1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
25 #include "LinkerInternals.h"
29 #include "RtsGlobals.h"
33 #ifdef HAVE_SYS_TYPES_H
34 #include <sys/types.h>
40 #ifdef HAVE_SYS_STAT_H
44 #if defined(HAVE_DLFCN_H)
48 #if defined(cygwin32_HOST_OS)
53 #ifdef HAVE_SYS_TIME_H
57 #include <sys/fcntl.h>
58 #include <sys/termios.h>
59 #include <sys/utime.h>
60 #include <sys/utsname.h>
64 #if defined(ia64_HOST_ARCH) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
69 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
77 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
78 # define OBJFORMAT_ELF
79 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
80 # define OBJFORMAT_PEi386
83 #elif defined(darwin_HOST_OS)
84 # define OBJFORMAT_MACHO
85 # include <mach-o/loader.h>
86 # include <mach-o/nlist.h>
87 # include <mach-o/reloc.h>
88 #if !defined(HAVE_DLFCN_H)
89 # include <mach-o/dyld.h>
91 #if defined(powerpc_HOST_ARCH)
92 # include <mach-o/ppc/reloc.h>
94 #if defined(x86_64_HOST_ARCH)
95 # include <mach-o/x86_64/reloc.h>
99 /* Hash table mapping symbol names to Symbol */
100 static /*Str*/HashTable *symhash;
102 /* Hash table mapping symbol names to StgStablePtr */
103 static /*Str*/HashTable *stablehash;
105 /* List of currently loaded objects */
106 ObjectCode *objects = NULL; /* initially empty */
108 #if defined(OBJFORMAT_ELF)
109 static int ocVerifyImage_ELF ( ObjectCode* oc );
110 static int ocGetNames_ELF ( ObjectCode* oc );
111 static int ocResolve_ELF ( ObjectCode* oc );
112 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
113 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
115 #elif defined(OBJFORMAT_PEi386)
116 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
117 static int ocGetNames_PEi386 ( ObjectCode* oc );
118 static int ocResolve_PEi386 ( ObjectCode* oc );
119 static void *lookupSymbolInDLLs ( unsigned char *lbl );
120 static void zapTrailingAtSign ( unsigned char *sym );
121 #elif defined(OBJFORMAT_MACHO)
122 static int ocVerifyImage_MachO ( ObjectCode* oc );
123 static int ocGetNames_MachO ( ObjectCode* oc );
124 static int ocResolve_MachO ( ObjectCode* oc );
126 static int machoGetMisalignment( FILE * );
127 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
128 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
130 #ifdef powerpc_HOST_ARCH
131 static void machoInitSymbolsWithoutUnderscore( void );
135 /* on x86_64 we have a problem with relocating symbol references in
136 * code that was compiled without -fPIC. By default, the small memory
137 * model is used, which assumes that symbol references can fit in a
138 * 32-bit slot. The system dynamic linker makes this work for
139 * references to shared libraries by either (a) allocating a jump
140 * table slot for code references, or (b) moving the symbol at load
141 * time (and copying its contents, if necessary) for data references.
143 * We unfortunately can't tell whether symbol references are to code
144 * or data. So for now we assume they are code (the vast majority
145 * are), and allocate jump-table slots. Unfortunately this will
146 * SILENTLY generate crashing code for data references. This hack is
147 * enabled by X86_64_ELF_NONPIC_HACK.
149 * One workaround is to use shared Haskell libraries. This is
150 * coming. Another workaround is to keep the static libraries but
151 * compile them with -fPIC, because that will generate PIC references
152 * to data which can be relocated. The PIC code is still too green to
153 * do this systematically, though.
156 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
158 * Naming Scheme for Symbol Macros
160 * SymI_*: symbol is internal to the RTS. It resides in an object
161 * file/library that is statically.
162 * SymE_*: symbol is external to the RTS library. It might be linked
165 * Sym*_HasProto : the symbol prototype is imported in an include file
166 * or defined explicitly
167 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
168 * default proto extern void sym(void);
170 #define X86_64_ELF_NONPIC_HACK 1
172 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
173 * small memory model on this architecture (see gcc docs,
176 * MAP_32BIT not available on OpenBSD/amd64
178 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
179 #define TRY_MAP_32BIT MAP_32BIT
181 #define TRY_MAP_32BIT 0
185 * Due to the small memory model (see above), on x86_64 we have to map
186 * all our non-PIC object files into the low 2Gb of the address space
187 * (why 2Gb and not 4Gb? Because all addresses must be reachable
188 * using a 32-bit signed PC-relative offset). On Linux we can do this
189 * using the MAP_32BIT flag to mmap(), however on other OSs
190 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
191 * can't do this. So on these systems, we have to pick a base address
192 * in the low 2Gb of the address space and try to allocate memory from
195 * We pick a default address based on the OS, but also make this
196 * configurable via an RTS flag (+RTS -xm)
198 #if defined(x86_64_HOST_ARCH)
200 #if defined(MAP_32BIT)
201 // Try to use MAP_32BIT
202 #define MMAP_32BIT_BASE_DEFAULT 0
205 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
208 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
211 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
212 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
213 #define MAP_ANONYMOUS MAP_ANON
216 /* -----------------------------------------------------------------------------
217 * Built-in symbols from the RTS
220 typedef struct _RtsSymbolVal {
225 #define Maybe_Stable_Names SymI_HasProto(mkWeakzh_fast) \
226 SymI_HasProto(mkWeakForeignEnvzh_fast) \
227 SymI_HasProto(makeStableNamezh_fast) \
228 SymI_HasProto(finalizzeWeakzh_fast)
230 #if !defined (mingw32_HOST_OS)
231 #define RTS_POSIX_ONLY_SYMBOLS \
232 SymI_HasProto(shutdownHaskellAndSignal) \
233 SymI_NeedsProto(lockFile) \
234 SymI_NeedsProto(unlockFile) \
235 SymI_HasProto(signal_handlers) \
236 SymI_HasProto(stg_sig_install) \
237 SymI_NeedsProto(nocldstop)
240 #if defined (cygwin32_HOST_OS)
241 #define RTS_MINGW_ONLY_SYMBOLS /**/
242 /* Don't have the ability to read import libs / archives, so
243 * we have to stupidly list a lot of what libcygwin.a
246 #define RTS_CYGWIN_ONLY_SYMBOLS \
247 SymI_HasProto(regfree) \
248 SymI_HasProto(regexec) \
249 SymI_HasProto(regerror) \
250 SymI_HasProto(regcomp) \
251 SymI_HasProto(__errno) \
252 SymI_HasProto(access) \
253 SymI_HasProto(chmod) \
254 SymI_HasProto(chdir) \
255 SymI_HasProto(close) \
256 SymI_HasProto(creat) \
258 SymI_HasProto(dup2) \
259 SymI_HasProto(fstat) \
260 SymI_HasProto(fcntl) \
261 SymI_HasProto(getcwd) \
262 SymI_HasProto(getenv) \
263 SymI_HasProto(lseek) \
264 SymI_HasProto(open) \
265 SymI_HasProto(fpathconf) \
266 SymI_HasProto(pathconf) \
267 SymI_HasProto(stat) \
269 SymI_HasProto(tanh) \
270 SymI_HasProto(cosh) \
271 SymI_HasProto(sinh) \
272 SymI_HasProto(atan) \
273 SymI_HasProto(acos) \
274 SymI_HasProto(asin) \
280 SymI_HasProto(sqrt) \
281 SymI_HasProto(localtime_r) \
282 SymI_HasProto(gmtime_r) \
283 SymI_HasProto(mktime) \
284 SymI_NeedsProto(_imp___tzname) \
285 SymI_HasProto(gettimeofday) \
286 SymI_HasProto(timezone) \
287 SymI_HasProto(tcgetattr) \
288 SymI_HasProto(tcsetattr) \
289 SymI_HasProto(memcpy) \
290 SymI_HasProto(memmove) \
291 SymI_HasProto(realloc) \
292 SymI_HasProto(malloc) \
293 SymI_HasProto(free) \
294 SymI_HasProto(fork) \
295 SymI_HasProto(lstat) \
296 SymI_HasProto(isatty) \
297 SymI_HasProto(mkdir) \
298 SymI_HasProto(opendir) \
299 SymI_HasProto(readdir) \
300 SymI_HasProto(rewinddir) \
301 SymI_HasProto(closedir) \
302 SymI_HasProto(link) \
303 SymI_HasProto(mkfifo) \
304 SymI_HasProto(pipe) \
305 SymI_HasProto(read) \
306 SymI_HasProto(rename) \
307 SymI_HasProto(rmdir) \
308 SymI_HasProto(select) \
309 SymI_HasProto(system) \
310 SymI_HasProto(write) \
311 SymI_HasProto(strcmp) \
312 SymI_HasProto(strcpy) \
313 SymI_HasProto(strncpy) \
314 SymI_HasProto(strerror) \
315 SymI_HasProto(sigaddset) \
316 SymI_HasProto(sigemptyset) \
317 SymI_HasProto(sigprocmask) \
318 SymI_HasProto(umask) \
319 SymI_HasProto(uname) \
320 SymI_HasProto(unlink) \
321 SymI_HasProto(utime) \
322 SymI_HasProto(waitpid)
324 #elif !defined(mingw32_HOST_OS)
325 #define RTS_MINGW_ONLY_SYMBOLS /**/
326 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
327 #else /* defined(mingw32_HOST_OS) */
328 #define RTS_POSIX_ONLY_SYMBOLS /**/
329 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
331 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
333 #define RTS_MINGW_EXTRA_SYMS \
334 SymI_NeedsProto(_imp____mb_cur_max) \
335 SymI_NeedsProto(_imp___pctype)
337 #define RTS_MINGW_EXTRA_SYMS
340 #if HAVE_GETTIMEOFDAY
341 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
343 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
346 /* These are statically linked from the mingw libraries into the ghc
347 executable, so we have to employ this hack. */
348 #define RTS_MINGW_ONLY_SYMBOLS \
349 SymI_HasProto(asyncReadzh_fast) \
350 SymI_HasProto(asyncWritezh_fast) \
351 SymI_HasProto(asyncDoProczh_fast) \
352 SymI_HasProto(memset) \
353 SymI_HasProto(inet_ntoa) \
354 SymI_HasProto(inet_addr) \
355 SymI_HasProto(htonl) \
356 SymI_HasProto(recvfrom) \
357 SymI_HasProto(listen) \
358 SymI_HasProto(bind) \
359 SymI_HasProto(shutdown) \
360 SymI_HasProto(connect) \
361 SymI_HasProto(htons) \
362 SymI_HasProto(ntohs) \
363 SymI_HasProto(getservbyname) \
364 SymI_HasProto(getservbyport) \
365 SymI_HasProto(getprotobynumber) \
366 SymI_HasProto(getprotobyname) \
367 SymI_HasProto(gethostbyname) \
368 SymI_HasProto(gethostbyaddr) \
369 SymI_HasProto(gethostname) \
370 SymI_HasProto(strcpy) \
371 SymI_HasProto(strncpy) \
372 SymI_HasProto(abort) \
373 SymI_NeedsProto(_alloca) \
374 SymI_NeedsProto(isxdigit) \
375 SymI_NeedsProto(isupper) \
376 SymI_NeedsProto(ispunct) \
377 SymI_NeedsProto(islower) \
378 SymI_NeedsProto(isspace) \
379 SymI_NeedsProto(isprint) \
380 SymI_NeedsProto(isdigit) \
381 SymI_NeedsProto(iscntrl) \
382 SymI_NeedsProto(isalpha) \
383 SymI_NeedsProto(isalnum) \
384 SymI_NeedsProto(isascii) \
385 SymI_HasProto(__mingw_vfprintf) \
386 SymI_HasProto(strcmp) \
387 SymI_HasProto(memmove) \
388 SymI_HasProto(realloc) \
389 SymI_HasProto(malloc) \
391 SymI_HasProto(tanh) \
392 SymI_HasProto(cosh) \
393 SymI_HasProto(sinh) \
394 SymI_HasProto(atan) \
395 SymI_HasProto(acos) \
396 SymI_HasProto(asin) \
402 SymI_HasProto(sqrt) \
403 SymI_HasProto(powf) \
404 SymI_HasProto(tanhf) \
405 SymI_HasProto(coshf) \
406 SymI_HasProto(sinhf) \
407 SymI_HasProto(atanf) \
408 SymI_HasProto(acosf) \
409 SymI_HasProto(asinf) \
410 SymI_HasProto(tanf) \
411 SymI_HasProto(cosf) \
412 SymI_HasProto(sinf) \
413 SymI_HasProto(expf) \
414 SymI_HasProto(logf) \
415 SymI_HasProto(sqrtf) \
416 SymI_HasProto(memcpy) \
417 SymI_HasProto(rts_InstallConsoleEvent) \
418 SymI_HasProto(rts_ConsoleHandlerDone) \
419 SymI_NeedsProto(mktime) \
420 SymI_NeedsProto(_imp___timezone) \
421 SymI_NeedsProto(_imp___tzname) \
422 SymI_NeedsProto(_imp__tzname) \
423 SymI_NeedsProto(_imp___iob) \
424 SymI_NeedsProto(_imp___osver) \
425 SymI_NeedsProto(localtime) \
426 SymI_NeedsProto(gmtime) \
427 SymI_NeedsProto(opendir) \
428 SymI_NeedsProto(readdir) \
429 SymI_NeedsProto(rewinddir) \
430 RTS_MINGW_EXTRA_SYMS \
431 RTS_MINGW_GETTIMEOFDAY_SYM \
432 SymI_NeedsProto(closedir)
435 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
436 #define RTS_DARWIN_ONLY_SYMBOLS \
437 SymI_NeedsProto(asprintf$LDBLStub) \
438 SymI_NeedsProto(err$LDBLStub) \
439 SymI_NeedsProto(errc$LDBLStub) \
440 SymI_NeedsProto(errx$LDBLStub) \
441 SymI_NeedsProto(fprintf$LDBLStub) \
442 SymI_NeedsProto(fscanf$LDBLStub) \
443 SymI_NeedsProto(fwprintf$LDBLStub) \
444 SymI_NeedsProto(fwscanf$LDBLStub) \
445 SymI_NeedsProto(printf$LDBLStub) \
446 SymI_NeedsProto(scanf$LDBLStub) \
447 SymI_NeedsProto(snprintf$LDBLStub) \
448 SymI_NeedsProto(sprintf$LDBLStub) \
