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"
25 #include "LinkerInternals.h"
28 #include "StgPrimFloat.h" // for __int_encodeFloat etc.
31 #if !defined(mingw32_HOST_OS)
32 #include "posix/Signals.h"
35 #ifdef HAVE_SYS_TYPES_H
36 #include <sys/types.h>
42 #ifdef HAVE_SYS_STAT_H
46 #if defined(HAVE_DLFCN_H)
50 #if defined(cygwin32_HOST_OS)
55 #ifdef HAVE_SYS_TIME_H
59 #include <sys/fcntl.h>
60 #include <sys/termios.h>
61 #include <sys/utime.h>
62 #include <sys/utsname.h>
66 #if defined(ia64_HOST_ARCH) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
71 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
79 #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)
80 # define OBJFORMAT_ELF
81 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
82 # define OBJFORMAT_PEi386
85 #elif defined(darwin_HOST_OS)
86 # define OBJFORMAT_MACHO
87 # include <mach-o/loader.h>
88 # include <mach-o/nlist.h>
89 # include <mach-o/reloc.h>
90 #if !defined(HAVE_DLFCN_H)
91 # include <mach-o/dyld.h>
93 #if defined(powerpc_HOST_ARCH)
94 # include <mach-o/ppc/reloc.h>
96 #if defined(x86_64_HOST_ARCH)
97 # include <mach-o/x86_64/reloc.h>
101 /* Hash table mapping symbol names to Symbol */
102 static /*Str*/HashTable *symhash;
104 /* Hash table mapping symbol names to StgStablePtr */
105 static /*Str*/HashTable *stablehash;
107 /* List of currently loaded objects */
108 ObjectCode *objects = NULL; /* initially empty */
110 #if defined(OBJFORMAT_ELF)
111 static int ocVerifyImage_ELF ( ObjectCode* oc );
112 static int ocGetNames_ELF ( ObjectCode* oc );
113 static int ocResolve_ELF ( ObjectCode* oc );
114 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
115 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
117 #elif defined(OBJFORMAT_PEi386)
118 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
119 static int ocGetNames_PEi386 ( ObjectCode* oc );
120 static int ocResolve_PEi386 ( ObjectCode* oc );
121 static void *lookupSymbolInDLLs ( unsigned char *lbl );
122 static void zapTrailingAtSign ( unsigned char *sym );
123 #elif defined(OBJFORMAT_MACHO)
124 static int ocVerifyImage_MachO ( ObjectCode* oc );
125 static int ocGetNames_MachO ( ObjectCode* oc );
126 static int ocResolve_MachO ( ObjectCode* oc );
128 static int machoGetMisalignment( FILE * );
129 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
130 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
132 #ifdef powerpc_HOST_ARCH
133 static void machoInitSymbolsWithoutUnderscore( void );
137 /* on x86_64 we have a problem with relocating symbol references in
138 * code that was compiled without -fPIC. By default, the small memory
139 * model is used, which assumes that symbol references can fit in a
140 * 32-bit slot. The system dynamic linker makes this work for
141 * references to shared libraries by either (a) allocating a jump
142 * table slot for code references, or (b) moving the symbol at load
143 * time (and copying its contents, if necessary) for data references.
145 * We unfortunately can't tell whether symbol references are to code
146 * or data. So for now we assume they are code (the vast majority
147 * are), and allocate jump-table slots. Unfortunately this will
148 * SILENTLY generate crashing code for data references. This hack is
149 * enabled by X86_64_ELF_NONPIC_HACK.
151 * One workaround is to use shared Haskell libraries. This is
152 * coming. Another workaround is to keep the static libraries but
153 * compile them with -fPIC, because that will generate PIC references
154 * to data which can be relocated. The PIC code is still too green to
155 * do this systematically, though.
158 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
160 * Naming Scheme for Symbol Macros
162 * SymI_*: symbol is internal to the RTS. It resides in an object
163 * file/library that is statically.
164 * SymE_*: symbol is external to the RTS library. It might be linked
167 * Sym*_HasProto : the symbol prototype is imported in an include file
168 * or defined explicitly
169 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
170 * default proto extern void sym(void);
172 #define X86_64_ELF_NONPIC_HACK 1
174 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
175 * small memory model on this architecture (see gcc docs,
178 * MAP_32BIT not available on OpenBSD/amd64
180 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
181 #define TRY_MAP_32BIT MAP_32BIT
183 #define TRY_MAP_32BIT 0
187 * Due to the small memory model (see above), on x86_64 we have to map
188 * all our non-PIC object files into the low 2Gb of the address space
189 * (why 2Gb and not 4Gb? Because all addresses must be reachable
190 * using a 32-bit signed PC-relative offset). On Linux we can do this
191 * using the MAP_32BIT flag to mmap(), however on other OSs
192 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
193 * can't do this. So on these systems, we have to pick a base address
194 * in the low 2Gb of the address space and try to allocate memory from
197 * We pick a default address based on the OS, but also make this
198 * configurable via an RTS flag (+RTS -xm)
200 #if defined(x86_64_HOST_ARCH)
202 #if defined(MAP_32BIT)
203 // Try to use MAP_32BIT
204 #define MMAP_32BIT_BASE_DEFAULT 0
207 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
210 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
213 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
214 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
215 #define MAP_ANONYMOUS MAP_ANON
218 /* -----------------------------------------------------------------------------
219 * Built-in symbols from the RTS
222 typedef struct _RtsSymbolVal {
227 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
228 SymI_HasProto(stg_mkWeakForeignEnvzh) \
229 SymI_HasProto(stg_makeStableNamezh) \
230 SymI_HasProto(stg_finalizzeWeakzh)
232 #if !defined (mingw32_HOST_OS)
233 #define RTS_POSIX_ONLY_SYMBOLS \
234 SymI_HasProto(__hscore_get_saved_termios) \
235 SymI_HasProto(__hscore_set_saved_termios) \
236 SymI_HasProto(shutdownHaskellAndSignal) \
237 SymI_HasProto(lockFile) \
238 SymI_HasProto(unlockFile) \
239 SymI_HasProto(signal_handlers) \
240 SymI_HasProto(stg_sig_install) \
241 SymI_NeedsProto(nocldstop)
244 #if defined (cygwin32_HOST_OS)
245 #define RTS_MINGW_ONLY_SYMBOLS /**/
246 /* Don't have the ability to read import libs / archives, so
247 * we have to stupidly list a lot of what libcygwin.a
250 #define RTS_CYGWIN_ONLY_SYMBOLS \
251 SymI_HasProto(regfree) \
252 SymI_HasProto(regexec) \
253 SymI_HasProto(regerror) \
254 SymI_HasProto(regcomp) \
255 SymI_HasProto(__errno) \
256 SymI_HasProto(access) \
257 SymI_HasProto(chmod) \
258 SymI_HasProto(chdir) \
259 SymI_HasProto(close) \
260 SymI_HasProto(creat) \
262 SymI_HasProto(dup2) \
263 SymI_HasProto(fstat) \
264 SymI_HasProto(fcntl) \
265 SymI_HasProto(getcwd) \
266 SymI_HasProto(getenv) \
267 SymI_HasProto(lseek) \
268 SymI_HasProto(open) \
269 SymI_HasProto(fpathconf) \
270 SymI_HasProto(pathconf) \
271 SymI_HasProto(stat) \
273 SymI_HasProto(tanh) \
274 SymI_HasProto(cosh) \
275 SymI_HasProto(sinh) \
276 SymI_HasProto(atan) \
277 SymI_HasProto(acos) \
278 SymI_HasProto(asin) \
284 SymI_HasProto(sqrt) \
285 SymI_HasProto(localtime_r) \
286 SymI_HasProto(gmtime_r) \
287 SymI_HasProto(mktime) \
288 SymI_NeedsProto(_imp___tzname) \
289 SymI_HasProto(gettimeofday) \
290 SymI_HasProto(timezone) \
291 SymI_HasProto(tcgetattr) \
292 SymI_HasProto(tcsetattr) \
293 SymI_HasProto(memcpy) \
294 SymI_HasProto(memmove) \
295 SymI_HasProto(realloc) \
296 SymI_HasProto(malloc) \
297 SymI_HasProto(free) \
298 SymI_HasProto(fork) \
299 SymI_HasProto(lstat) \
300 SymI_HasProto(isatty) \
301 SymI_HasProto(mkdir) \
302 SymI_HasProto(opendir) \
303 SymI_HasProto(readdir) \
304 SymI_HasProto(rewinddir) \
305 SymI_HasProto(closedir) \
306 SymI_HasProto(link) \
307 SymI_HasProto(mkfifo) \
308 SymI_HasProto(pipe) \
309 SymI_HasProto(read) \
310 SymI_HasProto(rename) \
311 SymI_HasProto(rmdir) \
312 SymI_HasProto(select) \
313 SymI_HasProto(system) \
314 SymI_HasProto(write) \
315 SymI_HasProto(strcmp) \
316 SymI_HasProto(strcpy) \
317 SymI_HasProto(strncpy) \
318 SymI_HasProto(strerror) \
319 SymI_HasProto(sigaddset) \
320 SymI_HasProto(sigemptyset) \
321 SymI_HasProto(sigprocmask) \
322 SymI_HasProto(umask) \
323 SymI_HasProto(uname) \
324 SymI_HasProto(unlink) \
325 SymI_HasProto(utime) \
326 SymI_HasProto(waitpid)
328 #elif !defined(mingw32_HOST_OS)
329 #define RTS_MINGW_ONLY_SYMBOLS /**/
330 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
331 #else /* defined(mingw32_HOST_OS) */
332 #define RTS_POSIX_ONLY_SYMBOLS /**/
333 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
335 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
337 #define RTS_MINGW_EXTRA_SYMS \
338 SymI_NeedsProto(_imp____mb_cur_max) \
339 SymI_NeedsProto(_imp___pctype)
341 #define RTS_MINGW_EXTRA_SYMS
344 #if HAVE_GETTIMEOFDAY
345 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
347 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
350 #if HAVE___MINGW_VFPRINTF
351 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
353 #define RTS___MINGW_VFPRINTF_SYM /**/
356 /* These are statically linked from the mingw libraries into the ghc
357 executable, so we have to employ this hack. */
358 #define RTS_MINGW_ONLY_SYMBOLS \
359 SymI_HasProto(stg_asyncReadzh) \
360 SymI_HasProto(stg_asyncWritezh) \
361 SymI_HasProto(stg_asyncDoProczh) \
362 SymI_HasProto(memset) \
363 SymI_HasProto(inet_ntoa) \
364 SymI_HasProto(inet_addr) \
365 SymI_HasProto(htonl) \
366 SymI_HasProto(recvfrom) \
367 SymI_HasProto(listen) \
368 SymI_HasProto(bind) \
369 SymI_HasProto(shutdown) \
370 SymI_HasProto(connect) \
371 SymI_HasProto(htons) \
372 SymI_HasProto(ntohs) \
373 SymI_HasProto(getservbyname) \
374 SymI_HasProto(getservbyport) \
375 SymI_HasProto(getprotobynumber) \
376 SymI_HasProto(getprotobyname) \
377 SymI_HasProto(gethostbyname) \
378 SymI_HasProto(gethostbyaddr) \
379 SymI_HasProto(gethostname) \
380 SymI_HasProto(strcpy) \
381 SymI_HasProto(strncpy) \
382 SymI_HasProto(abort) \
383 SymI_NeedsProto(_alloca) \
384 SymI_NeedsProto(isxdigit) \
385 SymI_NeedsProto(isupper) \
386 SymI_NeedsProto(ispunct) \
387 SymI_NeedsProto(islower) \
388 SymI_NeedsProto(isspace) \
389 SymI_NeedsProto(isprint) \
390 SymI_NeedsProto(isdigit) \
391 SymI_NeedsProto(iscntrl) \
392 SymI_NeedsProto(isalpha) \
393 SymI_NeedsProto(isalnum) \
394 SymI_NeedsProto(isascii) \
395 RTS___MINGW_VFPRINTF_SYM \
396 SymI_HasProto(strcmp) \
397 SymI_HasProto(memmove) \
398 SymI_HasProto(realloc) \
399 SymI_HasProto(malloc) \
401 SymI_HasProto(tanh) \
402 SymI_HasProto(cosh) \
403 SymI_HasProto(sinh) \
404 SymI_HasProto(atan) \
405 SymI_HasProto(acos) \
406 SymI_HasProto(asin) \
412 SymI_HasProto(sqrt) \
413 SymI_HasProto(powf) \
414 SymI_HasProto(tanhf) \
415 SymI_HasProto(coshf) \
416 SymI_HasProto(sinhf) \
417 SymI_HasProto(atanf) \
418 SymI_HasProto(acosf) \
419 SymI_HasProto(asinf) \
420 SymI_HasProto(tanf) \
421 SymI_HasProto(cosf) \
422 SymI_HasProto(sinf) \
423 SymI_HasProto(expf) \
424 SymI_HasProto(logf) \
425 SymI_HasProto(sqrtf) \
427 SymI_HasProto(erfc) \
428 SymI_HasProto(erff) \
429 SymI_HasProto(erfcf) \
430 SymI_HasProto(memcpy) \
431 SymI_HasProto(rts_InstallConsoleEvent) \
432 SymI_HasProto(rts_ConsoleHandlerDone) \
433 SymI_NeedsProto(mktime) \
434 SymI_NeedsProto(_imp___timezone) \
435 SymI_NeedsProto(_imp___tzname) \
436 SymI_NeedsProto(_imp__tzname) \
437 SymI_NeedsProto(_imp___iob) \
438 SymI_NeedsProto(_imp___osver) \
439 SymI_NeedsProto(localtime) \
440 SymI_NeedsProto(gmtime) \
441 SymI_NeedsProto(opendir) \
442 SymI_NeedsProto(readdir) \
443 SymI_NeedsProto(rewinddir) \
444 RTS_MINGW_EXTRA_SYMS \
445 RTS_MINGW_GETTIMEOFDAY_SYM \
446 SymI_NeedsProto(closedir)
449 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
450 #define RTS_DARWIN_ONLY_SYMBOLS \
