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_HasProto(strcmp) \
385 SymI_HasProto(memmove) \
386 SymI_HasProto(realloc) \
387 SymI_HasProto(malloc) \
389 SymI_HasProto(tanh) \
390 SymI_HasProto(cosh) \
391 SymI_HasProto(sinh) \
392 SymI_HasProto(atan) \
393 SymI_HasProto(acos) \
394 SymI_HasProto(asin) \
400 SymI_HasProto(sqrt) \
401 SymI_HasProto(powf) \
402 SymI_HasProto(tanhf) \
403 SymI_HasProto(coshf) \
404 SymI_HasProto(sinhf) \
405 SymI_HasProto(atanf) \
406 SymI_HasProto(acosf) \
407 SymI_HasProto(asinf) \
408 SymI_HasProto(tanf) \
409 SymI_HasProto(cosf) \
410 SymI_HasProto(sinf) \
411 SymI_HasProto(expf) \
412 SymI_HasProto(logf) \
413 SymI_HasProto(sqrtf) \
414 SymI_HasProto(memcpy) \
415 SymI_HasProto(rts_InstallConsoleEvent) \
416 SymI_HasProto(rts_ConsoleHandlerDone) \
417 SymI_NeedsProto(mktime) \
418 SymI_NeedsProto(_imp___timezone) \
419 SymI_NeedsProto(_imp___tzname) \
420 SymI_NeedsProto(_imp__tzname) \
421 SymI_NeedsProto(_imp___iob) \
422 SymI_NeedsProto(_imp___osver) \
423 SymI_NeedsProto(localtime) \
424 SymI_NeedsProto(gmtime) \
425 SymI_NeedsProto(opendir) \
426 SymI_NeedsProto(readdir) \
427 SymI_NeedsProto(rewinddir) \
428 RTS_MINGW_EXTRA_SYMS \
429 RTS_MINGW_GETTIMEOFDAY_SYM \
430 SymI_NeedsProto(closedir)
433 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
434 #define RTS_DARWIN_ONLY_SYMBOLS \
435 SymI_NeedsProto(asprintf$LDBLStub) \
436 SymI_NeedsProto(err$LDBLStub) \
437 SymI_NeedsProto(errc$LDBLStub) \
438 SymI_NeedsProto(errx$LDBLStub) \
439 SymI_NeedsProto(fprintf$LDBLStub) \
440 SymI_NeedsProto(fscanf$LDBLStub) \
441 SymI_NeedsProto(fwprintf$LDBLStub) \
442 SymI_NeedsProto(fwscanf$LDBLStub) \
443 SymI_NeedsProto(printf$LDBLStub) \
444 SymI_NeedsProto(scanf$LDBLStub) \
445 SymI_NeedsProto(snprintf$LDBLStub) \
446 SymI_NeedsProto(sprintf$LDBLStub) \
447 SymI_NeedsProto(sscanf$LDBLStub) \
448 SymI_NeedsProto(strtold$LDBLStub) \
449 SymI_NeedsProto(swprintf$LDBLStub) \
450 SymI_NeedsProto(swscanf$LDBLStub) \
451 SymI_NeedsProto(syslog$LDBLStub) \
452 SymI_NeedsProto(vasprintf$LDBLStub) \
453 SymI_NeedsProto(verr$LDBLStub) \
454 SymI_NeedsProto(verrc$LDBLStub) \
455 SymI_NeedsProto(verrx$LDBLStub) \
456 SymI_NeedsProto(vfprintf$LDBLStub) \
457 SymI_NeedsProto(vfscanf$LDBLStub) \
458 SymI_NeedsProto(vfwprintf$LDBLStub) \
459 SymI_NeedsProto(vfwscanf$LDBLStub) \
460 SymI_NeedsProto(vprintf$LDBLStub) \
461 SymI_NeedsProto(vscanf$LDBLStub) \
462 SymI_NeedsProto(vsnprintf$LDBLStub) \
463 SymI_NeedsProto(vsprintf$LDBLStub) \
464 SymI_NeedsProto(vsscanf$LDBLStub) \
465 SymI_NeedsProto(vswprintf$LDBLStub) \
466 SymI_NeedsProto(vswscanf$LDBLStub) \
467 SymI_NeedsProto(vsyslog$LDBLStub) \
468 SymI_NeedsProto(vwarn$LDBLStub) \
469 SymI_NeedsProto(vwarnc$LDBLStub) \
470 SymI_NeedsProto(vwarnx$LDBLStub) \
471 SymI_NeedsProto(vwprintf$LDBLStub) \
472 SymI_NeedsProto(vwscanf$LDBLStub) \
473 SymI_NeedsProto(warn$LDBLStub) \
474 SymI_NeedsProto(warnc$LDBLStub) \
475 SymI_NeedsProto(warnx$LDBLStub) \
476 SymI_NeedsProto(wcstold$LDBLStub) \
477 SymI_NeedsProto(wprintf$LDBLStub) \
478 SymI_NeedsProto(wscanf$LDBLStub)
480 #define RTS_DARWIN_ONLY_SYMBOLS
484 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
486 # define MAIN_CAP_SYM
489 #if !defined(mingw32_HOST_OS)
490 #define RTS_USER_SIGNALS_SYMBOLS \
491 SymI_HasProto(setIOManagerPipe) \
492 SymI_NeedsProto(blockUserSignals) \
493 SymI_NeedsProto(unblockUserSignals)
495 #define RTS_USER_SIGNALS_SYMBOLS \
496 SymI_HasProto(sendIOManagerEvent) \
497 SymI_HasProto(readIOManagerEvent) \
498 SymI_HasProto(getIOManagerEvent) \
499 SymI_HasProto(console_handler)
502 #define RTS_LIBFFI_SYMBOLS \
503 SymE_NeedsProto(ffi_prep_cif) \
504 SymE_NeedsProto(ffi_call) \
505 SymE_NeedsProto(ffi_type_void) \
506 SymE_NeedsProto(ffi_type_float) \
507 SymE_NeedsProto(ffi_type_double) \
508 SymE_NeedsProto(ffi_type_sint64) \
509 SymE_NeedsProto(ffi_type_uint64) \
510 SymE_NeedsProto(ffi_type_sint32) \
511 SymE_NeedsProto(ffi_type_uint32) \
512 SymE_NeedsProto(ffi_type_sint16) \
513 SymE_NeedsProto(ffi_type_uint16) \
514 SymE_NeedsProto(ffi_type_sint8) \
515 SymE_NeedsProto(ffi_type_uint8) \
516 SymE_NeedsProto(ffi_type_pointer)
518 #ifdef TABLES_NEXT_TO_CODE
519 #define RTS_RET_SYMBOLS /* nothing */
521 #define RTS_RET_SYMBOLS \
522 SymI_HasProto(stg_enter_ret) \
523 SymI_HasProto(stg_gc_fun_ret) \
524 SymI_HasProto(stg_ap_v_ret) \
525 SymI_HasProto(stg_ap_f_ret) \
526 SymI_HasProto(stg_ap_d_ret) \
527 SymI_HasProto(stg_ap_l_ret) \
528 SymI_HasProto(stg_ap_n_ret) \
529 SymI_HasProto(stg_ap_p_ret) \
530 SymI_HasProto(stg_ap_pv_ret) \
531 SymI_HasProto(stg_ap_pp_ret) \
532 SymI_HasProto(stg_ap_ppv_ret) \
533 SymI_HasProto(stg_ap_ppp_ret) \
534 SymI_HasProto(stg_ap_pppv_ret) \
535 SymI_HasProto(stg_ap_pppp_ret) \
536 SymI_HasProto(stg_ap_ppppp_ret) \
537 SymI_HasProto(stg_ap_pppppp_ret)
540 #define RTS_SYMBOLS \
542 SymI_HasProto(StgReturn) \
543 SymI_HasProto(stg_enter_info) \
544 SymI_HasProto(stg_gc_void_info) \
545 SymI_HasProto(__stg_gc_enter_1) \
546 SymI_HasProto(stg_gc_noregs) \
547 SymI_HasProto(stg_gc_unpt_r1_info) \
548 SymI_HasProto(stg_gc_unpt_r1) \
549 SymI_HasProto(stg_gc_unbx_r1_info) \
550 SymI_HasProto(stg_gc_unbx_r1) \
551 SymI_HasProto(stg_gc_f1_info) \
552 SymI_HasProto(stg_gc_f1) \
553 SymI_HasProto(stg_gc_d1_info) \
554 SymI_HasProto(stg_gc_d1) \
555 SymI_HasProto(stg_gc_l1_info) \
556 SymI_HasProto(stg_gc_l1) \
557 SymI_HasProto(__stg_gc_fun) \
558 SymI_HasProto(stg_gc_fun_info) \
559 SymI_HasProto(stg_gc_gen) \
560 SymI_HasProto(stg_gc_gen_info) \
561 SymI_HasProto(stg_gc_gen_hp) \
562 SymI_HasProto(stg_gc_ut) \
563 SymI_HasProto(stg_gen_yield) \
564 SymI_HasProto(stg_yield_noregs) \
565 SymI_HasProto(stg_yield_to_interpreter) \
566 SymI_HasProto(stg_gen_block) \
567 SymI_HasProto(stg_block_noregs) \
568 SymI_HasProto(stg_block_1) \
569 SymI_HasProto(stg_block_takemvar) \
570 SymI_HasProto(stg_block_putmvar) \
572 SymI_HasProto(MallocFailHook) \
573 SymI_HasProto(OnExitHook) \
574 SymI_HasProto(OutOfHeapHook) \
575 SymI_HasProto(StackOverflowHook) \
576 SymI_HasProto(addDLL) \
577 SymI_HasProto(__int_encodeDouble) \
578 SymI_HasProto(__word_encodeDouble) \
579 SymI_HasProto(__2Int_encodeDouble) \
580 SymI_HasProto(__int_encodeFloat) \
581 SymI_HasProto(__word_encodeFloat) \
582 SymI_HasProto(atomicallyzh_fast) \
583 SymI_HasProto(barf) \
584 SymI_HasProto(debugBelch) \
585 SymI_HasProto(errorBelch) \
586 SymI_HasProto(sysErrorBelch) \
587 SymI_HasProto(asyncExceptionsBlockedzh_fast) \
588 SymI_HasProto(blockAsyncExceptionszh_fast) \
589 SymI_HasProto(catchzh_fast) \
590 SymI_HasProto(catchRetryzh_fast) \
591 SymI_HasProto(catchSTMzh_fast) \
592 SymI_HasProto(checkzh_fast) \
593 SymI_HasProto(closure_flags) \
594 SymI_HasProto(cmp_thread) \
595 SymI_HasProto(createAdjustor) \
596 SymI_HasProto(decodeDoublezu2Intzh_fast) \
597 SymI_HasProto(decodeFloatzuIntzh_fast) \
598 SymI_HasProto(defaultsHook) \
599 SymI_HasProto(delayzh_fast) \
600 SymI_HasProto(deRefWeakzh_fast) \
601 SymI_HasProto(deRefStablePtrzh_fast) \
602 SymI_HasProto(dirty_MUT_VAR) \
603 SymI_HasProto(forkzh_fast) \
604 SymI_HasProto(forkOnzh_fast) \
605 SymI_HasProto(forkProcess) \
606 SymI_HasProto(forkOS_createThread) \
607 SymI_HasProto(freeHaskellFunctionPtr) \
608 SymI_HasProto(freeStablePtr) \
609 SymI_HasProto(getOrSetTypeableStore) \
610 SymI_HasProto(getOrSetSignalHandlerStore) \
611 SymI_HasProto(genSymZh) \
612 SymI_HasProto(genericRaise) \
613 SymI_HasProto(getProgArgv) \
614 SymI_HasProto(getFullProgArgv) \
615 SymI_HasProto(getStablePtr) \
616 SymI_HasProto(hs_init) \
617 SymI_HasProto(hs_exit) \
618 SymI_HasProto(hs_set_argv) \
619 SymI_HasProto(hs_add_root) \
620 SymI_HasProto(hs_perform_gc) \
621 SymI_HasProto(hs_free_stable_ptr) \
622 SymI_HasProto(hs_free_fun_ptr) \
623 SymI_HasProto(hs_hpc_rootModule) \
624 SymI_HasProto(hs_hpc_module) \
625 SymI_HasProto(initLinker) \
626 SymI_HasProto(unpackClosurezh_fast) \
627 SymI_HasProto(getApStackValzh_fast) \
628 SymI_HasProto(getSparkzh_fast) \
629 SymI_HasProto(isCurrentThreadBoundzh_fast) \
630 SymI_HasProto(isDoubleDenormalized) \
631 SymI_HasProto(isDoubleInfinite) \
632 SymI_HasProto(isDoubleNaN) \
633 SymI_HasProto(isDoubleNegativeZero) \
634 SymI_HasProto(isEmptyMVarzh_fast) \
635 SymI_HasProto(isFloatDenormalized) \
636 SymI_HasProto(isFloatInfinite) \
637 SymI_HasProto(isFloatNaN) \
638 SymI_HasProto(isFloatNegativeZero) \
639 SymI_HasProto(killThreadzh_fast) \
640 SymI_HasProto(loadObj) \
641 SymI_HasProto(insertStableSymbol) \
642 SymI_HasProto(insertSymbol) \
643 SymI_HasProto(lookupSymbol) \
644 SymI_HasProto(makeStablePtrzh_fast) \
645 SymI_HasProto(mkApUpd0zh_fast) \
646 SymI_HasProto(myThreadIdzh_fast) \
647 SymI_HasProto(labelThreadzh_fast) \
648 SymI_HasProto(newArrayzh_fast) \
649 SymI_HasProto(newBCOzh_fast) \
650 SymI_HasProto(newByteArrayzh_fast) \
651 SymI_HasProto_redirect(newCAF, newDynCAF) \
652 SymI_HasProto(newMVarzh_fast) \
653 SymI_HasProto(newMutVarzh_fast) \
654 SymI_HasProto(newTVarzh_fast) \
655 SymI_HasProto(noDuplicatezh_fast) \
656 SymI_HasProto(atomicModifyMutVarzh_fast) \
657 SymI_HasProto(newPinnedByteArrayzh_fast) \