449 SymI_NeedsProto(sscanf$LDBLStub) \
450 SymI_NeedsProto(strtold$LDBLStub) \
451 SymI_NeedsProto(swprintf$LDBLStub) \
452 SymI_NeedsProto(swscanf$LDBLStub) \
453 SymI_NeedsProto(syslog$LDBLStub) \
454 SymI_NeedsProto(vasprintf$LDBLStub) \
455 SymI_NeedsProto(verr$LDBLStub) \
456 SymI_NeedsProto(verrc$LDBLStub) \
457 SymI_NeedsProto(verrx$LDBLStub) \
458 SymI_NeedsProto(vfprintf$LDBLStub) \
459 SymI_NeedsProto(vfscanf$LDBLStub) \
460 SymI_NeedsProto(vfwprintf$LDBLStub) \
461 SymI_NeedsProto(vfwscanf$LDBLStub) \
462 SymI_NeedsProto(vprintf$LDBLStub) \
463 SymI_NeedsProto(vscanf$LDBLStub) \
464 SymI_NeedsProto(vsnprintf$LDBLStub) \
465 SymI_NeedsProto(vsprintf$LDBLStub) \
466 SymI_NeedsProto(vsscanf$LDBLStub) \
467 SymI_NeedsProto(vswprintf$LDBLStub) \
468 SymI_NeedsProto(vswscanf$LDBLStub) \
469 SymI_NeedsProto(vsyslog$LDBLStub) \
470 SymI_NeedsProto(vwarn$LDBLStub) \
471 SymI_NeedsProto(vwarnc$LDBLStub) \
472 SymI_NeedsProto(vwarnx$LDBLStub) \
473 SymI_NeedsProto(vwprintf$LDBLStub) \
474 SymI_NeedsProto(vwscanf$LDBLStub) \
475 SymI_NeedsProto(warn$LDBLStub) \
476 SymI_NeedsProto(warnc$LDBLStub) \
477 SymI_NeedsProto(warnx$LDBLStub) \
478 SymI_NeedsProto(wcstold$LDBLStub) \
479 SymI_NeedsProto(wprintf$LDBLStub) \
480 SymI_NeedsProto(wscanf$LDBLStub)
482 #define RTS_DARWIN_ONLY_SYMBOLS
486 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
488 # define MAIN_CAP_SYM
491 #if !defined(mingw32_HOST_OS)
492 #define RTS_USER_SIGNALS_SYMBOLS \
493 SymI_HasProto(setIOManagerPipe) \
494 SymI_NeedsProto(blockUserSignals) \
495 SymI_NeedsProto(unblockUserSignals)
497 #define RTS_USER_SIGNALS_SYMBOLS \
498 SymI_HasProto(sendIOManagerEvent) \
499 SymI_HasProto(readIOManagerEvent) \
500 SymI_HasProto(getIOManagerEvent) \
501 SymI_HasProto(console_handler)
504 #define RTS_LIBFFI_SYMBOLS \
505 SymE_NeedsProto(ffi_prep_cif) \
506 SymE_NeedsProto(ffi_call) \
507 SymE_NeedsProto(ffi_type_void) \
508 SymE_NeedsProto(ffi_type_float) \
509 SymE_NeedsProto(ffi_type_double) \
510 SymE_NeedsProto(ffi_type_sint64) \
511 SymE_NeedsProto(ffi_type_uint64) \
512 SymE_NeedsProto(ffi_type_sint32) \
513 SymE_NeedsProto(ffi_type_uint32) \
514 SymE_NeedsProto(ffi_type_sint16) \
515 SymE_NeedsProto(ffi_type_uint16) \
516 SymE_NeedsProto(ffi_type_sint8) \
517 SymE_NeedsProto(ffi_type_uint8) \
518 SymE_NeedsProto(ffi_type_pointer)
520 #ifdef TABLES_NEXT_TO_CODE
521 #define RTS_RET_SYMBOLS /* nothing */
523 #define RTS_RET_SYMBOLS \
524 SymI_HasProto(stg_enter_ret) \
525 SymI_HasProto(stg_gc_fun_ret) \
526 SymI_HasProto(stg_ap_v_ret) \
527 SymI_HasProto(stg_ap_f_ret) \
528 SymI_HasProto(stg_ap_d_ret) \
529 SymI_HasProto(stg_ap_l_ret) \
530 SymI_HasProto(stg_ap_n_ret) \
531 SymI_HasProto(stg_ap_p_ret) \
532 SymI_HasProto(stg_ap_pv_ret) \
533 SymI_HasProto(stg_ap_pp_ret) \
534 SymI_HasProto(stg_ap_ppv_ret) \
535 SymI_HasProto(stg_ap_ppp_ret) \
536 SymI_HasProto(stg_ap_pppv_ret) \
537 SymI_HasProto(stg_ap_pppp_ret) \
538 SymI_HasProto(stg_ap_ppppp_ret) \
539 SymI_HasProto(stg_ap_pppppp_ret)
542 #define RTS_SYMBOLS \
544 SymI_HasProto(StgReturn) \
545 SymI_HasProto(stg_enter_info) \
546 SymI_HasProto(stg_gc_void_info) \
547 SymI_HasProto(__stg_gc_enter_1) \
548 SymI_HasProto(stg_gc_noregs) \
549 SymI_HasProto(stg_gc_unpt_r1_info) \
550 SymI_HasProto(stg_gc_unpt_r1) \
551 SymI_HasProto(stg_gc_unbx_r1_info) \
552 SymI_HasProto(stg_gc_unbx_r1) \
553 SymI_HasProto(stg_gc_f1_info) \
554 SymI_HasProto(stg_gc_f1) \
555 SymI_HasProto(stg_gc_d1_info) \
556 SymI_HasProto(stg_gc_d1) \
557 SymI_HasProto(stg_gc_l1_info) \
558 SymI_HasProto(stg_gc_l1) \
559 SymI_HasProto(__stg_gc_fun) \
560 SymI_HasProto(stg_gc_fun_info) \
561 SymI_HasProto(stg_gc_gen) \
562 SymI_HasProto(stg_gc_gen_info) \
563 SymI_HasProto(stg_gc_gen_hp) \
564 SymI_HasProto(stg_gc_ut) \
565 SymI_HasProto(stg_gen_yield) \
566 SymI_HasProto(stg_yield_noregs) \
567 SymI_HasProto(stg_yield_to_interpreter) \
568 SymI_HasProto(stg_gen_block) \
569 SymI_HasProto(stg_block_noregs) \
570 SymI_HasProto(stg_block_1) \
571 SymI_HasProto(stg_block_takemvar) \
572 SymI_HasProto(stg_block_putmvar) \
574 SymI_HasProto(MallocFailHook) \
575 SymI_HasProto(OnExitHook) \
576 SymI_HasProto(OutOfHeapHook) \
577 SymI_HasProto(StackOverflowHook) \
578 SymI_HasProto(addDLL) \
579 SymI_HasProto(__int_encodeDouble) \
580 SymI_HasProto(__word_encodeDouble) \
581 SymI_HasProto(__2Int_encodeDouble) \
582 SymI_HasProto(__int_encodeFloat) \
583 SymI_HasProto(__word_encodeFloat) \
584 SymI_HasProto(atomicallyzh_fast) \
585 SymI_HasProto(barf) \
586 SymI_HasProto(debugBelch) \
587 SymI_HasProto(errorBelch) \
588 SymI_HasProto(sysErrorBelch) \
589 SymI_HasProto(asyncExceptionsBlockedzh_fast) \
590 SymI_HasProto(blockAsyncExceptionszh_fast) \
591 SymI_HasProto(catchzh_fast) \
592 SymI_HasProto(catchRetryzh_fast) \
593 SymI_HasProto(catchSTMzh_fast) \
594 SymI_HasProto(checkzh_fast) \
595 SymI_HasProto(closure_flags) \
596 SymI_HasProto(cmp_thread) \
597 SymI_HasProto(createAdjustor) \
598 SymI_HasProto(decodeDoublezu2Intzh_fast) \
599 SymI_HasProto(decodeFloatzuIntzh_fast) \
600 SymI_HasProto(defaultsHook) \
601 SymI_HasProto(delayzh_fast) \
602 SymI_HasProto(deRefWeakzh_fast) \
603 SymI_HasProto(deRefStablePtrzh_fast) \
604 SymI_HasProto(dirty_MUT_VAR) \
605 SymI_HasProto(forkzh_fast) \
606 SymI_HasProto(forkOnzh_fast) \
607 SymI_HasProto(forkProcess) \
608 SymI_HasProto(forkOS_createThread) \
609 SymI_HasProto(freeHaskellFunctionPtr) \
610 SymI_HasProto(freeStablePtr) \
611 SymI_HasProto(getOrSetTypeableStore) \
612 SymI_HasProto(getOrSetSignalHandlerStore) \
613 SymI_HasProto(genSymZh) \
614 SymI_HasProto(genericRaise) \
615 SymI_HasProto(getProgArgv) \
616 SymI_HasProto(getFullProgArgv) \
617 SymI_HasProto(getStablePtr) \
618 SymI_HasProto(hs_init) \
619 SymI_HasProto(hs_exit) \
620 SymI_HasProto(hs_set_argv) \
621 SymI_HasProto(hs_add_root) \
622 SymI_HasProto(hs_perform_gc) \
623 SymI_HasProto(hs_free_stable_ptr) \
624 SymI_HasProto(hs_free_fun_ptr) \
625 SymI_HasProto(hs_hpc_rootModule) \
626 SymI_HasProto(hs_hpc_module) \
627 SymI_HasProto(initLinker) \
628 SymI_HasProto(unpackClosurezh_fast) \
629 SymI_HasProto(getApStackValzh_fast) \
630 SymI_HasProto(getSparkzh_fast) \
631 SymI_HasProto(isCurrentThreadBoundzh_fast) \
632 SymI_HasProto(isDoubleDenormalized) \
633 SymI_HasProto(isDoubleInfinite) \
634 SymI_HasProto(isDoubleNaN) \
635 SymI_HasProto(isDoubleNegativeZero) \
636 SymI_HasProto(isEmptyMVarzh_fast) \
637 SymI_HasProto(isFloatDenormalized) \
638 SymI_HasProto(isFloatInfinite) \
639 SymI_HasProto(isFloatNaN) \
640 SymI_HasProto(isFloatNegativeZero) \
641 SymI_HasProto(killThreadzh_fast) \
642 SymI_HasProto(loadObj) \
643 SymI_HasProto(insertStableSymbol) \
644 SymI_HasProto(insertSymbol) \
645 SymI_HasProto(lookupSymbol) \
646 SymI_HasProto(makeStablePtrzh_fast) \
647 SymI_HasProto(mkApUpd0zh_fast) \
648 SymI_HasProto(myThreadIdzh_fast) \
649 SymI_HasProto(labelThreadzh_fast) \
650 SymI_HasProto(newArrayzh_fast) \
651 SymI_HasProto(newBCOzh_fast) \
652 SymI_HasProto(newByteArrayzh_fast) \
653 SymI_HasProto_redirect(newCAF, newDynCAF) \
654 SymI_HasProto(newMVarzh_fast) \
655 SymI_HasProto(newMutVarzh_fast) \
656 SymI_HasProto(newTVarzh_fast) \
657 SymI_HasProto(noDuplicatezh_fast) \
658 SymI_HasProto(atomicModifyMutVarzh_fast) \
659 SymI_HasProto(newPinnedByteArrayzh_fast) \
660 SymI_HasProto(newAlignedPinnedByteArrayzh_fast) \
661 SymI_HasProto(newSpark) \
662 SymI_HasProto(performGC) \
663 SymI_HasProto(performMajorGC) \
664 SymI_HasProto(prog_argc) \
665 SymI_HasProto(prog_argv) \
666 SymI_HasProto(putMVarzh_fast) \
667 SymI_HasProto(raisezh_fast) \
668 SymI_HasProto(raiseIOzh_fast) \
669 SymI_HasProto(readTVarzh_fast) \
670 SymI_HasProto(readTVarIOzh_fast) \
671 SymI_HasProto(resetNonBlockingFd) \
672 SymI_HasProto(resumeThread) \
673 SymI_HasProto(resolveObjs) \
674 SymI_HasProto(retryzh_fast) \
675 SymI_HasProto(rts_apply) \
676 SymI_HasProto(rts_checkSchedStatus) \
677 SymI_HasProto(rts_eval) \
678 SymI_HasProto(rts_evalIO) \
679 SymI_HasProto(rts_evalLazyIO) \
680 SymI_HasProto(rts_evalStableIO) \
681 SymI_HasProto(rts_eval_) \
682 SymI_HasProto(rts_getBool) \
683 SymI_HasProto(rts_getChar) \
684 SymI_HasProto(rts_getDouble) \
685 SymI_HasProto(rts_getFloat) \
686 SymI_HasProto(rts_getInt) \
687 SymI_HasProto(rts_getInt8) \
688 SymI_HasProto(rts_getInt16) \
689 SymI_HasProto(rts_getInt32) \
690 SymI_HasProto(rts_getInt64) \
691 SymI_HasProto(rts_getPtr) \
692 SymI_HasProto(rts_getFunPtr) \
693 SymI_HasProto(rts_getStablePtr) \
694 SymI_HasProto(rts_getThreadId) \
695 SymI_HasProto(rts_getWord) \
696 SymI_HasProto(rts_getWord8) \
697 SymI_HasProto(rts_getWord16) \
698 SymI_HasProto(rts_getWord32) \
699 SymI_HasProto(rts_getWord64) \
700 SymI_HasProto(rts_lock) \
701 SymI_HasProto(rts_mkBool) \
702 SymI_HasProto(rts_mkChar) \
703 SymI_HasProto(rts_mkDouble) \
704 SymI_HasProto(rts_mkFloat) \
705 SymI_HasProto(rts_mkInt) \
706 SymI_HasProto(rts_mkInt8) \
707 SymI_HasProto(rts_mkInt16) \
708 SymI_HasProto(rts_mkInt32) \
709 SymI_HasProto(rts_mkInt64) \
710 SymI_HasProto(rts_mkPtr) \
711 SymI_HasProto(rts_mkFunPtr) \
712 SymI_HasProto(rts_mkStablePtr) \
713 SymI_HasProto(rts_mkString) \
714 SymI_HasProto(rts_mkWord) \
715 SymI_HasProto(rts_mkWord8) \
716 SymI_HasProto(rts_mkWord16) \
717 SymI_HasProto(rts_mkWord32) \
718 SymI_HasProto(rts_mkWord64) \
719 SymI_HasProto(rts_unlock) \
720 SymI_HasProto(rts_unsafeGetMyCapability) \
721 SymI_HasProto(rtsSupportsBoundThreads) \
722 SymI_HasProto(__hscore_get_saved_termios) \
723 SymI_HasProto(__hscore_set_saved_termios) \
724 SymI_HasProto(setProgArgv) \
725 SymI_HasProto(startupHaskell) \
726 SymI_HasProto(shutdownHaskell) \
727 SymI_HasProto(shutdownHaskellAndExit) \
728 SymI_HasProto(stable_ptr_table) \
729 SymI_HasProto(stackOverflow) \
730 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
731 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
732 SymI_HasProto(awakenBlockedQueue) \
733 SymI_HasProto(startTimer) \
734 SymI_HasProto(stg_CHARLIKE_closure) \
735 SymI_HasProto(stg_MVAR_CLEAN_info) \
736 SymI_HasProto(stg_MVAR_DIRTY_info) \
737 SymI_HasProto(stg_IND_STATIC_info) \
738 SymI_HasProto(stg_INTLIKE_closure) \
739 SymI_HasProto(stg_ARR_WORDS_info) \
740 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
741 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
742 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
743 SymI_HasProto(stg_WEAK_info) \
744 SymI_HasProto(stg_ap_v_info) \
745 SymI_HasProto(stg_ap_f_info) \
746 SymI_HasProto(stg_ap_d_info) \
747 SymI_HasProto(stg_ap_l_info) \
748 SymI_HasProto(stg_ap_n_info) \
749 SymI_HasProto(stg_ap_p_info) \
750 SymI_HasProto(stg_ap_pv_info) \
751 SymI_HasProto(stg_ap_pp_info) \
752 SymI_HasProto(stg_ap_ppv_info) \
753 SymI_HasProto(stg_ap_ppp_info) \
754 SymI_HasProto(stg_ap_pppv_info) \
755 SymI_HasProto(stg_ap_pppp_info) \
756 SymI_HasProto(stg_ap_ppppp_info) \
757 SymI_HasProto(stg_ap_pppppp_info) \
758 SymI_HasProto(stg_ap_0_fast) \
759 SymI_HasProto(stg_ap_v_fast) \
760 SymI_HasProto(stg_ap_f_fast) \
761 SymI_HasProto(stg_ap_d_fast) \
762 SymI_HasProto(stg_ap_l_fast) \
763 SymI_HasProto(stg_ap_n_fast) \
764 SymI_HasProto(stg_ap_p_fast) \
765 SymI_HasProto(stg_ap_pv_fast) \
766 SymI_HasProto(stg_ap_pp_fast) \
767 SymI_HasProto(stg_ap_ppv_fast) \
768 SymI_HasProto(stg_ap_ppp_fast) \
769 SymI_HasProto(stg_ap_pppv_fast) \
770 SymI_HasProto(stg_ap_pppp_fast) \
771 SymI_HasProto(stg_ap_ppppp_fast) \
772 SymI_HasProto(stg_ap_pppppp_fast) \
773 SymI_HasProto(stg_ap_1_upd_info) \
774 SymI_HasProto(stg_ap_2_upd_info) \
775 SymI_HasProto(stg_ap_3_upd_info) \
776 SymI_HasProto(stg_ap_4_upd_info) \
777 SymI_HasProto(stg_ap_5_upd_info) \
778 SymI_HasProto(stg_ap_6_upd_info) \
779 SymI_HasProto(stg_ap_7_upd_info) \
780 SymI_HasProto(stg_exit) \
781 SymI_HasProto(stg_sel_0_upd_info) \
782 SymI_HasProto(stg_sel_10_upd_info) \
783 SymI_HasProto(stg_sel_11_upd_info) \
784 SymI_HasProto(stg_sel_12_upd_info) \
