetattr) \
316 SymI_HasProto(memcpy) \
317 SymI_HasProto(memmove) \
318 SymI_HasProto(realloc) \
319 SymI_HasProto(malloc) \
320 SymI_HasProto(free) \
321 SymI_HasProto(fork) \
322 SymI_HasProto(lstat) \
323 SymI_HasProto(isatty) \
324 SymI_HasProto(mkdir) \
325 SymI_HasProto(opendir) \
326 SymI_HasProto(readdir) \
327 SymI_HasProto(rewinddir) \
328 SymI_HasProto(closedir) \
329 SymI_HasProto(link) \
330 SymI_HasProto(mkfifo) \
331 SymI_HasProto(pipe) \
332 SymI_HasProto(read) \
333 SymI_HasProto(rename) \
334 SymI_HasProto(rmdir) \
335 SymI_HasProto(select) \
336 SymI_HasProto(system) \
337 SymI_HasProto(write) \
338 SymI_HasProto(strcmp) \
339 SymI_HasProto(strcpy) \
340 SymI_HasProto(strncpy) \
341 SymI_HasProto(strerror) \
342 SymI_HasProto(sigaddset) \
343 SymI_HasProto(sigemptyset) \
344 SymI_HasProto(sigprocmask) \
345 SymI_HasProto(umask) \
346 SymI_HasProto(uname) \
347 SymI_HasProto(unlink) \
348 SymI_HasProto(utime) \
349 SymI_HasProto(waitpid)
351 #elif !defined(mingw32_HOST_OS)
352 #define RTS_MINGW_ONLY_SYMBOLS /**/
353 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
354 #else /* defined(mingw32_HOST_OS) */
355 #define RTS_POSIX_ONLY_SYMBOLS /**/
356 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
358 #if HAVE_GETTIMEOFDAY
359 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
361 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
364 #if HAVE___MINGW_VFPRINTF
365 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
367 #define RTS___MINGW_VFPRINTF_SYM /**/
370 /* These are statically linked from the mingw libraries into the ghc
371 executable, so we have to employ this hack. */
372 #define RTS_MINGW_ONLY_SYMBOLS \
373 SymI_HasProto(stg_asyncReadzh) \
374 SymI_HasProto(stg_asyncWritezh) \
375 SymI_HasProto(stg_asyncDoProczh) \
376 SymI_HasProto(memset) \
377 SymI_HasProto(inet_ntoa) \
378 SymI_HasProto(inet_addr) \
379 SymI_HasProto(htonl) \
380 SymI_HasProto(recvfrom) \
381 SymI_HasProto(listen) \
382 SymI_HasProto(bind) \
383 SymI_HasProto(shutdown) \
384 SymI_HasProto(connect) \
385 SymI_HasProto(htons) \
386 SymI_HasProto(ntohs) \
387 SymI_HasProto(getservbyname) \
388 SymI_HasProto(getservbyport) \
389 SymI_HasProto(getprotobynumber) \
390 SymI_HasProto(getprotobyname) \
391 SymI_HasProto(gethostbyname) \
392 SymI_HasProto(gethostbyaddr) \
393 SymI_HasProto(gethostname) \
394 SymI_HasProto(strcpy) \
395 SymI_HasProto(strncpy) \
396 SymI_HasProto(abort) \
397 SymI_NeedsProto(_alloca) \
398 SymI_HasProto(isxdigit) \
399 SymI_HasProto(isupper) \
400 SymI_HasProto(ispunct) \
401 SymI_HasProto(islower) \
402 SymI_HasProto(isspace) \
403 SymI_HasProto(isprint) \
404 SymI_HasProto(isdigit) \
405 SymI_HasProto(iscntrl) \
406 SymI_HasProto(isalpha) \
407 SymI_HasProto(isalnum) \
408 SymI_HasProto(isascii) \
409 RTS___MINGW_VFPRINTF_SYM \
410 SymI_HasProto(strcmp) \
411 SymI_HasProto(memmove) \
412 SymI_HasProto(realloc) \
413 SymI_HasProto(malloc) \
415 SymI_HasProto(tanh) \
416 SymI_HasProto(cosh) \
417 SymI_HasProto(sinh) \
418 SymI_HasProto(atan) \
419 SymI_HasProto(acos) \
420 SymI_HasProto(asin) \
426 SymI_HasProto(sqrt) \
427 SymI_HasProto(powf) \
428 SymI_HasProto(tanhf) \
429 SymI_HasProto(coshf) \
430 SymI_HasProto(sinhf) \
431 SymI_HasProto(atanf) \
432 SymI_HasProto(acosf) \
433 SymI_HasProto(asinf) \
434 SymI_HasProto(tanf) \
435 SymI_HasProto(cosf) \
436 SymI_HasProto(sinf) \
437 SymI_HasProto(expf) \
438 SymI_HasProto(logf) \
439 SymI_HasProto(sqrtf) \
441 SymI_HasProto(erfc) \
442 SymI_HasProto(erff) \
443 SymI_HasProto(erfcf) \
444 SymI_HasProto(memcpy) \
445 SymI_HasProto(rts_InstallConsoleEvent) \
446 SymI_HasProto(rts_ConsoleHandlerDone) \
447 SymI_NeedsProto(mktime) \
448 SymI_NeedsProto(_imp___timezone) \
449 SymI_NeedsProto(_imp___tzname) \
450 SymI_NeedsProto(_imp__tzname) \
451 SymI_NeedsProto(_imp___iob) \
452 SymI_NeedsProto(_imp___osver) \
453 SymI_NeedsProto(localtime) \
454 SymI_NeedsProto(gmtime) \
455 SymI_NeedsProto(opendir) \
456 SymI_NeedsProto(readdir) \
457 SymI_NeedsProto(rewinddir) \
458 SymI_NeedsProto(_imp____mb_cur_max) \
459 SymI_NeedsProto(_imp___pctype) \
460 SymI_NeedsProto(__chkstk) \
461 RTS_MINGW_GETTIMEOFDAY_SYM \
462 SymI_NeedsProto(closedir)
466 #if defined(darwin_HOST_OS) && HAVE_PRINTF_LDBLSTUB
467 #define RTS_DARWIN_ONLY_SYMBOLS \
468 SymI_NeedsProto(asprintf$LDBLStub) \
469 SymI_NeedsProto(err$LDBLStub) \
470 SymI_NeedsProto(errc$LDBLStub) \
471 SymI_NeedsProto(errx$LDBLStub) \
472 SymI_NeedsProto(fprintf$LDBLStub) \
473 SymI_NeedsProto(fscanf$LDBLStub) \
474 SymI_NeedsProto(fwprintf$LDBLStub) \
475 SymI_NeedsProto(fwscanf$LDBLStub) \
476 SymI_NeedsProto(printf$LDBLStub) \
477 SymI_NeedsProto(scanf$LDBLStub) \
478 SymI_NeedsProto(snprintf$LDBLStub) \
479 SymI_NeedsProto(sprintf$LDBLStub) \
480 SymI_NeedsProto(sscanf$LDBLStub) \
481 SymI_NeedsProto(strtold$LDBLStub) \
482 SymI_NeedsProto(swprintf$LDBLStub) \
483 SymI_NeedsProto(swscanf$LDBLStub) \
484 SymI_NeedsProto(syslog$LDBLStub) \
485 SymI_NeedsProto(vasprintf$LDBLStub) \
486 SymI_NeedsProto(verr$LDBLStub) \
487 SymI_NeedsProto(verrc$LDBLStub) \
488 SymI_NeedsProto(verrx$LDBLStub) \
489 SymI_NeedsProto(vfprintf$LDBLStub) \
490 SymI_NeedsProto(vfscanf$LDBLStub) \
491 SymI_NeedsProto(vfwprintf$LDBLStub) \
492 SymI_NeedsProto(vfwscanf$LDBLStub) \
493 SymI_NeedsProto(vprintf$LDBLStub) \
494 SymI_NeedsProto(vscanf$LDBLStub) \
495 SymI_NeedsProto(vsnprintf$LDBLStub) \
496 SymI_NeedsProto(vsprintf$LDBLStub) \
497 SymI_NeedsProto(vsscanf$LDBLStub) \
498 SymI_NeedsProto(vswprintf$LDBLStub) \
499 SymI_NeedsProto(vswscanf$LDBLStub) \
500 SymI_NeedsProto(vsyslog$LDBLStub) \
501 SymI_NeedsProto(vwarn$LDBLStub) \
502 SymI_NeedsProto(vwarnc$LDBLStub) \
503 SymI_NeedsProto(vwarnx$LDBLStub) \
504 SymI_NeedsProto(vwprintf$LDBLStub) \
505 SymI_NeedsProto(vwscanf$LDBLStub) \
506 SymI_NeedsProto(warn$LDBLStub) \
507 SymI_NeedsProto(warnc$LDBLStub) \
508 SymI_NeedsProto(warnx$LDBLStub) \
509 SymI_NeedsProto(wcstold$LDBLStub) \
510 SymI_NeedsProto(wprintf$LDBLStub) \
511 SymI_NeedsProto(wscanf$LDBLStub)
513 #define RTS_DARWIN_ONLY_SYMBOLS
517 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
519 # define MAIN_CAP_SYM
522 #if !defined(mingw32_HOST_OS)
523 #define RTS_USER_SIGNALS_SYMBOLS \
524 SymI_HasProto(setIOManagerControlFd) \
525 SymI_HasProto(setIOManagerWakeupFd) \
526 SymI_HasProto(ioManagerWakeup) \
527 SymI_HasProto(blockUserSignals) \
528 SymI_HasProto(unblockUserSignals)
530 #define RTS_USER_SIGNALS_SYMBOLS \
531 SymI_HasProto(ioManagerWakeup) \
532 SymI_HasProto(sendIOManagerEvent) \
533 SymI_HasProto(readIOManagerEvent) \
534 SymI_HasProto(getIOManagerEvent) \
535 SymI_HasProto(console_handler)
538 #define RTS_LIBFFI_SYMBOLS \
539 SymE_NeedsProto(ffi_prep_cif) \
540 SymE_NeedsProto(ffi_call) \
541 SymE_NeedsProto(ffi_type_void) \
542 SymE_NeedsProto(ffi_type_float) \
543 SymE_NeedsProto(ffi_type_double) \
544 SymE_NeedsProto(ffi_type_sint64) \
545 SymE_NeedsProto(ffi_type_uint64) \
546 SymE_NeedsProto(ffi_type_sint32) \
547 SymE_NeedsProto(ffi_type_uint32) \
548 SymE_NeedsProto(ffi_type_sint16) \
549 SymE_NeedsProto(ffi_type_uint16) \
550 SymE_NeedsProto(ffi_type_sint8) \
551 SymE_NeedsProto(ffi_type_uint8) \
552 SymE_NeedsProto(ffi_type_pointer)
554 #ifdef TABLES_NEXT_TO_CODE
555 #define RTS_RET_SYMBOLS /* nothing */
557 #define RTS_RET_SYMBOLS \
558 SymI_HasProto(stg_enter_ret) \
559 SymI_HasProto(stg_gc_fun_ret) \
560 SymI_HasProto(stg_ap_v_ret) \
561 SymI_HasProto(stg_ap_f_ret) \
562 SymI_HasProto(stg_ap_d_ret) \
563 SymI_HasProto(stg_ap_l_ret) \
564 SymI_HasProto(stg_ap_n_ret) \
565 SymI_HasProto(stg_ap_p_ret) \
566 SymI_HasProto(stg_ap_pv_ret) \
567 SymI_HasProto(stg_ap_pp_ret) \
568 SymI_HasProto(stg_ap_ppv_ret) \
569 SymI_HasProto(stg_ap_ppp_ret) \
570 SymI_HasProto(stg_ap_pppv_ret) \
571 SymI_HasProto(stg_ap_pppp_ret) \
572 SymI_HasProto(stg_ap_ppppp_ret) \
573 SymI_HasProto(stg_ap_pppppp_ret)
576 /* Modules compiled with -ticky may mention ticky counters */
577 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
578 #define RTS_TICKY_SYMBOLS \
579 SymI_NeedsProto(ticky_entry_ctrs) \
580 SymI_NeedsProto(top_ct) \
582 SymI_HasProto(ENT_VIA_NODE_ctr) \
583 SymI_HasProto(ENT_STATIC_THK_ctr) \
584 SymI_HasProto(ENT_DYN_THK_ctr) \
585 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
586 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
587 SymI_HasProto(ENT_STATIC_CON_ctr) \
588 SymI_HasProto(ENT_DYN_CON_ctr) \
589 SymI_HasProto(ENT_STATIC_IND_ctr) \
590 SymI_HasProto(ENT_DYN_IND_ctr) \
591 SymI_HasProto(ENT_PERM_IND_ctr) \
592 SymI_HasProto(ENT_PAP_ctr) \
593 SymI_HasProto(ENT_AP_ctr) \
594 SymI_HasProto(ENT_AP_STACK_ctr) \
595 SymI_HasProto(ENT_BH_ctr) \
596 SymI_HasProto(UNKNOWN_CALL_ctr) \
597 SymI_HasProto(SLOW_CALL_v_ctr) \
598 SymI_HasProto(SLOW_CALL_f_ctr) \
599 SymI_HasProto(SLOW_CALL_d_ctr) \
600 SymI_HasProto(SLOW_CALL_l_ctr) \
601 SymI_HasProto(SLOW_CALL_n_ctr) \
602 SymI_HasProto(SLOW_CALL_p_ctr) \
603 SymI_HasProto(SLOW_CALL_pv_ctr) \
604 SymI_HasProto(SLOW_CALL_pp_ctr) \
605 SymI_HasProto(SLOW_CALL_ppv_ctr) \
606 SymI_HasProto(SLOW_CALL_ppp_ctr) \
607 SymI_HasProto(SLOW_CALL_pppv_ctr) \
608 SymI_HasProto(SLOW_CALL_pppp_ctr) \
609 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
610 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
611 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
612 SymI_HasProto(ticky_slow_call_unevald) \
613 SymI_HasProto(SLOW_CALL_ctr) \
614 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
615 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
616 SymI_HasProto(KNOWN_CALL_ctr) \
617 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
618 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
619 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
620 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
621 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
622 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
623 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
624 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
625 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
626 SymI_HasProto(UPDF_OMITTED_ctr) \
627 SymI_HasProto(UPDF_PUSHED_ctr) \
628 SymI_HasProto(CATCHF_PUSHED_ctr) \
629 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
630 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
631 SymI_HasProto(UPD_SQUEEZED_ctr) \
632 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
633 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
634 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
635 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
636 SymI_HasProto(ALLOC_HEAP_ctr) \
637 SymI_HasProto(ALLOC_HEAP_tot) \
638 SymI_HasProto(ALLOC_FUN_ctr) \
639 SymI_HasProto(ALLOC_FUN_adm) \
640 SymI_HasProto(ALLOC_FUN_gds) \
641 SymI_HasProto(ALLOC_FUN_slp) \
642 SymI_HasProto(UPD_NEW_IND_ctr) \
643 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
644 SymI_HasProto(UPD_OLD_IND_ctr) \
645 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
646 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
647 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
648 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
649 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
650 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
651 SymI_HasProto(GC_SEL_MINOR_ctr) \
652 SymI_HasProto(GC_SEL_MAJOR_ctr) \
653 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
654 SymI_HasProto(ALLOC_UP_THK_ctr) \
655 SymI_HasProto(ALLOC_SE_THK_ctr) \
656 SymI_HasProto(ALLOC_THK_adm) \
657 SymI_HasProto(ALLOC_THK_gds) \
658 SymI_HasProto(ALLOC_THK_slp) \
659 SymI_HasProto(ALLOC_CON_ctr) \
660 SymI_HasProto(ALLOC_CON_adm) \
661 SymI_HasProto(ALLOC_CON_gds) \
662 SymI_HasProto(ALLOC_CON_slp) \
663 SymI_HasProto(ALLOC_TUP_ctr) \
664 SymI_HasProto(ALLOC_TUP_adm) \
665 SymI_HasProto(ALLOC_TUP_gds) \
666 SymI_HasProto(ALLOC_TUP_slp) \
667 SymI_HasProto(ALLOC_BH_ctr) \
668 SymI_HasProto(ALLOC_BH_adm) \
669 SymI_HasProto(ALLOC_BH_gds) \
670 SymI_HasProto(ALLOC_BH_slp) \
671 SymI_HasProto(ALLOC_PRIM_ctr) \
672 SymI_HasProto(ALLOC_PRIM_adm) \
673 SymI_HasProto(ALLOC_PRIM_gds) \
674 SymI_HasProto(ALLOC_PRIM_slp) \
675 SymI_HasProto(ALLOC_PAP_ctr) \
676 SymI_HasProto(ALLOC_PAP_adm) \
677 SymI_HasProto(ALLOC_PAP_gds) \
678 SymI_HasProto(ALLOC_PAP_slp) \
679 SymI_HasProto(ALLOC_TSO_ctr) \
680 SymI_HasProto(ALLOC_TSO_adm) \
681 SymI_HasProto(ALLOC_TSO_gds) \
682 SymI_HasProto(ALLOC_TSO_slp) \
683 SymI_HasProto(RET_NEW_ctr) \
684 SymI_HasProto(RET_OLD_ctr) \
685 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
686 SymI_HasProto(RET_SEMI_loads_avoided)
689 // On most platforms, the garbage collector rewrites references
690 // to small integer and char objects to a set of common, shared ones.