451 SymI_NeedsProto(asprintf$LDBLStub) \
452 SymI_NeedsProto(err$LDBLStub) \
453 SymI_NeedsProto(errc$LDBLStub) \
454 SymI_NeedsProto(errx$LDBLStub) \
455 SymI_NeedsProto(fprintf$LDBLStub) \
456 SymI_NeedsProto(fscanf$LDBLStub) \
457 SymI_NeedsProto(fwprintf$LDBLStub) \
458 SymI_NeedsProto(fwscanf$LDBLStub) \
459 SymI_NeedsProto(printf$LDBLStub) \
460 SymI_NeedsProto(scanf$LDBLStub) \
461 SymI_NeedsProto(snprintf$LDBLStub) \
462 SymI_NeedsProto(sprintf$LDBLStub) \
463 SymI_NeedsProto(sscanf$LDBLStub) \
464 SymI_NeedsProto(strtold$LDBLStub) \
465 SymI_NeedsProto(swprintf$LDBLStub) \
466 SymI_NeedsProto(swscanf$LDBLStub) \
467 SymI_NeedsProto(syslog$LDBLStub) \
468 SymI_NeedsProto(vasprintf$LDBLStub) \
469 SymI_NeedsProto(verr$LDBLStub) \
470 SymI_NeedsProto(verrc$LDBLStub) \
471 SymI_NeedsProto(verrx$LDBLStub) \
472 SymI_NeedsProto(vfprintf$LDBLStub) \
473 SymI_NeedsProto(vfscanf$LDBLStub) \
474 SymI_NeedsProto(vfwprintf$LDBLStub) \
475 SymI_NeedsProto(vfwscanf$LDBLStub) \
476 SymI_NeedsProto(vprintf$LDBLStub) \
477 SymI_NeedsProto(vscanf$LDBLStub) \
478 SymI_NeedsProto(vsnprintf$LDBLStub) \
479 SymI_NeedsProto(vsprintf$LDBLStub) \
480 SymI_NeedsProto(vsscanf$LDBLStub) \
481 SymI_NeedsProto(vswprintf$LDBLStub) \
482 SymI_NeedsProto(vswscanf$LDBLStub) \
483 SymI_NeedsProto(vsyslog$LDBLStub) \
484 SymI_NeedsProto(vwarn$LDBLStub) \
485 SymI_NeedsProto(vwarnc$LDBLStub) \
486 SymI_NeedsProto(vwarnx$LDBLStub) \
487 SymI_NeedsProto(vwprintf$LDBLStub) \
488 SymI_NeedsProto(vwscanf$LDBLStub) \
489 SymI_NeedsProto(warn$LDBLStub) \
490 SymI_NeedsProto(warnc$LDBLStub) \
491 SymI_NeedsProto(warnx$LDBLStub) \
492 SymI_NeedsProto(wcstold$LDBLStub) \
493 SymI_NeedsProto(wprintf$LDBLStub) \
494 SymI_NeedsProto(wscanf$LDBLStub)
496 #define RTS_DARWIN_ONLY_SYMBOLS
500 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
502 # define MAIN_CAP_SYM
505 #if !defined(mingw32_HOST_OS)
506 #define RTS_USER_SIGNALS_SYMBOLS \
507 SymI_HasProto(setIOManagerPipe) \
508 SymI_HasProto(ioManagerWakeup) \
509 SymI_HasProto(ioManagerSync) \
510 SymI_HasProto(blockUserSignals) \
511 SymI_HasProto(unblockUserSignals)
513 #define RTS_USER_SIGNALS_SYMBOLS \
514 SymI_HasProto(ioManagerWakeup) \
515 SymI_HasProto(sendIOManagerEvent) \
516 SymI_HasProto(readIOManagerEvent) \
517 SymI_HasProto(getIOManagerEvent) \
518 SymI_HasProto(console_handler)
521 #define RTS_LIBFFI_SYMBOLS \
522 SymE_NeedsProto(ffi_prep_cif) \
523 SymE_NeedsProto(ffi_call) \
524 SymE_NeedsProto(ffi_type_void) \
525 SymE_NeedsProto(ffi_type_float) \
526 SymE_NeedsProto(ffi_type_double) \
527 SymE_NeedsProto(ffi_type_sint64) \
528 SymE_NeedsProto(ffi_type_uint64) \
529 SymE_NeedsProto(ffi_type_sint32) \
530 SymE_NeedsProto(ffi_type_uint32) \
531 SymE_NeedsProto(ffi_type_sint16) \
532 SymE_NeedsProto(ffi_type_uint16) \
533 SymE_NeedsProto(ffi_type_sint8) \
534 SymE_NeedsProto(ffi_type_uint8) \
535 SymE_NeedsProto(ffi_type_pointer)
537 #ifdef TABLES_NEXT_TO_CODE
538 #define RTS_RET_SYMBOLS /* nothing */
540 #define RTS_RET_SYMBOLS \
541 SymI_HasProto(stg_enter_ret) \
542 SymI_HasProto(stg_gc_fun_ret) \
543 SymI_HasProto(stg_ap_v_ret) \
544 SymI_HasProto(stg_ap_f_ret) \
545 SymI_HasProto(stg_ap_d_ret) \
546 SymI_HasProto(stg_ap_l_ret) \
547 SymI_HasProto(stg_ap_n_ret) \
548 SymI_HasProto(stg_ap_p_ret) \
549 SymI_HasProto(stg_ap_pv_ret) \
550 SymI_HasProto(stg_ap_pp_ret) \
551 SymI_HasProto(stg_ap_ppv_ret) \
552 SymI_HasProto(stg_ap_ppp_ret) \
553 SymI_HasProto(stg_ap_pppv_ret) \
554 SymI_HasProto(stg_ap_pppp_ret) \
555 SymI_HasProto(stg_ap_ppppp_ret) \
556 SymI_HasProto(stg_ap_pppppp_ret)
559 /* Modules compiled with -ticky may mention ticky counters */
560 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
561 #define RTS_TICKY_SYMBOLS \
562 SymI_NeedsProto(ticky_entry_ctrs) \
563 SymI_NeedsProto(top_ct) \
565 SymI_HasProto(ENT_VIA_NODE_ctr) \
566 SymI_HasProto(ENT_STATIC_THK_ctr) \
567 SymI_HasProto(ENT_DYN_THK_ctr) \
568 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
569 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
570 SymI_HasProto(ENT_STATIC_CON_ctr) \
571 SymI_HasProto(ENT_DYN_CON_ctr) \
572 SymI_HasProto(ENT_STATIC_IND_ctr) \
573 SymI_HasProto(ENT_DYN_IND_ctr) \
574 SymI_HasProto(ENT_PERM_IND_ctr) \
575 SymI_HasProto(ENT_PAP_ctr) \
576 SymI_HasProto(ENT_AP_ctr) \
577 SymI_HasProto(ENT_AP_STACK_ctr) \
578 SymI_HasProto(ENT_BH_ctr) \
579 SymI_HasProto(UNKNOWN_CALL_ctr) \
580 SymI_HasProto(SLOW_CALL_v_ctr) \
581 SymI_HasProto(SLOW_CALL_f_ctr) \
582 SymI_HasProto(SLOW_CALL_d_ctr) \
583 SymI_HasProto(SLOW_CALL_l_ctr) \
584 SymI_HasProto(SLOW_CALL_n_ctr) \
585 SymI_HasProto(SLOW_CALL_p_ctr) \
586 SymI_HasProto(SLOW_CALL_pv_ctr) \
587 SymI_HasProto(SLOW_CALL_pp_ctr) \
588 SymI_HasProto(SLOW_CALL_ppv_ctr) \
589 SymI_HasProto(SLOW_CALL_ppp_ctr) \
590 SymI_HasProto(SLOW_CALL_pppv_ctr) \
591 SymI_HasProto(SLOW_CALL_pppp_ctr) \
592 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
593 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
594 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
595 SymI_HasProto(ticky_slow_call_unevald) \
596 SymI_HasProto(SLOW_CALL_ctr) \
597 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
598 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
599 SymI_HasProto(KNOWN_CALL_ctr) \
600 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
601 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
602 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
603 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
604 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
605 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
606 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
607 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
608 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
609 SymI_HasProto(UPDF_OMITTED_ctr) \
610 SymI_HasProto(UPDF_PUSHED_ctr) \
611 SymI_HasProto(CATCHF_PUSHED_ctr) \
612 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
613 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
614 SymI_HasProto(UPD_SQUEEZED_ctr) \
615 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
616 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
617 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
618 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
619 SymI_HasProto(ALLOC_HEAP_ctr) \
620 SymI_HasProto(ALLOC_HEAP_tot) \
621 SymI_HasProto(ALLOC_FUN_ctr) \
622 SymI_HasProto(ALLOC_FUN_adm) \
623 SymI_HasProto(ALLOC_FUN_gds) \
624 SymI_HasProto(ALLOC_FUN_slp) \
625 SymI_HasProto(UPD_NEW_IND_ctr) \
626 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
627 SymI_HasProto(UPD_OLD_IND_ctr) \
628 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
629 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
630 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
631 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
632 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
633 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
634 SymI_HasProto(GC_SEL_MINOR_ctr) \
635 SymI_HasProto(GC_SEL_MAJOR_ctr) \
636 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
637 SymI_HasProto(ALLOC_UP_THK_ctr) \
638 SymI_HasProto(ALLOC_SE_THK_ctr) \
639 SymI_HasProto(ALLOC_THK_adm) \
640 SymI_HasProto(ALLOC_THK_gds) \
641 SymI_HasProto(ALLOC_THK_slp) \
642 SymI_HasProto(ALLOC_CON_ctr) \
643 SymI_HasProto(ALLOC_CON_adm) \
644 SymI_HasProto(ALLOC_CON_gds) \
645 SymI_HasProto(ALLOC_CON_slp) \
646 SymI_HasProto(ALLOC_TUP_ctr) \
647 SymI_HasProto(ALLOC_TUP_adm) \
648 SymI_HasProto(ALLOC_TUP_gds) \
649 SymI_HasProto(ALLOC_TUP_slp) \
650 SymI_HasProto(ALLOC_BH_ctr) \
651 SymI_HasProto(ALLOC_BH_adm) \
652 SymI_HasProto(ALLOC_BH_gds) \
653 SymI_HasProto(ALLOC_BH_slp) \
654 SymI_HasProto(ALLOC_PRIM_ctr) \
655 SymI_HasProto(ALLOC_PRIM_adm) \
656 SymI_HasProto(ALLOC_PRIM_gds) \
657 SymI_HasProto(ALLOC_PRIM_slp) \
658 SymI_HasProto(ALLOC_PAP_ctr) \
659 SymI_HasProto(ALLOC_PAP_adm) \
660 SymI_HasProto(ALLOC_PAP_gds) \
661 SymI_HasProto(ALLOC_PAP_slp) \
662 SymI_HasProto(ALLOC_TSO_ctr) \
663 SymI_HasProto(ALLOC_TSO_adm) \
664 SymI_HasProto(ALLOC_TSO_gds) \
665 SymI_HasProto(ALLOC_TSO_slp) \
666 SymI_HasProto(RET_NEW_ctr) \
667 SymI_HasProto(RET_OLD_ctr) \
668 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
669 SymI_HasProto(RET_SEMI_loads_avoided)
672 // On most platforms, the garbage collector rewrites references
673 // to small integer and char objects to a set of common, shared ones.
675 // We don't do this when compiling to Windows DLLs at the moment because
676 // it doesn't support cross package data references well.
678 #if defined(__PIC__) && defined(mingw32_HOST_OS)
679 #define RTS_INTCHAR_SYMBOLS
681 #define RTS_INTCHAR_SYMBOLS \
682 SymI_HasProto(stg_CHARLIKE_closure) \
683 SymI_HasProto(stg_INTLIKE_closure)
687 #define RTS_SYMBOLS \
690 SymI_HasProto(StgReturn) \
691 SymI_HasProto(stg_enter_info) \
692 SymI_HasProto(stg_gc_void_info) \
693 SymI_HasProto(__stg_gc_enter_1) \
694 SymI_HasProto(stg_gc_noregs) \
695 SymI_HasProto(stg_gc_unpt_r1_info) \
696 SymI_HasProto(stg_gc_unpt_r1) \
697 SymI_HasProto(stg_gc_unbx_r1_info) \
698 SymI_HasProto(stg_gc_unbx_r1) \
699 SymI_HasProto(stg_gc_f1_info) \
700 SymI_HasProto(stg_gc_f1) \
701 SymI_HasProto(stg_gc_d1_info) \
702 SymI_HasProto(stg_gc_d1) \
703 SymI_HasProto(stg_gc_l1_info) \
704 SymI_HasProto(stg_gc_l1) \
705 SymI_HasProto(__stg_gc_fun) \
706 SymI_HasProto(stg_gc_fun_info) \
707 SymI_HasProto(stg_gc_gen) \
708 SymI_HasProto(stg_gc_gen_info) \
709 SymI_HasProto(stg_gc_gen_hp) \
710 SymI_HasProto(stg_gc_ut) \
711 SymI_HasProto(stg_gen_yield) \
712 SymI_HasProto(stg_yield_noregs) \
713 SymI_HasProto(stg_yield_to_interpreter) \
714 SymI_HasProto(stg_gen_block) \
715 SymI_HasProto(stg_block_noregs) \
716 SymI_HasProto(stg_block_1) \
717 SymI_HasProto(stg_block_takemvar) \
718 SymI_HasProto(stg_block_putmvar) \
720 SymI_HasProto(MallocFailHook) \
721 SymI_HasProto(OnExitHook) \
722 SymI_HasProto(OutOfHeapHook) \
723 SymI_HasProto(StackOverflowHook) \
724 SymI_HasProto(addDLL) \
725 SymI_HasProto(__int_encodeDouble) \
726 SymI_HasProto(__word_encodeDouble) \
727 SymI_HasProto(__2Int_encodeDouble) \
728 SymI_HasProto(__int_encodeFloat) \
729 SymI_HasProto(__word_encodeFloat) \
730 SymI_HasProto(stg_atomicallyzh) \
731 SymI_HasProto(barf) \
732 SymI_HasProto(debugBelch) \
733 SymI_HasProto(errorBelch) \
734 SymI_HasProto(sysErrorBelch) \
735 SymI_HasProto(stg_asyncExceptionsBlockedzh) \
736 SymI_HasProto(stg_blockAsyncExceptionszh) \
737 SymI_HasProto(stg_catchzh) \
738 SymI_HasProto(stg_catchRetryzh) \
739 SymI_HasProto(stg_catchSTMzh) \
740 SymI_HasProto(stg_checkzh) \
741 SymI_HasProto(closure_flags) \
742 SymI_HasProto(cmp_thread) \
743 SymI_HasProto(createAdjustor) \
744 SymI_HasProto(stg_decodeDoublezu2Intzh) \
745 SymI_HasProto(stg_decodeFloatzuIntzh) \
746 SymI_HasProto(defaultsHook) \
747 SymI_HasProto(stg_delayzh) \
748 SymI_HasProto(stg_deRefWeakzh) \
749 SymI_HasProto(stg_deRefStablePtrzh) \
750 SymI_HasProto(dirty_MUT_VAR) \
751 SymI_HasProto(stg_forkzh) \
752 SymI_HasProto(stg_forkOnzh) \
753 SymI_HasProto(forkProcess) \
754 SymI_HasProto(forkOS_createThread) \
755 SymI_HasProto(freeHaskellFunctionPtr) \
756 SymI_HasProto(getOrSetTypeableStore) \
757 SymI_HasProto(getOrSetGHCConcSignalHandlerStore) \
758 SymI_HasProto(getOrSetGHCConcPendingEventsStore) \
759 SymI_HasProto(getOrSetGHCConcPendingDelaysStore) \
760 SymI_HasProto(getOrSetGHCConcIOManagerThreadStore) \
761 SymI_HasProto(getOrSetGHCConcProddingStore) \
762 SymI_HasProto(genSymZh) \
763 SymI_HasProto(genericRaise) \
764 SymI_HasProto(getProgArgv) \
765 SymI_HasProto(getFullProgArgv) \
766 SymI_HasProto(getStablePtr) \
767 SymI_HasProto(hs_init) \
768 SymI_HasProto(hs_exit) \
769 SymI_HasProto(hs_set_argv) \
770 SymI_HasProto(hs_add_root) \
771 SymI_HasProto(hs_perform_gc) \
772 SymI_HasProto(hs_free_stable_ptr) \