785 SymI_HasProto(stg_sel_13_upd_info) \
786 SymI_HasProto(stg_sel_14_upd_info) \
787 SymI_HasProto(stg_sel_15_upd_info) \
788 SymI_HasProto(stg_sel_1_upd_info) \
789 SymI_HasProto(stg_sel_2_upd_info) \
790 SymI_HasProto(stg_sel_3_upd_info) \
791 SymI_HasProto(stg_sel_4_upd_info) \
792 SymI_HasProto(stg_sel_5_upd_info) \
793 SymI_HasProto(stg_sel_6_upd_info) \
794 SymI_HasProto(stg_sel_7_upd_info) \
795 SymI_HasProto(stg_sel_8_upd_info) \
796 SymI_HasProto(stg_sel_9_upd_info) \
797 SymI_HasProto(stg_upd_frame_info) \
798 SymI_HasProto(suspendThread) \
799 SymI_HasProto(takeMVarzh_fast) \
800 SymI_HasProto(threadStatuszh_fast) \
801 SymI_HasProto(tryPutMVarzh_fast) \
802 SymI_HasProto(tryTakeMVarzh_fast) \
803 SymI_HasProto(unblockAsyncExceptionszh_fast) \
804 SymI_HasProto(unloadObj) \
805 SymI_HasProto(unsafeThawArrayzh_fast) \
806 SymI_HasProto(waitReadzh_fast) \
807 SymI_HasProto(waitWritezh_fast) \
808 SymI_HasProto(writeTVarzh_fast) \
809 SymI_HasProto(yieldzh_fast) \
810 SymI_NeedsProto(stg_interp_constr_entry) \
811 SymI_HasProto(alloc_blocks) \
812 SymI_HasProto(alloc_blocks_lim) \
813 SymI_HasProto(allocateLocal) \
814 SymI_HasProto(allocateExec) \
815 SymI_HasProto(freeExec) \
816 SymI_HasProto(getAllocations) \
817 SymI_HasProto(revertCAFs) \
818 SymI_HasProto(RtsFlags) \
819 SymI_NeedsProto(rts_breakpoint_io_action) \
820 SymI_NeedsProto(rts_stop_next_breakpoint) \
821 SymI_NeedsProto(rts_stop_on_exception) \
822 SymI_HasProto(stopTimer) \
823 SymI_HasProto(n_capabilities) \
824 SymI_HasProto(traceCcszh_fast) \
825 RTS_USER_SIGNALS_SYMBOLS
828 // 64-bit support functions in libgcc.a
829 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
830 #define RTS_LIBGCC_SYMBOLS \
831 SymI_NeedsProto(__divdi3) \
832 SymI_NeedsProto(__udivdi3) \
833 SymI_NeedsProto(__moddi3) \
834 SymI_NeedsProto(__umoddi3) \
835 SymI_NeedsProto(__muldi3) \
836 SymI_NeedsProto(__ashldi3) \
837 SymI_NeedsProto(__ashrdi3) \
838 SymI_NeedsProto(__lshrdi3) \
839 SymI_NeedsProto(__eprintf)
840 #elif defined(ia64_HOST_ARCH)
841 #define RTS_LIBGCC_SYMBOLS \
842 SymI_NeedsProto(__divdi3) \
843 SymI_NeedsProto(__udivdi3) \
844 SymI_NeedsProto(__moddi3) \
845 SymI_NeedsProto(__umoddi3) \
846 SymI_NeedsProto(__divsf3) \
847 SymI_NeedsProto(__divdf3)
849 #define RTS_LIBGCC_SYMBOLS
852 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
853 // Symbols that don't have a leading underscore
854 // on Mac OS X. They have to receive special treatment,
855 // see machoInitSymbolsWithoutUnderscore()
856 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
857 SymI_NeedsProto(saveFP) \
858 SymI_NeedsProto(restFP)
861 /* entirely bogus claims about types of these symbols */
862 #define SymI_NeedsProto(vvv) extern void vvv(void);
863 #if defined(__PIC__) && defined(mingw32_TARGET_OS)
864 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
865 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
867 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
868 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
870 #define SymI_HasProto(vvv) /**/
871 #define SymI_HasProto_redirect(vvv,xxx) /**/
874 RTS_POSIX_ONLY_SYMBOLS
875 RTS_MINGW_ONLY_SYMBOLS
876 RTS_CYGWIN_ONLY_SYMBOLS
877 RTS_DARWIN_ONLY_SYMBOLS
880 #undef SymI_NeedsProto
882 #undef SymI_HasProto_redirect
884 #undef SymE_NeedsProto
886 #ifdef LEADING_UNDERSCORE
887 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
889 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
892 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
894 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
895 (void*)DLL_IMPORT_DATA_REF(vvv) },
897 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
898 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
900 // SymI_HasProto_redirect allows us to redirect references to one symbol to
901 // another symbol. See newCAF/newDynCAF for an example.
902 #define SymI_HasProto_redirect(vvv,xxx) \
903 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
906 static RtsSymbolVal rtsSyms[] = {
909 RTS_POSIX_ONLY_SYMBOLS
910 RTS_MINGW_ONLY_SYMBOLS
911 RTS_CYGWIN_ONLY_SYMBOLS
912 RTS_DARWIN_ONLY_SYMBOLS
915 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
916 // dyld stub code contains references to this,
917 // but it should never be called because we treat
918 // lazy pointers as nonlazy.
919 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
921 { 0, 0 } /* sentinel */
926 /* -----------------------------------------------------------------------------
927 * Insert symbols into hash tables, checking for duplicates.
930 static void ghciInsertStrHashTable ( char* obj_name,
936 if (lookupHashTable(table, (StgWord)key) == NULL)
938 insertStrHashTable(table, (StgWord)key, data);
943 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
945 "whilst processing object file\n"
947 "This could be caused by:\n"
948 " * Loading two different object files which export the same symbol\n"
949 " * Specifying the same object file twice on the GHCi command line\n"
950 " * An incorrect `package.conf' entry, causing some object to be\n"
952 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
959 /* -----------------------------------------------------------------------------
960 * initialize the object linker
964 static int linker_init_done = 0 ;
966 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
967 static void *dl_prog_handle;
975 /* Make initLinker idempotent, so we can call it
976 before evey relevant operation; that means we
977 don't need to initialise the linker separately */
978 if (linker_init_done == 1) { return; } else {
979 linker_init_done = 1;
982 stablehash = allocStrHashTable();
983 symhash = allocStrHashTable();
985 /* populate the symbol table with stuff from the RTS */
986 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
987 ghciInsertStrHashTable("(GHCi built-in symbols)",
988 symhash, sym->lbl, sym->addr);
990 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
991 machoInitSymbolsWithoutUnderscore();
994 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
995 # if defined(RTLD_DEFAULT)
996 dl_prog_handle = RTLD_DEFAULT;
998 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
999 # endif /* RTLD_DEFAULT */
1002 #if defined(x86_64_HOST_ARCH)
1003 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1004 // User-override for mmap_32bit_base
1005 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1009 #if defined(mingw32_HOST_OS)
1011 * These two libraries cause problems when added to the static link,
1012 * but are necessary for resolving symbols in GHCi, hence we load
1013 * them manually here.
1020 /* -----------------------------------------------------------------------------
1021 * Loading DLL or .so dynamic libraries
1022 * -----------------------------------------------------------------------------
1024 * Add a DLL from which symbols may be found. In the ELF case, just
1025 * do RTLD_GLOBAL-style add, so no further messing around needs to
1026 * happen in order that symbols in the loaded .so are findable --
1027 * lookupSymbol() will subsequently see them by dlsym on the program's
1028 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1030 * In the PEi386 case, open the DLLs and put handles to them in a
1031 * linked list. When looking for a symbol, try all handles in the
1032 * list. This means that we need to load even DLLs that are guaranteed
1033 * to be in the ghc.exe image already, just so we can get a handle
1034 * to give to loadSymbol, so that we can find the symbols. For such
1035 * libraries, the LoadLibrary call should be a no-op except for returning
1040 #if defined(OBJFORMAT_PEi386)
1041 /* A record for storing handles into DLLs. */
1046 struct _OpenedDLL* next;
1051 /* A list thereof. */
1052 static OpenedDLL* opened_dlls = NULL;
1056 addDLL( char *dll_name )
1058 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1059 /* ------------------- ELF DLL loader ------------------- */
1065 // omitted: RTLD_NOW
1066 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1067 hdl= dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1070 /* dlopen failed; return a ptr to the error msg. */
1072 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1079 # elif defined(OBJFORMAT_PEi386)
1080 /* ------------------- Win32 DLL loader ------------------- */
1088 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1090 /* See if we've already got it, and ignore if so. */
1091 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1092 if (0 == strcmp(o_dll->name, dll_name))
1096 /* The file name has no suffix (yet) so that we can try
1097 both foo.dll and foo.drv
1099 The documentation for LoadLibrary says:
1100 If no file name extension is specified in the lpFileName
1101 parameter, the default library extension .dll is
1102 appended. However, the file name string can include a trailing
1103 point character (.) to indicate that the module name has no
1106 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1107 sprintf(buf, "%s.DLL", dll_name);
1108 instance = LoadLibrary(buf);
1109 if (instance == NULL) {
1110 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1111 // KAA: allow loading of drivers (like winspool.drv)
1112 sprintf(buf, "%s.DRV", dll_name);
1113 instance = LoadLibrary(buf);
1114 if (instance == NULL) {
1115 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1116 // #1883: allow loading of unix-style libfoo.dll DLLs
1117 sprintf(buf, "lib%s.DLL", dll_name);
1118 instance = LoadLibrary(buf);
1119 if (instance == NULL) {
1126 /* Add this DLL to the list of DLLs in which to search for symbols. */
1127 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1128 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1129 strcpy(o_dll->name, dll_name);
1130 o_dll->instance = instance;
1131 o_dll->next = opened_dlls;
1132 opened_dlls = o_dll;
1138 sysErrorBelch(dll_name);
1140 /* LoadLibrary failed; return a ptr to the error msg. */
1141 return "addDLL: could not load DLL";
1144 barf("addDLL: not implemented on this platform");
1148 /* -----------------------------------------------------------------------------
1149 * insert a stable symbol in the hash table
1153 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1155 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1159 /* -----------------------------------------------------------------------------
1160 * insert a symbol in the hash table
1163 insertSymbol(char* obj_name, char* key, void* data)
1165 ghciInsertStrHashTable(obj_name, symhash, key, data);
1168 /* -----------------------------------------------------------------------------
1169 * lookup a symbol in the hash table
1172 lookupSymbol( char *lbl )
1176 ASSERT(symhash != NULL);
1177 val = lookupStrHashTable(symhash, lbl);
1180 # if defined(OBJFORMAT_ELF)
1181 return dlsym(dl_prog_handle, lbl);
1182 # elif defined(OBJFORMAT_MACHO)
1184 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1187 HACK: On OS X, global symbols are prefixed with an underscore.
1188 However, dlsym wants us to omit the leading underscore from the
1189 symbol name. For now, we simply strip it off here (and ONLY
1192 ASSERT(lbl[0] == '_');
1193 return dlsym(dl_prog_handle, lbl+1);
1195 if(NSIsSymbolNameDefined(lbl)) {
1196 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1197 return NSAddressOfSymbol(symbol);
1201 # endif /* HAVE_DLFCN_H */
1202 # elif defined(OBJFORMAT_PEi386)
1205 sym = lookupSymbolInDLLs(lbl);
1206 if (sym != NULL) { return sym; };
1208 // Also try looking up the symbol without the @N suffix. Some
1209 // DLLs have the suffixes on their symbols, some don't.
1210 zapTrailingAtSign ( lbl );
1211 sym = lookupSymbolInDLLs(lbl);
1212 if (sym != NULL) { return sym; };
1224 /* -----------------------------------------------------------------------------
1225 * Debugging aid: look in GHCi's object symbol tables for symbols
1226 * within DELTA bytes of the specified address, and show their names.