658 SymI_HasProto(newAlignedPinnedByteArrayzh_fast) \
659 SymI_HasProto(newSpark) \
660 SymI_HasProto(performGC) \
661 SymI_HasProto(performMajorGC) \
662 SymI_HasProto(prog_argc) \
663 SymI_HasProto(prog_argv) \
664 SymI_HasProto(putMVarzh_fast) \
665 SymI_HasProto(raisezh_fast) \
666 SymI_HasProto(raiseIOzh_fast) \
667 SymI_HasProto(readTVarzh_fast) \
668 SymI_HasProto(readTVarIOzh_fast) \
669 SymI_HasProto(resetNonBlockingFd) \
670 SymI_HasProto(resumeThread) \
671 SymI_HasProto(resolveObjs) \
672 SymI_HasProto(retryzh_fast) \
673 SymI_HasProto(rts_apply) \
674 SymI_HasProto(rts_checkSchedStatus) \
675 SymI_HasProto(rts_eval) \
676 SymI_HasProto(rts_evalIO) \
677 SymI_HasProto(rts_evalLazyIO) \
678 SymI_HasProto(rts_evalStableIO) \
679 SymI_HasProto(rts_eval_) \
680 SymI_HasProto(rts_getBool) \
681 SymI_HasProto(rts_getChar) \
682 SymI_HasProto(rts_getDouble) \
683 SymI_HasProto(rts_getFloat) \
684 SymI_HasProto(rts_getInt) \
685 SymI_HasProto(rts_getInt8) \
686 SymI_HasProto(rts_getInt16) \
687 SymI_HasProto(rts_getInt32) \
688 SymI_HasProto(rts_getInt64) \
689 SymI_HasProto(rts_getPtr) \
690 SymI_HasProto(rts_getFunPtr) \
691 SymI_HasProto(rts_getStablePtr) \
692 SymI_HasProto(rts_getThreadId) \
693 SymI_HasProto(rts_getWord) \
694 SymI_HasProto(rts_getWord8) \
695 SymI_HasProto(rts_getWord16) \
696 SymI_HasProto(rts_getWord32) \
697 SymI_HasProto(rts_getWord64) \
698 SymI_HasProto(rts_lock) \
699 SymI_HasProto(rts_mkBool) \
700 SymI_HasProto(rts_mkChar) \
701 SymI_HasProto(rts_mkDouble) \
702 SymI_HasProto(rts_mkFloat) \
703 SymI_HasProto(rts_mkInt) \
704 SymI_HasProto(rts_mkInt8) \
705 SymI_HasProto(rts_mkInt16) \
706 SymI_HasProto(rts_mkInt32) \
707 SymI_HasProto(rts_mkInt64) \
708 SymI_HasProto(rts_mkPtr) \
709 SymI_HasProto(rts_mkFunPtr) \
710 SymI_HasProto(rts_mkStablePtr) \
711 SymI_HasProto(rts_mkString) \
712 SymI_HasProto(rts_mkWord) \
713 SymI_HasProto(rts_mkWord8) \
714 SymI_HasProto(rts_mkWord16) \
715 SymI_HasProto(rts_mkWord32) \
716 SymI_HasProto(rts_mkWord64) \
717 SymI_HasProto(rts_unlock) \
718 SymI_HasProto(rts_unsafeGetMyCapability) \
719 SymI_HasProto(rtsSupportsBoundThreads) \
720 SymI_HasProto(__hscore_get_saved_termios) \
721 SymI_HasProto(__hscore_set_saved_termios) \
722 SymI_HasProto(setProgArgv) \
723 SymI_HasProto(startupHaskell) \
724 SymI_HasProto(shutdownHaskell) \
725 SymI_HasProto(shutdownHaskellAndExit) \
726 SymI_HasProto(stable_ptr_table) \
727 SymI_HasProto(stackOverflow) \
728 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
729 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
730 SymI_HasProto(awakenBlockedQueue) \
731 SymI_HasProto(startTimer) \
732 SymI_HasProto(stg_CHARLIKE_closure) \
733 SymI_HasProto(stg_MVAR_CLEAN_info) \
734 SymI_HasProto(stg_MVAR_DIRTY_info) \
735 SymI_HasProto(stg_IND_STATIC_info) \
736 SymI_HasProto(stg_INTLIKE_closure) \
737 SymI_HasProto(stg_ARR_WORDS_info) \
738 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
739 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
740 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
741 SymI_HasProto(stg_WEAK_info) \
742 SymI_HasProto(stg_ap_v_info) \
743 SymI_HasProto(stg_ap_f_info) \
744 SymI_HasProto(stg_ap_d_info) \
745 SymI_HasProto(stg_ap_l_info) \
746 SymI_HasProto(stg_ap_n_info) \
747 SymI_HasProto(stg_ap_p_info) \
748 SymI_HasProto(stg_ap_pv_info) \
749 SymI_HasProto(stg_ap_pp_info) \
750 SymI_HasProto(stg_ap_ppv_info) \
751 SymI_HasProto(stg_ap_ppp_info) \
752 SymI_HasProto(stg_ap_pppv_info) \
753 SymI_HasProto(stg_ap_pppp_info) \
754 SymI_HasProto(stg_ap_ppppp_info) \
755 SymI_HasProto(stg_ap_pppppp_info) \
756 SymI_HasProto(stg_ap_0_fast) \
757 SymI_HasProto(stg_ap_v_fast) \
758 SymI_HasProto(stg_ap_f_fast) \
759 SymI_HasProto(stg_ap_d_fast) \
760 SymI_HasProto(stg_ap_l_fast) \
761 SymI_HasProto(stg_ap_n_fast) \
762 SymI_HasProto(stg_ap_p_fast) \
763 SymI_HasProto(stg_ap_pv_fast) \
764 SymI_HasProto(stg_ap_pp_fast) \
765 SymI_HasProto(stg_ap_ppv_fast) \
766 SymI_HasProto(stg_ap_ppp_fast) \
767 SymI_HasProto(stg_ap_pppv_fast) \
768 SymI_HasProto(stg_ap_pppp_fast) \
769 SymI_HasProto(stg_ap_ppppp_fast) \
770 SymI_HasProto(stg_ap_pppppp_fast) \
771 SymI_HasProto(stg_ap_1_upd_info) \
772 SymI_HasProto(stg_ap_2_upd_info) \
773 SymI_HasProto(stg_ap_3_upd_info) \
774 SymI_HasProto(stg_ap_4_upd_info) \
775 SymI_HasProto(stg_ap_5_upd_info) \
776 SymI_HasProto(stg_ap_6_upd_info) \
777 SymI_HasProto(stg_ap_7_upd_info) \
778 SymI_HasProto(stg_exit) \
779 SymI_HasProto(stg_sel_0_upd_info) \
780 SymI_HasProto(stg_sel_10_upd_info) \
781 SymI_HasProto(stg_sel_11_upd_info) \
782 SymI_HasProto(stg_sel_12_upd_info) \
783 SymI_HasProto(stg_sel_13_upd_info) \
784 SymI_HasProto(stg_sel_14_upd_info) \
785 SymI_HasProto(stg_sel_15_upd_info) \
786 SymI_HasProto(stg_sel_1_upd_info) \
787 SymI_HasProto(stg_sel_2_upd_info) \
788 SymI_HasProto(stg_sel_3_upd_info) \
789 SymI_HasProto(stg_sel_4_upd_info) \
790 SymI_HasProto(stg_sel_5_upd_info) \
791 SymI_HasProto(stg_sel_6_upd_info) \
792 SymI_HasProto(stg_sel_7_upd_info) \
793 SymI_HasProto(stg_sel_8_upd_info) \
794 SymI_HasProto(stg_sel_9_upd_info) \
795 SymI_HasProto(stg_upd_frame_info) \
796 SymI_HasProto(suspendThread) \
797 SymI_HasProto(takeMVarzh_fast) \
798 SymI_HasProto(threadStatuszh_fast) \
799 SymI_HasProto(tryPutMVarzh_fast) \
800 SymI_HasProto(tryTakeMVarzh_fast) \
801 SymI_HasProto(unblockAsyncExceptionszh_fast) \
802 SymI_HasProto(unloadObj) \
803 SymI_HasProto(unsafeThawArrayzh_fast) \
804 SymI_HasProto(waitReadzh_fast) \
805 SymI_HasProto(waitWritezh_fast) \
806 SymI_HasProto(writeTVarzh_fast) \
807 SymI_HasProto(yieldzh_fast) \
808 SymI_NeedsProto(stg_interp_constr_entry) \
809 SymI_HasProto(alloc_blocks) \
810 SymI_HasProto(alloc_blocks_lim) \
811 SymI_HasProto(allocateLocal) \
812 SymI_HasProto(allocateExec) \
813 SymI_HasProto(freeExec) \
814 SymI_HasProto(getAllocations) \
815 SymI_HasProto(revertCAFs) \
816 SymI_HasProto(RtsFlags) \
817 SymI_NeedsProto(rts_breakpoint_io_action) \
818 SymI_NeedsProto(rts_stop_next_breakpoint) \
819 SymI_NeedsProto(rts_stop_on_exception) \
820 SymI_HasProto(stopTimer) \
821 SymI_HasProto(n_capabilities) \
822 SymI_HasProto(traceCcszh_fast) \
823 RTS_USER_SIGNALS_SYMBOLS
826 // 64-bit support functions in libgcc.a
827 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
828 #define RTS_LIBGCC_SYMBOLS \
829 SymI_NeedsProto(__divdi3) \
830 SymI_NeedsProto(__udivdi3) \
831 SymI_NeedsProto(__moddi3) \
832 SymI_NeedsProto(__umoddi3) \
833 SymI_NeedsProto(__muldi3) \
834 SymI_NeedsProto(__ashldi3) \
835 SymI_NeedsProto(__ashrdi3) \
836 SymI_NeedsProto(__lshrdi3) \
837 SymI_NeedsProto(__eprintf)
838 #elif defined(ia64_HOST_ARCH)
839 #define RTS_LIBGCC_SYMBOLS \
840 SymI_NeedsProto(__divdi3) \
841 SymI_NeedsProto(__udivdi3) \
842 SymI_NeedsProto(__moddi3) \
843 SymI_NeedsProto(__umoddi3) \
844 SymI_NeedsProto(__divsf3) \
845 SymI_NeedsProto(__divdf3)
847 #define RTS_LIBGCC_SYMBOLS
850 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
851 // Symbols that don't have a leading underscore
852 // on Mac OS X. They have to receive special treatment,
853 // see machoInitSymbolsWithoutUnderscore()
854 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
855 SymI_NeedsProto(saveFP) \
856 SymI_NeedsProto(restFP)
859 /* entirely bogus claims about types of these symbols */
860 #define SymI_NeedsProto(vvv) extern void vvv(void);
861 #if defined(__PIC__) && defined(mingw32_TARGET_OS)
862 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
863 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
865 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
866 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
868 #define SymI_HasProto(vvv) /**/
869 #define SymI_HasProto_redirect(vvv,xxx) /**/
872 RTS_POSIX_ONLY_SYMBOLS
873 RTS_MINGW_ONLY_SYMBOLS
874 RTS_CYGWIN_ONLY_SYMBOLS
875 RTS_DARWIN_ONLY_SYMBOLS
878 #undef SymI_NeedsProto
880 #undef SymI_HasProto_redirect
882 #undef SymE_NeedsProto
884 #ifdef LEADING_UNDERSCORE
885 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
887 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
890 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
892 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
893 (void*)DLL_IMPORT_DATA_REF(vvv) },
895 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
896 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
898 // SymI_HasProto_redirect allows us to redirect references to one symbol to
899 // another symbol. See newCAF/newDynCAF for an example.
900 #define SymI_HasProto_redirect(vvv,xxx) \
901 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
904 static RtsSymbolVal rtsSyms[] = {
907 RTS_POSIX_ONLY_SYMBOLS
908 RTS_MINGW_ONLY_SYMBOLS
909 RTS_CYGWIN_ONLY_SYMBOLS
910 RTS_DARWIN_ONLY_SYMBOLS
913 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
914 // dyld stub code contains references to this,
915 // but it should never be called because we treat
916 // lazy pointers as nonlazy.
917 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
919 { 0, 0 } /* sentinel */
924 /* -----------------------------------------------------------------------------
925 * Insert symbols into hash tables, checking for duplicates.