773 SymI_HasProto(hs_free_fun_ptr) \
774 SymI_HasProto(hs_hpc_rootModule) \
775 SymI_HasProto(hs_hpc_module) \
776 SymI_HasProto(initLinker) \
777 SymI_HasProto(stg_unpackClosurezh) \
778 SymI_HasProto(stg_getApStackValzh) \
779 SymI_HasProto(stg_getSparkzh) \
780 SymI_HasProto(stg_isCurrentThreadBoundzh) \
781 SymI_HasProto(stg_isEmptyMVarzh) \
782 SymI_HasProto(stg_killThreadzh) \
783 SymI_HasProto(loadObj) \
784 SymI_HasProto(insertStableSymbol) \
785 SymI_HasProto(insertSymbol) \
786 SymI_HasProto(lookupSymbol) \
787 SymI_HasProto(stg_makeStablePtrzh) \
788 SymI_HasProto(stg_mkApUpd0zh) \
789 SymI_HasProto(stg_myThreadIdzh) \
790 SymI_HasProto(stg_labelThreadzh) \
791 SymI_HasProto(stg_newArrayzh) \
792 SymI_HasProto(stg_newBCOzh) \
793 SymI_HasProto(stg_newByteArrayzh) \
794 SymI_HasProto_redirect(newCAF, newDynCAF) \
795 SymI_HasProto(stg_newMVarzh) \
796 SymI_HasProto(stg_newMutVarzh) \
797 SymI_HasProto(stg_newTVarzh) \
798 SymI_HasProto(stg_noDuplicatezh) \
799 SymI_HasProto(stg_atomicModifyMutVarzh) \
800 SymI_HasProto(stg_newPinnedByteArrayzh) \
801 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
802 SymI_HasProto(newSpark) \
803 SymI_HasProto(performGC) \
804 SymI_HasProto(performMajorGC) \
805 SymI_HasProto(prog_argc) \
806 SymI_HasProto(prog_argv) \
807 SymI_HasProto(stg_putMVarzh) \
808 SymI_HasProto(stg_raisezh) \
809 SymI_HasProto(stg_raiseIOzh) \
810 SymI_HasProto(stg_readTVarzh) \
811 SymI_HasProto(stg_readTVarIOzh) \
812 SymI_HasProto(resumeThread) \
813 SymI_HasProto(resolveObjs) \
814 SymI_HasProto(stg_retryzh) \
815 SymI_HasProto(rts_apply) \
816 SymI_HasProto(rts_checkSchedStatus) \
817 SymI_HasProto(rts_eval) \
818 SymI_HasProto(rts_evalIO) \
819 SymI_HasProto(rts_evalLazyIO) \
820 SymI_HasProto(rts_evalStableIO) \
821 SymI_HasProto(rts_eval_) \
822 SymI_HasProto(rts_getBool) \
823 SymI_HasProto(rts_getChar) \
824 SymI_HasProto(rts_getDouble) \
825 SymI_HasProto(rts_getFloat) \
826 SymI_HasProto(rts_getInt) \
827 SymI_HasProto(rts_getInt8) \
828 SymI_HasProto(rts_getInt16) \
829 SymI_HasProto(rts_getInt32) \
830 SymI_HasProto(rts_getInt64) \
831 SymI_HasProto(rts_getPtr) \
832 SymI_HasProto(rts_getFunPtr) \
833 SymI_HasProto(rts_getStablePtr) \
834 SymI_HasProto(rts_getThreadId) \
835 SymI_HasProto(rts_getWord) \
836 SymI_HasProto(rts_getWord8) \
837 SymI_HasProto(rts_getWord16) \
838 SymI_HasProto(rts_getWord32) \
839 SymI_HasProto(rts_getWord64) \
840 SymI_HasProto(rts_lock) \
841 SymI_HasProto(rts_mkBool) \
842 SymI_HasProto(rts_mkChar) \
843 SymI_HasProto(rts_mkDouble) \
844 SymI_HasProto(rts_mkFloat) \
845 SymI_HasProto(rts_mkInt) \
846 SymI_HasProto(rts_mkInt8) \
847 SymI_HasProto(rts_mkInt16) \
848 SymI_HasProto(rts_mkInt32) \
849 SymI_HasProto(rts_mkInt64) \
850 SymI_HasProto(rts_mkPtr) \
851 SymI_HasProto(rts_mkFunPtr) \
852 SymI_HasProto(rts_mkStablePtr) \
853 SymI_HasProto(rts_mkString) \
854 SymI_HasProto(rts_mkWord) \
855 SymI_HasProto(rts_mkWord8) \
856 SymI_HasProto(rts_mkWord16) \
857 SymI_HasProto(rts_mkWord32) \
858 SymI_HasProto(rts_mkWord64) \
859 SymI_HasProto(rts_unlock) \
860 SymI_HasProto(rts_unsafeGetMyCapability) \
861 SymI_HasProto(rtsSupportsBoundThreads) \
862 SymI_HasProto(setProgArgv) \
863 SymI_HasProto(startupHaskell) \
864 SymI_HasProto(shutdownHaskell) \
865 SymI_HasProto(shutdownHaskellAndExit) \
866 SymI_HasProto(stable_ptr_table) \
867 SymI_HasProto(stackOverflow) \
868 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
869 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
870 SymI_HasProto(startTimer) \
871 SymI_HasProto(stg_MVAR_CLEAN_info) \
872 SymI_HasProto(stg_MVAR_DIRTY_info) \
873 SymI_HasProto(stg_IND_STATIC_info) \
874 SymI_HasProto(stg_ARR_WORDS_info) \
875 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
876 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
877 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
878 SymI_HasProto(stg_WEAK_info) \
879 SymI_HasProto(stg_ap_v_info) \
880 SymI_HasProto(stg_ap_f_info) \
881 SymI_HasProto(stg_ap_d_info) \
882 SymI_HasProto(stg_ap_l_info) \
883 SymI_HasProto(stg_ap_n_info) \
884 SymI_HasProto(stg_ap_p_info) \
885 SymI_HasProto(stg_ap_pv_info) \
886 SymI_HasProto(stg_ap_pp_info) \
887 SymI_HasProto(stg_ap_ppv_info) \
888 SymI_HasProto(stg_ap_ppp_info) \
889 SymI_HasProto(stg_ap_pppv_info) \
890 SymI_HasProto(stg_ap_pppp_info) \
891 SymI_HasProto(stg_ap_ppppp_info) \
892 SymI_HasProto(stg_ap_pppppp_info) \
893 SymI_HasProto(stg_ap_0_fast) \
894 SymI_HasProto(stg_ap_v_fast) \
895 SymI_HasProto(stg_ap_f_fast) \
896 SymI_HasProto(stg_ap_d_fast) \
897 SymI_HasProto(stg_ap_l_fast) \
898 SymI_HasProto(stg_ap_n_fast) \
899 SymI_HasProto(stg_ap_p_fast) \
900 SymI_HasProto(stg_ap_pv_fast) \
901 SymI_HasProto(stg_ap_pp_fast) \
902 SymI_HasProto(stg_ap_ppv_fast) \
903 SymI_HasProto(stg_ap_ppp_fast) \
904 SymI_HasProto(stg_ap_pppv_fast) \
905 SymI_HasProto(stg_ap_pppp_fast) \
906 SymI_HasProto(stg_ap_ppppp_fast) \
907 SymI_HasProto(stg_ap_pppppp_fast) \
908 SymI_HasProto(stg_ap_1_upd_info) \
909 SymI_HasProto(stg_ap_2_upd_info) \
910 SymI_HasProto(stg_ap_3_upd_info) \
911 SymI_HasProto(stg_ap_4_upd_info) \
912 SymI_HasProto(stg_ap_5_upd_info) \
913 SymI_HasProto(stg_ap_6_upd_info) \
914 SymI_HasProto(stg_ap_7_upd_info) \
915 SymI_HasProto(stg_exit) \
916 SymI_HasProto(stg_sel_0_upd_info) \
917 SymI_HasProto(stg_sel_10_upd_info) \
918 SymI_HasProto(stg_sel_11_upd_info) \
919 SymI_HasProto(stg_sel_12_upd_info) \
920 SymI_HasProto(stg_sel_13_upd_info) \
921 SymI_HasProto(stg_sel_14_upd_info) \
922 SymI_HasProto(stg_sel_15_upd_info) \
923 SymI_HasProto(stg_sel_1_upd_info) \
924 SymI_HasProto(stg_sel_2_upd_info) \
925 SymI_HasProto(stg_sel_3_upd_info) \
926 SymI_HasProto(stg_sel_4_upd_info) \
927 SymI_HasProto(stg_sel_5_upd_info) \
928 SymI_HasProto(stg_sel_6_upd_info) \
929 SymI_HasProto(stg_sel_7_upd_info) \
930 SymI_HasProto(stg_sel_8_upd_info) \
931 SymI_HasProto(stg_sel_9_upd_info) \
932 SymI_HasProto(stg_upd_frame_info) \
933 SymI_HasProto(suspendThread) \
934 SymI_HasProto(stg_takeMVarzh) \
935 SymI_HasProto(stg_threadStatuszh) \
936 SymI_HasProto(stg_tryPutMVarzh) \
937 SymI_HasProto(stg_tryTakeMVarzh) \
938 SymI_HasProto(stg_unblockAsyncExceptionszh) \
939 SymI_HasProto(unloadObj) \
940 SymI_HasProto(stg_unsafeThawArrayzh) \
941 SymI_HasProto(stg_waitReadzh) \
942 SymI_HasProto(stg_waitWritezh) \
943 SymI_HasProto(stg_writeTVarzh) \
944 SymI_HasProto(stg_yieldzh) \
945 SymI_NeedsProto(stg_interp_constr_entry) \
946 SymI_HasProto(alloc_blocks_lim) \
947 SymI_HasProto(allocate) \
948 SymI_HasProto(allocateExec) \
949 SymI_HasProto(freeExec) \
950 SymI_HasProto(getAllocations) \
951 SymI_HasProto(revertCAFs) \
952 SymI_HasProto(RtsFlags) \
953 SymI_NeedsProto(rts_breakpoint_io_action) \
954 SymI_NeedsProto(rts_stop_next_breakpoint) \
955 SymI_NeedsProto(rts_stop_on_exception) \
956 SymI_HasProto(stopTimer) \
957 SymI_HasProto(n_capabilities) \
958 SymI_HasProto(stg_traceCcszh) \
959 SymI_HasProto(stg_traceEventzh) \
960 RTS_USER_SIGNALS_SYMBOLS \
964 // 64-bit support functions in libgcc.a
965 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
966 #define RTS_LIBGCC_SYMBOLS \
967 SymI_NeedsProto(__divdi3) \
968 SymI_NeedsProto(__udivdi3) \
969 SymI_NeedsProto(__moddi3) \
970 SymI_NeedsProto(__umoddi3) \
971 SymI_NeedsProto(__muldi3) \
972 SymI_NeedsProto(__ashldi3) \
973 SymI_NeedsProto(__ashrdi3) \
974 SymI_NeedsProto(__lshrdi3) \
975 SymI_NeedsProto(__eprintf)
977 #define RTS_LIBGCC_SYMBOLS
980 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
981 // Symbols that don't have a leading underscore
982 // on Mac OS X. They have to receive special treatment,
983 // see machoInitSymbolsWithoutUnderscore()
984 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
985 SymI_NeedsProto(saveFP) \
986 SymI_NeedsProto(restFP)
989 /* entirely bogus claims about types of these symbols */
990 #define SymI_NeedsProto(vvv) extern void vvv(void);
991 #if defined(__PIC__) && defined(mingw32_TARGET_OS)
992 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
993 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
995 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
996 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
998 #define SymI_HasProto(vvv) /**/
999 #define SymI_HasProto_redirect(vvv,xxx) /**/
1002 RTS_POSIX_ONLY_SYMBOLS
1003 RTS_MINGW_ONLY_SYMBOLS
1004 RTS_CYGWIN_ONLY_SYMBOLS
1005 RTS_DARWIN_ONLY_SYMBOLS
1008 #undef SymI_NeedsProto
1009 #undef SymI_HasProto
1010 #undef SymI_HasProto_redirect
1011 #undef SymE_HasProto
1012 #undef SymE_NeedsProto
1014 #ifdef LEADING_UNDERSCORE
1015 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1017 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1020 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1022 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1023 (void*)DLL_IMPORT_DATA_REF(vvv) },
1025 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1026 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1028 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1029 // another symbol. See newCAF/newDynCAF for an example.
1030 #define SymI_HasProto_redirect(vvv,xxx) \
1031 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1034 static RtsSymbolVal rtsSyms[] = {
1037 RTS_POSIX_ONLY_SYMBOLS
1038 RTS_MINGW_ONLY_SYMBOLS
1039 RTS_CYGWIN_ONLY_SYMBOLS
1040 RTS_DARWIN_ONLY_SYMBOLS
1043 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1044 // dyld stub code contains references to this,
1045 // but it should never be called because we treat
1046 // lazy pointers as nonlazy.
1047 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1049 { 0, 0 } /* sentinel */
1054 /* -----------------------------------------------------------------------------
1055 * Insert symbols into hash tables, checking for duplicates.
1058 static void ghciInsertStrHashTable ( char* obj_name,
1064 if (lookupHashTable(table, (StgWord)key) == NULL)
1066 insertStrHashTable(table, (StgWord)key, data);
1071 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1073 "whilst processing object file\n"
1075 "This could be caused by:\n"
1076 " * Loading two different object files which export the same symbol\n"
1077 " * Specifying the same object file twice on the GHCi command line\n"
1078 " * An incorrect `package.conf' entry, causing some object to be\n"
1080 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1087 /* -----------------------------------------------------------------------------
1088 * initialize the object linker
1092 static int linker_init_done = 0 ;
1094 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1095 static void *dl_prog_handle;
1103 /* Make initLinker idempotent, so we can call it
1104 before evey relevant operation; that means we
1105 don't need to initialise the linker separately */
1106 if (linker_init_done == 1) { return; } else {
1107 linker_init_done = 1;
1110 stablehash = allocStrHashTable();
1111 symhash = allocStrHashTable();
1113 /* populate the symbol table with stuff from the RTS */
1114 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1115 ghciInsertStrHashTable("(GHCi built-in symbols)",
1116 symhash, sym->lbl, sym->addr);
1118 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1119 machoInitSymbolsWithoutUnderscore();
1122 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1123 # if defined(RTLD_DEFAULT)
1124 dl_prog_handle = RTLD_DEFAULT;
1126 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1127 # endif /* RTLD_DEFAULT */
1130 #if defined(x86_64_HOST_ARCH)
1131 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1132 // User-override for mmap_32bit_base
1133 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1137 #if defined(mingw32_HOST_OS)
1139 * These two libraries cause problems when added to the static link,
1140 * but are necessary for resolving symbols in GHCi, hence we load
1141 * them manually here.