1229 void ghci_enquire ( char* addr );
1231 void ghci_enquire ( char* addr )
1236 const int DELTA = 64;
1241 for (oc = objects; oc; oc = oc->next) {
1242 for (i = 0; i < oc->n_symbols; i++) {
1243 sym = oc->symbols[i];
1244 if (sym == NULL) continue;
1247 a = lookupStrHashTable(symhash, sym);
1250 // debugBelch("ghci_enquire: can't find %s\n", sym);
1252 else if (addr-DELTA <= a && a <= addr+DELTA) {
1253 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1260 #ifdef ia64_HOST_ARCH
1261 static unsigned int PLTSize(void);
1265 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1268 mmapForLinker (size_t bytes, nat flags, int fd)
1270 void *map_addr = NULL;
1273 static nat fixed = 0;
1275 pagesize = getpagesize();
1276 size = ROUND_UP(bytes, pagesize);
1278 #if defined(x86_64_HOST_ARCH)
1281 if (mmap_32bit_base != 0) {
1282 map_addr = mmap_32bit_base;
1286 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1287 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1289 if (result == MAP_FAILED) {
1290 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1291 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1292 stg_exit(EXIT_FAILURE);
1295 #if defined(x86_64_HOST_ARCH)
1296 if (mmap_32bit_base != 0) {
1297 if (result == map_addr) {
1298 mmap_32bit_base = map_addr + size;
1300 if ((W_)result > 0x80000000) {
1301 // oops, we were given memory over 2Gb
1302 #if defined(freebsd_HOST_OS)
1303 // Some platforms require MAP_FIXED. This is normally
1304 // a bad idea, because MAP_FIXED will overwrite
1305 // existing mappings.
1306 munmap(result,size);
1310 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);
1313 // hmm, we were given memory somewhere else, but it's
1314 // still under 2Gb so we can use it. Next time, ask
1315 // for memory right after the place we just got some
1316 mmap_32bit_base = (void*)result + size;
1320 if ((W_)result > 0x80000000) {
1321 // oops, we were given memory over 2Gb
1322 // ... try allocating memory somewhere else?;
1323 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1324 munmap(result, size);
1326 // Set a base address and try again... (guess: 1Gb)
1327 mmap_32bit_base = (void*)0x40000000;
1337 /* -----------------------------------------------------------------------------
1338 * Load an obj (populate the global symbol table, but don't resolve yet)
1340 * Returns: 1 if ok, 0 on error.
1343 loadObj( char *path )
1355 /* debugBelch("loadObj %s\n", path ); */
1357 /* Check that we haven't already loaded this object.
1358 Ignore requests to load multiple times */
1362 for (o = objects; o; o = o->next) {
1363 if (0 == strcmp(o->fileName, path)) {
1365 break; /* don't need to search further */
1369 IF_DEBUG(linker, debugBelch(
1370 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1371 "same object file twice:\n"
1373 "GHCi will ignore this, but be warned.\n"
1375 return 1; /* success */
1379 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1381 # if defined(OBJFORMAT_ELF)
1382 oc->formatName = "ELF";
1383 # elif defined(OBJFORMAT_PEi386)
1384 oc->formatName = "PEi386";
1385 # elif defined(OBJFORMAT_MACHO)
1386 oc->formatName = "Mach-O";
1389 barf("loadObj: not implemented on this platform");
1392 r = stat(path, &st);
1393 if (r == -1) { return 0; }
1395 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1396 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1397 strcpy(oc->fileName, path);
1399 oc->fileSize = st.st_size;
1401 oc->sections = NULL;
1402 oc->proddables = NULL;
1404 /* chain it onto the list of objects */
1409 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1411 #if defined(openbsd_HOST_OS)
1412 fd = open(path, O_RDONLY, S_IRUSR);
1414 fd = open(path, O_RDONLY);
1417 barf("loadObj: can't open `%s'", path);
1419 #ifdef ia64_HOST_ARCH
1420 /* The PLT needs to be right before the object */
1423 pagesize = getpagesize();
1424 n = ROUND_UP(PLTSize(), pagesize);
1425 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1426 if (oc->plt == MAP_FAILED)
1427 barf("loadObj: can't allocate PLT");
1430 map_addr = oc->plt + n;
1432 n = ROUND_UP(oc->fileSize, pagesize);
1433 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1434 MAP_PRIVATE|TRY_MAP_32BIT, fd, 0);
1435 if (oc->image == MAP_FAILED)
1436 barf("loadObj: can't map `%s'", path);
1439 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1444 #else /* !USE_MMAP */
1445 /* load the image into memory */
1446 f = fopen(path, "rb");
1448 barf("loadObj: can't read `%s'", path);
1450 # if defined(mingw32_HOST_OS)
1451 // TODO: We would like to use allocateExec here, but allocateExec
1452 // cannot currently allocate blocks large enough.
1453 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1454 PAGE_EXECUTE_READWRITE);
1455 # elif defined(darwin_HOST_OS)
1456 // In a Mach-O .o file, all sections can and will be misaligned
1457 // if the total size of the headers is not a multiple of the
1458 // desired alignment. This is fine for .o files that only serve
1459 // as input for the static linker, but it's not fine for us,
1460 // as SSE (used by gcc for floating point) and Altivec require
1461 // 16-byte alignment.
1462 // We calculate the correct alignment from the header before
1463 // reading the file, and then we misalign oc->image on purpose so
1464 // that the actual sections end up aligned again.
1465 oc->misalignment = machoGetMisalignment(f);
1466 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1467 oc->image += oc->misalignment;
1469 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1474 n = fread ( oc->image, 1, oc->fileSize, f );
1475 if (n != oc->fileSize)
1476 barf("loadObj: error whilst reading `%s'", path);
1479 #endif /* USE_MMAP */
1481 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1482 r = ocAllocateSymbolExtras_MachO ( oc );
1483 if (!r) { return r; }
1484 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1485 r = ocAllocateSymbolExtras_ELF ( oc );
1486 if (!r) { return r; }
1489 /* verify the in-memory image */
1490 # if defined(OBJFORMAT_ELF)
1491 r = ocVerifyImage_ELF ( oc );
1492 # elif defined(OBJFORMAT_PEi386)
1493 r = ocVerifyImage_PEi386 ( oc );
1494 # elif defined(OBJFORMAT_MACHO)
1495 r = ocVerifyImage_MachO ( oc );
1497 barf("loadObj: no verify method");
1499 if (!r) { return r; }
1501 /* build the symbol list for this image */
1502 # if defined(OBJFORMAT_ELF)
1503 r = ocGetNames_ELF ( oc );
1504 # elif defined(OBJFORMAT_PEi386)
1505 r = ocGetNames_PEi386 ( oc );
1506 # elif defined(OBJFORMAT_MACHO)
1507 r = ocGetNames_MachO ( oc );
1509 barf("loadObj: no getNames method");
1511 if (!r) { return r; }
1513 /* loaded, but not resolved yet */
1514 oc->status = OBJECT_LOADED;
1519 /* -----------------------------------------------------------------------------
1520 * resolve all the currently unlinked objects in memory
1522 * Returns: 1 if ok, 0 on error.
1532 for (oc = objects; oc; oc = oc->next) {
1533 if (oc->status != OBJECT_RESOLVED) {
1534 # if defined(OBJFORMAT_ELF)
1535 r = ocResolve_ELF ( oc );
1536 # elif defined(OBJFORMAT_PEi386)
1537 r = ocResolve_PEi386 ( oc );
1538 # elif defined(OBJFORMAT_MACHO)
1539 r = ocResolve_MachO ( oc );
1541 barf("resolveObjs: not implemented on this platform");
1543 if (!r) { return r; }
1544 oc->status = OBJECT_RESOLVED;
1550 /* -----------------------------------------------------------------------------
1551 * delete an object from the pool
1554 unloadObj( char *path )
1556 ObjectCode *oc, *prev;
1558 ASSERT(symhash != NULL);
1559 ASSERT(objects != NULL);
1564 for (oc = objects; oc; prev = oc, oc = oc->next) {
1565 if (!strcmp(oc->fileName,path)) {
1567 /* Remove all the mappings for the symbols within this
1572 for (i = 0; i < oc->n_symbols; i++) {
1573 if (oc->symbols[i] != NULL) {
1574 removeStrHashTable(symhash, oc->symbols[i], NULL);
1582 prev->next = oc->next;
1585 // We're going to leave this in place, in case there are
1586 // any pointers from the heap into it:
1587 // #ifdef mingw32_HOST_OS
1588 // VirtualFree(oc->image);
1590 // stgFree(oc->image);
1592 stgFree(oc->fileName);
1593 stgFree(oc->symbols);
1594 stgFree(oc->sections);
1600 errorBelch("unloadObj: can't find `%s' to unload", path);
1604 /* -----------------------------------------------------------------------------
1605 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1606 * which may be prodded during relocation, and abort if we try and write
1607 * outside any of these.
1609 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1612 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1613 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1617 pb->next = oc->proddables;
1618 oc->proddables = pb;
1621 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1624 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1625 char* s = (char*)(pb->start);
1626 char* e = s + pb->size - 1;
1627 char* a = (char*)addr;
1628 /* Assumes that the biggest fixup involves a 4-byte write. This
1629 probably needs to be changed to 8 (ie, +7) on 64-bit
1631 if (a >= s && (a+3) <= e) return;
1633 barf("checkProddableBlock: invalid fixup in runtime linker");
1636 /* -----------------------------------------------------------------------------
1637 * Section management.
1639 static void addSection ( ObjectCode* oc, SectionKind kind,
1640 void* start, void* end )
1642 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1646 s->next = oc->sections;
1649 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1650 start, ((char*)end)-1, end - start + 1, kind );
1655 /* --------------------------------------------------------------------------
1657 * This is about allocating a small chunk of memory for every symbol in the
1658 * object file. We make sure that the SymboLExtras are always "in range" of
1659 * limited-range PC-relative instructions on various platforms by allocating
1660 * them right next to the object code itself.
1663 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1666 ocAllocateSymbolExtras
1668 Allocate additional space at the end of the object file image to make room
1669 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1671 PowerPC relative branch instructions have a 24 bit displacement field.
1672 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1673 If a particular imported symbol is outside this range, we have to redirect
1674 the jump to a short piece of new code that just loads the 32bit absolute
1675 address and jumps there.
1676 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1679 This function just allocates space for one SymbolExtra for every
1680 undefined symbol in the object file. The code for the jump islands is
1681 filled in by makeSymbolExtra below.
1684 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1691 int misalignment = 0;
1692 #ifdef darwin_HOST_OS
1693 misalignment = oc->misalignment;
1699 // round up to the nearest 4
1700 aligned = (oc->fileSize + 3) & ~3;
1703 pagesize = getpagesize();
1704 n = ROUND_UP( oc->fileSize, pagesize );
1705 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1707 /* we try to use spare space at the end of the last page of the
1708 * image for the jump islands, but if there isn't enough space
1709 * then we have to map some (anonymously, remembering MAP_32BIT).
1711 if( m > n ) // we need to allocate more pages
1713 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1718 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1721 oc->image -= misalignment;
1722 oc->image = stgReallocBytes( oc->image,
1724 aligned + sizeof (SymbolExtra) * count,
1725 "ocAllocateSymbolExtras" );
1726 oc->image += misalignment;
1728 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1729 #endif /* USE_MMAP */
1731 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1734 oc->symbol_extras = NULL;
1736 oc->first_symbol_extra = first;
1737 oc->n_symbol_extras = count;
1742 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1743 unsigned long symbolNumber,
1744 unsigned long target )
1748 ASSERT( symbolNumber >= oc->first_symbol_extra
1749 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1751 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1753 #ifdef powerpc_HOST_ARCH
1754 // lis r12, hi16(target)
1755 extra->jumpIsland.lis_r12 = 0x3d80;
1756 extra->jumpIsland.hi_addr = target >> 16;
1758 // ori r12, r12, lo16(target)
1759 extra->jumpIsland.ori_r12_r12 = 0x618c;
1760 extra->jumpIsland.lo_addr = target & 0xffff;
1763 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1766 extra->jumpIsland.bctr = 0x4e800420;
1768 #ifdef x86_64_HOST_ARCH
1770 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1771 extra->addr = target;
1772 memcpy(extra->jumpIsland, jmp, 6);
1780 /* --------------------------------------------------------------------------
1781 * PowerPC specifics (instruction cache flushing)
1782 * ------------------------------------------------------------------------*/
1784 #ifdef powerpc_TARGET_ARCH
1786 ocFlushInstructionCache
1788 Flush the data & instruction caches.
1789 Because the PPC has split data/instruction caches, we have to
1790 do that whenever we modify code at runtime.
1793 static void ocFlushInstructionCache( ObjectCode *oc )
1795 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1796 unsigned long *p = (unsigned long *) oc->image;
1800 __asm__ volatile ( "dcbf 0,%0\n\t"
1808 __asm__ volatile ( "sync\n\t"
1814 /* --------------------------------------------------------------------------
1815 * PEi386 specifics (Win32 targets)
1816 * ------------------------------------------------------------------------*/
1818 /* The information for this linker comes from
1819 Microsoft Portable Executable
1820 and Common Object File Format Specification
1821 revision 5.1 January 1998
1822 which SimonM says comes from the MS Developer Network CDs.
1824 It can be found there (on older CDs), but can also be found
1827 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1829 (this is Rev 6.0 from February 1999).
1831 Things move, so if that fails, try searching for it via
1833 http://www.google.com/search?q=PE+COFF+specification
1835 The ultimate reference for the PE format is the Winnt.h
1836 header file that comes with the Platform SDKs; as always,
1837 implementations will drift wrt their documentation.
1839 A good background article on the PE format is Matt Pietrek's
1840 March 1994 article in Microsoft System Journal (MSJ)
1841 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1842 Win32 Portable Executable File Format." The info in there
1843 has recently been updated in a two part article in
1844 MSDN magazine, issues Feb and March 2002,
1845 "Inside Windows: An In-Depth Look into the Win32 Portable
1846 Executable File Format"
1848 John Levine's book "Linkers and Loaders" contains useful
1853 #if defined(OBJFORMAT_PEi386)
1857 typedef unsigned char UChar;
1858 typedef unsigned short UInt16;
1859 typedef unsigned int UInt32;
1866 UInt16 NumberOfSections;
1867 UInt32 TimeDateStamp;
1868 UInt32 PointerToSymbolTable;
1869 UInt32 NumberOfSymbols;
1870 UInt16 SizeOfOptionalHeader;
1871 UInt16 Characteristics;
1875 #define sizeof_COFF_header 20
1882 UInt32 VirtualAddress;
1883 UInt32 SizeOfRawData;
1884 UInt32 PointerToRawData;
1885 UInt32 PointerToRelocations;
1886 UInt32 PointerToLinenumbers;
1887 UInt16 NumberOfRelocations;
1888 UInt16 NumberOfLineNumbers;
1889 UInt32 Characteristics;
1893 #define sizeof_COFF_section 40
1900 UInt16 SectionNumber;
1903 UChar NumberOfAuxSymbols;
1907 #define sizeof_COFF_symbol 18
1912 UInt32 VirtualAddress;
1913 UInt32 SymbolTableIndex;
1918 #define sizeof_COFF_reloc 10
1921 /* From PE spec doc, section 3.3.2 */
1922 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1923 windows.h -- for the same purpose, but I want to know what I'm
1925 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1926 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1927 #define MYIMAGE_FILE_DLL 0x2000
1928 #define MYIMAGE_FILE_SYSTEM 0x1000
1929 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1930 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1931 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1933 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1934 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1935 #define MYIMAGE_SYM_CLASS_STATIC 3
1936 #define MYIMAGE_SYM_UNDEFINED 0
1938 /* From PE spec doc, section 4.1 */
1939 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1940 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1941 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1943 /* From PE spec doc, section 5.2.1 */
1944 #define MYIMAGE_REL_I386_DIR32 0x0006
1945 #define MYIMAGE_REL_I386_REL32 0x0014
1948 /* We use myindex to calculate array addresses, rather than
1949 simply doing the normal subscript thing. That's because
1950 some of the above structs have sizes which are not
1951 a whole number of words. GCC rounds their sizes up to a
1952 whole number of words, which means that the address calcs
1953 arising from using normal C indexing or pointer arithmetic
1954 are just plain wrong. Sigh.