692 // We don't do this when compiling to Windows DLLs at the moment because
693 // it doesn't support cross package data references well.
695 #if defined(__PIC__) && defined(mingw32_HOST_OS)
696 #define RTS_INTCHAR_SYMBOLS
698 #define RTS_INTCHAR_SYMBOLS \
699 SymI_HasProto(stg_CHARLIKE_closure) \
700 SymI_HasProto(stg_INTLIKE_closure)
704 #define RTS_SYMBOLS \
707 SymI_HasProto(StgReturn) \
708 SymI_HasProto(stg_enter_info) \
709 SymI_HasProto(stg_gc_void_info) \
710 SymI_HasProto(__stg_gc_enter_1) \
711 SymI_HasProto(stg_gc_noregs) \
712 SymI_HasProto(stg_gc_unpt_r1_info) \
713 SymI_HasProto(stg_gc_unpt_r1) \
714 SymI_HasProto(stg_gc_unbx_r1_info) \
715 SymI_HasProto(stg_gc_unbx_r1) \
716 SymI_HasProto(stg_gc_f1_info) \
717 SymI_HasProto(stg_gc_f1) \
718 SymI_HasProto(stg_gc_d1_info) \
719 SymI_HasProto(stg_gc_d1) \
720 SymI_HasProto(stg_gc_l1_info) \
721 SymI_HasProto(stg_gc_l1) \
722 SymI_HasProto(__stg_gc_fun) \
723 SymI_HasProto(stg_gc_fun_info) \
724 SymI_HasProto(stg_gc_gen) \
725 SymI_HasProto(stg_gc_gen_info) \
726 SymI_HasProto(stg_gc_gen_hp) \
727 SymI_HasProto(stg_gc_ut) \
728 SymI_HasProto(stg_gen_yield) \
729 SymI_HasProto(stg_yield_noregs) \
730 SymI_HasProto(stg_yield_to_interpreter) \
731 SymI_HasProto(stg_gen_block) \
732 SymI_HasProto(stg_block_noregs) \
733 SymI_HasProto(stg_block_1) \
734 SymI_HasProto(stg_block_takemvar) \
735 SymI_HasProto(stg_block_putmvar) \
737 SymI_HasProto(MallocFailHook) \
738 SymI_HasProto(OnExitHook) \
739 SymI_HasProto(OutOfHeapHook) \
740 SymI_HasProto(StackOverflowHook) \
741 SymI_HasProto(addDLL) \
742 SymI_HasProto(__int_encodeDouble) \
743 SymI_HasProto(__word_encodeDouble) \
744 SymI_HasProto(__2Int_encodeDouble) \
745 SymI_HasProto(__int_encodeFloat) \
746 SymI_HasProto(__word_encodeFloat) \
747 SymI_HasProto(stg_atomicallyzh) \
748 SymI_HasProto(barf) \
749 SymI_HasProto(debugBelch) \
750 SymI_HasProto(errorBelch) \
751 SymI_HasProto(sysErrorBelch) \
752 SymI_HasProto(stg_getMaskingStatezh) \
753 SymI_HasProto(stg_maskAsyncExceptionszh) \
754 SymI_HasProto(stg_maskUninterruptiblezh) \
755 SymI_HasProto(stg_catchzh) \
756 SymI_HasProto(stg_catchRetryzh) \
757 SymI_HasProto(stg_catchSTMzh) \
758 SymI_HasProto(stg_checkzh) \
759 SymI_HasProto(closure_flags) \
760 SymI_HasProto(cmp_thread) \
761 SymI_HasProto(createAdjustor) \
762 SymI_HasProto(stg_decodeDoublezu2Intzh) \
763 SymI_HasProto(stg_decodeFloatzuIntzh) \
764 SymI_HasProto(defaultsHook) \
765 SymI_HasProto(stg_delayzh) \
766 SymI_HasProto(stg_deRefWeakzh) \
767 SymI_HasProto(stg_deRefStablePtrzh) \
768 SymI_HasProto(dirty_MUT_VAR) \
769 SymI_HasProto(stg_forkzh) \
770 SymI_HasProto(stg_forkOnzh) \
771 SymI_HasProto(forkProcess) \
772 SymI_HasProto(forkOS_createThread) \
773 SymI_HasProto(freeHaskellFunctionPtr) \
774 SymI_HasProto(getOrSetTypeableStore) \
775 SymI_HasProto(getOrSetGHCConcSignalSignalHandlerStore) \
776 SymI_HasProto(getOrSetGHCConcWindowsPendingDelaysStore) \
777 SymI_HasProto(getOrSetGHCConcWindowsIOManagerThreadStore) \
778 SymI_HasProto(getOrSetGHCConcWindowsProddingStore) \
779 SymI_HasProto(getOrSetSystemEventThreadEventManagerStore) \
780 SymI_HasProto(getOrSetSystemEventThreadIOManagerThreadStore) \
781 SymI_HasProto(genSymZh) \
782 SymI_HasProto(genericRaise) \
783 SymI_HasProto(getProgArgv) \
784 SymI_HasProto(getFullProgArgv) \
785 SymI_HasProto(getStablePtr) \
786 SymI_HasProto(hs_init) \
787 SymI_HasProto(hs_exit) \
788 SymI_HasProto(hs_set_argv) \
789 SymI_HasProto(hs_add_root) \
790 SymI_HasProto(hs_perform_gc) \
791 SymI_HasProto(hs_free_stable_ptr) \
792 SymI_HasProto(hs_free_fun_ptr) \
793 SymI_HasProto(hs_hpc_rootModule) \
794 SymI_HasProto(hs_hpc_module) \
795 SymI_HasProto(initLinker) \
796 SymI_HasProto(stg_unpackClosurezh) \
797 SymI_HasProto(stg_getApStackValzh) \
798 SymI_HasProto(stg_getSparkzh) \
799 SymI_HasProto(stg_numSparkszh) \
800 SymI_HasProto(stg_isCurrentThreadBoundzh) \
801 SymI_HasProto(stg_isEmptyMVarzh) \
802 SymI_HasProto(stg_killThreadzh) \
803 SymI_HasProto(loadArchive) \
804 SymI_HasProto(loadObj) \
805 SymI_HasProto(insertStableSymbol) \
806 SymI_HasProto(insertSymbol) \
807 SymI_HasProto(lookupSymbol) \
808 SymI_HasProto(stg_makeStablePtrzh) \
809 SymI_HasProto(stg_mkApUpd0zh) \
810 SymI_HasProto(stg_myThreadIdzh) \
811 SymI_HasProto(stg_labelThreadzh) \
812 SymI_HasProto(stg_newArrayzh) \
813 SymI_HasProto(stg_newBCOzh) \
814 SymI_HasProto(stg_newByteArrayzh) \
815 SymI_HasProto_redirect(newCAF, newDynCAF) \
816 SymI_HasProto(stg_newMVarzh) \
817 SymI_HasProto(stg_newMutVarzh) \
818 SymI_HasProto(stg_newTVarzh) \
819 SymI_HasProto(stg_noDuplicatezh) \
820 SymI_HasProto(stg_atomicModifyMutVarzh) \
821 SymI_HasProto(stg_newPinnedByteArrayzh) \
822 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
823 SymI_HasProto(newSpark) \
824 SymI_HasProto(performGC) \
825 SymI_HasProto(performMajorGC) \
826 SymI_HasProto(prog_argc) \
827 SymI_HasProto(prog_argv) \
828 SymI_HasProto(stg_putMVarzh) \
829 SymI_HasProto(stg_raisezh) \
830 SymI_HasProto(stg_raiseIOzh) \
831 SymI_HasProto(stg_readTVarzh) \
832 SymI_HasProto(stg_readTVarIOzh) \
833 SymI_HasProto(resumeThread) \
834 SymI_HasProto(resolveObjs) \
835 SymI_HasProto(stg_retryzh) \
836 SymI_HasProto(rts_apply) \
837 SymI_HasProto(rts_checkSchedStatus) \
838 SymI_HasProto(rts_eval) \
839 SymI_HasProto(rts_evalIO) \
840 SymI_HasProto(rts_evalLazyIO) \
841 SymI_HasProto(rts_evalStableIO) \
842 SymI_HasProto(rts_eval_) \
843 SymI_HasProto(rts_getBool) \
844 SymI_HasProto(rts_getChar) \
845 SymI_HasProto(rts_getDouble) \
846 SymI_HasProto(rts_getFloat) \
847 SymI_HasProto(rts_getInt) \
848 SymI_HasProto(rts_getInt8) \
849 SymI_HasProto(rts_getInt16) \
850 SymI_HasProto(rts_getInt32) \
851 SymI_HasProto(rts_getInt64) \
852 SymI_HasProto(rts_getPtr) \
853 SymI_HasProto(rts_getFunPtr) \
854 SymI_HasProto(rts_getStablePtr) \
855 SymI_HasProto(rts_getThreadId) \
856 SymI_HasProto(rts_getWord) \
857 SymI_HasProto(rts_getWord8) \
858 SymI_HasProto(rts_getWord16) \
859 SymI_HasProto(rts_getWord32) \
860 SymI_HasProto(rts_getWord64) \
861 SymI_HasProto(rts_lock) \
862 SymI_HasProto(rts_mkBool) \
863 SymI_HasProto(rts_mkChar) \
864 SymI_HasProto(rts_mkDouble) \
865 SymI_HasProto(rts_mkFloat) \
866 SymI_HasProto(rts_mkInt) \
867 SymI_HasProto(rts_mkInt8) \
868 SymI_HasProto(rts_mkInt16) \
869 SymI_HasProto(rts_mkInt32) \
870 SymI_HasProto(rts_mkInt64) \
871 SymI_HasProto(rts_mkPtr) \
872 SymI_HasProto(rts_mkFunPtr) \
873 SymI_HasProto(rts_mkStablePtr) \
874 SymI_HasProto(rts_mkString) \
875 SymI_HasProto(rts_mkWord) \
876 SymI_HasProto(rts_mkWord8) \
877 SymI_HasProto(rts_mkWord16) \
878 SymI_HasProto(rts_mkWord32) \
879 SymI_HasProto(rts_mkWord64) \
880 SymI_HasProto(rts_unlock) \
881 SymI_HasProto(rts_unsafeGetMyCapability) \
882 SymI_HasProto(rtsSupportsBoundThreads) \
883 SymI_HasProto(rts_isProfiled) \
884 SymI_HasProto(setProgArgv) \
885 SymI_HasProto(startupHaskell) \
886 SymI_HasProto(shutdownHaskell) \
887 SymI_HasProto(shutdownHaskellAndExit) \
888 SymI_HasProto(stable_ptr_table) \
889 SymI_HasProto(stackOverflow) \
890 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
891 SymI_HasProto(stg_BLACKHOLE_info) \
892 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
893 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
894 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
895 SymI_HasProto(startTimer) \
896 SymI_HasProto(stg_MVAR_CLEAN_info) \
897 SymI_HasProto(stg_MVAR_DIRTY_info) \
898 SymI_HasProto(stg_IND_STATIC_info) \
899 SymI_HasProto(stg_ARR_WORDS_info) \
900 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
901 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
902 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
903 SymI_HasProto(stg_WEAK_info) \
904 SymI_HasProto(stg_ap_v_info) \
905 SymI_HasProto(stg_ap_f_info) \
906 SymI_HasProto(stg_ap_d_info) \
907 SymI_HasProto(stg_ap_l_info) \
908 SymI_HasProto(stg_ap_n_info) \
909 SymI_HasProto(stg_ap_p_info) \
910 SymI_HasProto(stg_ap_pv_info) \
911 SymI_HasProto(stg_ap_pp_info) \
912 SymI_HasProto(stg_ap_ppv_info) \
913 SymI_HasProto(stg_ap_ppp_info) \
914 SymI_HasProto(stg_ap_pppv_info) \
915 SymI_HasProto(stg_ap_pppp_info) \
916 SymI_HasProto(stg_ap_ppppp_info) \
917 SymI_HasProto(stg_ap_pppppp_info) \
918 SymI_HasProto(stg_ap_0_fast) \
919 SymI_HasProto(stg_ap_v_fast) \
920 SymI_HasProto(stg_ap_f_fast) \
921 SymI_HasProto(stg_ap_d_fast) \
922 SymI_HasProto(stg_ap_l_fast) \
923 SymI_HasProto(stg_ap_n_fast) \
924 SymI_HasProto(stg_ap_p_fast) \
925 SymI_HasProto(stg_ap_pv_fast) \
926 SymI_HasProto(stg_ap_pp_fast) \
927 SymI_HasProto(stg_ap_ppv_fast) \
928 SymI_HasProto(stg_ap_ppp_fast) \
929 SymI_HasProto(stg_ap_pppv_fast) \
930 SymI_HasProto(stg_ap_pppp_fast) \
931 SymI_HasProto(stg_ap_ppppp_fast) \
932 SymI_HasProto(stg_ap_pppppp_fast) \
933 SymI_HasProto(stg_ap_1_upd_info) \
934 SymI_HasProto(stg_ap_2_upd_info) \
935 SymI_HasProto(stg_ap_3_upd_info) \
936 SymI_HasProto(stg_ap_4_upd_info) \
937 SymI_HasProto(stg_ap_5_upd_info) \
938 SymI_HasProto(stg_ap_6_upd_info) \
939 SymI_HasProto(stg_ap_7_upd_info) \
940 SymI_HasProto(stg_exit) \
941 SymI_HasProto(stg_sel_0_upd_info) \
942 SymI_HasProto(stg_sel_10_upd_info) \
943 SymI_HasProto(stg_sel_11_upd_info) \
944 SymI_HasProto(stg_sel_12_upd_info) \
945 SymI_HasProto(stg_sel_13_upd_info) \
946 SymI_HasProto(stg_sel_14_upd_info) \
947 SymI_HasProto(stg_sel_15_upd_info) \
948 SymI_HasProto(stg_sel_1_upd_info) \
949 SymI_HasProto(stg_sel_2_upd_info) \
950 SymI_HasProto(stg_sel_3_upd_info) \
951 SymI_HasProto(stg_sel_4_upd_info) \
952 SymI_HasProto(stg_sel_5_upd_info) \
953 SymI_HasProto(stg_sel_6_upd_info) \
954 SymI_HasProto(stg_sel_7_upd_info) \
955 SymI_HasProto(stg_sel_8_upd_info) \
956 SymI_HasProto(stg_sel_9_upd_info) \
957 SymI_HasProto(stg_upd_frame_info) \
958 SymI_HasProto(stg_bh_upd_frame_info) \
959 SymI_HasProto(suspendThread) \
960 SymI_HasProto(stg_takeMVarzh) \
961 SymI_HasProto(stg_threadStatuszh) \
962 SymI_HasProto(stg_tryPutMVarzh) \
963 SymI_HasProto(stg_tryTakeMVarzh) \
964 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
965 SymI_HasProto(unloadObj) \
966 SymI_HasProto(stg_unsafeThawArrayzh) \
967 SymI_HasProto(stg_waitReadzh) \
968 SymI_HasProto(stg_waitWritezh) \
969 SymI_HasProto(stg_writeTVarzh) \
970 SymI_HasProto(stg_yieldzh) \
971 SymI_NeedsProto(stg_interp_constr_entry) \
972 SymI_HasProto(stg_arg_bitmaps) \
973 SymI_HasProto(alloc_blocks_lim) \
975 SymI_HasProto(allocate) \
976 SymI_HasProto(allocateExec) \
977 SymI_HasProto(freeExec) \
978 SymI_HasProto(getAllocations) \
979 SymI_HasProto(revertCAFs) \
980 SymI_HasProto(RtsFlags) \
981 SymI_NeedsProto(rts_breakpoint_io_action) \
982 SymI_NeedsProto(rts_stop_next_breakpoint) \
983 SymI_NeedsProto(rts_stop_on_exception) \
984 SymI_HasProto(stopTimer) \
985 SymI_HasProto(n_capabilities) \
986 SymI_HasProto(stg_traceCcszh) \
987 SymI_HasProto(stg_traceEventzh) \
988 RTS_USER_SIGNALS_SYMBOLS \
992 // 64-bit support functions in libgcc.a
993 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
994 #define RTS_LIBGCC_SYMBOLS \
995 SymI_NeedsProto(__divdi3) \
996 SymI_NeedsProto(__udivdi3) \
997 SymI_NeedsProto(__moddi3) \
998 SymI_NeedsProto(__umoddi3) \
999 SymI_NeedsProto(__muldi3) \
1000 SymI_NeedsProto(__ashldi3) \
1001 SymI_NeedsProto(__ashrdi3) \
1002 SymI_NeedsProto(__lshrdi3)
1004 #define RTS_LIBGCC_SYMBOLS
1007 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
1008 // Symbols that don't have a leading underscore
1009 // on Mac OS X. They have to receive special treatment,
1010 // see machoInitSymbolsWithoutUnderscore()
1011 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1012 SymI_NeedsProto(saveFP) \
1013 SymI_NeedsProto(restFP)
1016 /* entirely bogus claims about types of these symbols */
1017 #define SymI_NeedsProto(vvv) extern void vvv(void);
1018 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1019 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1020 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1022 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1023 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1025 #define SymI_HasProto(vvv) /**/
1026 #define SymI_HasProto_redirect(vvv,xxx) /**/
1029 RTS_POSIX_ONLY_SYMBOLS
1030 RTS_MINGW_ONLY_SYMBOLS
1031 RTS_CYGWIN_ONLY_SYMBOLS
1032 RTS_DARWIN_ONLY_SYMBOLS
1035 #undef SymI_NeedsProto
1036 #undef SymI_HasProto
1037 #undef SymI_HasProto_redirect
1038 #undef SymE_HasProto
1039 #undef SymE_NeedsProto
1041 #ifdef LEADING_UNDERSCORE
1042 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1044 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1047 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1049 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1050 (void*)DLL_IMPORT_DATA_REF(vvv) },
1052 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1053 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1055 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1056 // another symbol. See newCAF/newDynCAF for an example.