928 static void ghciInsertStrHashTable ( char* obj_name,
934 if (lookupHashTable(table, (StgWord)key) == NULL)
936 insertStrHashTable(table, (StgWord)key, data);
941 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
943 "whilst processing object file\n"
945 "This could be caused by:\n"
946 " * Loading two different object files which export the same symbol\n"
947 " * Specifying the same object file twice on the GHCi command line\n"
948 " * An incorrect `package.conf' entry, causing some object to be\n"
950 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
957 /* -----------------------------------------------------------------------------
958 * initialize the object linker
962 static int linker_init_done = 0 ;
964 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
965 static void *dl_prog_handle;
973 /* Make initLinker idempotent, so we can call it
974 before evey relevant operation; that means we
975 don't need to initialise the linker separately */
976 if (linker_init_done == 1) { return; } else {
977 linker_init_done = 1;
980 stablehash = allocStrHashTable();
981 symhash = allocStrHashTable();
983 /* populate the symbol table with stuff from the RTS */
984 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
985 ghciInsertStrHashTable("(GHCi built-in symbols)",
986 symhash, sym->lbl, sym->addr);
988 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
989 machoInitSymbolsWithoutUnderscore();
992 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
993 # if defined(RTLD_DEFAULT)
994 dl_prog_handle = RTLD_DEFAULT;
996 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
997 # endif /* RTLD_DEFAULT */
1000 #if defined(x86_64_HOST_ARCH)
1001 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1002 // User-override for mmap_32bit_base
1003 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1007 #if defined(mingw32_HOST_OS)
1009 * These two libraries cause problems when added to the static link,
1010 * but are necessary for resolving symbols in GHCi, hence we load
1011 * them manually here.
1018 /* -----------------------------------------------------------------------------
1019 * Loading DLL or .so dynamic libraries
1020 * -----------------------------------------------------------------------------
1022 * Add a DLL from which symbols may be found. In the ELF case, just
1023 * do RTLD_GLOBAL-style add, so no further messing around needs to
1024 * happen in order that symbols in the loaded .so are findable --
1025 * lookupSymbol() will subsequently see them by dlsym on the program's
1026 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1028 * In the PEi386 case, open the DLLs and put handles to them in a
1029 * linked list. When looking for a symbol, try all handles in the
1030 * list. This means that we need to load even DLLs that are guaranteed
1031 * to be in the ghc.exe image already, just so we can get a handle
1032 * to give to loadSymbol, so that we can find the symbols. For such
1033 * libraries, the LoadLibrary call should be a no-op except for returning
1038 #if defined(OBJFORMAT_PEi386)
1039 /* A record for storing handles into DLLs. */
1044 struct _OpenedDLL* next;
1049 /* A list thereof. */
1050 static OpenedDLL* opened_dlls = NULL;
1054 addDLL( char *dll_name )
1056 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1057 /* ------------------- ELF DLL loader ------------------- */
1063 // omitted: RTLD_NOW
1064 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1065 hdl= dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1068 /* dlopen failed; return a ptr to the error msg. */
1070 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1077 # elif defined(OBJFORMAT_PEi386)
1078 /* ------------------- Win32 DLL loader ------------------- */
1086 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1088 /* See if we've already got it, and ignore if so. */
1089 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1090 if (0 == strcmp(o_dll->name, dll_name))
1094 /* The file name has no suffix (yet) so that we can try
1095 both foo.dll and foo.drv
1097 The documentation for LoadLibrary says:
1098 If no file name extension is specified in the lpFileName
1099 parameter, the default library extension .dll is
1100 appended. However, the file name string can include a trailing
1101 point character (.) to indicate that the module name has no
1104 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1105 sprintf(buf, "%s.DLL", dll_name);
1106 instance = LoadLibrary(buf);
1107 if (instance == NULL) {
1108 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1109 // KAA: allow loading of drivers (like winspool.drv)
1110 sprintf(buf, "%s.DRV", dll_name);
1111 instance = LoadLibrary(buf);
1112 if (instance == NULL) {
1113 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1114 // #1883: allow loading of unix-style libfoo.dll DLLs
1115 sprintf(buf, "lib%s.DLL", dll_name);
1116 instance = LoadLibrary(buf);
1117 if (instance == NULL) {
1124 /* Add this DLL to the list of DLLs in which to search for symbols. */
1125 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1126 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1127 strcpy(o_dll->name, dll_name);
1128 o_dll->instance = instance;
1129 o_dll->next = opened_dlls;
1130 opened_dlls = o_dll;
1136 sysErrorBelch(dll_name);
1138 /* LoadLibrary failed; return a ptr to the error msg. */
1139 return "addDLL: could not load DLL";
1142 barf("addDLL: not implemented on this platform");
1146 /* -----------------------------------------------------------------------------
1147 * insert a stable symbol in the hash table
1151 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1153 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1157 /* -----------------------------------------------------------------------------
1158 * insert a symbol in the hash table
1161 insertSymbol(char* obj_name, char* key, void* data)
1163 ghciInsertStrHashTable(obj_name, symhash, key, data);
1166 /* -----------------------------------------------------------------------------
1167 * lookup a symbol in the hash table
1170 lookupSymbol( char *lbl )
1174 ASSERT(symhash != NULL);
1175 val = lookupStrHashTable(symhash, lbl);
1178 # if defined(OBJFORMAT_ELF)
1179 return dlsym(dl_prog_handle, lbl);
1180 # elif defined(OBJFORMAT_MACHO)
1182 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1185 HACK: On OS X, global symbols are prefixed with an underscore.
1186 However, dlsym wants us to omit the leading underscore from the
1187 symbol name. For now, we simply strip it off here (and ONLY
1190 ASSERT(lbl[0] == '_');
1191 return dlsym(dl_prog_handle, lbl+1);
1193 if(NSIsSymbolNameDefined(lbl)) {
1194 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1195 return NSAddressOfSymbol(symbol);
1199 # endif /* HAVE_DLFCN_H */
1200 # elif defined(OBJFORMAT_PEi386)
1203 sym = lookupSymbolInDLLs(lbl);
1204 if (sym != NULL) { return sym; };
1206 // Also try looking up the symbol without the @N suffix. Some
1207 // DLLs have the suffixes on their symbols, some don't.
1208 zapTrailingAtSign ( lbl );
1209 sym = lookupSymbolInDLLs(lbl);
1210 if (sym != NULL) { return sym; };
1222 /* -----------------------------------------------------------------------------
1223 * Debugging aid: look in GHCi's object symbol tables for symbols
1224 * within DELTA bytes of the specified address, and show their names.
1227 void ghci_enquire ( char* addr );
1229 void ghci_enquire ( char* addr )
1234 const int DELTA = 64;
1239 for (oc = objects; oc; oc = oc->next) {
1240 for (i = 0; i < oc->n_symbols; i++) {
1241 sym = oc->symbols[i];
1242 if (sym == NULL) continue;
1245 a = lookupStrHashTable(symhash, sym);
1248 // debugBelch("ghci_enquire: can't find %s\n", sym);
1250 else if (addr-DELTA <= a && a <= addr+DELTA) {
1251 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1258 #ifdef ia64_HOST_ARCH
1259 static unsigned int PLTSize(void);
1263 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1266 mmapForLinker (size_t bytes, nat flags, int fd)
1268 void *map_addr = NULL;
1271 static nat fixed = 0;
1273 pagesize = getpagesize();
1274 size = ROUND_UP(bytes, pagesize);
1276 #if defined(x86_64_HOST_ARCH)
1279 if (mmap_32bit_base != 0) {
1280 map_addr = mmap_32bit_base;
1284 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1285 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1287 if (result == MAP_FAILED) {
1288 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1289 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1290 stg_exit(EXIT_FAILURE);
1293 #if defined(x86_64_HOST_ARCH)
1294 if (mmap_32bit_base != 0) {
1295 if (result == map_addr) {
1296 mmap_32bit_base = map_addr + size;
1298 if ((W_)result > 0x80000000) {
1299 // oops, we were given memory over 2Gb
1300 #if defined(freebsd_HOST_OS)
1301 // Some platforms require MAP_FIXED. This is normally
1302 // a bad idea, because MAP_FIXED will overwrite
1303 // existing mappings.
1304 munmap(result,size);
1308 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);
1311 // hmm, we were given memory somewhere else, but it's
1312 // still under 2Gb so we can use it. Next time, ask
1313 // for memory right after the place we just got some
1314 mmap_32bit_base = (void*)result + size;
1318 if ((W_)result > 0x80000000) {
1319 // oops, we were given memory over 2Gb
1320 // ... try allocating memory somewhere else?;
1321 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1322 munmap(result, size);
1324 // Set a base address and try again... (guess: 1Gb)
1325 mmap_32bit_base = (void*)0x40000000;
1335 /* -----------------------------------------------------------------------------
1336 * Load an obj (populate the global symbol table, but don't resolve yet)
1338 * Returns: 1 if ok, 0 on error.
1341 loadObj( char *path )
1353 /* debugBelch("loadObj %s\n", path ); */
1355 /* Check that we haven't already loaded this object.
1356 Ignore requests to load multiple times */
1360 for (o = objects; o; o = o->next) {
1361 if (0 == strcmp(o->fileName, path)) {
1363 break; /* don't need to search further */
1367 IF_DEBUG(linker, debugBelch(
1368 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1369 "same object file twice:\n"
1371 "GHCi will ignore this, but be warned.\n"
1373 return 1; /* success */
1377 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1379 # if defined(OBJFORMAT_ELF)
1380 oc->formatName = "ELF";
1381 # elif defined(OBJFORMAT_PEi386)
1382 oc->formatName = "PEi386";
1383 # elif defined(OBJFORMAT_MACHO)
1384 oc->formatName = "Mach-O";
1387 barf("loadObj: not implemented on this platform");
1390 r = stat(path, &st);
1391 if (r == -1) { return 0; }
1393 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1394 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1395 strcpy(oc->fileName, path);
1397 oc->fileSize = st.st_size;
1399 oc->sections = NULL;
1400 oc->proddables = NULL;
1402 /* chain it onto the list of objects */
1407 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1409 #if defined(openbsd_HOST_OS)
1410 fd = open(path, O_RDONLY, S_IRUSR);
1412 fd = open(path, O_RDONLY);
1415 barf("loadObj: can't open `%s'", path);
1417 #ifdef ia64_HOST_ARCH
1418 /* The PLT needs to be right before the object */
1421 pagesize = getpagesize();
1422 n = ROUND_UP(PLTSize(), pagesize);
1423 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1424 if (oc->plt == MAP_FAILED)
1425 barf("loadObj: can't allocate PLT");
1428 map_addr = oc->plt + n;
1430 n = ROUND_UP(oc->fileSize, pagesize);
1431 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1432 MAP_PRIVATE|TRY_MAP_32BIT, fd, 0);
1433 if (oc->image == MAP_FAILED)
1434 barf("loadObj: can't map `%s'", path);
1437 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1442 #else /* !USE_MMAP */
1443 /* load the image into memory */
1444 f = fopen(path, "rb");
1446 barf("loadObj: can't read `%s'", path);
1448 # if defined(mingw32_HOST_OS)
1449 // TODO: We would like to use allocateExec here, but allocateExec
1450 // cannot currently allocate blocks large enough.
1451 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1452 PAGE_EXECUTE_READWRITE);
1453 # elif defined(darwin_HOST_OS)
1454 // In a Mach-O .o file, all sections can and will be misaligned
1455 // if the total size of the headers is not a multiple of the
1456 // desired alignment. This is fine for .o files that only serve
1457 // as input for the static linker, but it's not fine for us,
1458 // as SSE (used by gcc for floating point) and Altivec require
1459 // 16-byte alignment.
1460 // We calculate the correct alignment from the header before
1461 // reading the file, and then we misalign oc->image on purpose so
1462 // that the actual sections end up aligned again.
1463 oc->misalignment = machoGetMisalignment(f);
1464 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1465 oc->image += oc->misalignment;
1467 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1472 n = fread ( oc->image, 1, oc->fileSize, f );
1473 if (n != oc->fileSize)
1474 barf("loadObj: error whilst reading `%s'", path);
1477 #endif /* USE_MMAP */
1479 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1480 r = ocAllocateSymbolExtras_MachO ( oc );
1481 if (!r) { return r; }
1482 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1483 r = ocAllocateSymbolExtras_ELF ( oc );
1484 if (!r) { return r; }
1487 /* verify the in-memory image */
1488 # if defined(OBJFORMAT_ELF)
1489 r = ocVerifyImage_ELF ( oc );
1490 # elif defined(OBJFORMAT_PEi386)
1491 r = ocVerifyImage_PEi386 ( oc );
1492 # elif defined(OBJFORMAT_MACHO)
1493 r = ocVerifyImage_MachO ( oc );
1495 barf("loadObj: no verify method");
1497 if (!r) { return r; }
1499 /* build the symbol list for this image */
1500 # if defined(OBJFORMAT_ELF)
1501 r = ocGetNames_ELF ( oc );
1502 # elif defined(OBJFORMAT_PEi386)
1503 r = ocGetNames_PEi386 ( oc );
1504 # elif defined(OBJFORMAT_MACHO)
1505 r = ocGetNames_MachO ( oc );
1507 barf("loadObj: no getNames method");
1509 if (!r) { return r; }
1511 /* loaded, but not resolved yet */
1512 oc->status = OBJECT_LOADED;
1517 /* -----------------------------------------------------------------------------
1518 * resolve all the currently unlinked objects in memory
1520 * Returns: 1 if ok, 0 on error.