1148 /* -----------------------------------------------------------------------------
1149 * Loading DLL or .so dynamic libraries
1150 * -----------------------------------------------------------------------------
1152 * Add a DLL from which symbols may be found. In the ELF case, just
1153 * do RTLD_GLOBAL-style add, so no further messing around needs to
1154 * happen in order that symbols in the loaded .so are findable --
1155 * lookupSymbol() will subsequently see them by dlsym on the program's
1156 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1158 * In the PEi386 case, open the DLLs and put handles to them in a
1159 * linked list. When looking for a symbol, try all handles in the
1160 * list. This means that we need to load even DLLs that are guaranteed
1161 * to be in the ghc.exe image already, just so we can get a handle
1162 * to give to loadSymbol, so that we can find the symbols. For such
1163 * libraries, the LoadLibrary call should be a no-op except for returning
1168 #if defined(OBJFORMAT_PEi386)
1169 /* A record for storing handles into DLLs. */
1174 struct _OpenedDLL* next;
1179 /* A list thereof. */
1180 static OpenedDLL* opened_dlls = NULL;
1184 addDLL( char *dll_name )
1186 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1187 /* ------------------- ELF DLL loader ------------------- */
1193 // omitted: RTLD_NOW
1194 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1195 hdl= dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1198 /* dlopen failed; return a ptr to the error msg. */
1200 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1207 # elif defined(OBJFORMAT_PEi386)
1208 /* ------------------- Win32 DLL loader ------------------- */
1216 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1218 /* See if we've already got it, and ignore if so. */
1219 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1220 if (0 == strcmp(o_dll->name, dll_name))
1224 /* The file name has no suffix (yet) so that we can try
1225 both foo.dll and foo.drv
1227 The documentation for LoadLibrary says:
1228 If no file name extension is specified in the lpFileName
1229 parameter, the default library extension .dll is
1230 appended. However, the file name string can include a trailing
1231 point character (.) to indicate that the module name has no
1234 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1235 sprintf(buf, "%s.DLL", dll_name);
1236 instance = LoadLibrary(buf);
1237 if (instance == NULL) {
1238 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1239 // KAA: allow loading of drivers (like winspool.drv)
1240 sprintf(buf, "%s.DRV", dll_name);
1241 instance = LoadLibrary(buf);
1242 if (instance == NULL) {
1243 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1244 // #1883: allow loading of unix-style libfoo.dll DLLs
1245 sprintf(buf, "lib%s.DLL", dll_name);
1246 instance = LoadLibrary(buf);
1247 if (instance == NULL) {
1254 /* Add this DLL to the list of DLLs in which to search for symbols. */
1255 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1256 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1257 strcpy(o_dll->name, dll_name);
1258 o_dll->instance = instance;
1259 o_dll->next = opened_dlls;
1260 opened_dlls = o_dll;
1266 sysErrorBelch(dll_name);
1268 /* LoadLibrary failed; return a ptr to the error msg. */
1269 return "addDLL: could not load DLL";
1272 barf("addDLL: not implemented on this platform");
1276 /* -----------------------------------------------------------------------------
1277 * insert a stable symbol in the hash table
1281 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1283 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1287 /* -----------------------------------------------------------------------------
1288 * insert a symbol in the hash table
1291 insertSymbol(char* obj_name, char* key, void* data)
1293 ghciInsertStrHashTable(obj_name, symhash, key, data);
1296 /* -----------------------------------------------------------------------------
1297 * lookup a symbol in the hash table
1300 lookupSymbol( char *lbl )
1304 ASSERT(symhash != NULL);
1305 val = lookupStrHashTable(symhash, lbl);
1308 # if defined(OBJFORMAT_ELF)
1309 return dlsym(dl_prog_handle, lbl);
1310 # elif defined(OBJFORMAT_MACHO)
1312 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1315 HACK: On OS X, global symbols are prefixed with an underscore.
1316 However, dlsym wants us to omit the leading underscore from the
1317 symbol name. For now, we simply strip it off here (and ONLY
1320 ASSERT(lbl[0] == '_');
1321 return dlsym(dl_prog_handle, lbl+1);
1323 if(NSIsSymbolNameDefined(lbl)) {
1324 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1325 return NSAddressOfSymbol(symbol);
1329 # endif /* HAVE_DLFCN_H */
1330 # elif defined(OBJFORMAT_PEi386)
1333 sym = lookupSymbolInDLLs(lbl);
1334 if (sym != NULL) { return sym; };
1336 // Also try looking up the symbol without the @N suffix. Some
1337 // DLLs have the suffixes on their symbols, some don't.
1338 zapTrailingAtSign ( lbl );
1339 sym = lookupSymbolInDLLs(lbl);
1340 if (sym != NULL) { return sym; };
1352 /* -----------------------------------------------------------------------------
1353 * Debugging aid: look in GHCi's object symbol tables for symbols
1354 * within DELTA bytes of the specified address, and show their names.
1357 void ghci_enquire ( char* addr );
1359 void ghci_enquire ( char* addr )
1364 const int DELTA = 64;
1369 for (oc = objects; oc; oc = oc->next) {
1370 for (i = 0; i < oc->n_symbols; i++) {
1371 sym = oc->symbols[i];
1372 if (sym == NULL) continue;
1375 a = lookupStrHashTable(symhash, sym);
1378 // debugBelch("ghci_enquire: can't find %s\n", sym);
1380 else if (addr-DELTA <= a && a <= addr+DELTA) {
1381 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1389 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1392 mmapForLinker (size_t bytes, nat flags, int fd)
1394 void *map_addr = NULL;
1397 static nat fixed = 0;
1399 pagesize = getpagesize();
1400 size = ROUND_UP(bytes, pagesize);
1402 #if defined(x86_64_HOST_ARCH)
1405 if (mmap_32bit_base != 0) {
1406 map_addr = mmap_32bit_base;
1410 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1411 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1413 if (result == MAP_FAILED) {
1414 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1415 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1416 stg_exit(EXIT_FAILURE);
1419 #if defined(x86_64_HOST_ARCH)
1420 if (mmap_32bit_base != 0) {
1421 if (result == map_addr) {
1422 mmap_32bit_base = (StgWord8*)map_addr + size;
1424 if ((W_)result > 0x80000000) {
1425 // oops, we were given memory over 2Gb
1426 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1427 // Some platforms require MAP_FIXED. This is normally
1428 // a bad idea, because MAP_FIXED will overwrite
1429 // existing mappings.
1430 munmap(result,size);
1434 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);
1437 // hmm, we were given memory somewhere else, but it's
1438 // still under 2Gb so we can use it. Next time, ask
1439 // for memory right after the place we just got some
1440 mmap_32bit_base = (StgWord8*)result + size;
1444 if ((W_)result > 0x80000000) {
1445 // oops, we were given memory over 2Gb
1446 // ... try allocating memory somewhere else?;
1447 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1448 munmap(result, size);
1450 // Set a base address and try again... (guess: 1Gb)
1451 mmap_32bit_base = (void*)0x40000000;
1461 /* -----------------------------------------------------------------------------
1462 * Load an obj (populate the global symbol table, but don't resolve yet)
1464 * Returns: 1 if ok, 0 on error.
1467 loadObj( char *path )
1479 /* debugBelch("loadObj %s\n", path ); */
1481 /* Check that we haven't already loaded this object.
1482 Ignore requests to load multiple times */
1486 for (o = objects; o; o = o->next) {
1487 if (0 == strcmp(o->fileName, path)) {
1489 break; /* don't need to search further */
1493 IF_DEBUG(linker, debugBelch(
1494 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1495 "same object file twice:\n"
1497 "GHCi will ignore this, but be warned.\n"
1499 return 1; /* success */
1503 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1505 # if defined(OBJFORMAT_ELF)
1506 oc->formatName = "ELF";
1507 # elif defined(OBJFORMAT_PEi386)
1508 oc->formatName = "PEi386";
1509 # elif defined(OBJFORMAT_MACHO)
1510 oc->formatName = "Mach-O";
1513 barf("loadObj: not implemented on this platform");
1516 r = stat(path, &st);
1517 if (r == -1) { return 0; }
1519 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1520 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1521 strcpy(oc->fileName, path);
1523 oc->fileSize = st.st_size;
1525 oc->sections = NULL;
1526 oc->proddables = NULL;
1528 /* chain it onto the list of objects */
1533 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1535 #if defined(openbsd_HOST_OS)
1536 fd = open(path, O_RDONLY, S_IRUSR);
1538 fd = open(path, O_RDONLY);
1541 barf("loadObj: can't open `%s'", path);
1543 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1547 #else /* !USE_MMAP */
1548 /* load the image into memory */
1549 f = fopen(path, "rb");
1551 barf("loadObj: can't read `%s'", path);
1553 # if defined(mingw32_HOST_OS)
1554 // TODO: We would like to use allocateExec here, but allocateExec
1555 // cannot currently allocate blocks large enough.
1556 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1557 PAGE_EXECUTE_READWRITE);
1558 # elif defined(darwin_HOST_OS)
1559 // In a Mach-O .o file, all sections can and will be misaligned
1560 // if the total size of the headers is not a multiple of the
1561 // desired alignment. This is fine for .o files that only serve
1562 // as input for the static linker, but it's not fine for us,
1563 // as SSE (used by gcc for floating point) and Altivec require
1564 // 16-byte alignment.
1565 // We calculate the correct alignment from the header before
1566 // reading the file, and then we misalign oc->image on purpose so
1567 // that the actual sections end up aligned again.
1568 oc->misalignment = machoGetMisalignment(f);
1569 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1570 oc->image += oc->misalignment;
1572 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1577 n = fread ( oc->image, 1, oc->fileSize, f );
1578 if (n != oc->fileSize)
1579 barf("loadObj: error whilst reading `%s'", path);
1582 #endif /* USE_MMAP */
1584 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1585 r = ocAllocateSymbolExtras_MachO ( oc );
1586 if (!r) { return r; }
1587 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1588 r = ocAllocateSymbolExtras_ELF ( oc );
1589 if (!r) { return r; }
1592 /* verify the in-memory image */
1593 # if defined(OBJFORMAT_ELF)
1594 r = ocVerifyImage_ELF ( oc );
1595 # elif defined(OBJFORMAT_PEi386)
1596 r = ocVerifyImage_PEi386 ( oc );
1597 # elif defined(OBJFORMAT_MACHO)
1598 r = ocVerifyImage_MachO ( oc );
1600 barf("loadObj: no verify method");
1602 if (!r) { return r; }
1604 /* build the symbol list for this image */
1605 # if defined(OBJFORMAT_ELF)
1606 r = ocGetNames_ELF ( oc );
1607 # elif defined(OBJFORMAT_PEi386)
1608 r = ocGetNames_PEi386 ( oc );
1609 # elif defined(OBJFORMAT_MACHO)
1610 r = ocGetNames_MachO ( oc );
1612 barf("loadObj: no getNames method");
1614 if (!r) { return r; }
1616 /* loaded, but not resolved yet */
1617 oc->status = OBJECT_LOADED;
1622 /* -----------------------------------------------------------------------------
1623 * resolve all the currently unlinked objects in memory
1625 * Returns: 1 if ok, 0 on error.
1635 for (oc = objects; oc; oc = oc->next) {
1636 if (oc->status != OBJECT_RESOLVED) {
1637 # if defined(OBJFORMAT_ELF)
1638 r = ocResolve_ELF ( oc );
1639 # elif defined(OBJFORMAT_PEi386)
1640 r = ocResolve_PEi386 ( oc );
1641 # elif defined(OBJFORMAT_MACHO)
1642 r = ocResolve_MachO ( oc );
1644 barf("resolveObjs: not implemented on this platform");
1646 if (!r) { return r; }
1647 oc->status = OBJECT_RESOLVED;
1653 /* -----------------------------------------------------------------------------
1654 * delete an object from the pool
1657 unloadObj( char *path )
1659 ObjectCode *oc, *prev;
1661 ASSERT(symhash != NULL);
1662 ASSERT(objects != NULL);
1667 for (oc = objects; oc; prev = oc, oc = oc->next) {
1668 if (!strcmp(oc->fileName,path)) {
1670 /* Remove all the mappings for the symbols within this
1675 for (i = 0; i < oc->n_symbols; i++) {
1676 if (oc->symbols[i] != NULL) {
1677 removeStrHashTable(symhash, oc->symbols[i], NULL);
1685 prev->next = oc->next;
1688 // We're going to leave this in place, in case there are
1689 // any pointers from the heap into it:
1690 // #ifdef mingw32_HOST_OS
1691 // VirtualFree(oc->image);
1693 // stgFree(oc->image);
1695 stgFree(oc->fileName);
1696 stgFree(oc->symbols);
1697 stgFree(oc->sections);
1703 errorBelch("unloadObj: can't find `%s' to unload", path);
1707 /* -----------------------------------------------------------------------------
1708 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1709 * which may be prodded during relocation, and abort if we try and write
1710 * outside any of these.
1712 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1715 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1716 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1720 pb->next = oc->proddables;
1721 oc->proddables = pb;
1724 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1727 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1728 char* s = (char*)(pb->start);
1729 char* e = s + pb->size - 1;
1730 char* a = (char*)addr;
1731 /* Assumes that the biggest fixup involves a 4-byte write. This
1732 probably needs to be changed to 8 (ie, +7) on 64-bit
1734 if (a >= s && (a+3) <= e) return;
1736 barf("checkProddableBlock: invalid fixup in runtime linker");
1739 /* -----------------------------------------------------------------------------
1740 * Section management.
1742 static void addSection ( ObjectCode* oc, SectionKind kind,
1743 void* start, void* end )
1745 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1749 s->next = oc->sections;
1752 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1753 start, ((char*)end)-1, end - start + 1, kind );
1758 /* --------------------------------------------------------------------------
1760 * This is about allocating a small chunk of memory for every symbol in the
1761 * object file. We make sure that the SymboLExtras are always "in range" of
1762 * limited-range PC-relative instructions on various platforms by allocating
1763 * them right next to the object code itself.
1766 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1769 ocAllocateSymbolExtras
1771 Allocate additional space at the end of the object file image to make room
1772 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1774 PowerPC relative branch instructions have a 24 bit displacement field.
1775 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1776 If a particular imported symbol is outside this range, we have to redirect
1777 the jump to a short piece of new code that just loads the 32bit absolute
1778 address and jumps there.
1779 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1782 This function just allocates space for one SymbolExtra for every
1783 undefined symbol in the object file. The code for the jump islands is
1784 filled in by makeSymbolExtra below.
1787 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1794 int misalignment = 0;
1795 #ifdef darwin_HOST_OS
1796 misalignment = oc->misalignment;
1802 // round up to the nearest 4
1803 aligned = (oc->fileSize + 3) & ~3;
1806 pagesize = getpagesize();
1807 n = ROUND_UP( oc->fileSize, pagesize );
1808 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1810 /* we try to use spare space at the end of the last page of the
1811 * image for the jump islands, but if there isn't enough space
1812 * then we have to map some (anonymously, remembering MAP_32BIT).
1814 if( m > n ) // we need to allocate more pages
1816 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1821 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1824 oc->image -= misalignment;
1825 oc->image = stgReallocBytes( oc->image,
1827 aligned + sizeof (SymbolExtra) * count,
1828 "ocAllocateSymbolExtras" );
1829 oc->image += misalignment;
1831 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1832 #endif /* USE_MMAP */
1834 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1837 oc->symbol_extras = NULL;
1839 oc->first_symbol_extra = first;
1840 oc->n_symbol_extras = count;
1845 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1846 unsigned long symbolNumber,
1847 unsigned long target )
1851 ASSERT( symbolNumber >= oc->first_symbol_extra
1852 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1854 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1856 #ifdef powerpc_HOST_ARCH
1857 // lis r12, hi16(target)
1858 extra->jumpIsland.lis_r12 = 0x3d80;
1859 extra->jumpIsland.hi_addr = target >> 16;
1861 // ori r12, r12, lo16(target)
1862 extra->jumpIsland.ori_r12_r12 = 0x618c;
1863 extra->jumpIsland.lo_addr = target & 0xffff;
1866 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1869 extra->jumpIsland.bctr = 0x4e800420;
1871 #ifdef x86_64_HOST_ARCH
1873 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1874 extra->addr = target;
1875 memcpy(extra->jumpIsland, jmp, 6);
1883 /* --------------------------------------------------------------------------
1884 * PowerPC specifics (instruction cache flushing)
1885 * ------------------------------------------------------------------------*/
1887 #ifdef powerpc_TARGET_ARCH
1889 ocFlushInstructionCache
1891 Flush the data & instruction caches.
1892 Because the PPC has split data/instruction caches, we have to
1893 do that whenever we modify code at runtime.