1057 #define SymI_HasProto_redirect(vvv,xxx) \
1058 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1061 static RtsSymbolVal rtsSyms[] = {
1064 RTS_POSIX_ONLY_SYMBOLS
1065 RTS_MINGW_ONLY_SYMBOLS
1066 RTS_CYGWIN_ONLY_SYMBOLS
1067 RTS_DARWIN_ONLY_SYMBOLS
1070 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1071 // dyld stub code contains references to this,
1072 // but it should never be called because we treat
1073 // lazy pointers as nonlazy.
1074 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1076 { 0, 0 } /* sentinel */
1081 /* -----------------------------------------------------------------------------
1082 * Insert symbols into hash tables, checking for duplicates.
1085 static void ghciInsertStrHashTable ( char* obj_name,
1091 if (lookupHashTable(table, (StgWord)key) == NULL)
1093 insertStrHashTable(table, (StgWord)key, data);
1098 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1100 "whilst processing object file\n"
1102 "This could be caused by:\n"
1103 " * Loading two different object files which export the same symbol\n"
1104 " * Specifying the same object file twice on the GHCi command line\n"
1105 " * An incorrect `package.conf' entry, causing some object to be\n"
1107 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1114 /* -----------------------------------------------------------------------------
1115 * initialize the object linker
1119 static int linker_init_done = 0 ;
1121 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1122 static void *dl_prog_handle;
1123 static regex_t re_invalid;
1124 static regex_t re_realso;
1126 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1134 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1138 IF_DEBUG(linker, debugBelch("initLinker: start\n"));
1140 /* Make initLinker idempotent, so we can call it
1141 before evey relevant operation; that means we
1142 don't need to initialise the linker separately */
1143 if (linker_init_done == 1) {
1144 IF_DEBUG(linker, debugBelch("initLinker: idempotent return\n"));
1147 linker_init_done = 1;
1150 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1151 initMutex(&dl_mutex);
1153 stablehash = allocStrHashTable();
1154 symhash = allocStrHashTable();
1156 /* populate the symbol table with stuff from the RTS */
1157 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1158 ghciInsertStrHashTable("(GHCi built-in symbols)",
1159 symhash, sym->lbl, sym->addr);
1160 IF_DEBUG(linker, debugBelch("initLinker: inserting rts symbol %s, %p\n", sym->lbl, sym->addr));
1162 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1163 machoInitSymbolsWithoutUnderscore();
1166 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1167 # if defined(RTLD_DEFAULT)
1168 dl_prog_handle = RTLD_DEFAULT;
1170 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1171 # endif /* RTLD_DEFAULT */
1173 compileResult = regcomp(&re_invalid,
1174 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1176 ASSERT( compileResult == 0 );
1177 compileResult = regcomp(&re_realso,
1178 "GROUP *\\( *(([^ )])+)",
1180 ASSERT( compileResult == 0 );
1183 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1184 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1185 // User-override for mmap_32bit_base
1186 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1190 #if defined(mingw32_HOST_OS)
1192 * These two libraries cause problems when added to the static link,
1193 * but are necessary for resolving symbols in GHCi, hence we load
1194 * them manually here.
1200 IF_DEBUG(linker, debugBelch("initLinker: done\n"));
1205 exitLinker( void ) {
1206 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1207 if (linker_init_done == 1) {
1208 regfree(&re_invalid);
1209 regfree(&re_realso);
1211 closeMutex(&dl_mutex);
1217 /* -----------------------------------------------------------------------------
1218 * Loading DLL or .so dynamic libraries
1219 * -----------------------------------------------------------------------------
1221 * Add a DLL from which symbols may be found. In the ELF case, just
1222 * do RTLD_GLOBAL-style add, so no further messing around needs to
1223 * happen in order that symbols in the loaded .so are findable --
1224 * lookupSymbol() will subsequently see them by dlsym on the program's
1225 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1227 * In the PEi386 case, open the DLLs and put handles to them in a
1228 * linked list. When looking for a symbol, try all handles in the
1229 * list. This means that we need to load even DLLs that are guaranteed
1230 * to be in the ghc.exe image already, just so we can get a handle
1231 * to give to loadSymbol, so that we can find the symbols. For such
1232 * libraries, the LoadLibrary call should be a no-op except for returning
1237 #if defined(OBJFORMAT_PEi386)
1238 /* A record for storing handles into DLLs. */
1243 struct _OpenedDLL* next;
1248 /* A list thereof. */
1249 static OpenedDLL* opened_dlls = NULL;
1252 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1255 internal_dlopen(const char *dll_name)
1261 // omitted: RTLD_NOW
1262 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1264 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1266 //-------------- Begin critical section ------------------
1267 // This critical section is necessary because dlerror() is not
1268 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1269 // Also, the error message returned must be copied to preserve it
1272 ACQUIRE_LOCK(&dl_mutex);
1273 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1277 /* dlopen failed; return a ptr to the error msg. */
1279 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1280 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1281 strcpy(errmsg_copy, errmsg);
1282 errmsg = errmsg_copy;
1284 RELEASE_LOCK(&dl_mutex);
1285 //--------------- End critical section -------------------
1292 addDLL( char *dll_name )
1294 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1295 /* ------------------- ELF DLL loader ------------------- */
1298 regmatch_t match[NMATCH];
1301 size_t match_length;
1302 #define MAXLINE 1000
1308 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1309 errmsg = internal_dlopen(dll_name);
1311 if (errmsg == NULL) {
1315 // GHC Trac ticket #2615
1316 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1317 // contain linker scripts rather than ELF-format object code. This
1318 // code handles the situation by recognizing the real object code
1319 // file name given in the linker script.
1321 // If an "invalid ELF header" error occurs, it is assumed that the
1322 // .so file contains a linker script instead of ELF object code.
1323 // In this case, the code looks for the GROUP ( ... ) linker
1324 // directive. If one is found, the first file name inside the
1325 // parentheses is treated as the name of a dynamic library and the
1326 // code attempts to dlopen that file. If this is also unsuccessful,
1327 // an error message is returned.
1329 // see if the error message is due to an invalid ELF header
1330 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1331 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1332 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1334 // success -- try to read the named file as a linker script
1335 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1337 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1338 line[match_length] = '\0'; // make sure string is null-terminated
1339 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1340 if ((fp = fopen(line, "r")) == NULL) {
1341 return errmsg; // return original error if open fails
1343 // try to find a GROUP ( ... ) command
1344 while (fgets(line, MAXLINE, fp) != NULL) {
1345 IF_DEBUG(linker, debugBelch("input line = %s", line));
1346 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1347 // success -- try to dlopen the first named file
1348 IF_DEBUG(linker, debugBelch("match%s\n",""));
1349 line[match[1].rm_eo] = '\0';
1350 errmsg = internal_dlopen(line+match[1].rm_so);
1353 // if control reaches here, no GROUP ( ... ) directive was found
1354 // and the original error message is returned to the caller
1360 # elif defined(OBJFORMAT_PEi386)
1361 /* ------------------- Win32 DLL loader ------------------- */
1369 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1371 /* See if we've already got it, and ignore if so. */
1372 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1373 if (0 == strcmp(o_dll->name, dll_name))
1377 /* The file name has no suffix (yet) so that we can try
1378 both foo.dll and foo.drv
1380 The documentation for LoadLibrary says:
1381 If no file name extension is specified in the lpFileName
1382 parameter, the default library extension .dll is
1383 appended. However, the file name string can include a trailing
1384 point character (.) to indicate that the module name has no
1387 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1388 sprintf(buf, "%s.DLL", dll_name);
1389 instance = LoadLibrary(buf);
1390 if (instance == NULL) {
1391 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1392 // KAA: allow loading of drivers (like winspool.drv)
1393 sprintf(buf, "%s.DRV", dll_name);
1394 instance = LoadLibrary(buf);
1395 if (instance == NULL) {
1396 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1397 // #1883: allow loading of unix-style libfoo.dll DLLs
1398 sprintf(buf, "lib%s.DLL", dll_name);
1399 instance = LoadLibrary(buf);
1400 if (instance == NULL) {
1407 /* Add this DLL to the list of DLLs in which to search for symbols. */
1408 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1409 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1410 strcpy(o_dll->name, dll_name);
1411 o_dll->instance = instance;
1412 o_dll->next = opened_dlls;
1413 opened_dlls = o_dll;
1419 sysErrorBelch(dll_name);
1421 /* LoadLibrary failed; return a ptr to the error msg. */
1422 return "addDLL: could not load DLL";
1425 barf("addDLL: not implemented on this platform");
1429 /* -----------------------------------------------------------------------------
1430 * insert a stable symbol in the hash table
1434 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1436 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1440 /* -----------------------------------------------------------------------------
1441 * insert a symbol in the hash table
1444 insertSymbol(char* obj_name, char* key, void* data)
1446 ghciInsertStrHashTable(obj_name, symhash, key, data);
1449 /* -----------------------------------------------------------------------------
1450 * lookup a symbol in the hash table
1453 lookupSymbol( char *lbl )
1456 IF_DEBUG(linker, debugBelch("lookupSymbol: looking up %s\n", lbl));
1458 ASSERT(symhash != NULL);
1459 val = lookupStrHashTable(symhash, lbl);
1462 IF_DEBUG(linker, debugBelch("lookupSymbol: symbol not found\n"));
1463 # if defined(OBJFORMAT_ELF)
1464 return dlsym(dl_prog_handle, lbl);
1465 # elif defined(OBJFORMAT_MACHO)
1467 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1470 HACK: On OS X, global symbols are prefixed with an underscore.
1471 However, dlsym wants us to omit the leading underscore from the
1472 symbol name. For now, we simply strip it off here (and ONLY
1475 IF_DEBUG(linker, debugBelch("lookupSymbol: looking up %s with dlsym\n", lbl));
1476 ASSERT(lbl[0] == '_');
1477 return dlsym(dl_prog_handle, lbl+1);
1479 if(NSIsSymbolNameDefined(lbl)) {
1480 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1481 return NSAddressOfSymbol(symbol);
1485 # endif /* HAVE_DLFCN_H */
1486 # elif defined(OBJFORMAT_PEi386)
1489 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1490 if (sym != NULL) { return sym; };
1492 // Also try looking up the symbol without the @N suffix. Some
1493 // DLLs have the suffixes on their symbols, some don't.
1494 zapTrailingAtSign ( (unsigned char*)lbl );
1495 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1496 if (sym != NULL) { return sym; };
1504 IF_DEBUG(linker, debugBelch("lookupSymbol: value of %s is %p\n", lbl, val));
1509 /* -----------------------------------------------------------------------------
1510 * Debugging aid: look in GHCi's object symbol tables for symbols
1511 * within DELTA bytes of the specified address, and show their names.
1514 void ghci_enquire ( char* addr );
1516 void ghci_enquire ( char* addr )
1521 const int DELTA = 64;
1526 for (oc = objects; oc; oc = oc->next) {
1527 for (i = 0; i < oc->n_symbols; i++) {
1528 sym = oc->symbols[i];
1529 if (sym == NULL) continue;
1532 a = lookupStrHashTable(symhash, sym);
1535 // debugBelch("ghci_enquire: can't find %s\n", sym);
1537 else if (addr-DELTA <= a && a <= addr+DELTA) {
1538 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1546 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1549 mmapForLinker (size_t bytes, nat flags, int fd)
1551 void *map_addr = NULL;
1554 static nat fixed = 0;
1556 pagesize = getpagesize();
1557 size = ROUND_UP(bytes, pagesize);
1559 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1562 if (mmap_32bit_base != 0) {
1563 map_addr = mmap_32bit_base;
1567 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1568 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1570 if (result == MAP_FAILED) {
1571 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1572 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1573 stg_exit(EXIT_FAILURE);
1576 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1577 if (mmap_32bit_base != 0) {
1578 if (result == map_addr) {
1579 mmap_32bit_base = (StgWord8*)map_addr + size;
1581 if ((W_)result > 0x80000000) {
1582 // oops, we were given memory over 2Gb
1583 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1584 // Some platforms require MAP_FIXED. This is normally
1585 // a bad idea, because MAP_FIXED will overwrite
1586 // existing mappings.