1530 for (oc = objects; oc; oc = oc->next) {
1531 if (oc->status != OBJECT_RESOLVED) {
1532 # if defined(OBJFORMAT_ELF)
1533 r = ocResolve_ELF ( oc );
1534 # elif defined(OBJFORMAT_PEi386)
1535 r = ocResolve_PEi386 ( oc );
1536 # elif defined(OBJFORMAT_MACHO)
1537 r = ocResolve_MachO ( oc );
1539 barf("resolveObjs: not implemented on this platform");
1541 if (!r) { return r; }
1542 oc->status = OBJECT_RESOLVED;
1548 /* -----------------------------------------------------------------------------
1549 * delete an object from the pool
1552 unloadObj( char *path )
1554 ObjectCode *oc, *prev;
1556 ASSERT(symhash != NULL);
1557 ASSERT(objects != NULL);
1562 for (oc = objects; oc; prev = oc, oc = oc->next) {
1563 if (!strcmp(oc->fileName,path)) {
1565 /* Remove all the mappings for the symbols within this
1570 for (i = 0; i < oc->n_symbols; i++) {
1571 if (oc->symbols[i] != NULL) {
1572 removeStrHashTable(symhash, oc->symbols[i], NULL);
1580 prev->next = oc->next;
1583 // We're going to leave this in place, in case there are
1584 // any pointers from the heap into it:
1585 // #ifdef mingw32_HOST_OS
1586 // VirtualFree(oc->image);
1588 // stgFree(oc->image);
1590 stgFree(oc->fileName);
1591 stgFree(oc->symbols);
1592 stgFree(oc->sections);
1598 errorBelch("unloadObj: can't find `%s' to unload", path);
1602 /* -----------------------------------------------------------------------------
1603 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1604 * which may be prodded during relocation, and abort if we try and write
1605 * outside any of these.
1607 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1610 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1611 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1615 pb->next = oc->proddables;
1616 oc->proddables = pb;
1619 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1622 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1623 char* s = (char*)(pb->start);
1624 char* e = s + pb->size - 1;
1625 char* a = (char*)addr;
1626 /* Assumes that the biggest fixup involves a 4-byte write. This
1627 probably needs to be changed to 8 (ie, +7) on 64-bit
1629 if (a >= s && (a+3) <= e) return;
1631 barf("checkProddableBlock: invalid fixup in runtime linker");
1634 /* -----------------------------------------------------------------------------
1635 * Section management.
1637 static void addSection ( ObjectCode* oc, SectionKind kind,
1638 void* start, void* end )
1640 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1644 s->next = oc->sections;
1647 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1648 start, ((char*)end)-1, end - start + 1, kind );
1653 /* --------------------------------------------------------------------------
1655 * This is about allocating a small chunk of memory for every symbol in the
1656 * object file. We make sure that the SymboLExtras are always "in range" of
1657 * limited-range PC-relative instructions on various platforms by allocating
1658 * them right next to the object code itself.
1661 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1664 ocAllocateSymbolExtras
1666 Allocate additional space at the end of the object file image to make room
1667 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1669 PowerPC relative branch instructions have a 24 bit displacement field.
1670 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1671 If a particular imported symbol is outside this range, we have to redirect
1672 the jump to a short piece of new code that just loads the 32bit absolute
1673 address and jumps there.
1674 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1677 This function just allocates space for one SymbolExtra for every
1678 undefined symbol in the object file. The code for the jump islands is
1679 filled in by makeSymbolExtra below.
1682 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1689 int misalignment = 0;
1690 #ifdef darwin_HOST_OS
1691 misalignment = oc->misalignment;
1697 // round up to the nearest 4
1698 aligned = (oc->fileSize + 3) & ~3;
1701 pagesize = getpagesize();
1702 n = ROUND_UP( oc->fileSize, pagesize );
1703 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1705 /* we try to use spare space at the end of the last page of the
1706 * image for the jump islands, but if there isn't enough space
1707 * then we have to map some (anonymously, remembering MAP_32BIT).
1709 if( m > n ) // we need to allocate more pages
1711 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1716 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1719 oc->image -= misalignment;
1720 oc->image = stgReallocBytes( oc->image,
1722 aligned + sizeof (SymbolExtra) * count,
1723 "ocAllocateSymbolExtras" );
1724 oc->image += misalignment;
1726 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1727 #endif /* USE_MMAP */
1729 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1732 oc->symbol_extras = NULL;
1734 oc->first_symbol_extra = first;
1735 oc->n_symbol_extras = count;
1740 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1741 unsigned long symbolNumber,
1742 unsigned long target )
1746 ASSERT( symbolNumber >= oc->first_symbol_extra
1747 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1749 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1751 #ifdef powerpc_HOST_ARCH
1752 // lis r12, hi16(target)
1753 extra->jumpIsland.lis_r12 = 0x3d80;
1754 extra->jumpIsland.hi_addr = target >> 16;
1756 // ori r12, r12, lo16(target)
1757 extra->jumpIsland.ori_r12_r12 = 0x618c;
1758 extra->jumpIsland.lo_addr = target & 0xffff;
1761 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1764 extra->jumpIsland.bctr = 0x4e800420;
1766 #ifdef x86_64_HOST_ARCH
1768 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1769 extra->addr = target;
1770 memcpy(extra->jumpIsland, jmp, 6);
1778 /* --------------------------------------------------------------------------
1779 * PowerPC specifics (instruction cache flushing)
1780 * ------------------------------------------------------------------------*/
1782 #ifdef powerpc_TARGET_ARCH
1784 ocFlushInstructionCache
1786 Flush the data & instruction caches.
1787 Because the PPC has split data/instruction caches, we have to
1788 do that whenever we modify code at runtime.
1791 static void ocFlushInstructionCache( ObjectCode *oc )
1793 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1794 unsigned long *p = (unsigned long *) oc->image;
1798 __asm__ volatile ( "dcbf 0,%0\n\t"
1806 __asm__ volatile ( "sync\n\t"
1812 /* --------------------------------------------------------------------------
1813 * PEi386 specifics (Win32 targets)
1814 * ------------------------------------------------------------------------*/
1816 /* The information for this linker comes from
1817 Microsoft Portable Executable
1818 and Common Object File Format Specification
1819 revision 5.1 January 1998
1820 which SimonM says comes from the MS Developer Network CDs.
1822 It can be found there (on older CDs), but can also be found
1825 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1827 (this is Rev 6.0 from February 1999).
1829 Things move, so if that fails, try searching for it via
1831 http://www.google.com/search?q=PE+COFF+specification
1833 The ultimate reference for the PE format is the Winnt.h
1834 header file that comes with the Platform SDKs; as always,
1835 implementations will drift wrt their documentation.
1837 A good background article on the PE format is Matt Pietrek's
1838 March 1994 article in Microsoft System Journal (MSJ)
1839 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1840 Win32 Portable Executable File Format." The info in there
1841 has recently been updated in a two part article in
1842 MSDN magazine, issues Feb and March 2002,
1843 "Inside Windows: An In-Depth Look into the Win32 Portable
1844 Executable File Format"
1846 John Levine's book "Linkers and Loaders" contains useful
1851 #if defined(OBJFORMAT_PEi386)
1855 typedef unsigned char UChar;
1856 typedef unsigned short UInt16;
1857 typedef unsigned int UInt32;
1864 UInt16 NumberOfSections;
1865 UInt32 TimeDateStamp;
1866 UInt32 PointerToSymbolTable;
1867 UInt32 NumberOfSymbols;
1868 UInt16 SizeOfOptionalHeader;
1869 UInt16 Characteristics;
1873 #define sizeof_COFF_header 20
1880 UInt32 VirtualAddress;
1881 UInt32 SizeOfRawData;
1882 UInt32 PointerToRawData;
1883 UInt32 PointerToRelocations;
1884 UInt32 PointerToLinenumbers;
1885 UInt16 NumberOfRelocations;
1886 UInt16 NumberOfLineNumbers;
1887 UInt32 Characteristics;
1891 #define sizeof_COFF_section 40
1898 UInt16 SectionNumber;
1901 UChar NumberOfAuxSymbols;
1905 #define sizeof_COFF_symbol 18
1910 UInt32 VirtualAddress;
1911 UInt32 SymbolTableIndex;
1916 #define sizeof_COFF_reloc 10
1919 /* From PE spec doc, section 3.3.2 */
1920 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1921 windows.h -- for the same purpose, but I want to know what I'm
1923 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1924 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1925 #define MYIMAGE_FILE_DLL 0x2000
1926 #define MYIMAGE_FILE_SYSTEM 0x1000
1927 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1928 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1929 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1931 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1932 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1933 #define MYIMAGE_SYM_CLASS_STATIC 3
1934 #define MYIMAGE_SYM_UNDEFINED 0
1936 /* From PE spec doc, section 4.1 */
1937 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1938 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1939 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1941 /* From PE spec doc, section 5.2.1 */
1942 #define MYIMAGE_REL_I386_DIR32 0x0006
1943 #define MYIMAGE_REL_I386_REL32 0x0014
1946 /* We use myindex to calculate array addresses, rather than
1947 simply doing the normal subscript thing. That's because
1948 some of the above structs have sizes which are not
1949 a whole number of words. GCC rounds their sizes up to a
1950 whole number of words, which means that the address calcs
1951 arising from using normal C indexing or pointer arithmetic
1952 are just plain wrong. Sigh.
1955 myindex ( int scale, void* base, int index )
1958 ((UChar*)base) + scale * index;
1963 printName ( UChar* name, UChar* strtab )
1965 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1966 UInt32 strtab_offset = * (UInt32*)(name+4);
1967 debugBelch("%s", strtab + strtab_offset );
1970 for (i = 0; i < 8; i++) {
1971 if (name[i] == 0) break;
1972 debugBelch("%c", name[i] );
1979 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
1981 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1982 UInt32 strtab_offset = * (UInt32*)(name+4);
1983 strncpy ( dst, strtab+strtab_offset, dstSize );
1989 if (name[i] == 0) break;
1999 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2002 /* If the string is longer than 8 bytes, look in the
2003 string table for it -- this will be correctly zero terminated.
2005 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2006 UInt32 strtab_offset = * (UInt32*)(name+4);
2007 return ((UChar*)strtab) + strtab_offset;
2009 /* Otherwise, if shorter than 8 bytes, return the original,
2010 which by defn is correctly terminated.
2012 if (name[7]==0) return name;
2013 /* The annoying case: 8 bytes. Copy into a temporary
2014 (which is never freed ...)
2016 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2018 strncpy(newstr,name,8);
2024 /* Just compares the short names (first 8 chars) */
2025 static COFF_section *
2026 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2030 = (COFF_header*)(oc->image);
2031 COFF_section* sectab
2033 ((UChar*)(oc->image))
2034 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2036 for (i = 0; i < hdr->NumberOfSections; i++) {
2039 COFF_section* section_i
2041 myindex ( sizeof_COFF_section, sectab, i );
2042 n1 = (UChar*) &(section_i->Name);
2044 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2045 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2046 n1[6]==n2[6] && n1[7]==n2[7])
2055 zapTrailingAtSign ( UChar* sym )
2057 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2059 if (sym[0] == 0) return;
2061 while (sym[i] != 0) i++;
2064 while (j > 0 && my_isdigit(sym[j])) j--;
2065 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2070 lookupSymbolInDLLs ( UChar *lbl )
2075 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2076 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2078 if (lbl[0] == '_') {
2079 /* HACK: if the name has an initial underscore, try stripping
2080 it off & look that up first. I've yet to verify whether there's
2081 a Rule that governs whether an initial '_' *should always* be
2082 stripped off when mapping from import lib name to the DLL name.
2084 sym = GetProcAddress(o_dll->instance, (lbl+1));
2086 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2090 sym = GetProcAddress(o_dll->instance, lbl);
2092 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2101 ocVerifyImage_PEi386 ( ObjectCode* oc )
2106 COFF_section* sectab;
2107 COFF_symbol* symtab;
2109 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2110 hdr = (COFF_header*)(oc->image);
2111 sectab = (COFF_section*) (
2112 ((UChar*)(oc->image))
2113 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2115 symtab = (COFF_symbol*) (
2116 ((UChar*)(oc->image))
2117 + hdr->PointerToSymbolTable
2119 strtab = ((UChar*)symtab)
2120 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2122 if (hdr->Machine != 0x14c) {
2123 errorBelch("%s: Not x86 PEi386", oc->fileName);
2126 if (hdr->SizeOfOptionalHeader != 0) {
2127 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2130 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2131 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2132 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2133 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2134 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2137 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2138 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2139 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2141 (int)(hdr->Characteristics));
2144 /* If the string table size is way crazy, this might indicate that
2145 there are more than 64k relocations, despite claims to the
2146 contrary. Hence this test. */
2147 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2149 if ( (*(UInt32*)strtab) > 600000 ) {
2150 /* Note that 600k has no special significance other than being
2151 big enough to handle the almost-2MB-sized lumps that
2152 constitute HSwin32*.o. */
2153 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2158 /* No further verification after this point; only debug printing. */
2160 IF_DEBUG(linker, i=1);
2161 if (i == 0) return 1;
2163 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2164 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2165 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2168 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2169 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2170 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2171 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2172 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2173 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2174 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2176 /* Print the section table. */
2178 for (i = 0; i < hdr->NumberOfSections; i++) {
2180 COFF_section* sectab_i
2182 myindex ( sizeof_COFF_section, sectab, i );
2189 printName ( sectab_i->Name, strtab );
2199 sectab_i->VirtualSize,
2200 sectab_i->VirtualAddress,
2201 sectab_i->SizeOfRawData,
2202 sectab_i->PointerToRawData,
2203 sectab_i->NumberOfRelocations,
2204 sectab_i->PointerToRelocations,
2205 sectab_i->PointerToRawData
2207 reltab = (COFF_reloc*) (
2208 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2211 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2212 /* If the relocation field (a short) has overflowed, the
2213 * real count can be found in the first reloc entry.
2215 * See Section 4.1 (last para) of the PE spec (rev6.0).