1957 myindex ( int scale, void* base, int index )
1960 ((UChar*)base) + scale * index;
1965 printName ( UChar* name, UChar* strtab )
1967 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1968 UInt32 strtab_offset = * (UInt32*)(name+4);
1969 debugBelch("%s", strtab + strtab_offset );
1972 for (i = 0; i < 8; i++) {
1973 if (name[i] == 0) break;
1974 debugBelch("%c", name[i] );
1981 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1983 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1984 UInt32 strtab_offset = * (UInt32*)(name+4);
1985 strncpy ( dst, strtab+strtab_offset, dstSize );
1991 if (name[i] == 0) break;
2001 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2004 /* If the string is longer than 8 bytes, look in the
2005 string table for it -- this will be correctly zero terminated.
2007 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2008 UInt32 strtab_offset = * (UInt32*)(name+4);
2009 return ((UChar*)strtab) + strtab_offset;
2011 /* Otherwise, if shorter than 8 bytes, return the original,
2012 which by defn is correctly terminated.
2014 if (name[7]==0) return name;
2015 /* The annoying case: 8 bytes. Copy into a temporary
2016 (which is never freed ...)
2018 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2020 strncpy(newstr,name,8);
2026 /* Just compares the short names (first 8 chars) */
2027 static COFF_section *
2028 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2032 = (COFF_header*)(oc->image);
2033 COFF_section* sectab
2035 ((UChar*)(oc->image))
2036 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2038 for (i = 0; i < hdr->NumberOfSections; i++) {
2041 COFF_section* section_i
2043 myindex ( sizeof_COFF_section, sectab, i );
2044 n1 = (UChar*) &(section_i->Name);
2046 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2047 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2048 n1[6]==n2[6] && n1[7]==n2[7])
2057 zapTrailingAtSign ( UChar* sym )
2059 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2061 if (sym[0] == 0) return;
2063 while (sym[i] != 0) i++;
2066 while (j > 0 && my_isdigit(sym[j])) j--;
2067 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2072 lookupSymbolInDLLs ( UChar *lbl )
2077 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2078 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2080 if (lbl[0] == '_') {
2081 /* HACK: if the name has an initial underscore, try stripping
2082 it off & look that up first. I've yet to verify whether there's
2083 a Rule that governs whether an initial '_' *should always* be
2084 stripped off when mapping from import lib name to the DLL name.
2086 sym = GetProcAddress(o_dll->instance, (lbl+1));
2088 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2092 sym = GetProcAddress(o_dll->instance, lbl);
2094 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2103 ocVerifyImage_PEi386 ( ObjectCode* oc )
2108 COFF_section* sectab;
2109 COFF_symbol* symtab;
2111 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2112 hdr = (COFF_header*)(oc->image);
2113 sectab = (COFF_section*) (
2114 ((UChar*)(oc->image))
2115 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2117 symtab = (COFF_symbol*) (
2118 ((UChar*)(oc->image))
2119 + hdr->PointerToSymbolTable
2121 strtab = ((UChar*)symtab)
2122 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2124 if (hdr->Machine != 0x14c) {
2125 errorBelch("%s: Not x86 PEi386", oc->fileName);
2128 if (hdr->SizeOfOptionalHeader != 0) {
2129 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2132 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2133 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2134 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2135 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2136 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2139 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2140 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2141 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2143 (int)(hdr->Characteristics));
2146 /* If the string table size is way crazy, this might indicate that
2147 there are more than 64k relocations, despite claims to the
2148 contrary. Hence this test. */
2149 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2151 if ( (*(UInt32*)strtab) > 600000 ) {
2152 /* Note that 600k has no special significance other than being
2153 big enough to handle the almost-2MB-sized lumps that
2154 constitute HSwin32*.o. */
2155 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2160 /* No further verification after this point; only debug printing. */
2162 IF_DEBUG(linker, i=1);
2163 if (i == 0) return 1;
2165 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2166 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2167 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2170 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2171 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2172 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2173 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2174 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2175 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2176 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2178 /* Print the section table. */
2180 for (i = 0; i < hdr->NumberOfSections; i++) {
2182 COFF_section* sectab_i
2184 myindex ( sizeof_COFF_section, sectab, i );
2191 printName ( sectab_i->Name, strtab );
2201 sectab_i->VirtualSize,
2202 sectab_i->VirtualAddress,
2203 sectab_i->SizeOfRawData,
2204 sectab_i->PointerToRawData,
2205 sectab_i->NumberOfRelocations,
2206 sectab_i->PointerToRelocations,
2207 sectab_i->PointerToRawData
2209 reltab = (COFF_reloc*) (
2210 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2213 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2214 /* If the relocation field (a short) has overflowed, the
2215 * real count can be found in the first reloc entry.
2217 * See Section 4.1 (last para) of the PE spec (rev6.0).
2219 COFF_reloc* rel = (COFF_reloc*)
2220 myindex ( sizeof_COFF_reloc, reltab, 0 );
2221 noRelocs = rel->VirtualAddress;
2224 noRelocs = sectab_i->NumberOfRelocations;
2228 for (; j < noRelocs; j++) {
2230 COFF_reloc* rel = (COFF_reloc*)
2231 myindex ( sizeof_COFF_reloc, reltab, j );
2233 " type 0x%-4x vaddr 0x%-8x name `",
2235 rel->VirtualAddress );
2236 sym = (COFF_symbol*)
2237 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2238 /* Hmm..mysterious looking offset - what's it for? SOF */
2239 printName ( sym->Name, strtab -10 );
2246 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2247 debugBelch("---START of string table---\n");
2248 for (i = 4; i < *(Int32*)strtab; i++) {
2250 debugBelch("\n"); else
2251 debugBelch("%c", strtab[i] );
2253 debugBelch("--- END of string table---\n");
2258 COFF_symbol* symtab_i;
2259 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2260 symtab_i = (COFF_symbol*)
2261 myindex ( sizeof_COFF_symbol, symtab, i );
2267 printName ( symtab_i->Name, strtab );
2276 (Int32)(symtab_i->SectionNumber),
2277 (UInt32)symtab_i->Type,
2278 (UInt32)symtab_i->StorageClass,
2279 (UInt32)symtab_i->NumberOfAuxSymbols
2281 i += symtab_i->NumberOfAuxSymbols;
2291 ocGetNames_PEi386 ( ObjectCode* oc )
2294 COFF_section* sectab;
2295 COFF_symbol* symtab;
2302 hdr = (COFF_header*)(oc->image);
2303 sectab = (COFF_section*) (
2304 ((UChar*)(oc->image))
2305 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2307 symtab = (COFF_symbol*) (
2308 ((UChar*)(oc->image))
2309 + hdr->PointerToSymbolTable
2311 strtab = ((UChar*)(oc->image))
2312 + hdr->PointerToSymbolTable
2313 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2315 /* Allocate space for any (local, anonymous) .bss sections. */
2317 for (i = 0; i < hdr->NumberOfSections; i++) {
2320 COFF_section* sectab_i
2322 myindex ( sizeof_COFF_section, sectab, i );
2323 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2324 /* sof 10/05: the PE spec text isn't too clear regarding what
2325 * the SizeOfRawData field is supposed to hold for object
2326 * file sections containing just uninitialized data -- for executables,
2327 * it is supposed to be zero; unclear what it's supposed to be
2328 * for object files. However, VirtualSize is guaranteed to be
2329 * zero for object files, which definitely suggests that SizeOfRawData
2330 * will be non-zero (where else would the size of this .bss section be
2331 * stored?) Looking at the COFF_section info for incoming object files,
2332 * this certainly appears to be the case.
2334 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2335 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2336 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2337 * variable decls into to the .bss section. (The specific function in Q which
2338 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2340 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2341 /* This is a non-empty .bss section. Allocate zeroed space for
2342 it, and set its PointerToRawData field such that oc->image +
2343 PointerToRawData == addr_of_zeroed_space. */
2344 bss_sz = sectab_i->VirtualSize;
2345 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2346 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2347 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2348 addProddableBlock(oc, zspace, bss_sz);
2349 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2352 /* Copy section information into the ObjectCode. */
2354 for (i = 0; i < hdr->NumberOfSections; i++) {
2360 = SECTIONKIND_OTHER;
2361 COFF_section* sectab_i
2363 myindex ( sizeof_COFF_section, sectab, i );
2364 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2367 /* I'm sure this is the Right Way to do it. However, the
2368 alternative of testing the sectab_i->Name field seems to
2369 work ok with Cygwin.
2371 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2372 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2373 kind = SECTIONKIND_CODE_OR_RODATA;
2376 if (0==strcmp(".text",sectab_i->Name) ||
2377 0==strcmp(".rdata",sectab_i->Name)||
2378 0==strcmp(".rodata",sectab_i->Name))
2379 kind = SECTIONKIND_CODE_OR_RODATA;
2380 if (0==strcmp(".data",sectab_i->Name) ||
2381 0==strcmp(".bss",sectab_i->Name))
2382 kind = SECTIONKIND_RWDATA;
2384 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2385 sz = sectab_i->SizeOfRawData;
2386 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2388 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2389 end = start + sz - 1;
2391 if (kind == SECTIONKIND_OTHER
2392 /* Ignore sections called which contain stabs debugging
2394 && 0 != strcmp(".stab", sectab_i->Name)
2395 && 0 != strcmp(".stabstr", sectab_i->Name)
2396 /* ignore constructor section for now */
2397 && 0 != strcmp(".ctors", sectab_i->Name)
2398 /* ignore section generated from .ident */
2399 && 0!= strcmp("/4", sectab_i->Name)
2400 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2401 && 0!= strcmp(".reloc", sectab_i->Name)
2403 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2407 if (kind != SECTIONKIND_OTHER && end >= start) {
2408 addSection(oc, kind, start, end);
2409 addProddableBlock(oc, start, end - start + 1);
2413 /* Copy exported symbols into the ObjectCode. */
2415 oc->n_symbols = hdr->NumberOfSymbols;
2416 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2417 "ocGetNames_PEi386(oc->symbols)");
2418 /* Call me paranoid; I don't care. */
2419 for (i = 0; i < oc->n_symbols; i++)
2420 oc->symbols[i] = NULL;
2424 COFF_symbol* symtab_i;
2425 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2426 symtab_i = (COFF_symbol*)
2427 myindex ( sizeof_COFF_symbol, symtab, i );
2431 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2432 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2433 /* This symbol is global and defined, viz, exported */
2434 /* for MYIMAGE_SYMCLASS_EXTERNAL
2435 && !MYIMAGE_SYM_UNDEFINED,
2436 the address of the symbol is:
2437 address of relevant section + offset in section
2439 COFF_section* sectabent
2440 = (COFF_section*) myindex ( sizeof_COFF_section,
2442 symtab_i->SectionNumber-1 );
2443 addr = ((UChar*)(oc->image))
2444 + (sectabent->PointerToRawData
2448 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2449 && symtab_i->Value > 0) {
2450 /* This symbol isn't in any section at all, ie, global bss.
2451 Allocate zeroed space for it. */
2452 addr = stgCallocBytes(1, symtab_i->Value,
2453 "ocGetNames_PEi386(non-anonymous bss)");
2454 addSection(oc, SECTIONKIND_RWDATA, addr,
2455 ((UChar*)addr) + symtab_i->Value - 1);
2456 addProddableBlock(oc, addr, symtab_i->Value);
2457 /* debugBelch("BSS section at 0x%x\n", addr); */
2460 if (addr != NULL ) {
2461 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2462 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2463 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2464 ASSERT(i >= 0 && i < oc->n_symbols);
2465 /* cstring_from_COFF_symbol_name always succeeds. */
2466 oc->symbols[i] = sname;
2467 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2471 "IGNORING symbol %d\n"
2475 printName ( symtab_i->Name, strtab );
2484 (Int32)(symtab_i->SectionNumber),
2485 (UInt32)symtab_i->Type,
2486 (UInt32)symtab_i->StorageClass,
2487 (UInt32)symtab_i->NumberOfAuxSymbols
2492 i += symtab_i->NumberOfAuxSymbols;
2501 ocResolve_PEi386 ( ObjectCode* oc )
2504 COFF_section* sectab;
2505 COFF_symbol* symtab;
2515 /* ToDo: should be variable-sized? But is at least safe in the
2516 sense of buffer-overrun-proof. */
2518 /* debugBelch("resolving for %s\n", oc->fileName); */
2520 hdr = (COFF_header*)(oc->image);
2521 sectab = (COFF_section*) (
2522 ((UChar*)(oc->image))
2523 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2525 symtab = (COFF_symbol*) (
2526 ((UChar*)(oc->image))
2527 + hdr->PointerToSymbolTable
2529 strtab = ((UChar*)(oc->image))
2530 + hdr->PointerToSymbolTable
2531 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2533 for (i = 0; i < hdr->NumberOfSections; i++) {
2534 COFF_section* sectab_i
2536 myindex ( sizeof_COFF_section, sectab, i );
2539 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2542 /* Ignore sections called which contain stabs debugging
2544 if (0 == strcmp(".stab", sectab_i->Name)
2545 || 0 == strcmp(".stabstr", sectab_i->Name)
2546 || 0 == strcmp(".ctors", sectab_i->Name))
2549 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2550 /* If the relocation field (a short) has overflowed, the
2551 * real count can be found in the first reloc entry.
2553 * See Section 4.1 (last para) of the PE spec (rev6.0).
2555 * Nov2003 update: the GNU linker still doesn't correctly
2556 * handle the generation of relocatable object files with
2557 * overflown relocations. Hence the output to warn of potential
2560 COFF_reloc* rel = (COFF_reloc*)
2561 myindex ( sizeof_COFF_reloc, reltab, 0 );
2562 noRelocs = rel->VirtualAddress;
2564 /* 10/05: we now assume (and check for) a GNU ld that is capable
2565 * of handling object files with (>2^16) of relocs.