1587 munmap(result,size);
1591 barf("loadObj: failed to mmap() memory below 2Gb; asked for %lu bytes at %p. Try specifying an address with +RTS -xm<addr> -RTS", size, map_addr, result);
1594 // hmm, we were given memory somewhere else, but it's
1595 // still under 2Gb so we can use it. Next time, ask
1596 // for memory right after the place we just got some
1597 mmap_32bit_base = (StgWord8*)result + size;
1601 if ((W_)result > 0x80000000) {
1602 // oops, we were given memory over 2Gb
1603 // ... try allocating memory somewhere else?;
1604 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1605 munmap(result, size);
1607 // Set a base address and try again... (guess: 1Gb)
1608 mmap_32bit_base = (void*)0x40000000;
1619 mkOc( char *path, char *image, int imageSize,
1620 char *archiveMemberName
1622 #ifdef darwin_HOST_OS
1629 oc = stgMallocBytes(sizeof(ObjectCode), "loadArchive(oc)");
1631 # if defined(OBJFORMAT_ELF)
1632 oc->formatName = "ELF";
1633 # elif defined(OBJFORMAT_PEi386)
1634 oc->formatName = "PEi386";
1635 # elif defined(OBJFORMAT_MACHO)
1636 oc->formatName = "Mach-O";
1639 barf("loadObj: not implemented on this platform");
1643 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1644 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1645 strcpy(oc->fileName, path);
1647 if (archiveMemberName) {
1648 oc->archiveMemberName = stgMallocBytes( strlen(archiveMemberName)+1, "loadObj" );
1649 strcpy(oc->archiveMemberName, archiveMemberName);
1652 oc->archiveMemberName = NULL;
1655 oc->fileSize = imageSize;
1657 oc->sections = NULL;
1658 oc->proddables = NULL;
1661 #ifdef darwin_HOST_OS
1662 oc->misalignment = misalignment;
1666 /* chain it onto the list of objects */
1674 loadArchive( char *path )
1681 size_t thisFileNameSize;
1683 size_t fileNameSize;
1684 int isObject, isGnuIndex;
1687 int gnuFileIndexSize;
1689 IF_DEBUG(linker, debugBelch("loadArchive: Loading archive `%s'\n", path));
1691 gnuFileIndex = NULL;
1692 gnuFileIndexSize = 0;
1695 fileName = stgMallocBytes(fileNameSize, "loadArchive(fileName)");
1697 f = fopen(path, "rb");
1699 barf("loadObj: can't read `%s'", path);
1701 n = fread ( tmp, 1, 8, f );
1702 if (strncmp(tmp, "!<arch>\n", 8) != 0)
1703 barf("loadArchive: Not an archive: `%s'", path);
1706 n = fread ( fileName, 1, 16, f );
1712 barf("loadArchive: Failed reading file name from `%s'", path);
1715 n = fread ( tmp, 1, 12, f );
1717 barf("loadArchive: Failed reading mod time from `%s'", path);
1718 n = fread ( tmp, 1, 6, f );
1720 barf("loadArchive: Failed reading owner from `%s'", path);
1721 n = fread ( tmp, 1, 6, f );
1723 barf("loadArchive: Failed reading group from `%s'", path);
1724 n = fread ( tmp, 1, 8, f );
1726 barf("loadArchive: Failed reading mode from `%s'", path);
1727 n = fread ( tmp, 1, 10, f );
1729 barf("loadArchive: Failed reading size from `%s'", path);
1731 for (n = 0; isdigit(tmp[n]); n++);
1733 memberSize = atoi(tmp);
1734 n = fread ( tmp, 1, 2, f );
1735 if (strncmp(tmp, "\x60\x0A", 2) != 0)
1736 barf("loadArchive: Failed reading magic from `%s' at %ld. Got %c%c",
1737 path, ftell(f), tmp[0], tmp[1]);
1740 /* Check for BSD-variant large filenames */
1741 if (0 == strncmp(fileName, "#1/", 3)) {
1742 fileName[16] = '\0';
1743 if (isdigit(fileName[3])) {
1744 for (n = 4; isdigit(fileName[n]); n++);
1746 thisFileNameSize = atoi(fileName + 3);
1747 memberSize -= thisFileNameSize;
1748 if (thisFileNameSize >= fileNameSize) {
1749 /* Double it to avoid potentially continually
1750 increasing it by 1 */
1751 fileNameSize = thisFileNameSize * 2;
1752 fileName = stgReallocBytes(fileName, fileNameSize, "loadArchive(fileName)");
1754 n = fread ( fileName, 1, thisFileNameSize, f );
1755 if (n != (int)thisFileNameSize) {
1756 barf("loadArchive: Failed reading filename from `%s'",
1759 fileName[thisFileNameSize] = 0;
1762 barf("loadArchive: BSD-variant filename size not found while reading filename from `%s'", path);
1765 /* Check for GNU file index file */
1766 else if (0 == strncmp(fileName, "//", 2)) {
1768 thisFileNameSize = 0;
1771 /* Check for a file in the GNU file index */
1772 else if (fileName[0] == '/') {
1773 if (isdigit(fileName[1])) {
1776 for (n = 2; isdigit(fileName[n]); n++);
1778 n = atoi(fileName + 1);
1780 if (gnuFileIndex == NULL) {
1781 barf("loadArchive: GNU-variant filename without an index while reading from `%s'", path);
1783 if (n < 0 || n > gnuFileIndexSize) {
1784 barf("loadArchive: GNU-variant filename offset %d out of range [0..%d] while reading filename from `%s'", n, gnuFileIndexSize, path);
1786 if (n != 0 && gnuFileIndex[n - 1] != '\n') {
1787 barf("loadArchive: GNU-variant filename offset %d invalid (range [0..%d]) while reading filename from `%s'", n, gnuFileIndexSize, path);
1789 for (i = n; gnuFileIndex[i] != '/'; i++);
1790 thisFileNameSize = i - n;
1791 if (thisFileNameSize >= fileNameSize) {
1792 /* Double it to avoid potentially continually
1793 increasing it by 1 */
1794 fileNameSize = thisFileNameSize * 2;
1795 fileName = stgReallocBytes(fileName, fileNameSize, "loadArchive(fileName)");
1797 memcpy(fileName, gnuFileIndex + n, thisFileNameSize);
1798 fileName[thisFileNameSize] = '\0';
1800 else if (fileName[1] == ' ') {
1802 thisFileNameSize = 0;
1805 barf("loadArchive: GNU-variant filename offset not found while reading filename from `%s'", path);
1808 /* Finally, the case where the filename field actually contains
1811 /* GNU ar terminates filenames with a '/', this allowing
1812 spaces in filenames. So first look to see if there is a
1814 for (thisFileNameSize = 0;
1815 thisFileNameSize < 16;
1816 thisFileNameSize++) {
1817 if (fileName[thisFileNameSize] == '/') {
1818 fileName[thisFileNameSize] = '\0';
1822 /* If we didn't find a '/', then a space teminates the
1823 filename. Note that if we don't find one, then
1824 thisFileNameSize ends up as 16, and we already have the
1826 if (thisFileNameSize == 16) {
1827 for (thisFileNameSize = 0;
1828 thisFileNameSize < 16;
1829 thisFileNameSize++) {
1830 if (fileName[thisFileNameSize] == ' ') {
1831 fileName[thisFileNameSize] = '\0';
1839 debugBelch("loadArchive: Found member file `%s'\n", fileName));
1841 isObject = thisFileNameSize >= 2
1842 && fileName[thisFileNameSize - 2] == '.'
1843 && fileName[thisFileNameSize - 1] == 'o';
1846 char *archiveMemberName;
1848 IF_DEBUG(linker, debugBelch("loadArchive: Member is an object file...loading...\n"));
1850 /* We can't mmap from the archive directly, as object
1851 files need to be 8-byte aligned but files in .ar
1852 archives are 2-byte aligned. When possible we use mmap
1853 to get some anonymous memory, as on 64-bit platforms if
1854 we use malloc then we can be given memory above 2^32.
1855 In the mmap case we're probably wasting lots of space;
1856 we could do better. */
1858 image = mmapForLinker(memberSize, MAP_ANONYMOUS, -1);
1860 image = stgMallocBytes(memberSize, "loadArchive(image)");
1862 n = fread ( image, 1, memberSize, f );
1863 if (n != memberSize) {
1864 barf("loadArchive: error whilst reading `%s'", path);
1867 archiveMemberName = stgMallocBytes(strlen(path) + thisFileNameSize + 3,
1868 "loadArchive(file)");
1869 sprintf(archiveMemberName, "%s(%.*s)",
1870 path, (int)thisFileNameSize, fileName);
1872 oc = mkOc(path, image, memberSize, archiveMemberName
1874 #ifdef darwin_HOST_OS
1880 stgFree(archiveMemberName);
1882 if (0 == loadOc(oc)) {
1887 else if (isGnuIndex) {
1888 if (gnuFileIndex != NULL) {
1889 barf("loadArchive: GNU-variant index found, but already have an index, while reading filename from `%s'", path);
1891 IF_DEBUG(linker, debugBelch("loadArchive: Found GNU-variant file index\n"));
1893 gnuFileIndex = mmapForLinker(memberSize + 1, MAP_ANONYMOUS, -1);
1895 gnuFileIndex = stgMallocBytes(memberSize + 1, "loadArchive(image)");
1897 n = fread ( gnuFileIndex, 1, memberSize, f );
1898 if (n != memberSize) {
1899 barf("loadArchive: error whilst reading `%s'", path);
1901 gnuFileIndex[memberSize] = '/';
1902 gnuFileIndexSize = memberSize;
1905 n = fseek(f, memberSize, SEEK_CUR);
1907 barf("loadArchive: error whilst seeking by %d in `%s'",
1910 /* .ar files are 2-byte aligned */
1911 if (memberSize % 2) {
1912 n = fread ( tmp, 1, 1, f );
1918 barf("loadArchive: Failed reading padding from `%s'", path);
1927 if (gnuFileIndex != NULL) {
1929 munmap(gnuFileIndex, gnuFileIndexSize + 1);
1931 stgFree(gnuFileIndex);
1938 /* -----------------------------------------------------------------------------
1939 * Load an obj (populate the global symbol table, but don't resolve yet)
1941 * Returns: 1 if ok, 0 on error.
1944 loadObj( char *path )
1956 IF_DEBUG(linker, debugBelch("loadObj %s\n", path));
1960 /* debugBelch("loadObj %s\n", path ); */
1962 /* Check that we haven't already loaded this object.
1963 Ignore requests to load multiple times */
1967 for (o = objects; o; o = o->next) {
1968 if (0 == strcmp(o->fileName, path)) {
1970 break; /* don't need to search further */
1974 IF_DEBUG(linker, debugBelch(
1975 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1976 "same object file twice:\n"
1978 "GHCi will ignore this, but be warned.\n"
1980 return 1; /* success */
1984 r = stat(path, &st);
1986 IF_DEBUG(linker, debugBelch("File doesn't exist\n"));
1990 fileSize = st.st_size;
1993 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1995 #if defined(openbsd_HOST_OS)
1996 fd = open(path, O_RDONLY, S_IRUSR);
1998 fd = open(path, O_RDONLY);
2001 barf("loadObj: can't open `%s'", path);
2003 image = mmapForLinker(fileSize, 0, fd);
2007 #else /* !USE_MMAP */
2008 /* load the image into memory */
2009 f = fopen(path, "rb");
2011 barf("loadObj: can't read `%s'", path);
2013 # if defined(mingw32_HOST_OS)
2014 // TODO: We would like to use allocateExec here, but allocateExec
2015 // cannot currently allocate blocks large enough.
2016 image = VirtualAlloc(NULL, fileSize, MEM_RESERVE | MEM_COMMIT,
2017 PAGE_EXECUTE_READWRITE);
2018 # elif defined(darwin_HOST_OS)
2019 // In a Mach-O .o file, all sections can and will be misaligned
2020 // if the total size of the headers is not a multiple of the
2021 // desired alignment. This is fine for .o files that only serve
2022 // as input for the static linker, but it's not fine for us,
2023 // as SSE (used by gcc for floating point) and Altivec require
2024 // 16-byte alignment.
2025 // We calculate the correct alignment from the header before
2026 // reading the file, and then we misalign image on purpose so
2027 // that the actual sections end up aligned again.
2028 misalignment = machoGetMisalignment(f);
2029 image = stgMallocBytes(fileSize + misalignment, "loadObj(image)");
2030 image += misalignment;
2032 image = stgMallocBytes(fileSize, "loadObj(image)");
2037 n = fread ( image, 1, fileSize, f );
2039 barf("loadObj: error whilst reading `%s'", path);
2042 #endif /* USE_MMAP */
2044 oc = mkOc(path, image, fileSize, NULL
2046 #ifdef darwin_HOST_OS
2056 loadOc( ObjectCode* oc ) {
2059 IF_DEBUG(linker, debugBelch("loadOc\n"));
2061 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
2062 r = ocAllocateSymbolExtras_MachO ( oc );
2064 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO failed\n"));
2067 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
2068 r = ocAllocateSymbolExtras_ELF ( oc );
2070 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_ELF failed\n"));
2075 /* verify the in-memory image */
2076 # if defined(OBJFORMAT_ELF)
2077 r = ocVerifyImage_ELF ( oc );
2078 # elif defined(OBJFORMAT_PEi386)
2079 r = ocVerifyImage_PEi386 ( oc );
2080 # elif defined(OBJFORMAT_MACHO)
2081 r = ocVerifyImage_MachO ( oc );
2083 barf("loadObj: no verify method");
2086 IF_DEBUG(linker, debugBelch("ocVerifyImage_* failed\n"));
2090 /* build the symbol list for this image */
2091 # if defined(OBJFORMAT_ELF)
2092 r = ocGetNames_ELF ( oc );
2093 # elif defined(OBJFORMAT_PEi386)
2094 r = ocGetNames_PEi386 ( oc );
2095 # elif defined(OBJFORMAT_MACHO)
2096 r = ocGetNames_MachO ( oc );
2098 barf("loadObj: no getNames method");
2101 IF_DEBUG(linker, debugBelch("ocGetNames_* failed\n"));
2105 /* loaded, but not resolved yet */
2106 oc->status = OBJECT_LOADED;
2107 IF_DEBUG(linker, debugBelch("loadObj done.\n"));
2112 /* -----------------------------------------------------------------------------
2113 * resolve all the currently unlinked objects in memory
2115 * Returns: 1 if ok, 0 on error.
2123 IF_DEBUG(linker, debugBelch("resolveObjs: start\n"));
2126 for (oc = objects; oc; oc = oc->next) {
2127 if (oc->status != OBJECT_RESOLVED) {
2128 # if defined(OBJFORMAT_ELF)
2129 r = ocResolve_ELF ( oc );
2130 # elif defined(OBJFORMAT_PEi386)
2131 r = ocResolve_PEi386 ( oc );
2132 # elif defined(OBJFORMAT_MACHO)
2133 r = ocResolve_MachO ( oc );
2135 barf("resolveObjs: not implemented on this platform");
2137 if (!r) { return r; }
2138 oc->status = OBJECT_RESOLVED;
2141 IF_DEBUG(linker, debugBelch("resolveObjs: done\n"));
2145 /* -----------------------------------------------------------------------------
2146 * delete an object from the pool
2149 unloadObj( char *path )
2151 ObjectCode *oc, *prev;
2152 HsBool unloadedAnyObj = HS_BOOL_FALSE;
2154 ASSERT(symhash != NULL);
2155 ASSERT(objects != NULL);
2160 for (oc = objects; oc; prev = oc, oc = oc->next) {
2161 if (!strcmp(oc->fileName,path)) {
2163 /* Remove all the mappings for the symbols within this
2168 for (i = 0; i < oc->n_symbols; i++) {
2169 if (oc->symbols[i] != NULL) {
2170 removeStrHashTable(symhash, oc->symbols[i], NULL);
2178 prev->next = oc->next;
2181 // We're going to leave this in place, in case there are
2182 // any pointers from the heap into it:
2183 // #ifdef mingw32_HOST_OS
2184 // VirtualFree(oc->image);
2186 // stgFree(oc->image);
2188 stgFree(oc->fileName);
2189 stgFree(oc->symbols);
2190 stgFree(oc->sections);
2193 /* This could be a member of an archive so continue
2194 * unloading other members. */
2195 unloadedAnyObj = HS_BOOL_TRUE;
2199 if (unloadedAnyObj) {
2203 errorBelch("unloadObj: can't find `%s' to unload", path);
2208 /* -----------------------------------------------------------------------------
2209 * Sanity checking. For each ObjectCode, maintain a list of address ranges
2210 * which may be prodded during relocation, and abort if we try and write
2211 * outside any of these.
2213 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
2216 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
2217 IF_DEBUG(linker, debugBelch("addProddableBlock %p %p %d\n", oc, start, size));
2221 pb->next = oc->proddables;
2222 oc->proddables = pb;
2225 static void checkProddableBlock ( ObjectCode* oc, void* addr )
2228 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
2229 char* s = (char*)(pb->start);
2230 char* e = s + pb->size - 1;
2231 char* a = (char*)addr;
2232 /* Assumes that the biggest fixup involves a 4-byte write. This
2233 probably needs to be changed to 8 (ie, +7) on 64-bit
2235 if (a >= s && (a+3) <= e) return;
2237 barf("checkProddableBlock: invalid fixup in runtime linker");
2240 /* -----------------------------------------------------------------------------
2241 * Section management.
2243 static void addSection ( ObjectCode* oc, SectionKind kind,
2244 void* start, void* end )
2246 Section* s = stgMallocBytes(sizeof(Section), "addSection");
2250 s->next = oc->sections;
2253 debugBelch("addSection: %p-%p (size %d), kind %d\n",
2254 start, ((char*)end)-1, end - start + 1, kind );
2259 /* --------------------------------------------------------------------------
2261 * This is about allocating a small chunk of memory for every symbol in the
2262 * object file. We make sure that the SymboLExtras are always "in range" of
2263 * limited-range PC-relative instructions on various platforms by allocating
2264 * them right next to the object code itself.
2267 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
2270 ocAllocateSymbolExtras
2272 Allocate additional space at the end of the object file image to make room
2273 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
2275 PowerPC relative branch instructions have a 24 bit displacement field.
2276 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
2277 If a particular imported symbol is outside this range, we have to redirect
2278 the jump to a short piece of new code that just loads the 32bit absolute
2279 address and jumps there.
2280 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
2283 This function just allocates space for one SymbolExtra for every
2284 undefined symbol in the object file. The code for the jump islands is
2285 filled in by makeSymbolExtra below.
2288 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
2295 int misalignment = 0;
2296 #ifdef darwin_HOST_OS
2297 misalignment = oc->misalignment;
2303 // round up to the nearest 4
2304 aligned = (oc->fileSize + 3) & ~3;
2307 pagesize = getpagesize();
2308 n = ROUND_UP( oc->fileSize, pagesize );
2309 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
2311 /* we try to use spare space at the end of the last page of the
2312 * image for the jump islands, but if there isn't enough space
2313 * then we have to map some (anonymously, remembering MAP_32BIT).
2315 if( m > n ) // we need to allocate more pages
2317 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
2322 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
2325 oc->image -= misalignment;
2326 oc->image = stgReallocBytes( oc->image,
2328 aligned + sizeof (SymbolExtra) * count,
2329 "ocAllocateSymbolExtras" );
2330 oc->image += misalignment;
2332 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
2333 #endif /* USE_MMAP */
2335 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
2338 oc->symbol_extras = NULL;
2340 oc->first_symbol_extra = first;
2341 oc->n_symbol_extras = count;
2346 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
2347 unsigned long symbolNumber,
2348 unsigned long target )
2352 ASSERT( symbolNumber >= oc->first_symbol_extra
2353 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
2355 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
2357 #ifdef powerpc_HOST_ARCH
2358 // lis r12, hi16(target)
2359 extra->jumpIsland.lis_r12 = 0x3d80;
2360 extra->jumpIsland.hi_addr = target >> 16;
2362 // ori r12, r12, lo16(target)
2363 extra->jumpIsland.ori_r12_r12 = 0x618c;
2364 extra->jumpIsland.lo_addr = target & 0xffff;
2367 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2370 extra->jumpIsland.bctr = 0x4e800420;
2372 #ifdef x86_64_HOST_ARCH
2374 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2375 extra->addr = target;
2376 memcpy(extra->jumpIsland, jmp, 6);
2384 /* --------------------------------------------------------------------------
2385 * PowerPC specifics (instruction cache flushing)
2386 * ------------------------------------------------------------------------*/
2388 #ifdef powerpc_HOST_ARCH
2390 ocFlushInstructionCache
2392 Flush the data & instruction caches.
2393 Because the PPC has split data/instruction caches, we have to
2394 do that whenever we modify code at runtime.