1896 static void ocFlushInstructionCache( ObjectCode *oc )
1898 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1899 unsigned long *p = (unsigned long *) oc->image;
1903 __asm__ volatile ( "dcbf 0,%0\n\t"
1911 __asm__ volatile ( "sync\n\t"
1917 /* --------------------------------------------------------------------------
1918 * PEi386 specifics (Win32 targets)
1919 * ------------------------------------------------------------------------*/
1921 /* The information for this linker comes from
1922 Microsoft Portable Executable
1923 and Common Object File Format Specification
1924 revision 5.1 January 1998
1925 which SimonM says comes from the MS Developer Network CDs.
1927 It can be found there (on older CDs), but can also be found
1930 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1932 (this is Rev 6.0 from February 1999).
1934 Things move, so if that fails, try searching for it via
1936 http://www.google.com/search?q=PE+COFF+specification
1938 The ultimate reference for the PE format is the Winnt.h
1939 header file that comes with the Platform SDKs; as always,
1940 implementations will drift wrt their documentation.
1942 A good background article on the PE format is Matt Pietrek's
1943 March 1994 article in Microsoft System Journal (MSJ)
1944 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1945 Win32 Portable Executable File Format." The info in there
1946 has recently been updated in a two part article in
1947 MSDN magazine, issues Feb and March 2002,
1948 "Inside Windows: An In-Depth Look into the Win32 Portable
1949 Executable File Format"
1951 John Levine's book "Linkers and Loaders" contains useful
1956 #if defined(OBJFORMAT_PEi386)
1960 typedef unsigned char UChar;
1961 typedef unsigned short UInt16;
1962 typedef unsigned int UInt32;
1969 UInt16 NumberOfSections;
1970 UInt32 TimeDateStamp;
1971 UInt32 PointerToSymbolTable;
1972 UInt32 NumberOfSymbols;
1973 UInt16 SizeOfOptionalHeader;
1974 UInt16 Characteristics;
1978 #define sizeof_COFF_header 20
1985 UInt32 VirtualAddress;
1986 UInt32 SizeOfRawData;
1987 UInt32 PointerToRawData;
1988 UInt32 PointerToRelocations;
1989 UInt32 PointerToLinenumbers;
1990 UInt16 NumberOfRelocations;
1991 UInt16 NumberOfLineNumbers;
1992 UInt32 Characteristics;
1996 #define sizeof_COFF_section 40
2003 UInt16 SectionNumber;
2006 UChar NumberOfAuxSymbols;
2010 #define sizeof_COFF_symbol 18
2015 UInt32 VirtualAddress;
2016 UInt32 SymbolTableIndex;
2021 #define sizeof_COFF_reloc 10
2024 /* From PE spec doc, section 3.3.2 */
2025 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2026 windows.h -- for the same purpose, but I want to know what I'm
2028 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2029 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2030 #define MYIMAGE_FILE_DLL 0x2000
2031 #define MYIMAGE_FILE_SYSTEM 0x1000
2032 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2033 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2034 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2036 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2037 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2038 #define MYIMAGE_SYM_CLASS_STATIC 3
2039 #define MYIMAGE_SYM_UNDEFINED 0
2041 /* From PE spec doc, section 4.1 */
2042 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2043 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2044 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2046 /* From PE spec doc, section 5.2.1 */
2047 #define MYIMAGE_REL_I386_DIR32 0x0006
2048 #define MYIMAGE_REL_I386_REL32 0x0014
2051 /* We use myindex to calculate array addresses, rather than
2052 simply doing the normal subscript thing. That's because
2053 some of the above structs have sizes which are not
2054 a whole number of words. GCC rounds their sizes up to a
2055 whole number of words, which means that the address calcs
2056 arising from using normal C indexing or pointer arithmetic
2057 are just plain wrong. Sigh.
2060 myindex ( int scale, void* base, int index )
2063 ((UChar*)base) + scale * index;
2068 printName ( UChar* name, UChar* strtab )
2070 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2071 UInt32 strtab_offset = * (UInt32*)(name+4);
2072 debugBelch("%s", strtab + strtab_offset );
2075 for (i = 0; i < 8; i++) {
2076 if (name[i] == 0) break;
2077 debugBelch("%c", name[i] );
2084 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2086 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2087 UInt32 strtab_offset = * (UInt32*)(name+4);
2088 strncpy ( dst, strtab+strtab_offset, dstSize );
2094 if (name[i] == 0) break;
2104 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2107 /* If the string is longer than 8 bytes, look in the
2108 string table for it -- this will be correctly zero terminated.
2110 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2111 UInt32 strtab_offset = * (UInt32*)(name+4);
2112 return ((UChar*)strtab) + strtab_offset;
2114 /* Otherwise, if shorter than 8 bytes, return the original,
2115 which by defn is correctly terminated.
2117 if (name[7]==0) return name;
2118 /* The annoying case: 8 bytes. Copy into a temporary
2119 (which is never freed ...)
2121 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2123 strncpy(newstr,name,8);
2129 /* Just compares the short names (first 8 chars) */
2130 static COFF_section *
2131 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2135 = (COFF_header*)(oc->image);
2136 COFF_section* sectab
2138 ((UChar*)(oc->image))
2139 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2141 for (i = 0; i < hdr->NumberOfSections; i++) {
2144 COFF_section* section_i
2146 myindex ( sizeof_COFF_section, sectab, i );
2147 n1 = (UChar*) &(section_i->Name);
2149 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2150 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2151 n1[6]==n2[6] && n1[7]==n2[7])
2160 zapTrailingAtSign ( UChar* sym )
2162 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2164 if (sym[0] == 0) return;
2166 while (sym[i] != 0) i++;
2169 while (j > 0 && my_isdigit(sym[j])) j--;
2170 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2175 lookupSymbolInDLLs ( UChar *lbl )
2180 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2181 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2183 if (lbl[0] == '_') {
2184 /* HACK: if the name has an initial underscore, try stripping
2185 it off & look that up first. I've yet to verify whether there's
2186 a Rule that governs whether an initial '_' *should always* be
2187 stripped off when mapping from import lib name to the DLL name.
2189 sym = GetProcAddress(o_dll->instance, (lbl+1));
2191 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2195 sym = GetProcAddress(o_dll->instance, lbl);
2197 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2206 ocVerifyImage_PEi386 ( ObjectCode* oc )
2211 COFF_section* sectab;
2212 COFF_symbol* symtab;
2214 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2215 hdr = (COFF_header*)(oc->image);
2216 sectab = (COFF_section*) (
2217 ((UChar*)(oc->image))
2218 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2220 symtab = (COFF_symbol*) (
2221 ((UChar*)(oc->image))
2222 + hdr->PointerToSymbolTable
2224 strtab = ((UChar*)symtab)
2225 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2227 if (hdr->Machine != 0x14c) {
2228 errorBelch("%s: Not x86 PEi386", oc->fileName);
2231 if (hdr->SizeOfOptionalHeader != 0) {
2232 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2235 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2236 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2237 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2238 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2239 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2242 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2243 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2244 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2246 (int)(hdr->Characteristics));
2249 /* If the string table size is way crazy, this might indicate that
2250 there are more than 64k relocations, despite claims to the
2251 contrary. Hence this test. */
2252 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2254 if ( (*(UInt32*)strtab) > 600000 ) {
2255 /* Note that 600k has no special significance other than being
2256 big enough to handle the almost-2MB-sized lumps that
2257 constitute HSwin32*.o. */
2258 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2263 /* No further verification after this point; only debug printing. */
2265 IF_DEBUG(linker, i=1);
2266 if (i == 0) return 1;
2268 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2269 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2270 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2273 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2274 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2275 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2276 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2277 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2278 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2279 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2281 /* Print the section table. */
2283 for (i = 0; i < hdr->NumberOfSections; i++) {
2285 COFF_section* sectab_i
2287 myindex ( sizeof_COFF_section, sectab, i );
2294 printName ( sectab_i->Name, strtab );
2304 sectab_i->VirtualSize,
2305 sectab_i->VirtualAddress,
2306 sectab_i->SizeOfRawData,
2307 sectab_i->PointerToRawData,
2308 sectab_i->NumberOfRelocations,
2309 sectab_i->PointerToRelocations,
2310 sectab_i->PointerToRawData
2312 reltab = (COFF_reloc*) (
2313 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2316 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2317 /* If the relocation field (a short) has overflowed, the
2318 * real count can be found in the first reloc entry.
2320 * See Section 4.1 (last para) of the PE spec (rev6.0).
2322 COFF_reloc* rel = (COFF_reloc*)
2323 myindex ( sizeof_COFF_reloc, reltab, 0 );
2324 noRelocs = rel->VirtualAddress;
2327 noRelocs = sectab_i->NumberOfRelocations;
2331 for (; j < noRelocs; j++) {
2333 COFF_reloc* rel = (COFF_reloc*)
2334 myindex ( sizeof_COFF_reloc, reltab, j );
2336 " type 0x%-4x vaddr 0x%-8x name `",
2338 rel->VirtualAddress );
2339 sym = (COFF_symbol*)
2340 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2341 /* Hmm..mysterious looking offset - what's it for? SOF */
2342 printName ( sym->Name, strtab -10 );
2349 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2350 debugBelch("---START of string table---\n");
2351 for (i = 4; i < *(Int32*)strtab; i++) {
2353 debugBelch("\n"); else
2354 debugBelch("%c", strtab[i] );
2356 debugBelch("--- END of string table---\n");
2361 COFF_symbol* symtab_i;
2362 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2363 symtab_i = (COFF_symbol*)
2364 myindex ( sizeof_COFF_symbol, symtab, i );
2370 printName ( symtab_i->Name, strtab );
2379 (Int32)(symtab_i->SectionNumber),
2380 (UInt32)symtab_i->Type,
2381 (UInt32)symtab_i->StorageClass,
2382 (UInt32)symtab_i->NumberOfAuxSymbols
2384 i += symtab_i->NumberOfAuxSymbols;
2394 ocGetNames_PEi386 ( ObjectCode* oc )
2397 COFF_section* sectab;
2398 COFF_symbol* symtab;
2405 hdr = (COFF_header*)(oc->image);
2406 sectab = (COFF_section*) (
2407 ((UChar*)(oc->image))
2408 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2410 symtab = (COFF_symbol*) (
2411 ((UChar*)(oc->image))
2412 + hdr->PointerToSymbolTable
2414 strtab = ((UChar*)(oc->image))
2415 + hdr->PointerToSymbolTable
2416 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2418 /* Allocate space for any (local, anonymous) .bss sections. */
2420 for (i = 0; i < hdr->NumberOfSections; i++) {
2423 COFF_section* sectab_i
2425 myindex ( sizeof_COFF_section, sectab, i );
2426 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2427 /* sof 10/05: the PE spec text isn't too clear regarding what
2428 * the SizeOfRawData field is supposed to hold for object
2429 * file sections containing just uninitialized data -- for executables,
2430 * it is supposed to be zero; unclear what it's supposed to be
2431 * for object files. However, VirtualSize is guaranteed to be
2432 * zero for object files, which definitely suggests that SizeOfRawData
2433 * will be non-zero (where else would the size of this .bss section be
2434 * stored?) Looking at the COFF_section info for incoming object files,
2435 * this certainly appears to be the case.
2437 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2438 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2439 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2440 * variable decls into to the .bss section. (The specific function in Q which
2441 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2443 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2444 /* This is a non-empty .bss section. Allocate zeroed space for
2445 it, and set its PointerToRawData field such that oc->image +
2446 PointerToRawData == addr_of_zeroed_space. */
2447 bss_sz = sectab_i->VirtualSize;
2448 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2449 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2450 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2451 addProddableBlock(oc, zspace, bss_sz);
2452 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2455 /* Copy section information into the ObjectCode. */
2457 for (i = 0; i < hdr->NumberOfSections; i++) {
2463 = SECTIONKIND_OTHER;
2464 COFF_section* sectab_i
2466 myindex ( sizeof_COFF_section, sectab, i );
2467 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2470 /* I'm sure this is the Right Way to do it. However, the
2471 alternative of testing the sectab_i->Name field seems to
2472 work ok with Cygwin.
2474 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2475 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2476 kind = SECTIONKIND_CODE_OR_RODATA;
2479 if (0==strcmp(".text",sectab_i->Name) ||
2480 0==strcmp(".rdata",sectab_i->Name)||
2481 0==strcmp(".rodata",sectab_i->Name))
2482 kind = SECTIONKIND_CODE_OR_RODATA;
2483 if (0==strcmp(".data",sectab_i->Name) ||
2484 0==strcmp(".bss",sectab_i->Name))
2485 kind = SECTIONKIND_RWDATA;
2487 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2488 sz = sectab_i->SizeOfRawData;
2489 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2491 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2492 end = start + sz - 1;
2494 if (kind == SECTIONKIND_OTHER
2495 /* Ignore sections called which contain stabs debugging
2497 && 0 != strcmp(".stab", sectab_i->Name)
2498 && 0 != strcmp(".stabstr", sectab_i->Name)
2499 /* ignore constructor section for now */
2500 && 0 != strcmp(".ctors", sectab_i->Name)
2501 /* ignore section generated from .ident */
2502 && 0!= strcmp("/4", sectab_i->Name)
2503 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2504 && 0!= strcmp(".reloc", sectab_i->Name)
2506 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2510 if (kind != SECTIONKIND_OTHER && end >= start) {
2511 addSection(oc, kind, start, end);
2512 addProddableBlock(oc, start, end - start + 1);
2516 /* Copy exported symbols into the ObjectCode. */
2518 oc->n_symbols = hdr->NumberOfSymbols;
2519 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2520 "ocGetNames_PEi386(oc->symbols)");
2521 /* Call me paranoid; I don't care. */
2522 for (i = 0; i < oc->n_symbols; i++)
2523 oc->symbols[i] = NULL;
2527 COFF_symbol* symtab_i;
2528 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2529 symtab_i = (COFF_symbol*)
2530 myindex ( sizeof_COFF_symbol, symtab, i );
2534 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2535 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2536 /* This symbol is global and defined, viz, exported */
2537 /* for MYIMAGE_SYMCLASS_EXTERNAL
2538 && !MYIMAGE_SYM_UNDEFINED,
2539 the address of the symbol is:
2540 address of relevant section + offset in section
2542 COFF_section* sectabent
2543 = (COFF_section*) myindex ( sizeof_COFF_section,
2545 symtab_i->SectionNumber-1 );
2546 addr = ((UChar*)(oc->image))
2547 + (sectabent->PointerToRawData
2551 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2552 && symtab_i->Value > 0) {
2553 /* This symbol isn't in any section at all, ie, global bss.