2568 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2573 noRelocs = sectab_i->NumberOfRelocations;
2578 for (; j < noRelocs; j++) {
2580 COFF_reloc* reltab_j
2582 myindex ( sizeof_COFF_reloc, reltab, j );
2584 /* the location to patch */
2586 ((UChar*)(oc->image))
2587 + (sectab_i->PointerToRawData
2588 + reltab_j->VirtualAddress
2589 - sectab_i->VirtualAddress )
2591 /* the existing contents of pP */
2593 /* the symbol to connect to */
2594 sym = (COFF_symbol*)
2595 myindex ( sizeof_COFF_symbol,
2596 symtab, reltab_j->SymbolTableIndex );
2599 "reloc sec %2d num %3d: type 0x%-4x "
2600 "vaddr 0x%-8x name `",
2602 (UInt32)reltab_j->Type,
2603 reltab_j->VirtualAddress );
2604 printName ( sym->Name, strtab );
2605 debugBelch("'\n" ));
2607 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2608 COFF_section* section_sym
2609 = findPEi386SectionCalled ( oc, sym->Name );
2611 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2614 S = ((UInt32)(oc->image))
2615 + (section_sym->PointerToRawData
2618 copyName ( sym->Name, strtab, symbol, 1000-1 );
2619 S = (UInt32) lookupSymbol( symbol );
2620 if ((void*)S != NULL) goto foundit;
2621 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2625 checkProddableBlock(oc, pP);
2626 switch (reltab_j->Type) {
2627 case MYIMAGE_REL_I386_DIR32:
2630 case MYIMAGE_REL_I386_REL32:
2631 /* Tricky. We have to insert a displacement at
2632 pP which, when added to the PC for the _next_
2633 insn, gives the address of the target (S).
2634 Problem is to know the address of the next insn
2635 when we only know pP. We assume that this
2636 literal field is always the last in the insn,
2637 so that the address of the next insn is pP+4
2638 -- hence the constant 4.
2639 Also I don't know if A should be added, but so
2640 far it has always been zero.
2642 SOF 05/2005: 'A' (old contents of *pP) have been observed
2643 to contain values other than zero (the 'wx' object file
2644 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2645 So, add displacement to old value instead of asserting
2646 A to be zero. Fixes wxhaskell-related crashes, and no other
2647 ill effects have been observed.
2649 Update: the reason why we're seeing these more elaborate
2650 relocations is due to a switch in how the NCG compiles SRTs
2651 and offsets to them from info tables. SRTs live in .(ro)data,
2652 while info tables live in .text, causing GAS to emit REL32/DISP32
2653 relocations with non-zero values. Adding the displacement is
2654 the right thing to do.
2656 *pP = S - ((UInt32)pP) - 4 + A;
2659 debugBelch("%s: unhandled PEi386 relocation type %d",
2660 oc->fileName, reltab_j->Type);
2667 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2671 #endif /* defined(OBJFORMAT_PEi386) */
2674 /* --------------------------------------------------------------------------
2676 * ------------------------------------------------------------------------*/
2678 #if defined(OBJFORMAT_ELF)
2683 #if defined(sparc_HOST_ARCH)
2684 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2685 #elif defined(i386_HOST_ARCH)
2686 # define ELF_TARGET_386 /* Used inside <elf.h> */
2687 #elif defined(x86_64_HOST_ARCH)
2688 # define ELF_TARGET_X64_64
2690 #elif defined (ia64_HOST_ARCH)
2691 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2693 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2694 # define ELF_NEED_GOT /* needs Global Offset Table */
2695 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2698 #if !defined(openbsd_HOST_OS)
2701 /* openbsd elf has things in different places, with diff names */
2702 # include <elf_abi.h>
2703 # include <machine/reloc.h>
2704 # define R_386_32 RELOC_32
2705 # define R_386_PC32 RELOC_PC32
2708 /* If elf.h doesn't define it */
2709 # ifndef R_X86_64_PC64
2710 # define R_X86_64_PC64 24
2714 * Define a set of types which can be used for both ELF32 and ELF64
2718 #define ELFCLASS ELFCLASS64
2719 #define Elf_Addr Elf64_Addr
2720 #define Elf_Word Elf64_Word
2721 #define Elf_Sword Elf64_Sword
2722 #define Elf_Ehdr Elf64_Ehdr
2723 #define Elf_Phdr Elf64_Phdr
2724 #define Elf_Shdr Elf64_Shdr
2725 #define Elf_Sym Elf64_Sym
2726 #define Elf_Rel Elf64_Rel
2727 #define Elf_Rela Elf64_Rela
2728 #define ELF_ST_TYPE ELF64_ST_TYPE
2729 #define ELF_ST_BIND ELF64_ST_BIND
2730 #define ELF_R_TYPE ELF64_R_TYPE
2731 #define ELF_R_SYM ELF64_R_SYM
2733 #define ELFCLASS ELFCLASS32
2734 #define Elf_Addr Elf32_Addr
2735 #define Elf_Word Elf32_Word
2736 #define Elf_Sword Elf32_Sword
2737 #define Elf_Ehdr Elf32_Ehdr
2738 #define Elf_Phdr Elf32_Phdr
2739 #define Elf_Shdr Elf32_Shdr
2740 #define Elf_Sym Elf32_Sym
2741 #define Elf_Rel Elf32_Rel
2742 #define Elf_Rela Elf32_Rela
2744 #define ELF_ST_TYPE ELF32_ST_TYPE
2747 #define ELF_ST_BIND ELF32_ST_BIND
2750 #define ELF_R_TYPE ELF32_R_TYPE
2753 #define ELF_R_SYM ELF32_R_SYM
2759 * Functions to allocate entries in dynamic sections. Currently we simply
2760 * preallocate a large number, and we don't check if a entry for the given
2761 * target already exists (a linear search is too slow). Ideally these
2762 * entries would be associated with symbols.
2765 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2766 #define GOT_SIZE 0x20000
2767 #define FUNCTION_TABLE_SIZE 0x10000
2768 #define PLT_SIZE 0x08000
2771 static Elf_Addr got[GOT_SIZE];
2772 static unsigned int gotIndex;
2773 static Elf_Addr gp_val = (Elf_Addr)got;
2776 allocateGOTEntry(Elf_Addr target)
2780 if (gotIndex >= GOT_SIZE)
2781 barf("Global offset table overflow");
2783 entry = &got[gotIndex++];
2785 return (Elf_Addr)entry;
2789 #ifdef ELF_FUNCTION_DESC
2795 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2796 static unsigned int functionTableIndex;
2799 allocateFunctionDesc(Elf_Addr target)
2801 FunctionDesc *entry;
2803 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2804 barf("Function table overflow");
2806 entry = &functionTable[functionTableIndex++];
2808 entry->gp = (Elf_Addr)gp_val;
2809 return (Elf_Addr)entry;
2813 copyFunctionDesc(Elf_Addr target)
2815 FunctionDesc *olddesc = (FunctionDesc *)target;
2816 FunctionDesc *newdesc;
2818 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2819 newdesc->gp = olddesc->gp;
2820 return (Elf_Addr)newdesc;
2825 #ifdef ia64_HOST_ARCH
2826 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2827 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2829 static unsigned char plt_code[] =
2831 /* taken from binutils bfd/elfxx-ia64.c */
2832 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2833 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2834 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2835 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2836 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2837 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2840 /* If we can't get to the function descriptor via gp, take a local copy of it */
2841 #define PLT_RELOC(code, target) { \
2842 Elf64_Sxword rel_value = target - gp_val; \
2843 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2844 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2846 ia64_reloc_gprel22((Elf_Addr)code, target); \
2851 unsigned char code[sizeof(plt_code)];
2855 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2857 PLTEntry *plt = (PLTEntry *)oc->plt;
2860 if (oc->pltIndex >= PLT_SIZE)
2861 barf("Procedure table overflow");
2863 entry = &plt[oc->pltIndex++];
2864 memcpy(entry->code, plt_code, sizeof(entry->code));
2865 PLT_RELOC(entry->code, target);
2866 return (Elf_Addr)entry;
2872 return (PLT_SIZE * sizeof(PLTEntry));
2878 * Generic ELF functions
2882 findElfSection ( void* objImage, Elf_Word sh_type )
2884 char* ehdrC = (char*)objImage;
2885 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2886 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2887 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2891 for (i = 0; i < ehdr->e_shnum; i++) {
2892 if (shdr[i].sh_type == sh_type
2893 /* Ignore the section header's string table. */
2894 && i != ehdr->e_shstrndx
2895 /* Ignore string tables named .stabstr, as they contain
2897 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2899 ptr = ehdrC + shdr[i].sh_offset;
2906 #if defined(ia64_HOST_ARCH)
2908 findElfSegment ( void* objImage, Elf_Addr vaddr )
2910 char* ehdrC = (char*)objImage;
2911 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2912 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2913 Elf_Addr segaddr = 0;
2916 for (i = 0; i < ehdr->e_phnum; i++) {
2917 segaddr = phdr[i].p_vaddr;
2918 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2926 ocVerifyImage_ELF ( ObjectCode* oc )
2930 int i, j, nent, nstrtab, nsymtabs;
2934 char* ehdrC = (char*)(oc->image);
2935 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2937 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2938 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2939 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2940 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2941 errorBelch("%s: not an ELF object", oc->fileName);
2945 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2946 errorBelch("%s: unsupported ELF format", oc->fileName);
2950 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2951 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2953 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2954 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2956 errorBelch("%s: unknown endiannness", oc->fileName);
2960 if (ehdr->e_type != ET_REL) {
2961 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2964 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2966 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2967 switch (ehdr->e_machine) {
2968 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2969 #ifdef EM_SPARC32PLUS
2970 case EM_SPARC32PLUS:
2972 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2974 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2976 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2978 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2979 #elif defined(EM_AMD64)
2980 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
2982 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2983 errorBelch("%s: unknown architecture (e_machine == %d)"
2984 , oc->fileName, ehdr->e_machine);
2988 IF_DEBUG(linker,debugBelch(
2989 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
2990 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2992 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2994 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2996 if (ehdr->e_shstrndx == SHN_UNDEF) {
2997 errorBelch("%s: no section header string table", oc->fileName);
3000 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3002 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3005 for (i = 0; i < ehdr->e_shnum; i++) {
3006 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3007 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3008 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3009 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3010 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3011 ehdrC + shdr[i].sh_offset,
3012 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3014 if (shdr[i].sh_type == SHT_REL) {
3015 IF_DEBUG(linker,debugBelch("Rel " ));
3016 } else if (shdr[i].sh_type == SHT_RELA) {
3017 IF_DEBUG(linker,debugBelch("RelA " ));
3019 IF_DEBUG(linker,debugBelch(" "));
3022 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3026 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3029 for (i = 0; i < ehdr->e_shnum; i++) {
3030 if (shdr[i].sh_type == SHT_STRTAB
3031 /* Ignore the section header's string table. */
3032 && i != ehdr->e_shstrndx
3033 /* Ignore string tables named .stabstr, as they contain
3035 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3037 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3038 strtab = ehdrC + shdr[i].sh_offset;
3043 errorBelch("%s: no string tables, or too many", oc->fileName);
3048 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3049 for (i = 0; i < ehdr->e_shnum; i++) {
3050 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3051 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3053 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3054 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3055 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3057 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3059 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3060 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3063 for (j = 0; j < nent; j++) {
3064 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3065 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3066 (int)stab[j].st_shndx,
3067 (int)stab[j].st_size,
3068 (char*)stab[j].st_value ));
3070 IF_DEBUG(linker,debugBelch("type=" ));
3071 switch (ELF_ST_TYPE(stab[j].st_info)) {
3072 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3073 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3074 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3075 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3076 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3077 default: IF_DEBUG(linker,debugBelch("? " )); break;
3079 IF_DEBUG(linker,debugBelch(" " ));
3081 IF_DEBUG(linker,debugBelch("bind=" ));
3082 switch (ELF_ST_BIND(stab[j].st_info)) {
3083 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3084 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3085 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3086 default: IF_DEBUG(linker,debugBelch("? " )); break;
3088 IF_DEBUG(linker,debugBelch(" " ));
3090 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3094 if (nsymtabs == 0) {
3095 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3102 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3106 if (hdr->sh_type == SHT_PROGBITS
3107 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3108 /* .text-style section */
3109 return SECTIONKIND_CODE_OR_RODATA;
3112 if (hdr->sh_type == SHT_PROGBITS
3113 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3114 /* .data-style section */
3115 return SECTIONKIND_RWDATA;
3118 if (hdr->sh_type == SHT_PROGBITS
3119 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3120 /* .rodata-style section */
3121 return SECTIONKIND_CODE_OR_RODATA;
3124 if (hdr->sh_type == SHT_NOBITS
3125 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3126 /* .bss-style section */
3128 return SECTIONKIND_RWDATA;
3131 return SECTIONKIND_OTHER;
3136 ocGetNames_ELF ( ObjectCode* oc )
3141 char* ehdrC = (char*)(oc->image);
3142 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3143 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3144 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3146 ASSERT(symhash != NULL);
3149 errorBelch("%s: no strtab", oc->fileName);
3154 for (i = 0; i < ehdr->e_shnum; i++) {
3155 /* Figure out what kind of section it is. Logic derived from
3156 Figure 1.14 ("Special Sections") of the ELF document
3157 ("Portable Formats Specification, Version 1.1"). */
3159 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3161 if (is_bss && shdr[i].sh_size > 0) {
3162 /* This is a non-empty .bss section. Allocate zeroed space for
3163 it, and set its .sh_offset field such that
3164 ehdrC + .sh_offset == addr_of_zeroed_space. */
3165 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3166 "ocGetNames_ELF(BSS)");
3167 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3169 debugBelch("BSS section at 0x%x, size %d\n",
3170 zspace, shdr[i].sh_size);
3174 /* fill in the section info */
3175 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3176 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3177 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3178 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3181 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3183 /* copy stuff into this module's object symbol table */
3184 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3185 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3187 oc->n_symbols = nent;
3188 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3189 "ocGetNames_ELF(oc->symbols)");
3191 for (j = 0; j < nent; j++) {
3193 char isLocal = FALSE; /* avoids uninit-var warning */
3195 char* nm = strtab + stab[j].st_name;
3196 int secno = stab[j].st_shndx;
3198 /* Figure out if we want to add it; if so, set ad to its
3199 address. Otherwise leave ad == NULL. */
3201 if (secno == SHN_COMMON) {
3203 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3205 debugBelch("COMMON symbol, size %d name %s\n",
3206 stab[j].st_size, nm);
3208 /* Pointless to do addProddableBlock() for this area,
3209 since the linker should never poke around in it. */
3212 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3213 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3215 /* and not an undefined symbol */
3216 && stab[j].st_shndx != SHN_UNDEF
3217 /* and not in a "special section" */
3218 && stab[j].st_shndx < SHN_LORESERVE
3220 /* and it's a not a section or string table or anything silly */
3221 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3222 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3223 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3226 /* Section 0 is the undefined section, hence > and not >=. */
3227 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3229 if (shdr[secno].sh_type == SHT_NOBITS) {
3230 debugBelch(" BSS symbol, size %d off %d name %s\n",
3231 stab[j].st_size, stab[j].st_value, nm);
3234 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3235 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3238 #ifdef ELF_FUNCTION_DESC
3239 /* dlsym() and the initialisation table both give us function
3240 * descriptors, so to be consistent we store function descriptors
3241 * in the symbol table */
3242 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3243 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3245 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3246 ad, oc->fileName, nm ));
3251 /* And the decision is ... */
3255 oc->symbols[j] = nm;
3258 /* Ignore entirely. */
3260 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3264 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3265 strtab + stab[j].st_name ));
3268 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3269 (int)ELF_ST_BIND(stab[j].st_info),
3270 (int)ELF_ST_TYPE(stab[j].st_info),
3271 (int)stab[j].st_shndx,
3272 strtab + stab[j].st_name
3275 oc->symbols[j] = NULL;
3284 /* Do ELF relocations which lack an explicit addend. All x86-linux
3285 relocations appear to be of this form. */
3287 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3288 Elf_Shdr* shdr, int shnum,
3289 Elf_Sym* stab, char* strtab )
3294 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3295 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3296 int target_shndx = shdr[shnum].sh_info;
3297 int symtab_shndx = shdr[shnum].sh_link;
3299 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3300 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3301 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3302 target_shndx, symtab_shndx ));
3304 /* Skip sections that we're not interested in. */
3307 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3308 if (kind == SECTIONKIND_OTHER) {
3309 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3314 for (j = 0; j < nent; j++) {
3315 Elf_Addr offset = rtab[j].r_offset;
3316 Elf_Addr info = rtab[j].r_info;
3318 Elf_Addr P = ((Elf_Addr)targ) + offset;
3319 Elf_Word* pP = (Elf_Word*)P;
3324 StgStablePtr stablePtr;
3327 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3328 j, (void*)offset, (void*)info ));
3330 IF_DEBUG(linker,debugBelch( " ZERO" ));
3333 Elf_Sym sym = stab[ELF_R_SYM(info)];
3334 /* First see if it is a local symbol. */
3335 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3336 /* Yes, so we can get the address directly from the ELF symbol
3338 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3340 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3341 + stab[ELF_R_SYM(info)].st_value);
3344 symbol = strtab + sym.st_name;
3345 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3346 if (NULL == stablePtr) {
3347 /* No, so look up the name in our global table. */
3348 S_tmp = lookupSymbol( symbol );
3349 S = (Elf_Addr)S_tmp;
3351 stableVal = deRefStablePtr( stablePtr );
3353 S = (Elf_Addr)S_tmp;
3357 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3360 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3363 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3364 (void*)P, (void*)S, (void*)A ));
3365 checkProddableBlock ( oc, pP );
3369 switch (ELF_R_TYPE(info)) {
3370 # ifdef i386_HOST_ARCH
3371 case R_386_32: *pP = value; break;
3372 case R_386_PC32: *pP = value - P; break;
3375 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3376 oc->fileName, (lnat)ELF_R_TYPE(info));
3384 /* Do ELF relocations for which explicit addends are supplied.