2396 static void ocFlushInstructionCacheFrom(void* begin, size_t length)
2398 size_t n = (length + 3) / 4;
2399 unsigned long* p = begin;
2403 __asm__ volatile ( "dcbf 0,%0\n\t"
2411 __asm__ volatile ( "sync\n\t"
2415 static void ocFlushInstructionCache( ObjectCode *oc )
2417 /* The main object code */
2418 ocFlushInstructionCacheFrom(oc->image + oc->misalignment, oc->fileSize);
2421 ocFlushInstructionCacheFrom(oc->symbol_extras, sizeof(SymbolExtra) * oc->n_symbol_extras);
2425 /* --------------------------------------------------------------------------
2426 * PEi386 specifics (Win32 targets)
2427 * ------------------------------------------------------------------------*/
2429 /* The information for this linker comes from
2430 Microsoft Portable Executable
2431 and Common Object File Format Specification
2432 revision 5.1 January 1998
2433 which SimonM says comes from the MS Developer Network CDs.
2435 It can be found there (on older CDs), but can also be found
2438 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2440 (this is Rev 6.0 from February 1999).
2442 Things move, so if that fails, try searching for it via
2444 http://www.google.com/search?q=PE+COFF+specification
2446 The ultimate reference for the PE format is the Winnt.h
2447 header file that comes with the Platform SDKs; as always,
2448 implementations will drift wrt their documentation.
2450 A good background article on the PE format is Matt Pietrek's
2451 March 1994 article in Microsoft System Journal (MSJ)
2452 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2453 Win32 Portable Executable File Format." The info in there
2454 has recently been updated in a two part article in
2455 MSDN magazine, issues Feb and March 2002,
2456 "Inside Windows: An In-Depth Look into the Win32 Portable
2457 Executable File Format"
2459 John Levine's book "Linkers and Loaders" contains useful
2464 #if defined(OBJFORMAT_PEi386)
2468 typedef unsigned char UChar;
2469 typedef unsigned short UInt16;
2470 typedef unsigned int UInt32;
2477 UInt16 NumberOfSections;
2478 UInt32 TimeDateStamp;
2479 UInt32 PointerToSymbolTable;
2480 UInt32 NumberOfSymbols;
2481 UInt16 SizeOfOptionalHeader;
2482 UInt16 Characteristics;
2486 #define sizeof_COFF_header 20
2493 UInt32 VirtualAddress;
2494 UInt32 SizeOfRawData;
2495 UInt32 PointerToRawData;
2496 UInt32 PointerToRelocations;
2497 UInt32 PointerToLinenumbers;
2498 UInt16 NumberOfRelocations;
2499 UInt16 NumberOfLineNumbers;
2500 UInt32 Characteristics;
2504 #define sizeof_COFF_section 40
2511 UInt16 SectionNumber;
2514 UChar NumberOfAuxSymbols;
2518 #define sizeof_COFF_symbol 18
2523 UInt32 VirtualAddress;
2524 UInt32 SymbolTableIndex;
2529 #define sizeof_COFF_reloc 10
2532 /* From PE spec doc, section 3.3.2 */
2533 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2534 windows.h -- for the same purpose, but I want to know what I'm
2536 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2537 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2538 #define MYIMAGE_FILE_DLL 0x2000
2539 #define MYIMAGE_FILE_SYSTEM 0x1000
2540 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2541 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2542 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2544 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2545 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2546 #define MYIMAGE_SYM_CLASS_STATIC 3
2547 #define MYIMAGE_SYM_UNDEFINED 0
2549 /* From PE spec doc, section 4.1 */
2550 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2551 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2552 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2554 /* From PE spec doc, section 5.2.1 */
2555 #define MYIMAGE_REL_I386_DIR32 0x0006
2556 #define MYIMAGE_REL_I386_REL32 0x0014
2559 /* We use myindex to calculate array addresses, rather than
2560 simply doing the normal subscript thing. That's because
2561 some of the above structs have sizes which are not
2562 a whole number of words. GCC rounds their sizes up to a
2563 whole number of words, which means that the address calcs
2564 arising from using normal C indexing or pointer arithmetic
2565 are just plain wrong. Sigh.
2568 myindex ( int scale, void* base, int index )
2571 ((UChar*)base) + scale * index;
2576 printName ( UChar* name, UChar* strtab )
2578 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2579 UInt32 strtab_offset = * (UInt32*)(name+4);
2580 debugBelch("%s", strtab + strtab_offset );
2583 for (i = 0; i < 8; i++) {
2584 if (name[i] == 0) break;
2585 debugBelch("%c", name[i] );
2592 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2594 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2595 UInt32 strtab_offset = * (UInt32*)(name+4);
2596 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2602 if (name[i] == 0) break;
2612 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2615 /* If the string is longer than 8 bytes, look in the
2616 string table for it -- this will be correctly zero terminated.
2618 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2619 UInt32 strtab_offset = * (UInt32*)(name+4);
2620 return ((UChar*)strtab) + strtab_offset;
2622 /* Otherwise, if shorter than 8 bytes, return the original,
2623 which by defn is correctly terminated.
2625 if (name[7]==0) return name;
2626 /* The annoying case: 8 bytes. Copy into a temporary
2627 (XXX which is never freed ...)
2629 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2631 strncpy((char*)newstr,(char*)name,8);
2636 /* Getting the name of a section is mildly tricky, so we make a
2637 function for it. Sadly, in one case we have to copy the string
2638 (when it is exactly 8 bytes long there's no trailing '\0'), so for
2639 consistency we *always* copy the string; the caller must free it
2642 cstring_from_section_name (UChar* name, UChar* strtab)
2647 int strtab_offset = strtol((char*)name+1,NULL,10);
2648 int len = strlen(((char*)strtab) + strtab_offset);
2650 newstr = stgMallocBytes(len, "cstring_from_section_symbol_name");
2651 strcpy((char*)newstr, (char*)((UChar*)strtab) + strtab_offset);
2656 newstr = stgMallocBytes(9, "cstring_from_section_symbol_name");
2658 strncpy((char*)newstr,(char*)name,8);
2664 /* Just compares the short names (first 8 chars) */
2665 static COFF_section *
2666 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2670 = (COFF_header*)(oc->image);
2671 COFF_section* sectab
2673 ((UChar*)(oc->image))
2674 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2676 for (i = 0; i < hdr->NumberOfSections; i++) {
2679 COFF_section* section_i
2681 myindex ( sizeof_COFF_section, sectab, i );
2682 n1 = (UChar*) &(section_i->Name);
2684 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2685 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2686 n1[6]==n2[6] && n1[7]==n2[7])
2695 zapTrailingAtSign ( UChar* sym )
2697 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2699 if (sym[0] == 0) return;
2701 while (sym[i] != 0) i++;
2704 while (j > 0 && my_isdigit(sym[j])) j--;
2705 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2710 lookupSymbolInDLLs ( UChar *lbl )
2715 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2716 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2718 if (lbl[0] == '_') {
2719 /* HACK: if the name has an initial underscore, try stripping
2720 it off & look that up first. I've yet to verify whether there's
2721 a Rule that governs whether an initial '_' *should always* be
2722 stripped off when mapping from import lib name to the DLL name.
2724 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2726 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2730 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2732 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2741 ocVerifyImage_PEi386 ( ObjectCode* oc )
2746 COFF_section* sectab;
2747 COFF_symbol* symtab;
2749 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2750 hdr = (COFF_header*)(oc->image);
2751 sectab = (COFF_section*) (
2752 ((UChar*)(oc->image))
2753 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2755 symtab = (COFF_symbol*) (
2756 ((UChar*)(oc->image))
2757 + hdr->PointerToSymbolTable
2759 strtab = ((UChar*)symtab)
2760 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2762 if (hdr->Machine != 0x14c) {
2763 errorBelch("%s: Not x86 PEi386", oc->fileName);
2766 if (hdr->SizeOfOptionalHeader != 0) {
2767 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2770 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2771 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2772 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2773 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2774 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2777 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2778 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2779 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2781 (int)(hdr->Characteristics));
2784 /* If the string table size is way crazy, this might indicate that
2785 there are more than 64k relocations, despite claims to the
2786 contrary. Hence this test. */
2787 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2789 if ( (*(UInt32*)strtab) > 600000 ) {
2790 /* Note that 600k has no special significance other than being
2791 big enough to handle the almost-2MB-sized lumps that
2792 constitute HSwin32*.o. */
2793 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2798 /* No further verification after this point; only debug printing. */
2800 IF_DEBUG(linker, i=1);
2801 if (i == 0) return 1;
2803 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2804 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2805 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2808 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2809 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2810 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2811 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2812 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2813 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2814 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2816 /* Print the section table. */
2818 for (i = 0; i < hdr->NumberOfSections; i++) {
2820 COFF_section* sectab_i
2822 myindex ( sizeof_COFF_section, sectab, i );
2829 printName ( sectab_i->Name, strtab );
2839 sectab_i->VirtualSize,
2840 sectab_i->VirtualAddress,
2841 sectab_i->SizeOfRawData,
2842 sectab_i->PointerToRawData,
2843 sectab_i->NumberOfRelocations,
2844 sectab_i->PointerToRelocations,
2845 sectab_i->PointerToRawData
2847 reltab = (COFF_reloc*) (
2848 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2851 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2852 /* If the relocation field (a short) has overflowed, the
2853 * real count can be found in the first reloc entry.
2855 * See Section 4.1 (last para) of the PE spec (rev6.0).
2857 COFF_reloc* rel = (COFF_reloc*)
2858 myindex ( sizeof_COFF_reloc, reltab, 0 );
2859 noRelocs = rel->VirtualAddress;
2862 noRelocs = sectab_i->NumberOfRelocations;
2866 for (; j < noRelocs; j++) {
2868 COFF_reloc* rel = (COFF_reloc*)
2869 myindex ( sizeof_COFF_reloc, reltab, j );
2871 " type 0x%-4x vaddr 0x%-8x name `",
2873 rel->VirtualAddress );
2874 sym = (COFF_symbol*)
2875 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2876 /* Hmm..mysterious looking offset - what's it for? SOF */
2877 printName ( sym->Name, strtab -10 );
2884 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2885 debugBelch("---START of string table---\n");
2886 for (i = 4; i < *(Int32*)strtab; i++) {
2888 debugBelch("\n"); else
2889 debugBelch("%c", strtab[i] );
2891 debugBelch("--- END of string table---\n");
2896 COFF_symbol* symtab_i;
2897 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2898 symtab_i = (COFF_symbol*)
2899 myindex ( sizeof_COFF_symbol, symtab, i );
2905 printName ( symtab_i->Name, strtab );
2914 (Int32)(symtab_i->SectionNumber),
2915 (UInt32)symtab_i->Type,
2916 (UInt32)symtab_i->StorageClass,
2917 (UInt32)symtab_i->NumberOfAuxSymbols
2919 i += symtab_i->NumberOfAuxSymbols;
2929 ocGetNames_PEi386 ( ObjectCode* oc )
2932 COFF_section* sectab;
2933 COFF_symbol* symtab;
2940 hdr = (COFF_header*)(oc->image);
2941 sectab = (COFF_section*) (
2942 ((UChar*)(oc->image))
2943 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2945 symtab = (COFF_symbol*) (
2946 ((UChar*)(oc->image))
2947 + hdr->PointerToSymbolTable
2949 strtab = ((UChar*)(oc->image))
2950 + hdr->PointerToSymbolTable
2951 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2953 /* Allocate space for any (local, anonymous) .bss sections. */
2955 for (i = 0; i < hdr->NumberOfSections; i++) {
2958 COFF_section* sectab_i
2960 myindex ( sizeof_COFF_section, sectab, i );
2962 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
2964 if (0 != strcmp(secname, ".bss")) {
2971 /* sof 10/05: the PE spec text isn't too clear regarding what
2972 * the SizeOfRawData field is supposed to hold for object
2973 * file sections containing just uninitialized data -- for executables,
2974 * it is supposed to be zero; unclear what it's supposed to be
2975 * for object files. However, VirtualSize is guaranteed to be
2976 * zero for object files, which definitely suggests that SizeOfRawData
2977 * will be non-zero (where else would the size of this .bss section be
2978 * stored?) Looking at the COFF_section info for incoming object files,
2979 * this certainly appears to be the case.
2981 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2982 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2983 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2984 * variable decls into to the .bss section. (The specific function in Q which
2985 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2987 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2988 /* This is a non-empty .bss section. Allocate zeroed space for
2989 it, and set its PointerToRawData field such that oc->image +
2990 PointerToRawData == addr_of_zeroed_space. */
2991 bss_sz = sectab_i->VirtualSize;
2992 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2993 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2994 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2995 addProddableBlock(oc, zspace, bss_sz);
2996 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2999 /* Copy section information into the ObjectCode. */
3001 for (i = 0; i < hdr->NumberOfSections; i++) {
3007 = SECTIONKIND_OTHER;
3008 COFF_section* sectab_i
3010 myindex ( sizeof_COFF_section, sectab, i );
3012 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
3014 IF_DEBUG(linker, debugBelch("section name = %s\n", secname ));
3017 /* I'm sure this is the Right Way to do it. However, the
3018 alternative of testing the sectab_i->Name field seems to
3019 work ok with Cygwin.
3021 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
3022 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
3023 kind = SECTIONKIND_CODE_OR_RODATA;
3026 if (0==strcmp(".text",(char*)secname) ||
3027 0==strcmp(".rdata",(char*)secname)||
3028 0==strcmp(".rodata",(char*)secname))
3029 kind = SECTIONKIND_CODE_OR_RODATA;
3030 if (0==strcmp(".data",(char*)secname) ||
3031 0==strcmp(".bss",(char*)secname))
3032 kind = SECTIONKIND_RWDATA;
3034 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
3035 sz = sectab_i->SizeOfRawData;
3036 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
3038 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
3039 end = start + sz - 1;
3041 if (kind == SECTIONKIND_OTHER
3042 /* Ignore sections called which contain stabs debugging
3044 && 0 != strcmp(".stab", (char*)secname)
3045 && 0 != strcmp(".stabstr", (char*)secname)
3046 /* ignore constructor section for now */
3047 && 0 != strcmp(".ctors", (char*)secname)
3048 /* ignore section generated from .ident */
3049 && 0!= strncmp(".debug", (char*)secname, 6)
3050 /* ignore unknown section that appeared in gcc 3.4.5(?) */
3051 && 0!= strcmp(".reloc", (char*)secname)
3052 && 0 != strcmp(".rdata$zzz", (char*)secname)
3054 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", secname, oc->fileName);
3059 if (kind != SECTIONKIND_OTHER && end >= start) {
3060 addSection(oc, kind, start, end);
3061 addProddableBlock(oc, start, end - start + 1);
3067 /* Copy exported symbols into the ObjectCode. */
3069 oc->n_symbols = hdr->NumberOfSymbols;
3070 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3071 "ocGetNames_PEi386(oc->symbols)");
3072 /* Call me paranoid; I don't care. */
3073 for (i = 0; i < oc->n_symbols; i++)
3074 oc->symbols[i] = NULL;
3078 COFF_symbol* symtab_i;
3079 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
3080 symtab_i = (COFF_symbol*)
3081 myindex ( sizeof_COFF_symbol, symtab, i );
3085 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
3086 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
3087 /* This symbol is global and defined, viz, exported */
3088 /* for MYIMAGE_SYMCLASS_EXTERNAL
3089 && !MYIMAGE_SYM_UNDEFINED,
3090 the address of the symbol is:
3091 address of relevant section + offset in section
3093 COFF_section* sectabent
3094 = (COFF_section*) myindex ( sizeof_COFF_section,
3096 symtab_i->SectionNumber-1 );
3097 addr = ((UChar*)(oc->image))
3098 + (sectabent->PointerToRawData
3102 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
3103 && symtab_i->Value > 0) {
3104 /* This symbol isn't in any section at all, ie, global bss.