2554 Allocate zeroed space for it. */
2555 addr = stgCallocBytes(1, symtab_i->Value,
2556 "ocGetNames_PEi386(non-anonymous bss)");
2557 addSection(oc, SECTIONKIND_RWDATA, addr,
2558 ((UChar*)addr) + symtab_i->Value - 1);
2559 addProddableBlock(oc, addr, symtab_i->Value);
2560 /* debugBelch("BSS section at 0x%x\n", addr); */
2563 if (addr != NULL ) {
2564 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2565 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2566 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2567 ASSERT(i >= 0 && i < oc->n_symbols);
2568 /* cstring_from_COFF_symbol_name always succeeds. */
2569 oc->symbols[i] = sname;
2570 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2574 "IGNORING symbol %d\n"
2578 printName ( symtab_i->Name, strtab );
2587 (Int32)(symtab_i->SectionNumber),
2588 (UInt32)symtab_i->Type,
2589 (UInt32)symtab_i->StorageClass,
2590 (UInt32)symtab_i->NumberOfAuxSymbols
2595 i += symtab_i->NumberOfAuxSymbols;
2604 ocResolve_PEi386 ( ObjectCode* oc )
2607 COFF_section* sectab;
2608 COFF_symbol* symtab;
2618 /* ToDo: should be variable-sized? But is at least safe in the
2619 sense of buffer-overrun-proof. */
2621 /* debugBelch("resolving for %s\n", oc->fileName); */
2623 hdr = (COFF_header*)(oc->image);
2624 sectab = (COFF_section*) (
2625 ((UChar*)(oc->image))
2626 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2628 symtab = (COFF_symbol*) (
2629 ((UChar*)(oc->image))
2630 + hdr->PointerToSymbolTable
2632 strtab = ((UChar*)(oc->image))
2633 + hdr->PointerToSymbolTable
2634 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2636 for (i = 0; i < hdr->NumberOfSections; i++) {
2637 COFF_section* sectab_i
2639 myindex ( sizeof_COFF_section, sectab, i );
2642 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2645 /* Ignore sections called which contain stabs debugging
2647 if (0 == strcmp(".stab", sectab_i->Name)
2648 || 0 == strcmp(".stabstr", sectab_i->Name)
2649 || 0 == strcmp(".ctors", sectab_i->Name))
2652 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2653 /* If the relocation field (a short) has overflowed, the
2654 * real count can be found in the first reloc entry.
2656 * See Section 4.1 (last para) of the PE spec (rev6.0).
2658 * Nov2003 update: the GNU linker still doesn't correctly
2659 * handle the generation of relocatable object files with
2660 * overflown relocations. Hence the output to warn of potential
2663 COFF_reloc* rel = (COFF_reloc*)
2664 myindex ( sizeof_COFF_reloc, reltab, 0 );
2665 noRelocs = rel->VirtualAddress;
2667 /* 10/05: we now assume (and check for) a GNU ld that is capable
2668 * of handling object files with (>2^16) of relocs.
2671 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2676 noRelocs = sectab_i->NumberOfRelocations;
2681 for (; j < noRelocs; j++) {
2683 COFF_reloc* reltab_j
2685 myindex ( sizeof_COFF_reloc, reltab, j );
2687 /* the location to patch */
2689 ((UChar*)(oc->image))
2690 + (sectab_i->PointerToRawData
2691 + reltab_j->VirtualAddress
2692 - sectab_i->VirtualAddress )
2694 /* the existing contents of pP */
2696 /* the symbol to connect to */
2697 sym = (COFF_symbol*)
2698 myindex ( sizeof_COFF_symbol,
2699 symtab, reltab_j->SymbolTableIndex );
2702 "reloc sec %2d num %3d: type 0x%-4x "
2703 "vaddr 0x%-8x name `",
2705 (UInt32)reltab_j->Type,
2706 reltab_j->VirtualAddress );
2707 printName ( sym->Name, strtab );
2708 debugBelch("'\n" ));
2710 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2711 COFF_section* section_sym
2712 = findPEi386SectionCalled ( oc, sym->Name );
2714 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2717 S = ((UInt32)(oc->image))
2718 + (section_sym->PointerToRawData
2721 copyName ( sym->Name, strtab, symbol, 1000-1 );
2722 S = (UInt32) lookupSymbol( symbol );
2723 if ((void*)S != NULL) goto foundit;
2724 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2728 checkProddableBlock(oc, pP);
2729 switch (reltab_j->Type) {
2730 case MYIMAGE_REL_I386_DIR32:
2733 case MYIMAGE_REL_I386_REL32:
2734 /* Tricky. We have to insert a displacement at
2735 pP which, when added to the PC for the _next_
2736 insn, gives the address of the target (S).
2737 Problem is to know the address of the next insn
2738 when we only know pP. We assume that this
2739 literal field is always the last in the insn,
2740 so that the address of the next insn is pP+4
2741 -- hence the constant 4.
2742 Also I don't know if A should be added, but so
2743 far it has always been zero.
2745 SOF 05/2005: 'A' (old contents of *pP) have been observed
2746 to contain values other than zero (the 'wx' object file
2747 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2748 So, add displacement to old value instead of asserting
2749 A to be zero. Fixes wxhaskell-related crashes, and no other
2750 ill effects have been observed.
2752 Update: the reason why we're seeing these more elaborate
2753 relocations is due to a switch in how the NCG compiles SRTs
2754 and offsets to them from info tables. SRTs live in .(ro)data,
2755 while info tables live in .text, causing GAS to emit REL32/DISP32
2756 relocations with non-zero values. Adding the displacement is
2757 the right thing to do.
2759 *pP = S - ((UInt32)pP) - 4 + A;
2762 debugBelch("%s: unhandled PEi386 relocation type %d",
2763 oc->fileName, reltab_j->Type);
2770 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2774 #endif /* defined(OBJFORMAT_PEi386) */
2777 /* --------------------------------------------------------------------------
2779 * ------------------------------------------------------------------------*/
2781 #if defined(OBJFORMAT_ELF)
2786 #if defined(sparc_HOST_ARCH)
2787 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2788 #elif defined(i386_HOST_ARCH)
2789 # define ELF_TARGET_386 /* Used inside <elf.h> */
2790 #elif defined(x86_64_HOST_ARCH)
2791 # define ELF_TARGET_X64_64
2795 #if !defined(openbsd_HOST_OS)
2798 /* openbsd elf has things in different places, with diff names */
2799 # include <elf_abi.h>
2800 # include <machine/reloc.h>
2801 # define R_386_32 RELOC_32
2802 # define R_386_PC32 RELOC_PC32
2805 /* If elf.h doesn't define it */
2806 # ifndef R_X86_64_PC64
2807 # define R_X86_64_PC64 24
2811 * Define a set of types which can be used for both ELF32 and ELF64
2815 #define ELFCLASS ELFCLASS64
2816 #define Elf_Addr Elf64_Addr
2817 #define Elf_Word Elf64_Word
2818 #define Elf_Sword Elf64_Sword
2819 #define Elf_Ehdr Elf64_Ehdr
2820 #define Elf_Phdr Elf64_Phdr
2821 #define Elf_Shdr Elf64_Shdr
2822 #define Elf_Sym Elf64_Sym
2823 #define Elf_Rel Elf64_Rel
2824 #define Elf_Rela Elf64_Rela
2825 #define ELF_ST_TYPE ELF64_ST_TYPE
2826 #define ELF_ST_BIND ELF64_ST_BIND
2827 #define ELF_R_TYPE ELF64_R_TYPE
2828 #define ELF_R_SYM ELF64_R_SYM
2830 #define ELFCLASS ELFCLASS32
2831 #define Elf_Addr Elf32_Addr
2832 #define Elf_Word Elf32_Word
2833 #define Elf_Sword Elf32_Sword
2834 #define Elf_Ehdr Elf32_Ehdr
2835 #define Elf_Phdr Elf32_Phdr
2836 #define Elf_Shdr Elf32_Shdr
2837 #define Elf_Sym Elf32_Sym
2838 #define Elf_Rel Elf32_Rel
2839 #define Elf_Rela Elf32_Rela
2841 #define ELF_ST_TYPE ELF32_ST_TYPE
2844 #define ELF_ST_BIND ELF32_ST_BIND
2847 #define ELF_R_TYPE ELF32_R_TYPE
2850 #define ELF_R_SYM ELF32_R_SYM
2856 * Functions to allocate entries in dynamic sections. Currently we simply
2857 * preallocate a large number, and we don't check if a entry for the given
2858 * target already exists (a linear search is too slow). Ideally these
2859 * entries would be associated with symbols.
2862 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2863 #define GOT_SIZE 0x20000
2864 #define FUNCTION_TABLE_SIZE 0x10000
2865 #define PLT_SIZE 0x08000
2868 static Elf_Addr got[GOT_SIZE];
2869 static unsigned int gotIndex;
2870 static Elf_Addr gp_val = (Elf_Addr)got;
2873 allocateGOTEntry(Elf_Addr target)
2877 if (gotIndex >= GOT_SIZE)
2878 barf("Global offset table overflow");
2880 entry = &got[gotIndex++];
2882 return (Elf_Addr)entry;
2886 #ifdef ELF_FUNCTION_DESC
2892 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2893 static unsigned int functionTableIndex;
2896 allocateFunctionDesc(Elf_Addr target)
2898 FunctionDesc *entry;
2900 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2901 barf("Function table overflow");
2903 entry = &functionTable[functionTableIndex++];
2905 entry->gp = (Elf_Addr)gp_val;
2906 return (Elf_Addr)entry;
2910 copyFunctionDesc(Elf_Addr target)
2912 FunctionDesc *olddesc = (FunctionDesc *)target;
2913 FunctionDesc *newdesc;
2915 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2916 newdesc->gp = olddesc->gp;
2917 return (Elf_Addr)newdesc;
2924 unsigned char code[sizeof(plt_code)];
2928 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2930 PLTEntry *plt = (PLTEntry *)oc->plt;
2933 if (oc->pltIndex >= PLT_SIZE)
2934 barf("Procedure table overflow");
2936 entry = &plt[oc->pltIndex++];
2937 memcpy(entry->code, plt_code, sizeof(entry->code));
2938 PLT_RELOC(entry->code, target);
2939 return (Elf_Addr)entry;
2945 return (PLT_SIZE * sizeof(PLTEntry));
2951 * Generic ELF functions
2955 findElfSection ( void* objImage, Elf_Word sh_type )
2957 char* ehdrC = (char*)objImage;
2958 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2959 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2960 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2964 for (i = 0; i < ehdr->e_shnum; i++) {
2965 if (shdr[i].sh_type == sh_type
2966 /* Ignore the section header's string table. */
2967 && i != ehdr->e_shstrndx
2968 /* Ignore string tables named .stabstr, as they contain
2970 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2972 ptr = ehdrC + shdr[i].sh_offset;
2980 ocVerifyImage_ELF ( ObjectCode* oc )
2984 int i, j, nent, nstrtab, nsymtabs;
2988 char* ehdrC = (char*)(oc->image);
2989 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2991 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2992 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2993 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2994 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2995 errorBelch("%s: not an ELF object", oc->fileName);
2999 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3000 errorBelch("%s: unsupported ELF format", oc->fileName);
3004 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3005 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3007 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3008 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3010 errorBelch("%s: unknown endiannness", oc->fileName);
3014 if (ehdr->e_type != ET_REL) {
3015 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3018 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3020 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3021 switch (ehdr->e_machine) {
3022 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3023 #ifdef EM_SPARC32PLUS
3024 case EM_SPARC32PLUS:
3026 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3028 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3030 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3032 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3033 #elif defined(EM_AMD64)
3034 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3036 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3037 errorBelch("%s: unknown architecture (e_machine == %d)"
3038 , oc->fileName, ehdr->e_machine);
3042 IF_DEBUG(linker,debugBelch(
3043 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3044 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3046 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3048 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3050 if (ehdr->e_shstrndx == SHN_UNDEF) {
3051 errorBelch("%s: no section header string table", oc->fileName);
3054 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3056 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3059 for (i = 0; i < ehdr->e_shnum; i++) {
3060 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3061 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3062 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3063 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3064 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3065 ehdrC + shdr[i].sh_offset,
3066 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3068 if (shdr[i].sh_type == SHT_REL) {
3069 IF_DEBUG(linker,debugBelch("Rel " ));
3070 } else if (shdr[i].sh_type == SHT_RELA) {
3071 IF_DEBUG(linker,debugBelch("RelA " ));
3073 IF_DEBUG(linker,debugBelch(" "));
3076 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3080 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3083 for (i = 0; i < ehdr->e_shnum; i++) {
3084 if (shdr[i].sh_type == SHT_STRTAB
3085 /* Ignore the section header's string table. */
3086 && i != ehdr->e_shstrndx
3087 /* Ignore string tables named .stabstr, as they contain
3089 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3091 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3092 strtab = ehdrC + shdr[i].sh_offset;
3097 errorBelch("%s: no string tables, or too many", oc->fileName);
3102 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3103 for (i = 0; i < ehdr->e_shnum; i++) {
3104 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3105 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3107 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3108 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3109 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3111 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3113 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3114 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3117 for (j = 0; j < nent; j++) {
3118 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3119 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3120 (int)stab[j].st_shndx,
3121 (int)stab[j].st_size,
3122 (char*)stab[j].st_value ));
3124 IF_DEBUG(linker,debugBelch("type=" ));
3125 switch (ELF_ST_TYPE(stab[j].st_info)) {
3126 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3127 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3128 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3129 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3130 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3131 default: IF_DEBUG(linker,debugBelch("? " )); break;
3133 IF_DEBUG(linker,debugBelch(" " ));
3135 IF_DEBUG(linker,debugBelch("bind=" ));
3136 switch (ELF_ST_BIND(stab[j].st_info)) {
3137 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3138 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3139 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3140 default: IF_DEBUG(linker,debugBelch("? " )); break;
3142 IF_DEBUG(linker,debugBelch(" " ));
3144 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3148 if (nsymtabs == 0) {
3149 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3156 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3160 if (hdr->sh_type == SHT_PROGBITS
3161 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3162 /* .text-style section */
3163 return SECTIONKIND_CODE_OR_RODATA;
3166 if (hdr->sh_type == SHT_PROGBITS
3167 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3168 /* .data-style section */
3169 return SECTIONKIND_RWDATA;
3172 if (hdr->sh_type == SHT_PROGBITS
3173 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3174 /* .rodata-style section */
3175 return SECTIONKIND_CODE_OR_RODATA;
3178 if (hdr->sh_type == SHT_NOBITS
3179 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3180 /* .bss-style section */
3182 return SECTIONKIND_RWDATA;
3185 return SECTIONKIND_OTHER;
3190 ocGetNames_ELF ( ObjectCode* oc )
3195 char* ehdrC = (char*)(oc->image);
3196 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3197 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3198 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3200 ASSERT(symhash != NULL);
3203 errorBelch("%s: no strtab", oc->fileName);
3208 for (i = 0; i < ehdr->e_shnum; i++) {
3209 /* Figure out what kind of section it is. Logic derived from
3210 Figure 1.14 ("Special Sections") of the ELF document
3211 ("Portable Formats Specification, Version 1.1"). */
3213 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3215 if (is_bss && shdr[i].sh_size > 0) {
3216 /* This is a non-empty .bss section. Allocate zeroed space for
3217 it, and set its .sh_offset field such that
3218 ehdrC + .sh_offset == addr_of_zeroed_space. */
3219 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3220 "ocGetNames_ELF(BSS)");
3221 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3223 debugBelch("BSS section at 0x%x, size %d\n",
3224 zspace, shdr[i].sh_size);
3228 /* fill in the section info */
3229 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3230 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3231 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3232 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3235 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3237 /* copy stuff into this module's object symbol table */
3238 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3239 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3241 oc->n_symbols = nent;
3242 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3243 "ocGetNames_ELF(oc->symbols)");
3245 for (j = 0; j < nent; j++) {
3247 char isLocal = FALSE; /* avoids uninit-var warning */
3249 char* nm = strtab + stab[j].st_name;
3250 int secno = stab[j].st_shndx;
3252 /* Figure out if we want to add it; if so, set ad to its
3253 address. Otherwise leave ad == NULL. */
3255 if (secno == SHN_COMMON) {
3257 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3259 debugBelch("COMMON symbol, size %d name %s\n",
3260 stab[j].st_size, nm);
3262 /* Pointless to do addProddableBlock() for this area,
3263 since the linker should never poke around in it. */
3266 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3267 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3269 /* and not an undefined symbol */
3270 && stab[j].st_shndx != SHN_UNDEF
3271 /* and not in a "special section" */
3272 && stab[j].st_shndx < SHN_LORESERVE
3274 /* and it's a not a section or string table or anything silly */
3275 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3276 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3277 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3280 /* Section 0 is the undefined section, hence > and not >=. */
3281 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3283 if (shdr[secno].sh_type == SHT_NOBITS) {
3284 debugBelch(" BSS symbol, size %d off %d name %s\n",
3285 stab[j].st_size, stab[j].st_value, nm);
3288 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3289 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3292 #ifdef ELF_FUNCTION_DESC
3293 /* dlsym() and the initialisation table both give us function
3294 * descriptors, so to be consistent we store function descriptors
3295 * in the symbol table */
3296 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3297 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3299 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3300 ad, oc->fileName, nm ));
3305 /* And the decision is ... */
3309 oc->symbols[j] = nm;
3312 /* Ignore entirely. */
3314 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3318 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3319 strtab + stab[j].st_name ));
3322 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3323 (int)ELF_ST_BIND(stab[j].st_info),
3324 (int)ELF_ST_TYPE(stab[j].st_info),
3325 (int)stab[j].st_shndx,
3326 strtab + stab[j].st_name
3329 oc->symbols[j] = NULL;
3338 /* Do ELF relocations which lack an explicit addend. All x86-linux
3339 relocations appear to be of this form. */
3341 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3342 Elf_Shdr* shdr, int shnum,
3343 Elf_Sym* stab, char* strtab )
3348 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3349 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3350 int target_shndx = shdr[shnum].sh_info;
3351 int symtab_shndx = shdr[shnum].sh_link;
3353 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3354 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3355 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3356 target_shndx, symtab_shndx ));
3358 /* Skip sections that we're not interested in. */
3361 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3362 if (kind == SECTIONKIND_OTHER) {
3363 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3368 for (j = 0; j < nent; j++) {
3369 Elf_Addr offset = rtab[j].r_offset;
3370 Elf_Addr info = rtab[j].r_info;
3372 Elf_Addr P = ((Elf_Addr)targ) + offset;
3373 Elf_Word* pP = (Elf_Word*)P;
3378 StgStablePtr stablePtr;
3381 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3382 j, (void*)offset, (void*)info ));
3384 IF_DEBUG(linker,debugBelch( " ZERO" ));
3387 Elf_Sym sym = stab[ELF_R_SYM(info)];
3388 /* First see if it is a local symbol. */
3389 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3390 /* Yes, so we can get the address directly from the ELF symbol
3392 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3394 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3395 + stab[ELF_R_SYM(info)].st_value);
3398 symbol = strtab + sym.st_name;
3399 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3400 if (NULL == stablePtr) {
3401 /* No, so look up the name in our global table. */
3402 S_tmp = lookupSymbol( symbol );
3403 S = (Elf_Addr)S_tmp;
3405 stableVal = deRefStablePtr( stablePtr );
3407 S = (Elf_Addr)S_tmp;
3411 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3414 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3417 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3418 (void*)P, (void*)S, (void*)A ));
3419 checkProddableBlock ( oc, pP );
3423 switch (ELF_R_TYPE(info)) {
3424 # ifdef i386_HOST_ARCH
3425 case R_386_32: *pP = value; break;
3426 case R_386_PC32: *pP = value - P; break;
3429 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3430 oc->fileName, (lnat)ELF_R_TYPE(info));
3438 /* Do ELF relocations for which explicit addends are supplied.