2217 COFF_reloc* rel = (COFF_reloc*)
2218 myindex ( sizeof_COFF_reloc, reltab, 0 );
2219 noRelocs = rel->VirtualAddress;
2222 noRelocs = sectab_i->NumberOfRelocations;
2226 for (; j < noRelocs; j++) {
2228 COFF_reloc* rel = (COFF_reloc*)
2229 myindex ( sizeof_COFF_reloc, reltab, j );
2231 " type 0x%-4x vaddr 0x%-8x name `",
2233 rel->VirtualAddress );
2234 sym = (COFF_symbol*)
2235 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2236 /* Hmm..mysterious looking offset - what's it for? SOF */
2237 printName ( sym->Name, strtab -10 );
2244 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2245 debugBelch("---START of string table---\n");
2246 for (i = 4; i < *(Int32*)strtab; i++) {
2248 debugBelch("\n"); else
2249 debugBelch("%c", strtab[i] );
2251 debugBelch("--- END of string table---\n");
2256 COFF_symbol* symtab_i;
2257 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2258 symtab_i = (COFF_symbol*)
2259 myindex ( sizeof_COFF_symbol, symtab, i );
2265 printName ( symtab_i->Name, strtab );
2274 (Int32)(symtab_i->SectionNumber),
2275 (UInt32)symtab_i->Type,
2276 (UInt32)symtab_i->StorageClass,
2277 (UInt32)symtab_i->NumberOfAuxSymbols
2279 i += symtab_i->NumberOfAuxSymbols;
2289 ocGetNames_PEi386 ( ObjectCode* oc )
2292 COFF_section* sectab;
2293 COFF_symbol* symtab;
2300 hdr = (COFF_header*)(oc->image);
2301 sectab = (COFF_section*) (
2302 ((UChar*)(oc->image))
2303 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2305 symtab = (COFF_symbol*) (
2306 ((UChar*)(oc->image))
2307 + hdr->PointerToSymbolTable
2309 strtab = ((UChar*)(oc->image))
2310 + hdr->PointerToSymbolTable
2311 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2313 /* Allocate space for any (local, anonymous) .bss sections. */
2315 for (i = 0; i < hdr->NumberOfSections; i++) {
2318 COFF_section* sectab_i
2320 myindex ( sizeof_COFF_section, sectab, i );
2321 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2322 /* sof 10/05: the PE spec text isn't too clear regarding what
2323 * the SizeOfRawData field is supposed to hold for object
2324 * file sections containing just uninitialized data -- for executables,
2325 * it is supposed to be zero; unclear what it's supposed to be
2326 * for object files. However, VirtualSize is guaranteed to be
2327 * zero for object files, which definitely suggests that SizeOfRawData
2328 * will be non-zero (where else would the size of this .bss section be
2329 * stored?) Looking at the COFF_section info for incoming object files,
2330 * this certainly appears to be the case.
2332 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2333 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2334 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2335 * variable decls into to the .bss section. (The specific function in Q which
2336 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2338 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2339 /* This is a non-empty .bss section. Allocate zeroed space for
2340 it, and set its PointerToRawData field such that oc->image +
2341 PointerToRawData == addr_of_zeroed_space. */
2342 bss_sz = sectab_i->VirtualSize;
2343 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2344 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2345 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2346 addProddableBlock(oc, zspace, bss_sz);
2347 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2350 /* Copy section information into the ObjectCode. */
2352 for (i = 0; i < hdr->NumberOfSections; i++) {
2358 = SECTIONKIND_OTHER;
2359 COFF_section* sectab_i
2361 myindex ( sizeof_COFF_section, sectab, i );
2362 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2365 /* I'm sure this is the Right Way to do it. However, the
2366 alternative of testing the sectab_i->Name field seems to
2367 work ok with Cygwin.
2369 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2370 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2371 kind = SECTIONKIND_CODE_OR_RODATA;
2374 if (0==strcmp(".text",sectab_i->Name) ||
2375 0==strcmp(".rdata",sectab_i->Name)||
2376 0==strcmp(".rodata",sectab_i->Name))
2377 kind = SECTIONKIND_CODE_OR_RODATA;
2378 if (0==strcmp(".data",sectab_i->Name) ||
2379 0==strcmp(".bss",sectab_i->Name))
2380 kind = SECTIONKIND_RWDATA;
2382 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2383 sz = sectab_i->SizeOfRawData;
2384 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2386 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2387 end = start + sz - 1;
2389 if (kind == SECTIONKIND_OTHER
2390 /* Ignore sections called which contain stabs debugging
2392 && 0 != strcmp(".stab", sectab_i->Name)
2393 && 0 != strcmp(".stabstr", sectab_i->Name)
2394 /* ignore constructor section for now */
2395 && 0 != strcmp(".ctors", sectab_i->Name)
2396 /* ignore section generated from .ident */
2397 && 0!= strcmp("/4", sectab_i->Name)
2398 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2399 && 0!= strcmp(".reloc", sectab_i->Name)
2401 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2405 if (kind != SECTIONKIND_OTHER && end >= start) {
2406 addSection(oc, kind, start, end);
2407 addProddableBlock(oc, start, end - start + 1);
2411 /* Copy exported symbols into the ObjectCode. */
2413 oc->n_symbols = hdr->NumberOfSymbols;
2414 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2415 "ocGetNames_PEi386(oc->symbols)");
2416 /* Call me paranoid; I don't care. */
2417 for (i = 0; i < oc->n_symbols; i++)
2418 oc->symbols[i] = NULL;
2422 COFF_symbol* symtab_i;
2423 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2424 symtab_i = (COFF_symbol*)
2425 myindex ( sizeof_COFF_symbol, symtab, i );
2429 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2430 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2431 /* This symbol is global and defined, viz, exported */
2432 /* for MYIMAGE_SYMCLASS_EXTERNAL
2433 && !MYIMAGE_SYM_UNDEFINED,
2434 the address of the symbol is:
2435 address of relevant section + offset in section
2437 COFF_section* sectabent
2438 = (COFF_section*) myindex ( sizeof_COFF_section,
2440 symtab_i->SectionNumber-1 );
2441 addr = ((UChar*)(oc->image))
2442 + (sectabent->PointerToRawData
2446 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2447 && symtab_i->Value > 0) {
2448 /* This symbol isn't in any section at all, ie, global bss.
2449 Allocate zeroed space for it. */
2450 addr = stgCallocBytes(1, symtab_i->Value,
2451 "ocGetNames_PEi386(non-anonymous bss)");
2452 addSection(oc, SECTIONKIND_RWDATA, addr,
2453 ((UChar*)addr) + symtab_i->Value - 1);
2454 addProddableBlock(oc, addr, symtab_i->Value);
2455 /* debugBelch("BSS section at 0x%x\n", addr); */
2458 if (addr != NULL ) {
2459 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2460 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2461 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2462 ASSERT(i >= 0 && i < oc->n_symbols);
2463 /* cstring_from_COFF_symbol_name always succeeds. */
2464 oc->symbols[i] = sname;
2465 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2469 "IGNORING symbol %d\n"
2473 printName ( symtab_i->Name, strtab );
2482 (Int32)(symtab_i->SectionNumber),
2483 (UInt32)symtab_i->Type,
2484 (UInt32)symtab_i->StorageClass,
2485 (UInt32)symtab_i->NumberOfAuxSymbols
2490 i += symtab_i->NumberOfAuxSymbols;
2499 ocResolve_PEi386 ( ObjectCode* oc )
2502 COFF_section* sectab;
2503 COFF_symbol* symtab;
2513 /* ToDo: should be variable-sized? But is at least safe in the
2514 sense of buffer-overrun-proof. */
2516 /* debugBelch("resolving for %s\n", oc->fileName); */
2518 hdr = (COFF_header*)(oc->image);
2519 sectab = (COFF_section*) (
2520 ((UChar*)(oc->image))
2521 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2523 symtab = (COFF_symbol*) (
2524 ((UChar*)(oc->image))
2525 + hdr->PointerToSymbolTable
2527 strtab = ((UChar*)(oc->image))
2528 + hdr->PointerToSymbolTable
2529 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2531 for (i = 0; i < hdr->NumberOfSections; i++) {
2532 COFF_section* sectab_i
2534 myindex ( sizeof_COFF_section, sectab, i );
2537 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2540 /* Ignore sections called which contain stabs debugging
2542 if (0 == strcmp(".stab", sectab_i->Name)
2543 || 0 == strcmp(".stabstr", sectab_i->Name)
2544 || 0 == strcmp(".ctors", sectab_i->Name))
2547 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2548 /* If the relocation field (a short) has overflowed, the
2549 * real count can be found in the first reloc entry.
2551 * See Section 4.1 (last para) of the PE spec (rev6.0).
2553 * Nov2003 update: the GNU linker still doesn't correctly
2554 * handle the generation of relocatable object files with
2555 * overflown relocations. Hence the output to warn of potential
2558 COFF_reloc* rel = (COFF_reloc*)
2559 myindex ( sizeof_COFF_reloc, reltab, 0 );
2560 noRelocs = rel->VirtualAddress;
2562 /* 10/05: we now assume (and check for) a GNU ld that is capable
2563 * of handling object files with (>2^16) of relocs.
2566 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2571 noRelocs = sectab_i->NumberOfRelocations;
2576 for (; j < noRelocs; j++) {
2578 COFF_reloc* reltab_j
2580 myindex ( sizeof_COFF_reloc, reltab, j );
2582 /* the location to patch */
2584 ((UChar*)(oc->image))
2585 + (sectab_i->PointerToRawData
2586 + reltab_j->VirtualAddress
2587 - sectab_i->VirtualAddress )
2589 /* the existing contents of pP */
2591 /* the symbol to connect to */
2592 sym = (COFF_symbol*)
2593 myindex ( sizeof_COFF_symbol,
2594 symtab, reltab_j->SymbolTableIndex );
2597 "reloc sec %2d num %3d: type 0x%-4x "
2598 "vaddr 0x%-8x name `",
2600 (UInt32)reltab_j->Type,
2601 reltab_j->VirtualAddress );
2602 printName ( sym->Name, strtab );
2603 debugBelch("'\n" ));
2605 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2606 COFF_section* section_sym
2607 = findPEi386SectionCalled ( oc, sym->Name );
2609 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2612 S = ((UInt32)(oc->image))
2613 + (section_sym->PointerToRawData
2616 copyName ( sym->Name, strtab, symbol, 1000-1 );
2617 S = (UInt32) lookupSymbol( symbol );
2618 if ((void*)S != NULL) goto foundit;
2619 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2623 checkProddableBlock(oc, pP);
2624 switch (reltab_j->Type) {
2625 case MYIMAGE_REL_I386_DIR32:
2628 case MYIMAGE_REL_I386_REL32:
2629 /* Tricky. We have to insert a displacement at
2630 pP which, when added to the PC for the _next_
2631 insn, gives the address of the target (S).
2632 Problem is to know the address of the next insn
2633 when we only know pP. We assume that this
2634 literal field is always the last in the insn,
2635 so that the address of the next insn is pP+4
2636 -- hence the constant 4.
2637 Also I don't know if A should be added, but so
2638 far it has always been zero.
2640 SOF 05/2005: 'A' (old contents of *pP) have been observed
2641 to contain values other than zero (the 'wx' object file
2642 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2643 So, add displacement to old value instead of asserting
2644 A to be zero. Fixes wxhaskell-related crashes, and no other
2645 ill effects have been observed.
2647 Update: the reason why we're seeing these more elaborate
2648 relocations is due to a switch in how the NCG compiles SRTs
2649 and offsets to them from info tables. SRTs live in .(ro)data,
2650 while info tables live in .text, causing GAS to emit REL32/DISP32
2651 relocations with non-zero values. Adding the displacement is
2652 the right thing to do.
2654 *pP = S - ((UInt32)pP) - 4 + A;
2657 debugBelch("%s: unhandled PEi386 relocation type %d",
2658 oc->fileName, reltab_j->Type);
2665 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2669 #endif /* defined(OBJFORMAT_PEi386) */
2672 /* --------------------------------------------------------------------------
2674 * ------------------------------------------------------------------------*/
2676 #if defined(OBJFORMAT_ELF)
2681 #if defined(sparc_HOST_ARCH)
2682 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2683 #elif defined(i386_HOST_ARCH)
2684 # define ELF_TARGET_386 /* Used inside <elf.h> */
2685 #elif defined(x86_64_HOST_ARCH)
2686 # define ELF_TARGET_X64_64
2688 #elif defined (ia64_HOST_ARCH)
2689 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2691 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2692 # define ELF_NEED_GOT /* needs Global Offset Table */
2693 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2696 #if !defined(openbsd_HOST_OS)
2699 /* openbsd elf has things in different places, with diff names */
2700 # include <elf_abi.h>
2701 # include <machine/reloc.h>
2702 # define R_386_32 RELOC_32
2703 # define R_386_PC32 RELOC_PC32
2706 /* If elf.h doesn't define it */
2707 # ifndef R_X86_64_PC64
2708 # define R_X86_64_PC64 24
2712 * Define a set of types which can be used for both ELF32 and ELF64
2716 #define ELFCLASS ELFCLASS64
2717 #define Elf_Addr Elf64_Addr
2718 #define Elf_Word Elf64_Word
2719 #define Elf_Sword Elf64_Sword
2720 #define Elf_Ehdr Elf64_Ehdr
2721 #define Elf_Phdr Elf64_Phdr
2722 #define Elf_Shdr Elf64_Shdr
2723 #define Elf_Sym Elf64_Sym
2724 #define Elf_Rel Elf64_Rel
2725 #define Elf_Rela Elf64_Rela
2726 #define ELF_ST_TYPE ELF64_ST_TYPE
2727 #define ELF_ST_BIND ELF64_ST_BIND
2728 #define ELF_R_TYPE ELF64_R_TYPE
2729 #define ELF_R_SYM ELF64_R_SYM
2731 #define ELFCLASS ELFCLASS32
2732 #define Elf_Addr Elf32_Addr
2733 #define Elf_Word Elf32_Word
2734 #define Elf_Sword Elf32_Sword
2735 #define Elf_Ehdr Elf32_Ehdr
2736 #define Elf_Phdr Elf32_Phdr
2737 #define Elf_Shdr Elf32_Shdr
2738 #define Elf_Sym Elf32_Sym
2739 #define Elf_Rel Elf32_Rel
2740 #define Elf_Rela Elf32_Rela
2742 #define ELF_ST_TYPE ELF32_ST_TYPE
2745 #define ELF_ST_BIND ELF32_ST_BIND
2748 #define ELF_R_TYPE ELF32_R_TYPE
2751 #define ELF_R_SYM ELF32_R_SYM
2757 * Functions to allocate entries in dynamic sections. Currently we simply
2758 * preallocate a large number, and we don't check if a entry for the given
2759 * target already exists (a linear search is too slow). Ideally these
2760 * entries would be associated with symbols.