3105 Allocate zeroed space for it. */
3106 addr = stgCallocBytes(1, symtab_i->Value,
3107 "ocGetNames_PEi386(non-anonymous bss)");
3108 addSection(oc, SECTIONKIND_RWDATA, addr,
3109 ((UChar*)addr) + symtab_i->Value - 1);
3110 addProddableBlock(oc, addr, symtab_i->Value);
3111 /* debugBelch("BSS section at 0x%x\n", addr); */
3114 if (addr != NULL ) {
3115 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
3116 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
3117 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
3118 ASSERT(i >= 0 && i < oc->n_symbols);
3119 /* cstring_from_COFF_symbol_name always succeeds. */
3120 oc->symbols[i] = (char*)sname;
3121 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
3125 "IGNORING symbol %d\n"
3129 printName ( symtab_i->Name, strtab );
3138 (Int32)(symtab_i->SectionNumber),
3139 (UInt32)symtab_i->Type,
3140 (UInt32)symtab_i->StorageClass,
3141 (UInt32)symtab_i->NumberOfAuxSymbols
3146 i += symtab_i->NumberOfAuxSymbols;
3155 ocResolve_PEi386 ( ObjectCode* oc )
3158 COFF_section* sectab;
3159 COFF_symbol* symtab;
3169 /* ToDo: should be variable-sized? But is at least safe in the
3170 sense of buffer-overrun-proof. */
3172 /* debugBelch("resolving for %s\n", oc->fileName); */
3174 hdr = (COFF_header*)(oc->image);
3175 sectab = (COFF_section*) (
3176 ((UChar*)(oc->image))
3177 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
3179 symtab = (COFF_symbol*) (
3180 ((UChar*)(oc->image))
3181 + hdr->PointerToSymbolTable
3183 strtab = ((UChar*)(oc->image))
3184 + hdr->PointerToSymbolTable
3185 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
3187 for (i = 0; i < hdr->NumberOfSections; i++) {
3188 COFF_section* sectab_i
3190 myindex ( sizeof_COFF_section, sectab, i );
3193 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
3196 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
3198 /* Ignore sections called which contain stabs debugging
3200 if (0 == strcmp(".stab", (char*)secname)
3201 || 0 == strcmp(".stabstr", (char*)secname)
3202 || 0 == strcmp(".ctors", (char*)secname)
3203 || 0 == strncmp(".debug", (char*)secname, 6)
3204 || 0 == strcmp(".rdata$zzz", (char*)secname)) {
3211 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
3212 /* If the relocation field (a short) has overflowed, the
3213 * real count can be found in the first reloc entry.
3215 * See Section 4.1 (last para) of the PE spec (rev6.0).
3217 * Nov2003 update: the GNU linker still doesn't correctly
3218 * handle the generation of relocatable object files with
3219 * overflown relocations. Hence the output to warn of potential
3222 COFF_reloc* rel = (COFF_reloc*)
3223 myindex ( sizeof_COFF_reloc, reltab, 0 );
3224 noRelocs = rel->VirtualAddress;
3226 /* 10/05: we now assume (and check for) a GNU ld that is capable
3227 * of handling object files with (>2^16) of relocs.
3230 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
3235 noRelocs = sectab_i->NumberOfRelocations;
3240 for (; j < noRelocs; j++) {
3242 COFF_reloc* reltab_j
3244 myindex ( sizeof_COFF_reloc, reltab, j );
3246 /* the location to patch */
3248 ((UChar*)(oc->image))
3249 + (sectab_i->PointerToRawData
3250 + reltab_j->VirtualAddress
3251 - sectab_i->VirtualAddress )
3253 /* the existing contents of pP */
3255 /* the symbol to connect to */
3256 sym = (COFF_symbol*)
3257 myindex ( sizeof_COFF_symbol,
3258 symtab, reltab_j->SymbolTableIndex );
3261 "reloc sec %2d num %3d: type 0x%-4x "
3262 "vaddr 0x%-8x name `",
3264 (UInt32)reltab_j->Type,
3265 reltab_j->VirtualAddress );
3266 printName ( sym->Name, strtab );
3267 debugBelch("'\n" ));
3269 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
3270 COFF_section* section_sym
3271 = findPEi386SectionCalled ( oc, sym->Name );
3273 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
3276 S = ((UInt32)(oc->image))
3277 + (section_sym->PointerToRawData
3280 copyName ( sym->Name, strtab, symbol, 1000-1 );
3281 S = (UInt32) lookupSymbol( (char*)symbol );
3282 if ((void*)S != NULL) goto foundit;
3283 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3287 checkProddableBlock(oc, pP);
3288 switch (reltab_j->Type) {
3289 case MYIMAGE_REL_I386_DIR32:
3292 case MYIMAGE_REL_I386_REL32:
3293 /* Tricky. We have to insert a displacement at
3294 pP which, when added to the PC for the _next_
3295 insn, gives the address of the target (S).
3296 Problem is to know the address of the next insn
3297 when we only know pP. We assume that this
3298 literal field is always the last in the insn,
3299 so that the address of the next insn is pP+4
3300 -- hence the constant 4.
3301 Also I don't know if A should be added, but so
3302 far it has always been zero.
3304 SOF 05/2005: 'A' (old contents of *pP) have been observed
3305 to contain values other than zero (the 'wx' object file
3306 that came with wxhaskell-0.9.4; dunno how it was compiled..).
3307 So, add displacement to old value instead of asserting
3308 A to be zero. Fixes wxhaskell-related crashes, and no other
3309 ill effects have been observed.
3311 Update: the reason why we're seeing these more elaborate
3312 relocations is due to a switch in how the NCG compiles SRTs
3313 and offsets to them from info tables. SRTs live in .(ro)data,
3314 while info tables live in .text, causing GAS to emit REL32/DISP32
3315 relocations with non-zero values. Adding the displacement is
3316 the right thing to do.
3318 *pP = S - ((UInt32)pP) - 4 + A;
3321 debugBelch("%s: unhandled PEi386 relocation type %d",
3322 oc->fileName, reltab_j->Type);
3329 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
3333 #endif /* defined(OBJFORMAT_PEi386) */
3336 /* --------------------------------------------------------------------------
3338 * ------------------------------------------------------------------------*/
3340 #if defined(OBJFORMAT_ELF)
3345 #if defined(sparc_HOST_ARCH)
3346 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
3347 #elif defined(i386_HOST_ARCH)
3348 # define ELF_TARGET_386 /* Used inside <elf.h> */
3349 #elif defined(x86_64_HOST_ARCH)
3350 # define ELF_TARGET_X64_64
3354 #if !defined(openbsd_HOST_OS)
3357 /* openbsd elf has things in different places, with diff names */
3358 # include <elf_abi.h>
3359 # include <machine/reloc.h>
3360 # define R_386_32 RELOC_32
3361 # define R_386_PC32 RELOC_PC32
3364 /* If elf.h doesn't define it */
3365 # ifndef R_X86_64_PC64
3366 # define R_X86_64_PC64 24
3370 * Define a set of types which can be used for both ELF32 and ELF64
3374 #define ELFCLASS ELFCLASS64
3375 #define Elf_Addr Elf64_Addr
3376 #define Elf_Word Elf64_Word
3377 #define Elf_Sword Elf64_Sword
3378 #define Elf_Ehdr Elf64_Ehdr
3379 #define Elf_Phdr Elf64_Phdr
3380 #define Elf_Shdr Elf64_Shdr
3381 #define Elf_Sym Elf64_Sym
3382 #define Elf_Rel Elf64_Rel
3383 #define Elf_Rela Elf64_Rela
3385 #define ELF_ST_TYPE ELF64_ST_TYPE
3388 #define ELF_ST_BIND ELF64_ST_BIND
3391 #define ELF_R_TYPE ELF64_R_TYPE
3394 #define ELF_R_SYM ELF64_R_SYM
3397 #define ELFCLASS ELFCLASS32
3398 #define Elf_Addr Elf32_Addr
3399 #define Elf_Word Elf32_Word
3400 #define Elf_Sword Elf32_Sword
3401 #define Elf_Ehdr Elf32_Ehdr
3402 #define Elf_Phdr Elf32_Phdr
3403 #define Elf_Shdr Elf32_Shdr
3404 #define Elf_Sym Elf32_Sym
3405 #define Elf_Rel Elf32_Rel
3406 #define Elf_Rela Elf32_Rela
3408 #define ELF_ST_TYPE ELF32_ST_TYPE
3411 #define ELF_ST_BIND ELF32_ST_BIND
3414 #define ELF_R_TYPE ELF32_R_TYPE
3417 #define ELF_R_SYM ELF32_R_SYM
3423 * Functions to allocate entries in dynamic sections. Currently we simply
3424 * preallocate a large number, and we don't check if a entry for the given
3425 * target already exists (a linear search is too slow). Ideally these
3426 * entries would be associated with symbols.
3429 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3430 #define GOT_SIZE 0x20000
3431 #define FUNCTION_TABLE_SIZE 0x10000
3432 #define PLT_SIZE 0x08000
3435 static Elf_Addr got[GOT_SIZE];
3436 static unsigned int gotIndex;
3437 static Elf_Addr gp_val = (Elf_Addr)got;
3440 allocateGOTEntry(Elf_Addr target)
3444 if (gotIndex >= GOT_SIZE)
3445 barf("Global offset table overflow");
3447 entry = &got[gotIndex++];
3449 return (Elf_Addr)entry;
3453 #ifdef ELF_FUNCTION_DESC
3459 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3460 static unsigned int functionTableIndex;
3463 allocateFunctionDesc(Elf_Addr target)
3465 FunctionDesc *entry;
3467 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3468 barf("Function table overflow");
3470 entry = &functionTable[functionTableIndex++];
3472 entry->gp = (Elf_Addr)gp_val;
3473 return (Elf_Addr)entry;
3477 copyFunctionDesc(Elf_Addr target)
3479 FunctionDesc *olddesc = (FunctionDesc *)target;
3480 FunctionDesc *newdesc;
3482 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3483 newdesc->gp = olddesc->gp;
3484 return (Elf_Addr)newdesc;
3491 unsigned char code[sizeof(plt_code)];
3495 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3497 PLTEntry *plt = (PLTEntry *)oc->plt;
3500 if (oc->pltIndex >= PLT_SIZE)
3501 barf("Procedure table overflow");
3503 entry = &plt[oc->pltIndex++];
3504 memcpy(entry->code, plt_code, sizeof(entry->code));
3505 PLT_RELOC(entry->code, target);
3506 return (Elf_Addr)entry;
3512 return (PLT_SIZE * sizeof(PLTEntry));
3518 * Generic ELF functions
3522 findElfSection ( void* objImage, Elf_Word sh_type )
3524 char* ehdrC = (char*)objImage;
3525 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3526 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3527 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3531 for (i = 0; i < ehdr->e_shnum; i++) {
3532 if (shdr[i].sh_type == sh_type
3533 /* Ignore the section header's string table. */
3534 && i != ehdr->e_shstrndx
3535 /* Ignore string tables named .stabstr, as they contain
3537 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3539 ptr = ehdrC + shdr[i].sh_offset;
3547 ocVerifyImage_ELF ( ObjectCode* oc )
3551 int i, j, nent, nstrtab, nsymtabs;
3555 char* ehdrC = (char*)(oc->image);
3556 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3558 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3559 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3560 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3561 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3562 errorBelch("%s: not an ELF object", oc->fileName);
3566 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3567 errorBelch("%s: unsupported ELF format", oc->fileName);
3571 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3572 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3574 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3575 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3577 errorBelch("%s: unknown endiannness", oc->fileName);
3581 if (ehdr->e_type != ET_REL) {
3582 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3585 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3587 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3588 switch (ehdr->e_machine) {
3589 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3590 #ifdef EM_SPARC32PLUS
3591 case EM_SPARC32PLUS:
3593 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3595 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3597 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3599 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3600 #elif defined(EM_AMD64)
3601 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3603 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3604 errorBelch("%s: unknown architecture (e_machine == %d)"
3605 , oc->fileName, ehdr->e_machine);
3609 IF_DEBUG(linker,debugBelch(
3610 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3611 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3613 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3615 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3617 if (ehdr->e_shstrndx == SHN_UNDEF) {
3618 errorBelch("%s: no section header string table", oc->fileName);
3621 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3623 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3626 for (i = 0; i < ehdr->e_shnum; i++) {
3627 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3628 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3629 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3630 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3631 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3632 ehdrC + shdr[i].sh_offset,
3633 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3635 if (shdr[i].sh_type == SHT_REL) {
3636 IF_DEBUG(linker,debugBelch("Rel " ));
3637 } else if (shdr[i].sh_type == SHT_RELA) {
3638 IF_DEBUG(linker,debugBelch("RelA " ));
3640 IF_DEBUG(linker,debugBelch(" "));
3643 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3647 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3650 for (i = 0; i < ehdr->e_shnum; i++) {
3651 if (shdr[i].sh_type == SHT_STRTAB
3652 /* Ignore the section header's string table. */
3653 && i != ehdr->e_shstrndx
3654 /* Ignore string tables named .stabstr, as they contain
3656 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3658 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3659 strtab = ehdrC + shdr[i].sh_offset;
3664 errorBelch("%s: no string tables, or too many", oc->fileName);
3669 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3670 for (i = 0; i < ehdr->e_shnum; i++) {
3671 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3672 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3674 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3675 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3676 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3678 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3680 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3681 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3684 for (j = 0; j < nent; j++) {
3685 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3686 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3687 (int)stab[j].st_shndx,
3688 (int)stab[j].st_size,
3689 (char*)stab[j].st_value ));
3691 IF_DEBUG(linker,debugBelch("type=" ));
3692 switch (ELF_ST_TYPE(stab[j].st_info)) {
3693 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3694 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3695 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3696 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3697 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3698 default: IF_DEBUG(linker,debugBelch("? " )); break;
3700 IF_DEBUG(linker,debugBelch(" " ));
3702 IF_DEBUG(linker,debugBelch("bind=" ));
3703 switch (ELF_ST_BIND(stab[j].st_info)) {
3704 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3705 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3706 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3707 default: IF_DEBUG(linker,debugBelch("? " )); break;
3709 IF_DEBUG(linker,debugBelch(" " ));
3711 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3715 if (nsymtabs == 0) {
3716 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3723 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3727 if (hdr->sh_type == SHT_PROGBITS
3728 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3729 /* .text-style section */
3730 return SECTIONKIND_CODE_OR_RODATA;
3733 if (hdr->sh_type == SHT_PROGBITS
3734 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3735 /* .data-style section */
3736 return SECTIONKIND_RWDATA;
3739 if (hdr->sh_type == SHT_PROGBITS
3740 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3741 /* .rodata-style section */
3742 return SECTIONKIND_CODE_OR_RODATA;
3745 if (hdr->sh_type == SHT_NOBITS
3746 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3747 /* .bss-style section */
3749 return SECTIONKIND_RWDATA;
3752 return SECTIONKIND_OTHER;
3757 ocGetNames_ELF ( ObjectCode* oc )
3762 char* ehdrC = (char*)(oc->image);
3763 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3764 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3765 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3767 ASSERT(symhash != NULL);
3770 errorBelch("%s: no strtab", oc->fileName);
3775 for (i = 0; i < ehdr->e_shnum; i++) {
3776 /* Figure out what kind of section it is. Logic derived from
3777 Figure 1.14 ("Special Sections") of the ELF document
3778 ("Portable Formats Specification, Version 1.1"). */
3780 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3782 if (is_bss && shdr[i].sh_size > 0) {
3783 /* This is a non-empty .bss section. Allocate zeroed space for
3784 it, and set its .sh_offset field such that
3785 ehdrC + .sh_offset == addr_of_zeroed_space. */
3786 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3787 "ocGetNames_ELF(BSS)");
3788 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3790 debugBelch("BSS section at 0x%x, size %d\n",
3791 zspace, shdr[i].sh_size);
3795 /* fill in the section info */
3796 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3797 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3798 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3799 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3802 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3804 /* copy stuff into this module's object symbol table */
3805 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3806 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3808 oc->n_symbols = nent;
3809 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3810 "ocGetNames_ELF(oc->symbols)");
3812 for (j = 0; j < nent; j++) {
3814 char isLocal = FALSE; /* avoids uninit-var warning */
3816 char* nm = strtab + stab[j].st_name;
3817 int secno = stab[j].st_shndx;
3819 /* Figure out if we want to add it; if so, set ad to its
3820 address. Otherwise leave ad == NULL. */
3822 if (secno == SHN_COMMON) {
3824 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3826 debugBelch("COMMON symbol, size %d name %s\n",
3827 stab[j].st_size, nm);
3829 /* Pointless to do addProddableBlock() for this area,
3830 since the linker should never poke around in it. */
3833 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3834 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3836 /* and not an undefined symbol */
3837 && stab[j].st_shndx != SHN_UNDEF
3838 /* and not in a "special section" */
3839 && stab[j].st_shndx < SHN_LORESERVE
3841 /* and it's a not a section or string table or anything silly */
3842 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3843 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3844 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3847 /* Section 0 is the undefined section, hence > and not >=. */
3848 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3850 if (shdr[secno].sh_type == SHT_NOBITS) {
3851 debugBelch(" BSS symbol, size %d off %d name %s\n",
3852 stab[j].st_size, stab[j].st_value, nm);
3855 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3856 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3859 #ifdef ELF_FUNCTION_DESC
3860 /* dlsym() and the initialisation table both give us function
3861 * descriptors, so to be consistent we store function descriptors
3862 * in the symbol table */
3863 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3864 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3866 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3867 ad, oc->fileName, nm ));
3872 /* And the decision is ... */
3876 oc->symbols[j] = nm;
3879 /* Ignore entirely. */
3881 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3885 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3886 strtab + stab[j].st_name ));
3889 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3890 (int)ELF_ST_BIND(stab[j].st_info),
3891 (int)ELF_ST_TYPE(stab[j].st_info),
3892 (int)stab[j].st_shndx,
3893 strtab + stab[j].st_name
3896 oc->symbols[j] = NULL;
3905 /* Do ELF relocations which lack an explicit addend. All x86-linux
3906 relocations appear to be of this form. */
3908 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3909 Elf_Shdr* shdr, int shnum,
3910 Elf_Sym* stab, char* strtab )
3915 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3916 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3917 int target_shndx = shdr[shnum].sh_info;
3918 int symtab_shndx = shdr[shnum].sh_link;
3920 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3921 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3922 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3923 target_shndx, symtab_shndx ));
3925 /* Skip sections that we're not interested in. */
3928 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3929 if (kind == SECTIONKIND_OTHER) {
3930 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3935 for (j = 0; j < nent; j++) {
3936 Elf_Addr offset = rtab[j].r_offset;
3937 Elf_Addr info = rtab[j].r_info;
3939 Elf_Addr P = ((Elf_Addr)targ) + offset;
3940 Elf_Word* pP = (Elf_Word*)P;
3945 StgStablePtr stablePtr;
3948 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3949 j, (void*)offset, (void*)info ));
3951 IF_DEBUG(linker,debugBelch( " ZERO" ));
3954 Elf_Sym sym = stab[ELF_R_SYM(info)];
3955 /* First see if it is a local symbol. */
3956 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3957 /* Yes, so we can get the address directly from the ELF symbol
3959 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3961 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3962 + stab[ELF_R_SYM(info)].st_value);
3965 symbol = strtab + sym.st_name;
3966 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3967 if (NULL == stablePtr) {
3968 /* No, so look up the name in our global table. */
3969 S_tmp = lookupSymbol( symbol );
3970 S = (Elf_Addr)S_tmp;
3972 stableVal = deRefStablePtr( stablePtr );
3974 S = (Elf_Addr)S_tmp;
3978 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3981 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3984 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3985 (void*)P, (void*)S, (void*)A ));
3986 checkProddableBlock ( oc, pP );
3990 switch (ELF_R_TYPE(info)) {
3991 # ifdef i386_HOST_ARCH
3992 case R_386_32: *pP = value; break;
3993 case R_386_PC32: *pP = value - P; break;
3996 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3997 oc->fileName, (lnat)ELF_R_TYPE(info));
4005 /* Do ELF relocations for which explicit addends are supplied.