3439 sparc-solaris relocations appear to be of this form. */
3441 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3442 Elf_Shdr* shdr, int shnum,
3443 Elf_Sym* stab, char* strtab )
3446 char *symbol = NULL;
3448 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3449 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3450 int target_shndx = shdr[shnum].sh_info;
3451 int symtab_shndx = shdr[shnum].sh_link;
3453 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3454 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3455 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3456 target_shndx, symtab_shndx ));
3458 for (j = 0; j < nent; j++) {
3459 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3460 /* This #ifdef only serves to avoid unused-var warnings. */
3461 Elf_Addr offset = rtab[j].r_offset;
3462 Elf_Addr P = targ + offset;
3464 Elf_Addr info = rtab[j].r_info;
3465 Elf_Addr A = rtab[j].r_addend;
3469 # if defined(sparc_HOST_ARCH)
3470 Elf_Word* pP = (Elf_Word*)P;
3472 # elif defined(powerpc_HOST_ARCH)
3476 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3477 j, (void*)offset, (void*)info,
3480 IF_DEBUG(linker,debugBelch( " ZERO" ));
3483 Elf_Sym sym = stab[ELF_R_SYM(info)];
3484 /* First see if it is a local symbol. */
3485 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3486 /* Yes, so we can get the address directly from the ELF symbol
3488 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3490 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3491 + stab[ELF_R_SYM(info)].st_value);
3492 #ifdef ELF_FUNCTION_DESC
3493 /* Make a function descriptor for this function */
3494 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3495 S = allocateFunctionDesc(S + A);
3500 /* No, so look up the name in our global table. */
3501 symbol = strtab + sym.st_name;
3502 S_tmp = lookupSymbol( symbol );
3503 S = (Elf_Addr)S_tmp;
3505 #ifdef ELF_FUNCTION_DESC
3506 /* If a function, already a function descriptor - we would
3507 have to copy it to add an offset. */
3508 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3509 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3513 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3516 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3519 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3520 (void*)P, (void*)S, (void*)A ));
3521 /* checkProddableBlock ( oc, (void*)P ); */
3525 switch (ELF_R_TYPE(info)) {
3526 # if defined(sparc_HOST_ARCH)
3527 case R_SPARC_WDISP30:
3528 w1 = *pP & 0xC0000000;
3529 w2 = (Elf_Word)((value - P) >> 2);
3530 ASSERT((w2 & 0xC0000000) == 0);
3535 w1 = *pP & 0xFFC00000;
3536 w2 = (Elf_Word)(value >> 10);
3537 ASSERT((w2 & 0xFFC00000) == 0);
3543 w2 = (Elf_Word)(value & 0x3FF);
3544 ASSERT((w2 & ~0x3FF) == 0);
3549 /* According to the Sun documentation:
3551 This relocation type resembles R_SPARC_32, except it refers to an
3552 unaligned word. That is, the word to be relocated must be treated
3553 as four separate bytes with arbitrary alignment, not as a word
3554 aligned according to the architecture requirements.
3557 w2 = (Elf_Word)value;
3559 // SPARC doesn't do misaligned writes of 32 bit words,
3560 // so we have to do this one byte-at-a-time.
3561 char *pPc = (char*)pP;
3562 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3563 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3564 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3565 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3569 w2 = (Elf_Word)value;
3572 # elif defined(powerpc_HOST_ARCH)
3573 case R_PPC_ADDR16_LO:
3574 *(Elf32_Half*) P = value;
3577 case R_PPC_ADDR16_HI:
3578 *(Elf32_Half*) P = value >> 16;
3581 case R_PPC_ADDR16_HA:
3582 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3586 *(Elf32_Word *) P = value;
3590 *(Elf32_Word *) P = value - P;
3596 if( delta << 6 >> 6 != delta )
3598 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3602 if( value == 0 || delta << 6 >> 6 != delta )
3604 barf( "Unable to make SymbolExtra for #%d",
3610 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3611 | (delta & 0x3fffffc);
3615 #if x86_64_HOST_ARCH
3617 *(Elf64_Xword *)P = value;
3622 StgInt64 off = value - P;
3623 if (off >= 0x7fffffffL || off < -0x80000000L) {
3624 #if X86_64_ELF_NONPIC_HACK
3625 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3627 off = pltAddress + A - P;
3629 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3630 symbol, off, oc->fileName );
3633 *(Elf64_Word *)P = (Elf64_Word)off;
3639 StgInt64 off = value - P;
3640 *(Elf64_Word *)P = (Elf64_Word)off;
3645 if (value >= 0x7fffffffL) {
3646 #if X86_64_ELF_NONPIC_HACK
3647 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3649 value = pltAddress + A;
3651 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3652 symbol, value, oc->fileName );
3655 *(Elf64_Word *)P = (Elf64_Word)value;
3659 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3660 #if X86_64_ELF_NONPIC_HACK
3661 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3663 value = pltAddress + A;
3665 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3666 symbol, value, oc->fileName );
3669 *(Elf64_Sword *)P = (Elf64_Sword)value;
3672 case R_X86_64_GOTPCREL:
3674 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3675 StgInt64 off = gotAddress + A - P;
3676 *(Elf64_Word *)P = (Elf64_Word)off;
3680 case R_X86_64_PLT32:
3682 StgInt64 off = value - P;
3683 if (off >= 0x7fffffffL || off < -0x80000000L) {
3684 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3686 off = pltAddress + A - P;
3688 *(Elf64_Word *)P = (Elf64_Word)off;
3694 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3695 oc->fileName, (lnat)ELF_R_TYPE(info));
3704 ocResolve_ELF ( ObjectCode* oc )
3708 Elf_Sym* stab = NULL;
3709 char* ehdrC = (char*)(oc->image);
3710 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3711 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3713 /* first find "the" symbol table */
3714 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3716 /* also go find the string table */
3717 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3719 if (stab == NULL || strtab == NULL) {
3720 errorBelch("%s: can't find string or symbol table", oc->fileName);
3724 /* Process the relocation sections. */
3725 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3726 if (shdr[shnum].sh_type == SHT_REL) {
3727 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3728 shnum, stab, strtab );
3732 if (shdr[shnum].sh_type == SHT_RELA) {
3733 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3734 shnum, stab, strtab );
3739 #if defined(powerpc_HOST_ARCH)
3740 ocFlushInstructionCache( oc );
3747 * PowerPC & X86_64 ELF specifics
3750 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3752 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3758 ehdr = (Elf_Ehdr *) oc->image;
3759 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3761 for( i = 0; i < ehdr->e_shnum; i++ )
3762 if( shdr[i].sh_type == SHT_SYMTAB )
3765 if( i == ehdr->e_shnum )
3767 errorBelch( "This ELF file contains no symtab" );
3771 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3773 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3774 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3779 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3782 #endif /* powerpc */
3786 /* --------------------------------------------------------------------------
3788 * ------------------------------------------------------------------------*/
3790 #if defined(OBJFORMAT_MACHO)
3793 Support for MachO linking on Darwin/MacOS X
3794 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3796 I hereby formally apologize for the hackish nature of this code.
3797 Things that need to be done:
3798 *) implement ocVerifyImage_MachO
3799 *) add still more sanity checks.
3802 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3803 #define mach_header mach_header_64
3804 #define segment_command segment_command_64
3805 #define section section_64
3806 #define nlist nlist_64
3809 #ifdef powerpc_HOST_ARCH
3810 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3812 struct mach_header *header = (struct mach_header *) oc->image;
3813 struct load_command *lc = (struct load_command *) (header + 1);
3816 for( i = 0; i < header->ncmds; i++ )
3818 if( lc->cmd == LC_SYMTAB )
3820 // Find out the first and last undefined external
3821 // symbol, so we don't have to allocate too many
3823 struct symtab_command *symLC = (struct symtab_command *) lc;
3824 unsigned min = symLC->nsyms, max = 0;
3825 struct nlist *nlist =
3826 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3828 for(i=0;i<symLC->nsyms;i++)
3830 if(nlist[i].n_type & N_STAB)
3832 else if(nlist[i].n_type & N_EXT)
3834 if((nlist[i].n_type & N_TYPE) == N_UNDF
3835 && (nlist[i].n_value == 0))
3845 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3850 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3852 return ocAllocateSymbolExtras(oc,0,0);
3855 #ifdef x86_64_HOST_ARCH
3856 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3858 struct mach_header *header = (struct mach_header *) oc->image;
3859 struct load_command *lc = (struct load_command *) (header + 1);
3862 for( i = 0; i < header->ncmds; i++ )
3864 if( lc->cmd == LC_SYMTAB )
3866 // Just allocate one entry for every symbol
3867 struct symtab_command *symLC = (struct symtab_command *) lc;
3869 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3872 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3874 return ocAllocateSymbolExtras(oc,0,0);
3878 static int ocVerifyImage_MachO(ObjectCode* oc)
3880 char *image = (char*) oc->image;
3881 struct mach_header *header = (struct mach_header*) image;
3883 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3884 if(header->magic != MH_MAGIC_64)
3887 if(header->magic != MH_MAGIC)
3890 // FIXME: do some more verifying here
3894 static int resolveImports(
3897 struct symtab_command *symLC,
3898 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3899 unsigned long *indirectSyms,
3900 struct nlist *nlist)
3903 size_t itemSize = 4;
3906 int isJumpTable = 0;
3907 if(!strcmp(sect->sectname,"__jump_table"))
3911 ASSERT(sect->reserved2 == itemSize);
3915 for(i=0; i*itemSize < sect->size;i++)
3917 // according to otool, reserved1 contains the first index into the indirect symbol table
3918 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3919 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3922 if((symbol->n_type & N_TYPE) == N_UNDF
3923 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3924 addr = (void*) (symbol->n_value);
3926 addr = lookupSymbol(nm);
3929 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
3937 checkProddableBlock(oc,image + sect->offset + i*itemSize);
3938 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
3939 *(unsigned*)(image + sect->offset + i*itemSize + 1)
3940 = (char*)addr - (image + sect->offset + i*itemSize + 5);
3945 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
3946 ((void**)(image + sect->offset))[i] = addr;
3953 static unsigned long relocateAddress(
3956 struct section* sections,
3957 unsigned long address)
3960 for(i = 0; i < nSections; i++)
3962 if(sections[i].addr <= address
3963 && address < sections[i].addr + sections[i].size)
3965 return (unsigned long)oc->image
3966 + sections[i].offset + address - sections[i].addr;
3969 barf("Invalid Mach-O file:"
3970 "Address out of bounds while relocating object file");
3974 static int relocateSection(
3977 struct symtab_command *symLC, struct nlist *nlist,
3978 int nSections, struct section* sections, struct section *sect)
3980 struct relocation_info *relocs;
3983 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
3985 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
3987 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
3989 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
3993 relocs = (struct relocation_info*) (image + sect->reloff);
3997 #ifdef x86_64_HOST_ARCH
3998 struct relocation_info *reloc = &relocs[i];
4000 char *thingPtr = image + sect->offset + reloc->r_address;
4002 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4003 complains that it may be used uninitialized if we don't */
4006 int type = reloc->r_type;
4008 checkProddableBlock(oc,thingPtr);
4009 switch(reloc->r_length)
4012 thing = *(uint8_t*)thingPtr;
4013 baseValue = (uint64_t)thingPtr + 1;
4016 thing = *(uint16_t*)thingPtr;
4017 baseValue = (uint64_t)thingPtr + 2;
4020 thing = *(uint32_t*)thingPtr;
4021 baseValue = (uint64_t)thingPtr + 4;
4024 thing = *(uint64_t*)thingPtr;
4025 baseValue = (uint64_t)thingPtr + 8;
4028 barf("Unknown size.");
4031 if(type == X86_64_RELOC_GOT
4032 || type == X86_64_RELOC_GOT_LOAD)
4034 ASSERT(reloc->r_extern);
4035 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4037 type = X86_64_RELOC_SIGNED;
4039 else if(reloc->r_extern)
4041 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4042 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4043 if(symbol->n_value == 0)
4044 value = (uint64_t) lookupSymbol(nm);
4046 value = relocateAddress(oc, nSections, sections,
4051 value = sections[reloc->r_symbolnum-1].offset
4052 - sections[reloc->r_symbolnum-1].addr
4056 if(type == X86_64_RELOC_BRANCH)
4058 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4060 ASSERT(reloc->r_extern);
4061 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4064 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4065 type = X86_64_RELOC_SIGNED;
4070 case X86_64_RELOC_UNSIGNED:
4071 ASSERT(!reloc->r_pcrel);
4074 case X86_64_RELOC_SIGNED:
4075 ASSERT(reloc->r_pcrel);
4076 thing += value - baseValue;
4078 case X86_64_RELOC_SUBTRACTOR:
4079 ASSERT(!reloc->r_pcrel);
4083 barf("unkown relocation");
4086 switch(reloc->r_length)
4089 *(uint8_t*)thingPtr = thing;
4092 *(uint16_t*)thingPtr = thing;
4095 *(uint32_t*)thingPtr = thing;
4098 *(uint64_t*)thingPtr = thing;
4102 if(relocs[i].r_address & R_SCATTERED)
4104 struct scattered_relocation_info *scat =
4105 (struct scattered_relocation_info*) &relocs[i];
4109 if(scat->r_length == 2)
4111 unsigned long word = 0;
4112 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4113 checkProddableBlock(oc,wordPtr);
4115 // Note on relocation types:
4116 // i386 uses the GENERIC_RELOC_* types,
4117 // while ppc uses special PPC_RELOC_* types.