3385 sparc-solaris relocations appear to be of this form. */
3387 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3388 Elf_Shdr* shdr, int shnum,
3389 Elf_Sym* stab, char* strtab )
3392 char *symbol = NULL;
3394 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3395 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3396 int target_shndx = shdr[shnum].sh_info;
3397 int symtab_shndx = shdr[shnum].sh_link;
3399 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3400 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3401 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3402 target_shndx, symtab_shndx ));
3404 for (j = 0; j < nent; j++) {
3405 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3406 /* This #ifdef only serves to avoid unused-var warnings. */
3407 Elf_Addr offset = rtab[j].r_offset;
3408 Elf_Addr P = targ + offset;
3410 Elf_Addr info = rtab[j].r_info;
3411 Elf_Addr A = rtab[j].r_addend;
3415 # if defined(sparc_HOST_ARCH)
3416 Elf_Word* pP = (Elf_Word*)P;
3418 # elif defined(ia64_HOST_ARCH)
3419 Elf64_Xword *pP = (Elf64_Xword *)P;
3421 # elif defined(powerpc_HOST_ARCH)
3425 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3426 j, (void*)offset, (void*)info,
3429 IF_DEBUG(linker,debugBelch( " ZERO" ));
3432 Elf_Sym sym = stab[ELF_R_SYM(info)];
3433 /* First see if it is a local symbol. */
3434 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3435 /* Yes, so we can get the address directly from the ELF symbol
3437 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3439 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3440 + stab[ELF_R_SYM(info)].st_value);
3441 #ifdef ELF_FUNCTION_DESC
3442 /* Make a function descriptor for this function */
3443 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3444 S = allocateFunctionDesc(S + A);
3449 /* No, so look up the name in our global table. */
3450 symbol = strtab + sym.st_name;
3451 S_tmp = lookupSymbol( symbol );
3452 S = (Elf_Addr)S_tmp;
3454 #ifdef ELF_FUNCTION_DESC
3455 /* If a function, already a function descriptor - we would
3456 have to copy it to add an offset. */
3457 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3458 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3462 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3465 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3468 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3469 (void*)P, (void*)S, (void*)A ));
3470 /* checkProddableBlock ( oc, (void*)P ); */
3474 switch (ELF_R_TYPE(info)) {
3475 # if defined(sparc_HOST_ARCH)
3476 case R_SPARC_WDISP30:
3477 w1 = *pP & 0xC0000000;
3478 w2 = (Elf_Word)((value - P) >> 2);
3479 ASSERT((w2 & 0xC0000000) == 0);
3484 w1 = *pP & 0xFFC00000;
3485 w2 = (Elf_Word)(value >> 10);
3486 ASSERT((w2 & 0xFFC00000) == 0);
3492 w2 = (Elf_Word)(value & 0x3FF);
3493 ASSERT((w2 & ~0x3FF) == 0);
3498 /* According to the Sun documentation:
3500 This relocation type resembles R_SPARC_32, except it refers to an
3501 unaligned word. That is, the word to be relocated must be treated
3502 as four separate bytes with arbitrary alignment, not as a word
3503 aligned according to the architecture requirements.
3506 w2 = (Elf_Word)value;
3508 // SPARC doesn't do misaligned writes of 32 bit words,
3509 // so we have to do this one byte-at-a-time.
3510 char *pPc = (char*)pP;
3511 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3512 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3513 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3514 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3518 w2 = (Elf_Word)value;
3521 # elif defined(ia64_HOST_ARCH)
3522 case R_IA64_DIR64LSB:
3523 case R_IA64_FPTR64LSB:
3526 case R_IA64_PCREL64LSB:
3529 case R_IA64_SEGREL64LSB:
3530 addr = findElfSegment(ehdrC, value);
3533 case R_IA64_GPREL22:
3534 ia64_reloc_gprel22(P, value);
3536 case R_IA64_LTOFF22:
3537 case R_IA64_LTOFF22X:
3538 case R_IA64_LTOFF_FPTR22:
3539 addr = allocateGOTEntry(value);
3540 ia64_reloc_gprel22(P, addr);
3542 case R_IA64_PCREL21B:
3543 ia64_reloc_pcrel21(P, S, oc);
3546 /* This goes with R_IA64_LTOFF22X and points to the load to
3547 * convert into a move. We don't implement relaxation. */
3549 # elif defined(powerpc_HOST_ARCH)
3550 case R_PPC_ADDR16_LO:
3551 *(Elf32_Half*) P = value;
3554 case R_PPC_ADDR16_HI:
3555 *(Elf32_Half*) P = value >> 16;
3558 case R_PPC_ADDR16_HA:
3559 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3563 *(Elf32_Word *) P = value;
3567 *(Elf32_Word *) P = value - P;
3573 if( delta << 6 >> 6 != delta )
3575 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3579 if( value == 0 || delta << 6 >> 6 != delta )
3581 barf( "Unable to make SymbolExtra for #%d",
3587 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3588 | (delta & 0x3fffffc);
3592 #if x86_64_HOST_ARCH
3594 *(Elf64_Xword *)P = value;
3599 StgInt64 off = value - P;
3600 if (off >= 0x7fffffffL || off < -0x80000000L) {
3601 #if X86_64_ELF_NONPIC_HACK
3602 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3604 off = pltAddress + A - P;
3606 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3607 symbol, off, oc->fileName );
3610 *(Elf64_Word *)P = (Elf64_Word)off;
3616 StgInt64 off = value - P;
3617 *(Elf64_Word *)P = (Elf64_Word)off;
3622 if (value >= 0x7fffffffL) {
3623 #if X86_64_ELF_NONPIC_HACK
3624 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3626 value = pltAddress + A;
3628 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3629 symbol, value, oc->fileName );
3632 *(Elf64_Word *)P = (Elf64_Word)value;
3636 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3637 #if X86_64_ELF_NONPIC_HACK
3638 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3640 value = pltAddress + A;
3642 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3643 symbol, value, oc->fileName );
3646 *(Elf64_Sword *)P = (Elf64_Sword)value;
3649 case R_X86_64_GOTPCREL:
3651 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3652 StgInt64 off = gotAddress + A - P;
3653 *(Elf64_Word *)P = (Elf64_Word)off;
3657 case R_X86_64_PLT32:
3659 StgInt64 off = value - P;
3660 if (off >= 0x7fffffffL || off < -0x80000000L) {
3661 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3663 off = pltAddress + A - P;
3665 *(Elf64_Word *)P = (Elf64_Word)off;
3671 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3672 oc->fileName, (lnat)ELF_R_TYPE(info));
3681 ocResolve_ELF ( ObjectCode* oc )
3685 Elf_Sym* stab = NULL;
3686 char* ehdrC = (char*)(oc->image);
3687 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3688 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3690 /* first find "the" symbol table */
3691 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3693 /* also go find the string table */
3694 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3696 if (stab == NULL || strtab == NULL) {
3697 errorBelch("%s: can't find string or symbol table", oc->fileName);
3701 /* Process the relocation sections. */
3702 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3703 if (shdr[shnum].sh_type == SHT_REL) {
3704 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3705 shnum, stab, strtab );
3709 if (shdr[shnum].sh_type == SHT_RELA) {
3710 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3711 shnum, stab, strtab );
3716 #if defined(powerpc_HOST_ARCH)
3717 ocFlushInstructionCache( oc );
3725 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3726 * at the front. The following utility functions pack and unpack instructions, and
3727 * take care of the most common relocations.
3730 #ifdef ia64_HOST_ARCH
3733 ia64_extract_instruction(Elf64_Xword *target)
3736 int slot = (Elf_Addr)target & 3;
3737 target = (Elf_Addr)target & ~3;
3745 return ((w1 >> 5) & 0x1ffffffffff);
3747 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3751 barf("ia64_extract_instruction: invalid slot %p", target);
3756 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3758 int slot = (Elf_Addr)target & 3;
3759 target = (Elf_Addr)target & ~3;
3764 *target |= value << 5;
3767 *target |= value << 46;
3768 *(target+1) |= value >> 18;
3771 *(target+1) |= value << 23;
3777 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3779 Elf64_Xword instruction;
3780 Elf64_Sxword rel_value;
3782 rel_value = value - gp_val;
3783 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3784 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3786 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3787 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3788 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3789 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3790 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3791 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3795 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3797 Elf64_Xword instruction;
3798 Elf64_Sxword rel_value;
3801 entry = allocatePLTEntry(value, oc);
3803 rel_value = (entry >> 4) - (target >> 4);
3804 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3805 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3807 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3808 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3809 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3810 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3816 * PowerPC & X86_64 ELF specifics
3819 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3821 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3827 ehdr = (Elf_Ehdr *) oc->image;
3828 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3830 for( i = 0; i < ehdr->e_shnum; i++ )
3831 if( shdr[i].sh_type == SHT_SYMTAB )
3834 if( i == ehdr->e_shnum )
3836 errorBelch( "This ELF file contains no symtab" );
3840 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3842 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3843 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3848 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3851 #endif /* powerpc */
3855 /* --------------------------------------------------------------------------
3857 * ------------------------------------------------------------------------*/
3859 #if defined(OBJFORMAT_MACHO)
3862 Support for MachO linking on Darwin/MacOS X
3863 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3865 I hereby formally apologize for the hackish nature of this code.
3866 Things that need to be done:
3867 *) implement ocVerifyImage_MachO
3868 *) add still more sanity checks.