4006 sparc-solaris relocations appear to be of this form. */
4008 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
4009 Elf_Shdr* shdr, int shnum,
4010 Elf_Sym* stab, char* strtab )
4013 char *symbol = NULL;
4015 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
4016 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
4017 int target_shndx = shdr[shnum].sh_info;
4018 int symtab_shndx = shdr[shnum].sh_link;
4020 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
4021 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
4022 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
4023 target_shndx, symtab_shndx ));
4025 for (j = 0; j < nent; j++) {
4026 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
4027 /* This #ifdef only serves to avoid unused-var warnings. */
4028 Elf_Addr offset = rtab[j].r_offset;
4029 Elf_Addr P = targ + offset;
4031 Elf_Addr info = rtab[j].r_info;
4032 Elf_Addr A = rtab[j].r_addend;
4036 # if defined(sparc_HOST_ARCH)
4037 Elf_Word* pP = (Elf_Word*)P;
4039 # elif defined(powerpc_HOST_ARCH)
4043 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
4044 j, (void*)offset, (void*)info,
4047 IF_DEBUG(linker,debugBelch( " ZERO" ));
4050 Elf_Sym sym = stab[ELF_R_SYM(info)];
4051 /* First see if it is a local symbol. */
4052 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
4053 /* Yes, so we can get the address directly from the ELF symbol
4055 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
4057 (ehdrC + shdr[ sym.st_shndx ].sh_offset
4058 + stab[ELF_R_SYM(info)].st_value);
4059 #ifdef ELF_FUNCTION_DESC
4060 /* Make a function descriptor for this function */
4061 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
4062 S = allocateFunctionDesc(S + A);
4067 /* No, so look up the name in our global table. */
4068 symbol = strtab + sym.st_name;
4069 S_tmp = lookupSymbol( symbol );
4070 S = (Elf_Addr)S_tmp;
4072 #ifdef ELF_FUNCTION_DESC
4073 /* If a function, already a function descriptor - we would
4074 have to copy it to add an offset. */
4075 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
4076 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
4080 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
4083 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
4086 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
4087 (void*)P, (void*)S, (void*)A ));
4088 /* checkProddableBlock ( oc, (void*)P ); */
4092 switch (ELF_R_TYPE(info)) {
4093 # if defined(sparc_HOST_ARCH)
4094 case R_SPARC_WDISP30:
4095 w1 = *pP & 0xC0000000;
4096 w2 = (Elf_Word)((value - P) >> 2);
4097 ASSERT((w2 & 0xC0000000) == 0);
4102 w1 = *pP & 0xFFC00000;
4103 w2 = (Elf_Word)(value >> 10);
4104 ASSERT((w2 & 0xFFC00000) == 0);
4110 w2 = (Elf_Word)(value & 0x3FF);
4111 ASSERT((w2 & ~0x3FF) == 0);
4116 /* According to the Sun documentation:
4118 This relocation type resembles R_SPARC_32, except it refers to an
4119 unaligned word. That is, the word to be relocated must be treated
4120 as four separate bytes with arbitrary alignment, not as a word
4121 aligned according to the architecture requirements.
4124 w2 = (Elf_Word)value;
4126 // SPARC doesn't do misaligned writes of 32 bit words,
4127 // so we have to do this one byte-at-a-time.
4128 char *pPc = (char*)pP;
4129 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
4130 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
4131 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
4132 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
4136 w2 = (Elf_Word)value;
4139 # elif defined(powerpc_HOST_ARCH)
4140 case R_PPC_ADDR16_LO:
4141 *(Elf32_Half*) P = value;
4144 case R_PPC_ADDR16_HI:
4145 *(Elf32_Half*) P = value >> 16;
4148 case R_PPC_ADDR16_HA:
4149 *(Elf32_Half*) P = (value + 0x8000) >> 16;
4153 *(Elf32_Word *) P = value;
4157 *(Elf32_Word *) P = value - P;
4163 if( delta << 6 >> 6 != delta )
4165 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
4169 if( value == 0 || delta << 6 >> 6 != delta )
4171 barf( "Unable to make SymbolExtra for #%d",
4177 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
4178 | (delta & 0x3fffffc);
4182 #if x86_64_HOST_ARCH
4184 *(Elf64_Xword *)P = value;
4189 #if defined(ALWAYS_PIC)
4190 barf("R_X86_64_PC32 relocation, but ALWAYS_PIC.");
4192 StgInt64 off = value - P;
4193 if (off >= 0x7fffffffL || off < -0x80000000L) {
4194 #if X86_64_ELF_NONPIC_HACK
4195 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4197 off = pltAddress + A - P;
4199 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4200 symbol, off, oc->fileName );
4203 *(Elf64_Word *)P = (Elf64_Word)off;
4210 StgInt64 off = value - P;
4211 *(Elf64_Word *)P = (Elf64_Word)off;
4216 #if defined(ALWAYS_PIC)
4217 barf("R_X86_64_32 relocation, but ALWAYS_PIC.");
4219 if (value >= 0x7fffffffL) {
4220 #if X86_64_ELF_NONPIC_HACK
4221 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4223 value = pltAddress + A;
4225 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4226 symbol, value, oc->fileName );
4229 *(Elf64_Word *)P = (Elf64_Word)value;
4234 #if defined(ALWAYS_PIC)
4235 barf("R_X86_64_32S relocation, but ALWAYS_PIC.");
4237 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
4238 #if X86_64_ELF_NONPIC_HACK
4239 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4241 value = pltAddress + A;
4243 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4244 symbol, value, oc->fileName );
4247 *(Elf64_Sword *)P = (Elf64_Sword)value;
4251 case R_X86_64_GOTPCREL:
4253 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
4254 StgInt64 off = gotAddress + A - P;
4255 *(Elf64_Word *)P = (Elf64_Word)off;
4259 case R_X86_64_PLT32:
4261 #if defined(ALWAYS_PIC)
4262 barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC.");
4264 StgInt64 off = value - P;
4265 if (off >= 0x7fffffffL || off < -0x80000000L) {
4266 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4268 off = pltAddress + A - P;
4270 *(Elf64_Word *)P = (Elf64_Word)off;
4277 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
4278 oc->fileName, (lnat)ELF_R_TYPE(info));
4287 ocResolve_ELF ( ObjectCode* oc )
4291 Elf_Sym* stab = NULL;
4292 char* ehdrC = (char*)(oc->image);
4293 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
4294 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
4296 /* first find "the" symbol table */
4297 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
4299 /* also go find the string table */
4300 strtab = findElfSection ( ehdrC, SHT_STRTAB );
4302 if (stab == NULL || strtab == NULL) {
4303 errorBelch("%s: can't find string or symbol table", oc->fileName);
4307 /* Process the relocation sections. */
4308 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
4309 if (shdr[shnum].sh_type == SHT_REL) {
4310 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
4311 shnum, stab, strtab );
4315 if (shdr[shnum].sh_type == SHT_RELA) {
4316 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
4317 shnum, stab, strtab );
4322 #if defined(powerpc_HOST_ARCH)
4323 ocFlushInstructionCache( oc );
4330 * PowerPC & X86_64 ELF specifics
4333 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
4335 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
4341 ehdr = (Elf_Ehdr *) oc->image;
4342 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
4344 for( i = 0; i < ehdr->e_shnum; i++ )
4345 if( shdr[i].sh_type == SHT_SYMTAB )
4348 if( i == ehdr->e_shnum )
4350 errorBelch( "This ELF file contains no symtab" );
4354 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
4356 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
4357 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
4362 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
4365 #endif /* powerpc */
4369 /* --------------------------------------------------------------------------
4371 * ------------------------------------------------------------------------*/
4373 #if defined(OBJFORMAT_MACHO)
4376 Support for MachO linking on Darwin/MacOS X
4377 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
4379 I hereby formally apologize for the hackish nature of this code.
4380 Things that need to be done:
4381 *) implement ocVerifyImage_MachO
4382 *) add still more sanity checks.
4385 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4386 #define mach_header mach_header_64
4387 #define segment_command segment_command_64
4388 #define section section_64
4389 #define nlist nlist_64
4392 #ifdef powerpc_HOST_ARCH
4393 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4395 struct mach_header *header = (struct mach_header *) oc->image;
4396 struct load_command *lc = (struct load_command *) (header + 1);
4399 for( i = 0; i < header->ncmds; i++ )
4401 if( lc->cmd == LC_SYMTAB )
4403 // Find out the first and last undefined external
4404 // symbol, so we don't have to allocate too many
4406 struct symtab_command *symLC = (struct symtab_command *) lc;
4407 unsigned min = symLC->nsyms, max = 0;
4408 struct nlist *nlist =
4409 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
4411 for(i=0;i<symLC->nsyms;i++)
4413 if(nlist[i].n_type & N_STAB)
4415 else if(nlist[i].n_type & N_EXT)
4417 if((nlist[i].n_type & N_TYPE) == N_UNDF
4418 && (nlist[i].n_value == 0))
4428 return ocAllocateSymbolExtras(oc, max - min + 1, min);
4433 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4435 return ocAllocateSymbolExtras(oc,0,0);
4438 #ifdef x86_64_HOST_ARCH
4439 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4441 struct mach_header *header = (struct mach_header *) oc->image;
4442 struct load_command *lc = (struct load_command *) (header + 1);
4445 for( i = 0; i < header->ncmds; i++ )
4447 if( lc->cmd == LC_SYMTAB )
4449 // Just allocate one entry for every symbol
4450 struct symtab_command *symLC = (struct symtab_command *) lc;
4452 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4455 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4457 return ocAllocateSymbolExtras(oc,0,0);
4461 static int ocVerifyImage_MachO(ObjectCode* oc)
4463 char *image = (char*) oc->image;
4464 struct mach_header *header = (struct mach_header*) image;
4466 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4467 if(header->magic != MH_MAGIC_64) {
4468 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4469 oc->fileName, MH_MAGIC_64, header->magic);
4473 if(header->magic != MH_MAGIC) {
4474 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4475 oc->fileName, MH_MAGIC, header->magic);
4479 // FIXME: do some more verifying here
4483 static int resolveImports(
4486 struct symtab_command *symLC,
4487 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
4488 unsigned long *indirectSyms,
4489 struct nlist *nlist)
4492 size_t itemSize = 4;
4494 IF_DEBUG(linker, debugBelch("resolveImports: start\n"));
4497 int isJumpTable = 0;
4498 if(!strcmp(sect->sectname,"__jump_table"))
4502 ASSERT(sect->reserved2 == itemSize);
4506 for(i=0; i*itemSize < sect->size;i++)
4508 // according to otool, reserved1 contains the first index into the indirect symbol table
4509 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4510 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4513 IF_DEBUG(linker, debugBelch("resolveImports: resolving %s\n", nm));
4514 if ((symbol->n_type & N_TYPE) == N_UNDF
4515 && (symbol->n_type & N_EXT) && (symbol->n_value != 0)) {
4516 addr = (void*) (symbol->n_value);
4517 IF_DEBUG(linker, debugBelch("resolveImports: undefined external %s has value %p\n", nm, addr));
4519 addr = lookupSymbol(nm);
4520 IF_DEBUG(linker, debugBelch("resolveImports: looking up %s, %p\n", nm, addr));
4524 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4532 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4533 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4534 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4535 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4540 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4541 ((void**)(image + sect->offset))[i] = addr;
4545 IF_DEBUG(linker, debugBelch("resolveImports: done\n"));
4549 static unsigned long relocateAddress(
4552 struct section* sections,
4553 unsigned long address)
4556 IF_DEBUG(linker, debugBelch("relocateAddress: start\n"));
4557 for (i = 0; i < nSections; i++)
4559 IF_DEBUG(linker, debugBelch(" relocating address in section %d\n", i));
4560 if (sections[i].addr <= address
4561 && address < sections[i].addr + sections[i].size)
4563 return (unsigned long)oc->image
4564 + sections[i].offset + address - sections[i].addr;
4567 barf("Invalid Mach-O file:"
4568 "Address out of bounds while relocating object file");
4572 static int relocateSection(
4575 struct symtab_command *symLC, struct nlist *nlist,
4576 int nSections, struct section* sections, struct section *sect)
4578 struct relocation_info *relocs;
4581 IF_DEBUG(linker, debugBelch("relocateSection: start\n"));
4583 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4585 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4587 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4589 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4593 IF_DEBUG(linker, debugBelch("relocateSection: number of relocations: %d\n", n));
4595 relocs = (struct relocation_info*) (image + sect->reloff);
4599 #ifdef x86_64_HOST_ARCH
4600 struct relocation_info *reloc = &relocs[i];
4602 char *thingPtr = image + sect->offset + reloc->r_address;
4604 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4605 complains that it may be used uninitialized if we don't */
4608 int type = reloc->r_type;
4610 checkProddableBlock(oc,thingPtr);
4611 switch(reloc->r_length)
4614 thing = *(uint8_t*)thingPtr;
4615 baseValue = (uint64_t)thingPtr + 1;
4618 thing = *(uint16_t*)thingPtr;
4619 baseValue = (uint64_t)thingPtr + 2;
4622 thing = *(uint32_t*)thingPtr;
4623 baseValue = (uint64_t)thingPtr + 4;
4626 thing = *(uint64_t*)thingPtr;
4627 baseValue = (uint64_t)thingPtr + 8;
4630 barf("Unknown size.");
4634 debugBelch("relocateSection: length = %d, thing = %d, baseValue = %p\n",
4635 reloc->r_length, thing, baseValue));
4637 if (type == X86_64_RELOC_GOT
4638 || type == X86_64_RELOC_GOT_LOAD)
4640 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4641 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4643 IF_DEBUG(linker, debugBelch("relocateSection: making jump island for %s, extern = %d, X86_64_RELOC_GOT\n", nm, reloc->r_extern));
4644 ASSERT(reloc->r_extern);
4645 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, (unsigned long)lookupSymbol(nm))->addr;
4647 type = X86_64_RELOC_SIGNED;
4649 else if(reloc->r_extern)
4651 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4652 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4654 IF_DEBUG(linker, debugBelch("relocateSection: looking up external symbol %s\n", nm));
4655 IF_DEBUG(linker, debugBelch(" : type = %d\n", symbol->n_type));
4656 IF_DEBUG(linker, debugBelch(" : sect = %d\n", symbol->n_sect));
4657 IF_DEBUG(linker, debugBelch(" : desc = %d\n", symbol->n_desc));
4658 IF_DEBUG(linker, debugBelch(" : value = %d\n", symbol->n_value));
4659 if ((symbol->n_type & N_TYPE) == N_SECT) {
4660 value = relocateAddress(oc, nSections, sections,
4662 IF_DEBUG(linker, debugBelch("relocateSection, defined external symbol %s, relocated address %p\n", nm, value));
4665 value = (uint64_t) lookupSymbol(nm);
4666 IF_DEBUG(linker, debugBelch("relocateSection: external symbol %s, address %p\n", nm, value));
4671 value = sections[reloc->r_symbolnum-1].offset
4672 - sections[reloc->r_symbolnum-1].addr
4676 IF_DEBUG(linker, debugBelch("relocateSection: value = %p\n", value));
4678 if (type == X86_64_RELOC_BRANCH)
4680 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4682 ASSERT(reloc->r_extern);
4683 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4686 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4687 type = X86_64_RELOC_SIGNED;
4692 case X86_64_RELOC_UNSIGNED:
4693 ASSERT(!reloc->r_pcrel);
4696 case X86_64_RELOC_SIGNED:
4697 case X86_64_RELOC_SIGNED_1:
4698 case X86_64_RELOC_SIGNED_2:
4699 case X86_64_RELOC_SIGNED_4:
4700 ASSERT(reloc->r_pcrel);
4701 thing += value - baseValue;
4703 case X86_64_RELOC_SUBTRACTOR:
4704 ASSERT(!reloc->r_pcrel);
4708 barf("unkown relocation");
4711 switch(reloc->r_length)
4714 *(uint8_t*)thingPtr = thing;
4717 *(uint16_t*)thingPtr = thing;
4720 *(uint32_t*)thingPtr = thing;
4723 *(uint64_t*)thingPtr = thing;
4727 if(relocs[i].r_address & R_SCATTERED)
4729 struct scattered_relocation_info *scat =
4730 (struct scattered_relocation_info*) &relocs[i];
4734 if(scat->r_length == 2)
4736 unsigned long word = 0;
4737 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4738 checkProddableBlock(oc,wordPtr);
4740 // Note on relocation types:
4741 // i386 uses the GENERIC_RELOC_* types,
4742 // while ppc uses special PPC_RELOC_* types.