2763 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2764 #define GOT_SIZE 0x20000
2765 #define FUNCTION_TABLE_SIZE 0x10000
2766 #define PLT_SIZE 0x08000
2769 static Elf_Addr got[GOT_SIZE];
2770 static unsigned int gotIndex;
2771 static Elf_Addr gp_val = (Elf_Addr)got;
2774 allocateGOTEntry(Elf_Addr target)
2778 if (gotIndex >= GOT_SIZE)
2779 barf("Global offset table overflow");
2781 entry = &got[gotIndex++];
2783 return (Elf_Addr)entry;
2787 #ifdef ELF_FUNCTION_DESC
2793 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2794 static unsigned int functionTableIndex;
2797 allocateFunctionDesc(Elf_Addr target)
2799 FunctionDesc *entry;
2801 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2802 barf("Function table overflow");
2804 entry = &functionTable[functionTableIndex++];
2806 entry->gp = (Elf_Addr)gp_val;
2807 return (Elf_Addr)entry;
2811 copyFunctionDesc(Elf_Addr target)
2813 FunctionDesc *olddesc = (FunctionDesc *)target;
2814 FunctionDesc *newdesc;
2816 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2817 newdesc->gp = olddesc->gp;
2818 return (Elf_Addr)newdesc;
2823 #ifdef ia64_HOST_ARCH
2824 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2825 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2827 static unsigned char plt_code[] =
2829 /* taken from binutils bfd/elfxx-ia64.c */
2830 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2831 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2832 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2833 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2834 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2835 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2838 /* If we can't get to the function descriptor via gp, take a local copy of it */
2839 #define PLT_RELOC(code, target) { \
2840 Elf64_Sxword rel_value = target - gp_val; \
2841 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2842 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2844 ia64_reloc_gprel22((Elf_Addr)code, target); \
2849 unsigned char code[sizeof(plt_code)];
2853 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2855 PLTEntry *plt = (PLTEntry *)oc->plt;
2858 if (oc->pltIndex >= PLT_SIZE)
2859 barf("Procedure table overflow");
2861 entry = &plt[oc->pltIndex++];
2862 memcpy(entry->code, plt_code, sizeof(entry->code));
2863 PLT_RELOC(entry->code, target);
2864 return (Elf_Addr)entry;
2870 return (PLT_SIZE * sizeof(PLTEntry));
2876 * Generic ELF functions
2880 findElfSection ( void* objImage, Elf_Word sh_type )
2882 char* ehdrC = (char*)objImage;
2883 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2884 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2885 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2889 for (i = 0; i < ehdr->e_shnum; i++) {
2890 if (shdr[i].sh_type == sh_type
2891 /* Ignore the section header's string table. */
2892 && i != ehdr->e_shstrndx
2893 /* Ignore string tables named .stabstr, as they contain
2895 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2897 ptr = ehdrC + shdr[i].sh_offset;
2904 #if defined(ia64_HOST_ARCH)
2906 findElfSegment ( void* objImage, Elf_Addr vaddr )
2908 char* ehdrC = (char*)objImage;
2909 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2910 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2911 Elf_Addr segaddr = 0;
2914 for (i = 0; i < ehdr->e_phnum; i++) {
2915 segaddr = phdr[i].p_vaddr;
2916 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2924 ocVerifyImage_ELF ( ObjectCode* oc )
2928 int i, j, nent, nstrtab, nsymtabs;
2932 char* ehdrC = (char*)(oc->image);
2933 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2935 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2936 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2937 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2938 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2939 errorBelch("%s: not an ELF object", oc->fileName);
2943 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2944 errorBelch("%s: unsupported ELF format", oc->fileName);
2948 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2949 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2951 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2952 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2954 errorBelch("%s: unknown endiannness", oc->fileName);
2958 if (ehdr->e_type != ET_REL) {
2959 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2962 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2964 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2965 switch (ehdr->e_machine) {
2966 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2967 #ifdef EM_SPARC32PLUS
2968 case EM_SPARC32PLUS:
2970 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2972 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2974 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
2976 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
2977 #elif defined(EM_AMD64)
2978 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
2980 default: IF_DEBUG(linker,debugBelch( "unknown" ));
2981 errorBelch("%s: unknown architecture (e_machine == %d)"
2982 , oc->fileName, ehdr->e_machine);
2986 IF_DEBUG(linker,debugBelch(
2987 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
2988 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
2990 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
2992 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
2994 if (ehdr->e_shstrndx == SHN_UNDEF) {
2995 errorBelch("%s: no section header string table", oc->fileName);
2998 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3000 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3003 for (i = 0; i < ehdr->e_shnum; i++) {
3004 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3005 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3006 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3007 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3008 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3009 ehdrC + shdr[i].sh_offset,
3010 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3012 if (shdr[i].sh_type == SHT_REL) {
3013 IF_DEBUG(linker,debugBelch("Rel " ));
3014 } else if (shdr[i].sh_type == SHT_RELA) {
3015 IF_DEBUG(linker,debugBelch("RelA " ));
3017 IF_DEBUG(linker,debugBelch(" "));
3020 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3024 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3027 for (i = 0; i < ehdr->e_shnum; i++) {
3028 if (shdr[i].sh_type == SHT_STRTAB
3029 /* Ignore the section header's string table. */
3030 && i != ehdr->e_shstrndx
3031 /* Ignore string tables named .stabstr, as they contain
3033 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3035 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3036 strtab = ehdrC + shdr[i].sh_offset;
3041 errorBelch("%s: no string tables, or too many", oc->fileName);
3046 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3047 for (i = 0; i < ehdr->e_shnum; i++) {
3048 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3049 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3051 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3052 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3053 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3055 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3057 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3058 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3061 for (j = 0; j < nent; j++) {
3062 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3063 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3064 (int)stab[j].st_shndx,
3065 (int)stab[j].st_size,
3066 (char*)stab[j].st_value ));
3068 IF_DEBUG(linker,debugBelch("type=" ));
3069 switch (ELF_ST_TYPE(stab[j].st_info)) {
3070 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3071 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3072 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3073 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3074 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3075 default: IF_DEBUG(linker,debugBelch("? " )); break;
3077 IF_DEBUG(linker,debugBelch(" " ));
3079 IF_DEBUG(linker,debugBelch("bind=" ));
3080 switch (ELF_ST_BIND(stab[j].st_info)) {
3081 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3082 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3083 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3084 default: IF_DEBUG(linker,debugBelch("? " )); break;
3086 IF_DEBUG(linker,debugBelch(" " ));
3088 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3092 if (nsymtabs == 0) {
3093 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3100 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3104 if (hdr->sh_type == SHT_PROGBITS
3105 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3106 /* .text-style section */
3107 return SECTIONKIND_CODE_OR_RODATA;
3110 if (hdr->sh_type == SHT_PROGBITS
3111 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3112 /* .data-style section */
3113 return SECTIONKIND_RWDATA;
3116 if (hdr->sh_type == SHT_PROGBITS
3117 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3118 /* .rodata-style section */
3119 return SECTIONKIND_CODE_OR_RODATA;
3122 if (hdr->sh_type == SHT_NOBITS
3123 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3124 /* .bss-style section */
3126 return SECTIONKIND_RWDATA;
3129 return SECTIONKIND_OTHER;
3134 ocGetNames_ELF ( ObjectCode* oc )
3139 char* ehdrC = (char*)(oc->image);
3140 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3141 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3142 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3144 ASSERT(symhash != NULL);
3147 errorBelch("%s: no strtab", oc->fileName);
3152 for (i = 0; i < ehdr->e_shnum; i++) {
3153 /* Figure out what kind of section it is. Logic derived from
3154 Figure 1.14 ("Special Sections") of the ELF document
3155 ("Portable Formats Specification, Version 1.1"). */
3157 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3159 if (is_bss && shdr[i].sh_size > 0) {
3160 /* This is a non-empty .bss section. Allocate zeroed space for
3161 it, and set its .sh_offset field such that
3162 ehdrC + .sh_offset == addr_of_zeroed_space. */
3163 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3164 "ocGetNames_ELF(BSS)");
3165 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3167 debugBelch("BSS section at 0x%x, size %d\n",
3168 zspace, shdr[i].sh_size);
3172 /* fill in the section info */
3173 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3174 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3175 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3176 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3179 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3181 /* copy stuff into this module's object symbol table */
3182 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3183 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3185 oc->n_symbols = nent;
3186 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3187 "ocGetNames_ELF(oc->symbols)");
3189 for (j = 0; j < nent; j++) {
3191 char isLocal = FALSE; /* avoids uninit-var warning */
3193 char* nm = strtab + stab[j].st_name;
3194 int secno = stab[j].st_shndx;
3196 /* Figure out if we want to add it; if so, set ad to its
3197 address. Otherwise leave ad == NULL. */
3199 if (secno == SHN_COMMON) {
3201 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3203 debugBelch("COMMON symbol, size %d name %s\n",
3204 stab[j].st_size, nm);
3206 /* Pointless to do addProddableBlock() for this area,
3207 since the linker should never poke around in it. */
3210 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3211 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3213 /* and not an undefined symbol */
3214 && stab[j].st_shndx != SHN_UNDEF
3215 /* and not in a "special section" */
3216 && stab[j].st_shndx < SHN_LORESERVE
3218 /* and it's a not a section or string table or anything silly */
3219 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3220 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3221 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3224 /* Section 0 is the undefined section, hence > and not >=. */
3225 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3227 if (shdr[secno].sh_type == SHT_NOBITS) {
3228 debugBelch(" BSS symbol, size %d off %d name %s\n",
3229 stab[j].st_size, stab[j].st_value, nm);
3232 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3233 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3236 #ifdef ELF_FUNCTION_DESC
3237 /* dlsym() and the initialisation table both give us function
3238 * descriptors, so to be consistent we store function descriptors
3239 * in the symbol table */
3240 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3241 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3243 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3244 ad, oc->fileName, nm ));
3249 /* And the decision is ... */
3253 oc->symbols[j] = nm;
3256 /* Ignore entirely. */
3258 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3262 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3263 strtab + stab[j].st_name ));
3266 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3267 (int)ELF_ST_BIND(stab[j].st_info),
3268 (int)ELF_ST_TYPE(stab[j].st_info),
3269 (int)stab[j].st_shndx,
3270 strtab + stab[j].st_name
3273 oc->symbols[j] = NULL;
3282 /* Do ELF relocations which lack an explicit addend. All x86-linux
3283 relocations appear to be of this form. */
3285 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3286 Elf_Shdr* shdr, int shnum,
3287 Elf_Sym* stab, char* strtab )
3292 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3293 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3294 int target_shndx = shdr[shnum].sh_info;
3295 int symtab_shndx = shdr[shnum].sh_link;
3297 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3298 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3299 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3300 target_shndx, symtab_shndx ));
3302 /* Skip sections that we're not interested in. */
3305 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3306 if (kind == SECTIONKIND_OTHER) {
3307 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3312 for (j = 0; j < nent; j++) {
3313 Elf_Addr offset = rtab[j].r_offset;
3314 Elf_Addr info = rtab[j].r_info;
3316 Elf_Addr P = ((Elf_Addr)targ) + offset;
3317 Elf_Word* pP = (Elf_Word*)P;
3322 StgStablePtr stablePtr;
3325 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3326 j, (void*)offset, (void*)info ));
3328 IF_DEBUG(linker,debugBelch( " ZERO" ));
3331 Elf_Sym sym = stab[ELF_R_SYM(info)];
3332 /* First see if it is a local symbol. */
3333 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3334 /* Yes, so we can get the address directly from the ELF symbol
3336 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3338 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3339 + stab[ELF_R_SYM(info)].st_value);
3342 symbol = strtab + sym.st_name;
3343 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3344 if (NULL == stablePtr) {
3345 /* No, so look up the name in our global table. */
3346 S_tmp = lookupSymbol( symbol );
3347 S = (Elf_Addr)S_tmp;
3349 stableVal = deRefStablePtr( stablePtr );
3351 S = (Elf_Addr)S_tmp;
3355 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3358 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3361 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3362 (void*)P, (void*)S, (void*)A ));
3363 checkProddableBlock ( oc, pP );
3367 switch (ELF_R_TYPE(info)) {
3368 # ifdef i386_HOST_ARCH
3369 case R_386_32: *pP = value; break;
3370 case R_386_PC32: *pP = value - P; break;
3373 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3374 oc->fileName, (lnat)ELF_R_TYPE(info));
3382 /* Do ELF relocations for which explicit addends are supplied.