3871 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3872 #define mach_header mach_header_64
3873 #define segment_command segment_command_64
3874 #define section section_64
3875 #define nlist nlist_64
3878 #ifdef powerpc_HOST_ARCH
3879 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3881 struct mach_header *header = (struct mach_header *) oc->image;
3882 struct load_command *lc = (struct load_command *) (header + 1);
3885 for( i = 0; i < header->ncmds; i++ )
3887 if( lc->cmd == LC_SYMTAB )
3889 // Find out the first and last undefined external
3890 // symbol, so we don't have to allocate too many
3892 struct symtab_command *symLC = (struct symtab_command *) lc;
3893 unsigned min = symLC->nsyms, max = 0;
3894 struct nlist *nlist =
3895 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3897 for(i=0;i<symLC->nsyms;i++)
3899 if(nlist[i].n_type & N_STAB)
3901 else if(nlist[i].n_type & N_EXT)
3903 if((nlist[i].n_type & N_TYPE) == N_UNDF
3904 && (nlist[i].n_value == 0))
3914 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3919 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3921 return ocAllocateSymbolExtras(oc,0,0);
3924 #ifdef x86_64_HOST_ARCH
3925 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3927 struct mach_header *header = (struct mach_header *) oc->image;
3928 struct load_command *lc = (struct load_command *) (header + 1);
3931 for( i = 0; i < header->ncmds; i++ )
3933 if( lc->cmd == LC_SYMTAB )
3935 // Just allocate one entry for every symbol
3936 struct symtab_command *symLC = (struct symtab_command *) lc;
3938 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3941 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3943 return ocAllocateSymbolExtras(oc,0,0);
3947 static int ocVerifyImage_MachO(ObjectCode* oc)
3949 char *image = (char*) oc->image;
3950 struct mach_header *header = (struct mach_header*) image;
3952 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3953 if(header->magic != MH_MAGIC_64)
3956 if(header->magic != MH_MAGIC)
3959 // FIXME: do some more verifying here
3963 static int resolveImports(
3966 struct symtab_command *symLC,
3967 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3968 unsigned long *indirectSyms,
3969 struct nlist *nlist)
3972 size_t itemSize = 4;
3975 int isJumpTable = 0;
3976 if(!strcmp(sect->sectname,"__jump_table"))
3980 ASSERT(sect->reserved2 == itemSize);
3984 for(i=0; i*itemSize < sect->size;i++)
3986 // according to otool, reserved1 contains the first index into the indirect symbol table
3987 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3988 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3991 if((symbol->n_type & N_TYPE) == N_UNDF
3992 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3993 addr = (void*) (symbol->n_value);
3995 addr = lookupSymbol(nm);
3998 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4006 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4007 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4008 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4009 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4014 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4015 ((void**)(image + sect->offset))[i] = addr;
4022 static unsigned long relocateAddress(
4025 struct section* sections,
4026 unsigned long address)
4029 for(i = 0; i < nSections; i++)
4031 if(sections[i].addr <= address
4032 && address < sections[i].addr + sections[i].size)
4034 return (unsigned long)oc->image
4035 + sections[i].offset + address - sections[i].addr;
4038 barf("Invalid Mach-O file:"
4039 "Address out of bounds while relocating object file");
4043 static int relocateSection(
4046 struct symtab_command *symLC, struct nlist *nlist,
4047 int nSections, struct section* sections, struct section *sect)
4049 struct relocation_info *relocs;
4052 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4054 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4056 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4058 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4062 relocs = (struct relocation_info*) (image + sect->reloff);
4066 #ifdef x86_64_HOST_ARCH
4067 struct relocation_info *reloc = &relocs[i];
4069 char *thingPtr = image + sect->offset + reloc->r_address;
4071 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4072 complains that it may be used uninitialized if we don't */
4075 int type = reloc->r_type;
4077 checkProddableBlock(oc,thingPtr);
4078 switch(reloc->r_length)
4081 thing = *(uint8_t*)thingPtr;
4082 baseValue = (uint64_t)thingPtr + 1;
4085 thing = *(uint16_t*)thingPtr;
4086 baseValue = (uint64_t)thingPtr + 2;
4089 thing = *(uint32_t*)thingPtr;
4090 baseValue = (uint64_t)thingPtr + 4;
4093 thing = *(uint64_t*)thingPtr;
4094 baseValue = (uint64_t)thingPtr + 8;
4097 barf("Unknown size.");
4100 if(type == X86_64_RELOC_GOT
4101 || type == X86_64_RELOC_GOT_LOAD)
4103 ASSERT(reloc->r_extern);
4104 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4106 type = X86_64_RELOC_SIGNED;
4108 else if(reloc->r_extern)
4110 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4111 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4112 if(symbol->n_value == 0)
4113 value = (uint64_t) lookupSymbol(nm);
4115 value = relocateAddress(oc, nSections, sections,
4120 value = sections[reloc->r_symbolnum-1].offset
4121 - sections[reloc->r_symbolnum-1].addr
4125 if(type == X86_64_RELOC_BRANCH)
4127 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4129 ASSERT(reloc->r_extern);
4130 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4133 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4134 type = X86_64_RELOC_SIGNED;
4139 case X86_64_RELOC_UNSIGNED:
4140 ASSERT(!reloc->r_pcrel);
4143 case X86_64_RELOC_SIGNED:
4144 ASSERT(reloc->r_pcrel);
4145 thing += value - baseValue;
4147 case X86_64_RELOC_SUBTRACTOR:
4148 ASSERT(!reloc->r_pcrel);
4152 barf("unkown relocation");
4155 switch(reloc->r_length)
4158 *(uint8_t*)thingPtr = thing;
4161 *(uint16_t*)thingPtr = thing;
4164 *(uint32_t*)thingPtr = thing;
4167 *(uint64_t*)thingPtr = thing;
4171 if(relocs[i].r_address & R_SCATTERED)
4173 struct scattered_relocation_info *scat =
4174 (struct scattered_relocation_info*) &relocs[i];
4178 if(scat->r_length == 2)
4180 unsigned long word = 0;
4181 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4182 checkProddableBlock(oc,wordPtr);
4184 // Note on relocation types:
4185 // i386 uses the GENERIC_RELOC_* types,
4186 // while ppc uses special PPC_RELOC_* types.
4187 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4188 // in both cases, all others are different.
4189 // Therefore, we use GENERIC_RELOC_VANILLA
4190 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4191 // and use #ifdefs for the other types.
4193 // Step 1: Figure out what the relocated value should be
4194 if(scat->r_type == GENERIC_RELOC_VANILLA)
4196 word = *wordPtr + (unsigned long) relocateAddress(
4203 #ifdef powerpc_HOST_ARCH
4204 else if(scat->r_type == PPC_RELOC_SECTDIFF
4205 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4206 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4207 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4209 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4212 struct scattered_relocation_info *pair =
4213 (struct scattered_relocation_info*) &relocs[i+1];
4215 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4216 barf("Invalid Mach-O file: "
4217 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4219 word = (unsigned long)
4220 (relocateAddress(oc, nSections, sections, scat->r_value)
4221 - relocateAddress(oc, nSections, sections, pair->r_value));
4224 #ifdef powerpc_HOST_ARCH
4225 else if(scat->r_type == PPC_RELOC_HI16
4226 || scat->r_type == PPC_RELOC_LO16
4227 || scat->r_type == PPC_RELOC_HA16
4228 || scat->r_type == PPC_RELOC_LO14)
4229 { // these are generated by label+offset things
4230 struct relocation_info *pair = &relocs[i+1];
4231 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4232 barf("Invalid Mach-O file: "
4233 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4235 if(scat->r_type == PPC_RELOC_LO16)
4237 word = ((unsigned short*) wordPtr)[1];
4238 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4240 else if(scat->r_type == PPC_RELOC_LO14)
4242 barf("Unsupported Relocation: PPC_RELOC_LO14");
4243 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4244 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4246 else if(scat->r_type == PPC_RELOC_HI16)
4248 word = ((unsigned short*) wordPtr)[1] << 16;
4249 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4251 else if(scat->r_type == PPC_RELOC_HA16)
4253 word = ((unsigned short*) wordPtr)[1] << 16;
4254 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4258 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4265 continue; // ignore the others
4267 #ifdef powerpc_HOST_ARCH
4268 if(scat->r_type == GENERIC_RELOC_VANILLA
4269 || scat->r_type == PPC_RELOC_SECTDIFF)
4271 if(scat->r_type == GENERIC_RELOC_VANILLA
4272 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4277 #ifdef powerpc_HOST_ARCH
4278 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4280 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4282 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4284 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4286 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4288 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4289 + ((word & (1<<15)) ? 1 : 0);
4295 continue; // FIXME: I hope it's OK to ignore all the others.
4299 struct relocation_info *reloc = &relocs[i];
4300 if(reloc->r_pcrel && !reloc->r_extern)
4303 if(reloc->r_length == 2)
4305 unsigned long word = 0;
4306 #ifdef powerpc_HOST_ARCH
4307 unsigned long jumpIsland = 0;
4308 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4309 // to avoid warning and to catch
4313 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4314 checkProddableBlock(oc,wordPtr);
4316 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4320 #ifdef powerpc_HOST_ARCH
4321 else if(reloc->r_type == PPC_RELOC_LO16)
4323 word = ((unsigned short*) wordPtr)[1];
4324 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4326 else if(reloc->r_type == PPC_RELOC_HI16)
4328 word = ((unsigned short*) wordPtr)[1] << 16;
4329 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4331 else if(reloc->r_type == PPC_RELOC_HA16)
4333 word = ((unsigned short*) wordPtr)[1] << 16;
4334 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4336 else if(reloc->r_type == PPC_RELOC_BR24)
4339 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4343 if(!reloc->r_extern)
4346 sections[reloc->r_symbolnum-1].offset
4347 - sections[reloc->r_symbolnum-1].addr
4354 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4355 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4356 void *symbolAddress = lookupSymbol(nm);
4359 errorBelch("\nunknown symbol `%s'", nm);
4365 #ifdef powerpc_HOST_ARCH
4366 // In the .o file, this should be a relative jump to NULL
4367 // and we'll change it to a relative jump to the symbol
4368 ASSERT(word + reloc->r_address == 0);
4369 jumpIsland = (unsigned long)
4370 &makeSymbolExtra(oc,
4372 (unsigned long) symbolAddress)
4376 offsetToJumpIsland = word + jumpIsland
4377 - (((long)image) + sect->offset - sect->addr);
4380 word += (unsigned long) symbolAddress
4381 - (((long)image) + sect->offset - sect->addr);
4385 word += (unsigned long) symbolAddress;
4389 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4394 #ifdef powerpc_HOST_ARCH
4395 else if(reloc->r_type == PPC_RELOC_LO16)
4397 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4400 else if(reloc->r_type == PPC_RELOC_HI16)
4402 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4405 else if(reloc->r_type == PPC_RELOC_HA16)
4407 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4408 + ((word & (1<<15)) ? 1 : 0);
4411 else if(reloc->r_type == PPC_RELOC_BR24)
4413 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4415 // The branch offset is too large.
4416 // Therefore, we try to use a jump island.
4419 barf("unconditional relative branch out of range: "
4420 "no jump island available");
4423 word = offsetToJumpIsland;
4424 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4425 barf("unconditional relative branch out of range: "
4426 "jump island out of range");
4428 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4433 barf("\nunknown relocation %d",reloc->r_type);
4441 static int ocGetNames_MachO(ObjectCode* oc)
4443 char *image = (char*) oc->image;
4444 struct mach_header *header = (struct mach_header*) image;
4445 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4446 unsigned i,curSymbol = 0;
4447 struct segment_command *segLC = NULL;
4448 struct section *sections;
4449 struct symtab_command *symLC = NULL;
4450 struct nlist *nlist;
4451 unsigned long commonSize = 0;
4452 char *commonStorage = NULL;
4453 unsigned long commonCounter;
4455 for(i=0;i<header->ncmds;i++)
4457 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4458 segLC = (struct segment_command*) lc;
4459 else if(lc->cmd == LC_SYMTAB)
4460 symLC = (struct symtab_command*) lc;
4461 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4464 sections = (struct section*) (segLC+1);
4465 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4469 barf("ocGetNames_MachO: no segment load command");
4471 for(i=0;i<segLC->nsects;i++)
4473 if(sections[i].size == 0)
4476 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4478 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4479 "ocGetNames_MachO(common symbols)");
4480 sections[i].offset = zeroFillArea - image;
4483 if(!strcmp(sections[i].sectname,"__text"))
4484 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4485 (void*) (image + sections[i].offset),
4486 (void*) (image + sections[i].offset + sections[i].size));
4487 else if(!strcmp(sections[i].sectname,"__const"))
4488 addSection(oc, SECTIONKIND_RWDATA,
4489 (void*) (image + sections[i].offset),
4490 (void*) (image + sections[i].offset + sections[i].size));
4491 else if(!strcmp(sections[i].sectname,"__data"))
4492 addSection(oc, SECTIONKIND_RWDATA,
4493 (void*) (image + sections[i].offset),
4494 (void*) (image + sections[i].offset + sections[i].size));
4495 else if(!strcmp(sections[i].sectname,"__bss")
4496 || !strcmp(sections[i].sectname,"__common"))
4497 addSection(oc, SECTIONKIND_RWDATA,
4498 (void*) (image + sections[i].offset),
4499 (void*) (image + sections[i].offset + sections[i].size));
4501 addProddableBlock(oc, (void*) (image + sections[i].offset),
4505 // count external symbols defined here
4509 for(i=0;i<symLC->nsyms;i++)
4511 if(nlist[i].n_type & N_STAB)
4513 else if(nlist[i].n_type & N_EXT)
4515 if((nlist[i].n_type & N_TYPE) == N_UNDF
4516 && (nlist[i].n_value != 0))
4518 commonSize += nlist[i].n_value;
4521 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4526 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4527 "ocGetNames_MachO(oc->symbols)");
4531 for(i=0;i<symLC->nsyms;i++)
4533 if(nlist[i].n_type & N_STAB)
4535 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4537 if(nlist[i].n_type & N_EXT)
4539 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4540 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4541 ; // weak definition, and we already have a definition
4544 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4546 + sections[nlist[i].n_sect-1].offset
4547 - sections[nlist[i].n_sect-1].addr
4548 + nlist[i].n_value);
4549 oc->symbols[curSymbol++] = nm;
4556 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4557 commonCounter = (unsigned long)commonStorage;
4560 for(i=0;i<symLC->nsyms;i++)
4562 if((nlist[i].n_type & N_TYPE) == N_UNDF
4563 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4565 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4566 unsigned long sz = nlist[i].n_value;
4568 nlist[i].n_value = commonCounter;
4570 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4571 (void*)commonCounter);
4572 oc->symbols[curSymbol++] = nm;
4574 commonCounter += sz;
4581 static int ocResolve_MachO(ObjectCode* oc)
4583 char *image = (char*) oc->image;
4584 struct mach_header *header = (struct mach_header*) image;
4585 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4587 struct segment_command *segLC = NULL;
4588 struct section *sections;
4589 struct symtab_command *symLC = NULL;
4590 struct dysymtab_command *dsymLC = NULL;
4591 struct nlist *nlist;
4593 for(i=0;i<header->ncmds;i++)
4595 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4596 segLC = (struct segment_command*) lc;
4597 else if(lc->cmd == LC_SYMTAB)
4598 symLC = (struct symtab_command*) lc;
4599 else if(lc->cmd == LC_DYSYMTAB)
4600 dsymLC = (struct dysymtab_command*) lc;
4601 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4604 sections = (struct section*) (segLC+1);
4605 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4610 unsigned long *indirectSyms
4611 = (unsigned long*) (image + dsymLC->indirectsymoff);
4613 for(i=0;i<segLC->nsects;i++)
4615 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4616 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4617 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4619 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4622 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4623 || !strcmp(sections[i].sectname,"__pointers"))
4625 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4628 else if(!strcmp(sections[i].sectname,"__jump_table"))
4630 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4636 for(i=0;i<segLC->nsects;i++)
4638 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4642 #if defined (powerpc_HOST_ARCH)
4643 ocFlushInstructionCache( oc );
4649 #ifdef powerpc_HOST_ARCH
4651 * The Mach-O object format uses leading underscores. But not everywhere.
4652 * There is a small number of runtime support functions defined in
4653 * libcc_dynamic.a whose name does not have a leading underscore.
4654 * As a consequence, we can't get their address from C code.
4655 * We have to use inline assembler just to take the address of a function.
4659 static void machoInitSymbolsWithoutUnderscore()
4661 extern void* symbolsWithoutUnderscore[];
4662 void **p = symbolsWithoutUnderscore;
4663 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4665 #undef SymI_NeedsProto
4666 #define SymI_NeedsProto(x) \
4667 __asm__ volatile(".long " # x);
4669 RTS_MACHO_NOUNDERLINE_SYMBOLS
4671 __asm__ volatile(".text");
4673 #undef SymI_NeedsProto
4674 #define SymI_NeedsProto(x) \
4675 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4677 RTS_MACHO_NOUNDERLINE_SYMBOLS
4679 #undef SymI_NeedsProto
4684 * Figure out by how much to shift the entire Mach-O file in memory
4685 * when loading so that its single segment ends up 16-byte-aligned
4687 static int machoGetMisalignment( FILE * f )
4689 struct mach_header header;
4692 fread(&header, sizeof(header), 1, f);
4695 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4696 if(header.magic != MH_MAGIC_64)
4699 if(header.magic != MH_MAGIC)
4703 misalignment = (header.sizeofcmds + sizeof(header))
4706 return misalignment ? (16 - misalignment) : 0;