4118 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4119 // in both cases, all others are different.
4120 // Therefore, we use GENERIC_RELOC_VANILLA
4121 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4122 // and use #ifdefs for the other types.
4124 // Step 1: Figure out what the relocated value should be
4125 if(scat->r_type == GENERIC_RELOC_VANILLA)
4127 word = *wordPtr + (unsigned long) relocateAddress(
4134 #ifdef powerpc_HOST_ARCH
4135 else if(scat->r_type == PPC_RELOC_SECTDIFF
4136 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4137 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4138 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4140 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4141 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4144 struct scattered_relocation_info *pair =
4145 (struct scattered_relocation_info*) &relocs[i+1];
4147 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4148 barf("Invalid Mach-O file: "
4149 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4151 word = (unsigned long)
4152 (relocateAddress(oc, nSections, sections, scat->r_value)
4153 - relocateAddress(oc, nSections, sections, pair->r_value));
4156 #ifdef powerpc_HOST_ARCH
4157 else if(scat->r_type == PPC_RELOC_HI16
4158 || scat->r_type == PPC_RELOC_LO16
4159 || scat->r_type == PPC_RELOC_HA16
4160 || scat->r_type == PPC_RELOC_LO14)
4161 { // these are generated by label+offset things
4162 struct relocation_info *pair = &relocs[i+1];
4163 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4164 barf("Invalid Mach-O file: "
4165 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4167 if(scat->r_type == PPC_RELOC_LO16)
4169 word = ((unsigned short*) wordPtr)[1];
4170 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4172 else if(scat->r_type == PPC_RELOC_LO14)
4174 barf("Unsupported Relocation: PPC_RELOC_LO14");
4175 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4176 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4178 else if(scat->r_type == PPC_RELOC_HI16)
4180 word = ((unsigned short*) wordPtr)[1] << 16;
4181 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4183 else if(scat->r_type == PPC_RELOC_HA16)
4185 word = ((unsigned short*) wordPtr)[1] << 16;
4186 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4190 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4198 barf ("Don't know how to handle this Mach-O "
4199 "scattered relocation entry: "
4200 "object file %s; entry type %ld; "
4202 oc->fileName, scat->r_type, scat->r_address);
4206 #ifdef powerpc_HOST_ARCH
4207 if(scat->r_type == GENERIC_RELOC_VANILLA
4208 || scat->r_type == PPC_RELOC_SECTDIFF)
4210 if(scat->r_type == GENERIC_RELOC_VANILLA
4211 || scat->r_type == GENERIC_RELOC_SECTDIFF
4212 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4217 #ifdef powerpc_HOST_ARCH
4218 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4220 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4222 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4224 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4226 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4228 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4229 + ((word & (1<<15)) ? 1 : 0);
4235 barf("Can't handle Mach-O scattered relocation entry "
4236 "with this r_length tag: "
4237 "object file %s; entry type %ld; "
4238 "r_length tag %ld; address %#lx\n",
4239 oc->fileName, scat->r_type, scat->r_length,
4244 else /* scat->r_pcrel */
4246 barf("Don't know how to handle *PC-relative* Mach-O "
4247 "scattered relocation entry: "
4248 "object file %s; entry type %ld; address %#lx\n",
4249 oc->fileName, scat->r_type, scat->r_address);
4254 else /* !(relocs[i].r_address & R_SCATTERED) */
4256 struct relocation_info *reloc = &relocs[i];
4257 if(reloc->r_pcrel && !reloc->r_extern)
4260 if(reloc->r_length == 2)
4262 unsigned long word = 0;
4263 #ifdef powerpc_HOST_ARCH
4264 unsigned long jumpIsland = 0;
4265 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4266 // to avoid warning and to catch
4270 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4271 checkProddableBlock(oc,wordPtr);
4273 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4277 #ifdef powerpc_HOST_ARCH
4278 else if(reloc->r_type == PPC_RELOC_LO16)
4280 word = ((unsigned short*) wordPtr)[1];
4281 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4283 else if(reloc->r_type == PPC_RELOC_HI16)
4285 word = ((unsigned short*) wordPtr)[1] << 16;
4286 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4288 else if(reloc->r_type == PPC_RELOC_HA16)
4290 word = ((unsigned short*) wordPtr)[1] << 16;
4291 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4293 else if(reloc->r_type == PPC_RELOC_BR24)
4296 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4301 barf("Can't handle this Mach-O relocation entry "
4303 "object file %s; entry type %ld; address %#lx\n",
4304 oc->fileName, reloc->r_type, reloc->r_address);
4308 if(!reloc->r_extern)
4311 sections[reloc->r_symbolnum-1].offset
4312 - sections[reloc->r_symbolnum-1].addr
4319 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4320 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4321 void *symbolAddress = lookupSymbol(nm);
4324 errorBelch("\nunknown symbol `%s'", nm);
4330 #ifdef powerpc_HOST_ARCH
4331 // In the .o file, this should be a relative jump to NULL
4332 // and we'll change it to a relative jump to the symbol
4333 ASSERT(word + reloc->r_address == 0);
4334 jumpIsland = (unsigned long)
4335 &makeSymbolExtra(oc,
4337 (unsigned long) symbolAddress)
4341 offsetToJumpIsland = word + jumpIsland
4342 - (((long)image) + sect->offset - sect->addr);
4345 word += (unsigned long) symbolAddress
4346 - (((long)image) + sect->offset - sect->addr);
4350 word += (unsigned long) symbolAddress;
4354 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4359 #ifdef powerpc_HOST_ARCH
4360 else if(reloc->r_type == PPC_RELOC_LO16)
4362 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4365 else if(reloc->r_type == PPC_RELOC_HI16)
4367 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4370 else if(reloc->r_type == PPC_RELOC_HA16)
4372 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4373 + ((word & (1<<15)) ? 1 : 0);
4376 else if(reloc->r_type == PPC_RELOC_BR24)
4378 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4380 // The branch offset is too large.
4381 // Therefore, we try to use a jump island.
4384 barf("unconditional relative branch out of range: "
4385 "no jump island available");
4388 word = offsetToJumpIsland;
4389 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4390 barf("unconditional relative branch out of range: "
4391 "jump island out of range");
4393 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4400 barf("Can't handle Mach-O relocation entry (not scattered) "
4401 "with this r_length tag: "
4402 "object file %s; entry type %ld; "
4403 "r_length tag %ld; address %#lx\n",
4404 oc->fileName, reloc->r_type, reloc->r_length,
4414 static int ocGetNames_MachO(ObjectCode* oc)
4416 char *image = (char*) oc->image;
4417 struct mach_header *header = (struct mach_header*) image;
4418 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4419 unsigned i,curSymbol = 0;
4420 struct segment_command *segLC = NULL;
4421 struct section *sections;
4422 struct symtab_command *symLC = NULL;
4423 struct nlist *nlist;
4424 unsigned long commonSize = 0;
4425 char *commonStorage = NULL;
4426 unsigned long commonCounter;
4428 for(i=0;i<header->ncmds;i++)
4430 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4431 segLC = (struct segment_command*) lc;
4432 else if(lc->cmd == LC_SYMTAB)
4433 symLC = (struct symtab_command*) lc;
4434 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4437 sections = (struct section*) (segLC+1);
4438 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4442 barf("ocGetNames_MachO: no segment load command");
4444 for(i=0;i<segLC->nsects;i++)
4446 if(sections[i].size == 0)
4449 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4451 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4452 "ocGetNames_MachO(common symbols)");
4453 sections[i].offset = zeroFillArea - image;
4456 if(!strcmp(sections[i].sectname,"__text"))
4457 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4458 (void*) (image + sections[i].offset),
4459 (void*) (image + sections[i].offset + sections[i].size));
4460 else if(!strcmp(sections[i].sectname,"__const"))
4461 addSection(oc, SECTIONKIND_RWDATA,
4462 (void*) (image + sections[i].offset),
4463 (void*) (image + sections[i].offset + sections[i].size));
4464 else if(!strcmp(sections[i].sectname,"__data"))
4465 addSection(oc, SECTIONKIND_RWDATA,
4466 (void*) (image + sections[i].offset),
4467 (void*) (image + sections[i].offset + sections[i].size));
4468 else if(!strcmp(sections[i].sectname,"__bss")
4469 || !strcmp(sections[i].sectname,"__common"))
4470 addSection(oc, SECTIONKIND_RWDATA,
4471 (void*) (image + sections[i].offset),
4472 (void*) (image + sections[i].offset + sections[i].size));
4474 addProddableBlock(oc, (void*) (image + sections[i].offset),
4478 // count external symbols defined here
4482 for(i=0;i<symLC->nsyms;i++)
4484 if(nlist[i].n_type & N_STAB)
4486 else if(nlist[i].n_type & N_EXT)
4488 if((nlist[i].n_type & N_TYPE) == N_UNDF
4489 && (nlist[i].n_value != 0))
4491 commonSize += nlist[i].n_value;
4494 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4499 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4500 "ocGetNames_MachO(oc->symbols)");
4504 for(i=0;i<symLC->nsyms;i++)
4506 if(nlist[i].n_type & N_STAB)
4508 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4510 if(nlist[i].n_type & N_EXT)
4512 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4513 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4514 ; // weak definition, and we already have a definition
4517 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4519 + sections[nlist[i].n_sect-1].offset
4520 - sections[nlist[i].n_sect-1].addr
4521 + nlist[i].n_value);
4522 oc->symbols[curSymbol++] = nm;
4529 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4530 commonCounter = (unsigned long)commonStorage;
4533 for(i=0;i<symLC->nsyms;i++)
4535 if((nlist[i].n_type & N_TYPE) == N_UNDF
4536 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4538 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4539 unsigned long sz = nlist[i].n_value;
4541 nlist[i].n_value = commonCounter;
4543 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4544 (void*)commonCounter);
4545 oc->symbols[curSymbol++] = nm;
4547 commonCounter += sz;
4554 static int ocResolve_MachO(ObjectCode* oc)
4556 char *image = (char*) oc->image;
4557 struct mach_header *header = (struct mach_header*) image;
4558 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4560 struct segment_command *segLC = NULL;
4561 struct section *sections;
4562 struct symtab_command *symLC = NULL;
4563 struct dysymtab_command *dsymLC = NULL;
4564 struct nlist *nlist;
4566 for(i=0;i<header->ncmds;i++)
4568 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4569 segLC = (struct segment_command*) lc;
4570 else if(lc->cmd == LC_SYMTAB)
4571 symLC = (struct symtab_command*) lc;
4572 else if(lc->cmd == LC_DYSYMTAB)
4573 dsymLC = (struct dysymtab_command*) lc;
4574 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4577 sections = (struct section*) (segLC+1);
4578 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4583 unsigned long *indirectSyms
4584 = (unsigned long*) (image + dsymLC->indirectsymoff);
4586 for(i=0;i<segLC->nsects;i++)
4588 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4589 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4590 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4592 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4595 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4596 || !strcmp(sections[i].sectname,"__pointers"))
4598 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4601 else if(!strcmp(sections[i].sectname,"__jump_table"))
4603 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4609 for(i=0;i<segLC->nsects;i++)
4611 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4615 #if defined (powerpc_HOST_ARCH)
4616 ocFlushInstructionCache( oc );
4622 #ifdef powerpc_HOST_ARCH
4624 * The Mach-O object format uses leading underscores. But not everywhere.
4625 * There is a small number of runtime support functions defined in
4626 * libcc_dynamic.a whose name does not have a leading underscore.
4627 * As a consequence, we can't get their address from C code.
4628 * We have to use inline assembler just to take the address of a function.
4632 static void machoInitSymbolsWithoutUnderscore()
4634 extern void* symbolsWithoutUnderscore[];
4635 void **p = symbolsWithoutUnderscore;
4636 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4638 #undef SymI_NeedsProto
4639 #define SymI_NeedsProto(x) \
4640 __asm__ volatile(".long " # x);
4642 RTS_MACHO_NOUNDERLINE_SYMBOLS
4644 __asm__ volatile(".text");
4646 #undef SymI_NeedsProto
4647 #define SymI_NeedsProto(x) \
4648 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4650 RTS_MACHO_NOUNDERLINE_SYMBOLS
4652 #undef SymI_NeedsProto
4657 * Figure out by how much to shift the entire Mach-O file in memory
4658 * when loading so that its single segment ends up 16-byte-aligned
4660 static int machoGetMisalignment( FILE * f )
4662 struct mach_header header;
4665 fread(&header, sizeof(header), 1, f);
4668 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4669 if(header.magic != MH_MAGIC_64)
4672 if(header.magic != MH_MAGIC)
4676 misalignment = (header.sizeofcmds + sizeof(header))
4679 return misalignment ? (16 - misalignment) : 0;