3383 sparc-solaris relocations appear to be of this form. */
3385 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3386 Elf_Shdr* shdr, int shnum,
3387 Elf_Sym* stab, char* strtab )
3390 char *symbol = NULL;
3392 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3393 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3394 int target_shndx = shdr[shnum].sh_info;
3395 int symtab_shndx = shdr[shnum].sh_link;
3397 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3398 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3399 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3400 target_shndx, symtab_shndx ));
3402 for (j = 0; j < nent; j++) {
3403 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3404 /* This #ifdef only serves to avoid unused-var warnings. */
3405 Elf_Addr offset = rtab[j].r_offset;
3406 Elf_Addr P = targ + offset;
3408 Elf_Addr info = rtab[j].r_info;
3409 Elf_Addr A = rtab[j].r_addend;
3413 # if defined(sparc_HOST_ARCH)
3414 Elf_Word* pP = (Elf_Word*)P;
3416 # elif defined(ia64_HOST_ARCH)
3417 Elf64_Xword *pP = (Elf64_Xword *)P;
3419 # elif defined(powerpc_HOST_ARCH)
3423 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3424 j, (void*)offset, (void*)info,
3427 IF_DEBUG(linker,debugBelch( " ZERO" ));
3430 Elf_Sym sym = stab[ELF_R_SYM(info)];
3431 /* First see if it is a local symbol. */
3432 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3433 /* Yes, so we can get the address directly from the ELF symbol
3435 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3437 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3438 + stab[ELF_R_SYM(info)].st_value);
3439 #ifdef ELF_FUNCTION_DESC
3440 /* Make a function descriptor for this function */
3441 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3442 S = allocateFunctionDesc(S + A);
3447 /* No, so look up the name in our global table. */
3448 symbol = strtab + sym.st_name;
3449 S_tmp = lookupSymbol( symbol );
3450 S = (Elf_Addr)S_tmp;
3452 #ifdef ELF_FUNCTION_DESC
3453 /* If a function, already a function descriptor - we would
3454 have to copy it to add an offset. */
3455 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3456 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3460 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3463 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3466 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3467 (void*)P, (void*)S, (void*)A ));
3468 /* checkProddableBlock ( oc, (void*)P ); */
3472 switch (ELF_R_TYPE(info)) {
3473 # if defined(sparc_HOST_ARCH)
3474 case R_SPARC_WDISP30:
3475 w1 = *pP & 0xC0000000;
3476 w2 = (Elf_Word)((value - P) >> 2);
3477 ASSERT((w2 & 0xC0000000) == 0);
3482 w1 = *pP & 0xFFC00000;
3483 w2 = (Elf_Word)(value >> 10);
3484 ASSERT((w2 & 0xFFC00000) == 0);
3490 w2 = (Elf_Word)(value & 0x3FF);
3491 ASSERT((w2 & ~0x3FF) == 0);
3496 /* According to the Sun documentation:
3498 This relocation type resembles R_SPARC_32, except it refers to an
3499 unaligned word. That is, the word to be relocated must be treated
3500 as four separate bytes with arbitrary alignment, not as a word
3501 aligned according to the architecture requirements.
3504 w2 = (Elf_Word)value;
3506 // SPARC doesn't do misaligned writes of 32 bit words,
3507 // so we have to do this one byte-at-a-time.
3508 char *pPc = (char*)pP;
3509 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3510 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3511 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3512 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3516 w2 = (Elf_Word)value;
3519 # elif defined(ia64_HOST_ARCH)
3520 case R_IA64_DIR64LSB:
3521 case R_IA64_FPTR64LSB:
3524 case R_IA64_PCREL64LSB:
3527 case R_IA64_SEGREL64LSB:
3528 addr = findElfSegment(ehdrC, value);
3531 case R_IA64_GPREL22:
3532 ia64_reloc_gprel22(P, value);
3534 case R_IA64_LTOFF22:
3535 case R_IA64_LTOFF22X:
3536 case R_IA64_LTOFF_FPTR22:
3537 addr = allocateGOTEntry(value);
3538 ia64_reloc_gprel22(P, addr);
3540 case R_IA64_PCREL21B:
3541 ia64_reloc_pcrel21(P, S, oc);
3544 /* This goes with R_IA64_LTOFF22X and points to the load to
3545 * convert into a move. We don't implement relaxation. */
3547 # elif defined(powerpc_HOST_ARCH)
3548 case R_PPC_ADDR16_LO:
3549 *(Elf32_Half*) P = value;
3552 case R_PPC_ADDR16_HI:
3553 *(Elf32_Half*) P = value >> 16;
3556 case R_PPC_ADDR16_HA:
3557 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3561 *(Elf32_Word *) P = value;
3565 *(Elf32_Word *) P = value - P;
3571 if( delta << 6 >> 6 != delta )
3573 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3577 if( value == 0 || delta << 6 >> 6 != delta )
3579 barf( "Unable to make SymbolExtra for #%d",
3585 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3586 | (delta & 0x3fffffc);
3590 #if x86_64_HOST_ARCH
3592 *(Elf64_Xword *)P = value;
3597 StgInt64 off = value - P;
3598 if (off >= 0x7fffffffL || off < -0x80000000L) {
3599 #if X86_64_ELF_NONPIC_HACK
3600 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3602 off = pltAddress + A - P;
3604 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3605 symbol, off, oc->fileName );
3608 *(Elf64_Word *)P = (Elf64_Word)off;
3614 StgInt64 off = value - P;
3615 *(Elf64_Word *)P = (Elf64_Word)off;
3620 if (value >= 0x7fffffffL) {
3621 #if X86_64_ELF_NONPIC_HACK
3622 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3624 value = pltAddress + A;
3626 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3627 symbol, value, oc->fileName );
3630 *(Elf64_Word *)P = (Elf64_Word)value;
3634 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3635 #if X86_64_ELF_NONPIC_HACK
3636 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3638 value = pltAddress + A;
3640 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3641 symbol, value, oc->fileName );
3644 *(Elf64_Sword *)P = (Elf64_Sword)value;
3647 case R_X86_64_GOTPCREL:
3649 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3650 StgInt64 off = gotAddress + A - P;
3651 *(Elf64_Word *)P = (Elf64_Word)off;
3655 case R_X86_64_PLT32:
3657 StgInt64 off = value - P;
3658 if (off >= 0x7fffffffL || off < -0x80000000L) {
3659 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3661 off = pltAddress + A - P;
3663 *(Elf64_Word *)P = (Elf64_Word)off;
3669 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3670 oc->fileName, (lnat)ELF_R_TYPE(info));
3679 ocResolve_ELF ( ObjectCode* oc )
3683 Elf_Sym* stab = NULL;
3684 char* ehdrC = (char*)(oc->image);
3685 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3686 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3688 /* first find "the" symbol table */
3689 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3691 /* also go find the string table */
3692 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3694 if (stab == NULL || strtab == NULL) {
3695 errorBelch("%s: can't find string or symbol table", oc->fileName);
3699 /* Process the relocation sections. */
3700 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3701 if (shdr[shnum].sh_type == SHT_REL) {
3702 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3703 shnum, stab, strtab );
3707 if (shdr[shnum].sh_type == SHT_RELA) {
3708 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3709 shnum, stab, strtab );
3714 #if defined(powerpc_HOST_ARCH)
3715 ocFlushInstructionCache( oc );
3723 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3724 * at the front. The following utility functions pack and unpack instructions, and
3725 * take care of the most common relocations.
3728 #ifdef ia64_HOST_ARCH
3731 ia64_extract_instruction(Elf64_Xword *target)
3734 int slot = (Elf_Addr)target & 3;
3735 target = (Elf_Addr)target & ~3;
3743 return ((w1 >> 5) & 0x1ffffffffff);
3745 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3749 barf("ia64_extract_instruction: invalid slot %p", target);
3754 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3756 int slot = (Elf_Addr)target & 3;
3757 target = (Elf_Addr)target & ~3;
3762 *target |= value << 5;
3765 *target |= value << 46;
3766 *(target+1) |= value >> 18;
3769 *(target+1) |= value << 23;
3775 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3777 Elf64_Xword instruction;
3778 Elf64_Sxword rel_value;
3780 rel_value = value - gp_val;
3781 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3782 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3784 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3785 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3786 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3787 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3788 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3789 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3793 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3795 Elf64_Xword instruction;
3796 Elf64_Sxword rel_value;
3799 entry = allocatePLTEntry(value, oc);
3801 rel_value = (entry >> 4) - (target >> 4);
3802 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3803 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3805 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3806 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3807 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3808 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3814 * PowerPC & X86_64 ELF specifics
3817 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3819 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3825 ehdr = (Elf_Ehdr *) oc->image;
3826 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3828 for( i = 0; i < ehdr->e_shnum; i++ )
3829 if( shdr[i].sh_type == SHT_SYMTAB )
3832 if( i == ehdr->e_shnum )
3834 errorBelch( "This ELF file contains no symtab" );
3838 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3840 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3841 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3846 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3849 #endif /* powerpc */
3853 /* --------------------------------------------------------------------------
3855 * ------------------------------------------------------------------------*/
3857 #if defined(OBJFORMAT_MACHO)
3860 Support for MachO linking on Darwin/MacOS X
3861 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3863 I hereby formally apologize for the hackish nature of this code.
3864 Things that need to be done:
3865 *) implement ocVerifyImage_MachO
3866 *) add still more sanity checks.
3869 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3870 #define mach_header mach_header_64
3871 #define segment_command segment_command_64
3872 #define section section_64
3873 #define nlist nlist_64
3876 #ifdef powerpc_HOST_ARCH
3877 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3879 struct mach_header *header = (struct mach_header *) oc->image;
3880 struct load_command *lc = (struct load_command *) (header + 1);
3883 for( i = 0; i < header->ncmds; i++ )
3885 if( lc->cmd == LC_SYMTAB )
3887 // Find out the first and last undefined external
3888 // symbol, so we don't have to allocate too many
3890 struct symtab_command *symLC = (struct symtab_command *) lc;
3891 unsigned min = symLC->nsyms, max = 0;
3892 struct nlist *nlist =
3893 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3895 for(i=0;i<symLC->nsyms;i++)
3897 if(nlist[i].n_type & N_STAB)
3899 else if(nlist[i].n_type & N_EXT)
3901 if((nlist[i].n_type & N_TYPE) == N_UNDF
3902 && (nlist[i].n_value == 0))
3912 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3917 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3919 return ocAllocateSymbolExtras(oc,0,0);
3922 #ifdef x86_64_HOST_ARCH
3923 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3925 struct mach_header *header = (struct mach_header *) oc->image;
3926 struct load_command *lc = (struct load_command *) (header + 1);
3929 for( i = 0; i < header->ncmds; i++ )
3931 if( lc->cmd == LC_SYMTAB )
3933 // Just allocate one entry for every symbol
3934 struct symtab_command *symLC = (struct symtab_command *) lc;
3936 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3939 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3941 return ocAllocateSymbolExtras(oc,0,0);
3945 static int ocVerifyImage_MachO(ObjectCode* oc)
3947 char *image = (char*) oc->image;
3948 struct mach_header *header = (struct mach_header*) image;
3950 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3951 if(header->magic != MH_MAGIC_64)
3954 if(header->magic != MH_MAGIC)
3957 // FIXME: do some more verifying here
3961 static int resolveImports(
3964 struct symtab_command *symLC,
3965 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3966 unsigned long *indirectSyms,
3967 struct nlist *nlist)
3970 size_t itemSize = 4;
3973 int isJumpTable = 0;
3974 if(!strcmp(sect->sectname,"__jump_table"))
3978 ASSERT(sect->reserved2 == itemSize);
3982 for(i=0; i*itemSize < sect->size;i++)
3984 // according to otool, reserved1 contains the first index into the indirect symbol table
3985 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
3986 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
3989 if((symbol->n_type & N_TYPE) == N_UNDF
3990 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
3991 addr = (void*) (symbol->n_value);
3993 addr = lookupSymbol(nm);
3996 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4004 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4005 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4006 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4007 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4012 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4013 ((void**)(image + sect->offset))[i] = addr;
4020 static unsigned long relocateAddress(
4023 struct section* sections,
4024 unsigned long address)
4027 for(i = 0; i < nSections; i++)
4029 if(sections[i].addr <= address
4030 && address < sections[i].addr + sections[i].size)
4032 return (unsigned long)oc->image
4033 + sections[i].offset + address - sections[i].addr;
4036 barf("Invalid Mach-O file:"
4037 "Address out of bounds while relocating object file");
4041 static int relocateSection(
4044 struct symtab_command *symLC, struct nlist *nlist,
4045 int nSections, struct section* sections, struct section *sect)
4047 struct relocation_info *relocs;
4050 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4052 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4054 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4056 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4060 relocs = (struct relocation_info*) (image + sect->reloff);
4064 #ifdef x86_64_HOST_ARCH
4065 struct relocation_info *reloc = &relocs[i];
4067 char *thingPtr = image + sect->offset + reloc->r_address;
4069 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4070 complains that it may be used uninitialized if we don't */
4073 int type = reloc->r_type;
4075 checkProddableBlock(oc,thingPtr);
4076 switch(reloc->r_length)
4079 thing = *(uint8_t*)thingPtr;
4080 baseValue = (uint64_t)thingPtr + 1;
4083 thing = *(uint16_t*)thingPtr;
4084 baseValue = (uint64_t)thingPtr + 2;
4087 thing = *(uint32_t*)thingPtr;
4088 baseValue = (uint64_t)thingPtr + 4;
4091 thing = *(uint64_t*)thingPtr;
4092 baseValue = (uint64_t)thingPtr + 8;
4095 barf("Unknown size.");
4098 if(type == X86_64_RELOC_GOT
4099 || type == X86_64_RELOC_GOT_LOAD)
4101 ASSERT(reloc->r_extern);
4102 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4104 type = X86_64_RELOC_SIGNED;
4106 else if(reloc->r_extern)
4108 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4109 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4110 if(symbol->n_value == 0)
4111 value = (uint64_t) lookupSymbol(nm);
4113 value = relocateAddress(oc, nSections, sections,
4118 value = sections[reloc->r_symbolnum-1].offset
4119 - sections[reloc->r_symbolnum-1].addr
4123 if(type == X86_64_RELOC_BRANCH)
4125 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4127 ASSERT(reloc->r_extern);
4128 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4131 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4132 type = X86_64_RELOC_SIGNED;
4137 case X86_64_RELOC_UNSIGNED:
4138 ASSERT(!reloc->r_pcrel);
4141 case X86_64_RELOC_SIGNED:
4142 ASSERT(reloc->r_pcrel);
4143 thing += value - baseValue;
4145 case X86_64_RELOC_SUBTRACTOR:
4146 ASSERT(!reloc->r_pcrel);
4150 barf("unkown relocation");
4153 switch(reloc->r_length)
4156 *(uint8_t*)thingPtr = thing;
4159 *(uint16_t*)thingPtr = thing;
4162 *(uint32_t*)thingPtr = thing;
4165 *(uint64_t*)thingPtr = thing;
4169 if(relocs[i].r_address & R_SCATTERED)
4171 struct scattered_relocation_info *scat =
4172 (struct scattered_relocation_info*) &relocs[i];
4176 if(scat->r_length == 2)
4178 unsigned long word = 0;
4179 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4180 checkProddableBlock(oc,wordPtr);
4182 // Note on relocation types:
4183 // i386 uses the GENERIC_RELOC_* types,
4184 // while ppc uses special PPC_RELOC_* types.