4743 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4744 // in both cases, all others are different.
4745 // Therefore, we use GENERIC_RELOC_VANILLA
4746 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4747 // and use #ifdefs for the other types.
4749 // Step 1: Figure out what the relocated value should be
4750 if(scat->r_type == GENERIC_RELOC_VANILLA)
4752 word = *wordPtr + (unsigned long) relocateAddress(
4759 #ifdef powerpc_HOST_ARCH
4760 else if(scat->r_type == PPC_RELOC_SECTDIFF
4761 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4762 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4763 || scat->r_type == PPC_RELOC_HA16_SECTDIFF
4764 || scat->r_type == PPC_RELOC_LOCAL_SECTDIFF)
4766 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4767 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4770 struct scattered_relocation_info *pair =
4771 (struct scattered_relocation_info*) &relocs[i+1];
4773 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4774 barf("Invalid Mach-O file: "
4775 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4777 word = (unsigned long)
4778 (relocateAddress(oc, nSections, sections, scat->r_value)
4779 - relocateAddress(oc, nSections, sections, pair->r_value));
4782 #ifdef powerpc_HOST_ARCH
4783 else if(scat->r_type == PPC_RELOC_HI16
4784 || scat->r_type == PPC_RELOC_LO16
4785 || scat->r_type == PPC_RELOC_HA16
4786 || scat->r_type == PPC_RELOC_LO14)
4787 { // these are generated by label+offset things
4788 struct relocation_info *pair = &relocs[i+1];
4789 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4790 barf("Invalid Mach-O file: "
4791 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4793 if(scat->r_type == PPC_RELOC_LO16)
4795 word = ((unsigned short*) wordPtr)[1];
4796 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4798 else if(scat->r_type == PPC_RELOC_LO14)
4800 barf("Unsupported Relocation: PPC_RELOC_LO14");
4801 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4802 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4804 else if(scat->r_type == PPC_RELOC_HI16)
4806 word = ((unsigned short*) wordPtr)[1] << 16;
4807 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4809 else if(scat->r_type == PPC_RELOC_HA16)
4811 word = ((unsigned short*) wordPtr)[1] << 16;
4812 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4816 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4824 barf ("Don't know how to handle this Mach-O "
4825 "scattered relocation entry: "
4826 "object file %s; entry type %ld; "
4828 OC_INFORMATIVE_FILENAME(oc),
4834 #ifdef powerpc_HOST_ARCH
4835 if(scat->r_type == GENERIC_RELOC_VANILLA
4836 || scat->r_type == PPC_RELOC_SECTDIFF)
4838 if(scat->r_type == GENERIC_RELOC_VANILLA
4839 || scat->r_type == GENERIC_RELOC_SECTDIFF
4840 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4845 #ifdef powerpc_HOST_ARCH
4846 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4848 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4850 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4852 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4854 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4856 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4857 + ((word & (1<<15)) ? 1 : 0);
4863 barf("Can't handle Mach-O scattered relocation entry "
4864 "with this r_length tag: "
4865 "object file %s; entry type %ld; "
4866 "r_length tag %ld; address %#lx\n",
4867 OC_INFORMATIVE_FILENAME(oc),
4874 else /* scat->r_pcrel */
4876 barf("Don't know how to handle *PC-relative* Mach-O "
4877 "scattered relocation entry: "
4878 "object file %s; entry type %ld; address %#lx\n",
4879 OC_INFORMATIVE_FILENAME(oc),
4886 else /* !(relocs[i].r_address & R_SCATTERED) */
4888 struct relocation_info *reloc = &relocs[i];
4889 if(reloc->r_pcrel && !reloc->r_extern)
4892 if(reloc->r_length == 2)
4894 unsigned long word = 0;
4895 #ifdef powerpc_HOST_ARCH
4896 unsigned long jumpIsland = 0;
4897 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4898 // to avoid warning and to catch
4902 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4903 checkProddableBlock(oc,wordPtr);
4905 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4909 #ifdef powerpc_HOST_ARCH
4910 else if(reloc->r_type == PPC_RELOC_LO16)
4912 word = ((unsigned short*) wordPtr)[1];
4913 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4915 else if(reloc->r_type == PPC_RELOC_HI16)
4917 word = ((unsigned short*) wordPtr)[1] << 16;
4918 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4920 else if(reloc->r_type == PPC_RELOC_HA16)
4922 word = ((unsigned short*) wordPtr)[1] << 16;
4923 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4925 else if(reloc->r_type == PPC_RELOC_BR24)
4928 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4933 barf("Can't handle this Mach-O relocation entry "
4935 "object file %s; entry type %ld; address %#lx\n",
4936 OC_INFORMATIVE_FILENAME(oc),
4942 if(!reloc->r_extern)
4945 sections[reloc->r_symbolnum-1].offset
4946 - sections[reloc->r_symbolnum-1].addr
4953 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4954 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4955 void *symbolAddress = lookupSymbol(nm);
4958 errorBelch("\nunknown symbol `%s'", nm);
4964 #ifdef powerpc_HOST_ARCH
4965 // In the .o file, this should be a relative jump to NULL
4966 // and we'll change it to a relative jump to the symbol
4967 ASSERT(word + reloc->r_address == 0);
4968 jumpIsland = (unsigned long)
4969 &makeSymbolExtra(oc,
4971 (unsigned long) symbolAddress)
4975 offsetToJumpIsland = word + jumpIsland
4976 - (((long)image) + sect->offset - sect->addr);
4979 word += (unsigned long) symbolAddress
4980 - (((long)image) + sect->offset - sect->addr);
4984 word += (unsigned long) symbolAddress;
4988 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4993 #ifdef powerpc_HOST_ARCH
4994 else if(reloc->r_type == PPC_RELOC_LO16)
4996 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4999 else if(reloc->r_type == PPC_RELOC_HI16)
5001 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
5004 else if(reloc->r_type == PPC_RELOC_HA16)
5006 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
5007 + ((word & (1<<15)) ? 1 : 0);
5010 else if(reloc->r_type == PPC_RELOC_BR24)
5012 if((word & 0x03) != 0)
5013 barf("%s: unconditional relative branch with a displacement "
5014 "which isn't a multiple of 4 bytes: %#lx",
5015 OC_INFORMATIVE_FILENAME(oc),
5018 if((word & 0xFE000000) != 0xFE000000 &&
5019 (word & 0xFE000000) != 0x00000000)
5021 // The branch offset is too large.
5022 // Therefore, we try to use a jump island.
5025 barf("%s: unconditional relative branch out of range: "
5026 "no jump island available: %#lx",
5027 OC_INFORMATIVE_FILENAME(oc),
5031 word = offsetToJumpIsland;
5032 if((word & 0xFE000000) != 0xFE000000 &&
5033 (word & 0xFE000000) != 0x00000000)
5034 barf("%s: unconditional relative branch out of range: "
5035 "jump island out of range: %#lx",
5036 OC_INFORMATIVE_FILENAME(oc),
5039 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
5046 barf("Can't handle Mach-O relocation entry (not scattered) "
5047 "with this r_length tag: "
5048 "object file %s; entry type %ld; "
5049 "r_length tag %ld; address %#lx\n",
5050 OC_INFORMATIVE_FILENAME(oc),
5059 IF_DEBUG(linker, debugBelch("relocateSection: done\n"));
5063 static int ocGetNames_MachO(ObjectCode* oc)
5065 char *image = (char*) oc->image;
5066 struct mach_header *header = (struct mach_header*) image;
5067 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
5068 unsigned i,curSymbol = 0;
5069 struct segment_command *segLC = NULL;
5070 struct section *sections;
5071 struct symtab_command *symLC = NULL;
5072 struct nlist *nlist;
5073 unsigned long commonSize = 0;
5074 char *commonStorage = NULL;
5075 unsigned long commonCounter;
5077 IF_DEBUG(linker,debugBelch("ocGetNames_MachO: start\n"));
5079 for(i=0;i<header->ncmds;i++)
5081 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
5082 segLC = (struct segment_command*) lc;
5083 else if(lc->cmd == LC_SYMTAB)
5084 symLC = (struct symtab_command*) lc;
5085 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
5088 sections = (struct section*) (segLC+1);
5089 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
5093 barf("ocGetNames_MachO: no segment load command");
5095 for(i=0;i<segLC->nsects;i++)
5097 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: segment %d\n", i));
5098 if (sections[i].size == 0)
5101 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
5103 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
5104 "ocGetNames_MachO(common symbols)");
5105 sections[i].offset = zeroFillArea - image;
5108 if(!strcmp(sections[i].sectname,"__text"))
5109 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
5110 (void*) (image + sections[i].offset),
5111 (void*) (image + sections[i].offset + sections[i].size));
5112 else if(!strcmp(sections[i].sectname,"__const"))
5113 addSection(oc, SECTIONKIND_RWDATA,
5114 (void*) (image + sections[i].offset),
5115 (void*) (image + sections[i].offset + sections[i].size));
5116 else if(!strcmp(sections[i].sectname,"__data"))
5117 addSection(oc, SECTIONKIND_RWDATA,
5118 (void*) (image + sections[i].offset),
5119 (void*) (image + sections[i].offset + sections[i].size));
5120 else if(!strcmp(sections[i].sectname,"__bss")
5121 || !strcmp(sections[i].sectname,"__common"))
5122 addSection(oc, SECTIONKIND_RWDATA,
5123 (void*) (image + sections[i].offset),
5124 (void*) (image + sections[i].offset + sections[i].size));
5126 addProddableBlock(oc, (void*) (image + sections[i].offset),
5130 // count external symbols defined here
5134 for(i=0;i<symLC->nsyms;i++)
5136 if(nlist[i].n_type & N_STAB)
5138 else if(nlist[i].n_type & N_EXT)
5140 if((nlist[i].n_type & N_TYPE) == N_UNDF
5141 && (nlist[i].n_value != 0))
5143 commonSize += nlist[i].n_value;
5146 else if((nlist[i].n_type & N_TYPE) == N_SECT)
5151 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: %d external symbols\n", oc->n_symbols));
5152 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
5153 "ocGetNames_MachO(oc->symbols)");
5157 for(i=0;i<symLC->nsyms;i++)
5159 if(nlist[i].n_type & N_STAB)
5161 else if((nlist[i].n_type & N_TYPE) == N_SECT)
5163 if(nlist[i].n_type & N_EXT)
5165 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
5166 if ((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm)) {
5167 // weak definition, and we already have a definition
5168 IF_DEBUG(linker, debugBelch(" weak: %s\n", nm));
5172 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: inserting %s\n", nm));
5173 ghciInsertStrHashTable(oc->fileName, symhash, nm,
5175 + sections[nlist[i].n_sect-1].offset
5176 - sections[nlist[i].n_sect-1].addr
5177 + nlist[i].n_value);
5178 oc->symbols[curSymbol++] = nm;
5185 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
5186 commonCounter = (unsigned long)commonStorage;
5189 for(i=0;i<symLC->nsyms;i++)
5191 if((nlist[i].n_type & N_TYPE) == N_UNDF
5192 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
5194 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
5195 unsigned long sz = nlist[i].n_value;
5197 nlist[i].n_value = commonCounter;
5199 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: inserting common symbol: %s\n", nm));
5200 ghciInsertStrHashTable(oc->fileName, symhash, nm,
5201 (void*)commonCounter);
5202 oc->symbols[curSymbol++] = nm;
5204 commonCounter += sz;
5211 static int ocResolve_MachO(ObjectCode* oc)
5213 char *image = (char*) oc->image;
5214 struct mach_header *header = (struct mach_header*) image;
5215 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
5217 struct segment_command *segLC = NULL;
5218 struct section *sections;
5219 struct symtab_command *symLC = NULL;
5220 struct dysymtab_command *dsymLC = NULL;
5221 struct nlist *nlist;
5223 IF_DEBUG(linker, debugBelch("ocResolve_MachO: start\n"));
5224 for (i = 0; i < header->ncmds; i++)
5226 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
5227 segLC = (struct segment_command*) lc;
5228 else if(lc->cmd == LC_SYMTAB)
5229 symLC = (struct symtab_command*) lc;
5230 else if(lc->cmd == LC_DYSYMTAB)
5231 dsymLC = (struct dysymtab_command*) lc;
5232 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
5235 sections = (struct section*) (segLC+1);
5236 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
5241 unsigned long *indirectSyms
5242 = (unsigned long*) (image + dsymLC->indirectsymoff);
5244 IF_DEBUG(linker, debugBelch("ocResolve_MachO: resolving dsymLC\n"));
5245 for (i = 0; i < segLC->nsects; i++)
5247 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
5248 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
5249 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
5251 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5254 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
5255 || !strcmp(sections[i].sectname,"__pointers"))
5257 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5260 else if(!strcmp(sections[i].sectname,"__jump_table"))
5262 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5267 IF_DEBUG(linker, debugBelch("ocResolve_MachO: unknown section\n"));
5272 for(i=0;i<segLC->nsects;i++)
5274 IF_DEBUG(linker, debugBelch("ocResolve_MachO: relocating section %d\n", i));
5276 if (!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
5280 #if defined (powerpc_HOST_ARCH)
5281 ocFlushInstructionCache( oc );
5287 #ifdef powerpc_HOST_ARCH
5289 * The Mach-O object format uses leading underscores. But not everywhere.
5290 * There is a small number of runtime support functions defined in
5291 * libcc_dynamic.a whose name does not have a leading underscore.
5292 * As a consequence, we can't get their address from C code.
5293 * We have to use inline assembler just to take the address of a function.
5297 extern void* symbolsWithoutUnderscore[];
5299 static void machoInitSymbolsWithoutUnderscore()
5301 void **p = symbolsWithoutUnderscore;
5302 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
5304 #undef SymI_NeedsProto
5305 #define SymI_NeedsProto(x) \
5306 __asm__ volatile(".long " # x);
5308 RTS_MACHO_NOUNDERLINE_SYMBOLS
5310 __asm__ volatile(".text");
5312 #undef SymI_NeedsProto
5313 #define SymI_NeedsProto(x) \
5314 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
5316 RTS_MACHO_NOUNDERLINE_SYMBOLS
5318 #undef SymI_NeedsProto
5324 * Figure out by how much to shift the entire Mach-O file in memory
5325 * when loading so that its single segment ends up 16-byte-aligned
5327 static int machoGetMisalignment( FILE * f )
5329 struct mach_header header;
5332 fread(&header, sizeof(header), 1, f);
5335 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
5336 if(header.magic != MH_MAGIC_64) {
5337 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
5338 MH_MAGIC_64, header->magic);
5342 if(header.magic != MH_MAGIC) {
5343 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
5344 MH_MAGIC, header->magic);
5349 misalignment = (header.sizeofcmds + sizeof(header))
5352 return misalignment ? (16 - misalignment) : 0;