4185 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4186 // in both cases, all others are different.
4187 // Therefore, we use GENERIC_RELOC_VANILLA
4188 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4189 // and use #ifdefs for the other types.
4191 // Step 1: Figure out what the relocated value should be
4192 if(scat->r_type == GENERIC_RELOC_VANILLA)
4194 word = *wordPtr + (unsigned long) relocateAddress(
4201 #ifdef powerpc_HOST_ARCH
4202 else if(scat->r_type == PPC_RELOC_SECTDIFF
4203 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4204 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4205 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4207 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4210 struct scattered_relocation_info *pair =
4211 (struct scattered_relocation_info*) &relocs[i+1];
4213 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4214 barf("Invalid Mach-O file: "
4215 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4217 word = (unsigned long)
4218 (relocateAddress(oc, nSections, sections, scat->r_value)
4219 - relocateAddress(oc, nSections, sections, pair->r_value));
4222 #ifdef powerpc_HOST_ARCH
4223 else if(scat->r_type == PPC_RELOC_HI16
4224 || scat->r_type == PPC_RELOC_LO16
4225 || scat->r_type == PPC_RELOC_HA16
4226 || scat->r_type == PPC_RELOC_LO14)
4227 { // these are generated by label+offset things
4228 struct relocation_info *pair = &relocs[i+1];
4229 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4230 barf("Invalid Mach-O file: "
4231 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4233 if(scat->r_type == PPC_RELOC_LO16)
4235 word = ((unsigned short*) wordPtr)[1];
4236 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4238 else if(scat->r_type == PPC_RELOC_LO14)
4240 barf("Unsupported Relocation: PPC_RELOC_LO14");
4241 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4242 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4244 else if(scat->r_type == PPC_RELOC_HI16)
4246 word = ((unsigned short*) wordPtr)[1] << 16;
4247 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4249 else if(scat->r_type == PPC_RELOC_HA16)
4251 word = ((unsigned short*) wordPtr)[1] << 16;
4252 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4256 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4263 continue; // ignore the others
4265 #ifdef powerpc_HOST_ARCH
4266 if(scat->r_type == GENERIC_RELOC_VANILLA
4267 || scat->r_type == PPC_RELOC_SECTDIFF)
4269 if(scat->r_type == GENERIC_RELOC_VANILLA
4270 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4275 #ifdef powerpc_HOST_ARCH
4276 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4278 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4280 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4282 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4284 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4286 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4287 + ((word & (1<<15)) ? 1 : 0);
4293 continue; // FIXME: I hope it's OK to ignore all the others.
4297 struct relocation_info *reloc = &relocs[i];
4298 if(reloc->r_pcrel && !reloc->r_extern)
4301 if(reloc->r_length == 2)
4303 unsigned long word = 0;
4304 #ifdef powerpc_HOST_ARCH
4305 unsigned long jumpIsland = 0;
4306 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4307 // to avoid warning and to catch
4311 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4312 checkProddableBlock(oc,wordPtr);
4314 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4318 #ifdef powerpc_HOST_ARCH
4319 else if(reloc->r_type == PPC_RELOC_LO16)
4321 word = ((unsigned short*) wordPtr)[1];
4322 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4324 else if(reloc->r_type == PPC_RELOC_HI16)
4326 word = ((unsigned short*) wordPtr)[1] << 16;
4327 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4329 else if(reloc->r_type == PPC_RELOC_HA16)
4331 word = ((unsigned short*) wordPtr)[1] << 16;
4332 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4334 else if(reloc->r_type == PPC_RELOC_BR24)
4337 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4341 if(!reloc->r_extern)
4344 sections[reloc->r_symbolnum-1].offset
4345 - sections[reloc->r_symbolnum-1].addr
4352 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4353 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4354 void *symbolAddress = lookupSymbol(nm);
4357 errorBelch("\nunknown symbol `%s'", nm);
4363 #ifdef powerpc_HOST_ARCH
4364 // In the .o file, this should be a relative jump to NULL
4365 // and we'll change it to a relative jump to the symbol
4366 ASSERT(word + reloc->r_address == 0);
4367 jumpIsland = (unsigned long)
4368 &makeSymbolExtra(oc,
4370 (unsigned long) symbolAddress)
4374 offsetToJumpIsland = word + jumpIsland
4375 - (((long)image) + sect->offset - sect->addr);
4378 word += (unsigned long) symbolAddress
4379 - (((long)image) + sect->offset - sect->addr);
4383 word += (unsigned long) symbolAddress;
4387 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4392 #ifdef powerpc_HOST_ARCH
4393 else if(reloc->r_type == PPC_RELOC_LO16)
4395 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4398 else if(reloc->r_type == PPC_RELOC_HI16)
4400 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4403 else if(reloc->r_type == PPC_RELOC_HA16)
4405 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4406 + ((word & (1<<15)) ? 1 : 0);
4409 else if(reloc->r_type == PPC_RELOC_BR24)
4411 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4413 // The branch offset is too large.
4414 // Therefore, we try to use a jump island.
4417 barf("unconditional relative branch out of range: "
4418 "no jump island available");
4421 word = offsetToJumpIsland;
4422 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4423 barf("unconditional relative branch out of range: "
4424 "jump island out of range");
4426 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4431 barf("\nunknown relocation %d",reloc->r_type);
4439 static int ocGetNames_MachO(ObjectCode* oc)
4441 char *image = (char*) oc->image;
4442 struct mach_header *header = (struct mach_header*) image;
4443 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4444 unsigned i,curSymbol = 0;
4445 struct segment_command *segLC = NULL;
4446 struct section *sections;
4447 struct symtab_command *symLC = NULL;
4448 struct nlist *nlist;
4449 unsigned long commonSize = 0;
4450 char *commonStorage = NULL;
4451 unsigned long commonCounter;
4453 for(i=0;i<header->ncmds;i++)
4455 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4456 segLC = (struct segment_command*) lc;
4457 else if(lc->cmd == LC_SYMTAB)
4458 symLC = (struct symtab_command*) lc;
4459 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4462 sections = (struct section*) (segLC+1);
4463 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4467 barf("ocGetNames_MachO: no segment load command");
4469 for(i=0;i<segLC->nsects;i++)
4471 if(sections[i].size == 0)
4474 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4476 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4477 "ocGetNames_MachO(common symbols)");
4478 sections[i].offset = zeroFillArea - image;
4481 if(!strcmp(sections[i].sectname,"__text"))
4482 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4483 (void*) (image + sections[i].offset),
4484 (void*) (image + sections[i].offset + sections[i].size));
4485 else if(!strcmp(sections[i].sectname,"__const"))
4486 addSection(oc, SECTIONKIND_RWDATA,
4487 (void*) (image + sections[i].offset),
4488 (void*) (image + sections[i].offset + sections[i].size));
4489 else if(!strcmp(sections[i].sectname,"__data"))
4490 addSection(oc, SECTIONKIND_RWDATA,
4491 (void*) (image + sections[i].offset),
4492 (void*) (image + sections[i].offset + sections[i].size));
4493 else if(!strcmp(sections[i].sectname,"__bss")
4494 || !strcmp(sections[i].sectname,"__common"))
4495 addSection(oc, SECTIONKIND_RWDATA,
4496 (void*) (image + sections[i].offset),
4497 (void*) (image + sections[i].offset + sections[i].size));
4499 addProddableBlock(oc, (void*) (image + sections[i].offset),
4503 // count external symbols defined here
4507 for(i=0;i<symLC->nsyms;i++)
4509 if(nlist[i].n_type & N_STAB)
4511 else if(nlist[i].n_type & N_EXT)
4513 if((nlist[i].n_type & N_TYPE) == N_UNDF
4514 && (nlist[i].n_value != 0))
4516 commonSize += nlist[i].n_value;
4519 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4524 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4525 "ocGetNames_MachO(oc->symbols)");
4529 for(i=0;i<symLC->nsyms;i++)
4531 if(nlist[i].n_type & N_STAB)
4533 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4535 if(nlist[i].n_type & N_EXT)
4537 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4538 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4539 ; // weak definition, and we already have a definition
4542 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4544 + sections[nlist[i].n_sect-1].offset
4545 - sections[nlist[i].n_sect-1].addr
4546 + nlist[i].n_value);
4547 oc->symbols[curSymbol++] = nm;
4554 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4555 commonCounter = (unsigned long)commonStorage;
4558 for(i=0;i<symLC->nsyms;i++)
4560 if((nlist[i].n_type & N_TYPE) == N_UNDF
4561 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4563 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4564 unsigned long sz = nlist[i].n_value;
4566 nlist[i].n_value = commonCounter;
4568 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4569 (void*)commonCounter);
4570 oc->symbols[curSymbol++] = nm;
4572 commonCounter += sz;
4579 static int ocResolve_MachO(ObjectCode* oc)
4581 char *image = (char*) oc->image;
4582 struct mach_header *header = (struct mach_header*) image;
4583 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4585 struct segment_command *segLC = NULL;
4586 struct section *sections;
4587 struct symtab_command *symLC = NULL;
4588 struct dysymtab_command *dsymLC = NULL;
4589 struct nlist *nlist;
4591 for(i=0;i<header->ncmds;i++)
4593 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4594 segLC = (struct segment_command*) lc;
4595 else if(lc->cmd == LC_SYMTAB)
4596 symLC = (struct symtab_command*) lc;
4597 else if(lc->cmd == LC_DYSYMTAB)
4598 dsymLC = (struct dysymtab_command*) lc;
4599 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4602 sections = (struct section*) (segLC+1);
4603 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4608 unsigned long *indirectSyms
4609 = (unsigned long*) (image + dsymLC->indirectsymoff);
4611 for(i=0;i<segLC->nsects;i++)
4613 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4614 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4615 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4617 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4620 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4621 || !strcmp(sections[i].sectname,"__pointers"))
4623 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4626 else if(!strcmp(sections[i].sectname,"__jump_table"))
4628 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4634 for(i=0;i<segLC->nsects;i++)
4636 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4640 #if defined (powerpc_HOST_ARCH)
4641 ocFlushInstructionCache( oc );
4647 #ifdef powerpc_HOST_ARCH
4649 * The Mach-O object format uses leading underscores. But not everywhere.
4650 * There is a small number of runtime support functions defined in
4651 * libcc_dynamic.a whose name does not have a leading underscore.
4652 * As a consequence, we can't get their address from C code.
4653 * We have to use inline assembler just to take the address of a function.
4657 static void machoInitSymbolsWithoutUnderscore()
4659 extern void* symbolsWithoutUnderscore[];
4660 void **p = symbolsWithoutUnderscore;
4661 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4663 #undef SymI_NeedsProto
4664 #define SymI_NeedsProto(x) \
4665 __asm__ volatile(".long " # x);
4667 RTS_MACHO_NOUNDERLINE_SYMBOLS
4669 __asm__ volatile(".text");
4671 #undef SymI_NeedsProto
4672 #define SymI_NeedsProto(x) \
4673 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4675 RTS_MACHO_NOUNDERLINE_SYMBOLS
4677 #undef SymI_NeedsProto
4682 * Figure out by how much to shift the entire Mach-O file in memory
4683 * when loading so that its single segment ends up 16-byte-aligned
4685 static int machoGetMisalignment( FILE * f )
4687 struct mach_header header;
4690 fread(&header, sizeof(header), 1, f);
4693 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4694 if(header.magic != MH_MAGIC_64)
4697 if(header.magic != MH_MAGIC)
4701 misalignment = (header.sizeofcmds + sizeof(header))
4704 return misalignment ? (16 - misalignment) : 0;